Citation
A psychometric evaluation of the correctional adjustment checklist

Material Information

Title:
A psychometric evaluation of the correctional adjustment checklist
Creator:
Hines, Brainard Willem, 1945-
Publication Date:
Language:
English
Physical Description:
x, 132 leaves : ; 28 cm.

Subjects

Subjects / Keywords:
Community colleges ( jstor )
Criminals ( jstor )
Estimate reliability ( jstor )
Graduates ( jstor )
Higher education ( jstor )
Human resources ( jstor )
Linear programming ( jstor )
Operations research ( jstor )
Psychometrics ( jstor )
Statistical discrepancies ( jstor )
Crime -- Classification ( lcsh )
Criminals -- Classification ( lcsh )
Dissertations, Academic -- Foundations of Education -- UF ( lcsh )
Foundations of Education thesis Ph. D ( lcsh )
Psychometrics ( lcsh )
Genre:
bibliography ( marcgt )
non-fiction ( marcgt )

Notes

Thesis:
Thesis--University of Florida.
Bibliography:
Bibliography: leaves 123-131.
General Note:
Typescript.
General Note:
Vita.
Statement of Responsibility:
by Brainard Willem Hines.

Record Information

Source Institution:
University of Florida
Holding Location:
University of Florida
Rights Management:
The University of Florida George A. Smathers Libraries respect the intellectual property rights of others and do not claim any copyright interest in this item. This item may be protected by copyright but is made available here under a claim of fair use (17 U.S.C. §107) for non-profit research and educational purposes. Users of this work have responsibility for determining copyright status prior to reusing, publishing or reproducing this item for purposes other than what is allowed by fair use or other copyright exemptions. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder. The Smathers Libraries would like to learn more about this item and invite individuals or organizations to contact the RDS coordinator (ufdissertations@uflib.ufl.edu) with any additional information they can provide.
Resource Identifier:
023378836 ( ALEPH )
06766157 ( OCLC )

Downloads

This item has the following downloads:


Full Text














A PSYCHOMETRIC EVALUATION CF THE CORRECTIONAL ADJUSTMENT CHECKLIST










BY

BRAINARD WILLEM HINES


A DISSERTATION PRESENTED TO THE GRADUATE
COUNCIL OF THE UNIVERSITY OF FLORIDA
IN PARTIAL FULFILLMENT OF THE REQUIREMENTS
FOR THE DOCTOR OF PHILOSOPHY DEGREE



UNIVERSITY OF FLORIDA


1980
















ACKNOWLEDGMENTS


My sincerest thanks go to the members of my doctoral committee for their patience and understanding. Particularly, I would like to thank the chairman of my committee, Dr. William Ware who has been a good friend and constructive influence throughout my academic career at the University of Florida. I also owe a special debt of gratitude to Dr. Linda Crocker and Dr. Richard Swanson for their support and encouragement.

In addition, I would like to express my gratitude to the University of North Carolina at Chapel Hill for allowing Dr. Ware to continue as my chairman for the past months.

Finally, I would li]~e to express my appreciation to

-my wife Magdalena Llabre, whose love and encouragement made this dissertation possible.


ii














TABLE OF CONTENTS


Page

ACKNOWLEDGMENTS.....................iii

LIST OF TABLES.......................V

ABSTRACT.......................Vii

Chapter

I. INTRODUCTION..................1

Psychometric Properties Investigated . . . 4
Definitions of Reliability.........4 Definitions of Validity ...........5
Construct Validation...............7
Statement of the Problem.............8
Significance of the Study...........10

II. REVIEW OF THE LITERATURE............12

Classification................13
Classification of Criminals .........15 Empirically Derived Typologies.......19 Current Reviews of Criminal Typologies 21 Psychometric Concepts.............30
Reliability Estimation in This Study ..37 Validity...................41
Types of Validity..............44
Construct Validity Estimates........46
Chapter Summary..................51

III. METHOD........................53

The Sample..................53
Selection of the Sample...........55
Instrumentation..............59
Data Collection................66
Data Analysis.................69
Reliability of the CACL...........69
Predictive Validity of the CACL........71


iii










Construct Validation of the CACL .......73 Postdiction of Crime Type..........74

Summary....................75

IV. RESULTS....................77

Inter-Rater Reliability of the CACL ....78 Construct Validation of the CACL ........79

Criterion-Related Validity of the CACL . . 85 Relationship of the CACL to Crime . . . . 91

Summary....................93

V. DISCUSSION....................94

The Inter-Rater Reliability of the CACL ..96 Construct Validation of the CACL ........96 Criterion Validity of the CACL.........101

Suicide Attempts..............102
Threats of Assault.............104
Assaults..................105
Infractions of Rules............105

Relation of the CACL to Crime Type. .....106 Summary of Psychometric Evaluation. .....107 Recommendations...............110

Appendix

A. CORRECTIONAL ADJUSTMENT CHECKLIST.......11.3

B. SUMMARY TABLES FOR INTER-RATER
RELIABILITY STUDIES..............116

REFERENCES........................123

BIOGRAPHICAL SKETCH..................132


iv















LIST OF TABLES


TABLE PAGE

1. NUMBER OF RESIDENTS BY UNIT ADMITTED TO
NFETC FROM ITS INCEPTION UNTIL JULY 1,
1978.......................58

2. DESCRIPTIVE STATISTICS FOR CONCURRENT
VALIDITY STUDY..................80

3. INTERCORRELATION MATRIX FOR CONCURRENT
VALIDATION STUDY................81

4. RESULTS OF CANONICAL CORRELATION ANALYSIS
OF THE CACL AND M.MPI..............83

5. CANON4ICAL WEIGHTS OF M.MPI AND CACL SUBTESTS FOR CANONICAL VARIATES 1 AND 2 .... 84

6. PRODUCT MOMENT CORRELATION BETWEEN SUBTESTS OF THE CACL AND M.MPI AND CANONICAL VARIATES...................86

7. DESCRIPTIVE STATISTICS FOR PREDICTIVE
VALIDITY STUDY......................87

8. INTERCORRELATION MATRIX FOR PREDICTIVE
VALIDITY STUDY..................88

9. RESULTS OF MULTIPLE REGRESSION OF
FREQUENCY OF SUICIDE ATTEMPTS ON
CACL SUBSCALES..................89

10. RESULTS OF MULTIPLE REGRESSION OF FREQUENCY OF ASSAULTS ON CACL SUBSCALES.....................89

11. RESULTS OF MULTIPLE REGRESSION OF
FREQUENCY OF THREATS OF ASSAULT
ON CACL SUBSCALES.................90

12. RESULTS OF MULTIPLE REGRESSION OF
FREQUENCY OF INFRACTIONS ON CACL
SUBSCALES...................90


V












13. RESULTS FOR DISCRIMINANT FUNCTION
ANALYSIS OF CRIME TYPE........ . .. ....


APPENDIX

14. DESCRIPTIVE STATISTICS FOR INTERRATER RELIABILITY STUDY: "INTAKE''
CONDITION ...............

15. DESCRIPTIVE STATISTICS FOR INTERRATER RELIABILITY STUDY: "CONTROLLED'
CONDITION . . . . . . . . . . . . . . .


92


116


116


117




. . . . 117


16. ANALYSIS OF VARIANCE TABLE FOR CACL
PA "CONTROLLED" CONDITION INTERRATER RELIABILITY STUDY

17. ANALYSIS OF VARIANCE SUMMARY TABLE
FOR CACL ID "CONTROLLED" CONDITION
INTER-RATER RELIABILITY STUDY ..

18. ANALYSIS OF VARIANCE TABLE FOR CACL
NA "CONTROLLED" CONDITION INTERRATER RELIABILITY STUDY .....

19. ANALYSIS OF VARIANCE TABLE FOR CACL
MA "CONTROLLED" CONDITION INTERRATER RELIABILITY STUDY .....

20. ANALYSIS OF VARIANCE SUMMARY TABLE
FOR CACL PA 'INTAKE" CONDITION
INTER-RATER RELIABILITY STUDY


118


118


119


21. ANALYSIS OF VARIANCE SUMMARY TABLE
FOR CACL ID "INTAKE" CONDITIONS
INTER-RATER RELIABILITY STUDY..............119

22. ANALYSIS OF VARIANCE SUMMARY TABLE
FOR CACL NA "INTAKE" CONDITION
INTER-RATER RELIABILITY STUDY............120

23. ANALYSIS OF VARIANCE SUMMI ARY TABLE
FOR CACL MA "INTAKE" CONDITION
INTER-RATER RELIABILITY STUDY.........120


24. INTER-RATER RELIABILITY COEFFICIENTS
FOR INTAKE AND CONTROLLED CONDITIONS,
INCLUDING SYSTEMATIC RATER BIAS IN
THE ERROR TERM............. 12


vi


. . . 121
















Abstract of Dissertation Presented to the
Graduate Council of the University of Florida
in Partial Fulfillment of the Requirements
for the Degree of Doctor of Philosophy

A PSYCHOMETRIC EVALUATION OF THE CORRECTIONAL ADJUSTMENT CHECKLIST

BY

Brainard Willem Hines

June 1980



Chairman: Professor William B. Ware Major Department: Foundations of Education


A variety of classification systems have been developed for use in the social sciences. These systems have become increasingly complex with the advent of modern statistical techniques. The use of classification systems in the field of criminology began with the effort to discriminate between criminals and "normals" on the basis of physical features. Although more recent classification systems in the field of criminology attempt to define types of criminals or criminal behavior, few have been adequately evaluated in terms of the psychometric properties of reliability and validity.

The Correctional Adjustment Checklist (CACL) is a

factor-analytically derived classification instrument which is designed to describe the behavior of incarcerated males vii












along four dimensions. These dimensions have been labelled Psychopathic-Aggressive (PA) , Neurotic-Anxious (NA), Immature-Dependent (ID) , and Manipulative (Ma). Ratings along these dimensions are intended to have differential implications for the management and treatment of individuals in close confinement. Although the instrument has been used in a variety of settings, little information is available on its inter-rater reliability or validity.

This study attempted to evaluate the CACL by assessing the degree of congruence among raters in a naturalistic setting and under conditions designed to provide maximal reliability estimates. Data gathered under both conditions provided reliability estimates for the average of three raters which ranged upwards of .60, with the exception of the Ma subscale, which showed a lower inter-rater reliability estimate in the "controlled" condition.

Assessment of the validity of the CACL with this

example involved estimating the relationship between it and other variables of interest. In this study, these variables were as follows: scores on the M4MPI administered concurrently with the CACL; the frequency of several types of disruptive behavior during the first sixty days after the CACL was administered; and the degree of violence involved in the crime with which the subjects had been most recently charged.


viii











These estimates of the CACL's relationship with other variables are statistically significant in several instances. First, a canonical variate analysis derived two sets of variables from the CACL and fAMPI. Although both canonical correlation coefficients are significant at the .05 level, a redundancy analysis indicates that the relationship between the two instruments is very modest. Also, scores on the CACL showed a statistically significant relationship to suicide attempts and threats

of violence which occurred within the first sixty days.

When the subscales of the CACL were used as the predictors in a multiple regression analysis, the NA subscale showed the highest degree of association with suicide attempts, followed by the PA subscale. Additionally, the PA subscale is the only subtest which accounts for a statistically significant amount of variance in verbal threats of physical violence. other disruptive behaviors (actual assaults and other infractions) were not significantly related to scores on any of the CACL's subscales.

A discriminant function analysis did not show any significant relationship between subscale scores on the CACL and the presence of physical violence in the subjects' most recent crime. This may have been due to the imprecise match between the charge (e.g. , armed robbery) and the actual degree of violence in the crime.












In summary, the CACL provides subscale scores which are reliable across raters, and which predict several behaviors of interest within a maximum security mental hospital setting. It shows a modest degree of redundancy with the MNMPI, indicating that it may well be measuring factors not being tapped by that instrument. Although the CACL was developed for the classification of a general prison population, it appears to have utility when used with individuals who are emotionally disturbed.


x















CHAPTER I


INTRODUCTION


The process of classification of objects or events

is essential to the development of any science. Although the origin of taxonomy or classification goes back to the ancient Greeks, the advent of modern statistical techniques and the use of high-speed computers have allowed for the use of more sophisticated classification methods than have been previously possible. In many areas, such as entomology, the development of more sophisticated taxonomic systems has contributed to the general advancement of the field in question.

The use of classification in criminology can be traced to a number of physiologists such as Lambroso who, in the nineteenth century, attempted to define a criminall type" on the basis of physical features. Such typologies were intended to discriminate between criminals and "normals," not to classify types of individuals who had been convicted of crimes nor relate those types to other measures of any sort.

With the modern emphasis on treatment rather than custody of criminals has come a concern for possible subtypes of offenders. Additionally, the more


1









2


humanitarian philosophy of our own era has encouraged the scientific study of criminal behavior and the types of individuals who become criminals. Such efforts to create meaningful offender typologies (Gibbons, 1975) have come about because of the failure of unitary treatment approaches and because of the observed variance in types of crimes, demographic and personal or behavioral characteristics of criminals.

Although a variety of offender typologies have been proposed, none has been widely accepted. Many typologies either are based on traditional psychological personality types or are concerned with classifying offenders based on the type of crime which they have committed. Other classification systems have been impressionistic and have included a variety of types which have not been found by other investigators (Gibbons, 1975) . Generally, no single typology of criminals or criminal behavior has been found to be of use in a variety of settings or with various age groups of individuals. Also, no single typology has been constructed which is of use in delineating both the etiology and diagnostic category of criminal behavior.

Most criminal typologies are based on the results of a single instrument or a series of descriptions of the crime or its etiology. Few classification systems used in criminology are based on empirically derived methods, but rather are derived from theoretical formulations. Quay







3


(1971) has made one of the few attempts to empirically construct a classification system for criminals.

The Quay Correctional Adjustment Checklist (CACL) is an instrument derived using factor analysis for the purpose of describing the behavior of incarcerated individuals on four dimensions (Quay, 1971). These dimensions are labelled Psychopathic-Aggressive (PA) , Neurotic-Anxious

(NA) , Immature-Dependent (ID) , and Manipulative (Ma) . It is intended not only to describe an individual's patterns of behavior within an institution, but also to provide information useful for differential treatment based on those patterns.

Although the CACL has been used in a variety of settings, its psychometric properties have never been thoroughly investigated. A review by Warren (1969) reported that the CACL appeared to have "adequate" reliability but gave no source for that statement. Another article by Quay (1971) has called for further study of the Checklist, but gave no validity estimates for the instrument.

The purpose of this study is to investigate the

inter-rater reliability and the validity of this instrument, based on the behavior of a sample of individuals who have been confined in a maximum security mental hospital in Gainesville, Florida. All of those individuals have either been convicted of a felony, have been found incompetent to stand trial for a felony, or are not guilty







4


by reason of insanity. Data on the CACL have never been gathered on a psychiatric population (Quay, personal communication, 1978) . If adguate reliability and validity estimates are obtained for the sample, the CACL should be used in other such settings.


Psychometric Properties Investigated Definitions of Reliability

As Kerlinger and Pedhazur (1973) pointed out, there are a variety of definitions of reliability. In general, they defined reliability as the consistency and accuracy of an instrument which are related to the absence of random or error variance in that instrument. Specifically, he wrote that

...reliability can be defined as the relative absence

of errors of measurement in a measuring instrument" (p. 443).

The reliability of any measure can be thought of as

existing in any of several dimensions. These dimensions may involve consistency (freedom from measurement error) across time, across items in a single measure, across other forms of the test or across raters or scorers on the same form of the test. The types of reliability which correspond to the degree of consistency in each of these dimensions are known as test-retest (time) , internal consistency (items) , parallel forms (forms) and inter-rater reliability.

The central focus of this study is the consistency of the CACL scores across raters who have observed the









5


individual under similar circumstances and who have received similar training in the -use of the instrument. This type of reliability (inter-rater) is of paramount importance to the CACL, since it is intended to measure the presence of observable behaviors. If equally trained observers cannot agree on whether a particular behavior is present, then tisefulness of this instrument for any practical purpose is highly questionable.


Definitions of Validity

The Standards for Educational and Psychological Tests and Manuals (1974) , published by the American Psychological Association, stated:

Validity information indicates the degree to which
the test is capable of achieving certain aims.
Tests are used for several types of judgment, and
for each type of judgment a different type of
investigation is required to establish validity.
(p. 13)

That is, validation is defined as a process or activity performed on the data arising from a test. The manner in which the data are treated is intended to parallel an aspect of the intended use of the measure, or of its interpretability. The Standards publication goes on to list three aims of testing which correspond to three types of validation procedures.

1. The test user wishes to determine how an individual performs at present in a -universe of
situations that the test situation is claimed
to represent.









6


2. The test user w..ishes to forecast an individual's future standing or to estimate an individual's present standing of some variable
of particular significance that is different
from the test.

3. The test user wishes to infer the degree to
which the individual possesses some hypothetical trait or quality (construct) preserved
to be reflected in the test performance.
(p. 13)

The American Psychological Association and the American Educational Research Association have defined three basic types of validity: content, construct, and criterionrelated (APA, 1974). Among these three types, criterionrelated and construct validity are most appropriate in assessing the potential usefulness of the CACL.

Criterion-related validity encompasses both predictive and concurrent validity which reflect the correlation between scores on a test and performance on a criterion variable. In concurrent validation, measures on both test and criterion are obtained at approximately the same point in time; predictive validation occurs when the criterion measure is taken after the test in question. For this study, predictive criterion-related validity will be investigated by relating CACL scores at the time of intake to the frequencies of several types of disruptive behavior, recorded during the first sixty days of confinement at the hospital.









7


Construct Validation

Construct validation is a complex procedure attempting to ascertain the degree to which a measure empirically relates to a number of other variables which logically and deductively derive from the constr-uct which the instrument purports to measure. To do this, a construct validation study often involves an attempt to demonstrate that the trait is related to other variables which are logically inherent from the construct, and that variables which do not logically derive from the construct do not empirically relate to it.

In the same way, this study will assess the degree to which the CACL's subscales relate to other variables which logically derive from the constructs they purport to measure. That is, we would expect that individuals who score highly on the Psychopathic-Aggressive subscale would be more violent and disruptive than those who score highly on the Immature-Dependent subscale. In addition, they should more often take a leadership role and rely less on staff for advice than other types of individuals. Also, we would expect such individuals to be more frequently threatening to others, and to score more highly on those subscales of another instrument which measure impulsivity and hostility. If such relationships were evident, this would help in defining the nature of the basic traits being assessed by the CACL.









8


The analyses conducted in this study provide estimates of the relationship between the CACL and several behaviors which are of interest in the institution. They also provide an estimate of the nature and degree of relationship between the CACL and the Minnesota Multiphasic Personality Inventory, a self-report diagnostic instrument which has been used in other criminal classification systems.


Statement of the Problem

This study will be addressed to determining the psychometric properties of the Quay Correctional Adjustment Checklist based on the performance of a sample of individuals in a maximum security mental hospital. It will not address the decision rules used to classify individuals, but will deal only with the psychometric properties of the instrument. Specifically, it will attempt to answer the following questions:D

1. What is the degree of agreement among raters with similar training who rate individuals on each of the four subscales of the CACL? (Inter-rater reliability)

2. What is the degree of association between the

subscales on the CACL and the subscales of the MM.PI, when both instruments are administered concurrently? (Construct validity)

3. Is there a relationship between the various subscales of the CACL and the type of crime which caused the individuals' incarceration? (Construct validity)









9


4. What is the relationship between scores on the CACL and an index of disruptiveness within the institution? (Criterion-related and Construct validity)

The questions above are concerned respectively with inter-rater reliability and validity, both of which are important to the use of the CACL in institutional settings. It is important to note that the study is not designed to explore the -use of particular decision rules for classification. Rather, it is, concerned with the consistency and "interpretability" of scores on the CACL. Also, the reliability estimates which are given are for the average of three raters, where, each rater rates all individuals. These estimates are considerably higher than those which would be obtained for a single rater.

At a time when there is a clear need for adequate classification techniques in the area of corrections (Warren, 1969), the CACL presents some unique advantages and disadvantages. Although it has been used in a variety of settings, often for the purpose of classification for treatment, its psychometric properties remain for the most part unknown. If its reliability and validity are low, its use should be discontinued. If they are acceptable, the instrument's utility could be explored in other settings. These possibilities are further discussed in the next section.





10


Significance of the Study

As Gibbons (1975) has pointed out, there is increasing dissatisfaction with the process of classification in criminology and criminal justice. Many of the current problems with the medical model, often used in criminal classification, may well be due to the ineffectiveness of current diagnostic instruments and the consequent misclassification of many individuals. Despite the apparent failure of treatment strategies which presume a single type of offender, no type of classification system other than the CACL has evolved which is based on actual patterns of behavior in an institution.

Quay (1971) noted that: "Additional research with

respect to reliability and construct validity (of the CACL) is in order" (p. 11). Although this need has been recognized since the initial development of the instrument, such studies have not been forthcoming. Despite the fact that there has never been adequate assessment of the instrument's psychometric properties, the CACL has been used in a variety of institutions, including the Robert Kennedy Federal Youth Center in Morgantown, West Virginia, the North Florida Evaluation and Treatment Center in Gainesville, Florida, and the Federal Correctional Institution in Miami, Florida.

Studies such as this one are important for several

reasons. First, if the instrument misclassifies offenders,


w










11


it may be hindering their effective treatment. Such misclassification is unfair to individual offenders and to the society which supports such treatment efforts. Second, the continued use of an instrument with unknown psychometric properties may well contribute to the increasing disenchantment with differential treatment strategies for offenders. Third, although the instrument may have an important function in the derivation of new theories in criminology, such functions will be of little use until validity is established. Additionally, the instrument may have potential use in the assessment of treatment effectiveness for individuals as well as groups of offenders. If it provides a reliable and valid measure of institutional behavior, it could allow for improved monitoring of those behavioral changes which occur during incarceration.

In this chapter an overview of classification as a process has been presented, and the application of this process to the field of criminoloty has been summarized.

The psychometric properties of the CACL, which are investigated in this study, are summarized along with the implications of the study. The following chapter includes a review of the literature on classification as a logical process, its use in criminology, and on the psychometric properties of reliability and validity.















CHAPTER II


REVIEW OF THE LITERATURE


This study is intended to provide reliability and

validity estimates for the CACL, based on the behavior of a sample of individuals incarcerated in a maximum security mental hospital. Since the CACL is an instrument intended to classify criminals, the literature review first includes a summary of articles on classification as a field of study in its own right. This is followed by a more extensive review of the development of criminal typologies. The trend towards empirically developed rather than theoretically oriented systems is discussed. Next, the development of the CACL is outlined, and the instrument is compared with other classification systems which are based on self-report instruments rather than behavioral observation. The need for psychometric evaluation of the CACL is pointed out, and the importance of inter-rater reliability and further validation studies is emphasized. The final section of the literature review provides a brief review of the theoretical definitions of reliability and validity as they pertain to the CACL. The need for consistent rating of individuals is stressed, along with the need for meaningfulness and utility in the classifications which are derived from the instrument's


12









13


use. Thus, the inter-rater reliability and construct validity of the CACL are the areas of primary interest which are explored in this study. Although the criterionrelated validity of the CACL is also investigated, the relationship of the criterion variables to the constructs measured by instrument is also explored.


Classification

Classification may be defined as the arrangement of objects or events into sets on the basis of their common characteristics. This process has been part of the natural sciences for centuries, but has only recently become a field of study in its own right. As such, the term taxonomy has been used to mean the theoretical study of classification as it occurs in a variety of specific disciplines.

One of the most comprehensive reviews of taxonomy appeared in 1974 in which Sokal reviewed the purposes, development, and structure of any classificatory activity. This review covered several general areas which are applicable to the use of classification in criminology, particularly the criteria for a desirable taxonomic system and the major purposes of classification.

Sokal (1974) said that,

The paramount purpose of a classification is to
describe the structure and relationship of the
constituent objects to each other and to simplify
these relationships in such a way that general statements can be made about classes of events.
(p. 1116)









14


Implicit in this definition are several purposes of taxonomy. First, classification may be used to reveal the "true" relationships between objects or events by ordering them on the basis of common characteristics. Second, classification can be used to achieve economy of memory. By grouping single cases it provides the capacity to summarize information and to avoid repetition. Third, classification provides for ease of manipulation and facilitates information retrieval. It may be used to simplify problems in routing or delivery, to define political districts or to allow for cataloging printed materials. Finally, Sokal noted that classification systems have the primary scientific purpose of generating hypotheses, in that they should "stimulate interest as a means of furthering investigation" (p. 1117).

Sokal made several important points about the purpose and types of classification systems. He noted that classification systems may serve the purpose of economy of memory, reveal "natural" relationships between elements in each taxon, provide for ease of manipulation, and generate interest in new scientific problems. Classification systems may vary in the number of salient dimensions which they include and may be monothetic or polythetic in nature, depending on whether the elements in each taxon must share a common trait (in the former case) or whether an element may possess any combination of the traits (in the latter case).









15


In general, Sokal pointed out that the classification is emerging as a distinct discipline and that a "metatypology" or classification of classification systems is possible. An ideal classification system should accommodate all elements of the set of objects or events to be classified, and should enable the typologist to match the dimensions and specificity of the system to its intended use. These criteria will be used later in this study to evaluate the CACL as a typological instrument in the field of criminology.


Classification of Criminals

An excellent overview of the classification of criminals is provided by Schafer (1968), who not only provided a historical narrative of the major typologists, but also defined several categories of criminal typologies (pp. 143144) . These include legal typologies of crime type, multiple cause typologies, typologies based on sociological or psychological theories, typologies which stress physiological factors, and those which describe the longitudinal development of criminal behavior.

Included in this section are an overview of current criminal typologies which fall into these categories and an explanation of the relationship between the CACL and these typologies. Generally, it is of interest that the CACL does not fall readily into any of Schafer's categories, since it deals with the behavior of felons while









16


incarcerated rather than the longitudinal development of criminal behavior.

The categories developed by Schafer emphasize either the hypothetical "causes" of criminal behavior, the type of crime(s) committed, or attempt to relate the two in a single description of a criminal "role career." Although some typologies are most often used for reporting frequencies of particular criminal acts in a given geographic area and are thus primarily empirical, most of the other typologies reflect on underlying theory of the causes of criminal behavior.

Legal typologies represent monothetic classification systems in which crimes rather than criminals are classified. The FBI Uniform Crime Reports for the United States represent such a system. Schafer noted that although such systems have historical and legal interest, "They are technical divisions for the use of the administration of of justice and are not conceived of as explanations for behavior" (p. 146) . Such a typological system will be used in this study to relate crime types to CACL classifications.

Multiple-cause typologies stress the interaction of biological, social and psychological causes of criminal behavior. Such systems were first developed in Germany in the nineteenth century by theorists who emphasized the affective and motivational components of criminal behavior as they exerted their influence across the criminal's life








'7


span. This historical perspective was later used by Gibbons (1965, 1970) in his typology of criminal role careers. Gibbons, along withClinnard and Quinney (1967) based a major criminological text on a multiple-cause typology. However, in a later article, Gibbons (1975) has emphasized the difficulties in the use of such a

typo logy.

Sociological and psychological typologies both emphasize the hypothetical causes for criminal behavior. Sociological typologies attempt to delineate the external forces which contribute to criminal behavior, while psychological classification systems reflect the inner dynamics which may lead to such acts. Although these typologies reflect some of the most productive and extensive areas of offender classification, they also elicit some of the more veheMEnt opposition (Schafer, 1968, p. 155).

Two of the most prominent socioloqical typologies are those of Tappan (1967) and Thrasher (1963). These two individuals have attempted to relate criminal behavior to its social causes, but frequently derived hypothetical multiple factor typologies with little empirical verification (Schafer, 1968). This is a general problem with etiological typologies, since they limit validation studies to ex post facto designs.

Psychological typologies are exemplified by the work of Alexander and Staub (1956) and Abrahamson (1960) . Such









18


systems suffer from some of the same problems as those depending on more sociological explanations, since they are limited to the assessment of current psychological functioning. Current functioning in criminals may not reflect the dynamics in operation at the time the crime was committed.

Constitutional typologies have the most lengthy history in criminology, dating to Galen (circa, 150 A.D.) . This group of typologies centers around the biopsychological causes of crime, especially the morphology of the offenders. The works of Lambroso (1911) , Kretschmer (1925) , and Sheldon (1949, 1954) are typical of this area.

Although these authors consistently have tried to

relate body type to crime type, George Vald (1958) points out that "there is no present evidence at all of physical type, as such, having any consistent relation to legal and sociologically defined crime" (p. 129) . Thus, constitutional typologies ma! allow for the classification of criminals, but the resultant categories have no empirical relationship to any manifest behavior. This problem is not unique to constitutional typologies, as will be shown later.

Normative typologies attempt to define the criminal's total personality in an effort to identify the "types" for which a particular sentence is appropriate. As such they incorporate a variety of legal, sociological, and psychological typologies. German authors have been primarily responsible for work in this area which has been little used in America.









19


Life-trend typologies are similar to multiple-facet typologies, but stress the dynamic structural coherence of the individual criminal's way of life. They are typically more complex than Gibbon's "role-careers" in that they attempt to follow the criminal behavior which is not part of a criminal life style.

Authors such as Reckless (1967) and Clinnard (1963)

have developed systems of this type and have generally made a large impact in the field of criminology (Schafer, 1968, p. 6). This may well be due to the comprehensiveness of the system itself and the polythetic, multi-dimensional process which they use to classify offenders.


Empirically Derived Typologies

Given the problem inherent in the classification systems previously discussed, individuals such as Quay (1964) , Gibbons (1975) and Megargee (1977) have recommended the use of empirically derived typologies. In such systems criminals are grouped on the basis of current behavior or demographic variables, without first theorizing about the causes for antisocial behavior. Accordingly, the development of such typologies differs from that involved in more "theory-oriented systems."

The effort to develop empirically keyed classificatory systems involves the administration of an instrument to a group of offenders and the development of a classificatory system based on the results of that instrument.








20


The important distinction to be made is that such typologies assess the current responses of the individual and are of limited scope and purpose. They are designed primarily for their immediate rather than long-term utility value, and may not have a predetermined underlying construct which they attempt to measure. Examples of such instruments are the classificatory subscales of the MPI developed by Panton (1965, 1966, 1968, 1970); Quay's work on the CACL (1971) ; and Megargee's recent work (1977), which also uses items on the MMPI.

These classificatory techniques have several common elements: first, they are not based on a single etiological or explanatory construct; second, they use a single instrument or subscale of an extent measure; and third, they describe current levels of functioning of the individual. These instruments are usually constructed by relating items to an external criterion (i.e. , behavior in the institution) or to an internal criteria of factorial homogeneity, as is the case with the CACL and the Megargee MMPI system.

The CACL was developed by Quay as par t of such an empirically derived classification system. Basing his classification system on the techniques developed by Dewitt and Jenkins (1946) , Quay (1971) developed instruments assessing both current functioning (the CACL) and life history (CALH).









21


Although the development of the CACL will be discussed in greater detail in the next chapter, it should be mentioned here that the CACL is a behavioral checklist intended to assess patterns of current functioning while incarcerated. The instrument was norined on a prison population, and provides normalized T scores in each of four dimensions: Psychopathic-Aggressive (PA) ; Neurotic-Anxious

(NA) ; Immature-Dependent (ID) ; and Manipulative (Ma) . It is of interest that the CALH, a life-history checklist also groups criminals into these categories, and provides a "'situational" dimension, where the CACL does not.


Current Rev,;iews of Criminal Typologies

Megarcree (1977) considered both the substance and form of a taxonomic system for offenders. He listed seven criteria for "usefulness" (P. 108) of such a classification scheme which are ap follows:

1. The system should classify all of the offenders under consideration.

2. It should have clear operational definitions of types.

3. It should be reliable, especially across raters.

4. It should be valid (construct validity is implied).

5. It should be dynamic, reflecting changes in the individual.

6. It should carry implications for treatment.

7. It should be economical to administer.









22


These criteria do not stress the "theory building" function

of such a system stressed by Sokal (1974), Schafer (1968) and others, but rather emphasize the practical significance of the system. In this, Megargee is following the point of view expoused by Gibbons (1975) in moving away from theoretically oriented taxonomic systems.

Commenting on the CACL diagnostic system, Megargee

said: "Systems (such as the CACL) can reflect changes in the individual and typically have clear implications for differential treatment strategies" (p. 110) . Later, he stresses the training and supervision of raters necessary to the CACL system, and said that the development of his MMPI system was intended to "retain the advantages of the Quay . . . system land to be], . . . widely implemented with less cost and fewer trained personnel" (p. 110). However, he is equating a system based on a self-report device intended for psychiatric classification with a more direct system of behavioral monitoring. Thus, the comparison does not seem adequate in its inference that both are based on "personality characteristics of the offender" (p. 112) , except in the broadest sense.

Examples of various typologies have been also discussed in a review of Warren (1969) and in the proceedings of an NIMH conference on criminal typologies (1967) . Both of these reviews group typologies differently than does Schafer and elaborate other characteristics than those which he emphasized.









23


Warren discussed five groups of offender typologies which provide the background for her own classification system (p. 241). These typologies include the following:

1. Prior probability systems, which rank offenders on the expectancy of some future behavior, usually recidivism.

2. Reference group typologies, relating criminal behavior to the social norms of a specific group.

3. Behavior classifications, which are oriented to some aspect of the offender's behavior.

4. Psychiatrically-oriented approaches which seek to define the nature of any mental disorder underlying crime.

5. Social perception and interaction systems. Such typologies relate criminal behavior to specific social interactions, and to the criminal's perceptions of those interactions.

It is obvious that these groupings are poorly defined and that they frequently overlap, as Warren has admitted (p. 241). The reviewer continued to make several more valid points about the structure and function of offender typologies. Generally, Warren made the point that "each of the . . . classification systems is not equally relevant for all purposes" (p. 242).

Warren saw typologies as serving the purposes of either management" or "treatment" (p. 242) . She said,









24


It is possible for certain purposes to use a
classification system which . . .has no etiological reference, one which has no implications
for treatment, or one which is specific to an
institutional setting. (p. 243)

Her review, like others, pointed out the difficulties with typologies which emphasize etiological dimensions, and argued for the use of more effective systems for treatment. In Warren's view, any combination of several factors ma have caused the crime and it is necessary to specify the exact cause in order to change the behavior (p. 243). This view, although widely shared, has not led to more effective treatment of criminals. Warren reviewed several studies which indicated that no form of differential treatment has effectively reduced recidivism rates (p. 245).

Despite this fact, Warren remained optimistic that adequate typologies will reveal that treatment outcomes depend on characteristics of the offender which interact with characteristics of the treatment program. It is not surprising that her interpersonal maturity system is oriented to such a purpose. Unfortunately, no later articles have been published which report on whether her system was more effective than others.

Warren (1969) summarized the results of an NIMH study (1967) which attempted a cross-tabulation of many existing classification systems, including that of Quay. The resulting configuration of typologies or composite system revealed six "bands" which were judged to represent a









25


stable set of underlying characteristics of offenders (Warren, 1969, p. 249). The six categories common to the sixteen classification systems reviewed are as follows:

1. Band 1 - labeled the asocial type, included the CACL Psychopathic-Aggressive type. Such individuals are characterized as "primitive, underinhibited, impulsive, hostile, insecure, inadequate, maladaptive, demanding of immediate gratification and attention, thoroughly egocentric, etc." (Warren, p. 251).

2. Band 2 - labeled the conformist type, incorporated the CACL Immature-Dependent type. Persons in this band are characterized as "concerned with power, searching for structure, dominated by the need for social approval, rule-oriented, unable to empathize, having low, selfesteem" (Warren, p. 251).

3. Band 3 - labeled the antisocial manipulator,

included the CACL Dganipulative type. These offenders are described as "guilt-free, power-oriented, self-satisfied, non-trusting, emotionally insulated, cynical . . . and extremely hostile" (Warren, p. 252).

4. Band 4 - identified as the neurotic subtype,

including the CACL Neurotic-Anxious type. Such individuals are characterized by high levels of anxiety and are described as "intimidated, disturbed, anxious, depressed, and withdrawn" (Warren, p. 254).









26


5. Band 5 - labelled as the subcultural identifier. Such individuals are presumed to commit their crimes because of their integration of subcultural values conducive to crime. Individuals of this type are described as "loyal to their group, psychologically healthy, proud, adequate, suspicious of the authority system, having a stable family, have criminal attitudes, and accessible to new experiences" (Warren, p. 254).

6. Band 6 - labelled the situational offender. This grouping included the CALH situational type, and is characterized as "relatively normal, exposed to acute, severe stress, having no evidence of neurosis, having little prior criminal records, etc." (Warren, p. 255). Such persons are seen as reacting to an overwhelming, non-recurring emotional stress which led to committing their crime.

Unfortunately, these "bands" or subtypes were identified by an informal comparison rather than on the basis of the measurement of a heterogeneous group of offenders with the same group of classificatory instruments. That is, the bands were constructed intuitively rather than empirically. However, since the various classification systems were developed independently, it is possible that this consensus reveals the existence of separate constructs which differ across the bands. As Warren noted (p. 245), until an empirical study is done on a single population, the diagnostic bands described above will remain somewhat hypothetical and tentative.









27


Generally, Warren used this review to provide background f or her own diagnostic system, but she made several points pertinent to criminal classification as it exists today. She pointed out that "the classification systems are not equally relevant for all purposes" (p. 241) . In addition, this review indicated that an ideal typology would provide "an explanatory theory with the resulting aid to prediction, implications for management and treatment, greater precision for research" (p. 240) . Thus it does seem that Warren believes that a single system can meet these needs.

As Gibbons (1975) pointed out, there is an increasing disenchantment with all of the taxonomic systems described above. Most have failed to show any real usefulness in the treatment of criminal behavior. Although the authors of these systems have hoped for empirical verification of their systems, little evidence has been forthcoming. Even though these systems have stimulated some new research, and do provide several of the benefits outlined by Sokal (1974) , they have failed to show pragmatic usefulness (Schafer, 1968, p. 177).

Gibbons (1975) was also pessimistic about usefulness of current offender typologies. He said: "It is by no means clear that existing typologies are empirically precise" (p. 254). The reasons for this lack of clarity are several, according to Gibbons (p. 299) . Firstly, no









28


single typology subsumes all types of criminality. Secondly, new forms of lawbreaking may be emerging which do not fit traditional typologies. Thirdly, the patterns of behavior or etiology which most typologies hypothesize have yet to be found in the actual study of offenders.

Gibbons argued that this lack of satisfactory classification systems is due either to the faults in the systems themselves or to the possibility that criminal behavior develops in a unique manner in each individual. It is difficult to assume the latter case, however, until the former has been eliminated as a potential problem.

He concluded by noting:

Insofar as the search for typologies turn out to
be profitable in corrections, it will be as a
consequence of the further development of statistical classifications . . . Which involve] . .
the development of classificatory devices based
on specific groups of offenders within certain
limited correctional settings. (p. 245)

Thus Gibbons recommended turning away from theoretically derived typologies, especially those which center on the etiology of criminal behavior. His discussion indicated that the more any typology depends on retrospective investigation or hypothetical constructs, the less likely it is to produce meaningful results.

In summary, the literature on classification systems for offenders seems to support several overall trends. First, no single system can perform all of the functions necessary in the criminal justice system. IMonothetic,







29


crime-based systems are best suited to the needs of law enforcement agencies, while polythetic, treatment-oriented systems meet the needs of prison officials and program planners. Systems which have sought to delineate the causes of criminal behavior or to trace recurring patterns of adjustment prior to the offense have failed.

Second, since little empirical verification has been found for the theoretical constructs underlying many classification systems, the trend has been to attempt to define coherent sets of variables and to them explore the relations between these "categories" and other variables. That is, the usual process in developing an offender taxonomy has been to group individuals with similar crimes and then explore their similarities on other variables. As Megargee and Bohn noted (1977) , this technique has been singularly unproductive, and the "psychometric" method which this system and the CACL use reverses this process. That is, offenders are categorized on variables related to current functioning, and the resulting types are related to past behavior or to predict future adjustment (p. 155).

Third, a tentative list of criteria for judging a taxonomic system for offenders emerges from all the articles in this area. These criteria are as follows:

1. It should relate to other variables of interest, and as a consequence may have theory building value.









30


2. It should serve a specific purpose for limited population.

3. It should assess current functioning rather than past behavior.

4. It should classify all individuals in question

and no individual should be classified into more than one category.

5. It should have specific, clear-cut decision rules allocating individuals to categories.

This study will investigate the CACLpgimarily on

related criteria (1) and (5) . That is, the clearness of decision rules is reflected in the consistency of raters in assessing the behaviors in question. Thus the interrater reliability study will assess the clarity of the CACL's definitions, and the various validity studies will delineate the CACL's relationship with other variables. The following section will review the literature which provides the background for establishing the reliability and validity of the CACL.


Psychometric Concepts

This section reviews the concepts of reliability and validity as they pertain to this study. The classical theory of reliability and the major types of reliability estimates are reviewed. The factors affecting reliability and validity are then presented and the specific types of







31


reliability and validity estimates obtained in this study are discussed at greater length.

Reliability. The concept of reliability of measurement refers to its consistency across any of several dimensions. As several authors have pointed out, reliability, like validity, takes on special significance in the measurement of traits or inferred constructs (Stanley, 1969; Cureton, 1958) . Reliability has been of central importance in areas such as psychology and education, where indirect measurement is frequently employed.

Definitions of reliability. Classical measurement

theory has based the concept of reliability on the assumption that any measurement contains a discrete amount of random fluctuation or error in addition to the influence of the actual variable under consideration. As Stanley (1969) pointed out:

When a feature or attribute of anything (in any
of the sciences) is measured, that measurement
contains a certain amount of chance error. The
amount of chance error may be large or small,
but it is universally present. (p. 356)

Thus it is assumed that any observed score for an individual is composed of a true score component and an error score component which are linearly additive. Since the error score components are presumed to be random, they should not show any relationship to each other or to the true or observed scores. Cureton (1958) said:









32


The basic theorem which underlies all formulas
of reliability, and of empirical validity as
well, may be stated as follows: In a population
of individuals, the errors of measurement in
different tests and the different forms of the
same test are uncorrelated with one another and
are uncorrelated with the true scores on all
tests and forms. (p. 103)

The error of measurement referred to by Cureton is an estimate which relates to the variability in a series of repeated testings of the same sample due to random (error) fluctuations. That is, if a number of independent measurements are taken on the sa-me individuals, the variability in those measures would reflect the random fluctuations, or amount of error variance present in the measurements. The shared or common variance would reflect the amount of true score variability which was present. Since repeated testings of the same sample of individuals are not practical for a variety of reasons (the interactive effects of measurement, practice, etc.) , the errors of measurement must be estimated indirectly.

It is possible to see that the notion of variability of scores across any of several dimensions is central to the definition of reliability. Without observed variance in scores, the estimation of reliability is not possible. Again, since the true score variance and error score variance can never be assessed directly, one must attempt to estimate them from the observed variance in test scores. As Stanley (1969) emphasized:







33


The basic problem in defining the reliability of a
testing procedure . . . becomes that of defining
what shall be thought of an error variance in
relation to the type of inference one wishes to make from the test scores. When this definition
has been made, the next step is to devise those
series of empirical and statistical operations that
will provide the best estimates of the defined
fractions of variance. (p. 362)

Since variability in test scores can arise from a

variety of sources, the selection of which of these are to be considered as sources of error variance depends on the purpose of the testing. Stanley emphasized this point when he said:

There is no single universal and absolute reliability coefficient for a test . . . . The allocation of variance from different sources calls
for practical judgment of what use is to be made
of the resulting statistical value. (p. 363)

The reliability coefficient for any measure can be defined as that proportion of observed score variance which is composed of true score variance. Theoretically, then, the reliability coefficient can range from 0 to 1.00, where a zero reliability coefficient indicates an absence of variability attributable to true score differences, and where a reliability coefficient of one results from a complete absence of random, extraneous variability. The formula for this relationship can be expressed as:


R S t 2
tt S2
0









34


where R ttis the reliability coefficient, S t2 is the

true score variance, and S 02is the observed score variance. Since the observed score variance is presumed to be composed of a linear combination of true score variance and error variance (S 02= S e2+ S t2) , the formula may also be written as,

S 2 _ 2 2
R -0 e ore
Rtt S 2 orRtt 2
0 0

Factors affecting R tt. The presumed random nature and normal distribution of the error component influences the magnitude of the reliability coefficient in several ways. As the number of items in any measure increases, the errors will tend to cancel each other out to a greater degree. That is, as the number of items approaches infinity, the sum of the errors will tend to approach zero. Magnusson (1967) also noted that the error variance increases arithmetically with the length of the test while the true-score variance increases with the square of the number of items. Thus, "when the test is lengthened, the true variance increases at a faster rate than the error variance. This . . means that the test will become more reliable" (p. 72).

In addition, the homogeneity or amount of total variance in the sample also determines the magnitude of the reliability estimate. As the sample becomes more homogeneous,








35


the amount of true score variance decreases, while the error remains unchanged. This decrease results in a reduction of the magnitude of the reliability coefficient, since the ratio of true score variance total score variance has been decreased.

Types of reliability estimates. As mentioned previously, the particular source of total test variance which is considered as error depends on intended use of the instrument. For example, if a measure is intended to measure a single, unitary trait it is highly desirable that the items share as much common variance as possible. Again, if a test is intended to measure an enduring characteristic of the individual, it should have as much stability across time as possible.

Reliability estimates can be thought of as approximations of true-to-total variance proportions, where the priority of the use of the test determines which of the above will be considered as most important sources of true variance. The various types of reliability coefficients can be thought of as falling into several broad classes, based on the type of error which is considered most important to the measure in question. Cronbach (1960) has defined three such classes of reliability coefficients which he calls coefficients of stability, coefficients of equivalence, and coefficients of internal consistency.









36


1. Coefficients of stability estimate the consistency of test scores across time, and are particularly important in measuring the lasting characteristics or traits of individuals. Such coefficients are generated in a test-retest paradigm where the same instrument is given on several occasions.

2. Coefficients of equivalence are intended to measure the similarity of several forms of a specific test. That is, equivalence estimates are intended to measure the degree to which two tests are parallel--that is, having the same means, variances, and average item intercorrelations. It is also possible to consider inter-rater reliability as a type of equivalence estimate, although the same form of the instrument is used. Inter-rater reliability estimates compare the shared variance across several individuals who assess the same person at the same time and under tha same conditions.

3. Coefficients of internal consistency assess the

degree to which the items in a test measure the same trait, construct or characteristic. One estimate of internal consistency is obtained by dividing the responses to a test into two parts, and correlating the two halves with each other, and approximating the reliability of the total test by the use of the Spearman-Brown prophecy formula. This is known as a "split-half" reliability coefficient.









37


Various other indices of internal consistency have been developed, such as Cronbach's coefficient alpha (Cronbach, 1951) and the Kuder-Richardson (1937) formulas, which estimate the average of all possible split-half reliability coefficients of a given test. These coefficients will not be further discussed, since this study is concerned only with the consistency across raters, rather than internal consistency or stability across time. Reliability Estimation in This Study

The specific type of reliability which is of the greatest concern in this study is the degree to which equally trained independent observers agree on the presence of the behaviors assessed by the CACL. Although the instrument itself will be discussed further in the methodology section, it is important to note that it is neither a rating scale por a traditional observational instrument. Rather than counting the frequency of occurrence of specific behaviors or rating the individual along a theoretical continuum, the CACL is designed to determine whether a specific behavioral trait is characteristic of the individual (Quay, 1964).

In a recent article, Prick and Semmel (1978) made

several important points in regard to inter-observer agreement (reliability). They said:







38


Minimal observer disagreement is a necessary but
insufficient condition for high reliability
coefficients, since there are other components
of the generic error variance that are theoretically independent from observer error variance
(e.g., in trasubject variance from occasion to
occasion) . (p. 159)

In addition, the authors also note that although

observer or rater agreement is only a part of the reliability of observational data, it does set the upper limit for the reliability of the data under consideration. That is, until the observational systems capacity for interobserver agreement has been defined, it is difficult to determine the degree to which other factors are limiting the reliability of the data (pp. 160-161).

Frick and Semniel also point out that the traditional definition of reliability as agreement between measures which have identical content, means, variance, and item intercorrelations is impractical when applied to human raters. That is, observers or raters do not have identical or equivalent observational skills. Accordingly, intraclass correlation coefficients or generalizability coefficients have been proposed as techniques to determine the reliability of a set of data without depending on the above assumptions. Such coefficients have often been used in the analysis of classroom observation data, but are equally applicable to measurements from other sources (Haggard, 1958; McGaw, Wardrop, & Burda, 1972).








39


Such coefficients estimate the ratio of true-to-total variance, but use an analysis of variance model to estimate the relative contributions of various sources of error variance. Although a more detailed description of the technique used in this study will be given in the procedures section, it is of importance to reiterate that such analytic techniques are used since the traditional assumptions underlying reliability are not applicable to data arising from ratings or observations.

An earlier article by Ebel (1951) compared the advantages of the intraclass correlation coefficient with other methods for assessing the reliability of ratings. In recommending the intraclass coefficient, Ebel listed three major advantages of such an approach.

First, the intraclass formula permits the investigator to choose whether to include
"between raters" variance as part of the
error variance . . . . Second, a convenient
means for estimating the precision of the
reliability coefficients is available to
the user of the intraclass formula. Third,
the intraclass formula uses the familiar statistics and routine computational procedures of analysis of variance. (p. 423)

In a position paper, McGaw et al. (1972) made a distinction between reliability coefficient as calculated from the internal structure of a test, from repeated testings, or from parallel forms, contrasting these with indices of observer agreement. Antedating the views of Frick and Semmel, they noted that agreement between observers has all









40


too often been considered the only important aspect of the reliability estimation of observational data. Specifically, they say:

The confusion introduced into the literature
through failure to clearly distinguish the
different sources of unreliability, and
through over-emphasis on inter-judge agreement has resulted from a confusion of the
importance of primacy with prime importance.
Inter-judge agreement is the first, but not
the most important issue to be faced. (p. 16)

Thus for the current study, it is most important to

note that the inter-rater reliability (agreement) which is calculated is not to be considered the only aspect of stability of data arising from the CACL which should be studied. However, because of its importance it is the type of reliability to be examined in this study.

The inter-class correlation coefficients which were derived in the study are for the average of three raters, where each rater rates all subjects. These estimates are considerably higher than those which would be obtained for a single rater.

These coefficients are also calculated differently

when absolute rather than comparative decisions are being made. When absolute decisions are involved, systematic rater bias is included in the error term of the model. For comparative decisions, such bias is not included along with the subjects by rater interaction in the error term.









41


Validity

Most authors agree that validity, like reliability, is a general term for a variety of related processes which assess the "usefulness" of a test. Brown (1970) pointed out that validity analysis may answer any of the following questions:

How well does the test do the job it is employed
to do? What traits are being measured by the test? Is the test actually measuring what it
was designed to measure? Does the test supply
information that can be used in making decisions?
What interpretation can be given to the scores
on a test? What can be predicted from the test
scores? (p. 99)

That is, validity studies generally attempt to relate test scores to other variables of interest. In terms of true and error score variance, Brown said:

Whereas reliability was defined by the proportions of true and error variance, validity is determined by the proportion of true variance
that is relevant to the purposes of testing.
3(p. 98)

Thus, the process of validation usually involves assessing the relationship between the test and some external criterion.

The definitions of validity, which have been given

in the Standards for Educational and Psychological Tests, center around the process of estimating the usefulness or meaningfulness of the data from a particular instrument. Each of these definitions will be discussed at a later point, but it is important here to compare the definitions of validity held by other authors.






42


Ebel (1961) suggested that defining validity is more

difficult than it may appear at first glance. He pointed

out that various authors diverge widely in their definitions of validity, and as examples notes that:

Gullikesen .. . has said: "The validity of a
test is the correlation of the test with some
criterion." Cureton writes: "The validity of
a test is an estimate of the correlation between the raw test scores and the 'true' (that is perfectly reliable) criterion scores." Lindquist
suggests: "The validity of a test . . . (is)
...the accuracy with which it measures that
which it is intended to measure..
Edgerton suggests: "By validity we refer to the
extent to which the measuring device is useful for a given purpose." Cronbach explains: "The more fully and confidently a test can be interpreted, the greater its validity." (p. 75)

Ebel continued by defining three other problem areas

in the area of validity:

The fact that it must assume diverse forms to
fit diverse situations, the discrepancy between
the importance of test validity and the state of
the art of validation, and the fact that the
question of validity doesn't arise in the physical sciences. (pp. 76-78)

In addition, he pointed out that the concept of validity is

not philosophically adequate, in that it is unlikely that,

"the naive faith in the pre-existence of a quantity to be

measured is basic to the general conception of validity"

(p. 79).

Ebel also mentioned that these difficulties may well

be due to a variety of causes. First, he suggested that

although the relation between a test and criterion is

central to validity theory, the criterion, like the test









43


itself is most often constructed and thus of limited validity itself. In addition to the philosophic problems of a "true" score, Ebel also saw the concept as frequently overgeneralized and used in inappropriate settings.

As a solution to these problems, Ebel (1961) suggests that the term "meaningfulness" be used to subsume the concept of validity. That is, he suggested that the assessment of the relationship between test scores and other measures be one of factors which contribute to the interpretability of test scores. He recommended the other factors to be considered should be the reliability of the measure, the norms used, and the operational definition of the score itself.

Following Ebel's recommendations, this study is an assessment of the meaningfulness of the CACL. That is, scores on the CACL are related to other measures for a sample which differs from the norms and the reliability of the instrument is assessed. In this way, we have an indication of the usefulness of the instrument with a population having a high degree of psychopathology.

Magnusson (1967) said that validity, like reliability, is an aspect of dependability, and that "the validity of a method is the accuracy with which meaningful and relevant measurements can be made with it" (p. 124).

As mentioned above, the criterion measure may be a

test which has less than perfect validity and reliability







44


itself. Magnusson pointed out that although imperfect reliability can be corrected, "low validity in the criterion data, however, can never be corrected for . .*" (p. 127). often the question of how best to define the criterion variable is left essentially unanswered.


Types of Validity

Other authors concur that validity is most often concerned with the relationship between the test and other variables. Like reliability, this relationship can exist in any of several dimensions. Each of these dimensions covers a different aspect of validity, and may be thought of as the relationship between the test and a larger domain, other measures of the same trait, or the degree of "meaningfulness" of the test. The types of validity which correspond to those dimensions have been mentioned above and labelled by the American Psychological Association as content validity, criterion-related validity, and construct validity (APA, 1974).

The first of these concepts, content validity, refers to the adequacy with which a measure reflects the domain of items in question. Although content validity is an important area in the construction of achievement tests, it has little bearing on this study. Therefore, it will not be discussed at length.









45


Criterion related validity has been defined by Gaion (1974) as "the extent to which scores on one variable, usually a predictor, may be used to infer performance on a different and operationally independent variable called a criterion" (p. 288) . If the criterion measure is taken at the same point in time, the process is known as concurrent validation. if the measure is taken later, the process is known as predictive validation.

As has been mentioned previously, validation studies are intended to specify the "usefulness" of the test, or the degree to which it successfully accomplishes a given purpose. In a general review of validation, Cronbach (1960) equated criterion related validity with usefulness in selection and placement, both of which he subsumes under the process of decision making (p. 446).

It is important to note that criterion-related validity may be conceptualized as existing for a specific purpose and is empirically determined by the relationship between the test scores in question and a second criterion measure. In a brief review, Cureton (1958) said that the criterion may exist in the present or future, and may be pre-existing or constructed (p. 105).

Pre-existing criteria include those that exist without any special effort made to predict them. Examples of such criteria include graduation from college, number of previous criminal convictions, etc. Constructed criteria are









46


usually developed on the basis of some hypothetical trait concept, and include rating scales, intelligence measures and personality tests.

Criterion-related validation studies often numerically express the relationship between this test score and external measures in the form of a validity index, which represents the amount of variance common to the two. However, it is often presumed that the criterion measure is an adequate measure of the criterion when in reality this may not be the case. In an article on the problems inherent in criterionrealted validation, Brogden and Taylor (1950) defined "criterion bias" as "any variable, except errors of measurement and sampling errors, producing a deviation of obtained criterion scores from a hypothetical 'true' score criterion" (p. 82).

Although bias in the criterion which is not correlated with the predictor may undesirably affect validity studies of this type, Brogden and Taylor point out, "it is the presence of test-correlated bias that 'makes' or 'breaks'

the criterion" (p. 82).


Construct Validity Estimates

Unlike criterion related validity, construct validation procedures are often more conceptual than statistical. They attempt to assess the degree to which an instrument reflects an underlying construct or hypothetical trait. In a classic article, Cronbach and Meehl (1955) stated:









47


Construct validation is involved whenever the
test is to be interpreted as a measure of some
attribute or quality which is not "operationally defined" . . . . Construct validity must be investigated whenever no criterion or universe of content is accepted as entirely adequate to define the quality to be measured.
(p. 282)

The authors continued to point out that construct

validity is "not to be identified solely by the particular investigative procedures, but by the orientation of the investigator" (p. 281). That is, the procedure may incorporate concurrent or predictive methodologies, factor analysis, or other techniques to be discussed in this section. It is the aim or intent of the investigator that uniquely defines construct validation.

A number of procedures have been used in an effort to determine the usefulness of a given construct in interpreting test data. Cronbach and Meehl listed several such techniques which provide the basis for inferring the existence of a trait. These techniques include the following:

1. Studies of group differences which would be
expected on the basis of the construct in
question.

2. Correlations between items or tests which
reflect the same trait. The covariation
between such items or tests may be -measured by means of factor analysis and
correlation matrices.

3. Studies of the internal structure of the
measure in question. For-many constructs, evidence of homogeneity within the test is
relevant in judging validity.









48


4. Studies of change over occasions (retest
reliability) may lend support to the logical
network defining the construct.

5. Studies of the process of performing on the
measure in question may also help to define
the construct in question. (p. 289)

In a reformulation of the techniques mentioned above, Campbell and Fiske (1959) point out that although we often use measures of association (correlation) to assess the presence of a construct, we also often look for divergences in test performance. They define the two processes as in the following manner:

1. Validation is typically convergent, a confirmation by independent measuring procedures. Independence of methods is a common denominator among major types of
validity (excepting content validity)
insofar as they are to be distinguished
from reliability.

2. For the justification of novel trait measures, for the validation of test interpretation, or for the establishment of
construct validity, divergent validation
as well as divergent validation is required.
Tests can be invalidated by too high correlations with other tests from which they
were intended to differ. (p. 82)

That is, the process incorporating convergent and

divergent validation indices aids specifically in the logical interpretation of validation data. By demonstrating that different techniques intended to measure the same trait correlate significantly with each other, and that similar methods intended to measure different traits do not, powerful logical evidence for the traits' presence has been presented.









49


Following Campbell and Fiske's logic, it is evident that construct validation relies on both statistical and logical inferential techniques. That is, it uses empirical evidence to logically deduce the presence or absence of a specific trait. Unlike criterion-related validity, which relies heavily on statistical measures of association, the construct validity of an instrument is demonstrated through a series of analyses which are logically incorporated into the overall validation process.

Factor analysis is widely used in the determination

of construct validity. An early article by Guilford (1948) stressed the use of factor analysis in assessing the construct validity of an instrument. Guilford seemed to be anticipating the distinction between criterion-related and construct validity when he wrote of practical and factorial validity. He defined the factorial validity of a test as being determined by "its loadings on meaningful, common, reference factors" (p. 428).

Cattell (1964) also discussed the use of factor analysis in the determination of construct validity. As a type of convergent validity, he believed that factor analysis can help to define a construct when it emerges as a simple factor across several studies. This technique "combines measurement precision with unitary character, as well as a meaning enriched beyond that of an empirical construct" (p. 22).






50


Although Anastasi (1976) indirectly accepted the use of factor analysis in construct validation, particularly with reference to the measurement of general versus specific abilities, her overall stance has been strongly against anything other than criterion-related validity. She referred to "the will-o'-the-wisp" of psychological processes which are distinct from performance" (p. 77). Cronbach and Meehl (1955) disagree with this position, and point out that inference based on patterns of association between variables "cannot be dismissed as pure speculation" (p. 290).

The CACL was not developed to measure a prespecified underlying trait, but rather was developed through the factor analysis of a set of behavioral descriptors. However, the four subscales of the CACL have been given labels based on their content, and these have been shown to correspond to broader traits which have appeared throughout various classification systems for offenders. Thus, the validation process in this study will attempt to relate scores on the subscales of the CACL to other measures which may be indicative of those traits. In this sense, estimates of construct validity are of primary importance in this study. That is, it is most important to define the nature of traits measured by the instrument, rather than to only establish its estimates of criterion-related validity.









51


Chapter Summary

Included in this chapter is a review of the literature in three major areas which are pertinent to this study. First, the process of classification in general has been stummarized. Generally, classification systems serve many purposes and no single system can meet all the needs in any one area. Next, in reviewing the history of classification in the field of criminology, the problems with theoretically oriented typologies have been noted. Empirically derived classification systems such as the CACL after the advantage of proven utility for a specific population but need to be reevaluated before they are used with a group which differs from the normative sample.

Since the purpose of this study is to evaluate the psychometric proportion of the CACL based on the behavior of a group of mentally disordered criminals, the area of reliability and validity were reviewed at some length in this chapter. Particular emphasis is given to the topic of inter-rater reliability, which sets the upper line for the reliability of rating scale such as the CACL.

Criterion-related validity was also discussed at some length since the CACL is intended to facilitate decisions about future custody and treatment of individuals in confinement. This study relates the CACL to several criterion measures, including the MMPI and behavioral measures of disruptiveness.









52


Since these behavioral measures are of interest in their relation to the hypothetical traits measured by the CACL, the area of construct validity is also reviewed. Although the CACL is designed to describe patterns of behavior within the institution, it also labels these patterns in accordance with existing theories of criminal behavior. Thus, it may be used in "theory-building" studies rather than as a descriptive tool.















CHAPTER III


METHOD


The purpose of this study was to investigate the psychometric properties of the Correctional Adjustment Checklist (CACL), based on ratings of the behavior of a group of individuals confined in a maximum security mental hospital. Specifically, this study was designed to assess the inter-rater reliability of the instrument and to provide estimates of its construct and criterion-related validity when used with individuals showing evidence of various types of mental disorders.

The procedures used to obtain these estimates are

detailed in a description of the subjects, the instruments, and the analytic techniques used. Since the emphasis of this study was to evaluate the instrument when used with a group which is different from the normative sample, the description of the subjects which follows is of considerable importance.


The Sample

All subjects included in this study were housed in the North Florida Evaluation and Treatment Center CNFETC), which is a 225-bed maximum security mental hospital located in Gainesville, Florida. It is operated and administered


53









54


by the Department of Health and Rehabilitative Services of the State of Florida and is currently the only mental hospital in the state which serves a purely forensic population.

The hospital is composed of eleven residential and

treatment buildings, consisting of one to three nine-person

living areas which are known as "pods." Each patient (known as a resident) has a private room, and shares bathing facilities and a living area with the other residents in his pod. The hospital is divided into three units, each of which serves a particular type of client. Based on diagnostic categories, these types are as follows: psychotic, behaviorally disordered, or mentally disordered sex offenders.

Although all of the residents have been charged and

arrested for a major felony, not all have been tried, convicted, or sentenced. Those individuals who have been found incompetent to stand trial or to be sentenced are placed in the psychotic unit for short-term (averaging two months) treatment. Also, individuals who become psychotic while incarcerated are given similar short-term care. The Psychotic Unit currently includes ninety beds.

The Behavior Disorders Unit is comprised of forty-five

beds and is intended for the behavioral managEment and treatment of antisocial, retarded, or neurologically impaired individuals. Such persons are usually management









55


problems in the traditional prison system, and are sent to NFETC for short-term treatment of recurring problem behaviors.

The Sex Offender treatment unit includes ninety beds and is oriented to the long-term (approximately two years) treatment of individuals who have been convicted of a sexual offense and been classified under Florida Statute 917 as Mentally Disordered Sex offenders. The individuals so classified must be manifestly non-psychotic, and be judged by at least two psychiatrists to have a predisposition to commit other sexual offenses.

Overall, the population of the North Florida Evaluation and Treatment Center can be described as a group of approximately 225 males, all of whom have been arrested for a major felony and most of whom have been either found incompetent to stand trial or incompetent to be sentenced; who have become psychotic or a management problem while incarcerated; or who have been adjudicated as Mentally Disordered Sex Offenders. The age of the residents at the time of this study ranged from seventeen to seventy-nine, with a median age of twenty-eight, and they came from a wide variety of ethnic and social backgrounds within the state of Florida.


Selection of the Sample

From October 1976, when NFETC first began receiving

residents, until July 1, 1978, approximately 550 individuals






56


have been treated or evaluated in the institution, of these, approximately 325 have been treated and returned to the referring agency, while the remainder are still confined at the hospital.

The data which are available on these individuals

are a function of events which were not under the control of this author. Since the emphasis at this hospital is on treatment and effective management of residents, changes in intake and diagnostic procedures were made which did not allow data collection procedures which would have been optimal for this study.

For the first 14 months of operation (until January

1978) , the hospital included a central intake and diagnostic unit where all incoming residents were housed for shortterm evaluation and diagnosis. During their stay in the intake and diagnostic unit, the residents were assessed on a battery of diagnostic tests including the Minnesota 1lultiphasic Personality Inventory (MMPI), the Incomplete

Sentences Test, the Social Reaction Inventory, the Quay Correctional Adjustment Checklist (CACL) and Checklist for the analysis of Life History (CALH).

Since January 1978, the Intake and Diagnostic Unit

has been concerned with the evaluation of incoming sex offenders only. Admission of residents to the Psychotic and Behavior Disorders Units has been directly to the building in which they were to be treated. This change has occurred










because of increased number of admissions to the Sex offender Unit and because of the increased need for more intensive evaluation of incoming residents.

Accordingly, the Intake and Diagnostic Unit has increased the number of evaluation instruments which are administered to sex offenders. All sex offenders are given the MMPI, CACL, Bipolar Psychological Inventory, a short form of the Wechsler Adult Intelligence Scale, the California Psychological Inventory (CPI) , and a complete and extensive social and demographic background information survey. Descriptive statistics for this sample are presented in Table 1.

Thus, most of the residents who have been admitted to NFETC have been tested during the first week of their stay in the institution. Unfortunately, since January of 1978, many residents who have been admitted to the Psychotic and Behavior Disorders Units have not been rated on the CACL. Since the residents were admitted directly into treatment, the staff in the buildings in which they were placed had not been trained in the use of the CACL or other diagnostic instruments.

Accordingly, the sample on which the following study

of the CACL is based includes higher proportions of Mentally Disordered Sex offenders than other treatment categories. Although some test data are available on all residents, with few exceptions, only those who were rated on the CACL during the first two weeks of their stay at NFETC are









58


included in this study. The exceptions to this sampling plan are those 27 individuals who were included in the inter-rater reliability study. Those persons had all been in treatment in the Sex Offender Unit for at least 60 days.



TABLE 1

NUMBER OF RESIDENTS BY UNIT ADMITTED TO NFETC
FROM ITS INCEPTION UNTIL JULY 1, 1978


Psychotic Unit Sex Offender Unit Behavior Disorders
Unit


In treatment Discharged
as of 7/1/78 prior to 7/1/78




Of those residents admitted, intake data on the CACL are available on 140 individuals. Of these, 73 have been treated in the Psychotic Unit, 47 in the Sex Offender Unit, and 20 in the Behavior Disorders Unit.

The number of residents included in each of the studies reported here varies to some degree as a function of the availability of CACL intake data. While the central Intake and Diagnostic unit was using the CACL, each resident was rated independently by three staff members, and an average


90 179

90 45

45 91









59


rating was used to describe the individual. The reliability of the ratings on the 140 individuals on whom such data are available will be computed and compared with that obtained on the twenty-seven residents who were included in the sex offender sample.

After January 1978, the CACL was administered only to those residents who were considered diagnostic problems or whose placement in a particular treatment unit was difficult. All residents were given the MMPI within two weeks of the date they entered NFETC, and often were retested if their responses were considered invalid. If this is the case, the second profile is used for the studies described here.


Instrumentation

The primary instrument of interest in this study is

the Quay Correctional Adjustment Checklist (CACL). This is a 41-item, factor analytically derived behavioral checklist. It was developed between 1964 and 1971 as a classi-fication instrument for incarcerated males. In form, it is neither a true rating scale nor behavioral checklist. Rather, it includes a number of statements which are said to be characteristic of the individual in question.

The CACL is related to the early work of Hewitt and Jenkins (1946) who conducted an analysis of clusters of traits common to juvenile delinquents referred to a child guidance clinic. The resulting groups of traits were used








60


to classify juvenile offenders into three categories: unsocialized-aggressive, socialized delinquent, and overinhibited.

Based on these results, Quay (1964) developed a 36item checklist which was used to quantify the life histories of approximately 100 juvenile offenders. The responses to this checklist were factor analyzed in order to determine whether patterns of developmental events could be used to classify juvenile offenders. The results of this study indicated that the categories developed by Hewitt and Jenkins also appeared in the data obtained by Quay (1964) . The checklist itself was later developed into the Checklist for the Analysis of Life Histories (CALH) , which is often used as a supplement to the CACL.

Subsequently, Quay reported on the development of the CACL and CALH in a 1971 paper. In describing the development of the CACL, Quay related that a pool of behavioral descriptors was assembled from correctional workers and from previous research. Approximately 1,000 inmates from four institutions were rated on the items which were derived from these traits, and the resulting data were analyzed by means of factor analysis in order to estimate the extent of any underlying traits in these results. Four factors emerged, three of which correspond to those found in the CALH.









61


In describing the item selection technique used, Quay

said that analyses were performed on three separate samples, each drawn from a different Federal Correctional Institution. He noted that,

Subsequent to the first analysis, items which did
not meet the frequency criterion (not more than 90% or less than 10% of the subjects were rated as exhibiting the trait) and items which loaded
less than .20 on alny of the factors were dropped, and other items were added for the second analysis. (Quay, 1971, p. 3)

All three analyses produced four principal dimensions. The first, labeled Aggressive-Psychopathic, reflects toughness, defiance, physical arnd verbal aggression, troublemaking, victimizing, and quick temperedness. The second dimension, labeled Immature-Dependent, is composed of such behaviors as inability to follow directions, sluggishness, daydreaming, preoccupation, passivity, moodiness, and dullness. The third factor, given the label NeuroticAnxious, reflects worry, tenseness, help seeking, fear of other inmates, sadness and emotional lability. The fourth dimension, measured by only five items is labeled as Manipulative and involves such characteristics as trying to "con" staff, lack of trust of staff, accusing staff of unfairness, and playing staff against one another.

According to Quay, the factors which emerged in the three samples were congruent with each other to a high degree in two cases (the Psychopathic-Aggressive and Immature-Dependent subscales) , and less so in the cases of









62


the Neurotic-Anxious and Manipulative subscales. The degree of congruence was measured by Tucker's congruency coefficients, but the numerical values of these coefficients were not presented by Quay.

In the final selection of items, two major criteria

were used: first, the item had to have a loading of .40 or higher in one or more of the analyses described previously; and second, the item had to load on the same factor in two of the three samples. After items were selected on these criteria, the results from all three groups were combined and factor scores were computed using unit weights. That is, each item checked as characteristic of the individual earned a value of one toward the score on that factor. Thus, the maximum score on each factor is the number of items contained on that subscale.

When the raw score distributions for each scale were plotted, Quay reported "gross departures from normality" (p. 5), which were evident by visual inspection. The raw scores were subsequently converted to normalized "T" scores.

As an estimate of the internal consistency reliability of the CACL, Quay reported that alpha coefficient was calculated for each of the subscales. For the total sample of 829 (all three groups combined) , the reliability estimates were as follows: .91 for the Psychopathic-Aggressive subscale; .82 for the Immature-Dependent subscale; .77 for the Neurotic-Anxious subscale, and .77 for the Manipulative subscale.









63


Quay also examined the intercorrelations of the four subscales. He noted that: "While the factor analytic procedure results in uncorrelated factors, the actual estimates of scores of individuals on the factors are not necessarily independent" (p. 5). The highest intercorrelation (.81) was found between subscales 1 and 4 (Psychopathic-Aggressive and Manipulative). A moderate correlation (.41) was also found between the ImmatureDependent and Neurotic-Anxious subscales. Quay speculates that this is probably due to rater's tendency to evaluate prisoners as being "totally troublesome" (Quay, 1971, p. 5).

Quay (1971) also reported a validation study in which CACL subscale scores were related to a variety of other variables, primarily demographic in nature. He reported that all of the subscales showed a "modest" relationship to other variables. The Psychopathic-Aggressive subscale correlated negatively with the age of the criminal and positively with the number of prior arrests. The ImmatureDependent subscale tended to relate negatively to I.Q. and years of education. Scores on the Manipulative subscale tended to relate negatively to number of prior arrests, but exact numerical values were not presented.

In general, the CACL has fairly high internal consistency, but unknown inter-ratEr reliability. Although it was designed to provide subscales which are independent of each other, modest subscale correlations are found in most









64


studies. Evidence of construct validity is slight; statistically significant correlations exist between CACL subscales and some other variables, particularly number of prior arrests, age at arrest, and intellectual level.

No specific suggestions for decision rules are

included with the instrument, forcing the user to choose whether to use scores on the CACL in making absolute or comparative decisions. When the instrument was used at NFETC, the highest subscale "T" score determined an individual's CACL classification type. This classification was supplemented by other tests, interviews and so forth.

Other instruments have been used to classify individuals who are incarcerated. Such instruments range from projective tests such as the Rorshach Ink Blots to selfreport inventories such as the 16 Personality Factor Inventory. It is of interest that these instruments were not created for the purpose of classifying criminals, but were developed as diagnostic aids in mental health settings.

The Minnesota Multiphasic Personality Inventory (MMPI) is a 556-item self-report personality inventory. It was developed as an aid to the classification of psychiatric patients, and each of its original nine subscales corresponds to a diagnostic category current at the time of the test's construction. Although these categories were originally presumed to be mutually exclusive, subsequent research has shown this not to be the case (Dahlstom, 19721)







65


Despite the intercorrelation of the subscales as well as the tests' sensitivity to the response set of the testtaker (Messick & Jackson, 1967) , the MMPI has been shown to be useful in assessing a variety of areas of functioning. Recent research has stressed the interpretation of profiles of subscale scores rather than classification into one of several psychiatric diagnostic categories (1eehl, 1955).

In addition to the original nine diagnostic subscales, three "validity" subscales were added to the instrument. These scales are intended to estimate the interpretability of the other subscales, and measure ego strength, naive lying to "fake good" and the frequency of items seldom endorsed by the normative population.

In general, the MMPI has been shown to be more valid for whites then blacks and to discriminate accurately between groups of psychiatric patients and prisoners with accuracy. Local norms are often more useful for behavioral predictions than are national norms (Palmer, 1970) , but both provide predictive ability at a level significantly above chance. The NMI has also been shown to relate to several other measures of criminal behavior, both in and out of incarceration (Panton, 1966).

By estimating the nature and extent of the relationship between the MMPI and the CACL, it is possible to assess the traits common to both and the overlap or redundancy in the instruments. It is also possible to estimate









66


the relationship between the CACL and other variables which are of interest in themselves, as well as for their logical relation to the traits measured by the CACL. Disruptive behaviors in the institution constitute such a criterion variable transition.

For the purpose of this study, disruptive behavior was defined as any act which was contrary to the resident rules of the North Florida Evaluation and Treatment Center and which disturbed the ongoing course of treatment. Such behaviors usually necessitate staff intervention and were limited to threats of aggression, aggressive acts, threats of self-injury, acts of selfinjury, destruction of property, and other unclassified infractions of rules (violation of curfew, refusal to take medication, etc.)


Data Collection

The data used in this study were collected at different times, by different staff members at NFECT, and on different individuals. As part of the normal intake procedure, ratings on the CACL as well as scores on the MMPI were obtained on 140 residents. In addition, CACL ratings were also obtained on a smaller, more homogeneous group of individuals who had been observed for at least eight weeks. Finally, the frequencies of six types of disruptive acts were recorded and included as a behavioral









67


adjustment to confinement. CACL intake data were collected from October 1, 1977, until July 1978, as were MMPI scores on the same individuals. Ratings on the CACL for the smaller group of residents were collected in May 1978.

To obtain the measures of behavioral adjustment, a tally of disruptive behaviors was made from the daily observation notes kept on each resident. These notes were written by the treatment staff in each building at least once every eight-hour shift. Since all significant behaviors, especially infractions of rules, were to be included in observation notes, it seems likely that most disruptive behaviors were so recorded.

For each of the 140 residents on whom CACL and MMPI

data were available, a survey was made of the 180 observation notes written during the first 60 days of his confinement. Those individuals who stayed less than 60 days were not included in this section of the study. Each note was inspected to determine if any disruptive behaviors were recorded. If more than one such behavior was mentioned, each was tallied separately. That is, if a resident threatened a staff member after receiving an infraction for face count, two disruptive behaviors were tallied. Only the specific mention of behaviors observed directly by staff were included. If one resident informed on another, the disruptive behavior was not tallied unless it was directly witnessed by a staff member.









68


For each resident included in this study, a record of the most recent arrest and conviction was made from the FBI "rap sheet." This is a listing of all prior arrests and convictions for the individual in question and is compiled from all arrest records throughout the United States. Arrests and convictions are matched by fingerprints as well as by name, so that crimes committed under an alias are also included. For this study, the following categories were used: murder, armed robbery, assault (including attempted murder) , breaking and entering, forgery, and other nonviolent property crimes, rape or sexual assault, and nonviolent child molestation.

Although the first two analyses included in this

study both assess the inter-rater reliability of the CACL, they differ in several respects. The first is based on the ratings made by three staff members of the intake and diagnostic unit. They were made after a relatively short (seven-day) period, which according to Quay (personal communication, 1978) may not allow for sufficient observation time. The second study is based on ratings made by three staff members in the sex offender treatment program. These ratings were based on the behavior of 27 residents who were being treated in that program, and who had been in treatment for at least eight weeks. The raters had observed the residents for the duration of their stay in treatment, and thus had the opportunity to







69


base their ratings on a larger sample of behavior than that in the first study. In both studies, each rater rated every subject.

The raters for the second study were trained over a seven-day period. Their training included operational definitions of the behaviors assessed by the CACL, as well as comparisons of their ratings on the same residents. That is, the raters filled out a CACL on two residents without discussing the results with each other. These ratings were then compared on an item-by-item basis, with group discussion of any discrepancies. After three such sessions, the raters agreed on 90% of the items on the CACL, and training was discontinued.


Data Analysis

All data were analyzed at the University of Miami

computing facility using a UNIVAC 1100 computer. All analyses requiring a "packaged" computer program used the Statistical Package for the Social Sciences (SPSS) which is available in several versions at the University of Miami. Reliability of the CACL

The inter-rater reliability of the CACL was estimated by the use of the intraclass correlation coefficient, which has been described by Ebel (1951) as well as by Bartko (1966) . Essentially, this method uses the analysis of variance to estimate the proportion of variance in a set









70


of measurements which can be attributed to individuals, raters, and error. In this case, a subject by rater design was used. The resulting mean squares from the analysis of variance were substituted into Ebel's (1951) formula. As expressed by Ebel, that formula is:


M- M
x
v1 M'L + (k-l)M
x


The v is the intraclass correlation coefficient, M S is the mean square for individuals, M is the mean square for error and k is the number of observers or raters. This formula is for estimating the reliability of a single rater, and does not include systematic rater bias in the error term.

When used in making absolute decisions, any systematic bias of the raters needs to be included in the error term of the formula. Thus, the formula is:



v x
2 M K+ (k-l)m +k(m -M)/



In this case, v 2 is the intraclass correlation coefficient, M R is the mean square for subjects, M is the residual mean square, k is the member of raters, MRis the mean square for raters, and N is the number of subjects.

Since the average of three raters scores was used

in placement decisions at NFETC, a third formula was used









71


to provide estimates of the reliability of the average. Including systematic rated bias, that formula is:


ML- M
x
v3 M X+ k(M~ - U



If we exclude systematic rater bias, the formula for the reliability of the average of 3 raters becomes:


M_- M
x
V =
4 M


This formula was used to estimate the reliability of the sum of all four subtests, as well as for each individual subtest. It should be noted that one sample on which these observations were drawn was fairly homogeneous, in that it consisted of only one type of offenders; i.e., Mentally Disordered Sex offenders. This homogeneity may have provided reliability estimates which are somewhat less than maximal. The other inter-rater reliability study was based on the ratings of all types of incoming residents over a nine-month interval. The training of the raters was not under the control of this author, and residents were rated after a relatively short period of observation.


Predictive Validity of the CACL

The ability of the CACL's subscale scores to predict institutional disruptiveness was estimated through a









72


multiple regression procedure. This technique analyzes "the collective and separate contributions of two or more independent variables . . . to the variation of a dependent variable" (Kerlinger & Pedhazur, 1973, p. 3) . That is, it estimates degree of relationship between a set of two or more variables and a single other-variable of interest. It provides approximations of the contributions to the variance of the dependent variable by a group of independent variables. This is accomplished by minimizing the sum of squared deviations between the predicted dependent variable values and those actual values obtained in the experiment. A linear combination is derived for the independent variables which minimizes those errors of prediction.

The model for this 'least squares"' solution can be expressed as:


Y =B 0+ B1X 1+ B2X 2+ +B kX k+ E,


where B 0is a constant value, and B 1B 2are the weights assigned to the independent variables X 1 * * X k.

The weights in a regression equation which are based on the raw scores of the independent variables are known as partial regression coefficients. They are scale dependent, in that they are not directly comparable with each other in absolute magnitude. These weights may be transformed into standard score format so that they are







73


directly comparable in size. In this case, the weights are known as standardized partial regression coefficients, and reflect the unique contribution of each independent variable to the variance explained in the dependent variable.

In this study, the total frequency of disruptive behavior during the first 60 days of incarceration was the dependent variable, and the subtests of the CACL were the independent variables entered in the multiple regression equation. The subscales of the PUMPI were also entered in a separate analysis in order to compare the predictive validity of the two instruments.


Construct Validation of the CACL

The construct validation of the CACL was carried out by means of a canonical correlation analysis. This technique, described by Timm (1975) and others, is an extension of the multiple regression method discussed previously. That is, it provides an estimate of the maximum correlation possible between two linear composites of two sets of variables. This is accomplished by including more than one dependent variable in a linear composite which maximizes the degree of relationship between that group and a linear composite of independent variables.

Two sets of predicted scores are generated by these

linear combinations. If the dependent variables are identified as 9 . 99t1fl=a + a9 . a9
1 Y n 1 a1 2 2 +an n'







74


where ai is the composite value. If the independent variables are labeled as x1+ x n' . then 7= b 1 + b n

where 7 is the composite of those values. Thus, a canonical correlation analysis provides the weights a 1. . an

and b 1 * . * b 2 such that the Pearson Product-Moment correlation between i and Q7 is a maximum.

In this study, canonical correlation analysis was performed to relate the subscale scores on the CACL to subscale scores on the MMPI, with data on both instruments being collected at the time of intake. The results of the canonical correlation analysis were used in a redundancy analysis, which has been described by Stewart and Love (1968). This technique provides a numerical estimate of the redundancy in one set of data, given the other. The redundancy coefficients were obtained by rating the proportion of variance in each set of variables extracted by each canonical variate. These proportions were then multiplied by the corresponding squared canonical correlation and summed across the significant canonical variates for each set separately. The resulting coefficients represent the proportion of variance in a set of variables that may be explained by the second set.


Postdiction of Crime Type

In order to further determine the validity of the CACL, a discriminant function analysis was performed using the CACL subscale scores as independent variables and type of crime as a categorical dependent variable.









75


A discriminant function analysis is an extension of the multiple regression procedure, in that it provides a set of weights for the independent variables which minimize the errors of prediction when the dependent variable is group membership. As in multiple regression, a linear combination of independent variables is formed such that y =a 0+ a y1 +a 2Y2. . . a yn'wherea, . . .a nare

the weights for the independent variables y n

This linear combination provides the best discrimination between the groups by maximizing the among group variance in relation to the within group variance.

In this study, only the most recent conviction was used to determine crime type. Nonviolent crimes were defined primarily as crimes against property (breaking and entering, etc.) , while violent crimes were defined as those which involved physical aggression toward another individual (rape, assault and battery, homicide, etc.). The subtests of the CACL were used as the independent variables in the equation.


Summary

In this chapter, the sample and instruments used in this study are described as well as the procedures for data collection. It is noted that the data collection procedures were not under the direct control of this investigator and thus introduce certain limitations.









76


Also included in this chapter is a description of the various procedures used in the analysis.

Separate analyses were conducted to obtain reliability and validity estimates of the CACL. Two estimates of inter-rater reliabilities were obtained on each subscale of the CACL. The first set of inter-rater reliabilities was computed using the ratings from observers who were not trained on a sample of incoming residents to the institution. Intraclass correlation coefficients were computed from the data. The second set of interrater reliabilities used the same method of computation on ratings by trained observers on a sample of sex offenders.

The relationship between the CACL and the MMPI was assessed using canonical variate analysis. Canonical correlations and redundancy indices were computed. In addition, multiple regression analysis was used in the prediction of institutional disruptiveness from the CACL. Finally, a discriminant function analysis was used to predict type of crime based on the CACL subscale scores.















CHAPTER IV


RES ULTS


The results of the analyses described previously are presented in this chapter. Generally, the results are given without interpretation, since their explanation and synthesis are presented in the following chapter. Descriptive statistics precede each section of this chapter.

In the first section of this chapter, the results of the two inter-rater reliability studies are presented. The first presents the reliability estimates for the ratings done on intake (after 4-7 days of observation) by raters whose training was not controlled by this writer. The second provides a summary of the "controlled" study in which the raters had been trained by this writer and where the subjects had been observed for a minimum of thirty days.

The second section of this chapter contains the results of the canonical correlation analysis between the average ratings on the CACL at intake and the scores on the MMPI administered at the same time. This section is followed by a presentation of the correlations between the canonical variates and the original variables to clarify the content of the canonical variates. The results of the redundancy analysis are also included.


77







78


The third section includes the results of the series

of multiple regression analyses relating scores on the CACL to several types of disruptive behavior. Results are given separately for suicide attempts, assaultive behavior, verbal threats and coercion, as well as for other infractions of program rules. This was done in order to relate CACL subscale scores to specific types of disruptive behavior, in order to assess the relationship between those behaviors and the CACL subtest which would be expected to relate most strongly to them.

The final section of this chapter contains the results of the discriminant function analysis, which relates scores in the CACL subscales to the presence of violence in the crime for which the subject had been most recently arrested and/or convicted. That is, scores on the CACL are weighted so that a linear combination of subtests best predicts group membership, where the criterion for group membership is the presence or absence of violence.


Inter-Rater Reliability of the CACL

As has been explained previously, two studies of the inter-rater reliability of the CACL were performed. These studies used separate samples of subjects and raters, and the reliability estimate from each was obtained through an analysis of variance procedure. The descriptive statistics for the "intake" and "controlled" studies are presented in







79

Tables 1-4 and 15 in Appendix B, respectively; Tables 16 through 23 include the corresponding analysis of variance summary tables for each subtest.

Separate analyses were performed for each of the four subtests of the CACL for the average of three raters, as well as for single raters. These estimates are given for the average rating first, followed by that for single raters. For the "controlled" condition the coefficients are: I-D Subscale, r=.37, .26; P-A Subscale, r=.76, .51; N-A Subscale, r=.73, .46; and Ma Subscale, r=.78, .59. For the "intake" condition the values are: I-D Subscale, r=.70, .42; P-A Subscale, r=.60, .36; N-A Subscale, r=.60, .41; and Ma Subscale, r=.60, .43. Table K in Appendix B presents the values including systematic rater bias in the error.


Construct Validation of the CACL

One construct validity estimate of the CACL was

obtained from intake ratings on the CACL and the results of the MNPI, when both were administered concurrently. The descriptive statistics for the CACL and M.MPI are presented in Table 2, while the intercorrelations are presented in Table 3.


















DESCRIPTIVE


TABLE 2

STATISTICS FOR CONCURRENT VALIDITY STUDY


Variable Mean Standard Deviation Number


CACL PA CACL ID CACL NA CACL Ma MMPI 11 MNP I F MMP I K M14PI H s MMPI D MMPI Hy

MMP I Pd I-MP I M f

MMPI Pa MNPI P t wJ~pi S c Mpi Mla MMPI S i


46.89

49.59 46.46 46.79 5.79 15.35 13.50

17.04 29. 82 23.90 29.77

26.49 16.23 32.17 39.60

24.52 31.16


4.95 6.32

5.34 4.10 4.58 10.76 6 .16

6.85

6. 84

7.09

4.53 4.68 6 .52

8.34 11.82 6.65 10.63


140 140 140 140 140 140 140 140 140 140 140 140 140 140 140 140 140


80






TABLE, 3

TNTF.RCORRFLATION MATRIX FOR CONCURRENT VALIDATION STUDY
(N-104)


Variable CACL Variable- IMPT

Variable PA 11) NA !"1 11 p K 11s D fly Pdi Mf pa Pt Sc fa Si


CACL PA 1.0

CAC[. ID .01 1.0

CACL NA .31 .4R 1.0

CAC1., Ma .68 -.06 .36 1.0

MMPI r, .11 -.04 -.03 .04 1.0

MMPT F .25 .38 .33 .18 -.13 1.0

MMPI K -.03 -.04 -.01 .02 .60 -.42 1.0

MMDI Hs- .09 .36 .36 .13 .10 .40 .07 1.0

MMDI D -.13 .37 .22 -.09 .0n .37 -.13 .59 1.0

MMPI fly .03 .15 .29 .09 .36 .26 .22 .79 .64 1.0

MMPI Pd .16 .09 .24 .10 .16 .31 .07 .47 .48 .60 1.0

MDIT Mf .02 .09 .17 -.05 .16 .37 -.07 .27 .39 .36 .32 1.0

WIPT Pla .10 .25 .28 .14 -.01 .66 -.33 .45 .44 .37 .35 .42 1.0

MMPT Pt .02 .20 .30 .03 -.23 .71 -.27 .51 .61 .41 .39 .39 .52 1.0

MMPT Sc .13 .35 .40 .09 -.11 .86 -.30 .54 .-,4 .43 .42 .50 .73 .82 1.0

MMPt Ma .30 .12 .20 .17 .26 .44 -.08 .13 .01 .14 .20 .21 .45 .21 .43 1.0

MMPI Si -.26 .20 .02 -.30 -.16 .36 -.40 .21 .S4 .17 .27 .39 .36 .45 .48 .05 1.0






82


The results of the canonical correlation analysis

which was performed on these data are presented in Table 4. The two canonical variates which appear in this table are the only two which produced a canonical correlation coefficient which was significant at least at the .05 level. The x2 values which are reported in this table test the significance of the cumulative canonical correlation coefficient as each canonical variate is removed. that is, the first X value tests the significance of all canonical correlation coefficients and the second X value tests the significance of all canonical correlation coefficients, after the first has been removed.

Table 5 includes the canonical weights for the 1MPI and CACL subtest or both canonical variates. Table 6 includes the product-moment correlations between the subtests of those two instruments and each canonical variate.

As was mentioned in the methodology section, a redundancy analysis was performed to provide estimates of the variance shared between the MMPI and CACL. Two redundancy coefficients were calculated: one estimating the redundancy of the MMPI, given the CACL ( RMPI/CACL) , and the other, that of the CACL given the MNPI (R CACL/MMPI) . As Stewart and Love (1968) pointed out, in a case such as this, both of these estimates are necessary since the total variances of the two instruments are not equal. For this study, the RMMPI/CACL was equal to .095, and the R CACL/MMPj was equal to .199.







83


TABLE 4

RESULTS OF CANONICAL CORRELATION ANALYSIS OF

THE CACL AND MMPI


Canonical Canonical Wilkes
Variate Eigenvalue Correlation Lambda x2 D.F.

1 .38 .61 .39 122.78 52*

2 .22 .47 .63 68.96 36*


*p < .05.






84


TABLE 5

CANONICAL WEIGHTS OF MIIPI AND CACL SUBTESTS FOR
CANONICAL VARIATES 1 AND 2 (N=140)


Canonical Canonical
Subtest Variate 1 - Weights Variate 2 - ',,eights

CACL-PA .144 1.028

IDEP .962 - .134

NA -.041 - .349

NA .234 .012


MMPI -L -.131 .433

F .936 .504

K .324 - .529

Hs .651 .091

D .537 - .201

Hy - .534 - .196

Dd -.001 .572

MF -.022 .019

Pa -.086 -.601

Pt -.654 -.074

Sc -.022 .008

Ma .022 .138

Si -. 1J)9 -.733







85


Criterion-Related Validity of the CACL

Several separate analyses were carried out to estimate the predictive validity of the CACL. First, a series of multiple regression analyses was carried out with the subtest scores on the CACL as the independent variable, and each measure of institutional disruptiveness as the dependent variable. In each analysis, the CACL subtest which theoretically should have shown the highest degree of association with the dependent variable was entered first in the regression. Because there was no rationale for the ordering of the remaining subtests, they were entered as a set on the second step.

Table 7 includes the descriptive statistics for this

analysis, and Table 8 gives the intercorrelation matrix for all variables. The results of the multiple regression analysis are presented in Tables 9 through 12, inclusive. These tables include the results for suicide attempts, assaults, threats, and interactions, respectively.

The multiple regression analysis reported in Tables

9 through 12 was performed by entering first the CACL subscale which logically was considered the best predictor of each type of disruptive behavior. The other three subscales were entered as a set on a second step. Accordingly, these tables include the multiple R, RP2 , and F value testing the significance of the R 2 on steps one and two. In addition, the standardized partial regression coefficients










86E


TABLE 6

PRODUCT MOMENT CORRELATIONS BETWEEN SUBTESTS OF THE CACL

AND MMPI AND CANONICAL VARIATES (N=139)



Canonical Variate 1 Canonical Variate 2

CACL MMPI CACL MMPI


CACL CACL CACL CACL




MNMP I MMP I PIMP I

MI MMP I M-71P I MM? I MMl I I MMP I MMP I MM PI MPlm I M? I


PA

IDE? NA

NA


.31

.94 .56 .30


L

F

K

H s

D

Hy Pd Mf

P a P t S c

Ma S i


-.01

.49

-.09

.40 .31 .19

. 14 .13 .32 .27

. 43 .21

.18


.19 .58

.34 . 19




-.03

.81

-.15

.65 . 50 .31 .23

.22 .52

.44 .71

.34 .29


.91

- .27

-.05

.55




.13

.04 .01

-.07

-.19

.04 .08

- .04

-.12

- .11

- .07

.17

-.28


.42

-.12

-.02

.25




.28 .07 .03

-.16

-.42



.17

-. 11

-.26

-.25

-.16

.37

-.62








87


TABLE 7

DESCRIPTIVE STATISTICS FOR PREDICTIVE VALIDITY STUDY

Standard
Mean Deviation N

Suicide Attempts .12 1.14 104
Assaults .60 1.15 104
Threats of Assault 1.08 1.67 104
Infractions of Rules .34 .97 104


CACL PA 47.12 5.18 104
CACL IDEP 50.54 6.33 104
CACL NA 46.77 5.32 104
CACL Ma 47.03 4.38 104
MMPI L 5.76 5.13 104
MMPI F 16.58 10.96 104
M.MPI K 13.22 6.05 104
MNPI Hs 17.54 7.15 104
MNPI D 25.64 6.72 104
MMPI Hy 24.26 7.56 104
M.MPI Pd 29.89 4.62 104
MNPI Mf 26.89 4.67 104
MNPI Pa 16.59 6.54 104
MNMPI Pt 32.53 8.46 104
M.MPI Sc 40.64 12.34 104
M11PI Ma 24.72 7.23 104
MNPI Si 31.99 10.50 104










88


TABLE 8

INTERCORRELATION MATRIX FOR PREDICTIVE VALIDITY STUDY



Suicide Threats of Other
Variable Attempts Assaults Assaults Infractions


CACL CACL CACL CACL MMP I MMP I MMP I M4MP I IUMP I

MMP I MM'P I

MMP I M.MP I MM4P I M-MP I MIAP I MMP I


PA I D NA

M a

L

F

K

H s

D

Hy P d Mt P a P t S c

M a S i


Suicide
Attempts Threats of
Assaults Assaults


.14

-.05

.14

-.08

-.04

.16

-.18

.08 .07 .07 .16

. 12 .23 .10

.18 .06 .07



1.00


.08 .13


.29 .06 .13

.14 .12

.02 .13

-.03

.18 .01 .06 .07

.02 .05 .07

.24

-.16


.16

-.04

.03 .03 .11

.05 .07

-.08

-.09

-.06

.05 .01 .01 .01 .05 . 11

-.07


.13

-.06

. 11

.14

- .09

.13 .06 . 11

.14

-.09

.01 .03

-.14

-.14

-.08

-.09

.07


1.00


.40


1.00


Other.2
Infractions.2


.25 .27 10


1.00









89


TAB LE 9

RESULTS OF MULTIPLE REGRESSION OF FREQUENCY
OF SUICIDE ATTEMPTS ON CACL SUBSCALES


CACL 2F
Subscale Step R _ F Beta -(unique)NA 1 .14 .02 2.09 .28 5.13*

1A2 -.41 9.60

PA 2 .32 5.83*

IDEP 2 .35 .13 3.63 -.21 3.55



*p2<.05, 1 and 99 df





TABLE 10

RESULTS OF MULTIPLE REGRESSION OF FREQUENCY
OF ASSAULTS ON CACL SUBSCALES


CACL2
Subscale Step R R2 Beta -(unique)

PA 1 .16 .03 2.62 .25 3.40

MA 2 -.15 1.18

IDEP 2 -.07 .30

NA 2 .20 .04 .99 .02 .40











90


TABLE 11

RESULTS OF MULTIPLE REGRESSION OF FREQUENCY
OF THREATS OF ASSAULT ON CACL SUBSCALES


CACL 2F
Subscale Step R R F Beta -(unique)

FA 1 .29 .08 9.25** .34 6;53*

MA 2 .92

IDEP 2 .96

NA 2 .31 .10 2.69 .10 .68


*P






TABLE 12

RESULTS OF MULTIPLE REGRESSION OF FREQUENCY
OF INFRACTIONS ON CACL SUBSCALES


CACL 2F
Subscale Step R R F aeta -(unique)PA 1 .1-3 .02 1.85 .05 .13

MIA 2 .05 .14

NA 2 .13 1.01

IDEP 2 .19 .04 .90 -.12 1.09




Full Text

PAGE 1

ANTICIPATING FUTURES: AN OPERATIONS RESEARCH MODEL FOR STATEWIDE POSTSECONDARY EDUCATION PLANNING BY THOMAS A. GAYLORD A DISSERTATION PRESENTED TO THE GRADUATE COUNCIL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY UNIVERSITY OF FLORIDA 1980

PAGE 2

Copyright 1980 by Thomas A. Gaylo

PAGE 3

ACKNOWLEDGEMENTS I wish to thank the members of my doctoral committee, Dr. James L. Wattenbarger , Dr. Ralph B. Kimbrough, and Dr. James F. Burns, for their constant guidance and support. The committee's efforts improved considerably the theoretical base, the model developed from it, and the manner in which it was presented. Thanks are extended to the staffs at the Florida Department of Education, the University of Florida College of Nursing, and the Bureau of Economic and Business Research for their assistance. In particular, I thank Dr. Wallace E. Bell of the Division of Community Colleges and Dr. James K. Umholtz of the Office of Educational Facilities Construction. I am appreciative of the financial support I received from the Florida Community Junior College Inter-institutional Research Council, the Department of Educational Administration, and the Florida Institute for Educational Linkage Development during the period of graduate work. Special thanks are extended to Deborah Hill for her excellent job of typing and proofing the dissertation. Lastly, appreciation of the unique role my family has played throughout my graduate studies is acknowledged. To ray wife, Lorrinda Gail, I bestow my greatest thanks of all. iii

PAGE 4

TABLE OF CONTENTS PAGE ACKNOWLEDGEMENTS iii LIST OF TABLES vi LIST OF FIGURES viii ABSTRACT ix CHAPTER I INTRODUCTION 1 The Problem 9 Assumptions 13 Definition of Terms 14 Procedures 17 Organization of the Research Report 20 II A SELECTIVE REVIEW OF RELATED LITERATURE .... 22 Introduction 22 Operations Research 23 Macro Educational Planning Models 29 Linear Programming and the MPSX Computer Code 45 Summary 60 III MODEL DEVELOPMENT 63 Introduction 63 Forecasting the Supply of Graduates Submodel 64 Forecasting Occupation Demand Submodel .... 81 The Student Flow Submodel 84 The Objective Function 88 Summary 89 iv

PAGE 5

TABLE OF CONTENTS (Continued) CHAPTER PAGE IV A LIMITED APPLICATION OF THE MODEL 91 Introduction Parameter Determination 9'* Results Strategy Development 128 Summary 134 V SUMMARY 135 Summary and Discussion 135 Recommendations 142 APPENDICES A SUMMARY OF MODEL VARIABLES, DEFINING EQUATIONS, CONSTRAINT EQUATIONS, AND OBJECTIVE FUNCTION 145 B FLORIDA COMMUNITY COLLEGE AND COUNTY KEY 149 C PROGRAM AND COMPUTER OUTPUT FOR SERIES II BASELINE SCENARIO 150 BIBLIOGRAPHY 163 BIOGRAPHICAL SKETCH 176 V

PAGE 6

LIST OF TABLES TABLE PAGE 1 Characteristics of Long-, Mediiim-, and Short-Range Planning 4 2 Number of Scenarios Generated 56 3 Calculation of the Program Headcount Enrollment Forecast for One Institution By the Ratio Method 69 4 Calculation of the Program Headcount Enrollment Forecast for One Institution By the Double Exponential Smoothing Technique. 73 5 The Student Flow Matrix — Participation Rates a.,(i,£) 86 6 RN Program Total Instructional Cost/FTE For Polk Community College 96 7 Total Cost/FTE for the RN Program at Polk Community College c^^ 98 8 Calculation of the Total Assignable Square Feet of Space for the RN Program at Polk Community College 100 9 Ratio of RN Graduates to RN FTE for Polk Community College 101 10 Forecasting the Supply of Graduates Submodel — Parameter Summarization 102 11 Projected RN Requirements for 1980 and 1985 in Florida 105 12 1985 State RN Requirements 107 13 1985 Requirements for Community College RN Program Graduates 110 14 Student Flow Matrix for RN Program Graduates . . Ill vi

PAGE 7

LIST OF TABLES (Continued) TABLE PAGE 15 Series I and II 1985 RN Enrollment Analysis . . . 115 16 Series I and II 1985 RN Program Space Analysis 117 17 Series I and II 1985 RN Program Cost Analysis 120 18 Series I and II 1985 Baseline Scenario RN Requirements Analysis 124 19 Series I and II 1985 NHI Scenario RN Requirements Analysis 126 20 Series I and II 1985 COMB Scenario RN Requirements Analysis 127 21 Series II Primary Limiting Constraints 130 22 Series II Secondary Limiting Constraints .... 133 vii

PAGE 8

1 LIST OF FIGURES FIGURE PAGE 1 The OR Process 26 2 A Mathematical Program 28 3 Use of OR Models in Central Planning 41 4 A Linear Program 47 5 MPSX Card Deck Layout 61 6 Model Constraint Factors 136 7 Three Tiers of the Limited State Resources Constraints 138 8 Developing Strategies for Model Results 141 viii

PAGE 9

Abstract of Dissertation Presented to the Graduate Council of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy ANTICIPATING FUTURES: AN OPERATIONS RESEARCH MODEL FOR STATEWIDE POSTSECONDARY EDUCATION PLANNING By Thomas A. Gaylord August, 1980 Chairman: Dr. James L. Wattenbarger Major Department: Educational Administration and Supervision This study presents the results of an initial research effort in the development of a statewide planning model for public postsecondary education. A prototype linear programming model was developed as a tool to assist state postsecondary education planners maximize selected system-wide policy objectives. The model could be used to generate alternative futures which could aid state-level planners in formulating strategies to adapt to, prevent, or achieve a possible system future. The focus of the model was not to predict the future, but rather to facilitate anticipation of many futures and preparation of alternative strategies as system conditions such as enrollment, social demands for educational opportunities, resource allocation patterns, and manpower requirements change .

PAGE 10

Limited application of the model was accomplished using the registered nurse programs offered in Florida's community colleges. A set of three manpower requirement scenarios for community college registered nurse program graduates by geographic area was devised for 1985. Each of the three manpower scenarios was applied to two different community college 1985 scenarios, one which allowed the community colleges complete freedom to respond to the manpower scenarios ; the other was constrained by trend analysis forecasts of student demand, available facilities, and appropriations. From the analysis of the resulting six futures, strategies were suggested for the community college system to prevent an anticipated 1985 shortage of registered nurses in Florida. X

PAGE 11

CHAPTER I INTRODUCTION With the evolution of a technologically complex society, traditional methods of planning, decision-making, and problem solving have become inadequate tools for the education administrator. It has become evident that exclusively verbal descriptions of complex systems and their interactions have resulted in generalizations which have presented problems for comparison, analysis, and evaluation. Numerous studies have documented the increasing importance of quantitative methods in several areas of administrative endeavor (Banghart, 1969; Bertalanffy, 1956; Churchman, Ackoff, & Arnoff, 1957; Goodlad, 0 'Toole, & Tyler, 1966; Hall, 1962; Optner, 1965; Van Dusseldorp, Richardson, & Foley, 1971; Wiener, 1950; 1961). Three principle areas in which quantitative techniques have been applied include: (1) planning, (2) organizational control and the decision-making process, and (3) the analysis of how organizational systems work and how their different parts or subsystems are interrelated (Bell, 1977; Martino, 1972; Optner, 1975). Though these three categories are hardly mutually exclusive, the fundamental utility of the study relates to the planning area. Planning has been characterized as being a continuous activity. Churchman (1968) expanded on this not ion and stated 1

PAGE 12

2 that planning was "a series of approximations in which each approximation in principle is better than its predecessor" (p. 146). Ackoff (1970) viewed planning as anticipatory decisionmaking, made necessary because of the interdependence of decisions, and future oriented by nature. In Ackoff 's words, planning is a process that involves making and evaluating each of a set of interrelated decisions before action is required, in a situation which it is believed that unless action is taken a desired future state is not likely to occur, and that, if appropriate action is taken, the likelihood of a favorable outcome can be increased. (p. 4) The determination of the possible "future states" of a system is accomplished through the development of a methodology and the resultant system model. Optner defined methodology as a "logically and procedurally organized arrangement of steps" (1965, p. 19). The methodology used for planning in futures research is called simply the forecast. Amara and Salancik defined forecast as a "probabilistic, reasonably definite statement about the future which is based upon an evaluation of alternatives" (1972, p. 112). Joseph (1974) offered a further elaboration and characterized a forecast as a formalized and systemic methodology for determining future possibilities that allows us to move beyond pure speculative conjecture about the future. It is a system of quantified estimates of changes and alternatives. The key is that a forecast uses a system of logic and yields reproducible and relatively consistent results independent of the forecaster. A forecast therefore rests upon an explicitly stated set of logical assumptions,

PAGE 13

3 data and relationships, and therefore differs from opinion and prophesy. (p. 4) Different planning orientations or planning approaches have been devised that incorporate different forecasting techniques. The three major approaches toward planning for the future in education have been identified by Ziegler (1972) as preventive, adaptive, and inventive. The preventive approach has been the most widely used approach. The approach is characterized by a lack of effort to examine the unanticipated consequences of intervention and to relate forecasts to a comprehensive analysis of other interacting variables. The future is essentially considered an extrapolation of the past. When utilizing the adaptive approach, immediate actions are taken to adapt to the forecasts made. The third planning approach is the inventive approach. One principal characteristic of the inventive approach is the recognition that the future is not a predetermined extrapolation of the past, but rather consists of an array of alternative futures which can be explained in some detail. Another characteristic is the recognition that it is both possible and desirable to intervene creatively in the present to bring about a more desirable future (p. 8). The planning approach, whether it be preventive, adaptive, or inventive has been one basis used to differentiate short-, medium-, and long-range planning. Table 1, adapted from Jarupanich (1978, p. 28), presents a more complete breakdown of planning type by planning characteristic.

PAGE 14

4 Just as there have been different planning types and different general planning approaches devised, different planning methodologies or forecasts have been developed also. Jantsch (1967, p. 18) has grouped the many available futures research techniques into four broad categories: intuitive, exploratory, normative, and feedback. Though other classification schemes have been devised (Joseph, 1974; Martino, 1972), the groupings offered by Jantsch facilitated the discussion of the procedures used throughout the study. Table 1 Characteristics of Long-, Medium-, and Short-Range Planning Planning Type Planning Characteristics Planning Objectives Planning Outccxnes Planning i^proach Planning Purpose Planning '"Criteria Long-Range Ideals Policies Inventive Establish Direction Normative or Policy Planning Medium-Range Objectives Programs Adaptive Establish Strategies Strategic Planning Short-Range (joals Methods and Procedures Preventive Allocate Resources Tactical Planning In Jantsch 's scheme, intuitive thinking has been used as one forecasting method. Jantsch described these intuitive techniques as the "gracious gift of the subconscious mind in return for the previous labors of the conscious

PAGE 15

5 mind" (1967, p. 133). Exploratory methods extend the past or current patterns into the future and attempt to create an awareness of the demands which these patterns are "hypothesized to create" (Heinmiller, 1977, p. 14). Normative techniques begin by determining future needs or goals and then, by working backward to the present, a path for the attainment of the stated goals is formulated (Sage & Chobot , 1974, p. 162). Feedback or cybernetic techniques integrate the normative and exploratory methods to create a future state; modifications are incorporated into the scheme as additional information is obtained (Jantsch, 1967, p. 113). In this study the mediumand long-range planning objectives of an educational system were incorporated into a system model that facilitated investigating a series of forecasts, any one of which could represent the eventual future. Emphasis was on taking the inventive approach toward the future. The system model combined both exploratory and normative methods to create a future state and was rooted in general systems theory. In brief, general systems theory was described by Miller (1965) as "a set of related definitions, assumptions, and propositions which deal with reality as an integrated hierarchy of organizations of matter and energy" (p. 193). All systems display certain properties, and although there have come to exist rather elaborate listings of these properties (see Miller, 1978), a few of the more basic Iramegart (1969) summarized as follows:

PAGE 16

6 1. All systems exist in time-space. 2. All systems tend toward a state of randomness and disorder, the ultimate of which is entropy, or inertia. 3. All systems have boundaries, which are more or less arbitrary demarcations of that included within and that excluded from the system. 4. All systems have environment, which is everything external to (without the boundary of) the system. 5. All systems have factors that affect the structure and function of the system. Factors within the system are variables; factors in the system's environment are parameters. 6. All but the largest systems have suprasystems . 7. All but the smallest systems have subsystems, (p. 167) A number of approaches have evolved that have applied concepts of general systems theory. The major approaches stated by Immegart and Pilecki (1973) have included: (1) cybernetics, (2) holism, (3) output analysis, (4) systems design, (5) information theory, (6) systems analysis, (7) systems engineering, (8) mathematical programming, (9) computer science, and (10) operations research (p. 9). Utilization of the operations research approach was made in the study . Operations research (OR) has somewhat different objectives than does general systems theory. Modar and Elmaghraby (1978) stated the different aims of each in the following manner : OR tends to be concerned with problems which can be represented by mathematical models, which, in turn, can be analytically studied and optimized. Systems theory, on the other hand, although formed

PAGE 17

in nature, is concerned with problems of greater complexity, and is more global and abstract in its approach. Its components may consist of mathematical models, but may also incorporate social and biological factors which have not been successfully quantified. OR tends to be mostly concerned with smaller scale problems of existing systems, as contrasted with systems theory which usually connotes a larger and more encompassing perspective of new systems, (pp. xiii-xiv) In OR, the terms model and model building have been defined as the structure and the act of formulating the structure of some specific concrete system. Dantzig (1963) stated that "model building refers to the process of putting together of symbols representing objects according to certain rules, to form a structure, the model, which corresponds to a system under study in the real world" (p. 6). Any OR model can be interpreted as being a forecast of many possible futures. Brown (1978) stated that OR is concerned with models that demonstrate the consequences of . . . alternatives . . . based on current information about the significant factors involved. While some of the pertinent information is observable and measurable, there is usually a time lag between the time when a decision is taken and the time when its consequences become apparent. During that time lag some of the information may change. Thus, the decision is based, not on observable information, but on a forecast. (p. 3) By varying the parameters of an OR model an array of possible futures can be generated. Strategies can then be developed to deal with each future that has been generated. Strategies are dependent upon the planning approach that has been taken, either preventive, adaptive, or inventive. Operations research is characterized by the application of the scientific approach, scientific tools, and scientific

PAGE 18

8 techniques to decision-making problems that involve the operation of an organizational system. Obtaining optimum solutions to the problems is the primary goal (Daellenbach & Bell, 1970, p. 1). Operations research techniques that have been used extensively include: (1) game theory, (2) queueing theory, (3) network analysis, (4) simulation, (5) dynamic programming, (6) inventory theory, (7) decision analysis, (8) integer programming, (9) nonlinear programming, and (10) linear programming (see Hillier & Lieberraan, 1974). Of the many techniques of OR, linear programming has been the most widely used. Simply put, linear programming "is concerned with describing the interrelations of the components of a system" (Dantzig, 1963, p. 6). The impact of linear programming has been succinctly stated by Hillier and Lieberman (1974): Many people rank the development of linear programming among the most important scientific advances of the mid-twentieth century, and we must agree with this assessment. Its impact since just 1950 has been extraordinary. Today it is a standard tool that has saved many thousands or millions of dollars for companies or businesses of even moderate size in the industrialized countries of the world, and its use in other sectors of society is rapidly spreading. (p. 15) In the study, the linear programming OR technique was used to incorporate possible future statewide needs and state-level policies into a prototype model of a statewide public postsecondary education system. Different scenarios were generated by changing policies and other model

PAGE 19

9 parameters thereby facilitating the development of alternative strategies to adapt to, prevent, or achieve each possible system future. The Problem Statement of the Problem The problem of the study was to develop a prototype linear programming model that will assist state public postsecondary education planners in maximizing selected or multiple system-wide policy objectives. The model was used to generate alternative futures which could aid state-level planners in formulating strategies to adapt to, prevent, or achieve a possible system future. The focus of the model was not to predict the future, but rather to facilitate anticipation of many futures and the preparation of alternative strategies as system conditions such as enrollment, manpower requirements, resource allocation patterns, and social demands for educational services change. Ultimate objectives of research such as this would include the following: 1. Establishment of long-range plans and priorities. 2. Coordination of mediumand long-range planning. 3. Program evaluation as a function of objectives, costs, and benefits. 4. Resource allocation as a function of systemwide needs and resources. 5. Identification and analysis of alternative methods of achieving system goals and anticipating different system futures.

PAGE 20

10 Delimitations The following were observed in the conduct of the study . 1. The development of the prototype model was confined to the analysis of a single occupational program. 2. A second delimitation is that the model application was restricted to the state of Florida. 3. The planning horizon was limited to five years. Various scenarios were created for the year 1985 and the model was used to assist the formulation of system strategies in response to the several scenarios. 4. If models are to simulate the entire public postsecondary education system of a state within various time, cost, and utility constraints, a system-wide management information system (MIS) becomes a necessity. In other words, standardized definitions of all data elements along the lines of the data element dictionary developed by the National Center for Higher Education Management Systems (Goddard, Martin, & Romney, 1973) must be formulated and each postsecondary institution in the state must submit requested MIS data in terms of the standardized definitions. In Florida, no such MIS has been developed. The objective of the study, however, was not the development of a standardized MIS. An MIS has been developed and used since 1969 to collect Florida community college system data. In view of this, a fourth delimination was that only Florida community college data were used for the public postsecondary education dimension of the prototype model. Limitations Although the prototype model developed in the study was restricted to a single occupational program and to the public community college system of the state of Florida, the model may be generalized to other programs, entire post secondary education systems, and to other states provided

PAGE 21

11 certain criteria are met. First, postsecondary education commissions must have the authority and responsibility to collect all relevant data from all postsecondary education institutions in the state on a regular basis. A comprehensive postsecondary education system MIS simply must exist. It should require little change or maintenance from year to year and include the maximiim amount of pertinent information available. Secondly, the model can be generalized to include other programs existing in postsecondary education institutions to the degree that these programs can be related to particular occupations. A study by Folger and Nam (1964) demonstrated the difficulty of relating education preparation to occupation and that in the United States the trend was toward increasing uncertainty concerning this relationship. Thirdly, the model is dependent on general manpower and economic variables and the breakdown of these variables by geographic areas of a state. Fourthly, generalizability will depend on the extent that the problem studied satisfies the theoretical assumptions of linear programming and modeling. Justification One major factor impelling the study centers on the investigation of the following proposition: Since OR has proven itself in industry, similar applications to education can result in more effective and efficient strategies being devised to attain educational goals and meet society's needs given a set of constraints. Coombs and Hallak (1972) offered

PAGE 22

12 the following comment on the need to apply a spectrum of different planning tools and include more factors to improve system efficiency and effectiveness: Planners and administrators will in the future have to take into greater account the economic aspects involved in their plans and explore every means of improving the efficiency of their educational system so as to get the best value from existing resources. (p. 5) The application of operations research (OR) models in educational planning has increased since 1970. McNamara (1973) cited several reasons for the expanded interest: During the past few years there has been a significant increasing interest placed on an examination of the extent to which models developed in operations research, management science and econometrics might be used to improve the methodology and procedures currently employed in various areas of public sector planning and policy analysis. This type of inquiry has led some social scientists and public administrators to conclude that the problem solving approach . . . can be used collectively to develop better measures for socioeconomic planning and resource allocation. (p. 19) Yet, OR models have had difficulty providing usable results. According to Johnstone (1974), models of many types have been developed over recent years for use in educational planning . . . very few . . . have given results which could be helpful to an educational planner. (p. 195) One reason has been the limited scope of past models. Relevant properties of the system under study failed to be represented in the model because of the difficulty of quantifying data. Areas which lacked any clear objective function were avoided (Page, Jarjoura, & Konopka, 1976). OR models in education, however, have been particularly useful when the plannin

PAGE 23

13 objective was optimum resource allocation, meeting manpower requirements, satisfying social demands for education, or minimizing the social cost-benefit ratio. Panitchpakdi (1977, p. 339) stated that the typical model has educational targets determined by only one of the above possibilities. The model developed in the study integrated aspects of each in order to devise a more realistic representation of an educational system. Few operations research models have utilized the inventive futures approach for mediumor long-range planning (Hufner, 1968; Johnstone, 1974; McNamara, 1973). The study was therefore heuristic. Extending OR models designed for a subsystem of a state post secondary education system to the entire system is a function of information availability and standardization (Harcleroad, 1971, pp. 34-38; Krauss,1970, pp. 86-91). The study, therefore, demonstrated the need that a statewide postsecondary education MIS exist. Assumptions For the purposes of this study a number of assumptions were made. The first set of assumptions relates to the formulation of the linear programming model. In order for the system to have been represented by a linear programming model, assumptions of proportionality, additivity, divisibility, nonnegativity , and deterministic problem nature had to be made. Chapter II includes a brief explanation of these basic linear programming assumptions.

PAGE 24

14 The second set of assumptions, adapted from Sachs (1977, pp. 165-166), relates to the general theory of modeling. The assumptions follow. 1. The model includes all relevant properties of the reality. 2. Model properties correspond to real world properties. 3. All relevant properties of the real world can be expressed mathematically. In light of the model assumptions, the single most pervasive assumption of the study was to "neglect the negligible." Dantzig (1963) confronted the issue as follows: It is important to realize in trying to construct models of real-life situations, that life seldom, if ever, presents a clearly defined linear programming problem, and the simplification and neglect of certain characteristics of reality are as necessary in the application of linear programming as they are in the use of any scientific tool in problem solving. (p. 7) Definition of Terms Community college . An educational institution supported by public tax funds, governed by a politically appointed or elected board, and which offers courses and/or programs confined to the first two years of post high school education. FTE student . Full-time equivalent student. Annualized for Florida community colleges, 30 semester hours equal one FTE student. Futures . Any of a number of probabilistic, reasonably definite statements about the future state of a system given a set of initial conditions specified at some earlier point in time.

PAGE 25

15 Futures research . Investigations which enhance the understanding of future consequences of current developments. Linear programming . A "mathematical tool for obtain ing optimum solutions that do not violate . . . constraints that cannot have negative activities, that require linearly proportional relationships, and that account for all inputs and outputs within the system" (Greenberg, 1978, p. 5). Manpower . The number of persons employed in a given occupation as function of geographical area. Model . A mathematical representation of a system to be utilized for the purposes of creating alternative future system states and facilitating the formulation of strategies to prevent, achieve, or adapt to each. MPSX . An abbreviated form of Mathematical Programming System Extended, an IBM computer code used for linear programming and closely related techniques (IBM, 1972). OR . An abbreviation for operations research, it is an approach used to assist in solving management problems through interpretation and use of available data (Schroeder k Adams, 1976 , p. 118) . Public postsecondary education . Education offered by public four-year and graduate institutions, community colleges, technical institutes, junior colleges, area vocational schools, and postsecondary vocational schools. Referred to, hereafter, as simply postsecondary education (U.S. Senate, 1972, p. 89).

PAGE 26

16 Registered nurses . This "group includes occupations concerned with administering nursing care to the ill or injured. Includes nursing administration and instruction, and public health, industrial, private duty, and surgical nursing. Licensing or registration is required" (U. S. Department of Labor, 1977, p. 56). Abbreviated as RN. Scenario . A combination of a set of model parameters or forecasts that affect the future state of a system. Sensitivity analysis . The analysis of the effect on the optimal solution of a linear programming model as model parameters are changed. State post secondary education commission . A state agency legally charged with supervising, evaluating, organizing, regulating, or compiling the policies for a state postsecondary education system for the purposes of planning and coordination, referred to as the "1202 commission." Statewide planning . The "identification of key problems, the accumulation of accurate data about those problems, the analysis of their interrelationships, the examination of alternatives which might emerge out of present conditions, the evaluation of the probable consequences of introducing new variables, the choice of the most desirable basic goals, a plan for implementing these goals, and a constant feedback system for periodically reevaluating both the goals selected and the appropriate means used to attain them" (Lyddy, 1975, p. 26). System . A set of interrelated activities, units, or variables (Miller, 1978, p. 16).

PAGE 27

17 Procedures The general procedures outlined below and detailed in subsequent sections were followed in the study: (1) review of related literature, (2) model development, and (3) development of alternative future scenarios and possible system strategies in response to scenario conditions. Phase I — Literature Review Literature reviewed related to statewide postsecondary education planning and research, operations research as a planning tool, linear programming, and linear programming computer codes. Sources included books, monographs, and articles identified through the following indices: Current Index of Journals in Education , Dissertation Abstracts International , and the Education Index . Articles taken from journals outside education came primarily from the following: Long Range Planning , Management Science , Operations Research , and Socio-Economic Planning Science . Review of the rise of operations research (OR) as a planning tool included a survey of selected mathematical programming techniques and probabilistic models that have been developed. Previous applications of OR techniques in postsecondary education at the state and national levels were discussed. Description of the general characteristics of linear programming included a discussion on sensitivity analysis. How linear programming was used to create alternative futures was also presented.

PAGE 28

18 The final section of the literature review included a general survey of linear programming computer codes. IBM's MPSX code was also introduced. Phase II — Model Development The development of the model was accomplished through the application of a series of steps recommended by Banghart (1969, p. 297): 1. Outline the problem. 2. Determine the constraints. 3. Put the problem in mathematical form. 4. Satisfy the linear programming assumptions. 5. Formulate the objective function. 6. Determine the optimal solution. 7. Perform a sensitivity analysis. Step 1. The outline of the problem has been presented in Chapter I. Rationale for selecting the registered nurse program in Florida's community college system as part of the general problem delimitation was determined and justified through the literature review and the analysis of Florida community college MIS. Step 2 and Step 3. Primary constraints and functional constraint equations were determined by Florida community college system policies and the data available through the Division of Community Colleges. Data for manpower and economic constraints were extracted mainly from U. S. government publications and the Florida Statistical Abstract. Whenever

PAGE 29

19 possible, existing variable values projected to the year 1985 were used. In the event no such information was available for a variable, the simple trend analysis or the double exponential smoothing technique was used to estimate the variable value for 1985. Step 4. It was assumed that the model and model variables satisfied the assumptions of linear programming. Step 5. From the wide range of objective functions that could have been formulated, the objective function selected was to maximize system-wide enrollment. Step 6 and Step 7. MPSX was the linear programming computer code used to determine the existence of an optimal solution. A sensitivity analysis was also accomplished using various 1985 RN manpower requirements scenarios as described in Phase III. Phase III — Alternative Future Scenarios and System Strategies Limited application of the model was accomplished using the registered nurse, RN, programs offered in Florida' community college system. A set of three future scenarios for 1985 based on trends identifiable in 1980 was devised to determine the 1985 manpower requirements for community college RN program graduates by state geographic area. Each of the three manpower scenarios was applied to two different community college 1985 scenarios. The first community college RN program scenario. Series I, was a normative scenario which described 1985

PAGE 30

20 community college RN programs as having the freedom to satisfy the manpower requirements optimally. By the application of three RN manpower scenarios, parameter ranges were determined for the resulting ideal community college RN program scenario futures. The second community college RN program scenario, Series II, was an exploratory scenario which described the 1985 community college RN program condition based on 1980 trends. Again, parameter ranges for the resulting straightline projection scenario futures were determined by the application of the three RN manpower scenarios. In sum, six different 1985 futures were investigated. The object of the analysis was to compare Series I and Series II futures to determine the primary and secondary constraints in the Series II futures that prevented the community college RN programs in 1985 from attaining the ideal Series I futures. Strategies were then suggested from the results. Organization of the Research Report The research report contained the following chapters: Chapter I: An introduction to the study, including a statement of the problem with delimitations, limitations, and justification for the study, comprise this chapter. Assiimptions are also stated, terms defined, and procedures described. Chapter II: A review of the related literature is made in Chapter II. The literature on statewide postsecondary

PAGE 31

21 education planning and research and general operations research techniques is reviewed. A general survey of linear programming is presented emphasizing the simplex method of solving linear programs and sensitivity analysis. IBM's MPSX computer code is also introduced. Chapter III: The model development is presented in this chapter. Chapter IV: A linear programming prototype model is developed dealing with possible state-level policies and needs as they relate to one occupational program offered in Florida's community college system. Data for the model are presented and analyzed. A sensitivity analysis of the model is also included in the chapter. Scenarios depicting conditions affecting Florida's community college system in 1985 are presented. Possible system strategies for each future described by the scenarios are suggested. Chapter V: In the final chapter, a summary, conclusions, and implications of the study are presented. In addition, directions for future research are suggested.

PAGE 32

CHAPTER II A SELECTIVE REVIEW OF RELATED LITERATURE Introduction In the preceding chapter a general survey of reasons that necessitated the use of increasingly qualitative methods for long-range educational planning was presented. By utilizing the techniques of operations research (OR), it was contended that a number of reasonable postsecondary education futures could be generated and then used to aid the development of planning strategies that either adapt to, prevent, or achieve each possible system future. Reviewed in this chapter is material more directly related to the precise topic explored by the study, the development of a linear programming model delimited to a single occupational program offered in Florida community colleges that can be used to measure system responses to various future conditions. Accordingly, the review of literature was divided into three major segments: Operations Research, Macro Educational Planning Models, and Linear Programming and the MPSX Code. First the development of operations research is presented along with the general characteristics of mathematical programming, the category of OR techniques into which linear programming falls. Next, background information on 22

PAGE 33

23 the rise of OR techniques in system-wide educational planning is given. Several macro models are reviewed followed by a critique of the use of models in educational planning. Finally the characteristics, assumptions, and utility of linear programming models are briefly reviewed and the MPSX computer code used in the study is explained. Operations Research Development As an organized activity in the United States, operations research (OR) began with the establishment of the Operations Research Society of America in 1952. The origin of OR, however, has generally been attributed to teams of military scientists working in Great Britain during World War II on strategic and tactical military problems (Modar k Elmaghraby, 1978, p . ix ) . Prior to 1962, information found in the literature on OR is scarce. Indeed, between 1957 and 1962, less than a half dozen books on OR were published in the United States. Books on OR since 1962 numbered in the hundreds and the application of OR techniques has expanded from just the military area to include the social sciences, government, business, economics, the natural sciences, engineering, and mathematics (Modar & Elmaghraby, 1978, pp. ix-xi ) . The rise of OR during this period has been attributed to two primary factors. The first was the development of

PAGE 34

24 the theoretical foundation of several OR techniques. Hillier and Lieberman (1974) stated that "many of the standard tools of operations research, e.g., linear programming, dynamic programming, queueing theory, and inventory theory, were relatively well developed before the end of the 1950's" (p. 3). The second factor involved the invention of the electronic computer and the rapid advances in computer technology that followed. Indeed, the "development of the electronic digital computers, with their ability to perform arithmetic calculations thousands or even millions of times faster than a human being can, was a tremendous boon to operations research" (Hillier k Lieberman, 1974, p. 3). The development of OR techniques has paralleled the overall development of OR as a process. In 1956, Ackoff stated at that time each operations researcher's "version of the operations research method (if recorded) would differ in some respects" (1956, p. 265). This led Ackoff to suggest the following six phases for the OR process: 1. Formulating the problem. 2. Constructing a mathematical model to represent the system under study. 3. Deriving a solution from the model. This involves finding the values of the variables that maximize the system's effectiveness. 4. Testing the model and the solution derived from it. This involves evaluating the variables, checking the model's predictions against reality, and comparing actual and forecasted results. 5. Establishing control over the solution. This involves developing tools for determining when

PAGE 35

25 significant changes occur in the variables and functions on which the solution depends, and determining how to modify the solution in light of such changes. 6. Putting the solution to work. Implementation. (1956, pp. 265-266) Schematically, these phases are shown in Figure 1. The development of OR both from a historical perspective and from the process perspective has been presented. The mathematical programming approach of OR and constructs pertinent to this study are presented next. Mathematical Programming Mathematical programming is a deterministic OR approach concerned with deriving optimal solutions to system problems subject to some set of constraints. As a branch of applied mathematics, mathematical programming is rooted in optimization theory. McMillan (1975) stated that the term optimum, in the mathematical sense, seems to have first been used by Leibniz in 1710. With the development of the calculus, "Newton and Leibniz made possible the proof that an optimiim solution exists, for a class of problems, and provided the means for finding it" (p. ix). Optimization theory grew to where any number of initial conditions or constraints could be incorporated into a problem while the problem remained theoretically solvable. As a tool for decision-making, mathematical programming began with Dantzig's development of the simplex method of solving linear programming problems in 1947 (see Dantzig,

PAGE 36

26 Problem Realization and Description Components (Symbols) ^ AssumptionsMathematical Model Formulations (Equations ) Model Solution too much time Examination of Results unaccept abl solution Continual updating and examination of the model and results usually necessary Report and Implementation Figure 1. The OR Process (Adapted from Salkin Saha, 1975, p. 2).

PAGE 37

27 1948; 1949). A number of prominent mathematicians also contributed significantly to the groundwork theory and extension of mathematical programming. These included Fourier, Gauss, Kantorovich, Koopmans, Kuhn , Tucker, and von Neuman (Dantzig, 1963, p. 13). A number of mathematical programming techniques have been developed. The most widely used techniques have included linear programming, dynamic programming, integer programming, inventory theory, nonlinear programming, game theory, and network analysis. Linear programming is discussed in a following section. General and specific treatments of all these techniques have been provided in a significant number of texts and articles existing in the literature and the reader is referred to these for elaboration (see Ford & Fulkerson, 1962; Gillett, 1976; Hadley, 1964; Hillier k Lieberman, 1974; Hu, 1969; McKinsey, 1952; Naddor, 1966; Nemhauser , 1966). A typical mathematical program is depicted in Figure 2. What is desired is the determination of activity levels or variable values that will produce optimal system effectiveness. The effectiveness criterion is an equation measuring the system's performance and is labeled the objective function. By introducing constraints and requiring all variable values to be positive, the basic mathematical program characteristics are complete. The intent of this section has been to review the development of the mathematical programming OR approach, demonstrate that a variety of mathematical programming 1

PAGE 38

28 From problem^ description , component analysis, and assumptions Activities or Variables Variable Coefficients and Relationships Between Variables Objective Function and Constraints Nonnegativity Constraint Mathematical Program Figure 2. A Mathematical Program (Adapted from Salkin & Saha, 1975, p. 3).

PAGE 39

29 techniques have been devised, and display the general characteristics of mathematical programs. The succeeding section is to serve as a survey of mathematical programming models developed for regional or national educational systems. Macro Educational Planning Models Introduction The use of mathematical programming OR models as planning tools in education has increased significantly since 1960. In a survey of Organization for Economic Cooperation and Development member countries conducted in 1969, 123 models were found in use or currently under development in 29 different countries (Kahn, 1970). Regardless of apparent widespread acceptance, the use of OR models in postsecondary education has not approached its full potential (Bogard, 1972). Greater utilization of mathematical programming models paralleled the growth of institutional, regional, and national educational planning. McNamara (1971) explained the increased attention to educational planning as follows: The large amount of resources devoted to education and the increasing emphasis on designing educational policy in relation to an overall set of objectives for economic and social development has resulted in the creation of educational plans in virtually all the major nations of the world. (p. 427) In the United States, the demand for more comprehensive and systematic planning in postsecondary education in the

PAGE 40

30 early 1970 's has been attributed to the two primary concerns of efficiency and effectiveness. Both issues fall within the purview of mathematical programming models. In a 1972 review of planning models in higher education, Fincher stated that the concern for efficiency may be said to reflect the public's concern for accountability. The concern for effectiveness, however, may be said to reflect a deeper, more pervasive concern with the effect that higher education has had on contemporary social, economic, political, and technological issues. The rapid expansion of higher education during the past twenty-five years has not been accompanied by the amelioration of social, economic, and political problems. It is the amelioration of these problems that demands more systematic planning in higher education. (p. 760) It was the rapid expansion of post secondary education from World War II through 1970 that was the original reason for increased statewide planning and which spawned an interest in the use of mathematical programming models in education. For example, higher education operating expenditures for the academic year 1972-73 totaled 30.2 billion dollars which represented a fourfold increase from just ten years before. Over this same period, higher education operating costs grew at an annual rate twice that of the gross national product (Halstead, 1974, p. 520). In addition, postsecondary education enrollments had grown to the extent that half of the traditionally college-aged youth were involved in some form of postsecondary education (Gleazer, 1974, p. 6). These factors and others showed the need for statewide postsecondary education planning and coordination, the urgency

PAGE 41

31 of requiring establishment of state-level coordinating boards, and the potential utility of comprehensive OR models in planning. Indeed, McConnell (1962) expressed the holistic view that given limited resources, the entire postsecondary education system of a state must act in collaboration in the pursuit of statewide goals if each institution hopes to have the necessary resources to provide quality education (p. 136). Similar sentiment provided the impetus for creating statelevel coordinating commissions for postsecondary education in the United States. In sum, the genesis of coordinating boards came from a felt need at the state level to know about budgets, enrollment, academic programs, off-campus programming, and long-term building and new campus plans to assure the best possible allocation of scarce state resources. (McTarnaghan , 1974, p. 28) McTarnaghan (1974) viewed the following seven developments necessary before statewide postsecondary education commissions could function effectively. 1. An accurate data base with commonly accepted definitions of terms must be maintained on a statewide basis with current accessibility. 2. A management information system must be developed to answer the "what if" questions of contingency planning and to project future needs, resources, utilization, and costs. 3. Management objectives must be established to specify the goals toward which higher education in the state is striving. 4. A state plan for higher education must be developed which can define state-level goals, missions, resource allocations and enrollment projections. It must also portray to the public, to the governor and legislature, and to all interested in higher education, where the

PAGE 42

32 state is going in terms of higher education, where it should be going, how it should get there, and who is going to do the job. 5. Clearly defined statements of mission and scope of each institution must be developed to focus upon the way in which each institution will contribute to the specified state goals and objectives. Of significant importance is recognition of the role and mission of the private sector. 6. A monitoring system must be established to evaluate the progress of each institution in achieving its mission. 7. Focus must be maintained on desired and needed legislative changes which can strengthen this process and can allocate resources to critical areas of need based on the state plan. (pp. 28-29) The environment necessary before statewide postsecondary education models could be of practical significance was critically dependent both upon the establishment of the state postsecondary education commissions responsible for statewide planning and coordination and the degree to which the commissions met these criteria. The trend to effective statewide coordination in postsecondary education demonstrated by satisfying the above prerequisites and the use of statewide OR models were developments not embraced by every state or educator. Opponents argued that statewide coordination would discourage institutional uniqueness and excellence, that the "bureaucracy" would take over, that system-wide savings would not materialize , and that demands made by the central staff on each institution would offset the services they provided (Harris, 1974, pp. 39-40) .

PAGE 43

33 The issue of how far statewide planning and coordination can go while institutional autonomy and diversity are maintained was addressed in an early study by Moos and Rourke (1959). They warned that "a tightly coordinated system of higher education can leach quality and originality out of state colleges and universities" (pp. 225-226). Yet, before statewide postsecondary education OR models could realistically be used in increasing system efficiency and effectiveness, the creation of state postsecondary education commissions was necessary. The eventual acceptance of the role of centralized state planning is illustrated by the fact that in 1940 there were 33 states out of 48 that had no state postsecondary education commission and in 1974 there remained only two out of 50 states (Berdahl, pp. 2-3, 1975). Pliner (1966) pinpointed the prevailing orientation of most states in the following terms: For the vast majority of states the question of coordination of higher education seems to be settled, and attention is now centered on how to make these agencies [state postsecondary education commissions] more effective in achieving a quality system of higher education that will meet the needs of today and tomorrow, (p. 7) In 1980 the postsecondary education climate in the United States was not one characterized by the growth and prosperity elements of the period when states first undertook the function of planning and coordinating postsecondary education. Rather, it was characterized by diminishing enrollments, rising costs, and personnel retraction (Purga,

PAGE 44

34 1979, pp. 25-33). It has been predicted that enrollment at this level of education would decline 22% between 1980 and 1990 (Glenny, 1974, p. 60). The task of coordinating post secondary education systems has been shown to be much more complicated and difficult under steady state, constrained conditions than under growth conditions (Millard, 1974, pp. 36-38) . Strategies developed to cope with the postsecondary education state of affairs during the 1980-1990 decade needed to broaden their scope to include diverse approaches to planning issues. Rodekohr and Rodekohr (1974) stressed that if educators are to adapt to and face the challenge of this change, they must acquire a new mode of thinking and planning. The greatest opportunity is to utilize the decline for better education, (p. 621) Ideally suited for deriving solutions to complex problems of this nature have been the techniques of OR. Indeed, OR mathematical programming models had become even more relevant as a planning tool by 1980. As Schroeder and Adams (1976) suggested, it has been said that higher education is entering an era of crises, an economic crisis due to dwindling enrollments and rising costs, a crisis of public confidence, and a crisis of student disenchantment. These crises are making our institutions more difficult to manage, and they call for improved management methods and procedures, a number of which fall within the purview of [operations research]. (p. 117) McNamara (1973) further displayed the extent OR mathematical programming models could contribute to statewide postsecondary education long-range planning under constraint conditions. He emphasized that

PAGE 45

35 since mathematical programming allows a planner to view an educational system as a set of input-output or production relationships, which can be controlled in a way that will optimize the use of scarce educational resources, it becomes a valuable technique to generate policy or "decision-oriented" information, (p. 20) To summarize, an attempt has been made to illustrate that the use of OR models in educational planning has grown significantly since the 1960 's. A prerequisite to meaningful statewide postsecondary education modeling in the United States was the creation of statewide postsecondary education commissions. These commissions had to satisfy certain criteria before effective planning and coordination could be realized. The 1980 's presented very different circumstances, both endogenous and exogenous to postsecondary education, than had previous years, making the application of OR mathematical programming models more relevant and essential for statewide planning. In the next section, existing macro educational models are reviewed. Models Postsecondary education planning models for the purposes of this study have been classified as being either micro or macro models. Micro educational planning models are primarily concerned with institutional operations. The scope of these models is confined to educational units ranging from institutional subdivisions, such as departments, to entire institutions. The only factors detailed relate to the internal characteristics of educational institutions. >

PAGE 46

36 Examples of these micro models that have been used extensively are CAMPUS (Judy & Levine, 1965), SEARCH (Keane & Daniel, 1970), RRPM (Hussain, 1971), HELP/PLANTRAN (Midwest Research Institute, 1972), and EFPM (Jones & Updegrove, 1978). The model developed in the study was, however, a macro educational planning model. Macro models relate entire educational systems to a system external to education such as the economic system. The initial interest in models of this nature began with the development of several such models by the Organization of Economic Cooperation and Development (O.E.C.D.) in the early 1960's (Chirikos & Wheeler, 1968, p. 267). In the United States, the two most widely used national planning models for postsecondary education have been the Federal Planning Model (FPM) (Huckfeldt, Weathersby, k Kirschling, 1973) and the Postsecondary Education Financing Model (PEFM) (National Commission on the Financing of Postsecondary Education, 1973). The model developed in this study was a state-level rather than national model. Since postsecondary education in 1980 was overwhelmingly a state responsibility, an aggregate model at the state-level seemed more appropriate as a planning tool. Models developed by the O.E.C.D. and others were surveyed, however, for relevance to this study. Macro educational planning models that have integrated educational and economic planning have been divided into three different approaches by Correa (1975). They are:

PAGE 47

37 (1) the manpower approach, (2) the rate-of-return approach, and (3) explicit optimization models (p. 27). The manpower approach was first used extensively as part of the O.E.C.D.'s Mediterranean Regional Project during the period 1960-75. In brief, the approach ascertains future needs for manpower from projections of the growth of the economy and inputs of labor of various skills; the resulting needs of the economy are compared with expected supplies of labor to determine educational production goals. (McNamara, 1971, pp. 427-428) Criticisms of manpower approach models have included: 1. A lack of attention to productivity changes. 2. Models have been concerned with the labor force required for production whereas available statistics have reported only the labor force being used. 3. Problems of underemployment of labor have not been accounted for since data have not shown the distinction between those who are fully employed and those who are underemployed (Correa, 1969b, pp. 119, 129). The rate-of-return approach has been used as a second technique to integrate macro educational and economic planning. The underlying assumption of this approach is that only one production technique exists. Models utilizing this assumption force all resource allocations (or system inputs) made to different system units to produce equal gross returns (or system outputs). Correa (1975) stated that in practical applications , current educational costs and wages are used to estimate the rate of return to investments in education . . . from this estimate it is concluded that additional investment should be made in the types of education with highest returns. (p. 29)

PAGE 48

38 Major weaknesses of this method have included: 1. The approach leaves out supply, demand, and market mechanisms that determine the costs and returns of education. 2. It explicitly excludes the influence on the rate of return of planned changes in the output of the economy and the technology used. 3. If data do not come from an economy in equilibrium, they do not reflect the actual costs and returns of education (Correa, 1975, p. 29). The third approach used to integrate educational and economic systems for planning purposes has been the use of models that have incorporated characteristics of both the manpower and rate-of -return approaches. These have been labeled optimization models (Correa, 1975, p. 30). Reviews of linear optimization models concerned with macro educational planning have been written by Golladay (1968, pp. 29-68), Johnstone (1974, pp. 177-201), and McNamara (1973, pp. 13-35). Models and Planning The orientation of this study toward planning emphasizes that planning for the future of a system encompasses anticipating all possible system futures that have reasonable probabilities of occurring and developing strategies to adapt , prevent, or achieve each future. A future that is seen as desirable results in devising strategies that increase its probability of occurrence. Contingency plans for less desirable futures are also made which in effect prepare the system for the unexpected. Ackoff (1970) expressed this view of planning as well as a word of caution as follows:

PAGE 49

39 Wisdom is the ability to see the long-run consequences of current actions, the willingness to sacrifice short-run gains for larger long-run benefits, and the ability to control what is controllable and not fret over what is not. Therefore the essence of wisdom is concern with the future. It is not the type of concern with the future that the fortune teller has; he only tries to predict it. The wise man tries to control it. Planning is the design of a desired future and of effective ways of bringing it about. It is an instrument that is used by the wise, but not by the wise alone. When conducted by lesser men it often becomes an irrelevant ritual that produces short-run peace of mind, but not the future that is longed for. (p. 1) The use of models in educational planning has been limited not only by the orientation of individual planners toward the future but also by the models used. A review of the literature by Johnstone (1974) on the capabilities of mathematical models in educational planning uncovered three generally recognized limitations. The first dealt with how well models can predict the future. Mathematical models "cannot account for contingencies which arise unexpectedly or which are out of character with past events" (p. 192). Models, therefore, will only be able to predict approximately the most likely future. Secondly, mathematical models which integrate mathematical tractability and a realistic system representation were seen as impossible to develop. Johnstone explained : Models at a high level of tractability (e.g. simple linear equations) are adequate in the study of certain physical phenomena, but their ability to describe behaviour in the social sciences is severely limited through the almost complete absence of similar linear relationships between variables in this field.

PAGE 50

40 If however, a realistic representation is only possible through extremely complex mathematics, the model is not easily managed or manipulated, and thus concise description is impossible. The model becomes more difficult to manipulate than the situation itself. (p. 192) The third limitation mentioned was that models rarely account for all the aspects of the planning process and therefore must be regarded as just one of the planning tools available (pp. 192-193). The last limitation found by Johnstone centered on the need for integration of mathematical models with the total planning process and to broaden the scope of models. Fincher (1972) summarized a survey and study by Casasco (1970a; 1970b) in which the majority of models in education suffered from a lack of comprehensiveness and understanding by planners: Casasco views the main problem as the limited scope of present models within the total planning needs of colleges and universities. Only 29 percent of the models described in his survey were found to have a sufficient degree of comprehensiveness, a fact that Casasco attributes to the failure of administrators to view institutional development within the framework of a total systems plan. Elsewhere, he has written that this has led to a preoccupation with sophisticated analytical tools that are becoming the ends of planning rather than the means. Dealing only with operational problems such as space requirements, student enrollments, and institutional budgets, these planning tools have not only been underutilized but have led to misunderstandings of their possibilities and limitations. (Fincher, 1972, p. 762) Schematically, Heigham (1969, p. 302) has shown how OR mathematical programming models have been used for systemwide educational planning, Figure 3. Models of the external

PAGE 52

42 environment, the educational system, and the decision-making process were shown to be interrelated and integrated before effective planning was realized. Criteria for the ideal planning process were identified by Ackoff (1970) as including five parts. They are: 1. Ends: specification of objectives and goals. 2. Means: selection of policies, programs, procedures, and practices by which objectives and goals are to be pursued. 3. Resources: determination of the types and amounts of resources required, how they are to be generated or acquired, and how they are to be allocated to activities. 4. Implementation: design of decision-making procedures and a way of organizing them so that the plan can be carried out. 5. Control: design of a procedure for anticipating or detecting errors in, or failures of, the plan and for preventing or correcting them on a continuing basis. (p. 6) A mathematical programming model that optimizes resource allocations is regarded, therefore, as being just one part of the total planning process. McNamara (1971) stated that it should be emphasized that mathematical programming models do not make decisions, nor can they replace judgment on the part of the decision-maker. These models are designed to aid and support decisionmakers by providing pertinent data on alternative programs and courses of actions. Planning models allow decision-makers to form expectations being based on known empirical relationships and the decision-makers' own judgments. (p. 441) One major difficulty recognized as contributing to the development of OR mathematical models in education and

PAGE 53

43 their integration into the total planning process has been the lack of effective communication between the model builder and the decision-maker. Indeed, contributing to this state of affairs is the obvious failure to realize that quantitative, computer-formed rationality is only one of the rationalities appropriate to a complex administrative problem. Political rationality, economic rationality, the valuing process or human relations concerns may be equally appropriate "rational" approaches to the same problem. Since the value of solutions depends on how adequately the model represents the real world problem, effective communications between the consultant and his client are essential and must precede the model building activities. (McNaraara, 1973, p. 29) The determination of how adequate a particular model is for a particular planning objective within the comprehensive planning process had been of vital concern to educators. McNamara (1971) proposed the following group of questions that should be addressed to analyze educational planning models : What is the single system to be modeled? Should a unique model be built for each subsystem? Who are the decision-makers? What assumptions are made about the decision-maker who would use the model? Who is responsible for the generation of alternatives? How are they developed and by what criteria are they compared? What are the goals, overall objectives or targets involved? How are these concepts defined and how are they measured? What is the pertinent time span for the construction of the model as well as for its implementation once it can be validated? What supportive information systems are required for the model? What are the controllable variables? (For which decision-makers? ) What are the uncontrollable variables? (For which decision-makers? ) Can multiple criteria be translated to a single feature of merit? Does the concept of optimization vanish without a single criterion?

PAGE 54

44 How would you implement the model in its intended environment? Are costs of translating an ideal model to a workable model in the real world 'prohibitive ' ? How would you teach people to do the things that the model asks people to do? (pp. 8-9) In sum, mathematical programming models in educational planning have been misunderstood, misused, abused, ignored, or oversold. Hard to quantify issues have been avoided. Personal leanings have been incorporated into models as a means to recruit model results for support. Models have not been effectively integrated into the total planning process. They have lacked a sufficient degree of comprehensiveness on the one hand, or have been so encompassing they have been rendered useless on the other. If the potential of mathematical programming models was to be realized in the 1980 's, increased emphasis had to be placed on achieving the following five objectives: 1. A closer communion between the technical model builder and the operative and policy making model users ; 2. An increasing integration and interaction between large-scale models as conceptual "master plans" and smaller scale problem-solving models dealing with immediate policy issues; 3. A deepening interface of modeling needs with the more routine transactional and stewardship reporting systems of institutional data base operat ions ; 4. A growing emphasis on measuring and recording the outputs, outcomes, products, and effects of the educational experience on individuals and society; and 5. Continuing experimentation especially with various approaches to modeling logic, such 1

PAGE 55

45 as optimized heuristic, normative, and interactive systems in which more emphasis is placed on the order-of -magnitude value judgments of the participants. (Mason, 1976, p. 106) Linear Programming and the MPSX Computer Code Introduction The OR technique used to develop the mathematical model in the study was linear programming. The first portion of this section, therefore, is structured as follows. First, the primary characteristics of linear programs were delimited. The basic mathematical nature of the technique was reviewed along with the necessary theoretical assumptions of linear programming. Next, the techniques of post optimality analysis were discussed. Finally, the utilization of post optimality analysis as a method of generating alternative system futures for use in long-range planning was presented. In the final portion, the general development of linear programming computer codes is reviewed. Since the MPSX code was used in the study, its description comprised the major part of the discussion. Linear Programming Linear programming has been referred to as basically a "mathematical tool for obtaining optimum solutions that do not violate various constraints, that cannot have negative activities, that require linearly proportional relationships, and that account for all inputs and outputs within the system"

PAGE 56

46 (Greenberg, 1978, p. 5). Figure 4 presents the general nature of a linear programming model. A typical linear programming model is presented in vector notation as follows: Maximize Z = cx, subject to Ax :< b, and X ^ o , where c is the row vector X, b, and o are the column vectors such that b. (2.1) (2.2) (2.3) X = b = n and A is the matrix n o = A = ^11 ^12 ^21 ^22 m m2 a In 2n a ran Equation 2.1 is called the objective function. It represents the function that is being maximized. Restrictions on the solutions to the objective function are described by equation 2.2. The restrictions are referred to as functional constraints. Equation 2.3 also represents a set of restrictions referred to as nonnegativity constraints. The x. variables represent decision variables, while the Cy b^ , a^^ are called model parameters (Hillier k Lieberman, 1974, p. 21).

PAGE 57

47 From problem description , component analysis, and assumptions Activities or Variables Variable Coefficients and Relationships Between Variables Objective Function and Constraints Linear Constraints Nonnegativity Constraints Linear Program Figure 4. A Linear Program (Adapted from Salkin & Saha, 1975, p. 3).

PAGE 58

48 The central mathematical problem of linear optimization is to find a vector x which satisfies equations 2.1 through 2.3 In other words, linear programming is a process of selecting the values of the decision variables within a feasible region described by the model constraints which will yield the optimal value of the objective function (McNamara, 1971, p. 423). The main characteristics common to all linear programming models were summarized by Daellenbach and Bell (1970). They identified the following common elements: 1. There exists an objective that is to be optimized, such as maximization of profits or minimization of costs. 2. There exist alternative courses of action, or decision variables, as we call them to achieve the desired objective. 3. There exist restrictions on the amount or extent of attaining the objective, that are expressed in the form of constraints on the values of the decision variables. (p. 5) Implicit in the linear programming model formulation given oy equations 2.1, 2.2, and 2.3 are the four model assumptions of proportionality, divisibility, additivity, and determinism. How well a particular problem can be put into the general model form outlined by the previously mentioned equations and also satisfy all four asstunptions becomes a measure of the appropriateness for using a linear programming model. The assumptions of linear programming were highlighted by Hillier and Lieberman (1974):

PAGE 59

49 Proportionality . In this case (1) the measure of effectiveness Z equals c^^Xj^, and (2) the usage of each resource i equals ^--i^Xj^j that is, both quantities are directly proportional to the level of each activity k conducted by itself (k = 1 ,2 , . . . ,n) . . . Divisibility . The divisibility assumption is that activity units can be divided into any fractional levels, so that noninteger values for the decision variables are permissible... Additivity . The additivity assumption requires that, given any activity levels (x^,X2, • • . ,x^) , the total usage of each resource and the resulting total measure of effectiveness equal the sum of the corresponding quantities generated by each activity conducted by itself... Deterministic . The deterministic assumption is that all the parameters of the model (the a . , b-, and c. values) are known constants. In real"^ problems this assumption is seldom satisfied precisely. Linear programming models usually are formulated in order to select some future course of action. Therefore the parameters used would be based on a prediction of future conditions, (pp. 22-24) In certain cases, where some but not all of the assumptions of linear programming have been met, constructing a linear programming model of the problem has still yielded valuable results. Daellenbach and Bell (1970) mentioned that even if some of the conditions [linear programming assumptions] cited are not satisfied by a particular problem we may still attempt to use an LP model as a first approximation. If some components of the problem exhibit some random behavior, we may substitute averages for these components. If in practice the decision variables are restricted to integers only, it might be possible to approximate them by continuous variables and suitably round the solution obtained to integers. If the objective or some of the constraints exhibit in fact nonlinear behavior, we might attempt to substitute suitable linear approximations for them. (pp. 5-6) I

PAGE 60

50 The algorithm used to solve linear programming models was developed by Dantzig in 1947 (Hadley, 1964, p. 14) and is referred to as the simplex method (see Dantzig, 1948; 1949). The simplex method is an iterative procedure that reaches an optimal solution to the model in a finite number of steps or gives an indication that the model solution is unbounded. Unbounded solutions arise in models where the objective function value Z can be made infinitely large. This situation generally indicates a misformulat ion of the model. A model that has an optimal solution must have an objective function value Z which is finite. The following two theorems siimmarize the discussion: 1. If a linear programming model has an optimal solution, at least one basic feasible solution will be optimal. (A basic solution is defined as a solution that has no more than m variables different from zero. Where m is the number of functional constraints. A basic feasible solution is defined as a basic solution where all m variables are non-negative and called degenerate if any of the m variables equals zero . ) 2. If a linear programming model has a basic feasible solution which is not optimal, it is possible to obtain an optimal basic solution in a finite number of steps by changing a single basic variable at each step, or to obtain an indication of an unbounded solution. (Hadley, 1964, p. 31) Proof of the two theorems and ultimately the use of the simplex method to solve a linear programming model is mathematically more convenient if the model has been converted from the form represented by equations 2.1, 2.2, and 2.3 to the following standard form:

PAGE 61

51 Maximize Z = cx, (2.4) subject to Ax = b, (2.5) and X > o, (2.6) where in general the A, b, c, and x matrices are different here than those used in equations 2.1, 2.2, and 2.3 Transforming the inequality form of equation 2.2 to the equality form of equation 2.5 is accomplished by introducing additional variables called slack variables in the case of "less than" inequalities and surplus variables in the case of "greater than" inequalities. In addition, it is usually convenient if all components of the b vector are positive. Artificial variables must be added in some cases to secure an initial basic feasible solution. Briefly, the simplex method begins with an initial basic feasible solution of the linear programming model in the standard form. It then computes the c.'s for all x.'s J J not in the basis. If all these x.'s have negative c.'s the J 3 simplex method concludes that the solution is optimal. If any x. not ^ in the basis has a positive c., the simplex first J J checks for the possibility of an unbounded solution. If there is no indication of an unbounded solution, the simplex method selects the x. not in the basis with the largest positive c. to enter the basis. The simplex method next removes an Xj in the basis in a manner to insure a new basic feasible solution. The procedure is repeated until an optimal solution or unbounded solution is reached.

PAGE 62

52 To siimmarize the discussion on linear programming to this point, it was stated that a number of assumptions must be satisfied before linear programming can be considered as the appropriate technique for approaching a particular problem. Next, the steps of the simplex method were discussed. Naylor and Bryne (1963) summarized the procedure as follows: 1. Define the problem in terms of a linear objective function subject to a set of linear restraints. 2. Introduce slack and artificial variables to convert inequalities to equalities and provide a basic feasible solution to the problem. 3. Construct the initial simplex table for a basic feasible solution. 4. Determine which variable should be introduced into the solution, if the solution is not optimiam. 5. Determine which variable should be removed from the solution. 6. Construct a new simplex table reflecting the changes that have been made in the solution. 7. Repeat steps 4 through 6 until an optimum solution is reached, i.e., until the solution cannot be further improved. (pp. 47-48) In fact , the first two phases of the OR process as suggested by Ackoff (1956) and previously mentioned in the chapter have been addressed thus far. That is, the formulation of the problem and the construction of the linear programming model. The third phase of the OR process is to derive a solution from the model. The simplex method for obtaining linear programming model solutions was presented. In addition, procedures to determine the stability of the solution are necessary before the third phase is complete. The

PAGE 63

53 procedures have been called sensitivity analysis and have been grouped under a broader process variously referred to as post optimality analysis (Randolph & Meeks, 1978, p. 196), post optimization (Simonnard, 1966, p. 138), or postoptimal analysis (Greenberg, 1978, p. 118). Sensitivity analysis has been referred to as the process for investigating the effect on the optimal solution of a linear programming model to changes in the model parameters — the a. .'s, b.'s, and c.'s. Sensitivity analysis is important for several reasons. First, in some instances a few of the a . . , b., and c. parameters may be controllable (Dantzig, 1963, pp. 266-267). The effects of changes in these parameter values would be desirable to know since, for example, advantages may be gained in the objective function value. Model parameters, particularly the b^'s, are sometimes "set as a result of policy decisions . . . and these decisions should be reviewed after seeing their consequences on what can be achieved" (Hillier & Lieberman, 1974, p. 182). Another reason for sensitivity analysis arises from the fact that the a^., b^, and c. values may only be estimates based on some approximation of conditions in the future. The deterministic assumption of linear programming models stated that the a^ . , b., and c. be known constants, but in reality this assiimption can rarely be said to hold. Sensitivity analysis is used to determine the stability of the optimum solution when the model parameters are assumed constant and the actual values of these parameters are only

PAGE 64

54 approximations. If the optimvim solution changes with small changes in any of the parameters the relationship between the sensitive parameters and the solution must be critically analyzed. On the other hand, an optimum solution that is insensitive to small parameter changes indicates a reliable linear programming model solution (Greenberg, 1978, p. 57). A third reason for sensitivity analysis relates also to the assumption that the a. ., b., and c. parameters are known constants. In this case the parameters are known but recognized as variable. Sensitivity analysis gives the range any parameter or set of parameters can vary within while the current solution remains optimal (Dantzig, 1963, p. 267). The importance of the discussion on sensitivity analysis relates directly to how the linear programming model developed in the study was used for long-range planning by generating multiple system futures and developing system strategies to adapt, prevent, or achieve each possible future. When, for example, sensitivity analysis is used to define the range of a model parameter (a^^, b^^, and c^ ) where the model solution remains optimal, it becomes a measure of the optimal solution evaluation criterion called resilience. Nair and Sarin (1979) explained that resilience can be defined as the ability of a plan [optimal solution] to perform adequately under a variety of futures. Resilience can be measured in a number of ways. The cost to change plans might be one measure. The variation from some desired performance under various future scenarios might be another measure. Therefore, an essential element

PAGE 65

55 of the evaluation process is the development of future scenarios under which plans [optimal solutions] should be evaluated. (p. 57) A technique frequently used by United States industrial companies to evaluate proposed strategies when subject to alternative futures has been multiple (or alternative) scenario analysis (MSA) (Linneman & Klein, 1979, p. 83). The following steps describe the general MSA procedure: (a) isolating assumptions about the future which you are sure will occur within your planning timeframe; (b) identifying key 'impact variables'; (c) specifying other environmental variables which might affect the behavior of the impact variables-what might be called 'cross impacts'; (d) constructing at least two descriptions of possible future — scenarios — which depict a range of behavior of the impact variables; and (e) developing strategies which are responsive to one or more of the scenarios. (Linneman k Klein, 1979, p. 83) How resilience and MSA relate to linear programming sensitivity analysis is obvious. Yet two critical points must be resolved during the early period of any scenario exercise. First a time scale or planning horizon must be determined and secondly the number of scenarios which will be investigated must be set (Chapman, 1976, p. 4). Selection of the five-year planning horizon for the linear programming model developed in the study was based on a priori determination of the period. A number of quantitative approaches to determine planning horizons have been developed but were not amenable to the study model (Friedman & Segev, 1976; Modigliani & Hohn, 1955). The planning horizon was necessarily set on the basis of the accuracy and availability of various future forecasts incorporated into the model.

PAGE 66

56 The other crucial decision involves setting the number of scenarios to develop and explore. Results of a survey conducted by Linneman & Klein (1979) on the number of scenarios generated by a sample of corporations are presented in Table 2. Table 2 Number of Scenarios Generated (Adapted from Linneman & Klein, 1979, p. 87) Number of Scenarios Number of Companies 2 11 23 1 3 20 34 4 4 2 more than 6 1 More than half the respondents listed the number of scenarios typically generated for their most distant planning horizon as three. In the study model three scenarios were also used. The reasons for using three included the following: (1) keeping the scenarios developed down to a manageable number, (2) time and cost constraints limited exploring many more scenarios, and (3) it was assumed the three scenarios used represented futures with the highest probabilities of occurrence. The last point need not be an assumption of course. Methods such as cross-impact analysis (Fontela, 1977), the delphi technique (Linstone & Turoff, 1975), and bayesian statistics (Morris, 1974; 1977) do indeed provide procedures

PAGE 67

57 for ranking scenarios. Rigorous application of a model such as the one developed in this study would include utilizing such techniques. In summary, the linear programming OR approach has been reviewed in general terms. The review of literature to this point has been developed after Naylor's (1977) sixstep approach for integrating planning models into the planning process: 1. Review of the planning environment. 2. Specification of planning requirements. 3. Definition of goals and objectives for planning. 4. Evaluation of existing planning resources. 5. Design of an integrated planning and modeling system. 6. Formulation of a strategy for integrating the planning model into the planning process, (p. 11) In addition to addressing the question of integrating the model into the overall planning process, the first five of the following eight basic elements Naylor and Mansfield (1977) considered essential for specifically designing computer based planning and modeling systems have been investigated: 1. Planning System. 2. Management Information System. 3. Modeling System. 4. Forecasting System. 5. Econometric Modeling.

PAGE 68

58 6. User Orientation of the System. 7. System Availability. 8. Software System. (p. 16) The last three elements are attended to in the final section of the literature review on the MPSX computer code. MPSX Computer Code The first linear programming model solved using a digital computer was on the National Bureau of Standards computer, the SEAC, in 1952 (Gass, 1975, p. 235). In the early 1950 's, models with as many as two hundred functional constraints and a thousand decision variables were accomodated and solved in approximately five hours by computers (Dantzig, 1963, p. 26). By 1975, successful solutions had been obtained for linear programming models with 50,000 functional constraints and 285,000 decision variables (Salkin & Saha, 1975, p. 40). Two approaches have been used to derive solutions from complex linear programming planning models on computers. The first has been to use scientific programming languages such as FORTRAN or APL. Another approach taken has been to use already existing planning and modeling software systems, or computer codes. Advantages of developing specific computer programs for every linear programming model have included the extreme flexibility of scientific programming languages and also the fact that the languages have been well-known. Disadvantages have included the following:

PAGE 69

59 1. The possibility planners are not familiar with any of the languages. 2. The main application strengths of languages such as FORTRAN and APL do not include databased management and report generation. 3. The languages cannot provide assistance in formulating and coding models. 4. Changes in policy assumptions, report formats, external assumptions, and forecasts cannot be implemented easily (Naylor & Mansfield, 1977, pp. 23-24). Disadvantages of using already existing planning and modeling softward systems have included the fees users must pay to the lending firm and the fact that the computer running costs are higher. The advantages of using planning and modeling software systems that have been designed to assist the planner formulate and code models have included the following: 1. The systems are easy to use. 2. The systems provide a conceptual framework for planning and modeling making model development easier . 3. They are reasonably flexible. 4. Implementing such things as forecasting are much easier. 5. Use of already existing systems usually lead to reduced total project costs (Naylor & Mansfield, 1977, pp. 23-24). In the study, an existing planning and modeling software system was used primarily to make model parameter changes, which represented multiple scenarios, easier. Additionally, using a software system represented a considerable savings in time and total study costs.

PAGE 70

60 In 1980, there existed many commercially available linear programming software systems, or computer codes. Salkin & Saha (1975, pp. 40-46) listed and described twelve such codes that have been frequently used. International Business Machines' (IBM) Mathematical Programming System Extended (MPSX) computer code was used in this study. Standardization and availability were the main reasons for choosing the MPSX code. The general MPSX card deck layout is presented in Figure 5. Cards 1-6 are the job control language cards (JCL) which provide control information to the computer system. The JCL depicted were for the University of Florida. The control program and data vary for each model. A number of publications have listed the formatting requirements and available MPSX options (see Greenberg, 1978, pp. 99127; Randolph k Meeks , 1978, pp. 41-231; Sposito, 1975, pp. 241-260). Summary The review of the literature demonstrated that the operations research technique of linear programming applied to macro educational planning is an appropriate approach to long-range planning, particularly in view of the advances made in statewide postsecondary education coordination. Additionally, linear programming and the MPSX computer code were shown to facilitate multiple scenario analysis. Subsequent

PAGE 71

61 /* DATA / /PROBLEM. SYSIN CONTROL PROGRAM )D^ //CONTROL. SYSIN \ — SDD* //EXEC MPS /^PASSWORD //JOB CARD Card 6 Card 5 Card 4 Card 3 Card 2 Card 1 Figure 5. MPSX Card Deck Layout

PAGE 72

62 chapters attempt to develop a general linear programming model for statewide postsecondary education systems and demonstrate the use of the model to devise long-range system strategies that adapt to, prevent, or achieve various system futures .

PAGE 73

CHAPTER III MODEL DEVELOPMENT Introduction The focus in this chapter is on the development of a prototype linear programming model for postsecondary education systems in the United States. The multiple scenario analysis approach is used to create different system constraint conditions. The model is employed, then, to devise alternative system strategies that consider each new scenario situation. The major aspects of the model are comprised of three submodels. Simply stated, the first submodel is the student vector or supply aspect of the model. The number of students in a particular program at a particular postsecondary education institution was controllable by constraint equations which incorporated variables such as the potential student pool, funding patterns, and square feet of assignable space. The second submodel is the future occupational demand aspect of the model. This future demand was described in terms of manpower requirements as a function of state geographic area and occupation. The last submodel integrated the flow of postsecondary education institution program graduates into "initially entered" occupations. The integration

PAGE 74

64 relation was represented by a probability matrix of where graduates of particular programs from particular postsecondary education institutions go in the state and what occupations they initially enter. The model has incorporated highly aggregated aspects of student flow modeling, fiscal planning, program planning, facilities planning, and manpower requirements modeling. In this chapter, factors that were used to construct linear constraint equations in each of the three submodels are described . Forecasting the Supply of Graduates Submodel Numerous enrollment projection techniques have been developed that vary considerably in sophistication and applicability. Various descriptions of enrollment projection techniques have been made by Lins (1960), Lyell and Toole (1974), and Mangelson, Norris, Poulton, and Seeley (1974). The most frequently used mathematical enrollment projection techniques have been trend analysis, curve fitting techniques, ratio method, cohort survival, regression analysis, and Markov chains. Render (1977) combined several of the techniques in a state system enrollment forecasting model (pp. 21-29). Indeed, as Render stated, the techniques listed above may not be considered mutually exclusive, for all could conceivably be present in a given study of national or statewide forecasting. (p. 21)

PAGE 75

65 Since a number of techniques have been developed and the possibility of using more than one technique in a comprehensive model of statewide enrollment projections has been recognized, the problem has been reduced to choosing the technique or combination of techniques best suited to the purposes and orientation of the planning. Evans (1975) cautioned, however, that the constraints inherent in the methods cited above are abundant, and the detriment incurred in their use (particularly the less sophisticated cohort, ratio, and curve-fitting approaches) varies with the context of the application. (p. 1) The classification system for categorizing enrollment projection techniques described by Gardner (1980) was used to integrate the overall planning objectives of the linear programming model with the supply relations. Gardner described four basic types of enrollment forecasting methods: !• Patterned based . Methods in this category attempt to identify an underlying pattern in the historical data, describe it in mathematical terms, and then extrapolate it into the future. . . . 2. Correlation analysis . This type of approach develops projections on the basis of the numerical values of "leading indicator" variables which have been demonstrated, e.g., through the use of multiple regression, to have a strong relationship to enrollment levels. . . . 3. Intention surveys . Intention surveys involve sampling specific groups to determine college attendance plans. . . . 4Professional judgment . . . . The lack of objectivity of this method may be more than offset by the benefit of experienced judgment and interpretation of subtle factors influencing enrollment not easily captured by mathematical models, (pp. 2-3)

PAGE 76

66 Patterned based techniques, correlation analyses, intention surveys, and the professional judgment technique have all been integrated in the construction of the supply segment of the model. The procedure to calculate the supply of postsecondary education institution program graduates was divided into three steps: (1) the highly probable student demand calculation, (2) formulation of the student access control equation, and (3) the calculation of the number of graduates. Step 1 — Determining the Probable Student Demand A description of what the probable student demand calculation is might best be accomplished by an examination of what it is not. It is not the number of students that will be attending public postsecondary education institutions at some specified time in the future. It is not a highly abstract enigma, nor an information-hungry simulation model. The calculation is a first approximation of what might result in the future with program enrollments derived from straightforward patterned based techniques. Both causal and trend dimensions were integrated using the ratio method and the double exponential smoothing technique in a linear combination that could be tempered with professional judgment by the selection of weighting factor values. The particular form of the calculation is as follows:

PAGE 77

67 X . 6 . . H. = ^ — R. + ^ — X. (3.1) 1 + 1 '^i ^' where time subscripts are understood and H. is the probable student demand by program headcount for institution i for future time t, j R. is the program headcount enrollment forecast i for institution i by the ratio method for future ' time t, is the program headcount enrollment forecast for institution i by the double exponential smoothing technique for future time t, , is the weighting factor for the ratio method forecast of program headcount enrollment for institution i for future time t, and 6^ is the weighting factor for the double exponential smoothing forecast of program headcount enrollment for institution i for future time t. The calculation of the program headcount enrollment forecast for institution i by the ratio method (R^) is straightforward. Briefly, a simple table in the form of Table 3 is employed to organize data. Data on program headcount by institution, age cohort, and county as well as total population by age cohort and county are compiled over a base period of x years from a year s through year (s + x-1). Here, (s + x-1) represents the most recent year that data were available for and is defined as the base year. For each year y in the base period, the ratio d^^ of the program headcount enrollment to total population by age cohort i is calculated. Next, all the d^^ ratios for the entire base period in a particular age cohort are considered to obtain an adjusted ratio d^. A degree of professional judgment is I

PAGE 78

68 interjected at this point to decide whether or not to use the median of the d„ ratios, the mean, the most recent value, or the trend pattern. The adjusted ratios are multiplied by the age cohort population forecast for future year t to yield enrollment forecasts by age cohort and county. By summing these forecasts over all counties and age cohorts the program headcount enrollment forecast for one institution is obtained. For example, if the d ratios are combined x.y to obtain adjusted ratios by taking their mean, the program headcount enrollment forecast for institution i by the ratio method for future time t becomes R. = 1 I I k I •£kt I [ I "i^ky]' (3.2) where R. 1 £kt is the program headcount enrollment forecast for institution i by the ratio method for future time t, is the population forecast by age cohort £ and county k for future time t, a„,„. is the program headcount enrollment for insti'Jiky tution i by age cohort Z and county k for base period year y, *^2,kv population by age cohort I and county k for base period year y, X is the number of years in the base period, and y is a base period year, where y = l,2,3,...,x. A further description of the ratio method as well as a listing of suggested population data sources have been provided by Lins (1960, pp. 11-15).

PAGE 79

69 Table 3 Calculation of the Program Headcount Enrollment Forecast for One Institution by the Ratio Method A B C D Enrolled in Programs % of Total Total Pop^i^ation Ratio Coionty (k) • • • Coiinty (k) 1,2 , . . . ,n Age Cohort ^iky ^5ky ^£y^^iacy I Repeat A,B,C, and D for each year y in the base period. Here, y=i,2, • • • y ^ * D' Adjusted Ratio E Total Population for Future Year t County (k) 1,2,. ..,n F Enrollment Forecast for Future Year t County (k) 1,2, . . . ,n Age Cohort I ^£kt ^J2kt"^' ®£kt Ratio Method Forecast for Institution i

PAGE 80

70 The ratio method is a patterned based enrollment forecasting technique that is causal in the sense that the forecast it yields is a function of the anticipated state population by age cohort. For postsecondary education institutions such as community colleges that have had a student market comprised of primarily in-state students, the ratio method as described has worked accurately. For an institution that draws heavily from out-of-state, however, a modification to is required where the median, mean, the most recent, or the trend determined value of the out-ofstate headcount enrollment for future time t is added to R. . 1 The overall linear trend in total program headcount enrollment is introduced into the probable student demand calculation by the second term in equation 3.1. The second term, X^, is the double exponential smoothing forecast of program headcount enrollment for institution i for future time t. Wheelwright and Makridakis (1973, pp. 44-47) have provided a detailed analysis of the theoretical foundation of the technique. The basic mathematical calculation is summarized as follows: X. = X 2 + a(t-s-x-M ) S' a-1 x+1 (3.3)

PAGE 81

71 with where X. is the third level double exponential smooth1 ing program enrollment headcount forecast for institution i for future time t, S' ^ is the single exponential smoothing forecast y+1 for base period year y+1, S"^^ is the second level double exponential smooth^ ing term for base period year y+1, (t-s-x+1) is the number of years into the future that the forecast is made from the base year, a is the exponential smoothing constant with a value from 0 through 1, X is the actual program enrollment headcount ^ for base period year y at school i, x is the number of years in the base period, and y is a base period year where y = l,2,3,...,x. The procedure for determining X^ is to first compute Sy+l and S^_|_^ for the first year in the base period, y = 1, and then repeat these calculations for the remainder of the base period years through y = x. Once S^_^^ and S^^^ have been obtained, simply substituting their values in equation 3.3 yields the double exponential smoothing forecast of program headcount enrollment for institution i for future year t. Of critical importance when using the double exponential smoothing technique is the value of a used in the calculation. The larger the value of a, the greater the

PAGE 82

72 weight is given to the more recent values of to determine the enrollment trend projected into the future. The selection of a is facilitated by the use of the "Average Absolute Difference" method (Gardner, 1980, pp. 10-11). Using the technique, different values of the exponential smoothing constant a are generated to compute different ^y+i each year y in the base period. For each value of a the absolute differences over the entire base period between ^y+i the actual program headcount enrollment value ^y+i each base year y are calculated. The a chosen to estimate future program headcount enrollments will be the a which has yielded the smallest average absolute error value. Expressed mathematically, X E(a ) = ^ I •7. V ^ y=i S"^T (a ) X y+1 y+1 (3.6) where E(a^) is the average absolute error value with a = where z = 1,2,3,. ..,ra, S" ^ is the second level double exponential smoothing term for base period year y+1, actual program headcount enrollment at school i for base period year y+1, X is the number of years in the base period, y is the base period year where y = 1,2,3, . . . ,x , and a is the exponential smoothing constant with a value from 0 through 1. Table 4 summarizes the double exponential smoothing calculation.

PAGE 83

73 Table 4 Calculation of the Program Headcount Enrollment Forecast for One Institution by the Double Exponential Smoothing Technique a = ^1 etc. a = a m Base Period Year Actual Enrollment Headcount Acadenic Year Single anoothing Second Level Shoot hing Single Smoothing Second Level Stoothing s 1 \ s + 1 2 Y s + 2 3 h f3 • ^3 • • s+x-1 x X X • S' X « S" X • S" X • S" X Average Absolute Error E(a^) E(aj Academic Year Base Period Year Use a from min[E(a )] Single Smoothing Second Level ShTKDthing S + X X + 1 ^x+l Sx+1 Double Exponential Smoothing Forecast X. For Institution i 1

PAGE 84

74 Once the ratio method and double exponential smoothing program headcount enrollment forecasts have been computed for an institution for future time t the values are inserted in equation 3.1 To complete the program headcount probable student demand calculation, , only the values of the weighting factors and 6^ are required. As mentioned earlier in this section, the weighting factor values are determined by professional judgment. Indeed, the forecasts obtained by the ratio method and the double exponential smoothing technique in reality set bounds on the probable student demand. The weighting factors allow the knowledgeable administrator to prejudice more toward one of the forecasts than the other. Since individual institutions and entire state postsecondary education systems have allocated resources and reported statistics on operations with the full-time equivalent (FTE) student and not the headcount student as the common denominator, the student demand calculation, H. , ' 1 ' is converted in the final step from a program headcount figure to a FTE student figure by the FTE to headcount ratio for the institution. This ratio is determined also by the double exponential smoothing technique previously described. To summarize, = g^H. , (3.7) where Fj^ is the probable student demand in program FTE's for institution i for future time t, is the probable student demand calculation by program headcount for institution i for future time t, and

PAGE 85

75 g. is the third level double exponential smoothing ratio of program FTE ' s to headcount for institution i for future time t. Step 2 — Controlling Student Access Equation 3.7 was derived to estimate the student demand in program FTE ' s for an institution at some future time t assuming that current trends are projected linearly into the future. In calculating it was assumed, for example, that the general characteristics of the student market for school i, the number of square feet of assignable space, the admission policy, scheduling patterns, and the funding, tuition, and aid trends all remain stable over the base year through future year t time period. Calculation of F^ was accomplished using an exploratory futures forecasting approach. One will recall from the first chapter that exploratory methods simply extend past or current patterns into the future. Yet the variables inherent in calculating F^ assumed to be linear projections of past trends are not inflexible, mandated entities but to a large degree reflect controllable administrative policies at the state and/or institutional level. In the first chapter, normative futures forecasting techniques were defined as beginning first by determining future needs or goals at time t and then formulating a strategy to attain the goals working forward from the base year. Any policy change different from the trend projection of that policy into the future therefore constitutes an alternative probable student demand future from the one depicted by equation 3.7.

PAGE 86

76 For purposes of discussion and analysis, only two policy dimensions were considered in the model. Ideally, the effect of altering policies from their trend projections on equation 3.7 would include investigating funding patterns, square feet of assignable space, admissions policy, tuition, and student aid dimensions. Only cost and square feet of assignable space for institutions by programs were included in the model. With just two dimensions available to create alternative strategies that adapt to, prevent, or achieve various future scenario conditions, a considerable number of possibilities exist. Demonstrating this point is the subject of the next chapter. The first constraint limits the number of full-time equivalent (FTE) students enrolled in programs at a particular institution to being less than or equal to the probable student demand calculation, , for the institution. By not changing admission policies, tuition, and student aid trends, F^ can be interpreted as setting an upper bound on the number of FTE's that will be generated by an institution at some future point in time. The degree to which the F^ ' s are satisfied for all postsecondary education institutions therefore indicates how well the system meets the anticipated needs of the citizenry. In sum, I Pij 1 F. , (3.8) where Pj^j is the number of FTE students enrolled at institution i in program j for future time t , and

PAGE 87

77 F. is the probable student demand calculation in program FTE ' s for institution i for future t ime t . A program cost constraint limits the number of program FTE's at an institution in each program area at a future point in time. For each program at an institution P. .c. . < C. . , (3.9) where P. . is the number of FTE students enrolled at "'"'^ institution i in program j for future time t, c. . is the cost per FTE student enrolled at i i institution i in program j in base year dollars, and . is the total dollar allocation in base year dollars for program j at institution i for future time t. By equation 3.8, P^^ is a finite number. Equation 3.9 facilitates the control of: (1) program FTE student distribution, (2) the extent that the probable student demand for educational services at an institution are satisfied, and (3) the manner in which institutional resources are distributed. Extensions of equation 3.9 from resource allocation at the program level to resource allocation at the institutional and at the state level are described by equations 3.10 and 3.11, I Plj<=ij 1 I , (3.10)

PAGE 88

78 Equation 3.10 expresses the statement that the total money allocated to all programs at an institution cannot exceed the money the institution has available. Equation 3.11 states that the total money spent by state postsecondary education institutions cannot exceed the amount allocated by the state legislature. The last constraint needs modification in the case where local tax support for postsecondary education institutions exists. To siimmarize, the program cost constraint set of equations 3.9, 3.10, and 3.11 address the following three questions : 1. How much money should the state provide for postsecondary education? 2. How should the state funds provided for postsecondary education be distributed to postsecondary education institutions? 3. How should the funds provided to postsecondary education institutions be distributed to programs? In similar fashion, the number of program FTE's at an institution is constrained by the physical facilities available for instruction. The facility constraints are where Pija,. < A.. , (3.12) I ^i.i^.1 ^ I hi ' (3.13) I I PiH^. . < I I A , (3.14) Pj^j is the number of FTE students enrolled at institution i in program j for future time t,

PAGE 89

79 a . is the number of assignable square feet of space required for instruction per FTE student enrolled in program j at institution i, is the total number of assignable square feet of space for program j at institution i for future time t. Both the program cost and facility constraints incorporate aspects of exploratory and normative futures forecasting techniques. For example, both c^^ and a^j are determined by historical data utilizing exploratory methods. On the normative approach to futures forecasting. For each different C. . and A. . selected, an alternative post secondary education system future state is achieved. Indeed, each and set chosen represents a specific postsecondary education system strategy to respond to a specific social demand-economic need future scenario. The social demand aspect of the scenario has already been delimited to the probable student demand calculation, F^ . The economic need aspect is represented by manpower requirements projections described in a later section of this chapter. Within the confines of total system fund and facility resources, the degree that a particular set of C. . and A. . values satisfy the social demand-economic need scenario is in fact a measure of the adequacy of the strategy. Step 3 — Estimating Graduates The third and final step in the supply of graduates submodel is the calculation of the expected number of other hand, C. . and A. . are set a priori and represent the

PAGE 90

80 graduates from a particular program at an institution at some future point in time. Patterned based techniques are used to derive the ratio of the number of program graduates to the number of program FTE students for future time t which, when multiplied by the anticipated number of program FTE students for future time t, yields the anticipated number of program graduates for that year. Simply put, G. . = r. .P. . , (3.15) where G. . is the niimber of anticipated graduates from '"^ program j at institution i for future time t, r. . is the ratio of the number of graduates from i T program j at institution i to the number of FTE students in program j at institution i for future time t, and P. . is the number of FTE students enrolled at """•^ institution i in program j for future time t. Equation 3.15 expresses the assertion that given an empirically determined ratio of the number of graduates to the FTE enrollment in a program at a postsecondary education institution, the expected number of graduates from that program for some future year can be estimated if the FTE program enrollment is given. Conversion to headcount graduates is necessary since manpower requirements projections on the whole have estimated the number of individuals needed by various occupations. The integration of future occupation demand and student flow constraints is the topic of the next two submodels.

PAGE 91

81 Forecasting Occupation Demand Submodel The occupation demand segment of the model involves the generation of a forecast of the demand for manpower by occupation and state geographic region for some future time. Manpower requirements models, such as those developed by Chance (1966), Correa (1969a), Le Vasseur (1969), Maki (1970), and Ritzen (1976), have modeled relationships between the education and economic sectors to optimize investments in education in order to satisfy labor demands. Indeed, the forecasting of manpower needs has become a routine matter in various divisions of federal and state agencies. The increasing level of sophistication and mathematical rigor of the state-of-the-art manpower models advises the integration of existing manpower forecasts with this model and not the development of a manpower submodel to generate original forecasts. Thus, the problem reduces to one of locating and integrating existing manpower forecasting efforts. A problem that has existed with the integration of federal manpower requirements forecasts into the model, for example the manpower forecasts available through a review of the Dictionary of Federal Statistics for Local Areas (U. S. Department of Commerce, 1978), has been that manpower requirements forecasts have been reported in statistics too highly aggregated by occupational group and by geographic region. Ideally, forecasts for occupations requiring postsecondary education training by state county are needed for the model.

PAGE 92

82 A review of the Dictionary of Occupational Titles (U. S. Department of Labor, 1977) showed that the federal government has broken occupations up into nine broad categories. The nine categories have been further subdivided into a nine-digit DOT number which delimits a specific job title. The first three digits of the DOT occupational titles represented occupational groups that best related to preparation programs offered in postsecondary education institutions. In 1980, there were 979 three-digit occupational group DOT numbers specified. Manpower requirements forecasts existed for a number of the groups, however no unified, comprehensive analysis of all 979 occupational groups was found. The model is handicapped to the extent that manpower needs forecasts for occupations requiring postsecondary education training are available. Manpower forecasts for highly delimited occupational groups by the federal government have used states as the local areas or regions for statistical reporting almost invariably. Breaking down the state totals for expected additional manpower by DOT number for some future time by county was accomplished by taking the arithmetic mean of two elements, one prorating the change in occupation need to the ratio of anticipated county population to anticipated state population change from the base year to the future year, the other prorating the manpower requirements using the ratio of future county to state population. To summarize.

PAGE 93

83 I 0 k£ t I = m=s+x-l I 0. k£m (t-s-x+1 ) , 'k£(s+x-l) I \n + ^ ^ d » 2 ^Ji (3.16) where 0 ^k J^Pk-Pk> Pk k ^ I 0 k£ (3.17) I I 0 in=s+x-l k£m I 0 k£ 0 k£ is total additional manpower needs for occupation £ in a particular state from the base year (s+x-1) through future time t, is the total additional manpower needs for occupation £ in a particular state for future time t, is the anticipated population of county k for future time t, is the population of county k in the base year, is the total additional manpower needs for occupation £ in county k for future time t, and d^ is the attrition rate for occupation £. Once the overall additional manpower needs by occupation and county for some future time have been computed, in-migration and private institution shares must be considered before a figure can be obtained for the maximiun number of program graduates that institutions in the public postsecondary education system should generate. Both factors can

PAGE 94

84 be obtained using patterned based forecasting techniques. Assuming that private institution graduates and in-migrating individuals distribute in a state in proportion to county population, the result is where Jj^^ is the anticipated need for public postsecondary education graduates for occupation £ in county k for future time t, Oj^^ is the total additional manpower needs for occupation I in county k for future time t, is the average number of jobs per county in occupation £ filled by in-migrating workers for future time t, and is the fraction of jobs in occupation £ filled by public institution graduates for future time t. The Student Flow Submodel The last model segment integrates the product of postsecondary education institutions, namely graduates, with the future manpower needs of the state. The overriding interest in developing constraints in this submodel is that a postsecondary education system should be in tune with the economic needs for an educated work force of specific composition. Too few graduates prepared for a certain occupational group produces manpower shortages, decreased service quality, and economic loss. Too many graduates prepared for a certain occupational group produces manpower oversupplies ,

PAGE 95

85 frustration, and results in wasted investment both by the state and the individual. Blame for either situation rests with ineffective educational planning. The very fact that both students and state funds earmarked for postsecondary education systems are scarce resources dictates that supply and demand factors be an integral part of responsible educational planning. Accomplishing the goal of supply and demand integration is achieved by first ascertaining which occupations graduates of particular programs have entered in the past. These quantities are defined as participation rates. Assuming that the participation rates remain stable until the planning horizon, when the rates are equated with future occupation needs, they gauge how many and in what program areas graduates from particular postsecondary education institutions are required in specific areas of a state. Participation rates are determined by an analysis of where and into what occupations the past graduates of particular programs at an institution have entered after some defined time lapse after graduation. The time lapse functions as a job hunting and adjustment period for graduates. Participation rates are necessary for relating each program an institution offers to each occupational group in each county in the state. Table 5 summarizes the information required where the aj^(i,£) are defined as the participation rates for graduates from program j at institution i that have found employment in occupation I in county k.

PAGE 96

86 Table 5 The Student Flow Matrix Participation Rates a.j^(i,£) Institution 1 Program k = 1 Count 2 y 3 etc. Occupation £=1 2.. . Occupation £=1 2.. . Occupation £=1 2.. . • i = 1 j = 1 2 3 • aj^(l,£) a^^Cl,^) • 2 1 2 3 • aj^(2,£) aj2(2,£) • • 3 1 2 3 aj^(3,£) ^ j 2 ^ *^ ' ^ ^ aj3(3,£) • • • • • • • • • • • • • •

PAGE 97

87 Once the participation ratios have been determined, constraint relations are formulated. Two constraints result and are summarized as follows: I I a., (i,£) < 1 , (3.19) I j G.jaj^(i,£) < J^^ , (3.20) where a.^(i,£) is the participation rate of graduates from program j at institution i in occupation a in county k, G. . is the number of anticipated graduates from program j at institution i for future time t, and Jj^^ is the anticipated need for public postsecondary education graduates for occupation I in county k for future time t. Equation 3.19 expresses a conservation of graduates relation. The participation rates describe what fraction of graduates from a particular program at an institution find employment in a particular occupation and where that employment is. Equation 3.19 states that the sum of all participation rates for graduates from a particular program at an institution must be less than or equal to one. This allows for the fact that program graduates from postsecondary education institutions continue going to school, leave the state, or fail to find employment after the lapse period in any one of the 979 defined occupational groups. Equation 3.20 is the manpower requirements constraint relation. Simply stated, it mathematically expresses the

PAGE 98

88 policy that the postsecondary education system will admit students in a particular program only to the point where there exists a reasonable probability of employment resulting from that program after program completion. Since most postsecondary education institutions have not regularly conducted follow-up studies of graduates, obtaining existing data for the submodel constraint relations will be severely handicapped until the situation changes. Ideally, a simple questionnaire sent to a sample of graduates one year after graduation would enable the calculation of participation rates. Approximations of the participation rates could be derived from data already existing at institution alumni offices as a starting point. The Objective Function The objective function for the model reflects the philosophy that the goal of a state's postsecondary education system is to provide the best possible education to the greatest number of citizens. Constraints developed in the three submodels of this chapter function as confining systemwide enrollment by the availability of resources, social demands for educational services, and manpower requirements for an educated and trained work force . Expressed mathematically, the objective function is max Z = I I P. , (3.21) i j

PAGE 99

89 where P. . is the number of FTE students enrolled ""•^ at institution i in program j for future t ime t . Maximizing system-wide enrollment may at first appear to be a peculiar choice for the objective function. For example, writing the model in a form where the objective function minimizes system expenditures subject to minimal standard enrollment, space use, student demand, and manpower requirements conditions may appear more appropriate. The same end, however, is achieved in the model as derived. An objective function of the form used also facilitates the analysis of the ideal situation where exactly the right amount of money, facilities, and students exist for each program and each institution to meet state manpower requirements in an optimal manner. Summary The model developed in this chapter used an objective function that maximized system-wide enrollment for some future time. Three submodels were developed to facilitate the development of relations that were used to constrain institution enrollments. The first submodel involved the determination of the probable student demand. A number of policy actions taken at either the state or institution level were mathematically related to a probable student demand calculation as a means for controlling program enrollments. The second submodel described how occupation demand

PAGE 100

90 forecasts were obtained and integrated into the model. The last submodel related occupations and educational programs through the development of a student flow matrix. The manpower requirements for graduates of particular educational programs formed constraint relations. A summary of the model variables, defining equations, constraint equations, and objective function is given in Appendix A. The model developed was applied to the analysis of the registered nurse programs in Florida's community college system using three registered nurse manpower requirements scenarios in the next chapter. As this was only a partial implementation of what would be required for meaningful results, the exercise was in reality a demonstration of the model procedures and feasibility.

PAGE 101

CHAPTER IV A LIMITED APPLICATION OF THE MODEL Introduction The model developed in Chapter III was applied in a limited fashion to the 21 public community colleges in Florida that offered complete registered nurse (RN) programs in 1980. The emphasis in this chapter was on implementation and analysis procedures which served to demonstrate how the model could be used as a planning tool. Since this was only a partial implementation of the model, a number of the defining and constraint equations developed in the previous chapter were necessarily modified. Strategies developed in response to various 1985 RN occupation demand scenarios that could be directly applied to the Florida community college system in 1980 were not the primary interest and should be viewed accordingly. An initial step required in setting up a linear programming model involves determining the form of the constraint equations and estimating the constraint and objective function parameter values. Recall that the standard form of a linear program from Chapter II is Maximize Z = cx , (2.1) subject to 91

PAGE 102

92 Ax < b , (2.2) and X ^ o , (2.3) where the x matrix is comprised of the model decision variables and the c, A, and b matrices contain the model parameters, In the notation developed in Chapter III, the P^^'s are the decision variables and comprise the x matrix. The A matrix contains the parameters c. ., a. ., and r. .a., (i,£) and the b matrix contains the F^, ^ , ^ , and J^^ parameters, When applied to the 21 Florida community colleges that offered complete RN programs in 1980, the model in standard form reduced to 21 Maximize Z = T P. , (4.1) 1 12303010 subject to P. < F. , (4.2) 1 12303010 — 112303010 P. c < C. , (4.3) 1 12303010 1 12303010 — ll2303010 P. a < A , (4.4) 1 12303010 1 12303010 " ll2303010 ^ a , (i, 075) < J, , -•_H 1 12 3 0 3 0 10 1 12 3 0 3 0 1 0 12 3 0 3 0 1 0 K — K 075 ' 21 I i=l (4.5) and P. > 0 . (4.6) 1 12303010 — Here, j = 12303010 was the community college Information Classification Structure (ICS) number for RN programs offered in Florida community colleges in 1980 (see Florida Department

PAGE 103

93 of Education, Division of Community Colleges, 1977a, pp. 2.12.6). The DOT ( Dictionary of Occupational Titles , U. S. Department of Labor, 1977) number used for the RN occupation was £ = 075 (p. 56). The first section of the chapter deals with how all parameter values in equations 4.2 through 4.5, except the C. parameters in equation 4.3 were derived. The 112303010^ ^ C. parameters were derived to respond to three alter112303010'^ native future RN occupation demand scenarios. Three RN occupation demand scenarios were constructed from RN manpower requirements projections for Florida obtained by a model developed for the Division of Nursing by Vector Research, Incorporated (U. S. Department of Health, Education, and Welfare, 1978). The model considered three major anticipated changes in the health care system: the introduction of national health insurance [NHI], the increased enrollment in health maintenance organizations [HMO] , and the reformulation of nursing roles. (p. iii) The three scenarios were incorporated into the model through the parameters in equation 4.5. In the second section of the chapter, effects were investigated on how well the anticipated 1985 RN social and economic demands are met by the Florida community college system. Funding strategies were represented in the model by the ^2303010 Parameters in equation 4.3.

PAGE 104

94 Parameter Determination The Supply of Graduates Submodel The first step in forecasting the supply of RN program graduates in 1985 from Florida community colleges involved the estimation of the probable student demand for RN programs in 1985 by institution. As described in Chapter III, equation 3.1 provides the total anticipated demand for educational services from a particular institution at some future time. It does not function to the program level. Approximations of RN program probable student demands were obtained by taking the 1978-79 RN program enrollments from the AA-1 reports for all the community colleges (Florida Department of Education, Division of Community Colleges, 1980a) and multiplying the figures by a constant factor of 1.5. The constant factor, in this case, was purely an estimation based on discussions with Division of Community Colleges staff on the average system-wide RN application to enrollment ratio. The major assumption was, therefore, that the social demand for RN program enrollment in 1985 for each of the community colleges offering complete RN programs in 1980 will be 1.5 times greater than 1978-79 enrollments. It was also assumed that new RN programs would not be started in the community college system prior to 1985. Converting the headcount demand to FTE demand was accomplished through equation 3.7 with the program to headcount ratios, the g^'s, set equal to one. The assumption

PAGE 105

95 was that RN program students at the community colleges have taken and will continue through 1985 to take an average of 30 semester credit hours a year. The 12303010 ^^l^^^ used for constraint equation 4.2 for all 21 community colleges are summarized in Table 10. In calculating the parameter values of the student access control constraints described by equations 4.3 and 4.4, Division of Community Colleges and Of f ice of Facilities Construction data were used. Cost per RN FTE parameter values, or the c. parameters in equation 4.3, were 112303010^ ^ ' derived as the sum of the 1978-79 instructional and support costs per RN FTE. Instructional cost data were obtained from the 1978-79 CA-3 reports for each community college (Florida Department of Education, Division of Community Colleges, 1980c). Since the CA-3 reports provided cost information by courses, it was necessary to compile the RN major required courses for each community college from the institution's catalog. The number of credits for each course was multiplied by the CA-3 report cost per student figure. Once the total institutional course cost per student was calculated by summing the individual course instructional costs, it was necessary to average the cost on a per year basis and then normalize the cost per student to a cost per FTE. Table 6 summarizes the procedure for one community college. Once the instructional costs per FTE were calculated, physical plant, institutional, student services, and academic

PAGE 106

96 Table 6 RN Program Total Instructional Cost/FTE For Polk Community College Rpmi ired Courses Credits Total Cost/ Credit for Instruction Total Course Cost for Instruction per Student APB 2190 5 $58.21 $291.05 2191 5 56 .45 282.25 ENC 1103 3 53.79 161.37 HUM. ELECTIVE 3 43.68 146.04 MCB 2013 5 54.11 270.55 MTB 1320 1 50.00 50.00 NUR 1211 10 74.25 742.50 1220 5 79.44 397.20 1620 1 63.75 63.75 2111 5 70.52 352.60 2120 5 54.71 273.55 2231 11 78.13 859.43 2622 1 67.45 67.45 NUU 1140 9 90.97 818.73 P.E. ELECTIVE 2 50.00 100.00 POS 1041 3 47. 79 143.37 PSY 2000 3 53.58 160.74 SOC 1000 3 47.37 142.11 TOTAL 80 $5322 Average Instructional Cost per FTE per Year $5322 ^ 60 _ ~2~~ ^ ~80~ " ^^^^^

PAGE 107

97 support costs per FTE were also added to obtain the c. 1 12303010 parameter values for each community college. Data for the support costs were obtained from the Report for Florida Community Colleges 1978-79 (Florida Department of Education, Division of Community Colleges, 1980d, pp. 84-88). An example of how the c. parameter for a single community 112303010^ ^ ^ college was arrived at after adding support costs to the instructional cost is presented in Table 7. The assumption was made that the relative costs of different RN programs remained fixed through 1985. To estimate the assignable square feet of space needed per RN FTE for each community college, the a. para' 1 12 3030 10 ^ meters in equation 4.4, two assumptions were made: (1) the RN FTE space requirements are the same system-wide, and (2) that the space requirement figures remain unchanged from 1980 through 1985. The occupational lab space factor derived by the Office of Facilities Construction for estimating facilities needs was used. The effect of the standard was to provide approximately 46 assignable square feet per occupational FTE (Florida Department of Education, Office of Facilities Construction, 1978, p. 2). The definition of annualized FTE used in the study has been 30 semester credit hours taken in one year, or a semester average of 15 hours. Dividing by five, the estimate of the number of contact hours per day was determined to be three. In one day, therefore, approximately 3 x 46 or 138 assignable square feet are consumed per RN FTE.

PAGE 108

98 Table 7 Total Cost/FTE for the RN Program at Polk Community College ^17 12 30 30 10 Instructional Cost/FTE $1996 Physical Plant Cost/FTE $183 Institutional Support Cost/FTE $270 Student Services Cost/FTE $158 Academic Support Cost/FTE $123 Total = c 17 12303010 = $2730/FTE

PAGE 109

99 The A. parameter values for equation 4.4 were 112 303010^ ^ arrived at using 1978-79 RF-1 reports from each community college (Florida Department of Education, Division of Community Colleges, 1980e), a unit space analysis computer run on the current room inventory by the Division of Community Colleges, and in some cases community college facilities personnel were contacted directly. The gross RN space figures obtained were multiplied by seven to provide an estimation of the total RN assignable square feet of space available in a day. The assximptions were that RN facilities have been and will continue to be utilized at most seven hours per day through 1985 and that the RN space available in 1979 will remain fixed in terms of assignable square feet of space. Using the 1978-79 RF-1 report for Polk Community College produced the figures listed in Table 8. The third step in the supply of graduates submodel was to estimate the number of graduates. To accomplish this, the r^ 12303010 ^^^ios in equation 3.15 were calculated using AA-1 reports for each community for each year from 1975-76 through 1978-79 (Florida Department of Education, Division of Community Colleges, 1977b; 1978; 1979; 1980a). A simple weighted average ratio of RN program completions to enrollments was used for the following reasons: (1) several RN programs were not in existence long enough to yield sufficient data entries to justify using the double exponential smoothing algorithm, (2) early RN enrollment figures skew a simple mean, and (3) inaccurate data. The

PAGE 110

100 Table 8 Calculation of the Total Assignable Square Feet of Space for the RN Program at Polk Community College Assignable Square Feet of Used by the RN Program Polk Community College by Space at Room 1295 104 98 94 376 113 96 336 101 96 25 214 201 96 25 192 99 458 259 113 96 729 23 135 96 505 96 98 45 Total ASF 6,369 17 12 3 0 30 10 44,600

PAGE 111

101 last point is demonstrated by the fact that Division of Community Colleges data seldom agreed with National League for Nursing data for the same time periods (National League for Nursing, 1976; 1977; 1978; 1979). A demonstration of the ratio of the number of RN graduates to FTE calculation for one community college is displayed in Table 9. The assumption here was that the ratios as determined represent the 1985 conditions. Table 9 Ratio of RN Graduates to RN FTE for Polk Community College Year (Y ) m ^1 1975-76 ^2 1976-77 ^3 1977-78 ^4 1978-79 Completions 120 114 89 81 Enrollments 306 418 364 294 Completions , Enrollments m . 39 .27 .24 .28 Weights (W ) ^ m 1 2 3 4 4 y u w L mm J. _ nT=l 17 1 2 3 0 3 0 1 0 -j^Q .28 A summary of all the parameters used in the supply of graduates submodel except the C. parameters is 1 12303010 ^ presented in Table 10. The C^^^^^^^^^ parameters are specified in a later section and are used as the primary elements in

PAGE 112

102 TJ O e w w * (D C V O ni -H ;3 -P T3 d ISI O -H H a ;h oS (D «H s rH O S X! :3 d >>!/! a ^( a CD 3 -p CO (D s fciD C •H +-> CO O 0) o fin -p •H o; 3 (D o o ooooooooooooooooooooo ooooooooooooooooooooo CO O T-l O 00 ^ CD CSI CO (M (M O 05 iH O H iHt>t>OlOt>'^0 (MOOCOCMCNCD^'^' CD 1> Oi H H 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 CO CO CO CO CO CO CO 0-5 00 CO CO 00 CO CO CO 00 CO 00 CO CO 00 tH iH iH rH tH rH rH rH H H rH rH tH rH rH rH rH rH rH rH rH o o O o o o o o o o o o o o o o o o o o o lO 00 00 LO o l> H o in C5 o o CO 00 00 est (M CD o 00 00 rH o 00 l> o (M CO rH o o l> rH 05 00 o CO CO CO (M oo CO 00 CO CSI 00 00 00 00 CO 00 00 CM CM CO CD O rH 00 t> 00 UO 00 CO CO CD 00 CD CO CM CO C55 00 l> CD I> 00 CM 00 CM CO CM 00 O CM o CO in rH CM m CO 00 CM rH CJl 05 CD t> o o ^ ^ CM 00 rH in 00 o CM CD 05 rH CD rH in in 00 05 00 rH t> CM CM rH in rH rH rH rH H rH 05 CO l> O CD 0> 00 05 t> C£5 C00 05O cMoO'*incDi>ooa5o rHrHrHrHrHrHrHrHCM t> 00 rH CM

PAGE 113

103 the construction of system strategies in response to the various RN occupation demand scenarios described in the next section. The Occupation Demand Submodel Three occupation demand future scenarios were derived for Florida registered nurse needs for 1985 from existing federal registered nurse manpower requirements forecasts. The three scenarios were extracted from the study, The Impact of Health System Changes on States' Requirements for Registered Nurses in 1985 (U. S. Department of Health, Education, and Welfare, 1978). The first scenario was the baseline scenario, defined a conservative extrapolation of the present trends in health care to the 1985 time frame. For the most part, the configuration of the future health care system is assumed to be an extension of the recent past. For this reason no NHI [national health insurance] program is instituted, no HMOs [health maintenance organizations] are added, and limited reformulation in nursing roles is preserved under the baseline scenario. Thus, the baseline can be considered a very conservative, albeit possible, projection of the future requirements for RNs [registered nurses]. The degree to which it becomes a realistic projection is dependent on the degree to which the system changes investigated here occur, or the extent of other changes not foreseen or treated in this analysis. (U. S. DREW, 1978, pp. 3-4) The baseline scenario functioned as the gauge by which results generated in response to other scenario conditions were measured. The fact that the baseline scenario projected registered nurse (RN) needs for the future based on the

PAGE 114

104 conservative assxixnption that the leading factors that have determined RN manpower requirements allowed for the assessment of the effects of deviations from this assiimption on future RN manpower requirements. The major assumptions inherent in the baseline scenario follow: 1. A national health insurance program will not occur prior to 1985. 2. The health maintenance organizations in existence in 1974 will continue to grow at rates derived from historical data. 3. There will be no further acceptance of nursing role reformulation from the estimated 1972 level (HEW, 1978, pp. 3-4). The second RN occupation demand future scenario was the national health insurance (NHI) scenario. The NHI scenario models the situation in which the uninsured portion of the population receives insurance benefits equivalent to those enjoyed by the insured population. This is intended to represent those NHI bills which guarantee some degree of protection to everyone and thereby plug the gap in current public and private insurance coverage. (U. S. DHEW , 1978, p. 11) The NHI scenario was different from the baseline scenario only in terms of NHI parameters. The principal assumptions were: 1. The effective implementation year for national health insurance is 1979, and the total impact of NHI is expected to occur within two years, by 1981. 2. As uncovered individuals become covered, they consume health services at the same rate as those in their respective cohorts who previously had health insurance (U. S. DHEW, 1978, pp. 11-12) .

PAGE 115

105 The third RN occupation demand future scenario was the combination (COMB) scenario. The COMB scenario consists of the NHI . . . scenario with moderate HMO [health maintenance organization] growth and moderate acceptance of role reformulation concepts. (U. S. DHEW, 1978, p. 22) The major assumptions different from the NHI scenario follow : 1. The rate of increase in the number of new HMO's per quarter will be identical to the average rate experienced between January 1974 and July 1975. 2. The percent of inpatient units in the country utilizing primary nursing grows linearly to 40 percent in 1985. 3. The number of clinical nurse specialists that would be in new roles by 1985 reaches 75 percent. 4. Demand for physicians' services in office visits satisfied by nurses in practitioner roles grows linearly to 65 percent by 1985 (U. S. DHEW, 1978, pp. 15-19). In the 1978 HEW report, 1980 and 1985 Florida RN requirements for each scenario were projected RN manpower needs for full-time workers. The full-time figures were converted to headcount requirements utilizing a ratio reported in a latter HEW publication (U. S. DHEW, 1979, pp. 152, 158). Table 11 summarizes the scenario RN requirements for Florida in 1980 and 1985. Table 11 Projected RN Requirements for 1980 and 1985 in Florida Year Baseline Scenario NHI Scenario COMB Scenario 1980 39,963 43,345 44,866 1985 49,662 53,848 56 ,995

PAGE 116

106 In the next step, equations 3.16 and 3.17 were used to break down the state requirements into county units. For the baseline scenario, equation 3.16 is k=l which represents the average number of total additional RN's needed each year in the five-year period from 1980 to 1985 in Florida. The annual net attrition rate value, d^„^, of 4.7% for RN's was taken from a Florida Board of u /o Regents' manpower study (1968, p. 12). Here, 67 is the number of Florida counties and 075 is the Dictionary of Occupational Titles , DOT, number for the registered nurse occupation. Medium midyear population projections by county were obtained from the Florida Statistical Abstract 79 (1979, pp. 11-16) and used in equation 3.17. Since a number of counties in Florida had relatively small populations and thus small RN yearly requirements, the number of counties specifically considered in the analysis was reduced from 67 to 28. Requirements from equation 3.16 for the 39 counties not specifically considered were summed and then evenly distributed to the 28 that were. For the baseline scenario, the total additional RN manpower requirements for 1985 by county k = 1,2, 3,..., 28 are summarized in Table 12. By a similar procedure, manpower requirements under the NHI and COMB scenarios were derived and are also listed in Table 12. A county key is supplied in Appendix B.

PAGE 117

107 Table 12 1985 State RN Requirements COMB O \^ O il d. X _L W County (k) 075 075 0 075 1 75 81 90 9 6R o •J 119 A 626 ^ v_/ c; O 6 50 55 60 7 P7 •^2 Q O o O I? ^ O Q 17*^ J# o 1 R8 -LOO 907 oo ins ±.\j %j 11 268 291 319 1 2 47 ^^2 o^ 1 3 \J J. \J\J 79 1 4 1 "=17 1 5 O J. 16 83 90 99 1 7 72 / o SIR 1 8 oo 19 61 66 72 20 191 207 227 21 317 344 378 22 124 134 148 23 350 379 417 24 130 141 155 25 59 64 71 26 119 128 141 27 115 125 137 28 124 134 147

PAGE 118

108 The anticipated need for public postsecondary education RN program graduates in 1985 were calculated from the data in Table 12 and by considering in-migration and private postsecondary education contributions to the RN work force in the form of equation 3.18. The in-migration factor, m^rj^, was approximated using 1970 U. S. census data (U. S. Department of Commerce, 1973, pp. 69, 212). The number of in-migrants from 1965 to 1970 from different states and abroad to Florida was summed and then divided by the number of years over which the migration occurred. It was assumed the in-migrant population age distribution was similar to Florida's as a whole. The proportion of over 65-year-olds and under 20-year-olds was discounted, which left an approximate figure of how many able workers were available for employment. This figure was then simply divided by the number of DOT occupational groups to yield the average approximate niimber of RN's that in-migrated to Florida in any year in the period from 1965 to 1970. The quantity obtained was used as the total niimber of in-migrating RN's in 1985 to the state of Florida. It was further assumed that in-migrating RN's distributed evenly throughout the state. A figure of 5 was obtained for m^r^^. The fraction of new 1985 RN positions expected to be filled by 1985 public community college RN program graduates, Hq^^, was derived utilizing National League for Nursing data from 1975 through 1979. To the nearest 10%, a mean figure of 70% was obtained for n„„_ (National League

PAGE 119

109 for Nursing, 1975, 1976, 1977, 1978, 1979). Table 13 summarizes the 1985 anticipated need for public community college RN program graduates for the RN occupation under three different scenario conditions. The Student Flow Submodel The parameters estimated in this section are the RN participation rates, the a,,,„ „,„ , (i,075) parameters, ^ ^ ' 12 303010k in equation 4.5. The 1978-79 AA-2 placements reports for each community college were surveyed and found inadequate for determining participation rates due to lack of detail concerning the occupations held by program graduates (Florida Department of Education, Division of Community Colleges, 1980b). The participation rates displayed in Table 14 were approximated by intuitive integration of past RN enrollments, county RN needs as a function of population, and the proximity of place of employment with community college. The emphasis was on accounting for the primary RN graduate flows. It was assumed that one year after graduation, approximately 80% of the RN program graduates from each community college find employment in the RN occupation in Florida and that this rate is constant over the planning period. Summary All parameters in equations 4.2 through 4.5 have been estimated except the C. parameters. A number 1 12303010 of simplifying assumptions were necessary due to the limited nature of the application example and to difficulties with

PAGE 120

110 Table 13 1985 Requirements for Community College RN Program Graduates Baseline NHI COMB Scenario Scenario Scenario County (k) \ 075 \ 075 075 1 48 52 58 2 35 38 43 3 67 73 81 4 263 394 433 5 30 33 36 6 30 33 37 7 14 16 18 8 376 408 488 9 117 127 140 10 56 62 68 11 183 198 219 12 25 28 31 13 38 41 46 14 87 95 105 15 47 52 57 16 53 58 64 17 45 49 55 18 16 18 20 1 Q 41 46 20 129 140 154 21 217 235 259 22 82 89 98 23 240 261 287 24 86 94 104 25 37 40 45 26 78 85 94 27 76 83 91 28 82 89 98

PAGE 121

Ill Table 14 Student Flow Matrix for RN Program Graduates* ** Comty (k) 1 2 3 4 5 Ccranunity College (i) 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 1 .4 2 .6 3 .6 .1 .1 • O .1 5 .1 6 .2 7 .3 8 .1 .1 .6 .1 .1 q .6 .1 • J• J. 10 .6 11 .1 .6 .1 .1 12 13 .2 14 a 15 1 .6 16 .2 17 .3 18 6 1 Q .1 .1 20 .1 .6 21 .1 .6 22 .6 23 .1 .1 .6 24 .1 .6 25 .3 26 .4 27 .1 .3 28 .1 .4 *A11 non-specified matrix elements have a value of 0.004. **Appendix B has community college and county keys.

PAGE 122

112 data. In the next section, total cost parameters were designated as the primary model parameters used to depict alternative system strategies that respond to the three different RN occupation demand scenarios. With all other parameters held constant, the varying of the RN manpower requirements parameters constitutes a sensitivity analysis of the optimal distribution of RN program FTE ' s throughout Florida's community college system in 1985. Results Description For the purposes of demonstrating how the model may be used as a planning tool for postsecondary education, six possible 1985 Florida community college system futures were considered dealing specifically with RN programs. The six futures were divided evenly into two broad categories. Three Series I futures represented situations where RN program costs, enrollments, and facilities parameters were determined solely by the anticipated 1985 RN manpower requirements scenarios. Series I futures were determined by normative futures forecasting techniques and were defined as the most desired, or ideal system future states. Three Series II futures integrated anticipated RN enrollments and facilities parameters, determined by explorative futures forecasting trend analysis methods, with anticipated 1985 RN manpower requirements scenarios. Series II futures

PAGE 123

113 represented simple projections of 1980 trends into the future . In this section, analyses of the baseline, NHI , and COMB 1985 RN manpower requirements scenarios within Series I and Series II Florida community college RN program scenarios are presented. The community college scenarios developed included community college RN program enrollment, facilities, and expenditure dimensions. Since Series I enrollment, facilities, and expenditure parameter values characterized the most desired system future states, strategies were developed for each community college RN program that specify what modifications are needed to increase the probability that a Series I future will occur and decrease the probability that the corresponding Series II future will occur. In Appendix C, the Series II baseline scenario future is provided for reference. Enrollment Analysis The analysis of the anticipated 1985 RN program enrollments at Florida's community colleges consisted of comparing Series I and Series II enrollments with each other and with the probable student demand figures derived in the previous section. Series I enrollments were obtained by letting the parameters in equations 4.2, 4.3, and 4.4 corresponding to the b matrix elements in equation 2.2 be free variables, thus allowing the remaining model constraint determine their values. Series II enrollments were obtained

PAGE 124

114 by using the parameter values determined in the previous section, letting only the total cost parameters, the C. 1 12 30 30 1 be free variables. In Table 15, the results of Series I and Series II conditions on enrollments are depicted. Comparing Series I enrollments with the anticipated demand, only the anticipated demand figure for community college 17 occurred between its Series I enrollment bounds. The anticipated student demand at school 17 therefore coincides with what will be needed. Approximately 38% of the schools have ideal Series I enrollments less than their anticipated demands while at 57% of the schools, the demand is less than the enrollment need. The most severe differences between anticipated student demand and Series I enrollments occur at schools 2, 6, 7, 11, and 15. The Series I enrollment needs for schools 2 and 15 range from 2.1 to 4.3 times the anticipated student demand. This means that unless the schools take measures to change the anticipated student demand for RN programs in 1985, serious discrepancies will exist between the number of students wanting to enter RN programs at these schools and the number of students needed in the programs to meet area manpower requirements. At schools 6, 7, and 11, more students are anticipated to want to enter RN programs in 1985 than are needed to fill area manpower requirements. The anticipated student demand for the schools range from 1.9 to 10.7 times the Series I enrollment needs. Since

PAGE 125

115 c s t> CO O CXi o O O lO 00 CM CO lO 05 CO in CO i> 05 CO i> 00 rH o CO CM in (M m CD 00 O O C:5 rH lO I> CO CM 05 H 05 H in rH CO tH CSI l-H rj< rH CD rH ^ 00 ^ rH CNj IT* ^ rji Cjj 0 U CO c w l-H rH CO O 00 O CD O lO 00 CM CO CO UJ Cjj CO G) CO I> CD o o csi m lO rH 00 O O t> rH lO t> CM CO CM 0^ CO tH CO CM CO i-l CM t-l rH CO rH ^ 00 ^ rH CM CO in ^ IS CO 1— 1 l-H tu •tH CO O 00 o CM CM lO 00 00 CO CO 05 00 CO CO l> rH CO H rH t> CM in csi lO 00 O O O CM lO rH in t> rH CM CJ5 C35 05 rH 00 0) rH CO iH CM iH CO rH CO rH 00 Ttl rH CM CO in 0 w CO i-Q +J CQ l-H I> rH rH lO O CO ^ ^ CD CM O CO C<1 00 o 00 in in CO rH (— ' 00 CM CM CXj CJ ^1 vi ~ vi; ^ CM CM CO ^ Cn t/^ on t^J UJ t'J [> [>• LQ 1— 1 o rH rH H H H C/3 r— 1 H >> 0 1 — 1 u d c C O rH CO t> CO t> 00 O rH CD -P r \^ T-i UN VipJ a-H \r\ I Uj ~ ~ uj rH CM rH C^ 00 rH O CM CO CM O C rH rH CM lO CO CO rH 00 rH CD O 05 rfl CM rH -vf* rH in O 0) * r> s rH rH rH O rH rH iH O f-l a Ul w lO ^ O CO O l> CM 00 o 00 00 00 CO CO 00 rH LO CO CO 05 CO CD CO 00 05 O rH I> rH CD ^ (35 t> CO T}< ^ 00 CO CO O rH CD CSl CO CO rH I> .H in O 00 rH CO O •<^< rH rf< 05 lO CX3 rH H CO 05 CQ tH T3 Q) +-> w -a an a O CO UO rH 05 CJ5 CO LO 00 05 rH CO in C:5 CO iH rH [> O CO l> rH •H fin rH CM in o o O ^ O ^ t> rH in I> CM CM 05 CO 05 H O o S rn UU rH CM CO T — 1 rH 00 '=7' rH in ^ CO ^ in H S 0) +-> « Q 00 •H CD C bJ3 CO 00 05 o H c 00 05 O H rH rH H rH rH rH H rH rH H H rH CM CM 0 Ol +-> 0 a u H

PAGE 126

116 Series I enrollment needs at schools 7 and 11 are very low, the possibility of phasing RN programs out of these schools must be a consideration. Series II enrollments were limited in 13 schools by the anticipated student demand. Twelve of these schools corresponded to Series I schools that needed to act on increasing their 1985 anticipated student demand. Only five schools had a Series II enrollment range that overlapped their Series I enrollment range. If the trend extrapolation enrollments represented by Series II figures persisted in 1985, the anticipated RN student demand will not be met. System totals show the average Series I enrollment need to be 18% greater than the anticipated student demand. The Series II enrollment need, however, is 20% less than the anticipated student demand. In sum, 38% fewer RN FTE's are expected by trend extrapolation in 1985 than are expected to be needed in Florida's community college system to meet the state's RN manpower requirements. Space Analysis The total 1985 anticipated RN program space for a seven hour day for each community college, you may recall, was taken to be equal the RN program space provided in the 1978-79 academic year. Comparisons of the Series I and Series II space needs with the anticipated available RN space were made from Table 16 data. The anticipated available spaces at only community colleges 12 and 14 were within their Series I ranges. Eleven

PAGE 127

117 o o •H CO X >> iH -P oj 0) C 0) I— I <5 (D (D CO Cl aj oJ CO 0 E ^ W as 0) Oj (U ;h T-H Eh -H bDX3 CD oS O. u . . CO a, bci •H en 00 05 •H CQ O O O CO u CD CO O a a CO z en 0) 00 0) > G bC ;3 CD E rH o o O CJ 1> CD [> lO O CSl ^ O H O 00 ^ C O H O CM t> 00 CD t> H lO rH CO H CO CO O ^ CM lO CO 'S* O CO t> CO <^< rH 00 rH in [> CO t> in t> O O rH O IM CO C 00 CO t>> rH in rH CO rH o CO o CO rj^ rH l> rH in ^ CD i> in i> O rH O CM CM CO CQ i> o in 05 1> CM O CO t> in rH rH 00 rH Tj* O rH O t> t> 00 CD in rH CD rH CO o 00 in 05 O t> CO I> CM Tt< rH CO H in 00 05 CM CM (35 in 00 o in CM CD 00 in CO CO CO t> rH O CO 03 O t> CD CM 00 rH CD CM in T}< CO t> rH CO 05 in t>00 rH CM CM CO CO 05 00 rH O O 00 05 o t> cq t> O t> rH 05 CO 00 CO 05 CM O CQ in 00 i> in 00 in t> cq in 05 00 in in rH rH H 00 rH in CO CQ in CO 00 CO 00 ^ C<1 CD 05 O CD C35 rH C35 05 CO in in c] rH CQ in in CO CQ t> CO rH CD rH ^ t> t> CO CM 05 00 CD ^ 00 O CD in 05 o in 00 05 CO in rH O rH Cq !> 00 (35 05 CQ 00 O 00 H t> th CO CM in 00 00 CQ CO I> rH CO rH O CD CM in ^ CO CM CD t> in t> CD CO O rH O CO CM 00 CQ I> rH in CM CD CO H O 05 CQ rH rH O rH CO CO in CQ CO rH o i> o in CM 00 00 CM CM 00 CO 05 CD 05 00 1> O CD 1> CJ5 CD rH rH T}< rH CM 00 ^ in CO t> 00 05 o CM CO in rH rH rH rH CD 00 05 O rH rH rH rH rH CQ CM

PAGE 128

118 schools needed more space than was anticipated to be available while 8 had Series I space needs less than their anticipated space. Of these last 19 schools, 6 had Series I space needs within 13% of their anticipated available space. The most severe Series I space shortages occurred at community colleges 2, 3, 4, 5, 8, and 15. Individual space increases of between 28% and 166% existed above the anticipated space levels. Space surpluses ranging from 37% to 90% of the anticipated space occurred at schools 1, 6, 7, 11, 16, 19, and 21. In a full scale model application, space surpluses occurring in one program area would be consumed by programs showing space shortages. A Series II space needs comparison with the anticipated available space showed 67% of the community colleges needed less than their anticipated space supply. Fourteen schools did not need all the space they had available provided Series II conditions prevailed in 1985. System totals show that the 1985 average total systemwide Series I space need for RN programs in Florida's community colleges equaled the total anticipated RN space available. The problem was, of course, redistributing this total anticipated space in the pattern of Series I space needs. Total Series II space need was 32% less than the anticipated RN space for 1985. If Series II conditions are maintained, one third of the system's RN anticipated space, therefore, would not be needed in 1985.

PAGE 129

119 Cost Analysis The total program cost parameters for Series I and Series II for each community college, you may recall, were allowed to be free variables. In the Series I situation, the RN program costs listed in Table 17 depict 1985 dollar amounts necessary to insure scenario manpower requirements for community college trained RN's are met. Series II program costs represented the ideal dollar amounts needed if space, student demand, and manpower requirements constraints were specified. Only in the situation of program quality improvement could an expenditure beyond the Series I or Series II recommendation be justified. Since the costs listed are in 1979 dollars, the figures themselves are meaningless as forecasts of the actual 1985 program costs for each institution. Only relative differences between community colleges within each series situation or between series, therefore, were investigated. A wide range of Series I program costs for community colleges constrained by three different RN manpower requirements scenarios existed. The maximum COMB program cost was greater than 90 times the minimum baseline program cost. The five highest cost programs under the baseline scenario in descending cost magnitude are at schools 13, 18, 9, 14, and 2. If the COMB scenario manifests itself, the order of the schools with the highest program costs changes to 2, 18, 13, 9, and 14. School 2 shows an estimated 79% increase in the Series I baseline to COMB scenario

PAGE 130

120 m •H ca >> H oS G O 1— ( <; o rH +J -o 03 X G 0 O M o h1 ^ rH bC 0 •H 0 T3 Oi rH lO 00 0) rH 05 CO O 'Ji lO 00 C oq CO o o o CO lo CD lO cq CO aa O 05 CM O CSJ CO in £> ^ CO CD rH 00 05 lO CM '*! 00 CD rH t> CD CO O in CO 05 O CO CD CO CD 05 00 CO CM t> LO LO o M o rH ^ CO LO CO CO CO CO UN CO 00 lO CO t> 00 lO 00 CSl LO CD 0 rH i~l II CM rH rH oq OI e Qi }< bD O to CO o m (Y\ PCI t> rH cq oq o o o CO lO CO LO cq o 1— 1 CO 05 oq o o5 CO 05 H 00 05 CD CM H CO rH LO in CO CO lO o CO 05 rH CO CD CO CO H 00 CO rH t> lO 2; CO rH CO lO CO CO CO CO CO cq CO 00 lO CO t> 00 LO CD CM LO o D5 rH rH CM rH CQ rH rH CO oq 1— 1 h-< M 0 G CD CO o in I> tH cm l> CO 00 CO oq o o O CO LO CD lO CO H •H •H Tj( (35 IM O CS] lO O Q> 0 W M lO r-i rp \n CO CO rH CO 00 LO CO CD 00 lO CM lO in rH CO r-\ C lO O t> rH CQ 05 MH rr^ i— i cn \\J J lO I> Li J (,\J l> CO 05 rH /^i ff\ t^m^ /T\ ty^ CM CJU L^ U3 ^ J Cj5 o CO 00 i> LO 00 lO CO l> 00 rH l> CJ5 rH CM CO 05 O CO o o «^ *\ CO 05 CO CO CO rH LO CO CO rH 00 oq lO t>CO 05 LO LO rH CM +-> ^ rH rH CO cq CO CO H rH CO H CO CO CO u E Oj U bH 00 05 rH O O CO 00 CD O CO H [> rH CO I> t> 00 O CO ^ LO CO O 1— 1 CSI rH (M (M 00 cNj in 00 00 O CO 00 00 o 00 in 00 tH 00 CO rfi CO (M CO CD 00 00 o CD O 00 CD 'S* in 00 CS3 CO CO *l *t *t «S LO O 00 lO (M rH cq CO 05 '53< 05 CO rH lO ^ C35 ^ rH rH CO H CO CO rH rH CO rH rH CO PC CO 1— ( CO •H 0 G (M CO CO rH lO m 05 H CO rH l> i> i> lo oq CO O CO CO CO t> rH •H ^ n 00 CO rH in I> 00 o 00 00 CD 00 LO rH C55 rH 05 rH LO O rH O Cvl rH CO in <^ LO cq O ^ •rf< 05 rH O CO CM 05 00 0 #t #S VI *t #t w in ^ 00 ^ iH iH rH O CO 00 Tj< i> CO LO o CO CO ^ i> CSI H CO H CO CO rH fH CO rH rH G5 m UN >» +j •H 0 G bC CQ 13 0 rH rH rH CM CO ^ in CO t> 00 O) o H cq CO rt< lO CD O 00 05 O H d rH rH rH r-{ r-i r-i r-i rH rH H rH CM CM -p O O 0 o o

PAGE 131

121 transition while the increases for the other four schools range from 12/„ to 19%. The five community colleges with the smallest Series I program costs under the baseline scenario in the order of increasing costs were schools, 11, 7, 10, 20, and 8. Under the COMB scenario, the same schools in the same order had the smallest program costs. The Series II program cost range was narrower than the Series I. The maximum COMB scenario program cost was 18 times the minimum baseline cost, a factor of five less than the Series I range. Greater stability existed among the highest cost programs in the Series II situation between the baseline and COMB scenarios. The six highest cost baseline scenario programs in the order of decreasing costs were at schools 13, 18, 9, 21, 14, and 19. Under the COMB scenario, the order changed to 13, 18, 9, 19, 21, and 14. The five Series II schools with minimum program costs under the baseline scenario in the order of increasing costs were at community colleges 11, 20, 7, 10, and 8. COMB scenario schools with the smallest program costs were 11, 20, 7, 8, and 15. Comparing Series I program costs with Series II costs revealed that only community colleges 1, 6, 16, and 19 had Series I program cost ranges that overlapped their corresponding Series II cost ranges. Schools 7, 10, and 11 had Series I cost ranges below their Series II ranges. The

PAGE 132

122 remaining 14 schools had Series I cost ranges above their corresponding Series II ranges. The five schools with the greatest discrepancies between their Series I and Series II program cost needs were 15, 2, 5, 4, and 14. The average costs for school 15 for the two series situations differed by a factor of four. In sum, wholly two-thirds of the community colleges needed to make efforts to increase RN program funds in order to move from the highly probably 1985 trend determined Series II situation toward the ideal Series I situation . System-wide, the total Series I program cost increased 24% from the baseline to the COMB 1985 RN manpower requirements scenario. For Series II, the increase was only 5%. The average total Series II system program cost was just 67% of the system average needed for the ideal Series I situation. In conclusion, the Series II cost projections at the system-level indicated that a serious under-funding of RN programs in 1985 would exist and that the more liberal RN manpower requirements scenarios considered produced even more pronounced discrepancies. Requirements Analysis The baseline, NHI , and COMB 1985 RN manpower requirements scenarios depicted in Table 12 indicate the ideal number of community college RN program graduates needed by county area to meet 1985 statewide needs. In this section. Series II RN program graduates anticipated from Florida's

PAGE 133

123 community colleges in 1985 that entered the RN occupation work force by county were analyzed for each manpower requirements scenario. How closely the Series II figures approached the scenario requirements indicated to what degree the economic demands for 1985 RN program graduates were satisfied. The Series II baseline scenario RN production from the community colleges in 1985 distributed to the 28 county areas as displayed in Table 18. Eight counties had their 1985 RN manpower requirements met. They were counties 1, 3, 4, 7, 9, 10, 12, and 18. Shortages occurred in the remaining 20 counties. Less than half the needs for 1985 community college RN graduates were satisfied in counties 5, 6, 14, 22, and 27. The most severe situation in terms of the percentage of unmet manpower need existed in county 22 where 66/o of the 1985 RN need for 1985 community college RN graduates was not met. The largest number shortage occurred in county 21 where 87 additional 1985 community college RN graduates were needed. The state total for the Series I RN production satisfied 97% of the baseline requirements. This was not 100% since the student flow matrix parameters were crude approximations. The small discrepancy, however, proved to be consistent throughout the occupation demand scenarios and therefore did not adversely affect results. The baseline Series II statewide total of 1985 community college RN program graduates entering Florida's RN work force in 1985 was 1,832. The number needed was 2,594.

PAGE 134

124 Table 18 Series I and II 1985 Baseline Scenario RN Requirements Analysis Base 1 ine Series I Series II RN RN RN County Requirements Production Production 1 48 48 48 9 OO o c OO 3 67 67 67 4 263 263 263 5 30 26 14 6 30 13 12 7 14 14 14 8 376 376 297 9 117 117 117 10 56 56 56 11 183 183 141 1 9 25 25 13 38 38 23 14 87 87 42 15 47 47 33 16 53 45 33 1 7 45 37 23 18 16 16 16 19 37 23 19 20 129 129 110 21 217 217 130 22 82 82 28 23 240 240 183 24 86 86 66 25 37 25 25 26 78 78 57 27 76 68 36 28 82 82 45 Totals 2,594 2,523 1,832

PAGE 135

125 Only 71% of the demand for 1985 community college RN program graduates was satisfied under the Series II baseline scenario. Series I NHI scenario results are presented in Table 19. In the NHI scenario, the number of counties that had their 1985 RN manpower requirements met was five, three less than in the Series II baseline scenario. They were counties 1, 3, 7, 9, and 10. For counties that had RN manpower shortages, the number that had less than half their 1985 RN manpower requirements met increased from five in the baseline scenario to seven. These hard hit counties were 4, 5, 6, 14, 17, 22, and 27. The greatest percentage of unmet manpower need occurred in county 22 where 69% of the 1985 RN requirement for 1985 community college RN program graduates was not filled. The largest number shortage existed in county 4 where 249 additional 1985 community college RN program graduates were needed. The NHI scenario Series II statewide total of 1985 community college RN program graduates that entered the RN occupation in Florida in 1985 was 1,876. The number needed was 2,932. Under the Series II NHI scenario, the shortage of RN program graduates intensified from the baseline scenario. Where 71% of the demand was met in the baseline scenario, 64% was met in the NHI scenario. The Series II COMB scenario RN requirements data are summarized in Table 20. The number of counties having their 1985 RN manpower requirements met remained unchanged from the NHI scenario. They were counties, 1, 3, 7, 9, and 10.

PAGE 136

126 Table 19 Series I and II 1985 NHI Scenario RN Requirements Analysis T* 1 *^ G T o c X X t; o X ^^^T'Ti^G TT 0\=;x Xs^i^ X X nil RN RN Cmi n 1" v xX w V_l Ll O U-Lwll irX V^LILIL/ L X.Wil 1 52 52 52 2 38 38 27 3 73 73 73 4 394 394 145 5 33 27 15 6 33 15 14 7 16 16 16 8 408 408 298 9 127 127 127 10 62 62 62 11 198 198 142 12 28 28 27 13 41 41 24 14 95 95 42 15 52 52 34 16 58 49 33 17 49 40 23 18 18 18 17 19 41 26 21 20 140 140 110 21 235 235 131 22 89 89 28 23 261 261 183 24 94 94 66 25 40 28 27 26 85 85 57 27 83 74 36 28 89 89 46 Totals 2 ,932 2,854 1,876

PAGE 137

127 Table 20 Series I and II 1985 COMB Scenario RN Requirements Analysis COMB Series I Series II RN RN RN Count V Requirement s Product ion Product ion 1 1 ^ Q OO Do Oo 2 43 43 27 3 81 81 81 4 433 433 145 5 36 30 15 D O / 1 7 X < X O 7 18 18 18 8 448 448 300 9 140 140 140 10 68 68 68 XX z xy zxy 1/1/1 12 31 31 27 13 46 46 24 14 105 105 43 15 57 57 35 04 o o OO 17 55 45 23 18 20 20 17 19 46 28 22 20 154 154 110 21 259 259 133 22 98 98 29 23 287 287 184 24 104 104 67 25 45 31 27 26 94 94 57 27 91 82 36 28 98 98 48 Totals 3,235 3,158 1,926

PAGE 138

128 The number of counties out of the remaining 23 with unmet RN manpower needs that had less than half of their needs for 1985 community college RN program graduates satisfied continue to increase in the COMB scenario. In the baseline scenario, five counties had less than half their needs met, while in the COMB scenario, the number was nine. The nine counties with severe RN shortages were counties 4, 5, 6, 14, 17, 19, 22, 27, and 28. County 22 had the greatest percentage of unmet manpower need where 70% of the 1985 community college RN program graduates requirement went unfilled. County 4 had the largest number shortage of RN's where 288 additional 1985 community college RN program graduates were needed. The number of 1985 community college RN program graduates that entered the RN profession in 1985 in Florida totaled 1,926 in the Series I COMB scenario. The niimber needed was 3,235. The shortage of RN program graduates continued to worsen from the baseline scenario where 29% of the demand went unfilled to the COMB scenario where the figure increased to 40%. Indeed, the community college system had increasing difficulty in producing adequate numbers of RN graduates the more the RN manpower requirements deviated from the baseline scenario. Strategy Development The major emphasis of the chapter to this point was to demonstrate how the model combines the exploratory and

PAGE 139

129 normative futures forecasting approaches in suggesting strategies that prevent or achieve trend analysis system future states or adapt to more desirable inventive system future states. The first section specified student demand, RN program space, and RN manpower requirements constraint parameter values used in the trend analysis 1985 long-range forecast involving Florida's community college RN programs. Ranges on the trend analysis forecast were derived from using three alternative 1985 RN manpower requirements scenarios. The trend analysis system future states were labeled Series II futures. The inventive future system states were labeled Series I futures. Student demand, RN program space, and RN program cost constraint equation maximum limits were totally relaxed and determined solely by the three alternative 1985 RN manpower requirements scenarios. The three manpower scenarios again set ranges on the Series I futures. Series I futures specify the desired system futures states. In the results section, it was shown to what extent the RN programs in Florida's community college system had to change by 1985 in order to go from Series II futures to Series I futures. Table 21 summarizes the primary limiting constraints for each RN program Series II future. The primary limiting constraint was the first constraint encountered that capped the RN FTE enrollment at a particular community college. Ideally, the primary limiting constraint desired was requirements. The RN program expenditures.

PAGE 140

130 Table 21 Series II Primary Limiting Constraints Community uo X xege Baseline NHI COMB 1 requiremen X s requiremen L s requireiucn uo 2 demand demand demand 3 space space space 4 space space space 5 space space space b requirement s requirements requirement s 7 requirements space space 8 space space space 9 demand demand demand 10 requirements space space ± J. requirement s requirement s requirements demand demand demand 13 demand demand demand 14 Homo n H 15 demand demand demand 16 requirements requirements requirements 17 space space space 18 demand demand demand 19 requirements requirements requirements 20 demand demand demand 21 demand demand demand

PAGE 141

131 available facilities space, and RN student demand should ideally be just enough to enable the requirements constraint to be the primary limiting constraint. The first point of interest concerning the primary limiting constraints listed in Table 21 is the stability of the constraints across the three manpower requirements scenarios. Only schools 7 and 10 had primary limiting constraints that changed. Primary limiting constraints were not, therefore, particularly sensitive to the manpower requirements in the baseline, NHI , and COMB scenarios. Since the objective was to reach the point where RN manpower requirements were the primary limiting constraints and since Table 21 indicates that only five community colleges had requirements as their primary limiting constraints through the scenario ranges, a significant amount of modification was necessary before Series I futures could be approached. Even the schools with requirements listed in Table 21 as their primary limiting constraints needed space and demand changes to avoid wasting these resources. For example, community college 11 has requirements listed as its primary limiting constraint in Table 21. Yet from Table 15 and Table 16, both demand and space anticipated in 1985 required considerable reduction to be more in line with the Series I recommended value ranges for demand and space. An initial step in developing strategies that prevent Series II futures and aid attaining Series I futures was the determination of the extent to which space and demand needed to

PAGE 142

132 be reduced for schools having requirements as their primary limiting constraint in Table 21. A community college that listed demand or space in Table 21 as a limiting constraint needed to increase the constraint value to its Series I level. This eliminated the constraint from preventing the school from satisfying state manpower requirements for its RN program graduates. In the case of community college 21, demand needed to be increased to a value in the range of its Series I enrollments in Table 15. Once demand was satisfied, the objective of having requirements as the new primary limiting constraint was attained. Checking the space constraint at this point, Table 16 reveals that to prevent wasting space, the 1985 anticipated available space for the RN program was excessive and needed reduction. Not all community colleges had requirements as a new primary limiting constraint after their Table 21 primary limiting constraints were eliminated. For example, once community college 2 had demand satisfied, space became the new primary limiting constraint. In sum, both student demand and facilities space were inadequate in the Series II futures to enable school 21 to meet state needs for its 1985 RN program graduates. Table 22 lists the Series II secondary limiting constraints defined to be the constraints that limit community college RN FTE enrollment at individual institutions after the non-requirements primary limiting constraints are eliminated in Table 21. Combining Table 21

PAGE 143

133 Table 22 Series II Secondary Limiting Constraints Community College CL O ^ X -1-11 NHI COMB 1 requirements requirements requirements 2 space space space f=>m n n H (i pmf3 n (i demand 111 C^XX 4 H pma n d ^ 111 Cell \JL demand demand 5 dpma nd VU. \^ ill CL Xi, demand demand 6 requirements requirements requirements 7 requirements requirements requirements 8 Tpmii Tpments X W U -iX \^ liiV^ 11 Lf O real] i remen t s X \J \A -X. JL \_> 111 XX \J ttj reouirement s q sna p p 10 reouirement s reouirement s requirement s 11 requirements requirements requirements 12 requirements space space 13 space space space 14 requirements space space 15 space space space 16 requirements requirements requirements 17 requirements requirements demand 18 space space space 19 requirements requirements requirements 20 space space space 21 requirements requirements requirements

PAGE 144

134 and Table 22 makes it possible to determine which community colleges face future shortages and/or surpluses of student demand and facilities space regarding RN programs. In this fashion, specific strategies for each community college may be formulated. Summary The limited application of the model to the RN programs in Florida's community college system revealed that with a number of simplifying assumptions the results obtained could be used to formulate strategies that adapt to, prevent, or achieve a possible system future. The most serious parameter estimations occurred when approximating the student flow matrix for RN program graduates. Specific results for community college RN programs indicated that a shortage of RN 1985 program graduates existed under a trend analysis 1985 projection of 1980 conditions. The shortage increased disproportionately as more liberal 1985 RN manpower requirements scenarios were considered.

PAGE 145

CHAPTER V SUMMARY Summary and Discussion This study presents the results of an initial research effort in the development of a statewide planning model for public post secondary education. A prototype linear programming model was developed as a tool to assist state postsecondary education planners maximize selected system-wide policy objectives. The model could be used to generate alternative futures to aid state-level planners formulate strategies that adapt to, prevent, or achieve a possible system future. The focus of the model was not to predict the future, but rather to facilitate anticipation of many futures and the preparation of alternative strategies in response to variable system conditions such as enrollment, social demands for educational opportunities, resource allocation patterns, and manpower requirements. The primary model constraint factors which functioned to determine the future program enrollments at post secondary education institutions were student demand, available state resources, and manpower requirements. How these constraint factors are interrelated is presented schematically in Figure 6 . 135

PAGE 146

136 Figure 6. Model Constraint Factors

PAGE 147

137 The student demand factor reflected the importance of integrating the need to keep postsecondary education accessible to a state's citizenry. The goal of public postsecondary education in the United States has historically been toward providing educational opportunities to the masses that elitist private institutions would not or could not provide. A system-wide plan that neglects considering this dimension of postsecondary education is really no plan at all. The state resources factor consisted of two limiting constraint sets. The first related to postsecondary education appropriations, the second related to instructional facilities. Both constraints were developed to parallel the three-tiered control /planning mechanism within the state postsecondary education systems after Heigham (1969, pp. 300-304). With feedback processes implied, the mechanism of control and planning is schematically represented in Figure 7. Equations 3.11 and 3.14 represent the total money allocated and facilities supported for the postsecondary education system at the state political level respectively. Equations 3.10 and 3.13 represent the fact that the sum of the money and space allocated to each postsecondary education institution in a state through the postsecondary education coordinating/governing board or boards must be less than the totals of these two resources that were provided by the state. Lastly, equations 3.9 and 3.12 state that the amount of money and space consumed by a particular

PAGE 148

138 Governor and State Legislature State Appropriations Post secondary Education Postsecondary Education Governing/Coordinating Board(s) Institution Appropriations An Institution President, Board, Committees Program Appropriations Figure 7. Three Tiers of the Limited State Resources Constraints

PAGE 149

139 program at an institution cannot exceed the amounts budgeted for that program by the institution. By their very simplicity, these model constraints allow for testing any set of if-then cost and space conditions at the specific control/ planning level where the conditions are most likely to occur. The third and final primary model constraint factor was the manpower requirements factor. This constraint factor corresponds to the manpower approach of mathematical programming models in education described by Correa (1975, p. 29). The overriding interest in developing constraints here was that a postsecondary education system should incorporate into its planning routine the economic needs for an educated work force of specific composition. The very fact that both students and state funds earmarked for postsecondary education systems are scarce resources dictates that economic demands for graduates from part icular programs be an integral part of responsible educational planning. With some modifications, limited application of the model was accomplished using the registered nurse, RN, programs offered in Florida's community college system. A set of three future scenarios for 1985 based on trends identifiable in 1980 were devised to determine the 1985 manpower requirements for community college RN program graduates by state geographic area. Each of the three manpower scenarios was applied to two different community college 1985 scenarios.

PAGE 150

140 The first community college RN program scenario, Series I, was a normative scenario which described 1985 community college RN programs as having the freedom to satisfy the manpower requirements optimally. By the application of three RN manpower scenarios, parameter ranges were determined for the resulting ideal community college RN program scenario futures. The second community college RN program scenario. Series II, was an exploratory scenario which described the 1985 community college RN program condition based on 1980 trends. Again, parameter ranges for the resulting straightline projection scenario futures were determined by the application of the three RN manpower scenarios. In sum, six different 1985 futures were investigated. The object of the analysis was to compare Series I and Series II futures to determine the primary and secondary constraints in the Series II futures that prevented the community college RN programs in 1985 from attaining the ideal Series I futures. Strategies were then suggested from the results. Figure 8 summarizes the model application procedure. The model, when applied to the limited situation, yielded results that appeared tenable. The fact that some data required for the model had to be approximated demonstrated the nature of the data problems that have confronted mathematical planning models in education. Indeed, Harcleroad (1971, pp. 34-38), Krause (1979, pp. 86-91), McNamara (1971, pp. 8-9), and McTarnaghan (1974, pp. 28-29) have all stressed

PAGE 151

Figure 8. Developing Strategies from Model Results

PAGE 152

142 the need for the creation and maintenance of comprehensive, accurate management information systems prior to any postsecondary education system modeling effort. By using a series of manpower scenarios which in effect functioned to specify ranges for the two community college scenarios, effects of imprecise input data were diminished. The major strengths of the model include a high level of tractability , ease in modeling any number of scenarios for analysis, low cost, and applicability to other state post secondary education systems. Model weaknesses are that stochastic processes were not considered, some post secondary education system data needed for the model may not be routinely collected in some states, the model is dependent on extensive manpower requirements forecasts from other governmental or private sectors, and to be effective the model needs to be understood by the policy making model users. The last point as well as much of this section may be summarized by the following quote from Ackoff (1970): Planning is the design of a desired future and of effective ways of bringing it about. It is an instrument that is used by the wise, but not by the wise alone. When conducted by lesser men it often becomes an irrelevant ritual that produces short-run peace of mind, but not the future that is longed for. (p. 1) Recommendat ions Model Related Research Prior to full model implementation to a state's public postsecondary education system, a number of substudies are recommended. The substudies are as follows:

PAGE 153

143 (1) The probable student demand relation should be tested for accuracy and for determining the effect of changing policies such as tuition rates, admissions standards, and student aid. (2) Nonlinear cost and assignable square feet of space per FTE parameters should be derived that incorporate economies of scale relations into the model. (3) Obtaining data for the student flow matrix should be made a routine procedure. The matrix should be tested for stability over time to determine its period of validity. (4) A study must be undertaken to determine the extent to which model compatible manpower requirements forecasts exist. (5) Modification of the linear programming model to a goal programming model should be investigated to increase the model's capacity to generate alternative futures due to changing policy scenarios. Florida Community College Recommendations During the period over which the research effort occurred, a number of system conditions were encountered that merit comment. Additionally, the model application to Florida community college registered nurse programs revealed that the direction programs were headed in 1980 would undoubtedly manifest manpower requirements problems in 1985. In sum, Florida community college recommendations are as follows : (1) Accuracy of the Florida community college management information system should be improved by incorporating routine data collection and recording checks both at the state and institution level. (2) Long-range comprehensive system planning should become a meaningful and ongoing effort.

PAGE 154

144 If anticipated shortages of community college trained registered nurses are to be prevented in 1985, steps suggested by the model application segment should be of valuable assistance. If community colleges do not meet the challenge, the following consequences may be incurred: (a) decreased health system quality with corresponding increased health service cost, (b) Florida will import nurses from elsewhere, and (c) other institutions, existing or newly created, will train nurses .

PAGE 155

APPENDIX A SUMMARY OF MODEL VARIABLES, DEFINING EQUATIONS, CONSTRAINT EQUATIONS, AND OBJECTIVE FUNCTION Variables a. . Niiraber of assignable square feet of space ""^ required for instruction per FTE student in program j at institution i. a.j^(i,£) Participation rate of graduates from program j at institution i in occupation Z in county k. A^. Total number of assignable square feet of space for program j at institution i for future time t. c^. Cost per FTE student enrolled at institution i in program j in base year dollars. C^. Total dollar allocation in base year dollars for program j at institution i for future time t . d^ Annual net attrition rate for occupation £. 6^ Weighting factor for the double exponential smoothing forecast of program headcount enrollment for institution i for future time t. Probable student demand in program FTE's for institution i for future time t. gj^ Third level double exponential smoothing ratio of program FTE's to headcount for institution i for future time t. Gj^Number of anticipated graduates from program j at institution i for future time t. Probable student demand by program headcount for institution i for future time t. J]j£ Anticipated need for public postsecondary education graduates for occupation £ in county k for future time t. 145

PAGE 156

146 Weighting factor for the ratio method forecast of program headcount enrollment for institution i for future time t. The average number of jobs per county in occupation I filled by in-migrating workers for future time t. Fraction of jobs in occupation £ filled by public institution graduates for future time t. Total additional manpower needs for occupation i in county k for future time t. Number of FTE students enrolled at institution i in program j for future time t. Anticipated population of county k for future time t. Population of county k in the base year. Ratio of the number of graduates from program j at institution i to the number of FTE students in program j at institution i for future time t. Program headcount enrollment forecast for institution i by the ratio method for future time t. First year in the base period. Planning horizon year. Number of years in the base period. Program headcount enrollment forecast for institution i by the double exponential smoothing technique for future time t. ng Equations X 6 . ^1=T-TT-\^1^\ (3.1) 11 XI Fi = g^H. (3.7) ^ij = "ij^j . (3-15)

PAGE 157

147 * ( I I 0 I 0 m=s+x-l kJlm k ° k£(s+x-l) J°k£t k£ (t-s-x+1) (3.16) 0 k£ 1 2 ^(Pk-Pk^ I Pk I 0. k£ (3.17) Constraint Equations I I a. (i,£) < 1 , (3.19) £ k I Pi-j 1 Fi (3.8) PijC. . < C. . (3.9) j; P. .c. . < y I I ^ij^ij 1 I I (3.11)

PAGE 158

148 Objective Function P..a..
PAGE 159

APPENDIX B FLORIDA COMMUNITY COLLEGE AND COUNTY KEY Number County Community College 1 Alachua Brevard 2 Bay Broward 3 Brevard Central Florida 4 Broward Daytona Beach 5 Charlotte Edison 6 Clay Fla. J. C. at Jacksonville 7 Columbia Florida Keys 8 Dade Gulf Coast 9 Duval Hillsborough 10 Escambia Indian River 11 Hillsborough Lake City 12 Indian River Manatee Jr. College 13 Lake Miami-Dade 14 Lee Palm Beach Jr. College 15 Leon Pasco-Hernando 16 Manatee Pensacola Jr. College 17 Marion Polk 18 Monroe St. Petersburg Jr. College 19 Okaloosa Santa Fe 20 Orange Tallahassee 21 Palm Beach Valencia 22 Pasco 23 Pinellas 24 Polk 25 St. Lucie 26 Sarasota 27 Seminole 28 Volusia 149

PAGE 160

APPENDIX C PROGRAM AND COMPUTER OUTPUT FOR SERIES II BASELINE SCENARIO Program Variable Key Cnn cost variables Dnn student demand variables Snn space variables Rnn requirements variables Xnn FTE enrollment variables 150

PAGE 161

151 • • VPSX-VI V7. . CCNTPCL PPCoSAM CC-^PILEP. M^SX RELEASE 3 C 3 I P P C G 1^ A V 30 3 2 IN I TI ALZ 00 96 TITLE('*^Y DISSERTATION') 3 0 9 7 MGV!Z ( XCMA , • M AXI MI ZE' ) 30 58 MO VE( XP3N AME, • A.^ I ZCNA • ) 309? CCN VEPT { • CHECK • . 'SUV MARY 3 13 3 gccnuT 313 1 3E'U."» ( 'MAX • ) 31 32 MCVEt XC=1J. • FTE' ) 3 13 3 MOVE{ XRHS , ' CAPACI TY' ) J X J ^ H I C J H c 3 135 C^ aSH 3 I 36 I MAL 3 1 3? SCL'JT ICN 3 13 1 EXIT 3 13^ 7 , , Y 3 1 ' A r ICN NAME MAXIMIZE = C'/«S » _ 1 L_ n 1 Lf L 1 I. r o 1 l_ u o 1 l_ 1 L. ! l1 1 L. 1 1 P 1 1 P 1 1 1 u. P 1 1 n 1 T U i J 1 p 1 /i L I '* « P 1 ct U 1 t3 IP 1 ^ 1 L. P 1 1 P 1 3 L PIP D I 9 L C2 3 L J <; 1 L 5 I L L S3 L S A L c c L S6 L £7 L SS L C P L SI 3 L SI 1 L 312 L £1 3 L S 1 4 L £ 1 5 L SI 5 L £ 1 7 L £ 1 3 L S 1 ^? L £ 23 L £2 I L CI L C2 L C3 L C4 L C5 L Cr L C7 L C =5

PAGE 162

152 • • M F £ X L C 1 0 L CI 1 L C 1 2 L C 1 3 L C 1 4 L C 1 3 L C I 6 L C I 7 L C 1 3 L CI 1 L C2 J L C2 1 L PI L h2 L 93 L P4 L F5 L R6 L F7 L CP L pg L F 1 3 L F 1 I L F 1 2 L K I 3 L Fl 4 L F 1 5 L K 1 5 L P 1 7 L F 1 L fi I 1 L F2 0 L F2 I L R22 L F2i 1 F2 4 L F25 L F26 L F27 L F2 3 CCLL = X 1 X I X 1 X 1 X 1 x 1 X I X 1 X 1 X I X I X 1 XI cT^ ^1 1.00000 125. OOo^. CI 32 5 0. : 0000 1 .00144 ^2 .00144 •^r .216CC F4 .00144 C7 '^^.h^" .00 144 .00144 F3 .00144 •^^l^'^ .00144 .00144 ^12 .00144 0 0 144 0 0 1 44 0 0 1 44 1.00144 314 ^ .0014-4 ^le 17 .0 0 144 ^13 -nf .00144 q2C .00144 ^21 .0014^ -22 . 00144

PAGE 163

153 • .MPSX-V I M7 • . DIS3ERTA7ICN > 1 "23 ,00144 f;24 .03144 X I R 25 • J J 1 4 4 12 6 . 3 3 144 X I p 2 . T r' T iT . J ^ <^ -J \j X2 FTF 1 • 00 3 2 3 ")2 1 • 0 3 3 3 0 X2 S 2 13a» 3300 0 C 2 362 3« 3303 0 >? S 1 • 0 0 2 0 3 c; 2 'J 'J \j X2 ~ 3 .00200 1 4 • 3 0 JO 0 X2 n 5 • 0 0 2 0 0 X 6 . 0 0 2 0 0 X2 ^7 •0020: " 3 • 0 " 0 0 0 X2 rio • 0 02 0 0 mo . -J ^ C >J J X2 ^ I I • 0 0 2 3 0 5 12 • 3 3 2 0 T X2 • 0 0 2 0 0 "14 W J J J ^ X2 P 1 5 •032:0 R I 6 V !>/ V C2 = 23 . C 5 2 3 0 =524 X2 -25 . 0 3 2 c : P 26 .33203 X2 27 • 0 0 2 30 R 2 ? . 0 32 0 0 X3 PTE 1 »0 0 0 3 3 •33 x:? 2 3 1 36. : :c : 3 C 3 3 0^0. 0 0333 X3 Rl • 3 0 12 4 1 2 X3 n 2 • 0 3 I 2 4 ~ 4 .33124 X3 r\ "5 • 0 0 I 2 % x6 • 0 0124 X3 — -T .03124 1 ^ • 3 3 12 4 X3 • 3 31 24 .^1 3 . 33' 1 24 X3 I 1 • 03 1 3 3 R I 2 X3 1 3 • C9cO 3 R 1 4 X3 R I 3 .33 124 R I fi .33124 X3 =; 1 7 .09103 R I e .00124 X3 19 . 3 01 2 R2 0 .33124 X3 = 2 1 .00124 R 22 . 33124 >3 -23 .00124 ^24 ,03124 X3 -3 C .00124 R26 . 33 124 X3 IT .02 1 : : R2a .3 3 124 X4 r TE 1 .00000 134 1 . 3 3 333 >4 S4. 1 33. : 0 33 0 C4 2383. 30000 X 4 X4 X4 X4 X4 fi 1 .03 122 ^ 2 . 03 132 .33132 -4 .03132 P. 3 .30 132 .00 122 R 6 .00132 .3 3172 .03132 Rl 3 .00 13? X4 >4 X 4 X4 X4 X4 >4 X4 X4 X5 P 1 I = 13 15 ^ 17 .03122 .0 0 132 .03132 .00122 R I 2 -14 R 1 c R I S .00 132 .0 3 1 32 .03132 . 3 3 1 32 1 9 .00132 R 2 0 . 3 3 1 32 12 I R 27 E .00122 .00 122 .3 3 122 . J 9 3 3 1 . C 0 3 C 3 1 3 3 . 3 3 C 0 3 . 3 0 1 4 R22 R24 ^^26 R 23 C5 C5 R 2 .30132 . 33 1 32 .33132 .132 30 X5 X5 S 3 1.33333 "^350 . 3303 3 . 3 3 W8

PAGE 164

154 • .MPSX-V I M7 . 1Y 01 S3E^T?\r ICN X5 X5 X5 >5 X5 X5 X5 X5 X5 X5 X3 XS X5 xe X6 X5 X6 X€ X6 >6 X6 X6 X6 X6 >D X6 X6 X6 xe XT X7 X7 X7 X7 X7 >7 X7 XT X7 X7 >7 X7 X7 X7 X? xa >8 X8 X3 X3 xa >3 xa k3 S5 PT =;5 Rl 1 K 1 3 "15 =? 17 ^ 1 9 ~2l R23 P25 27 FTE 36 !5 I P3 =?3 R • "9 ^ 1 I ~ 1 3 ^15 S 17 RI9 = 21 S 23 -25 =<27 FTE £ 7 5 1 S3 -< 3 -1 ~ 1 1 .^1 3 = 15 ^ 17 ^ 1 9 = 2 1 R23 = 25 = 27 FTE = a P 1 =,3 ?^~5 1 I I 3 . 0 I 4 ^ • C370 : .00143 .00143 • 0 0 1 4 3 .00143 .00 143 .0014-^ .0 0 1 43 .00 143 .00145 .00143 .03143 1 . 0 0 c : 0 1 38 . ]0 0 0 0 .00 16 4 .0016 4 .0 0 164 . C 0 I 5 4 .24600 .00 164 .00164 .0 D 1 54 .00 164 .00 16 4 .04 1 0 : .30 1=4 .00164 P4 =51 0 ^12 T 14 =5 16 =?i a R20 ^22 ^24 =?25 ^23 3c C5 =. 2 = 4 ^6 R3 ^1 0 = 1 2 = I 4 ^16 =513 =520 =522 x24 =526 .00164 =52 3 1.3000 3 17 133.0 0 003 C7 .30064 ^2 • 0 0 C c 4 = 4 . 33064 =56 . 0 0 C £ 4 ^3 .000 64. -1 0 . 00C64 =512 . 0 C C 6 4 =5 1 4 .0 0 0 34 16 . 3 C C 6 4 =5 1 a .33054 =523 .33054 ^22 . 3 0 C 6 4 =524 .O0CS4 ^25 . : 0 C64 ^28 I . 0 0 0 0 0 "33 133.30003 C3 .03112 H2 .03112 4 .03112 "56 .33112 '"3 • C 0 1 1 2 ^5 1 0 .03112 = 12 .03 112 = I 4 114 0 1. 4 350 1 34 3 0,

PAGE 165

155 '/IPSX-VIM?., .1Y OIS".ERTMICN r C ^ 1 • DTI 1 r< I C O 1 1 O • J >J i i <^ r< O C 1 7 ^. 1 / • u J I 1 .-: .-; I O • J J L 1 ^ AO Kir* "3 3 T •3 "3 ."^ • J J i I d A O • U J i i ^ — O O • J J I I ^ /* o w -J • J J 1 i ^ 2 0 A • J J 1 1 ^ S -'' ~ • w u I L C ^ fi-< C c • J J 1 1 £1 y p. « "5 7 T T t 1 "5 • vJ ^ 1 L ^ :3 "!> ^ • J J I 1 ^ X 9 PTP i • J J u w ^ I t J J J J J X9 Jr3'*J» JJ JUJ X9 K \ . T T 1 A • J J I J * XT _ "1 1 ^ A « A • J J I J X9 n T 1 T • U J I C A • J J I J xg 7 • DTI 1 ii J 1 3 4 xg q 1 g • T T 1 T i _ J . J J I J<* xg . n T 1 i • W i >^ -T C 0 . J J 1 J4 X 9 = 2 3 • -J J J CO A K £l 4^ .JO 104 xg — <; C . 0 01 04 xg ^27 • 0 J I J 4 X 1 2 FT5 1 . T T ^ ,^ 1 ^ I J 1 • J J J J J X 1 J = 10 I "> 3 . 1 r " I J 277 3. 30030 X I 3 . 0 1 1 •=! -7 • w J i O O 3 p • J J 1 oo > I D . ii ? T T • ^ ^ c, ^ J • ^ J 1 o o X 1 0 ~ ^ . o T 1 A 3 % \j i u D ^ C .001^3 X I D • vj J L C C . J J 1 OO X I 0 9 ^ ^ i \» . J J I C ^ X 1 0 PI 1 .03 163 ^ 1 2 .12 6 3 3 X 1 : S 13 .03163 ^14 .00168 XI 3 P 1 5 n 1 A v5 • O J i C *3 • 3 3 1 66 X I 0 ^ 1 7 .30165 = 1 a .03 163 X I 3 q I g • \J J I C o .'^ -V :i J . 3 0 163 XI 0 -2 I .03166 ^22 .3 3163 X 1 D .30 163 ^2 0. .33163 X I 3 • I c C J J -~ -i c . 3 3 163 X 1 3 ^27 • 0 3 1 6 3 =?2.3 .33163 X 1 1 F'E 1 • 0 0 C 0 c '3 11 I . T T T 1 T i • ^ -J Vj W ^J XI 1 S 1 1 13q,03C0 3 C I I 301 0. 3 3 303 XI 1 "1 .0312^ ^2 .00 124 X 1 I 5 ? .03124 ^4 .03124 >1 1 ~ 3 .00 124 P6 .36200 X I I X I I XI 1 p g = 1 1 .09330 .03 133 .00 124 '58 ^. I 0 R 1 2 .0 3 124 .03 124 .33124 X 1 I XI I R 1 3 S 1 5 . C 31 24 .03133 ^14 R 1 e .30 124 .33124 X 1 I X I I XI I X 1 1 X I 1 P 1 7 R I g P2 1 .0 0 124 .03124 .0 3 124 5 1 8 R2 C P22 .33124 .00 124 . 30124 =;2 3 P 23 .00124 .03124 -24 26 . 3 3 1 24 . 33 124

PAGE 166

156 • .VPSX-V IM 7 , , -lY D : S"E=5 'AT I CN XI 1 C27 .00124 9 2 5 .33124 X 1 2 F 11 1 . ODCCO ^1 2 1.33 303 X I 2 S I 2 13 3.30000 C I 2 3200. 30000 > 12 ^. I .00115 S2 .33115 X 1 2 R3 .001 16 ;^4 .03115 X 1 2 S5 .0011-3 9. 6 . 3 0 115 XI 2 ST .0011^ 93 .30116 X 1 2 • C 0 1 I 5 9 I 0 .03116 ><1 2 ?l 1 •CO 1 1'~ 912 .33115 X 1 2 S I 3 . 0 0 1 1 e 9 1 4 .03116 X 1 2 R 1 3 .0 3 1 15 I 5 . 05300 X12 -17 .03115 9 13 .33115 X 1 2 ^ 1-5 . COl 16 920 .30116 XI 2 k2 1 .0 3 I 15 92 2 ,33116 X 12 5 2 3 .32933 9 2 4 .33115 X I 2 S 25 .33115 -26 .115 0 0 XI 2 -y .03115 923 .33116 XI 3 F TE 1 . 3 0 C 0 3 313 1.30033 X 1 3 S t 2 n B . 3 3 C 3 3 Z 1 3 33 3 0. 3 3000 X 13 1 . C C 1 9 2 92 . 3 3 i 92 X 1 3 S3 .00192 94 . 34 1 3 3 X 1 3 ^ .03192 9 5 . 3 31 ^2 X 1 3 C7 .00192 93 . 2 J 3 3 3 X I 3 99 .00192 910 .33192 XI 3 1 1 .00 19 2 912 .00 192 X 1 3 12 .00192 1 4 .30 1 92 X I 3 ^ I 5 .0 3192 9 1 5 . 3 3192 X 1 3 =;i 7 .33 192 9 1 3 . 3 : 1 ->2 XI 3 ^ 19 .03192 9 20 . 3 3132 >1 3 n21 • 00192 922 . 30 192 XI 3 2 3 .03132 '^24 . 33 I ?2 X I 3 ,^2 5 .00192 "'2£ . 0 01)2 XI 3 27 .00 192 9 2 3 . 3 3192 X 1 4 F"rE 1.30033 014 1.00333 X 1 4. S 14 1 33. 30C00 C 1 4 3 15 3.00333 X 1 4 -. I .00143 — 2 .33143 X 1 4 F5.3 .03145 94 , 3 0 1 a 5 X 1 ^ c;5 . 0 3 1 4 P ~t . 3 3 1 4 d X 1 4 -7 .03143 93 . 3 370 3 X I 4 n r .33143 ^ I 0 .001 48 > 1 4 1 I .00143 91 2 .33142 X 1 4 R 1 3 .0314= r 1 4 . 3 3 I 43 XI 4 S 1 5 .0314=! I 6 .001 4 3 X 1 4. 0 1 .03143 9 1 3 .331 -"-B X I 4 5, 19 .33143 92 0 .33143 XI 4 K2 I .22230 92 2 . 3 0 1 4 S X 1 4 -23 .00145 924 • OOlJ-B X 1 4 25 .00 146 2c . 3 3 1 43 X 14 f.2 7 . 0 0 1 4 T 28 .3:14-1 X 1 3 F-H 1.00003 ">15 1.33333 X 1 5 3 I 5 I 33 .0 3C0 3 C 1 5 3 30 0. :)ri2 00 X I 5 = 1 .33134 . 3 : 1 3 i X 1 z S3 .0010 4 ^ 4 .301 34 XI 3 -5 .00104 .03134

PAGE 167

157 ,VPSX -V I ,V7 , , ,1Y C ISSEKTAT ICM X 15 =7 . 3 Jl 3* =;3 .301 34 Xl^ n-^, .0 3 134 ^10 .30104 X15 Sll .03134. ?12 .00104 X13 S13 . 331 04 =?14 .30104 XI 5 RIS .0310*+ Pie .00104 X15 ^17 .0310 4 KIS .03134 X15 Rll .00134 R20 .30134 X15 S2l .3 0 104 ^.2 2 .156 0 0 XIS R23 .00134 ^24 .32630 X15 .03104 126 .33134 >15 =27 .0010 4 rj^a .30104 X16 FT5 1.03CC3 T15 1.00000 X16 Sto 136.03CC0 C16 3000.00000 X16 ^1 ,03 143 «2 , 33143 X16 ^3 .0014a ^4 .0314 3 Xl-^) R5 .0314a Pe .3 0143 X16 =7 .0 014,:^ eg ,33143 Xlc R -. .30143 RIO .22200 X16 -11 .001-+3 =12 .03145 X16 =? 13 . 30143 r;i4 .00143 Xla nl5 .00143 ^16 .03 1 48 Xlo ^17 .00 143 ^13 .33143 XI '5 ^l i . 037 3 0 P2 0 . 33 143 "^21 .0314? -22 . 00149 Xlo ^23 .03143 -24 .33143 ^l^' . 03 143 ^26 .00143 >\o .00143 -2° .00143 Xt^ PTZ 1.030CO -17 1.00330 ^11 133.33C03 C17 2730.00000 XI 7 -1 .00 112 ^2 .0311^ XI 7 R3 .00112 14 .00112 l\l .03112 ^.3 .30112 -1 .00112 ^3 .0 0112 ^17 ^9 .00112 ^13 .30112 X17 ^11 .03112 -12 ^1"^ "13 .00112 ni4 .0 0112 Rie l\l III .03112 PI 3 .33112 Jl' .00112 R2C X 1 7 P 2 1 .03112 R 2 2 ^^'^ ^23 . 001 12 R24 ^-^ .00112 ,._ ...... ^^'^ ^27 .03112 R23 .3311'' Hi ^Tl i.ooooo" i33.o:cc: ci3 3130.00000 r', .03 124 R2 .33124 m5 '^^4* .03124 -^^3 .00 1 24 P6 .00124 .03112 .03112 00 112 3 3112 . 02 30 0 .30112 . 1 53 30 .33112 .33112 1.03000 i^t a.00124 eg .33124 .03124 lie .00 124 ^11 ^ I .03100 R12 .30124 ; I -3^12* ^1* . 3 3 24 .30 12 4 ^16 ..niPa C17 .00 124 R13 = .00 124 .03124

PAGE 168

158 :SX-Vl V7. , MY C ISS E.^T ^T ICN 3 9 5 9 X 1 3 X 1 '3 X 13 X 13 XI 3 XI 9 X I 1 > 1 9 X I -5 XI XI X 1 > I XI X 1 XIO XI 5 XI 9 X 1 9 X I '5 XI 9 X2 D X2 D >2 0 X2 : X2D X2 3 X? : X23 >2 J X2D X2 3 X 2 3 X20 X2 3 X2 3 >2 0 X21 X2 1 X2 1 X2 1 X2l X21 X2 I >2l X21 X2 1 X2 1 X2 ! X2 1 X2 1 X? I X2 1 R 19 S21 ^22 =5 2 5 -27 FTS 19 1 R3 ^5 ?7 P.O ~1 1 1 3 ~ I 5 S 1 7 i5 19 ^2 I 2 3 S 25 = 27 FTH 3 23 ~l = 3 !T f I ~ I 3 :^ I 5 = 17 ^ 1 3 P2 I ^^22 ^25 ^27 F r-£ 32 I ~ 1 R 1 P ^ ~l I =^ 1 3 R I 3 S 1 7 .s; 19 P2l -23 138 133 1 I 33 ;h3 3 312 4 0 3 12 4 18 6 0 3 0 0 12 4 0 0 12 4 -1 -» U J '.^ V 0 o 3 0 0 3 C 3 3 3c : c 0 3 C C =!G 0 3080 0 2 C 3 3 0 2000 ^ w W C W 3 3C = 3 3 3 0 3 3 w w w 3 0 0 0 3 0 00 330 0 0 C 5 0 0 0 0^3 ^ vj ^ J 3 30 3 0 00 155 00 15c 3 3 156 C 0 155 035 0 0 00 156 00 I 56 2340 3 00 136 0 01 56 0 0 15 6 0 3 1 56 00 1 56 0 3 15 6 0 0 0 0 3 3 3C03 0 014 3 3 3 7 3: 0 3 14 3 3 3 14 3 0 3 143 00 1 4 ^ 00 Mc 0 0 14 = 00 143 0 014 3 0 0 1 4 R 0 0 14 8 0 0 14? 0 0 14 3 ^ ^ -20 .00 124 2 2 .0012^ T 5 i • 0 0 I 2 4 rt 2 C .00124 n T !' .-i c .00124 -'19 1.00030 <~ 1 ^ 2990. 30000 ^ "3 .33033 • 0 3 03 0 • 0 0 0 8 0 ^•^ . 02 3 0 0 EJ 1 "> ^ I J .00080 C 1 5 .30080 • 0 0 3 3 0 I C . 0 3 03 0 5 1 ~ t o . 3 0 38 3 .00030 ~ ? . 0 0 0 3 0 ^ + .03333 K «i D . 0 0 030 C ^ .00333 1.30030 ' £: J 233 0. 003 0 0 • 0 3 1 56 R 4 .001 56 ~ ?5 • 3 015 6 ^3 .331 DC 1 1 . 3 0156 i. _ . 30 I 56 9 14 . 3 3 r -3 6 ^ i C . 0 0156 s 1 a . 3 3 156 T «V .00 156 " c c . 3 0 156 .331^6 o c ^ ^ c . 001 56 c a •-^ J T . 331 56 1.03300 372 3. 00 000 R2 .03143 ^4 • 00143 ^6 . 30143 R8 .33143 RIO .00143 Rl 2 .00 1 43 c 1 4 .331 43 R 16 . 0 0 1 43 R 13 . 33148 ~2 0 . ? 2 2 0 0 022 . 3 3 I 43 ==24 .33143 R26 .00143 R2 3 .00 148

PAGE 169

159 CAPACITY CA^ACI'Y CA=>AC I TY C APAC ITY CA3AC I 7Y CA^AC^TY CAPACITY CA^ACI T Y CA3 ACI TY CAPAC ITY CA^ACI ~Y CAPACITY CA^ACI'Y CA MCI TY CAPAC ITY CA =:aC I TY CA=> AC I TY CAPACI TY CA^AC I T Y CAPACITY CA !=AC ITY CA 3AC I'^Y C A^ACITY C A P A C I Y C AP ACI TY CAPAC IT Y CAPACI TY C APACI'Y CA34C ITY CAPACI T Y CAPACITY CA-ACI TY CAPACITY C A ^AC ITY CAPACI T Y C AP AC ITY CAPACI TY CAPAC ITY CAPACITY CA^ACI-Y CAP ACI T Y CA=AC I T Y C AOAC IT Y CAPAC ITY C A ^ A C I "^Y CAPAC I TY ATA •1Y Z I£SEKT\T ICN U -J •J J OJP»OJUJJ J 1 I C # 4 ^ \J 1 1 7 s ^ C ::, 5 7 S I J -J L C ~ I 41 w J U J • .J U u J C / O J J J J J 0 J S I 3 con 1 / U J • U 0 u J J r I r J J 0 J C J • 'w J r -J J C ^ L 0 0 » J J J C 1 3jCCGC0# CO D C -J ^ u \J \j J •L-J J C 1 1 C ! 3 C15 n J c 0 J 0 0 . : c 3 C I 7 C 19 3oCC300,033 C21 R 2 35. 0 3CC 3 -I 263,00 00 0 ^ 6 3 0 . 0 OC 0 0 R 3 376,00000 = 1 0 56 . 000 30 ~ 12 2 5.00003 ^ I 4 87.03C00 ~ 1 € 53 .000 0 0 P 1 3 1 e. ::o 00 -2 0 129.0000 3 P22 S2, 30COO q 24 So, 0 0000 -25 "'8,3003 3 -. ?3 62 . C 30 0 0 J d 323,0 0 003 J 471, 000 0 0 Jo 309,00 0 00 2 75, 00 00 0 J 1 149,0 0 000 'J L 416, 00000 J I 4 4 79, 0 0 000 J I o 521, 0003 0 1 a . J I o 797. 00 000 > J 96, 00000 D 1 '+6400.00000 c ~x 2070.>, 00000 2 0 7 0 ^ , 0 3 0 0 0 ^7 11100.00 0 00 I 00200. 0 0 00 -> i 1 216 0 3.03300 I 37o00. 0 0 00 r1 := J 4 C 2530 0, 00000 C 1 7 44600,0 0 0 00 " I Q 1 10000,00 0 0 C" O 1 1 1 49 300 , 0 0 0 0 /o i 30 0 000 0,000 3000000, 0 00 r ^ 3300w0^, 30 0 J 300 OuOO,000 — 1 J 300 0 00 3, 0 00 r 1 o ^ A . J J JO, 000 C 14 3000000,000 C 1 c 300003 0,0 00 CI a 3000 00 0,00 0 C 20 3 000 00 0. 0 00 Nl 43.00030 3 6 7, 00 00 0 T 3 0, 00 000 ^7 14,03330 ~9 1 1 7. 30 000 -! 1 1^3,30300 ^ I 3 33.0000 0 ^ 1 5 4 7, 00 00 0 =; 17 4 5.30300 ^ 19 3 7.00000 -2 1 217.0 0000 -23 240.00000 ^ P 5 37, 0 0 000 Ri7 76, 00000

PAGE 170

160 < I f\.' o ^ O fy e o o oI I I I — f J L" O o o no o o o oo o o o oo o • • * • • O * O * IB M O — OOO ooo ooo ooo ooo • • • d >» — ™ * — in K. iC oo oo o o oo o o • • oo o o «•* o e o r>oo ooo ooo O O • • • ooo o o ^ Nm N O-iO oeoo oooo o o ors o o oo oooo • • • • o o oo O O o o ^-•.ln(\J n»««-o N — — O oooo o o c o oooo oooo o O o o • • • • oooo OOoO ocnin OOO ooo ooo ooo ooo • • • OOo ooo r ' "n o oin»» fvj tO ooo ooo ooo ooo ooo • • • ooo 00-! ^ o. >c *o« oe c C' ? ~ c O O ^ 0* ^ ^ ^ oo ^ ^ 6s c ^ O O ^ 0* 0* oo ;> o ^ ? r> • •••••• o o 0> ^ 0^ c oo l> o» ^ K c» ^ ^ ^ -ff ^ {> (Js ^ ^ cv r» cvj N !L' LJ l; a Uj f ii; li; i rzrzzzzzr: rzzrzrzzr; c ill » c > K > o o t>>— in • • = 1 in — o" Ofine o — o— ae IT OCsO«>«K o n o » m r,' o o rj * p o* incjf. c r. — o rjK IT,'-' o oo ^w *N O ?~ SC oir o o • • • • • • *••••• o n c o — n o oo,' » c ni — o— oc CO • • o — OCPJOOOOO Of^CDOOOOt* on _ooook O'CcOoOO — oooooooc *JVO'CWO«J»t)C *o:e — cM—in r o — 0 c oo o 1 • e o o cc « O Cv CT e> o o p p r c O '> »C ^ • • • • • O 1^ C C"^ o ~. K (;> c » pj or»pjv — c> o cmn^r — lyr.T! — OCfN < IMO oj o i.»":o tn»«o • • • OOn O"oc oo Of nrg n in oe)— oc Ul o!n« ON n • • • O— p: — n >oo — »ro^ cy o — "OO oot• • • incp. OOpj — «— moo IV oo UJOO ooo «roo « -• • n«ii m — m ooo o o^ oorj 000 oec • • o> n tv) K Pg_ — O"ttjOO no — o — • • • n».— ooff>oooonirocf.'BOooooooe-.r>oori_or.oo.c-ooe' OOK-OOOO0>OOO r o o o o o o o eoo>eooof-noo«'r, Oooo— •-oil CKino>vn \0 I* OO CO OO ooc o> OOC P • • i." i\j t* ^ P^ c n — P in — I.' » n o t" o s; c o o n 1,1 r c N • • • • t> — o p.' y> IP o p; P n ,^ ^ -s. -.J « "UL O . * C T — "-P.-r.'~r;Ncvjpjcwr.-nr.T;n^n?;nr:.r»*^rc**««5;;^

PAGE 171

161 -0 IT '> e IT c r; o n « '0 O -i < ' c * • • c c 0« 0* f> c^ ^ rs O ~ • • • C» (M> ^ O* ^ o* ^ ^ ft; P.I a; o o ^ o ' c o ,oo> OCOOOOOOOOOO'SOOCOOOC 0oooorjoooooooooo03oo '^OOOOOoOOOOoOOOCOoOO **•«••••*•«••••«•** o c oo c o o o • • 3 oc e '5 oo ^ 3 O O o O C O O c c 3 C O O O • • • • • r» o >c r N. T . nj c u < 1 i" > » — 3 CM.' C uIT -P"iO — .O'^w-e PJ PJ r — PJ — c rT ^ ^ p^ : s K ^n K ^ 'C O 0 ?> t> o 1" PJ c:p.'.n P.' O fvJ ^ 'J' C ^3-, • c C o « PJ O PJ o u". c PJr^. O P. O f 1 PJ rf. n — f T L— >L' PJ — PJ — O rj fT— -^ O P^ C C*' C P^ 0> O » ? Lc 3' P> O' « 7> ^o ^ ^ P^ — o o • •••••• o ^» ^ L^ — o > o < c LTu c n c c fp ^ <; i\j ? T -J » O J -< O r c n s a) u ' ^ Pj IT. O p. c r n o c • • 3Pg \ o — O N PJ p; IP ;opr o pr 'Mr • • I -•r >o » • — (ji • 'J p? p* o o c o -: o c o — ' Pj PJ o — con 30 inf • -T r>. p; PJ c p* Aj Pj ^ o c >c — o pc • • . • PI r P' PJ rr Pipj o P* P^ o c * -5 C o r~ o u I in — p— « • • • ^ T Ip>j p« o p.' PI c p; O P* pp. 1 C P^ ij* (T O P* u: — — C t C C • • • • • OL« '1 LC p. uir , ^L. J 'T^ 'v! r ^ = -<•"f-' C 1^ r ^ P

PAGE 172

162 (A c u c UJ u 3 C Ui ir >L, :iJ u a: ;i; ij u u.' lii u LJ u UJ u iu li: iL' UJ I-' -ZZZZ22ZZZ2ZZZZ-ZZZZZ Z^ZZZZZZZZZZZZZZZZZZZ a 3 z J i c eooooooCoocooooooococ cooooooooooc^oooooooo z > coo'-'Ocr;p''-o — tcooof^TOcoo *CC)00i~i."C3^O'^'^oO'?OC'-O^-T^ — ."ot^c>^rO'^3>*ooo'5'C — o — O'-' > — C c o — rj o N n »c L". " f'J n N. 'u N ^N u' IT c r» O o l\J Ul — L" N — C
PAGE 173

BIBLIOGRAPHY Ackoff, R. L. The development of operations research as a science. Operations Research Quarterly , 1956, 4, 265-295. Ackoff, R. L. A concept of corporate planning . New York: John Wiley & Sons, 1970. Amara, R. C. , & Salancik, G. R. Forecasting: From conjectural art toward science. The Futurist , 1972, 6, 112-116. Banghart , F. W. Educational systems analysis . London: MacMillan, 1969. Bell, C. E. Quantitative methods for administration . Homewood, 111.: Richard D. Irwin, 1977. Berdahl , R, 0. Problems in evaluating statewide boards. In R. 0. Berdahl (Ed.), New directions for institutional research; Evaluating statewide boards . San Francisco: Jossey-Bass, 1975. Bertalanffy, L. V. General systems theory. In L. V. Bertalanffy & A. Rapoport (Eds.), General systems: Yearbook of the society for the advancement of general systems theory . Ann Arbor: Braun-Brumf ield , 1956. Bogard, L. Management in institutions of higher education. In A. M. Mood, C. Bell, L. Bogard, H. Brownlee, & J. McCloskey (Eds.), Paper on efficiency in the manage ment of higher education . Berkeley, Calif.: Carnegie Commission on Higher Education, 1972. Brevard Community College. Brevard Community College catalog 1979-81 . Cocoa, Florida: Author, 1978. Brown, R. B. Forecasting. In J. J. Modar & S. E. Elmaghraby (Eds.), Handbook of operations research: Models and applications (Vol. 2). New York: Van Nostrand Rinehold, 1978. Bureau of Economic and Business Research. Florida statistical abstract 79 . Gainesville, Florida: The University Presses of Florida, 1979. 163

PAGE 174

164 Casasco, J. A. Corporate planning models for university management . Washington, D. C. : ERIC Clearinghouse on Higher Education, 1970a. Casasco, J. A. Planning techniques for university management . Washington, D. C. : American Council on Education with the ERIC Clearinghouse on Higher Education, 1970b. Central Florida Community College. Central Florida Community College catalog 1980-81 . Ocala, Florida: Author, 1979. Chance, W. A. Long-term labor requirements and output of the educational system. The Southern Economic Journal , 1966, 32, 417-428. Chapman, P. F. A method for exploring the future. Long Range Planning , 1976, 9, 2-11. Chirikos, T. N. , & Wheeler, A. C. R. Concepts and techniques of educational planning. Review of Educational Research , 1968, 38, 264-276. Churchman, C. W. The systems approach . New York: Dell, 1968. Churchman, C. W., Ackoff, R. L. , k Arnoff, E. L. Introduction to operations research . New York: John Wiley & Sons, 1957. Coombs, P. H., & Hallak, J. Managing educational costs . London: Oxford University Press, 1972. Correa, H. Flows of students and manpower planning: Application to Italy. Comparative Education Review , 1969a, 13, 167-178. Correa, H. Quantitative methods of educational planning . Scranton, Pa.: International Textbook Co., 1969b. Correa, H. Analytical models in educational planning and administration . New York: David McKay, 1975. Daellenbach, H. G. , & Bell, E. J. User's guide to linear programming . Englewood Cliffs, N. J.: PrenticeHall, 1970. Dantzig, G. B. Programming in a linear structure . Washington, D. C: Comptroller, USAF, February, 1948. Dantzig, G. B. Programming of interdependent activities, II, mathematical model. Econometrica , 1949, 17, 200-211. —

PAGE 175

165 Dantzig, G. B. Linear programming and extensions . Princeton, N. J.: Princeton University Press, 1963. Daytona Beach Community College. Daytona Beach Community College 1978-80 general catalog . Daytona Beach, Florida: Author, 1978. Dresch, S. P. A critique of planning models for postsecondary education. Journal of Higher Education , 1975, 46, 245-286. Edison Community College. Edison Community College general catalog 1980-81 . Fort Myers, Florida: Author, 1979. Evans, W. K. Student flow modeling: An enrollment projection tool for administrators. Planning for Higher Education, 1975, 4 (6: 2/7). Fincher, C. Planning models and paradigms in higher education. Journal of Higher Education , 1972, 43, 754767. Florida Board of Regents, Florida State University System. Nursing education in Florida . Manpower Study Series, No. 10. Tallahassee: Author, 1969. Florida Department of Education, Division of Community Colleges. Community college management information system procedures manual . Tallahassee: Author, 1977a. Florida Department of Education, Division of Community Colleges. 1975-76 AA-IA program enrollments and completions reports (unpublished). Tallahassee: Author, 1977b. Florida Department of Education, Division of Community Colleges. 1976-77 AA-IA program enrollments and completions reports (unpublished). Tallahassee: Author. 1978. Florida Department of Education, Division of Community Colleges. Report of annual program enrollments and completions by ICS groups 1977-1978 . Tallahassee: Author, 1979. Florida Department of Education, Division of Community Colleges. 1978-79 AA-IA program enrollments and completions reports (unpublished). Tallahassee: Author. 1980a. Florida Department of Education, Division of Community Colleges. 1978-79 AA-2 placement reports (unpublished). Tallahassee: Author, 1980b.

PAGE 176

166 Florida Department of Education, Division of Community Colleges. 1978-79 CA-3 full cost of instruction reports (unpublished). Tallahassee: Author, 1980c. Florida Department of Education, Division of Community Colleges. Report for Florida community colleges 197879 . Tallahassee: Author, 1980d. Florida Department of Education, Division of Community Colleges. 1978-79 RF-1 space requirement reports (unpublished). Tallahassee: Author, 1980e. Florida Department of Education, Office of Educational Facilities Construction. 1978 space utilization standards for estimating facilities needs . Tallahassee: Author, 1978. Florida Department of Education, Office of Educational Facilities Construction. Survey of educational facilities : Pasco-Hernando Community College . Tallahassee: Author, 1980. Florida Junior College at Jacksonville. Florida Junior College at Jacksonville catalog 1980-1981 . Jacksonville, Florida: Author, 1979. Florida Keys Community College. Florida Keys Community College catalog 1980-81 . Key West, Florida: Author, 1979. Folger, J. K. , & Nam, C. B. Trends in education in relation to the occupational structure. Sociology of Education , 1964, 38, 19-33. Fontela, E. Scenario generation by cross-impact analysis. Futures , 1977, 9 (1), 87-89. Ford, L. R. , k Fulkerson, D. R. Flows in networks . Princeton, N. J.: Princeton University Press, 1962. Friedman, Y. , & Segev, E. Horizons for strategic planning. Long Range Planning , 1976, 9 (5), 84-89. Gardner, D. E. Enrollment forecasting with double exponential smoothing: Two methods for objective weight factor selectiolT ! A paper presented to the Association for Institutional Research Forum, Atlanta, April 1980. Gass, S. I. Linear programming: Methods and applications . (4th ed.) New York: McGraw-Hill, 1975. Gillett, B. E. Introduction to operations research: A computer-oriented algorithmic approach . New York: McGraw-Hill, 1976. ~~

PAGE 177

167 Gleazer, E. J., Jr. After the boom . . . what now for the community colleges? Community and Junior College Journal, 1974, 44(4), 6-11. Glenny, L. A. State coordination of two year college financing: A necessity. In E. J. Gleazer, Jr. & R. Yarrington (Eds.), New directions for community colleges: Coordinating state systems . San Francisco: Jossey-Bass, 1974. Goddard, S. , Martin, J. S., & Romney , L. Data element dictionary . (2nd ed. ) Boulder, Colorado: Western Interstate Commission on Higher Education, National Center for Higher Education Management Systems, 1973. Golladay, F. L. A dynamic linear programming model for educational planning with application to Morocco (Doctoral Dissertation, Northwestern University, 1968). Dissertation Abstracts International , 1969, 21, 3280. (University Microfilms No. 69-06928). Goodlad, J. I., O'Toole, J. F. , Jr., & Tyler, L. L. Compu ters and information systems in education . New York: Harcourt, Brace and World, 1966. Greenberg, M. R. Applied linear programming for the socioeconomic and environmental sciences . New York: Academic, 1978. Gulf Coast Community College. Gulf Coast Community College general catalog 1980-1981 . Panama City, Florida: Author, 1979. Hadley, G. Nonlinear and dynamic programming . Reading, Mass.: Addison-Wesley , 1964. Hall, A. D. A methodology for systems engineering . New York: Nostrand, 1962. Hall, A. D., k Fagan, R. E. Definition of a system. In L. V. Bertalanffy & A. Rapoport (Eds.), General systems: Yearbook of the society for the advancement of general systems theory . Ann Arbor: BraunBrumfield, 1956. Halstead, D. K. Statewide planning in higher education . Washington, D. C. : U. S. Government Printing Office. 1974. ' Harcleroad, F. F. Comprehensive information systems for statewide planning in higher education: Some prospects and critical concerns. In J. D. Boyd, A. D. Brown, F. F. Harcleroad, & B. Lawrence (Eds.), Comprehensive information systems for statewide planning in higher education . Iowa City: ACT, 1971.

PAGE 178

168 Harris, N. C. State-level leadership for occupational education. In J. L. Wattenbarger k L. W. Bender (Eds.), New directions for higher education; Improving statewide planning . San Francisco: Jossey-Bass, 1974. Heigham, D. A. C. Two O.E.C.D. feasibility studies on applications of operational research to education in England and Wales. In Organization for Economic Cooperation and Development, Efficiency in resource utilization in education . Paris: Author, 1969. Heinmiller, J. L. The diffusion of a new discipline. Educational Research Quarterly , 1977, 1, 13-21. Hillier, F. S. , & Lieberraan, G. J. Operations research . (2nd ed. ) San Francisco: Holden-Day, 1974. Hillsborough Community College. Hillsborough Community College catalog 1980-1981 . Tampa, Florida: Author, 1979. Hu, T. C. Integer programming and network flows . Reading, Mass.: Addison-Wesley , 1969. Huckfeldt, V., Weathersby, G. , & Kirschling, W. A design for a federal planning model for analysis of accessibility to higher education . Boulder, Colo.: Western Interstate Commission on Higher Education, National Center for Higher Education Management Systems, 1973. Hufner, K. Economics of higher education and educational planning — a bibliography . Socio-Economic Planning Science, 1968, 2, 25-101. Hussain, K. M. A resource requirements prediction model (RRPM-1): Guide for the pro.iect manager . Boulder , Colo. : Western Interstate Commission on Higher Education, National Center for Higher Education Management Systems, 1971. Immegart, G. L. Systems theory and taxonomic inquiry into organizational behavior in education. In D. E. Griffiths (Ed.), Developing taxonomies of organizational behavior in education admini stration. Chicago • Rand McNally, 1969. Immegart, G. L. , & Pilecki, F. J. An introduction to svstems for the educational administrator . Reading, Mass. • Addison-Wesley, 1973. Indian River Community College. Indian River Communit y College cata log 1980-1981 . Fort Pierce, Florida: Author 1979 . '

PAGE 179

169 International Business Machines. Mathematical programming systems-extended (MPSX), and generalized upper bounding (GUB) program description . White Plains, N. Y. : IBM Data Processing Division, 1972. Jantsch, E. Technological forecasting in perspective . Paris: Organization for Economic Cooperation and Development, 1967. Jarupanich, P. J. An alternative systems methodology for planning in higher education (Doctoral Dissertation, Oklahoma State University, 1978). Dissertation Abstracts International , 1979, 39, 4758A. (University Microfilms No. 7903685). Johnstone, J. N. Mathematical models developed for use in educational planning: A review. Review of Educational Research , 1974, 44, 177-201. Jones, A. M. , & Updegrove, D. A. EFPM — The EDUCOM financial planning model: In use over EDUNET . Paper presented at the CAUSE 78 National Conference, New Orleans, December 1978. Joseph, E. C. An introduction to studying the future. In S. Hencley & J. Yates (Eds.), Futurism in education . Berkeley: McCutchan, 1974. Judy, R. W., & Levine, J. A new tool for educational administrators . Toronto: University of Toronto Press, 1965. Kahn, A. G. A survey of educational planning models in the O.E.C.D. member countries. In A. R. Smith (Ed.), Models of manpower systems . London: The English Universities Press, 1970. Keane, G. F. , & Daniel, J. N. Systems for exploring alternative resource commitments in higher education (SEARCH) . New York: Unpublished report from Peat, Merwick and Mitchell, 1970. Krauss, L. I. Computer-based management information systems . New York; American Management Association, 1970. Lake City Community College. Lake City Community Coll ege catalog 1979-81. Lake City, Florida: Author, 1978. Lake-Sumter Community College. Lake-Sumter Community College catalog 1 979-1980 . Leesburg, Florida: Author, 1978? LeVasseur, P. M. A study of inter-relationships between education, manpower, and economy. SocioEconomic Planning Science . 1969, 2, 269-295."

PAGE 180

170 Linneman , R. E., & Klein, H. E. The use of multiple scenarios by U. S. industrial companies. Long Range Planning , 1979, 12 (1), 83-90. Lins, L. J. Methodology of enrollment projections for colleges and universities . American Association for Collegiate Registrars and Admissions Officers, 1960. Linstone, H. A. , & Turoff, M. (Eds.) The delphi method; Techniques and applications . Reading, Mass.: AddisonWesley, 1975. Lyddy, J. B. Statewide planning and coordination of postsecondary education: Relationship to private, nonprofit postsecondary educational institutions (Doctoral Dissertation, The Catholic University of America, 1975). Dissertation Abstracts International , 1975, 36, 1332A. (University Microfilms No. 75-19514). Lyell, E. H. , k Toole, P. Student flow modeling and enrollment forecasting. Planning for Higher Education , 1974, 3 (6: 2/6). Maki, D. A programming approach to manpower planning. Industrial and Labor Relations Review, 1970, 23, 397405: ~ — Manatee Junior College. 1980-1981 catalog of Manatee Junior College . Bradenton, Florida: Author, 1979. Mangelson, W. L. , Norris, D.' M. , Poulton, N. L. , & Seeley, J ' A. Projecting college and university enrollments: Analyzing the past and focusing the future . Ann Arbor, Michigan: Center for the Study of Higher Education, The University of Michigan, 1974. Martino, J. P. Technological forecasting for decision-making . New York: American Elsevier, 1972. Mason, T. R. New Directions for Modeling? In T. R. Mason (Ed.), New directions for institutional research: Assessing computer-based systems models "! San Francisco Jossey-Bass, 1976. McConnell, T. R. A general pattern for American public higher education . New York: McGraw-Hill, 1962. McKinsey, J. C. C. Introduction to the theory of gam es. New York: McGraw-Hill, 1952. McMillan, C. , Jr. Mathematical programming . (2nd ed. ) New York: John Wiley & Sons, 1975.

PAGE 181

171 McNamara, J. F. Mathematical programming models in educational planning. Review of Educational Research , 1971, 41, 419-446. McNamara, J. F. Operations research and educational planning Journal of Educational Data Processing , 1972, 9 (6), 1-12. McNamara, J. F. Mathematical programming applications in educational planning. Socio-Economic Planning Science 1973, 7, 19-35. McTarnaghan, R. E. A coordinating board view. In J. L. Wattenbarger &, L. W. Bender (Eds.), New directions for higher education: Improving statewide planning . San Francisco: Jossey-Bass, 1974. Miami-Dade Community College. Miami-Dade Community College catalog 1980-1981 . Miami, Florida: Author, 1979. Midwest Research Institute. PLANTRAN II computer-assisted institutional research and planning . Kansas City: Author, 1972. Millard, R. M. Integrating the strengths of private, public, and proprietary institutions. In E. J. Gleazer, Jr. & R. Yarrington (Eds.), New directions for community colleges: Coordinating state systems . San Francisco: Jossey-Bass, 1974. Miller, J, G. Living systems: Basic concepts. Behavioral Science , 1965, 10, 193-237. Miller, J. G. Living systems . New York: McGraw-Hill, 1978. Modar, J. J., & Elmaghraby, S. E. (Eds.) Handbook of operations research: Models and applications (Vol. 2). New York: Van Nostrand Rinehold, 1978. Modigliani, F. , & Hohn , F. Production planning over time and the nature of expectations and planning horizon. Econometrica . 1955, 23, 46-66. Moos, M. , k Rourke, F. The campus and the state . Baltimore: Johns Hopkins Press, 1959. Morris, P. A. Decision analysis expert use. Management Science . 1974, 20, 1233-1241. Morris, P. A. Combining expert judgments: A bayesian approach. Management Science . 1977, 23, 679-693. Naddor, E. Inventory systems . New York: John Wilev & Sons 1966. '

PAGE 182

172 Nair, K. , & Sarin, R. K. Generating future scenarios — their use in strategic planning. Long Range Planning , 1979, 12 (3), 57-61. National Commission on the Financing of Postsecondary Education. Financing postsecondary education in the United States . Washington, D. C. : U. S. Government Printing Office, 1973. National League for Nursing. State-approved schools of nursing — R.N. , 1975 . (33rd ed. ) New York: Author, 1975. National League for Nursing. State-approved schools of nursing — R.N. , 1976 . (34th ed. ) New York: Author, 1976. National League for Nursing. State-approved schools of nursing— R.N. . 1977 . (35th ed. ) New York: Author, 1977. National League for Nursing. State-approved schools of nursing — R.N., 1978 . (36th ed. ) New York: Author, 1978. National League for Nursing. State-approved schools of nursing— R.N. , 1979 . (37th ed. ) New York: Author, 1979. Naylor, T. H. Integrating models into the planning process. Long Range Planning . 1977, 10 (6), 11-15. Naylor, T. H. , h Byrne, E. T. Linear programming . Belmont, Calif.: Wadsworth, 1963. Naylor, T. H. , & Mansfield, M. J. The design of computer based planning and modeling systems. Long Range Planning . 1977, 10 (1), 16-26. Nemhauser, G. L. Introduction to dynamic programming . New York: John Wiley & Sons, 1966. Optner, S. L. Systems analysis for business and industrial problem solving . Englewood Cliffs, N. J.: PrenticeHall, 1965. Optner, S. L. Systems analysis for business management . Englewood Cliffs, N. J.: Prentice-Hall, 1975. Page, E. B., Jarjoura, D. , k Konopka, C. D. Curriculum design through operations research. American Educat ion Research Journal . 1976, 13, 31-49. ' Palm Beach Junior College. Palm Beach Junior College catal og 1980-1981 . Lake Worth, Florida: Author, 1979.

PAGE 183

173 Panitchpakdi , S. Demand and supply elements in educational planning. Socio-Economic Planning Science , 1977, 11, 339-346. Pasco-Hernando Community College. Pasco-Hernando Community College 1980-81 catalog . Dade City, Florida: Author, 1979. Pensacola Junior College. Pensacola Junior College catalog 1980-1981 . Pensacola, Florida: Author, 1979. Pliner, E. Coordination and planning . Baton Rouge: Public Affairs Research Council of Louisianna, 1966. Polk Community College. Polk Community College catalog 1980 81 . Winter Haven, Florida: Author, 1979. Purga, A. J., III. The impact of enrollment decline upon selected Florida community colleges (Doctoral Dissertation, University of Florida, 1979). Dissertation Abstracts International , 1980, 40, 4438A. (University Microfilms No. 8002889). Randolph, P. H. , & Meeks , H. D. Applied linear optimization . Columbus, Ohio: Grid, 1978. Render, B. Enrollment forecasting in a large state system. Planning for Higher Education , 1977, 6 (1), 21-29. Render, B. , & Shawhan, G. L. Statewide enrollment prediction models: A review and a new approach . Presented at the New Approaches to Planning in Higher Education Conference, Kent, Ohio, May 1974. Ritzen, J. M. Manpower targets and educational investments. Socio-Economic Planning Science , 1976, !£, 1-6. Rodekohr, M. , k Rodekohr, C. A study of the effects of enrollment decline. Phi Delta Kappan . 1974, 57, 621-623. Sachs, W. M. Some thoughts on the mathematical method and futures problems. In H, A. Linstone k W. H. C. Simmonds (Eds.), Futures research: New directions. Reading, Mass. : Addison-Wesley , 1977. Sage, D. E. , k Chobot , R. B. The scenario as an approach to studying the future. In S. P. Hencley k J. Yates (Eds.), Futurism in education . Berkeley: McCutchan, 1974. Salkin, H. , k Saha, J. An introduction to linear programming. In H. Salkin k J. Saha (Eds.), Studies in linear programming . New York: American Elsevier, 1975. Santa Fe Community College. Santa Fe Community College 1980-81 catalog . Gainesville, Florida: Author, 1979.

PAGE 184

174 Schroeder, R. G. , & Adams, C. R. The effective use of management science in university administration. Review of Educational Research , 1976, 46, 117-131. Simonnard, M. Linear programming . Englewood Cliffs, N. J.: Prentice-Hall, 1966. Sposito, V. A. Linear and nonlinear programming . Ames, Iowa: The Iowa State University Press, 1975. St. Petersburg Junior College. St. Petersburg Junior College bulletin 1979-80 . St. Petersburg, Florida: Author, 1978. Sutherland, J. W. A general systems philosophy for the social and behavioral sciences . New York, Braziller, 1973. Tallahassee Community College. Tallahassee Community College catalog 1980-81 . Tallahassee, Florida: Author, 1979. U. S. Department of Commerce, Bureau of the Census. 1970 census of population . (Vol. 1, Part II, Sect. 1). Washington, D. C. : U. S. Government Printing Office, 1973. U. S. Department of Commerce, Bureau of the Census. Directory of federal statistics for local areas: A guide to sources 1976 . Washington, D. C. : U. S. Government Printing Office, 1978. U. S. Department of Health, Education, and Welfare, Division of Nursing. The impact of health system changes on states' requirements for registered nurses in 1985 . DHEW Pub. No. (HRA) 79-8. Washington, D. C. : U. S. Government Printing Office, 1978. U. S. Department of Health, Education, and Welfare, Division of Nursing. Second report to the Congress, March 15^ 1979 (revised). Nurse training act of 1975 . DHEW Pub. No. (HRA) 79-45. Washington, D. C. : U. S. Government Printing Office, 1979. U. S. Department of Labor, Employment and Training Administration. Dictionary of occupational titles . (4th ed. ) Washington, D. C: U. S. Government Printing Office, 1977. ' U. S. Senate. An act to amend the higher education act of 1965. Washington, D. C. : U. S. Government Printine Office, 1972. ^ Valencia Community College. Valencia Community College catalog 1980-81 . Orlando, Florida: Author, 1979.

PAGE 185

175 Van Dusseldorp, R. A., Richardson, D. E., & Foley, W. J. Educational decision-making through operations research . Boston: Allyn and Bacon, 1971. Wiener, N. The use of human beings . Boston: Houghton Mifflin, 1950. Wiener, N. Cybernetics, or control and communication in the animal and the machine . New York: MIT Press and John Wiley & Sons, 1961. Wheelwright, S. C. , & Makridakis, S. Forecasting methods for management . New York: John Wiley & Sons, 1973. Wing, P. Higher education enrollment forecasting . Boulder, Colorado: National Center for Higher Education Management Systems, 1974. Ziegler, W. L. The potential of educational futures. In M. Marien & W. L. Ziegler (Eds.), The potential of educational futures . Worthington, Ohio: C. A. Jones, 1972.

PAGE 186

BIOGRAPHICAL SKETCH Thomas Alan Gay lord was born October 27, 1953, in St. Paul, Minnesota. He graduated from South St. Paul Senior High School in 1972 where he was active in student organizations and athletics. In 1972, Tom was President of the Minnesota Junior Academy of Science. That fall, he matriculated at Luther College in Decorah, Iowa. In 1975, he transferred to the University of Alaska in Fairbanks. He was honored as the outstanding physics student in 1976 by the faculty and was also nominated to Phi Kappa Phi. He graduated with a 3.S. degree in physics in 1976. The summers from 1973 through 1975, Tom worked for Wakefield Seafoods in Sand Point, Alaska. In 1976, he worked for the Alaska Department of Fish and Game collecting data on the salmon run in the Shumagin Islands and Bristol Bay areas . In 1976, Tom entered graduate school at the University of Florida and received the M.S.T. degree in physics in 1977. He entered the doctoral program in educational administration in 1978. Tom was a graduate instructor in physics and meteorology for two years at Florida. He was a graduate research assistant for the Florida Community Junior College Interinstitutional Research Council, the Department of Educational 176

PAGE 187

177 Administration, and the Florida Institute for Educational Linkage Development. He interned at Santa Fe Community College and at the University of Florida's Division of Planning and Analysis. Thomas Alan Gaylord is married to the former Lorrinda Gail Galovin of Sand Point, Alaska. He is a member of the Association for Institutional Research, Phi Delta Kappa, and the World Future Society.

PAGE 188

I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. /ames L. Wattenbarger , i^hairman yprofessor of Educational / Administration and Supervision I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. Professor of Educational Administration and Supervision I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. les F. Burns PlPofessor of Industrial and Systems Engineering This dissertation was submitted to the Graduate Faculty of the Department of Educational Administration and Supervision m the College of Education and to the Graduate Council, and was accepted as partial fulfillment of the requirements for the degree of Doctor of Philosophy. August, 1980 Dean, Graduate School

PAGE 189

I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. ^JSxnes L. Svattenbarger , jChairman professor of Educational Administration and Supervision I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. Professor of Educational Administration and Supervision I certify that I have read this study and that in my opinion it conforms to acceptable standards of scholarly presentation and is fully adequate, in scope and quality, as a dissertation for the degree of Doctor of Philosophy. ;s F. Burns PJ^ofessor of Industrial and Systems Engineering This dissertation was submitted to the Graduate Faculty of the Department of Educational Administration and Supervision m the College of Education and to the Graduate Council, and was accepted as partial fulfillment of the requirements for the degree of Doctor of Philosophy. August, 1980 Dean, Graduate School



PAGE 1

A PSYCHOMETRIC EVALUATION CF THE CORRECTIONAL ADJUSTMENT CHECKLIST BY BRAINARD WILLEM HINES A DISSERTATION PRESENTED TO THE GRADUATE COUNCIL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DOCTOR OF PHILOSOPHY DEGREE UNIVERSITY OF FLORIDA 1980

PAGE 2

ACKNOWLEDGMENTS My sincerest thanks go to the members of my doctoral committee for their patience and understanding. Particularly, I would like to thank the chairman of my committee, Dr. William Ware who has been a good friend and constructive influence throughout my academic career at the University of Florida. I also owe a special debt of gratitude to Dr. Linda Crocker and Dr. Richard Swanson for their support and encouragement. In addition, I would like to express my gratitude to the University of North Carolina at Chapel Hill for allowing Dr. Ware to continue as my chairman for the past months. Finally, I would lilse to express my appreciation to my wife Magdalena Llabre, whose love and encouragement made this dissertation possible.

PAGE 3

TABLE OF CONTENTS Page ACKNOWLEDGMENTS iii LIST OF TABLES V ABSTRACT vii Chapter I. INTRODUCTION 1 Psychometric Properties Investigated ... 4 Definitions of Reliability 4 Definitions of Validity 5 Construct Validation ... 7 Statement of the Problem 8 Significance of the Study 10 II. REVIEW OF THE LITERATURE 12 Classification 13 Classification of Criminals 15 Empirically Derived Typologies 19 Current Reviews of Criminal Typologies . 21 Psychometric Concepts 30 Reliability Estimation in This Study . . 37 Validity 41 Types of Validity 44 Construct Validity Estimates 46 Chapter Summary 51 * III. METHOD 53 The Sample 53 Selection of the Sample 55 Instrumentation 59 Data Collection 66 Data Analysis 69 Reliability of the CACL 69 Predictive Validity of the CACL 71 iii

PAGE 4

Construct Validation of the CACL Postdiction of Crime Type . . . 73 74 Summary 75 IV. RESULTS 7 7 Inter-Rater Reliability of the CACL .... 78 Construct Validation of the CACL 79 Criterion-Related Validity of the CACL . . 85 Relationship of the CACL to Crime .... 91 Summary 9 3 V. DISCUSSION 94 The Inter-Rater Reliability of the CACL . . 96 Construct Validation of the CACL 96 Criterion Validity of the CACL 101 Suicide Attempts 102 Threats of Assault 104 Assaults 105 Infractions of Rules 105 Relation of the CACL to Crime Type 106 Summary of Psychometric Evaluation 107 Recommendations 110 Appendix A. CORRECTIONAL ADJUSTMENT CHECKLIST 113 B. SUMMARY TABLES FOR INTER-RATER RELIABILITY STUDIES . 116 REFERENCES 12 3 BIOGRAPHICAL SKETCH 132 iv

PAGE 5

LIST OF TABLES TABLE PAGE 1. NUMBER OF RESIDENTS BY UNIT ADMITTED TO NFETC FROM ITS INCEPTION UNTIL JULY 1, 1978 58 2. DESCRIPTIVE STATISTICS FOR CONCURRENT VALIDITY STUDY 80 3. INTERCORRELATION MATRIX FOR CONCURRENT VALIDATION STUDY 81 4. RESULTS OF CANONICAL CORRELATION ANALYSIS OF THE CACL AND MMPI 8 3 5. CANONICAL WEIGHTS OF MT^PI AND CACL SUBTESTS FOR CANONICAL VARIATES 1 AND 2 . . . . 8 4 6. PRODUCT MOMENT CORRELATION BETWEEN SUBTESTS OF THE CACL AND MMPI AND CANONICAL VARIATES 8 6 7. DESCRIPTIVE STATISTICS FOR PREDICTIVE VALIDITY STUDY 8 7 8. INTERCORRELATION MATRIX FOR PREDICTIVE VALIDITY STUDY 88 9. RESULTS OF MULTIPLE REGRESSION OF FREQUENCY OF SUICIDE ATTEMPTS ON CACL SUBSCALES 89 10. RESULTS OF MULTIPLE REGRESSION OF FREQUENCY OF ASSAULTS ON CACL SUBSCALES 89 11. RESULTS OF MULTIPLE REGRESSION OF FREQUENCY OF THREATS OF ASSAULT ON CACL SUBSCALES 90 12. RESULTS OF MULTIPLE REGRESSION OF FREQUENCY OF INFRACTIONS ON CACL SUBSCALES 90 V

PAGE 6

13. RESULTS FOR DISCRIMINANT FUNCTION ANALYSIS OF CRIME TYPE .... 92 APPENDIX 14. DESCRIPTIVE STATISTICS FOR INTERRATER RELIABILITY STUDY: "INTAKE" CONDITION 116 15. DESCRIPTIVE STATISTICS FOR INTERRATER RELIABILITY STUDY: "CONTROLLED =' CONDITION • . 116 16. ANALYSIS OF VARIANCE TABLE FOR CACL PA "CONTROLLED" CONDITION INTERRATER RELIABILITY STUDY 117 17. ANALYSIS OF VARIANCE SUMTIARY TABLE FOR CACL ID "CONTROLLED" CONDITION INTER-RATER RELIABILITY STUDY 117 18. ANALYSIS OF VARIANCE TABLE FOR CACL NA "CONTROLLED" CONDITION INTERRATER RELIABILITY STUDY 118 19. ANALYSIS OF VARIANCE TABLE FOR CACL MA "CONTROLLED" CONDITION INTERRATER RELIABILITY STUDY 118 20. ANALYSIS OF VARIANCE SUMMARY TABLE FOR CACL PA "INTAKE" CONDITION INTER-RATER RELIABILITY STUDY 119 21. ANALYSIS OF VARIANCE SUMMARY TABLE FOR CACL ID "INTAKE" CONDITION INTER-RATER RELIABILITY STUDY 119 22. ANALYSIS OF VARIANCE SUMMARY TABLE FOR CACL NA "INTAKE" CONDITION INTER-RATER RELIABILITY STUDY 120 23. ANALYSIS OF VARIANCE SUM^^IARY TABLE FOR CACL MA "INTAKE" CONDITION INTER-RATER RELIABILITY STUDY 120 24. INTER-RATER RELIABILITY COEFFICIENTS FOR INTAKE AND CONTROLLED CONDITIONS, INCLUDING SYSTEMATIC RATER BIAS IN THE ERROR TERM 121 vi

PAGE 7

Abstract of Dissertation Presented to the Graduate Council of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Doctor of Philosophy A PSYCHOMETRIC EVALUATION OF THE CORRECTIONAL ADJUSTMENT CHECKLIST BY Brainard Willem Hines June 1980 Chairman: Professor William B. Ware Major Department: Foundations of Education A variety of classification systems have been developed for use in the social sciences. These systems have become increasingly complex with the advent of modern statistical techniques. The use of classification systems in the field of criminology began with the effort to discriminate between criminals and "normals" on the basis of physical features. Although more recent classification systems in the field of criminology attempt to define types of criminals or criminal behavior, few have been adequately evaluated in terms of the psychometric properties of reliability and validity . The Correctional Adjustment Checklist (CACL) is a factor-analytically derived classification instrument which is designed to describe the behavior of incarcerated males vii I

PAGE 8

along four dimensions. These dimensions have been labelled Psychopathic-Aggressive (PA) , Neurotic-T^xious (NA) , Immature-Dependent (ID), and Manipulative (Ma). Ratings along these dimensions are intended to have differential implications for the management and treatment of individuals in close confinement. Although the instrument has been used in a variety of settings, little information is available on its inter-rater reliability or validity. This study attempted to evaluate the CACL by assessing the degree of congruence among raters in a naturalistic setting and under conditions designed to provide maximal reliability estimates. Data gathered under both conditions provided reliability estimates for the average of three raters which ranged upwards of .60, with the exception of the Ma subscale, which showed a lower inter-rater reliability estimate in the "controlled" condition. Assessment of the validity of the CACL with this example involved estimating the relationship between it and other variables of interest. In this study, these variables were as follows: scores on the MMPI administered concurrently with the CACL; the frequency of several types of disruptive behavior during the first sixty days after the CACL was administered; and the degree of violence involved in the crime with which the subjects had been most recently charged. viii

PAGE 9

These estimates of the CACL's relationship with other variables are statistically significant in several instances. First, a canonical variate analysis derived two sets of variables from the CACL and f^MPI. Although both canonical correlation coefficients are significant at the .05 level, a redundancy analysis indicates that the relationship between the two instruments is very modest. Also, scores on the CACL showed a statistically significant relationship to suicide attempts and threats of violence which occurred within the first sixty days. When the subscales of the CACL were used as the predictors in a multiple regression analysis, the NA subscale showed the highest degree of association with suicide attempts, followed by the PA subscale. Additionally, the PA subscale is the only subtest which accounts for a statistically significant amount of variance in verbal threats of physical violence. Other disruptive behaviors (actual assaults and other infractions) were not significantly related to scores on any of the CACL's subscales. A discriminant function analysis did not show any significant relationship between subscale scores on the CACL and the presence of physical violence in the subjects' most recent crime. This may have been due to the imprecise match between the charge (e.g., armed robbery) and the actual degree of violence in the crime. ix

PAGE 10

In summary, the CACL provides subscale scores which are reliable across raters, and which predict several behaviors of interest within a maximum security mental hospital setting. It shows a modest degree of redundancy with the MMPI, indicating that it may well be measuring factors not being tapped by that instrument. Although the CACL was developed for the classification of a general prison population, it appears to have utility when used with individuals who are emotionally disturbed. X

PAGE 11

CHAPTER I INTRODUCTION The process of classification of objects or events is essential to the development of any science. Although the origin of taxonomy or classification goes back to the ancient Greeks, the advent of modern statistical techniques and the use of high-speed computers have allowed for the use of more sophisticated classification methods than have been previously possible. In many areas, such as entomology, the development of more sophisticated taxonomic systems has contributed to the general advancement of the field in question. The use of classification in criminology can be traced to a number of physiologists such as Lambroso s who, in the nineteenth century, attempted to define a "criminal type" on the basis of physical features. Such typologies were intended to discriminate between criminals and "normals," not to classify types of individuals who had been convicted of crimes nor relate those types to other measures of any sort. With the modern emphasis on treatment rather than custody of criminals has come a concern for possible subtypes of offenders. Additionally, the more 1

PAGE 12

humanitarian philosophy of our own era has encouraged the scientific study of criminal behavior and the types of individuals who become criminals. Such efforts to create meaningful offender typologies (Gibbons, 1975) have come about because of the failure of unitary treatment approaches and because of the observed variance in types of crimes, demographic and personal or behavioral characteristics of criminals. Although a variety of offender typologies have been proposed, none has been widely accepted. Many typologies either are based on traditional psychological personality types or are concerned with classifying offenders based on the type of crime which they have committed. Other classification systems have been impressionistic and have included a variety of types which have not been found by other investigators (Gibbons, 1975). Generally, no single typology of criminals or criminal behavior has been found to be of use in a variety of settings or with various age groups of individuals. Also, no single typology has been constructed which is of use in delineating both the etiology and diagnostic category of criminal behavior. Most criminal typologies are based on the results of a single instrument or a series of descriptions of the crime or its etiology. Few classification systems used in criminology are based on empirically derived methods, but rather are derived from theoretical formulations. Quay

PAGE 13

1 I 3 (1971) has made one of the few attempts to empirically construct a classification system for criminals. The Quay Correctional Adjustment Checklist (CACL) is an instrument derived using factor analysis for the purpose of describing the behavior of incarcerated individuals on four dimensions (Quay, 1971) . These dimensions are labelled Psychopathic-Aggressive (PA) , Neurotic-Anxious (NA) , Immature-Dependent (ID) , and Manipulative (Ma) . It is intended not only to describe an individual's patterns of behavior within an institution, but also to provide information useful for differential treatment based on those patterns. Although the CACL has been used in a variety of settings, its psychometric properties have never been thoroughly investigated. A review by Warren (1969) reported that the CACL appeared to have "adequate" reliability but gave no source for that statement. Another article by Quay (1971) has called for further study of the Checklist, but gave no validity estimates for the instrument. The purpose of this study is to investigate the inter-rater reliability and the validity of this instrument, based on the behavior of a sample of individuals who have been confined in a maximum security mental hospital in Gainesville, Florida. All of those individuals have either been convicted of a felony, have been found incompetent to stand trial for a felony, or are not guilty

PAGE 14

1 4 by reason of insanity. Data on the CACL have never been gathered on a psychiatric population (Quay, personal communication, 1978). If adquate reliability and validity estimates are obtained for the sample, the CACL should be used in other such settings. Psychometric Properties Investigated Definitions of Reliability As Kerlinger and Pedhazur (1973) pointed out, there are a variety of definitions of reliability. In general, they defined reliability as the consistency and accuracy of an instrument which are related to the absence of random or error variance in that instrument. Specifically, he wrote that ". . . reliability can be defined as the relative absence of errors of measurement in a measuring instrument" (p. 443) . The reliability of any measure can be thought of as existing in any of several dimensions. These dimensions may involve consistency (freedom from measurement error) across time , across items in a single measure, across other forms of the test or across raters or scorers on the same form of the test. The types of reliability which correspond to the degree of consistency in each of these dimensions are known as test-retest (time) , internal consistency (items) , parallel forms (forms) and inter-rater reliability. The central focus of this study is the consistency of the CACL scores across raters who have observed the J

PAGE 15

I 5 individual under similar circumstances and who have received similar training in the use of the instrument. This type of reliability (inter-rater) is of paramount importance to the CACL, since it is intended to measure the presence of observable behaviors. If equally trained observers cannot agree on whether a particular behavior is present, then usefulness of this instrximent for any practical purpose is highly questionable. Definitions of Validity The Standards for Educational and Psychological Tests and Manuals (1974) , published by the American Psychological Association, stated: Validity information indicates the degree to which the test is capable of achieving certain aims. Tests are used for several types of judgment, and for each type of judgment a different type of investigation is required to establish validity, (p. 13) That is, validation is defined as a process or activity performed on the data arising from a test. The manner in which the data are treated is intended to parallel an aspect of the intended use of the measure, or of its interpretability. The Standards publication goes on to list three aims of testing which correspond to three types of validation procedures . 1. The test user wishes to determine how an individual performs at present in a universe of situations that the test situation is claimed to represent.

PAGE 16

6 2. The test user wishes to forecast an individual's future standing or to estimate an individual's present standing of some variable of particular significance that is different from the test. 3. The test user wishes to infer the degree to which the individual possesses some hypothetical trait or quality (construct) preserved to be reflected in the test performance. (p. 13) The American Psychological Association and the American Educational Research Association have defined three basic types of validity: content, construct, and criterionrelated (APA, 1974) . Among these three types, criterionrelated and construct validity are most appropriate in assessing the potential usefulness of the CACL. Criterion-related validity encompasses both predictive and concurrent validity which reflect the correlation between scores on a test and performance on a criterion variable. In concurrent validation, measures on both test and criterion are obtained at approximately the same point in time; predictive validation occurs when the criterion measure is taken after the test in question. For this study, predictive criterion-related validity will be investigated by relating CACL scores at the time of intake to the frequencies of several types of disruptive behavior, recorded during the first sixty days of confinement at the hospital.

PAGE 17

7 Construct Validation Construct validation is a complex procedure attempting to ascertain the degree to which a measure empirically relates to a number of other variables which logically and deductively derive from the construct which the instrument purports to measure. To do this, a construct validation study often involves an attempt to demonstrate that the trait is related to other variables which are logically inherent from the construct, and that variables which do not logically derive from the construct do not empirically relate to it. In the same way, this study will assess the degree to which the CACL's subscales relate to other variables which logically derive from the constructs they purport to measure. That is, we would expect that individuals who score highly on the Psychopathic-Aggressive subscale would be more violent and disruptive than those who score highly on the Immature-Dependent subscale. In addition, they should more often take a leadership role and rely less on staff for advice than other types of individuals. Also, we would expect such individuals to be more frequently threatening to others, and to score more highly on those subscales of another instrument which measure impulsivity and hostility. If such relationships were evident, this would help in defining the nature of the basic traits being assessed by the CACL.

PAGE 18

The analyses conducted in this study provide estimates of the relationship between the CACL and several behaviors which are of interest in the institution. They also provide an estimate of the nature and degree of relationship between the CACL and the Minnesota Multiphasic Personality Inventory, a self-report diagnostic instrument which has been used in other criminal classification systems. Statement of the Problem This study will be addressed to determining the psychometric properties of the Quay Correctional Adjustment Checklist based on the performance of a sample of individuals in a maximum security mental hospital. It will not address the decision rules used to classify individuals, but will deal only with the psychometric properties of the instrument. Specifically, it will attempt to answer the following questions: a 1. What is the degree of agreement among raters with similar training who rate individuals on each of the four subscales of the CACL? (Inter-rater reliability) 2. What is the degree of association between the subscales on the CACL and the subscales of the MMPI, when both instruments are administered concurrently? (Construct validity) 3. Is there a relationship between the various subscales of the CACL and the type of crime which caused the individuals' incarceration? (Construct validity)

PAGE 19

4. What is the relationship between scores on the CACL and an index of disruptiveness within the institution? (Criterion-related and Construct validity) The questions above are concerned respectively with inter-rater reliability and validity, both of which are important to the use of the CACL in institutional settings. It is important to note that the study is not designed to explore the use of particular decision rules for classification. Rather, it is concerned with the consistency and "interpretability " of scores on the CACL. Also, the reliability estimates which are given are for the average of three raters, where each rater rates all individuals. These estimates are considerably higher than those which would be obtained for a single rater. At a time when there is a clear need for adequate classification techniques in the area of corrections (Warren, 1969) , the CACL presents some unique advantages and disadvantages. Although it has been used in a variety of settings, often for the purpose of classification for treatment, its psychometric properties remain for the most part unknown. If its reliability and validity are low, its use should be discontinued. If they are acceptable, the instrument's utility could be explored in other settings. These possibilities are further discussed in the next section.

PAGE 20

10 Significance of the Study As Gibbons (1975) has pointed out, there is increasing dissatisfaction with the process of classification in criminology and criminal justice. Many of the current problems with the medical model, often used in criminal classification, may well be due to the ineffectiveness of current diagnostic instruments and the consequent misclassif ication of many individuals. Despite the apparent failure of treatment strategies which presume a single type of offender, no type of classification system other than the CACL has evolved which is based on actual patterns of behavior in an institution. Quay (1971) noted that: "Additional research with respect to reliability and construct validity (of the CACL) is in order" (p. 11). Although this need has been recognized since the initial development of the instrument, such studies have not been forthcoming. Despite the fact that there has never been adequate assessment of the instrument's psychometric properties, the CACL has been used in a variety of institutions, including the Robert Kennedy Federal Youth Center in Morgantown, West Virginia, the North Florida Evaluation and Treatment Center in Gainesville, Florida, and the Federal Correctional Institution in Miami, Florida. Studies such as this one are important for several reasons. First, if the instrument misclassif ies offenders.

PAGE 21

11 it may be hindering their effective treatment. Such misclassification is unfair to individual offenders and to the society which supports such treatment efforts. Second, the continued use of an instrument with unknown psychometric properties may well contribute to the increasing disenchantment with differential treatment strategies for offenders. Third, although the instrument may have an important function in the derivation of new theories in criminology, such functions will be of little use until validity is established. Additionally, the instrument may have potential use in the assessment of treatment effectiveness for individuals as well as groups of offenders. If it provides a reliable and valid measure of institutional behavior, it could allow for improved monitoring of those behavioral changes which occur during incarceration. In this chapter an overview of classification as a process has been presented,' and the application of this process to the field of criminoloty has been summarized. The psychometric properties of the C ACL, which are investigated in this study, are summarized along with the implications of the study. The following chapter includes a review of the literature on classification as a logical process, its use in criminology, and on the psychometric properties of reliability and validity.

PAGE 22

CHAPTER II REVIEW OF THE LITERATURE This study is intended to provide reliability and validity estimates for the CACL, based on the behavior of a sample of individuals incarcerated in a maximum security mental hospital. Since the CACL is an instrument intended to classify criminals, the literature review first includes a summary of articles on classification as a field of study in its own right. This is followed by a more extensive review of the development of criminal typologies. The trend towards empirically developed rather than theoretically oriented systems is discussed. Next, the development of the CACL is outlined, and the instrument is compared with other classification systems which are based on self-report instruments rather than behavioral observation. The need for psychometric evaluation of the CACL is pointed out, and the importance of inter-rater reliability and further validation studies is emphasized. The final section of the literature review provides a brief review of the theoretical definitions of reliability and validity as they pertain to the CACL. The need for consistent rating of individuals is stressed, along with the need for meaningf ulness and utility in the classifications which are derived from the instrument' 12

PAGE 23

use. Thus, the inter-rater reliability and construct validity of the CACL are the areas of primary interest which are explored in this study. Although the criterionrelated validity of the CACL is also investigated, the relationship of the criterion variables to the constructs measured by instriiment is also explored. Classification Classification may be defined as the arrangement of objects or events into sets on the basis of their common characteristics. This process has been part of the natural sciences for centuries, but has only recently become a field of study in its own right. As such, the term taxonomy has been used to mean the theoretical study of classification as it occurs in a variety of specific disciplines. One of the most comprehensive reviews of taxonomy appeared in 19 74 in which Sokal reviewed the purposes, development, and structure of any classif icatory activity. This review covered several general areas which are applicable to the use of classification in criminology, particularly the criteria for a desirable taxonomic system and the major purposes of classification. Sokal (1974) said that. The paramount purpose of a classification is to describe the structure and relationship of the constituent objects to each other and to simplify these relationships in such a way that general statements can be made about classes of events, (p. 1116)

PAGE 24

14 Implicit in this definition are several purposes of taxonomy. First, classification may be used to reveal the "true" relationships between objects or events by ordering them on the basis of common characteristics. Second, classification can be used to achieve economy of memory. By grouping single cases it provides the capacity to sximmarize information and to avoid repetition. Third, classification provides for ease of manipulation and facilitates information retrieval. It may be used to simplify problems in routing or delivery, to define political districts or to allow for cataloging printed materials. Finally, Sokal noted that classification systems have the primary scientific purpose of generating hypotheses, in that they should "stimulate interest as a means of furthering investigation" (p. 1117) . Sokal made several important points about the purpose and types of classification systems. He noted that classia fication systems may serve the purpose of economy of memory, reveal "natural" relationships between elements in each taxon, provide for ease of manipulation, and generate interest in new scientific problems. Classification systems may vary in the number of salient dimensions which they include and may be monothetic or polythetic in nature, depending on whether the elements in each taxon must share a common trait (in the former case) or whether an element may possess any combination of the traits (in the latter case) .

PAGE 25

15 In general, Sokal pointed out that the classification is emerging as a distinct discipline and that a "metatypology" or classification of classification systems is possible. An ideal classification system should accommodate all elements of the set of objects or events to be classified, and should enable the typologist to match the dimensions and specificity of the system to its intended use. These criteria will be used later in this study to evaluate the CACL as a typological instrument in the field of criminology. Classification of Criminals An excellent overview of the classification of criminals is provided by Schafer (196 8) , who not only provided a historical narrative of the major typologists, but also defined several categories of criminal typologies (pp. 14314 4) . These include legal typologies of crime type, multiple 5 cause typologies, typologies based on sociological or psychological theories, typologies which stress physiological factors, and those which describe the longitudinal development of criminal behavior. Included in this section are an overview of current criminal typologies which fall into these categories and an explanation of the relationship between the CACL and these typologies. Generally, it is of interest that the CACL does not fall readily into any of Schafer 's categories, since it deals with the behavior of felons while

PAGE 26

1 16 incarcerated rather than the longitudinal development of criminal behavior. The categories developed by Schafer emphasize either the hypothetical "causes" of criminal behavior, the type of crime (s) committed, or attempt to relate the two in a single description of a criminal "role career." Although some typologies are most often used for reporting frequencies of particular criminal acts in a given geographic area and are thus primarily empirical, most of the other typologies reflect on underlying theory of the causes of criminal behavior. Legal typologies represent monothetic classification systems in which crimes rather than criminals are classified. The FBI Uniform Crime Reports for the United States represent such a system. Schafer noted that although such systems have historical and legal interest, "They are technical divisions for the use of the administration of of justice and are not conceived of as explanations for behavior" (p. 146) . Such a typological system will be used in this study to relate crime types to CACL classifications. Multiple-cause typologies stress the interaction of biological, social and psychological causes of criminal behavior. Such systems were first developed in Germany in the nineteenth century by theorists who emphasized the affective and motivational components of criminal behavior as they exerted their influence across the criminal's life

PAGE 27

17 span. This historical perspective was later used by Gibbons (1965, 1970) in his typology of criminal role careers. Gibbons, along with Clinnard and Quinney (1967) based a major criminological text on a multiple-cause typology. However, in a later article, Gibbons (1975) has emphasized the difficulties in the use of such a typology. Sociological and psychological typologies both emphasize the hypothetical causes for criminal behavior. Sociological typologies attempt to delineate the external forces which contribute to criminal behavior, while psychological classification systems reflect the inner dynamics which may lead to such acts. Although these typologies reflect some of the most productive and extensive areas of offender classification, they also elicit some of the more vehement opposition (Schafer, 1968, p. 155) . Two of the most prominent sociological typologies are those of Tappan (1967) and Thrasher (1963). These two individuals have attempted to relate criminal behavior to its social causes, but frequently derived hypothetical multiple factor typologies with little empirical verification (Schafer, 1.968). This is a general problem with etiological typologies, since they limit validation studies to ex post facto designs. Psychological typologies are exemplified by the work of Alexander and Staub (1956) and Abrahamson (1960). Such

PAGE 28

• 1 18 systems suffer from some of the same problems as those depending on more sociological explanations, since they are limited to the assessment of current psychological functioning. Current functioning in criminals may not reflect the dynamics in operation at the time the crime was committed. Constitutional typologies have the most lengthy history in criminology, dating to Galen (circa, 150 A.D.). This group of typologies centers around the biopsychological causes of crime, especially the morphology of the offenders. The works of Lambroso (1911) , Kretschmer (1925) , and Sheldon (1949, 1954) are typical of this area. Although these authors consistently have tried to relate body type to crime type, George Vald (1958) points out that "there is no present evidence at all of physical type, as such, having any consistent relation to legal and sociologically defined crime" (p. 129). Thus, constitutional typologies may" allow for the classification of criminals, but the resultant categories have no empirical relationship to any manifest behavior. This problem is not unique to constitutional typologies, as will be shown later. Normative typologies attempt to define the criminal's total personality in an effort to identify the "types" for which a particular sentence is appropriate. As such they incorporate a variety of legal, sociological, and psychological typologies. German authors have been primarily responsible for work in this area which has been little used in America.

PAGE 29

Life-trend typologies are similar to multiple-facet typologies, but stress the dynamic structural coherence of the individual criminal's way of life. They are typically more complex than Gibbon's "role-careers" in that they attempt to follow the criminal behavior which is not part of a criminal life style. Authors such as Reckless (1967) and Clinnard (1963) have developed systems of this type and have generally made a large impact in the field of criminology (Schafer, 1968, p. 6) . This may well be due to the comprehensiveness of the system itself and the polythetic, multi-dimensional process which they use to classify offenders. Empirically Derived Typologies Given the problem inherent in the classification systems previously discussed, individuals such as Quay (1964), Gibbons (1975.) and Megargee (1977) have recommended the use of empirically derived typologies. In such systems criminals are grouped on the basis of current behavior or demographic variables, without first theorizing about the causes for antisocial behavior. Accordingly, the development of such typologies differs from that involved in more "theory-oriented systems." The effort to develop empirically keyed classificatory systems involves the administration of an instrument to a group of offenders and the development of a classificatory system based on the results of that instrument.

PAGE 30

The important distinction to be made is that such typologies assess the current responses of the individual and are of limited scope and purpose. They are designed primarily for their immediate rather than long-term utility value, and may not have a predetermined underlying construct which they attempt to measure. Examples of such instruments are the classif icatory subscales of the MMPI developed by Panton (1965, 1966, 1968, 1970); Quay's work on the CACL (1971); and Megargee ' s recent work (1977), which also uses items on the MMPI. These classif icatory techniques have several common elements: first, they are not based on a single etiological or explanatory construct; second, they use a single instrument or subscale of an extent measure; and third, they describe current levels of functioning of the individual. These instruments are usually constructed by relating items to an external criterion (i.e., behavior in the institution) or to an internal criteria of factorial homogeneity, as is the case with the CACL and the Megargee MMPI system. The CACL was developed by Quay as part of such an empirically derived classification system. Basing his classification system on the techniques developed by Kewitt and Jenkins (1946) , Quay (1971) developed instruments assessing both current functioning (the CACL) and life history (CALH) .

PAGE 31

21 Although the development of the CACL will be discussed in greater detail in the next chapter, it should be mentioned here that the CACL is a behavioral checklist intended to assess patterns of current functioning while incarcerated. The instrument was normed on a prison population, and provides normalized T scores in each of four dimensions: Psychopathic-Aggressive (PA); Neurotic-Anxious (NA) ; Immature-Dependent (ID) ; and Manipulative (Ma) . It is of interest that the CALH, a life-history checklist also groups criminals into these categories, and provides a "situational" dimension, where the CACL does not. Current Reviews of Criminal Typologies Megargee (1977) considered both the substance and form of a taxonomic system for offenders. He listed seven criteria for "usefulness" (p. 108) of such a classification scheme which are ap follows: 1. The system should classify all of the offenders under consideration. 2. It should have clear operational definitions of types . 3. It should be reliable, especially across raters. 4. It should be valid (construct validity is implied). 5. It should be dynamic, reflecting changes in the individual. 6. It should carry implications for treatment. 7. It should be economical to administer.

PAGE 32

22 These criteria do not stress the "theory building" function of such a systein stressed by Sokal (1974), Schafer (196 8) and others, but rather emphasize the practical significance of the system. In this, Megargee is following the point of view expoused by Gibbons (1975) in moving away from theoretically oriented taxonomic systems. Commenting on the CACL diagnostic system, Megargee said: "Systems (such as the CACL) can reflect changes in the individual and typically have clear implications for differential treatment strategies" (p. 110) . Later, he stresses the training and supervision of raters necessary to the CACL system, and said that the development of his MMPI system was intended to "retain the advantages of the Quay . . . system land to be], . . . widely implemented with less cost and fewer trained personnel" (p. 110) . However, he is equating a system based on a self -report device intended for psychiatric classification with a more direct system of behavioral monitoring. Thus, the comparison does not seem adequate in its inference that both are based on "personality characteristics of the offender" (p. 112) , except in the broadest sense. Examples of various typologies have been also discussed in a review of Warren (1969) and in the proceedings of an NIMH conference on criminal typologies (1967) . Both of these reviews group typologies differently than does Schafer and elaborate other characteristics than those which he emphasized .

PAGE 33

Warren discussed five groups of offender typologies which provide the background for her own classification system (p. 241) . These typologies include the following: 1. Prior probability systems, which rank offenders on the expectancy of some future behavior, usually recidivism. 2. Reference group typologies, relating criminal behavior to the social norms of a specific group. 3. Behavior classifications, which are oriented to some aspect of the offender's behavior. 4. Psychiatrically-oriented approaches which seek to define the nature of any mental disorder underlying crime. 5. Social perception and interaction systems. Such typologies relate criminal behavior to specific social interactions, and to the criminal's perceptions of those interactions . s It is obvious that these groupings are poorly defined and that they frequently overlap, as Warren has admitted (p. 241) . The reviewer continued to make several more valid points about the structure and function of offender typologies. Generally, Warren made the point that "each of the . . . classification systems is not equally relevant for all purposes" (p. 242) . Warren saw typologies as serving the purposes of either "management" or "treatment" (p. 242). She said,

PAGE 34

It is possible for certain purposes to use a classification system which . . . has no etiological reference, one which has no implications for treatment, or one which is specific to an institutional setting. (p. 243) Her review, like others, pointed out the difficulties with typologies which emphasize etiological dimensions, and argued for the use of more effective systems for treatment. In Warren's view, any combination of several factors may have caused the crime and it is necessary to specify the exact cause in order to change the behavior (p. 243) . This view, although widely shared, has not led to more effective treatment of criminals. Warren reviewed several studies which indicated that no form of differential treatment has effectively reduced recidivism rates (p. 245) . Despite this fact. Warren remained optimistic that adequate typologies will reveal that treatment outcomes depend on characteristics of the offender which interact with characteristics of the treatment program. It is not surprising that her interpersonal maturity system is oriented to such a purpose. Unfortunately, no later articles have been published which report on whether her system was more effective than others. Warren (1969) summarized the results of an NIMH study (1967) which attempted a cross-tabulation of many existing classification systems, including that of Quay. The resulting configuration of typologies or composite system revealed six "bands" which were judged to represent a

PAGE 35

stable set of underlying characteristics of offenders (Warren, 1969, p. 249). The six categories coimnon to the sixteen classification systems reviewed are as follows 1. Band 1 labeled the asocial type, included the CACL Psychopathic-Aggressive type. Such individuals are characterized as "primitive, underinhibited , impulsive, hostile, insecure, inadequate, maladaptive, demanding of immediate gratification and attention, thoroughly egocentric, etc." (Warren, p. 251). 2. Band 2 labeled the conformist type, incorporated the CACL Immature-Dependent type. Persons in this band are characterized as "concerned with power, searching for structure, dominated by the need for social approval, rule-oriented, unable to empathize, having low, selfesteem" (Warren, p. 251) . 3. Band 3 labeled the antisocial manipulator , included the CACL Manipulative type. These offenders are described as "guilt-free, power-oriented, self-satisfied, non-trusting, emotionally insulated, cynical . . . and extremely hostile" (Warren, p. 252) . 4. Band 4 identified as the neurotic subtype, including the CACL Neurotic-Anxious type. Such individual are characterized by high levels of anxiety and are described as "intimidated, disturbed, anxious, depressed, and withdrawn" (Warren, p. 254) .

PAGE 36

5. Band 5 labelled as the subcultural identifier . Such individuals are presumed to coininit their crimes because of their integration of subcultural values conducive to crime. Individuals of this type are described as "loyal to their group, psychologically healthy, proud, adequate, suspicious of the authority system, having a stable family, have criminal attitudes, and accessible to new experiences" (VJarren, p. 254). 6. Band 6 labelled the situational offender . This grouping included the CALH situational type, and is characterized as "relatively normal, exposed to acute, severe stress, having no evidence of neurosis, having little prior criminal records, etc." (Warren, p. 255). Such persons are seen as reacting to an overwhelming, non-recurring emotional stress which led to committing their crime. Unfortunately, these "bands" or subtypes were identified by an informal comparison rather than on the basis of the measurement of a heterogeneous group of offenders with the same group of classif icatory instruments. That is, the bands were constructed intuitively rather than empirically. However, since the various classification systems were developed independently, it is possible that this consensus reveals the existence of separate constructs which differ across the bands. As Warren noted (p. 245) , until an empirical study is done on a single population, the diagnostic bands described above will remain somewhat hypothetical and tentative.

PAGE 37

Generally, Warren used this review to provide background for her own diagnostic system, but she made several points pertinent to criminal classification as it exists today. She pointed out that "the classification systems are not equally relevant for all purposes" (p. 241) . In addition, this review indicated that an ideal typology would provide "an explanatory theory with the resulting aid to prediction, implications for management and treatment, greater precision for research" (p. 240) . Thus it does seem that Warren believes that a single system can meet these needs. As Gibbons (1975) pointed out, there is an increasing disenchantment with all of the taxonomic systems described above. Most have failed to show any real usefulness in the treatment of criminal behavior. Although the authors of these systems have hoped for empirical verification of their systems, little evidence has been forthcoming. Even though these systems have stimulated some new research, and do provide several of the benefits outlined by Sokal (1974) , they have failed to show pragmatic usefulness (Schafer, 1968 , p. 177) . Gibbons (1975) was also pessimistic about usefulness of current offender typologies. He said: "It is by no means clear that existing typologies are empirically precise" (p. 254) . The reasons for this lack of clarity are several, according to Gibbons (p. 299) . Firstly, no

PAGE 38

28 single typology subsumes all types of criminality. Secondly, new forms of lawbreaking may be emerging which do not fit traditional typologies. Thirdly, the patterns of behavior or etiology which most typologies hypothesize have yet to be found in the actual study of offenders. Gibbons argued that this lack of satisfactory classification systems is due either to the faults in the systems themselves or to the possibility that criminal behavior develops in a unique manner in each individual. It is difficult to assume the latter case, however, until the former has been eliminated as a potential problem. He concluded by noting: Insofar as the search for typologies turn out to be profitable in corrections, it will be as a consequence of the further development of statistical classifications . . . Iwhich involve] . . . the development of classif icatory devices based on specific groups of offenders within certain limited correctional settings. (p. 245) Thus Gibbons recommended turning away from theoretically derived typologies, especially those which center on the etiology of criminal behavior. His discussion indicated that the more any typology depends on retrospective investigation or hypothetical constructs, the less likely it is to produce meaningful results. In summary, the literature on classification systems for offenders seems to support several overall trends. First, no single system can perform all of the functions necessary in the criminal justice system. Monothetic,

PAGE 39

29 crime-based systems are best suited to the needs of law enforcement agencies, while polythetic, treatment-oriented systems meet the needs of prison officials and program planners. Systems which have sought to delineate the causes of criminal behavior or to trace recurring patterns of adjustment prior to the offense have failed. Second, since little empirical verification has been found for the theoretical constructs underlying many classification systems, the trend has been to attempt to define coherent sets of variables and to them explore the relations between these "categories" and other variables. That is, the usual process in developing an offender taxonomy has been to group individuals with similar crimes and then explore their similarities on other variables. As Megargee and Bohn noted (19 77) , this technique has been singularly unproductive, and the "psychometric" method which this system and the CACL use reverses this process. That is, offenders are categorized on variables related to current functioning, and the resulting types are related to past behavior or to predict future adjustment (p. 155) . Third, a tentative list of criteria for judging a taxonomic system for offenders emerges from all the articles in this area. These criteria are as follows: 1. It should relate to other variables of interest, and as a consequence may have theory building value.

PAGE 40

2. It should serve a specific purpose for limited population. 3. It should assess current functioning rather than past behavior. 4. It should classify all individuals in question and no individual should be classified into more than one category. 5. It should have specific, clear-cut decision rules allocating individuals to categories. This study will investigate the CACL primarily on related criteria (1) and (5). That is, the clearness of decision rules is reflected in the consistency of raters in assessing the behaviors in question. Thus the interrater reliability study will assess the clarity of the CACL's definitions, and the various validity studies will delineate the CACL's relationship with other variables. The following section will review the literature which provides the background for establishing the reliability and validity of the CACL. Psychometric Concepts This section reviews the concepts of reliability and validity as they pertain to this study. The classical theory of reliability and the major types of reliability estimates are reviewed. The factors affecting reliability and validity are then presented and the specific types of

PAGE 41

31 reliability and validity estimates obtained in this study are discussed at greater length. Reliability . The concept of reliability of measurement refers to its consistency across any of several dimensions. As several authors have pointed out, reliability, like validity, takes on special significance in the measurement of traits or inferred constructs (Stanley, 19G9; Cureton, 1958). Reliability has been of central importance in areas such as psychology and education, where indirect measurement is frequently employed. Definitions of reliability . Classical measurement theory has based the concept of reliability on the assumption that any measurement contains a discrete amount of random fluctuation or error in addition to the influence of the actual variable under consideration. As Stanley (1965) pointed out: When a feature or attribute of anything (in any of the sciences) is measured, that measurement contains a certain amount of chance error. The amount of chance error may be large or small, but it is universally present. (p. 356) Thus it is assumed that any observed score for an individual is composed of a true score component and an error score component which are linearly additive. Since the error score components are presumed to be random, they should not show any relationship to each other or to the true or observed scores. Cureton (195R) said:

PAGE 42

The basic theorem which underlies all formulas of reliability, and of empirical validity as well, may be stated as follows: In a population of individuals, the errors of measurement in different tests and the different forms of the same test are uncorrelated with one another and are uncorrelated with the true scores on all tests and forms. (p. 103) The error of measurement referred to by Cureton is an estimate which relates to the variability in a series of repeated testings of the same sample due to random (error) fluctuations. That is, if a number of independent measurements are taken on the sa-ne individuals, the variability in those measures would reflect the random fluctuations, or amount of error variance present in the measurements. The shared or common variance would reflect the amount of true score variability which was present. Since repeated testings of the same sample of individuals are not practical for a variety of reasons (the interactive effects of measurement, pragtice, etc.), the errors of measurement must be estimated indirectly. It is possible to see that the notion of variability of scores across any of several dimensions is central to the definition of reliability. Without observed variance in scores, the estimation of reliability is not possible. Again, since the true score variance and error score variance can never be assessed directly, one must attempt to estimate them from the observed variance in test scores. As Stanley (196 9) emphasized:

PAGE 43

The basic problem in defining the reliability of a testing procedure . . . becomes that of defining what shall be thought of an error variance in relation to the type of inference one wishes to make from the test scores. VJhen this definition has been made, the next step is to devise those series of empirical and statistical operations that will provide the best estimates of the defined fractions of variance. (p. 362) Since variability in test scores can arise from a variety of sources, the selection of which of these are to be considered as sources of error variance depends on the purpose of the testing. Stanley emphasized this point when he said: There is no single universal and absolute reliability coefficient for a test .... The allocation of variance from different sources calls for practical judgment of what use is to be made of the resulting statistical value. (p. 363) The reliability coefficient for any measure can be defined as that proportion of observed score variance which is composed of true score variance. Theoretically, then, the reliability coefficient can range from 0 to 1.00, where a zero reliability coefficient indicates an absence of variability attributable to true score differences, and where a reliability coefficient of one results from a complete absence of random, extraneous variability. The formula for this relationship can be expressed as:

PAGE 44

34 2 where R^^ is the reliability coefficient, is the 2 true score variance, and is the observed score variance. Since the observed score variance is presumed to be composed of a linear combination of true score variance 2 2 2 and error variance (S^ = + ) , the formula may also be written as. Factors affecting R^^. The presumed random nature and normal distribution of the error component influences the magnitude of the reliability coefficient in several ways. As the number of items in any measure increases , the errors will tend to cancel each other out to a greater degree. That is, as the number of items approaches infinity, the sum of the errors will tend to approach zero. Magnusson (1967) also noted that the error variance increases arithmetically with the length of the test while the true-score variance increases with the square of the niimber of items. Thus, "when the test is lengthened, the true variance increases at a faster rate than the error variance. This . . . means that the test will become more reliable" (p. 72). In addition, the homogeneity or amount of total variance in the sample also determines the magnitude of the reliability estimate. As the sample becomes more homogeneous.

PAGE 45

1 35 the amount of true score variance decreases, while the error remains unchanged. This decrease results in a reduction of the magnitude of the reliability coefficient, since the ratio of true score variance total score variance has been decreased. Types of reliability estimates . As mentioned previously, the particular source of total test variance which is considered as error depends on intended use of the instrument. For example, if a measure is intended to measure a single, unitary trait it is highly desirable that the items share as much common variance as possible. Again, if a test is intended to measure an enduring characteristic of the individual, it should have as much stability across time as possible. Reliability estimates can be thought of as approximations of true-tototal variance proportions, where the priority of the use of the test determines which of the above will be considered as most important sources of true variance. The various types of reliability coefficients can be thought of as falling into several broad classes, based on the type of error which is considered most important to the measure in question. Cronbach (1960) has defined three such classes of reliability coefficients which he calls coefficients of stability , coefficients of equivalence , and coefficients of internal consistency . I

PAGE 46

1. Coefficients of stability estimate the consistency of test scores across time, and are particularly important in measuring the lasting characteristics or traits of individuals. Such coefficients are generated in a test-retest paradigm where the same instrument is given on several occasions . 2 . Coefficients of equivalence are intended to measure the similarity of several forms of a specific test. That is, equivalence estimates are intended to measure the degree to which two tests are parallel — that is, having the same means, variances, and average item intercorrelations. It is also possible to consider inter-rater reliability as a type of equivalence estimate, although the same form of the instrument is used. Inter-rater reliability estimates compare the shared variance across several individuals who assess the same person at the same time and under the' same conditions. 3. Coefficients of internal consistency assess the degree to which the items in a test measure the same trait, construct or characteristic. One estimate of internal consistency is obtained by dividing the responses to a test into two parts, and correlating the two halves with each other, and approximating the reliability of the total test by the use of the Spearman-Brov/n prophecy formula. This is known as a "split-half" reliability coefficient.

PAGE 47

37 Various other indices of internal consistency have been developed, such as Cronbach's coefficient alpha (Cronbach, 1951) and the Kuder-Richardson (1937) formulas, which estimate the average of all possible split-half reliability coefficients of a given test. These coefficients will not be further discussed, since this study is concerned only with the consistency across raters, rather than internal consistency or stability across time. Reliability Estimation in This Study The specific type of reliability which is of the greatest concern in this study is the degree to which equally trained independent observers agree on the presence of the behaviors assessed by the CACL. Although the instrument itself will be discussed further in the methodology section, it is important to note that it is neither a rating scale por a traditional observational instrximent. Rather than counting the frequency of occurrence of specific behaviors or rating the individual along a theoretical continuum, the CACL is designed to determine whether a specific behavioral trait is characteristic of the individual (Quay, 1964). In a recent article, Frick and Semmel (1978) made several important points in regard to inter-observer agreement (reliability). They said:

PAGE 48

38 Minimal observer disagreement is a necessary but insufficient condition for high reliability coefficients, since there are other components of the generic error variance that are theoretically independent from observer error variance (e.g., intrasubject variance from occasion to occasion) . (p. 159) In addition, the authors also note that although observer or rater agreement is only a part of the reliability of observational data, it does set the upper limit for the reliability of the data under consideration. That is, until the observational systems capacity for interobserver agreement has been defined, it is difficult to determine the degree to which other factors are limiting the reliability of the data (pp. 160-161) . Frick and Semmel also point out that the traditional definition of reliability as agreement between measures which have identical content, means, variance, and item intercorrelations is impractical when applied to human raters. That is, observers or raters do not have identical or equivalent observational skills. Accordingly, intraclass correlation coefficients or generalizability coefficients have been proposed as techniques to determine the reliability of a set of data without depending on the above assumptions. Such coefficients have often been used in the analysis of classroom observation data, but are equally applicable to measurements from other sources (Haggard, 1958; McGaw, Wardrop, & Burda, 1972).

PAGE 49

39 Such coefficients estimate the ratio of true-to-total variance, but use an analysis of variance model to estimate the relative contributions of various sources of error variance. Although a more detailed description of the technique used in this study will be given in the procedures section, it is of importance to reiterate that such analytic techniques are used since the traditional assumptions underlying reliability are not applicable to data arising from ratings or observations. An earlier article by Ebel (1951) compared the advantages of the intraclass correlation coefficient with other methods for assessing the reliability of ratings. In recommending the intraclass coefficient, Ebel listed three major advantages of such an approach. First, the intraclass formula permits the investigator to choose whether to include "between raters" variance as part of the error variance. . . . Second, a convenient means for estimating the precision of the reliability coefficients is available to the user of the intraclass formula. Third, the intraclass formula uses the familiar statistics and routine computational procedures of analysis of variance. (p. 423) In a position paper, McGaw et al. (1972) made a distinction between reliability coefficient as calculated from the internal structure of a test, from repeated testings, or from parallel forms, contrasting these with indices of observer agreement. Antedating the views of Frick and Semmel, they noted that agreement between observers has all

PAGE 50

40 too often been considered the only important aspect of the reliability estimation of observational data. Specifically, they say: The confusion introduced into the literature through failure to clearly distinguish the different sources of unreliability, and through over-emphasis on inter-judge agreement has resulted from a confusion of the importance of primacy with prime importance. Interjudge agreement is the first, but not the most important issue to be faced. (p. 16) Thus for the current study, it is most important to note that the inter-rater reliability (agreement) which is calculated is not to be considered the only aspect of stability of data arising from the CACL which should be studied. However, because of its importance it is the type of reliability to be examined in this study. The inter-class correlation coefficients which were derived in the study are for the average of three raters, where^each rater rates all subjects. These estimates are considerably higher than those which would be obtained for a single rater. These coefficients are also calculated differently when absolute rather than comparative decisions are being made. When absolute decisions are involved, systematic rater bias is included in the error term of the model. For comparative decisions, such bias is not included along with the subjects by rater interaction in the error term.

PAGE 51

Validity Most authors agree that validity, like reliability, is a general term for a variety of related processes which assess the "usefulness" of a test. Brown (1970) pointed out that validity analysis may answer any of the following questions : Kow well does the test do the job it is employed to do? What traits are being measured by the test? Is the test actually measuring what it was designed to measure? Does the test supply information that can be used in making decisions? What interpretation can be given to the scores on a test? What can be predicted from the test scores? (p. 99) That is, validity studies generally attempt to relate test scores to other variables of interest. In terms of true and error score variance, Brown said: Whereas reliability was defined by the proportions of true and error variance, validity is determined by the proportion of true variance that is relevant to the purposes of testing. . (p. 98) Thus, the process of validation usually involves assessing the relationship between the test and some external criterion . The definitions of validity, which have been given in the Standards for Educational and Psychological Tests , center around the process of estimating the usefulness or meaningfulness of the data from a particular instrument. Each of these definitions will be discussed at a later point, but it is important here to compare the definitions of validity held by other authors.

PAGE 52

42 Ebel (1961) suggested that defining validity is more difficult than it may appear at first glance. He pointed out that various authors diverge widely in their definitions of validity, and as examples notes that: Gullikesen . . . has said: "The validity of a test is the correlation of the test with some criterion." Cureton writes: "The validity of a test is an estimate of the correlation between the raw test scores and the 'true' (that is perfectly reliable) criterion scores." Lindquist suggests: "The validity of a test . . . (is) . . . the accuracy with which it measures that which it is intended to measure. ..." Edgerton suggests: "By validity we refer to the extent to which the measuring device is useful for a given purpose." Cronbach explains: "The more fully and confidently a test can be interpreted, the greater its validity." (p. 75) Ebel continued by defining three other problem areas in the area of validity: The fact that it must assume diverse forms to fit diverse situations, the discrepancy between the importance of test validity and the state of the art of validation, and the fact that the question of validity doesn't arise in the physical sciences. (pp. 76-78) In addition, he pointed out that the concept of validity is not philosophically adequate, in that it is unlikely that, "the naive faith in the pre-existence of a quantity to be measured is basic to the general conception of validity" (p. 79) . Ebel also mentioned that these difficulties may well be due to a variety of causes. First, he suggested that although the relation between a test and criterion is central to validity theory, the criterion, like the test

PAGE 53

itself is most often constructed and thus of limited validity itself. In addition to the philosophic problems of a "true" score, Ebel also saw the concept as frequently overgeneralized and used in inappropriate settings. As a solution to these problems, Ebel (1961) suggests that the term "meaningfulness" be used to subsume the concept of validity. That is, he suggested that the assessment of the relationship between test scores and other measures be one of factors which contribute to the interpretability of test scores. He recommended the other factors to be considered should be the reliability of the measure, the norms used, and the operational definition of the score itself. Following Ebel's recommendations, this study is an assessment of the meaningfulness of the CACL. That is, scores on the CACL are related to other measures for a sample which differs from the norms and the reliability of the instrument is assessed. In this way, we have an indication of the usefulness of the instrument with a population having a high degree of psychopathology . Magnusson (1967) said that validity, like reliability, is an aspect of dependability, and that "the validity of a method is the accuracy with which meaningful and relevant measurements can be made with it" (p. 124) . As mentioned above, the criterion measure may be a test which has less than perfect validity and reliability

PAGE 54

44 itself. Magnusson pointed out that although imperfect reliability can be corrected, "low validity in the criterion data, however, can never be corrected for . . ," (p. 127). Often the question of how best to define the criterion variable is left essentially unanswered. Types of Validity Other authors concur that validity is most often concerned with the relationship between the test and other variables. Like reliability, this relationship can exist in any of several dimensions. Each of these dimensions covers a different aspect of validity, and may be thought of as the relationship between the test and a larger domain, other measures of the same trait, or the degree of "meaningfulness" of the test. The types of validity which correspond to those dimensions have been mentioned above and labelled by the American Psychological Association as con tent validity , criterion-related validity , and construct validity (APA, 1974) . The first of these concepts, content validity, refers to the adequacy with which a measure reflects the domain of items in question. Although content validity is an important area in the construction of achievement tests, it has little bearing on this study. Therefore, it will not be discussed at length.

PAGE 55

1 45 Criterion related validity has been defined by Gaion (1974) as "the extent to which scores on one variable, usually a predictor, may be used to infer performance on a different and operationally independent variable called a criterion" (p. 288) . If the criterion measure is taken at the same point in time, the process is known as concurrent validation. If the measure is taken later, the process is known as predictive validation. As has been mentioned previously, validation studies are intended to specify the "usefulness" of the test, or the degree to which it successfully accomplishes a given purpose. In a general review of validation, Cronbach (1960) equated criterion related validity with usefulness in selection and placement, both of which he subsumes under the process of decision making (p. 446) . It is important to note that criterion-related validity may be conceptualized as existing for a specific purpose and is empirically determined by the relationship between the test scores in question and a second criterion measure. In a brief review, Cureton (1958) said that the criterion may exist in the present or future, and may be pre-existing or constructed (p. 105) . Pre-existing criteria include those that exist without any special effort made to predict them. Examples of such criteria include graduation from college, number of previous criminal convictions, etc. Constructed criteria are

PAGE 56

46 usually developed on the basis of some hypothetical trait concept, and include rating scales, intelligence measures and personality tests. Criterion-related validation studies often numerically express the relationship between this test score and external measures in the form of a validity index, which represents the amount of variance common to the two. However, it is often presumed that the criterion measure is an adequate measure of the criterion when in reality this may not be the case. In an article on the problems inherent in criterionrealted validation, Brogden and Taylor (1950) defined "criterion bias" as "any variable, except errors of measurement and sampling errors, producing a deviation of obtained criterion scores from a hypothetical 'true' score criterion" (p. 82) . Although bias in the criterion which is not correlated with the predictor may undesirably affect validity studies of this type, Brogden and Taylor point out, "it is the presence of test-correlated bias that 'makes' or 'breaks' the criterion" (p. 82) . Construct Validity Estimates Unlike criterion related validity, construct validation procedures are often more conceptual than statistical. They attempt to assess the degree to which an instrument reflects an underlying construct or hypothetical trait. In a classic article, Cronbach and Meehl (1955) stated:

PAGE 57

47 Construct validation is involved whenever the test is to be interpreted as a measure of some attribute or quality which is not "operationally defined" .... Construct validity must be investigated whenever no criterion or universe of content is accepted as entirely adequate to define the quality to be measured, (p. 282) The authors continued to point out that construct validity is "not to be identified solely by the particular investigative procedures, but by the orientation of the investigator" (p. 281). That is, the procedure may incorporate concurrent or predictive methodologies, factor analysis, or other techniques to be discussed in this section. It is the aim or intent of the investigator that uniquely defines construct validation. A number of procedures have been used in an effort to determine the usefulness of a given construct in interpreting test data. Cronbach and Meehl listed several such techniques which provide the basis for inferring the existence of a trait. These techniques include the following: 1Studies of group differences which would be expected on the basis of the construct in question. 2. Correlations between items or tests which reflect the same trait. The covariation between such items or tests may be measured by means of factor analysis and correlation matrices . 3. Studies of the internal structure of the measure in question. For many constructs , evidence of homogeneity within the test is relevant in judging validity.

PAGE 58

1 48 4. Studies of change over occasions (retest reliability) may lend support to the logical network defining the construct. 5. Studies of the process of performing on the measure in question may also help to define the construct in question. (p. 289) In a reformulation of the techniques mentioned above, Campbell and Fiske (1959) point out that although we often use measures of association (correlation) to assess the presence of a construct, we also often look for divergences in test performance. They define the two processes as in the following manner: 1. Validation is typically convergent , a confirmation by independent measuring procedures. Independence of methods is a common denominator among major types of validity (excepting content validity) insofar as they are to be distinguished from reliability. 2. For the justification of novel trait measures, for the validation of test interpretation, or for the establishment of construct validity, divergent validation as well as divergent validation is required. Tests can be invalidated by too high correlations with other tests from which they were intended to differ. (p. 82) That is, the process incorporating convergent and divergent validation indices aids specifically in the logical interpretation of validation data. By demonstrating that different techniques intended to measure the same trait correlate significantly with each other, and that similar methods intended to measure different traits do not, povzerful logical evidence for the traits' presence has been presented .

PAGE 59

Following Campbell and Fiske's logic, it is evident that construct validation relies on both statistical and logical inferential techniques. That is, it uses empirical evidence to logically deduce the presence or absence of a specific trait. Unlike criterion-related validity, which relies heavily on statistical measures of association, the construct validity of an instrument is demonstrated through a series of analyses which are logically incorporated into the overall validation process. Factor analysis is widely used in the determination of construct validity. An early article by Guilford (1948) stressed the use of factor analysis in assessing the construct validity of an instrument. Guilford seemed to be anticipating the distinction between criterion-related and construct validity when he wrote of practical and factorial validity. He defined the factorial validity of a test as being determined by "its loadings on meaningful, common, reference factors" (p. 428) . Cattell (1964) also discussed the use of factor analysis in the determination of construct validity. As a type of convergent validity, he believed that factor analysis can help to define a construct when it emerges as a simple factor across several studies. This technique "combines measurement precision with unitary character, as well as a meaning enriched beyond that of an empirical construct" (p. 22) .

PAGE 60

50 Although Anastasi (1976) indirectly accepted the use of factor analysis in construct validation, particularly with reference to the measurement of general versus specific abilities, her overall stance has been strongly against anything other than criterion-related validity. She referred to "the will-o'-the-wisp" of psychological processes which are distinct from performance" (p. 77) . Cronbach and Meehl (1955) disagree with this position, and point out that inference based on patterns of association between variables "cannot be dismissed as pure speculation" (p. 290) . The CACL was not developed to measure a prespecified underlying trait, but rather was developed through the factor analysis of a set of behavioral descriptors. However, the four subscales of the CACL have been given labels based on their content, and these have been shown to correspond to broader traits which have appeared throughout various classification systems for offenders. Thus, the validation process in this study will attempt to relate scores on the subscales of the CACL to other measures which may be indicative of those traits. In this sense, estimates of construct validity are of primary importance in this study. That is, it is most important to define the nature of traits measured by the instrument, rather than to only establish its estimates of criterion-related validity.

PAGE 61

51 Chapter Summary Included in this chapter is a review of the literature in three major areas which are pertinent to this study. First, the process of classification in general has been summarized. Generally, classification systems serve many purposes and no single system can meet all the needs in any one area. Next, in reviewing the history of classification in the field of criminology, the problems with theoretically oriented typologies have been noted. Empirically derived classification systems such as the CACL after the advantage of proven utility for a specific population but need to be reevaluated before they are used with a group which differs from the nojrmative sample. Since the purpose of this study is to evaluate the psychometric proportion of the CACL based on the behavior of a group of mentally disordered criminals, the area of reliability and validity were reviewed at some length in this chapter. Particular emphasis is given to the topic of inter-rater reliability, which sets the upper line for the reliability of rating scale such as the CACL. Criterion-related validity was also discussed at some length since the CACL is intended to facilitate decisions about future custody and treatment of individuals in confinement. This study relates the CACL to several criterion measures, including the MMPI and behavioral measures of disruptiveness .

PAGE 62

52 Since these behavioral measures are of interest in their relation to the hypothetical traits measured by the CACL, the area of construct validity is also reviewed. Although the CACL is designed to describe patterns of behavior within the institution, it also labels these patterns in accordance with existing theories of criminal behavior. Thus, it may be used in "theory-building" studies rather than as a descriptive tool.

PAGE 63

CHAPTER III METHOD The purpose of this study was to investigate the psychometric properties of the Correctional Adjustment Checklist (CACL) , based on ratings of the behavior of a group of individuals confined in a maximum security mental hospital. Specifically, this study was designed to assess the inter-rater reliability of the instrument and to pjrovide estimates of its construct and criterion-related validity when used with individuals showing evidence of various types of mental disorders. The procedures used to obtain these estimates are detailed in a description of the subjects, the instriiments, and the analytic techniques used. Since the emphasis of this study was to evaluate the instrument when used with a group which is different from the normative sample, the description of the subjects which follows is of considerable importance. The Sample All subjects included in this study were housed in the North Florida Evaluation and Treatment Center (NFETC) , which is a 225-bed maximum security mental hospital located in Gainesville, Florida. It is operated and administered 53

PAGE 64

by the Department of Health and Rehabilitative Services of the State of Florida and is currently the only mental hospital in the state which serves a purely forensic population. The hospital is composed of eleven residential and treatment buildings, consisting of one to three nine-person living areas which are known as "pods." Each patient (known as a resident) has a private room, and shares bathing facilities and a living area with the other residents in his pod. The hospital is divided into three units, each of which serves a particular type of client. Based on diagnostic categories, these types are as follows: psychotic, behaviorally disordered, or mentally disordered sex offenders. Although all of the residents have been charged and arrested for a major felony, not all have been tried, convicted, or sentenced. Those individuals who have been found incompetent to stand trial or to be sentenced are placed in the psychotic unit for short-term (averaging two months) treatment. Also, individuals who become psychotic while incarcerated are given similar short-term care. The Psychotic Unit currently includes ninety beds. The Behavior Disorders Unit is comprised of forty-five beds and is intended for the behavioral management and treatment of antisocial, retarded, or neurologically impaired individuals. Such persons are usually management

PAGE 65

55 problems in the traditional prison system, and are sent to NFETC for short-term treatment of recurring problem behaviors . The Sex Offender treatment unit includes ninety beds and is oriented to the long-term (approximately two years) treatment of individuals who have been convicted of a sexual offense and been classified under Florida Statute 917 as Mentally Disordered Sex Offenders. The individuals so classified must be manifestly non-psychotic, and be judged by at least two psychiatrists to have a predisposition to commit other sexual offenses. Overall, the population of the North Florida Evaluation and Treatment Center can be described as a group of approximately 225 males, all of whom have been arrested for a major felony and most of whom have been either found incompetent to stand trial or incompetent to be sentenced; who have become psychotic or a management problem while incarcerated; or who have been adjudicated as Mentally Disordered Sex Offenders. The age of the residents at the time of this study ranged from seventeen to seventy-nine, with a median age of twenty-eight, and they came from a wide variety of ethnic and social backgrounds within the state of Florida. Selection of the Sample From October 1976, when NFETC first began receiving residents, until July 1, 1978, approximately 550 individuals

PAGE 66

56 have been treated or evaluated in the institution. Of these, approximately 325 have been treated and returned to the referring agency, while the remainder are still confined at the hospital. The data which are available on these individuals are a function of events which were not under the control of this author. Since the emphasis at this hospital is on treatment and effective management of residents, changes in intake and diagnostic procedures were made which did not allow data collection procedures which would have been optimal for this study. For the first 14 months of operation (until January 1978) , the hospital included a central intake and diagnostic unit where all incoming residents were housed for shortterm evaluation and diagnosis. During their stay in the intake and diagnostic unit, the residents were assessed on a battery of diagnostic tests including the Minnesota Multiphasic Personality Inventory (MMPI) , the Incomplete Sentences Test, the Social Reaction Inventory, the Quay Correctional Adjustment Checklist (CACL) and Checklist for the analysis of Life History (CALH) . Since January 1978, the Intake and Diagnostic Unit has been concerned with the evaluation of incoming sex offenders only. Admission of residents to the Psychotic and Behavior Disorders Units has been directly to the building in which they were to be treated. This change has occurred

PAGE 67

57 because of increased number of admissions to the Sex Offender Unit and because of the increased need for more intensive evaluation of incoming residents. Accordingly, the Intake and Diagnostic Unit has increased the number of evaluation instruments which are administered to sex offenders. All sex offenders are given the MMPI, CACL, Bipolar Psychological Inventory, a short form of the Wechsler Adult Intelligence Scale, the California Psychological Inventory (CPI) , and a complete and extensive social and demographic background information survey. Descriptive statistics for this sample are presented in Table 1. Thus, most of the residents who have been admitted to NFETC have been tested during the first week of their stay in the institution. Unfortunately, since January of 1978, many residents who have been admitted to the Psychotic and Behavior Disorders Units have not been rated on the CACL. Since the residents were admitted directly into treatment, the staff in the buildings in which they were placed had not been trained in the use of the CACL or other diagnostic instruments . Accordingly, the sample on which the following study of the CACL is based includes higher proportions of Mentally Disordered Sex Offenders than other treatment categories. Although some test data are available on all residents, with few exceptions, only those who were rated on the CACL during the first two weeks of their stay at NFETC are

PAGE 68

58 included in this study. The exceptions to this sampling plan are those 27 individuals who were included in the inter-rater reliability study. Those persons had all been in treatment in the Sex Offender Unit for at least 60 days. TABLE 1 NUMBER OF RESIDENTS BY UNIT ADMITTED TO NFETC FROM ITS INCEPTION UNTIL JULY 1 , 1978 Psychotic Unit 90 179 . Sex Offender Unit 90 45 Behavior Disorders Unit 45 91 In treatment as of 7/1/78 Discharged prior to 7/1/78 Of those residents admitted, intake data on the CACL are available on 140 individuals. Of these, 73 have been treated in the Psychotic Unit, 4 7 in the Sex Offender Unit, and 20 in the Behavior Disorders Unit. The number of residents included in each of the studies reported here varies to some degree as a function of the availability of CACL intake data. While the central Intake and Diagnostic unit was using the CACL, each resident was rated independently by three staff members, and an average

PAGE 69

59 rating was used to describe the individual. The reliability of the ratings on the 140 individuals on whom such data are available will be computed and compared with that obtained on the twenty-seven residents who were included in the sex offender sample. After January 1978, the CACL was administered only to those residents who were considered diagnostic problems or whose placement in a particular treatment unit was difficult. All residents were given the MMPI within two weeks of the date they entered NFETC, and often were retested if their responses were considered invalid. If this is the case, the second profile is used for the studies described here. Instrumentation The primary instrument of interest in this study is the Quay Correctional Adjustment Checklist (CACL) . This is a 41-item, factor analytically derived behavioral checklist. It was developed between 1964 and 1971 as a classi-* fication instrument for incarcerated males. In form, it is neither a true rating scale nor behavioral checklist. Rather, it includes a number of statements which are said to be characteristic of the individual in question. The CACL is related to the early work of Hewitt and Jenkins (1946) who conducted an analysis of clusters of traits common to juvenile delinquents referred to a child guidance clinic. The resulting groups of traits were used

PAGE 70

60 to classify juvenile offenders into three categories: unsocialized-aggressive , socialized delinquent, and overinhibited . Based on these results, Quay (1964) developed a 36item checklist which was used to quantify the life histories of approximately 100 juvenile offenders. The responses to this checklist were factor analyzed in order to determine whether patterns of developmental events could be used to classify juvenile offenders. The results of this study indicated that the categories developed by Hewitt and Jenkins also appeared in the data obtained by Quay (1964). The checklist itself was later developed into the Checklist for the Analysis of Life Histories (CALH) , which is often used as a supplement to the CACL. Subsequently, Quay reported on the development of the CACL and CALH in a 1971 paper. In describing the development of the CACL, Quay related that a pool of behavioral descriptors was assembled from correctional workers and from previous research. Approximately 1,000 inmates from four institutions were rated on the items which were derived from these traits, and the resulting data were analyzed by means of factor analysis in order to estimate the extent of any underlying traits in these results. Four factors emerged, three of which correspond to those found in the CALH,

PAGE 71

61 In describing the item selection technique used. Quay said that analyses were performed on three separate samples, each drawn from a different Federal Correctional Institution He noted that, Subsequent to the first analysis, items which did not meet the frequency criterion (not more than 90% or less than 10% of the subjects were rated as exhibiting the trait) and items which loaded less than .20 on any of the factors were dropped, and other items were added for the second analysis. (Quay, 1971, p. 3) All three analyses produced four principal dimensions. The first, labeled Aggressive-Psychopathic, reflects toughness, defiance, physical and verbal aggression, troublemaking, victimizing, and quick teraperedness . The second dimension, labeled Immature-Dependent, is composed of such behaviors as inability to follow directions, sluggishness, daydreaming, preoccupation, passivity, moodiness, and dullness. The third factor, given the label NeuroticAnxious, reflects worry, tenseness, help seeking, fear of other inmates, sadness and emotional lability. The fourth dimension, measured by only five items is labeled as Manipulative and involves such characteristics as trying to "con" staff, lack of trust of staff, accusing staff of unfairness, and playing staff against one another. According to Quay, the factors which emerged in the three samples were congruent with each other to a high degree in two cases (the Psychopathic-Aggressive and Immature-Dependent subscales) , and less so in the cases of

PAGE 72

62 the Neurotic-Anxious and Manipulative sx±)scales. The degree of congruence was measured by Tucker's congruency coefficients, but the nxamerical values of these coefficients were not presented by Quay. In the final selection of items, two major criteria were used: first, the item had to have a loading of .40 or higher in one or more of the analyses described previously; and second, the item had to load on the same factor in two of the three samples. After items were selected on these criteria, the results from all three groups were combined and factor scores were computed using unit weights. That is, each item checked as characteristic of the individual earned a value of one toward the score on that factor. Thus, the maximum score on each factor is the number of items contained on that subscale. When the raw score distributions for each scale were plotted. Quay reported "gross departures from normality" (p. 5) , which were evident by visual inspection. The raw scores were subsequently converted to normalized "T" scores. As an estimate of the internal consistency reliability of the CACL, Quay reported that alpha coefficient was calculated for each of the subscales. For the total sample of 829 (all three groups combined) , the reliability estimates were as follows: .91 for the Psychopathic-Aggressive subscale; .82 for the Immature-Dependent subscale; .77 for the Neurotic-Anxious subscale, and .77 for the Manipulative subscale .

PAGE 73

63 Quay also examined the intercorrelations of the four subscales. He noted that: "While the factor analytic procedure results in uncorrelated factors, the actual estimates of scores of individuals on the factors are not necessarily independent" (p. 5) . The highest intercorrelation (.81) was found between soibscales 1 and 4 (Psychopathic-Aggressive and Manipulative) . A moderate correlation (.41) was also found between the ImmatureDependent and Neurotic-Anxious subscales. Quay speculates that this is probably due to rater's tendency to evaluate prisoners as being "totally troublesome" (Quay, 1971, p. 5) Quay (1971) also reported a validation study in which CACL subscale scores were related to a variety of other variables, primarily demographic in nature. He reported that all of the subscales showed a "modest" relationship to other variables. The Psychopathic-Aggressive subscale correlated negatively with the age of the criminal and positively with the number of prior arrests. The ImmatureDependent subscale tended to relate negatively to I.Q. and years of education. Scores on the Manipulative subscale tended to relate negatively to number of prior arrests, but exact numerical values were not presented. In general, the CACL has fairly high internal consistency, but unknown inter-rater reliability. Although it was designed to provide subscales which are independent of each other, modest subscale correlations are found in most

PAGE 74

64 studies. Evidence of construct validity is slight; statistically significant correlations exist between CACL subscales and some other variables, particularly number of prior arrests, age at arrest, and intellectual level. No specific suggestions for decision rules are included with the instrument, forcing the user to choose whether to use scores on the CACL in making absolute or comparative decisions. When the instrument was used at NFETC, the highest subscale "T" score determined an individual's CACL classification type. This classification was supplemented by other tests, interviews and so forth. Other instruments have been used to classify individuals who are incarcerated. Such instruments range from projective tests such as the Rorshach Ink Blots to selfreport inventories such as the 16 Personality Factor Inventory. It is of interest that these instruments were not created for the purpose of classifying criminals, but were developed as diagnostic aids in mental health settings. The Minnesota Multiphasic Personality Inventory (MMPI) is a 556-item self-report personality inventory. It was developed as an aid to the classification of psychiatric patients, and each of its original nine subscales corresponds to a diagnostic category current at the time of the test's construction. Although these categories were originally presumed to be mutually exclusive, subsequent research has shown this not to be the case (Dahlstom, 1972) .

PAGE 75

65 Despite the intercorrelation of the subscales as well as the tests' sensitivity to the response set of the testtaker (Messick & Jackson, 1967) , the MMPI has been shown to be useful in assessing a variety of areas of functioning. Recent research has stressed the interpretation of profiles of subscale scores rather than classification into one of several psychiatric diagnostic categories (Meehl, 1955). In addition to the original nine diagnostic subscales, three "validity" subscales were added to the instrument. These scales are intended to estimate the interpretability of the other subscales, and measure ego strength, naive lying to "fake good" and the frequency of items seldom endorsed by the normative population. In general, the MMPI has been shown to be more valid for whites then blacks and to discriminate accurately between groups of psychiatric patients and prisoners with accuracy. Local norms are often more useful for behavioral predictions than are national norms (Palmer, 1970), but both provide predictive ability at a level significantly above chance. The MMPI has also been shown to relate to several other measures of criminal behavior, both in and out of incarceration (Panton, 1966) . By estimating the nature and extent of the relationship between the MMPI and the CACL, it is possible to assess the traits common to both and the overlap or redundancy in the instruments. It is also possible to estimate

PAGE 76

66 the relationship between the CACL and other variables which are of interest in themselves, as well as for their logical relation to the traits measured by the CACL. Disruptive behaviors in the institution constitute such a criterion variable transition. For the purpose of this study, disruptive behavior was defined as any act which was contrary to the resident rules of the North Florida Evaluation and Treatment Center and which disturbed the ongoing course of treatment. Such behaviors usually necessitate staff intervention and were limited to threats of aggression, aggressive acts, threats of self-injury, acts of selfinjury, destruction of property, and other unclassified infractions of rules (violation of curfew, refusal to take medication, etc.) Data Collection The data used in this study were collected at different times, by different staff members at NFECT, and on different individuals. As part of the normal intake procedure, ratings on the CACL as well as scores on the MMPI were obtained on 140 residents. In addition, CACL ratings were also obtained on a smaller, more homogeneous group of individuals who had been observed for at least eight weeks. Finally, the frequencies of six types of disruptive acts were recorded and included as a behavioral

PAGE 77

67 adjustment to confinement. CACL intake data were collected from October 1, 1977, until July 1978, as were MMPI scores on the same individuals. Ratings on the CACL for the smaller group of residents were collected in May 1978. To obtain the measures of behavioral adjustment, a tally of disruptive behaviors was made from the daily observation notes kept on each resident. These notes were written by the treatment staff in each building at least once every eight-hour shift. Since all significant behaviors, especially infractions of rules, were to be included in observation notes, it seems likely that most disruptive behaviors were so recorded. For each of the 140 residents on whom CACL and MMPI data were available, a survey was made of the 180 observation notes written during the first 60 days of his confinement. Those individuals who stayed less than 60 days were not included in this section of the study. Each note was inspected to determine if any disruptive behaviors were recorded. If more than one such behavior was mentioned, each was tallied separately. That is, if a resident threatened a staff member after receiving an infraction for face count, two disruptive behaviors were tallied. Only the specific mention of behaviors observed directly by staff were included. If one resident informed on another, the disruptive behavior was not tallied unless it was directly witnessed by a staff member.

PAGE 78

For each resident included in this study, a record of the most recent arrest and conviction was made from the FBI "rap sheet." This is a listing of all prior arrests and convictions for the individual in question and is compiled from all arrest records throughout the United States. Arrests and convictions are matched by fingerprints as well as by name, so that crimes committed under an alias are also included. For this study, the following categories were used: murder, armed robbery, assault (including attempted murder) , breaking and entering, forgery, and other nonviolent property crimes, rape or sexual assault, and nonviolent child molestation. Although the first two analyses included in this study both assess the inter-rater reliability of the CACL, they differ in several respects. The first is based on the ratings made by three staff members of the intake and diagnostic unit. They were made after a relatively short (seven-day) period, which according to Quay (personal communication, 1978) may not allow for sufficient observation time. The second study is based on ratings made by three staff members in the sex offender treatment program. These ratings were based on the behavior of 27 residents who were being treated in that program, and who had been in treatir^ent for at least eight weeks. The raters had observed the residents for the duration of their stay in treatment, and thus had the opportunity to

PAGE 79

69 base their ratings on a larger sample of behavior than that in the first study. In both studies, each rater rated every subject. The raters for the second study were trained over a seven-day period. Their training included operational definitions of the behaviors assessed by the CACL, as well as comparisons of their ratings on the same residents. That is, the raters filled out a CACL on two residents without discussing the results with each other. These ratings were then compared on an item-by-item basis, with group discussion of any discrepancies. After three such sessions, the raters agreed on 90% of the items on the CACL, and training was discontinued. Data Analysis All data were analyzed at the University of Miami computing facility using a UNIVAC 1100 computer. All analyses requiring a "packaged" computer program used the Statistical Package for the Social Sciences (SPSS) which is available in several versions at the University of Miami. Reliability of the CACL The inter-rater reliability of the CACL was estimated by the use of the intraclass correlation coefficient, which has been described by Ebel (1951) as well as by Bartko (1966) . Essentially, this method uses the analysis of variance to estimate the proportion of variance in a set

PAGE 80

of measurements which can be attributed to individuals, raters, and error. In this case, a subject by rater design was used. The resulting mean squares from the analysis of variance were substituted into Ebel's (1951) formula. As expressed by Ebel, that formula is: M_ M X ^1 M_ + (k-l)M X The V is the intraclass correlation coefficient, M— is the mean square for individuals, M is the mean square for error and k is the number of observers or raters. This formula is for estimating the reliability of a single rater, and does not include systematic rater bias in the error term. When used in making absolute decisions, any systematic bias of the raters needs to be included in the error term of the formula. Thus, the formula is: MM V = ^ 2 M_ + (k-l)M + k(Mj^ M)/^ In this case, is the intraclass correlation coefficient, M_ is the mean square for subjects, M is the residual mean square, k is the member of raters, is the mean square for raters, and N is the number of subjects. Since the average of three raters scores was used in placement decisions at NFETC, a third formula was used

PAGE 81

71 to provide estimates of the reliability of the average. Including systematic rated bias, that formula is: M_ M ^3 " M+ k(M^ R)/^ If we exclude systematic rater bias, the formula for the reliability of the average of 3 raters becomes: M_ M This formula was used to estimate the reliability of the sum of all four subtests, as well as for each individual subtest. It should be noted that one sample on which these observations were drawn was fairly homogeneous, in that it consisted of only one type of offenders; i.e., Mentally Disordered Sex Offenders. This homogeneity may have provided reliability estimates which are somewhat less than maximal. The other inter-rater reliability study was based on the ratings of all types of incoming residents over a nine-month interval. The training of the raters was not under the control of this author, and residents were rated after a relatively short period of observation. Predictive Validity of the CACL The ability of the CACL's subscale scores to predict institutional disruptiveness was estimated through a

PAGE 82

72 multiple regression procedure. This technique analyzes "the collective and separate contributions of two or more independent variables ... to the variation of a dependent variable" (Kerlinger & Pedhazur, 1973, p. 3) . That is, it estimates degree of relationship between a set of two or more variables and a single other-variable of interest. It provides approximations of the contributions to the variance of the dependent variable by a group of independent variables. This is accomplished by minimizing the sum of squared deviations between the predicted dependent variable values and those actual values obtained in the experiment. A linear combination is derived for the independent variables which minimizes those errors of prediction. The model for this "least squares" solution can be expressed as: Y = + B^X^ + B^X^ -f . . . + B^X^ + E, where B^ is a constant value, and B^ . . . B2 are the weights assigned to the independent variables X ... X 1 k The weights in a regression equation which are based on the raw scores of the independent variables are known as partial regression coefficients. They are scale dependent, in that they are not directly comparable with each other in absolute magnitude. These weights may be transformed into standard score format so that they are

PAGE 83

73 directly comparable in size. In this case, the weights are known as standardized partial regression coefficients, and reflect the unique contribution of each independent variable to the variance explained in the dependent variable. In this study, the total frequency of disruptive behavior during the first 60 days of incarceration was the dependent variable, and the subtests of the CACL were the independent variables entered in the multiple regression equation. The subscales of the lAlAPl were also entered in a separate analysis in order to compare the predictive validity of the two instruments. Construct Validation of the CACL The construct validation of the CACL was carried out by means of a canonical correlation analysis. This technique, described by Timm (1975) and others, is an extension of the multiple regression method discussed previously. That is, it provides an estimate of the maximum correlation possible between two linear composites of two sets of variables. This is accomplished by including more than one dependent variable in a linear composite which maximizes the degree of relationship between that group and a linear composite of independent variables. Two sets of predicted scores are generated by these linear combinations. If the dependent variables are identified as ^ . . f then Q = a,^, + a_y_ + . . . a y , Jn 1^ ± 2'' 2 n n

PAGE 84

74 where u is the composite value. If the independent variables are labeled as x, + . . . x , then v = b,x, + b x , 1 n linn where v is the composite of those values. Thus, a canonical correlation analysis provides the weights a^^ . . . a^ and . . . such that the Pearson Product-Moment correlation between u and v is a maximum. In this study, canonical correlation analysis was performed to relate the subscale scores on the CACL to subscale scores on the MMPI , with data on both instruments being collected at the time of intake. The results of the canonical correlation analysis were used in a redundancy analysis, which has been described by Stewart and Love (1968) . This technique provides a numerical estimate of the redundancy in one set of data, given the other. The redundancy coefficients were obtained by rating the proportion of variance in each set of variables extracted by each canonical variate. These proportions were then multiplied by the corresponding squared canonical correlation and summed across the significant canonical variates for each set separately. The resulting coefficients represent the proportion of variance in a set of variables that may be explained by the second set. Postdiction of Crime Type In order to further determine the validity of the CACL, a discriminant function analysis was performed using the CACL subscale scores as independent variables and type of crime as a categorical dependent variable.

PAGE 85

75 A discriminant function analysis is an extension of the multiple regression procedure, in that it provides a set of weights for the independent variables which minimize the errors of prediction when the dependent variable is group membership. As in multiple regression, a linear combination of independent variables is formed such that ^ = ^o ^1^1 ^2^2 • • • Vn' ^^^^^ ^1 • • • ^n the weights for the independent variables • • ' This linear combination provides the best discrimination between the groups by maximizing the among group variance in relation to the within group variance. In this study, only the most recent conviction was used to determine crime type. Nonviolent crimes were defined primarily as crimes against property (breaking and entering, etc.), while violent crimes were defined as those which involved physical aggression toward another individual (rape, assault and battery, homicide, etc.). The subtests of the CACL were used as the independent variables in the equation. Summary In this chapter, the sample and instruments used in this study are described as well as the procedures for data collection. It is noted that the data collection procedures were not under the direct control of this investigator and thus introduce certain limitations.

PAGE 86

76 Also included in this chapter is a description of the various procedures used in the analysis. Separate analyses were conducted to obtain reliability and validity estimates of the CACL. Two estimates of inter-rater reliabilities were obtained on each subscale of the CACL. The first set of inter-rater reliabilities was computed using the ratings from observers who were not trained on a sample of incoming residents to the institution. Intraclass correlation coefficients were computed from the data. The second set of interrater reliabilities used the same method of computation on ratings by trained observers on a sample of sex offenders . The relationship between the CACL and the MMPI was assessed using canonical variate analysis. Canonical correlations and redundancy indices were computed. In addition, multiple regression analysis was used in the prediction of institutional disruptiveness from the CACL. Finally, a discriminant function analysis was used to predict type of crime based on the CACL subscale scores.

PAGE 87

CHAPTER IV RESULTS The results of the analyses described previously are presented in this chapter. Generally, the results are given without interpretation, since their explanation and synthesis are presented in the following chapter. Descriptive statistics precede each section of this chapter. In the first section of this chapter, the results of the two inter-rater reliability studies are presented. The first presents the reliability estimates for the ratings done on intake (after 4-7 days of observation) by raters whose training was not controlled by this writer. The second provides a summary of the "controlled" study in which the raters had been trained by this writer and where the subjects had been observed for a minimum of thirty days. The second section of this chapter contains the results of the canonical correlation analysis between the average ratings on the CACL at intake and the scores on the MMPI administered at the same time. This section is followed by a presentation of the correlations between the canonical variates and the original variables to clarify the content of the canonical variates. The results of the redundancy analysis are also included. 77

PAGE 88

78 The third section includes the results of the series of multiple regression analyses relating scores on the CACL to several types of disruptive behavior. Results are given separately for suicide attempts, assaultive behavior, verbal threats and coercion, as well as for other infractions of program rules. This was done in order to relate CACL subscale scores to specific types of disruptive behavior, in order to assess the relationship between those behaviors and the CACL subtest which would be expected to relate most strongly to them. The final section of this chapter contains the results of the discriminant function analysis, which relates scores in the CACL subscales to the presence of violence in the crime for which the subject had been most recently arrested and/or convicted. That is, scores on the CACL are weighted so that a linear combination of subtests best predicts group membership, where the criterion for group membership is the presence or absence of violence. Inter-Rater Reliability of the CACL As has been explained previously, two studies of the inter-rater reliability of the CACL were performed. These studies used separate samples of subjects and raters, and the reliability estimate from each was obtained through an analysis of variance procedure. The descriptive statistics for the "intake" and "controlled" studies are presented in

PAGE 89

i 79 Tables 14 and 15 in Appendix B, respectively; Tables ] 6 through 23 include the corresponding analysis of variance summary tables for each subtest. Separate analyses were performed for each of the four subtests of the CACL for the average of three raters, as well as for single raters. These estimates are given for the average rating first, followed by that for single raters. For the "controlled" condition the coefficients are: I-D Subscale, r=.37, .26; P-A Subscale, r=.76, .51; N-A Subscale, r-.73, .46; and Ma Subscale, r=.78, .59. For the "intake" condition the values are: I-D Subscale, r=.70, .42; P-A Subscale, r=.60, .36; N-A Subscale, r=.60, .41; and Ma Subscale, r=.60, .43. Table K in Appendix B presents the values including systematic rater bias in the error. Construct Validation of the CACL One construct validity estimate of the CACL was obtained from intake ratings on the CACL and the results of the MMPI, when both were administered concurrently. The descriptive statistics for the CACL and MMPI are presented in Table 2, while the intercorrelations are presented in Table 3.

PAGE 90

80 TABLE 2 DESCRIPTIVE STATISTICS FOR CONCURRENT VALIDITY STUDY Variable Mean Standard Deviation Number CACL PA 46. 89 4.95 140 CACL ID 49.59 6.32 140 CACL NA 46.46 5.34 140 CACL Ma 46.79 4.10 140 MMPI 1 5.79 4.58 140 MMPI F 15.35 10.76 140 MMPI K 13.50 6. 16 140 MMPI Hs 17.04 6. 85 140 MMPI D 29.82 6. 84 140 MMPI Hy 23.90 7.09 140 MMPI Pd 29.77 4.53 140 MMPI Mf 26.49 4.68 140 MMPI Pa 16.23 6.52 140 MMPI Pt 32.17 8.34 140 MMpi Sc 39.60 11.82 140 MT'lpi Ma 24.52 6.65 140 MMPI Si 31.16 10.63 140

PAGE 92

82 The results of the canonical correlation analysis which was performed on these data are presented in Table 4. The two canonical variates which appear in this table are the only two which produced a canonical correlation coefficient which was significant at least at the .05 level. The values which are reported in this table test the significance of the cumulative canonical correlation coefficient as each canonical variate is removed. That is, the first value tests the significance of all canonical correlation coefficients and the second X^ value tests the significance of all canonical correlation coefficients, after the first has been removed. Table 5 includes the canonical weights for the I4MPI and CACL subtest or both canonical variates. Table 6 includes the product-moment correlations between the subtests of those two instruments and each canonical variate. As was mentioned in the methodology section, a redundancy analysis was performed to provide estimates of the variance shared between the MlAPl and CACL. Two redundancy coefficients were calculated: one estimating the redundancy of the MMPI, given the CACL i^^j/cACL^ ' ^"^ other, that of the CACL given the m\PI (RcACL/MMPI^ ' Stewart and Love (1968) pointed out, in a case such as this, both of these estimates are necessary since the total variances of the two instruments are not equal. For this study, the ^-MPI/CACL ""^^ ^^"^1 to .095, and the RcACL/MMPI ^^"^^ to .199.

PAGE 93

83 TABLE 4 RESULTS OF CANONICAL CORRELATION ANALYSIS OF THE CACL AND MMPI Canonical Canonical Wilkes Variate Eigenvalue Correlation Lambda D.F. 1 .38 .61 .39 122.78 52* 2 .22 .47 .63 68.96 36* *p < .05.

PAGE 94

84 TABLE 5 CANONICAL WEIGHTS OF MTIPI AND CACL SUBTESTS FOR CANONICAL VARIATES 1 AND 2 (N=140) Subtest Canonical Variate 1 Weights Canonical Variate 2 V.'eights CACL-PA IDE? NA NA .144 .962 -.041 .234 1.028 .134 • .349 .012 MMPI L F K Hs D Hy Dd MF Pa Pt Sc Ma Si -.131 .936 .324 .651 .537 -.534 -.001 -.022 -.086 -.654 -.022 .022 -.10 9 .433 .504 .529 .091 .201 .196 .572 .019 .601 .074 .008 .138 .733

PAGE 95

85 Criterion-Related Validity of the CACL Several separate analyses were carried out to estimate the predictive validity of the CACL. First, a series of multiple regression analyses was carried out with the subtest scores on the CACL as the independent variable, and each measure of institutional disruptiveness as the dependent variable. In each analysis, the CACL subtest which theoretically should have shown the highest degree of association with the dependent variable was entered first in the regression. Because there was no rationale for the ordering of the remaining subtests, they were entered as a set on the second step. Table 7 includes the descriptive statistics for this analysis, and Table 8 gives the intercorrelation matrix for all variables. The results of the multiple regression analysis are presented in Tables 9 through 12, inclusive. These tables include the results for suicide attempts, assaults, threats, and interactions, respectively. The multiple regression analysis reported in Tables 9 through 12 was performed by entering first the CACL subscale which logically was considered the best predictor of each type of disruptive behavior. The other three subscales were entered as a set on a second step. Accordingly, these tables include the multiple R, R^ , and F value testing the significance of the R^ on steps one and two. In addition, the standardized partial regression coefficients

PAGE 96

8f; TABLE 6 PRODUCT MOriENT CORRELATIONS BETWEEN SUBTESTS OF THE CACL AND MMPI AND CANONICAL VARIATES (N=139) Canonical Variate 1 Canonical Variate 2 CACL mpi CACL MMPI CACL PA CACL IDEP CACL NA CACL NA 31 94 56 30 .19 . 58 .34 . 19 .91 .27 .05 . 55 .42 -.12 -.02 .25 mpi L MMPI F MMPI K M!4PI Hs MMPI D MJ4PI Hy MMPI Pd MJIPI Mf MMPI Pa MTIPI Pt MMPI Sc mPI Ma MMPI Si .01 .49 .09 .40 ,31 ,19 , 14 ,13 , 32 ,27 43 21 18 -.03 . 81 -.15 .65 .50 . 31 .23 .22 .52 .44 . 71 .34 .29 .13 .04 .01 -.07 -.19 .04 .08 .04 •.12 .11 .07 .17 .28 .28 .07 .03 -.16 -.42 -.11 .17 -.11 -.26 -.25 -.16 . 37 -.62

PAGE 97

87 TABLE 7 DESCRIPTIVE STATISTICS FOR PREDICTIVE VALIDITY STUDY Mean Ol^cillUciL LI Deviation N Suicide Attempts .12 1.14 104 Assaults .60 1. 15 104 Threats of Assault 1.08 1.67 10 4 Infractions of Rules .34 .97 104 CACL PA 47.12 5.18 104 CACL IDEP 50.54 6.33 10 4 CACL NA 46.77 5.32 104 CACL Ma 47.03 4.38 104 MMPI L 5.76 5.13 104 MMPI F 16.58 10.96 104 MMPI K 13.22 6.05 104 MMPI Hs 17.54 7.15 104 MMPI D 25.64 6.72 104 MMPI Hy 24.26 7.56 104 MI^PI Pd 29. 89 4.62 104 MMPI Mf 26. 89 4.67 104 MMPI Pa 16.59 6.54 104 MMPI Pt 32.53 8.46 104 MMPI Sc 40.64 12.34 104 MTIPI Ma 24.72 7.23 104 MMPI Si 31.99 10.50 104

PAGE 98

88 TABLE 8 INTERCORRELATION MATRIX FOR PREDICTIVE VALIDITY STUDY Suicide Threats of Other Variable Attempts Assaults Assaults Infractions ri\ 1 A . XH 0 0 • lb . io J. u _ nr n c . U D . (J 4 . U D P T\ PT In /A . 1 'i 1 "3 . lo . U J .11 CACL Ma -.08 .14 .03 .14 MMPI L -.04 .12 .11 -.09 MMPI F .16 .02 .05 .13 MMPI K -.18 .13 .07 .06 MMPI Hs . 0 8 -.03 -.08 .11 MMPI D .07 .18 -.09 . 14 MMPI Hy .07 .01 -.06 -.09 MMPI Pd .16 .06 .05 .01 mpi Mt .12 .07 .01 .03 MMPI Pa .23 .02 .01 -.14 MMPI Pt .10 .05 .01 -.14 MMPI Sc .18 .07 .05 -.08 MMPI Ma .06 .24 .11 -.09 MMPI Si .07 -.16 -.07 .07 Suicide Attempts 1.00 Threats of Assaults .08 1.00 Assaults .13 .40 1.00 Other Infractions .21 .25 .27 1.00

PAGE 99

B9 TABLE 9 RESULTS OF MULTIPLE REGRESSION OF FREQUENCY OF SUICIDE ATTEMPTS ON CACL SUBSCALES CACL Subscale Step R r2 F Beta —(unique) NA 1 .14 .02 2.09 .28 5.13* MA 2 -.41 9.60 PA 2 .32 5. 83* IDE? 2 .35 .13 • 3.63 -.21 3.55 *£<.05, 1 and 99 df TABLE 10 RESULTS OF MULTIPLE REGRESSION OF FREQUENCY OF ASSAULTS ON CACL SUBSCALES CACL Subscale Step R R • Beta -(unique) PA 1 .16 .03 2.62 .25 3.40 MA 2 -.15 1.18 IDEP 2 -.07 .30 NA 2 .20 .04 .99 .02 .40

PAGE 100

90 TABLE 11 RESULTS OF MULTIPLE REGRESSION OF FREQUENCY OF THREATS OF ASSAULT ON CACL SUBSCALES CACL 2 F Subscale Step B B Z Beta -(unique) FA 1 .29 .08 9.25** .34 f.53* MA 2 .92 IDEP 2 .96 NA 2 .31 .10 2.69 .10 .68 *P<.05, 1 and 99 df TABLE 12 RESULTS OF MULTIPLE REGRESSION OF FREQUENCY OF INFRACTIONS ON CACL SUBSCALES CACL Subscale Step ^ta -(unique) .13 .02 1.85 .05 .13 .05 .14 .13 1.01 .19 .04 .90 -.12 1.09 PA MA NA IDEP 1 2 2 2

PAGE 101

91 along with their corresponding tests of significance are reported for a model including all four subscales. Relationship of the CACL to Crime In order to assess the degree of relationship between the CACL and the presence of violence in the crime for which the subjects had been most recently convicted, a discriminant function analysis was performed. The CACL's subtests were the independent variables and charges at the time of arrest categorized into violent and nonviolent types constituted the grouping variable. Charges of property crime, drug charges, and others which did not involve physical contact were categorized as nonviolent. Those which involved a physical attack, battery, rape or murder v;ere categorized as violent. These data were analyzed by means of a discriminant function analysis, which provides weights for the independent variables such that the ratio of the sums of squares between groups to sums of squares within groups is maximized. In the procedure the dependent variable is categorical in nature, while the independent variables are continuous. This can be seen as a special case of the multiple regression procedure described previously, and therefore a multiple regression analysis was performed. The results of the analysis are presented in Table 13.

PAGE 102

92 TABLE 13 RESULTS FOR DISCRIMINANT FUNCTION ANALYSIS OF CRIME TYPE (N=104) Source D.F. S.S. M.S. F Regression 4 1.16 .29 1.18(n.s.) Residual 99 24.22 .24

PAGE 103

93 Summary In summary, this chapter included a presentation of the results of two inter-rater reliability studies on the CACL, as well as a number of validation procedures. These included assessments of the CACL's predictive validity in terms of four types of institutional disruptiveness , a concurrent validation with the MMPI, as well as a study of the instruments "post-dictive" relationship to the degree of violence involved in the subjects must recent crime. The following chapter will relate those findings to the overall utility of the CACL as a classification instrument in correctional settings. a

PAGE 104

CHAPTER V DISCUSSION The purpose of this study has been to investigate the psychometric properties of the Quay Correctional Adjustment Checklist (CACL) , which is a empirically derived classification instrument used with incarcerated criminals. In order to provide estimates of the instrument's reliability and validity, data were gathered on a sample of males who were being treated in a maximum security mental hospital. Because an average of three raters' scores was used for placement decisions within the institution, intraclass correlation coefficients were calculated for this average. For the purpose of comparison, these coefficients were also calculated for single raters. Since the CACL may be used for either absolute or comparative decisions, reliability coefficients were also calculated which included systematic rater bias in the error term (when the CACL is used for absolute decisions) as well as deleting it (when the CACL is used for comparative decisions) . These coefficients were calculated because although this study does not address the various decision rules or cutting scores which may be used in classifying individuals with the CACL, some readers may wish to use the CACL to compare individuals 94

PAGE 105

95 rather than making absolute placeinent decisions. In any case, reliability estimates will need to be calculated for the CACL for decision rules different than that used here. Two estimates of the CACL's reliability were obtained, one based on ratings of the subjects during the "intake and diagnostic" phase of treatment, and one after a longer period of observation and more controlled training of the raters. Both of these studies provided reliability estimates for the average of three raters which were larger than .50 in every instance except one. In order to assess the usefulness of the CACL with forensic psychiatric patients, three validity studies were conducted. First, a concurrent validation study, relating scores on the MMPI to those on the CACL was conducted. Second, subscale scores on the CACL were used to predict several types of disruptive behavior within the institution. Finally, scores on the CACL were related to the presence of violence in the subjects. This section contains a synthesis and interpretation, as well as a summary of results. Particularly in the validation analyses, an effort is made to place the results in a framework of meaning and applicability in other settings. The final section of this chapter contains suggestions for future research to improve the utility of those results obtained in this study.

PAGE 106

1 96 Construct Validation of the CACL The first inter-rater reliability study was based on data collected during the intake and evaluation phase of treatment. The subjects varied more in their crime types than did those in the second study and included a large percentage of individuals diagnosed as psychotic. The sex offenders included in the second study were by definition, nonpsychotic . Since the sample used in the second study was more homogeneous than that in the first, the reliability estimates obtained would tend to be somewhat lower than those obtained at intake. Thus, the reliability estimates from the "controlled" study do not represent the maximum possible for the instrument. Despite the homogeneity of the sample, three of the four scales on the CACL showed increases in the magnitude of obtained reliability estimates when the raters were thoroughly trained' in the definitions of terms and when a longer observation period was available before rating. Those subscales showing increases in the magnitude of obtained reliability estimates from "intake" to "controlled" settings, showed small gains in reliability. Estimates for single raters were much lower than those based on the average of three raters. Construct Validation of the CACL The canonical correlation analysis relating the CACL and MMPI was intended to determine the overall relationship

PAGE 107

97 between the two instruments , based on the analysis of data from a sample of 140 individuals who were assessed with both instruments during the first week of their stay at the institution. Two canonical variates were derived, and a product moment correlation coefficient between the weighted combination of CACL and MMPI subscale scores was calculated for each. The canonical correlation coefficients of .61 and .47 do not represent estimates of relationship between the total variance of each measure, but rather involve only that variance in each which was included in the particular linear combination of subscales (Stewart & Love, 1968) . Thus, these coefficients cannot be squared to determine the percentage of total variance common to both measures. The canonical correlation coefficients for both canonical variates are significant at the .05 level when tested with a chi-square statistic. Thus, for the two derived variates, a statistically significant relationship exists between the subscales of the MI'lPI and CACL. In addition to the canonical variate analysis, a redundancy analysis was performed in order to estimate the degree of congruence or overlap between the two measures. Two redundancy coefficients were derived by this analysis, one estimating the redundancy in the CACL given the MMPI, and a second estin-.ating the redundancy in the MMPI, given the CACL. The values for these coefficients are .19 and

PAGE 108

98 .12, respectively. The magnitude of these coefficients indicates a very modest relationship between the two sets of variables. In order to assess the concurrent validity of the CACL, it is necessary to examine the unique contribution of each subscale to the canonical variate in question, as well as to interpret the variates through the examination of the correlations between the original variables and their canonical weights. High positive weightings on the two canonical variates which were derived in this analysis appear to describe two distinct types of individuals within the population of the North Florida Evaluation and Treatment Center. That is, ratings on the CACL subscales do not correspond to four distinct patterns of canonical weights on the MMPI. Instead, the two variates which emerge load heavily on the Psychopathic-Aggressive (PA) and Manipulative (Ma) subscales in one case, and on the Immature-Dependent (ID) and NeuroticAnxious (NA) subscales in the second. For the purpose of convenience these canonical variates will be labelled according to the CACL subtest on which they load most heavily. The first variate will be called "ImmatureDependent" and the second, "Psychopathic-Aggressive." The "Immature 'Dependent" variate correlates highly (.60 and above) with MMPI subtests which relate to unusual responses (subscale F) , bodily discomfort or illness (subscale Hs) , and bizzare or psychotic symptoms (subscale Sc) .

PAGE 109

99 This variate also correlated .40 to .50 with subscales assessing hostility and suspiciousness (P-A) and overt symptoms of depression (subscale D) . The "Psychopathic-Aggressive" variate correlated +.55 with the Ma subscale. The MMPI subscales which showed correlation of the largest magnitude were those measuring social introversion (Si subscale) where r=.61; and the depression (D) subscale, where r=.42. Although dealing with a different population. Brown (1968) described the characteristics of a group of subjects at the Robert F. Kennedy Youth Center, who had been classified with the CACL as "inadequate-immature" delinquents. These descriptions correspond closely to the content of the "Immature-Dependent" variate. Brown (1968) also noted that such an individual is described as " . . . lazy, immature, a daydreamer, reticent, showing a lack of interest in things. His relationships are characterized by resentment (towards authority figures) or dependency. . ." (p. 3). Similarly, a group of individuals who are labelled as "psychopathic-aggressive" in the same study are described as "assaultive, cruel, defiant . . . wiley, deceitful and very untrustworthy . . . (such individuals) discount past mistakes and see their future without problems and themselves as great successes. . ."(p. 7). This description matches that given by Dahlstrom in the MMPI Handbook

PAGE 110

100 (1972) for individuals with low Si scale scores. He said that such individuals tend to be "active and vigorous, and competitive with their peers. They are persuasive and often win others over to their viewpoint. They also manipulated others in attempting to gain their own ends . . . they appeared unable to delay gratification and often acted with insufficient thought or deliberation . . . (which) . . . led to a destructive aggressiveness or hostility in their personal relations (p. 172) . It seems likely that this variate describes a group of individuals who tend to deny depression, are active and aggressive, and who have low impulse control. They tend to control others through verbal behavior or physical violence, and to have a great deal of energy and a rapid flight of ideas . This description fits those persons who were admitted to NFETC because "they were too assaultive or dangerous to others to be kept in other institutions. Their tendency to act out under minor stress and their need to control others were the likely cause of their confinement at the institution . The immature-dependent variate may well describe those individuals who are overtly psychotic but not highly agitated or assaultive. It also may include the individuals who were committed for evaluation and who are "faking bad" by pretending to be psychotic and or physically ill. m any case.

PAGE 111

101 such persons are seen as lethargic, withdrawn and passive individuals who depend highly on others to meet their needs. Whatever the reason, with this population, the CACL did not aggregate the subjects into four distinct groups. Rather, two groups emerged on both the CACL and MMPI , which seemed to differ primarily on the dimensions of activity control of others, and somatic complaints. For the "psychopathic aggressive" variate, this corresponds to Quay's earlier finding of the high degree of relationship between the PA and Ma subscales (Quay, 1971, p. 7). Criterion Validity of the CACL The second set of validity analyses on the CACL are concerned with the relationship between scores on its subtests and measures of disruptive behavior within the institution. Although the primary concern in this case is with predictive validity, inferences about construct a validity may also be made since the dependent variables are of interest in regard to their logical relationship to the CACL subtests, as well as being of concern in themselves. For example, we would expect the PsychopathicAggressive subscale to show a positive relationship of greater magnitude to threats of physical assault than do the other subscales, and this is in fact the case. This type of relationship provides partial confirmation of the validity of the trait names underlying the subtests of the

PAGE 112

102 CACL, and follows the distinction between predictive and construct validation made in the APA Standards for Psychological Tests. It should be noted that the four criterion measures which were chosen for the study showed very little variability. This may be because every effort was made to prevent the occurrence of those behaviors, and because many of the subjects were on their best behavior during the first sixty days of confinement. Whatever the reason, the lack of variability in the criterion tends to "cause" a decrease in validity estimates such as those presented here. Each category of disruptive behavior is described separately, and the results of the overall multiple regression of the CACL on each will be presented. Additionally, the multiple regression analysis is discussed in terms of the contribution of each individual subtest to the variance in each, dependent variable. Suicide Attempts The results of the multiple regression analysis indicate that the amount of variance in suicide attempts which is predicted by the CACL as a total test is significant at the .05 level. In addition, the unique contribution of each subtest is also significant at that level. It should be noted that with all the subtests in the prediction equation, only 13% of the variance in suicide attempts was predicted by the CACL.

PAGE 113

1 103 The Neurotic-Anxious subscale showed the largest degree of association with suicide attempts, and this is logically consistent with the hypothetical trait measured by this subscale. Individuals of this type have been characterized by Brown (1968) as " . . . fearful, anxious, withdrawn, hypersensitive, self-conscious, having feelings of inferiority and lacking self-confidence. . ." (p. 5). The question can be raised at this point as to the intent underlying the suicide attempts in question. As Samenow (1978) has pointed out, mental hospitals are preferable to prisons in terms of creature comforts. In his study, many individuals feigned psychopathology to be transferred to a more comfortable environment (usually a hospital) . Suicide threats and gestures were often used by the inmates to prevent their return to prison. Thus, the suicide attempts may have been either sincere or an effort to maintain status as a "patient" in need of treatment. Although it is not possible to retrospectively determine the reasons for such behavior, the results of the multiple regression analysis provide some confirmation of this hypothesis. The two highest positive beta weights in the prediction equation are for the Psychopathic-Aggressive and Neurotic-Anxious subscales. Thus, it may be that although these subscales relate significantly to the behavior in question, they may be discriminating between the two types of behavior (i.e., manipulative versus self-destructive) .

PAGE 114

104 The relatively high negative weighting on the Manipulative subscale (B=-.41) seems to be inconsistent with this hypothesis. However, the items on this subscale reflect behaviors such as lying and cheating, rather than less obvious manipulations. It is possible that such individuals have a repertoire of manipulative behaviors which are more effective than fake suicide attempts. Threats of Assault The overall F ratio for the multiple regression of the CACL on threats of assault is statistically significant at the .05 level. On inspection, however, only the PsychopathicAggressive subscale appears to accounting for a significant unique amount of variance to the dependent variable. The PA subscale predicts eight percent of the variance in threats of assault. The high loading on the PA subscale provides further evidence for the nature of the theoretical trait which it purports to measure. That is, we would expect such individuals to be aggressive, hostile and domineering in social interactions. Because such persons are thought to be easily frustrated, they are presumed to react to minor stress with a variety of aggressive manipulations, including threats of physical violence such as those recorded by the staff. It seems likely that threats of physical assault are less determined by the characteristics of the residents at the time of evaluation than they are determined by aspects

PAGE 115

105 of the environment of the time. Assault or threats of assault are intrinsically social behaviors, while suicide attempts are most often carried out in private. Thus, threats of assault are also partially determined by the behavior of the person being threatened. Assaults The overall F ratio for the multiple regression analysis of the CACL on instances of physical attack or assault is not statistically significant at the .05 level. This may well be due to the dyadic nature of the social interaction being measured. Although threats of assault may be seen as a manipulative style, a physical act of violence was usually followed by close confinement in a seclusion room. Again, repeated acts of violence were considered as grounds for transfer from the hospital, and may not have occurred with any frequency during the first two weeks of confinement. All of these factors may have contributed to the inability of the CACL to predict such behaviors. [it is of interest to note at this point that a recent review of dangerousness (Gottfredson, 1971) in a variety of settings showed a plethora of negative results in studies of individual characteristics, perhaps for the same reasons.] Infractions of Rules The overall multiple regression analysis for the CACL on minor infractions of rules was not significant at the .05

PAGE 116

106 level. None of the CACL subtests appears to relate significantly to the frequency of minor acts which were contrary to rules of the institution or unit where the residents were housed . This may be the result of several factors. First, the frequency of these infractions was the lowest of any of the recorded disruptive behaviors, having a mean recorded incidence of .34 for the sixty-day period. These infractions also showed the smallest variance of any of the recorded disruptive behaviors. This lack of variance may account for the minimal prediction of the CACL. Also, such behaviors may be the result of ignorance of the rules rather than of a prior condition, as measured by the CACL. Relationship of the CACL to Crime Type A final validity estimate on the CACL was derived from a discriminant function analysis in which the instrument's subscale scores were related to the presence of violence in the most recent crime for which they had been convicted. The results of this analysis were not statistically significant at the .05 level. Several factors may be responsible for this apparent lack of validity. First, the CACL was unable to predict violence within the institution, perhaps because of the short-time period that was involved in this study. As was mentioned before, violence is an interaction between two

PAGE 117

107 persons, and cannot be predicted well based on a knowledge of the characteristics of one individual. Second, the presence of violence was assessed by categorizing the crimes for which each subject had been convicted . Since these convictions often were based on plea bargaining, they may not have accurately measured the amount of violence present when the crime was actually committed. Summary of Psychometric Evaluation The purpose of this study was to evaluate the CACL in terms of its psychometric properties as measured in a variety of ways. Based on the various analyses which were performed, as well as the general characteristics of the instrument, several conclusions can be reached in regard to this evaluation . First, the instrument is polythetic in nature and provides rankings of individuals along several "behavioral dimensions." Although these dimensions were originally derived from a factor analysis, they have never been conclusively shown to the independent in later studies. This study and others have shown two clusters of traits rather than the four which were originally derived. The fact that these two groups of behaviors have been found in three separate studies with distinctly different populations indicates that they may well reflect actual methods of coping with a prison or hospital environment.

PAGE 118

108 Second, the CACL can be used to provide scores which produce inter-rater reliability estimates in the range of .60 to .70, for the average of three raters. These estimates are much lower for a single rater. It does not appear to be necessary for the raters to observe the subjects for two weeks to provide reliable scores on the instrument, but the short observation period used in these studies (3-6 days) may well have limited the validity estimates which were obtained. That is, this time period may not have been long enough to observe characteristic behavior patterns, since many persons may have been on their best behavior at the time of admission. Third, when used with a sample of mentally or behaviorally disordered individuals the CACL appears to be measuring some of the same traits as the MMPI. A construct validation study using a canonical correlation showed two underlying clusters of traits in this sample, and provided canonical correlation values of .61 and .47 for the two groups of subtests. For this sample and others, the CACL seems to be measuring dominance, aggression and mania in one dimension and feelings of distress, depression, social withdrawal and anxiety in the second. Since the Psychopathic-Aggressive and Manipulative subscales both load highly on the first dimension and the Immature-Dependent and Neurotic-Anxious subscales load on the second, it is possible that the raters were responding

PAGE 119

109 to more gross behavioral evidence than is desirable, and were tending to rate the subjects globally rather than specifically. Fourth, in the sample described above, the CACL shows a statistically significant relationship to the frequency of suicide attempts and to threats of violence, but not to other measures of disruptive behavior in the institution. The subtests which have the highest unique contribution to the prediction of those behaviors are the ones which would be expected to do so. That is, suicide attempts are most highly related to high scores on the Neurotic-Anxious and Psychopathic-Aggressive subscales, perhaps corresponding to real and feigned suicidality. The Psychopathic-Aggressive subscale showed the largest relationship in threats of violence, as would be expected. Finally, the CACL showed no statistically significant relationship to be presence of violence in the subjects most recent crime. However, the Manipulative subtest had a positive correlation of .17 with the presence of violence. While this is not statistically significant, it does provide some basis for speculation. It appears that more manipulative individuals tend to have more violence in their crime types than do other individuals. In general, the CACL appears to have some potential for useful classification in maximum security mental hospitals. Its inter-rater reliability is so low that it should not be

PAGE 120

110 used by a single rater to arrive at placement decisions. When three raters are used, the reliability estimates increase somewhat, but still do not provide much basis for placement decisions in the absence of other information. Although its value is limited by the extent of the raters' observation, it has the advantage of not being biased by the same response sets which influence self-report inventories such as the MMPI. Recommendations Several changes in the CACL mig-ht improve its reliability and general utility. It would be helpful if more precise definitions of the terms used were provided. During the course of training, several raters complained that no standards were given for decisions as to whether a particular behavior was included within a category on the CACL. Also, the instrument could be converted to a rating scale rather than a checklist. This would allow for more precise description of each individual than is currently possible, and by increasing the inter-individual variance would allow for more meaningful discriminations between individuals . Since the results of this study indicate that the CACL appears to have both construct and predictive validity, further efforts should be made to measure the stability of the behavioral traits which it measures. In this way, the instrument could be related to treatment outcomes and used

PAGE 121

Ill for measuring individual change across time. Since it is a nonreactive measure, it has the potential for serial administration without the reactive effects of other classification instruments. The question of the actual number of traits being measured by the CACL needs to be answered. It is possible that more precise behavioral definitions would allow for the assessment of whether two or four traits are being assessed. A larger sample of individuals should be assessed by raters who have been well trained, and who have observed the subjects in a variety of settings. A factor analysis of these results would provide more definitive evidence of this question . In general the CACL meets many of the requirements for an effective classification instrument. It provides a method for assessing behavioral styles in incarceration that have implications for both management and treatment. The lack of inter-rater reliability probably sets a limit on the validity of the instrument, and until this problem is improved it should be used with great caution.

PAGE 122

APPENDIX A CORRECTIONAL ADJUSTMENT CHECKLIST

PAGE 123

APPENDIX A CORRECTIONAL ADJUSTMENT CHECKLIST Marked for Final Factor Scales Scale I (Aggressive — Psychopathic) (N=18) Scale II (Immature — Dependent) (N=ll) + Scale III (Neurotic — Anxious) (N=7) Scale IV (Manipulative) (N=5) Col No. III (17) 0 1 1 II V Ju ; 0 a.xit;&, Dut cannot seem to roixow uirec III + (19 ) 0 1 •J • icii&t;, uiiauxe t_o rexax II (21) 0 1 A H • ouciaxxy wiunurawn T T T (ZZ) (J i 5 . Continually asks for help from staff I (24) 0 1 6. Gets along with the hoods II (25) 0 1 7. Seems to take no pleasure in anything III + (26) 0 1 8. Jittery, jumpy; seems afraid I (27) 0 1 9. Uses leisure time to cause trouble I (28) 0 1 10. Continually uses profane language; cur: and swears III + (29) 0 1 11. Easily upset II (30) 0 1 12. Sluggish and drowsy I (31) 0 1 13. Cannot be trusted at all II (32) 0 1 14. Moody; brooding I (34) 0 1 15. Needs constant supervision I (35) 0 1 16. Victimizes weaker inmates II (36) 0 1 17. Seems dull and unintelligent I (38) 0 1 18. Is an agitator about race IV (40) 0 1 19. Continually tries to con staff I (41) 0 1 20. Impulsive; unpredictable III + (42) 0 1 21. Afraid of other inmates 113

PAGE 124

114 COX 1 • J NO • I _(43) 0 1 22. II _ (44) 0 1 23. IV (46) 0 1 24. II — (49) 0 1 25. I (53) 0 1 26. I (55) 0 1 27. IV (56) 0 1 28. II (59) 0 1 29. I (62) 0 1 30 . I (64) 0 1 31 . I (65) 0 1 32. II (68) 0 1 33 . IV (70) 0 1 34 . II (71) 0 1 35 . I (72) 0 1 36. I (73) 0 1 37. III + (74) 0 1 38. I (75) 0 1 39. 1 (76) 0 1 40. IV (77) 0 1 41. Seems to seek excitement Never seems happy Doesn't trust staff Passive; easily led Talks aggressively to other inmates Accepts no blame for any of his troubles Continually complains; accuses staff of unfairness Daydreams; seems to be mentally off in space Talks aggressively to staff Has a quick temper Obviously holds grudges; seeks to "get even" Inattentive; seems preoccupied Attempts to play staff against one another Passively resistant; has to be forced to participate Tries to form a clique Openly defies regulations and rules Often sad and depressed Stirs up trouble among inmates Aiding or abetting others in breaking the rules Considers himself unjustly confined

PAGE 125

1 APPENDIX B SUMMARY TABLES FOR INTER-RATER RELIABILITY STUDIES

PAGE 126

TABLE 14 DESCRIPTIVE STATISTICS FOR INTER-RATER RELIABILITY STUDY: "INTAKE" CONDITION Variable Mean Standard Deviation Number CACL PA 46 . 89 4.95 140 CACL ID 49.59 6.32 140 CACL NA 46. 46 5.34 140 CALL Ma 46. 79 4. 10 140 TABLE 15 DESCRIPTIVE STATISTICS FOR INTER-RATER RELIABILITY STUDY: "CONTROLLED" CONDITION Variable Mean Standard Deviation Number CACL PA 47.12 5.18 69 CACL ID 50. 54 6.33 69 CACL NA 46. 77 5.32 69 CACL Ma 47.03 4.38 69 116

PAGE 127

117 TABLE 16 ANALYSIS OF VARIANCE TABLE FOR CACL PA "CONTROLLED" CONDITION INTER-RATER RELIABILITY STUDY Source Sums of Squares D.F. Mean Square Rater • 173.07 2 86.54 Subjects 2483.94 22 112.91 Residual 1216.93 44 27.66 Total 3873.94 68 TABLE 17 ANALYSIS OF VARIANCE SUMMARY TABLE FOR CACL ID "CONTROLLED" CONDITION INTER-RATER RELIABILITY STUDY Source Sum of Squares D.F . Mean Square Rater 314.46 2 157.23 Subjects 1837.28 22 83.51 Residual 2304.20 44 52.37 Total 4455.94 68

PAGE 128

118 TABLE 18 ANALYSIS OF VARIANCE TABLE FOR CACL NA "CONTROLLED" CONDITION INTER-RATER RELIABILITY STUDY Source Sum of Squares D.F. Mean Square Rater 71.04 2 35. 52 Subjects 2279. 30 22 103.61 Residual 1252.96 44 28. 47 Total 3603. 30 68 TABLE 19 ANALYSIS OF VARIANCE TABLE FOR CACL Ma "CONTROLLED" CONDITION INTER-RATER RELIABILITY STUDY Source Sum of Squares D.F . Mean Square Rater 81.48 2 40.74 Subjects 1954.44 22 88.84 Residual 880.52 44 20.01 Total 2916.44 68

PAGE 129

1 119 TABLE 2 0 ANALYSIS OF VARIANCE SUWIARY TABLE FOR CACL PA RELIABILITY STUDY Source U . r . Mean Square Raters 32. 43 2 16.21 Subjects 8298. 87 103 80.57 Residual 6703. 57 206 32.54 Total 15034. 87 311 TABLE 21 ANALYSIS OF VARIANCE SUMMARY TABLE FOR CACL ID "INTAKE" CONDITION INTER-RATER RELIABILITY STUDY Source Sum of Squares D.F. Mean Square Raters 92.55 2 46.28 Subjects 12372. 21 10 3 120.12 Residual 7392.78 206 35. 89 Total 19857. 54 311 I

PAGE 130

1 120 TABLE 22 ANALYSIS OF VARIANCE SUMTWRY TABLE FOR CACL NA "INTAKE" CONDITION INTER-RATER RELIABILITY STUDY Source Sum of Squares D.F. Mean Square Rater 7. 71 2 3. 86 Subjects 8733. 51 103 84. 79 Residual 6907.62 206 33.54 Total 15648. 84 311 TABLE 23 ANALYSIS OF VARIANCE SUMMARY TABLE FOR CACL Ma "INTAKE" CONDITION INTER-RATER RELIABILITY STUDY Source Sum of Squares D.F . Mean Square Raters 13.97 2 6.98 Subjects 5933.28 103 57.60 Residual 4815.36 206 23.38 Total 10762.61 311

PAGE 131

"1 121 TABLE 24 INTER-RATER RELIABILITY COEFFICIENTS FOR INTAKE AND CONTROLLED CONDITIONS, INCLUDING SYSTEMATIC RATER BIAS IN THE ERROR TERM Intake Controlled Average Average of Single of Single Three Raters Rater Three Raters Rater CACL ID .69 .40 .34 .26 CACL PA .56 .33 .71 . 49 CACL NA .57 .39 .72 .44 CACL Ma .59 . 43 .71 .58

PAGE 132

1 REFERENCES

PAGE 133

Abrahamson , D. The psychology of crime . New York: Holt, Rinehart and Winston, 1960. Alexander, F. G. Roots of crime; Psychoanalytic Studies . New York, W.Y.: Alfred A. Knopf, 1935. Alexander, F., & Staub, H. The criminal, the judge and the public. Glencoe, 111.: Glencoe Publishing Co., 1956. American Psychological Association. Standards for educa tional and psychological tests and manuals . Washington, D.C.: American Psychological Association, 1974. Anastasi, A. Psychological testing (4th ed . ) . New York: Macmillan Publishing Co., 1976. Argyle, D. C. A new approach to the classification of delinquents with implications for treatment. California State Board of Corrections. Monograph 2 ,1961, 15-26. Bartko, J, J. On various intraclass correlation reliability coefficients. Psychological Bulletin , 19f6, 83, 762-765. Blackburn, F. An empirical classification of psychopathic personality. British Journal of Psychiatry , 1975, 127, 456-460. Brogden, K. E., & Taylor, E. K. The theory and classification of criterion bias. Educational and Psychological Measurement , 1950, 10^, 159-186. Bromberg, W. , & Thompson, C. B. The relationship of psychosis, mental defect and personality types to crime. Journal of Criminal Law and Crimino logy, 1937, 28, 70-89 . " — Brown, D, E. The Robert Kennedy youth center: An interim report . Washington, D.C.: U.S. Department of Justice, 1968. Brown, F. G. Principles of educational and psychological testing . Hinsdale, 111.: Dryden Press, 1970. Campbell, D. T., & Fiske, D. W. Convergent and discriminant validation by the multi-trait, m.ul ti-method technique. Psychological Bulletin , 1959, 56, 81-105. 123

PAGE 134

124 Cattell, R. B. Validity and reliability: A proposed more basic set of concepts. Journal of Educational Psy chology , 1964, 55, 1-22, Clinnard, M. B. Sociology of deviant behavior . New York, N.Y.: Harper and Row Publishers, 1963. Clinnard, M. B., & Quinney, R. Criminal behavior systems : A typology . New York: Holt, Rinehart and Winston, 1973. Clinnard, M. B. , & Quinney, R. Criminal behavior systems (2nd ed. ) . New York: Holt, Rinehart and Winston, 1967. Cronbach, L. J. Test "reliability": Its meaning and determination. Psychome trika , 1947 , 12^, 1-16. Crornbach, L. J. Coefficient alpha and the interval structure of tests. Psychometrika , 1951, 16^, 297-334. Cronbach, L. J. Essentials of psychological testing (2nd ed,). New York: Harper and Row, 1960. Cronbach, L. J,, & Meehl, P, E, Construct validity in psychological tests. Psychological Bulletin , 1955, 5_2, 281-302. Cureton, E. E. Validity, reliability, and baloney. Educational and Psychological Measurement , 1950, 10 94-96. Cureton, E. E. The definition and estimation of test reliability. Educational and Psychological Measure ment , 1958, 18, 715-738. Dahlstrom, VI. G. An MMPI handbook (Vol. 1) . Clinical interpretations . Minneapolis: Univeristy of Minnesota Press, 1972. Dahlstrom, W. G. ; Welsh, G. S . ; & Dahlstrom, L. F. An MMPI handbook (Vol. 2). Minneapolis, Minnesota: University of Minnesota Press, 1972, Driver, E. A critique of typologies in criminology. Sociological Quarterly , 1968, 9_, 356-373 . Ebel, R. L, Estimation of the reliability of ratings. Psychometrika, 1951, 16, 407-424.

PAGE 135

125 Ebel, R. L. Must all tests be valid? American Psycholo gist , 1961, 16, 640-647. Ferdinand, T. N. Typologies of delinquency . New York: Random House, 1966. Fisher, S. Varieties of juvenile delinquency. British Journal of Criminology , 1962, _2, 251-261. Frick, T. , & Semmel, M. Observer agreement and reliabilities of classroom observational data. Review of Educational Research , 1978, £8(1), 157-184. Gaion, L. Criterion-related validity. Educational and Psychological Measurement , 1974, 3_2f 316-326. Gall, F. J. Craniology, and new discoveries about the head , the brain and the organs . Paris, France: publisher unknown, 18 07. Gibbons, D. C. Changing the lawbreaker . Englewood Cliffs, N.J.: Prentice-Hall, Inc., 1965. Gibbons, D. C. Society, crime and criminal careers (2nd ed.). Englewood Cliffs, N.J.: Prentice-Hall, Inc., 1970. Gibbons, D. C. Offender typologies: Two decades later. British Journal of Criminology , 1975, 15(2), 211-221. Glaser, D. The new correctional era-implications for manpower and training. Crime and Delinquency , 1964, 12^, 1-26. Glueck, S., & Glueck, E. Predicting delinquency and crime . Cambridge: Harvard University Press, 1959. Gottfredson, D. M. The base expectancy approach. In N. Johnson, L. Savirz, & M. Wolfgang (Eds.), The sociology of punishment and correction (2nd ed.). New York: John Wiley and Sons, 1971. Guikeson, H. Intrinsic validity. American Psychologist , 1960, 5, 511-517. Guilford, J. P. Factor analysis in a test development program. Psychological Review , 1948, 5^, 479-494. Guze, S. B. Criminality and psychiatric disorders . New York: Oxford University Press, 1976.

PAGE 136

126 Haggard, E. A. Intraclass correlation and the analysis of variance . New York: Dryden Publishers, 1958. Hewitt, L. E., & Jenkins, R. L. Functional patterns of maladjustment . Springfield, 111: State of Illinois, 1946. Hood, R. , & Sparks, R. Key issues in criminology . New York: McGraw-Hill, 1970. Hoyt, C. J. Test reliability estimated by analysis of variance. Psychometrika , 1941, 6, 153-160. Hunt, D., & Hardt, L. Developmental state, delinquency and differential treatment. Journal of Research in Crime and Delinquency , 1965, 2_3, 20-31. Hurwitz, L. Three delinquent types: A multivariate analysis. Journal of Criminal Law, Criminology and Police Science , 1965, 56, 328,334. Jenkins, R. A. , & Hewitt, L. C. Types of personality structure encountered in child guidence clinics. American Journal of Orthopsychiatry , 1949, 14, 84-94. Jenkins, R. L., & Glickman, S. Patterns of personality organization among delinquents. Nervous Child , 1947, 6, 329-339. Jesness, C. The Preston typology study. British Journal of Crime and Criminology , 1959, 2_3/ 112-128. Kerlinger, F. W. , & Pedhazur, E. J. Multiple regression in behavioral research . New YorFi Holt, Rinehart and Winston, 19 73. Killinger, G., & Cromwell, P. (Eds.). Penology . St. Paul, Minn.: West Publishing Co., 1973. Kinch, J. W. Continuities in the study of criminal types. Journal of Criminal Law, Criminology, and Police Science , 1962, 53, 323-328. Kinch, J. W. Continuities in the study of delinquent types. Journal of Criminology Law, Criminology and Pol ice Science , 1963, 54, 296-307. Kretschmer, E. Physique and character . London: W. J. H. Sprott, Publishers, 1925.

PAGE 137

127 Kuder, G. F., & Richardson, M. W. The theory of the estimation of test reliability. Psychometrika , 1937, 2, 151-160. Lambroso, C. Crime; Its causes and remedies . Boston, Mass.: Little, Brown and Co., 1911. Loveland, F. Classification in the prison system. In P. L. Tappan (Ed.), Contemporary corrections . New York: McGrav7-Hill , 1951, 91. Mack, J. L. The MMPI and recidivism. Journal of Abnormal Psychology , 1969, 74, 612-614. Maddocks, P. D. A five-year follow-up of untreated psychopaths. British Journal of Psychiatry , 1970, 116, 511-515. Magnusson, D. Test theory, translated by Hunter Mahon . Reading, Mass.: Addison Wesley Co., 1967. McGaw, B. L.; Wardrap, J. L.; and Burda, M. A. Classroom observation schemes: Where are the errors? American Educational Research Journal , 1972, 9 CD , 12-2T. Meehl, P. E. Antecedent probability and the efficiency of psychometric signs, patterns or cutting scores. Psychological Bulletin , 1955, 52, 194-216. Megargee, E. I. The need for a new classification system. Criminal Justice and Beh avior, June 1977, 4(2), 107-113. Magargee, E. I., & Bohn, M. J. Empirically derived characteristics of the ten types. Criminal Justice an d Behavior , June 1977, 4 (2), 149-210"^ ' Megargee, E. I.; Meyer, J.: Darhut, B.: & Bohn, M. J. A new classification system for offenders. Criminal Justice and Behavior , 1977, 4(2), 107-214. Messick, S., & Jackson, D. N. Problems in human assessme nt. New York: McGraw-Hill, Inc., 1967. Monachesi, E. P., & Hathaway, S. R. The personality of delinquents. In MMPI: Research development and clinical applications . Minneapolis: University of Minnesota Press, 1972.

PAGE 138

128 Morris, A. The comprehensive classification of adult offenders. Journal of Criminal Law, Criminology and Police Science , 1965, 36^, 197-202. Mosier, C. I. A critical examination of the concepts of face validity. Educational and Psychological Mea surement , 1957, 7, 191-205. National Institute of Mental Health. Typological approaches and delinquency control; An interim report . Washington, D.C.: U.S. Government of Health, Education and Welfare, 1967. Palmer, J., & Carlson, P. Problems with the use of regression analysis in prediction studies. Journal of Research in Crime and Delinquency , 1966, IS^Cl) , 64-79. Palmer, J. 0. The psychological assessment of children . New York: Wiley Publishers, 1970. Panton, J. H. The identification of predispositional factors in self-mutilation within a state prison population. Journal of Clinical Psycholo gy, 1965, 18, 63-67. — Panton, J. H. The longitudinal effects of first incarcera tion on MMPI profiles . Unpublished paper. Raleigh, N.C.: North Carolina Department of Social Rehabilitation, 1966. Panton, J. H. The identification of predispositional factors influencing prison adjustment . Unpublished paper. Raleigh, N.C.: North Carolina Department of Social Rehabilitation, 1968. Panton, J. H. Manual for a prison classification inventory for the MMPT ! Raleigh, N.C. : Department of Social Rehabilitation and Control. 1970. Peterson, H. R. : Quay, H. C; & Tiffany, T. L. Personality factors related to juvenile delinquency. Child Devel opment , 1961, 32, 355-372. Peterson, J.; Quay, J.; & Cameron, H. Personality and background factors in juvenile delinquency as inferred from questionnaire responses. Journal of Consu lting Psychology , 1959, 23^, 395-399. Peterson, R. A.; Pittraan, D. J.; and O'Neal, P. Stabilities in deviance: A study of assaultive and non-assaultive offenders. Journal of Criminal Law, Criminal and Police Science , March 1962. 44-48.

PAGE 139

129 Quay, K. Personality dimensions in delinquency males inferred from the factor analysis of behavior ratings. Journal of Research in Crime and Delinquency , 1964, 24, 33-37. Quay , H . C . The differential behavioral classification of the adult male offender: Interim results and pro cedures . Unpublished report, presented to the National Institute of Mental Health, Washington, D.C., 1971. Quinney, R. The social reality of crime . Boston, Mass.: Little, Brown and Co., 1970. Quinney, R. Criminology: Analysis and critique of crime in America . Boston, Mass.: Little, Brown and Co., 1972. Reckless, W. C. The crime problem (4th ed . ) . New York, N.Y.: Appleton-Century-Crof ts , 1967. Roebuck, J. The Negro numbers man as a criminal type: The construction and application of a typology. Journal of Criminal Law, Criminology and Police Science , 1963, 54, 48-60. Roebuck, J. H. Criminal typology . Springfield, 111.: C. C. Thomas Publishers, 1967. Rulon, P. J. A simplified procedure for determining the reliability of a test by split halves. Harvard Educational Review , 1939, 9, 97-103. Samenow, D. The criminal personality . New York: Wiley Publishers, 1978. Schafer, S. The criminal and his victim . New York, N.Y.: Random House, 1968. Schafer, S. Theories in criminology . Nev7 York, N.Y.: Random House Publishers, 1969. Schlapp, M. G. The new criminology: A consideration of the chemical causes of abnormal behavior . New York: N. Y. : Boni and Liverright, 1928. Schrag, C. B. A preliminary criminal typology. Pacific Sociological Review , 1961, 4, 11-16. Schrag, C. The correctional system: Problems and Prospects. The Annals, 1969, 11-20.

PAGE 140

130 Schulman, VJ. J. Personality and behavior characteristics of assaultive patients (Doctoral dissertation, University of Minnesota, 1975) . Dissertation Abstracts International, 1975, DAI 30:56953. Schulsinger, F. Psychopathy, heredity and environment. International Journal of Mental Health , 1972, 22 , 190-206. Sechrest, L. Incremental validity: A recommendation. Educational and Psychological Measurement , 1963, 23 , 153-158. Sheldon, W. H. Atlas of man. New York, N.Y.: Harper and Row, 1949. Sheldon, W. H. Varieties of delinquent youth: An introduc tion to constitutional psychiatry . New York, N.Y.: Harper and Row, 19 54. Shoham, S . ; Gutmann, L. : & Rahav, G. A two-dimensional space for classification of legal offenses. Journal of Research in Crime and Delinquency , July 1970, 219-243. Sokal, R. R. Classification: Purposes, principles, progress, prospects. Science, 1974, 185 (4157 ) , 1115-1123. Stanley, J. C. Reliability. In R. L. Thorndike (Ed.), Educational Measurement (Rev. ed . ) . Washington, D.C.: American Council in Education, 1969. Stein, K. B. ; Vadum, A. C . ; & Serbin, T. Socialization and delinquency: A study of false positive and false negatives in prediction. Psychological Record , 1970, 2^(3), 353-364. Stewart, D. K. , & Love, W. A. A general canonical correlation index. Psychological Bulletin , 1968, 7_0, 160-163. Sutherland, E. H. , & Cressey, D. R. Principals of criminology (7th ed.). Philadelphia^; Penn . : Lippincott and Co. , 1966. Tappan, P. W. Who is the criminal? American Sociological Review , 1967, 12, 96-102. Thrasher, F. M. The gang . Chicago, 111.: University of Chicago Press, 1963.

PAGE 141

131 Tinun, N. J. Multivariate analysis with applications in education and pyschology . Bellmont, Calif.: Woodsworth Publishers, 1975. Vald, G. B. Theoretical criminology . New York, N.Y.: Oxford University Press, 1958. Warren, M. Q. Classification of offenders: An aid to efficient management and effective treatment. Journal of Research in Crime and Delinquency , 1969, 62, 239-291. Widom, C. S. An empirical classification of female offenders. Criminal Justice and Behavior , 1978, 8 CD , 35-48. Wilkins, L. T. , & Smith, P. M. Predictive attribute analysis. In N. Johnston, L. Savitz, & M. VJolfgang (Eds.), The sociology of punishment and correction (2nd ed.). New York: John Wiley and Sons, 1974.

PAGE 142

BIOGRAPHICAL SKETCH Brainard Willem Hines was born April 4, 1945, in Maxton, North Carolina. He moved with his parents to Charleston, West Virginia, in 1949, and attended elementary, junior high and high school in that city. He attended West Virginia University from 1963 until 1969, and obtained a Bachelor of Arts degree in psychology, as well as a Master of Science degree in clinical psychology. From 1969 until 1974, he worked as a Program Evaluator at the Appalachia Educational Laboratory, and also was employed as Psychologist at the Charleston Guidance Clinic. He moved to Gainesville, Florida, to attend the University of Florida and to obtain a doctoral degree in foundations of education. He is currently residing in Miami, Florida. 132

PAGE 143

a. a dissertation for th^ o^f SSc?o?^^f ^?h'l?o"sX^: William a. ware. Chairman Professor of Foundations of Education as r^^^l^]^ ^Scto?''^f^?.^?o%^J^: Professor of Foundations of Education Ari W/ y . \. ^Ki chard M. Swan son — Professor of Psychology June 19 80 ^ ^ chairman. Foundations of Education" Deau, Graduate School