Title Page
 Table of Contents
 List of Tables
 Biographical sketch

Group Title: developmental relationship between neuropsychological and achievement variables
Title: The developmental relationship between neuropsychological and achievement variables
Full Citation
Permanent Link: http://ufdc.ufl.edu/UF00097462/00001
 Material Information
Title: The developmental relationship between neuropsychological and achievement variables a cluster analytic study
Physical Description: x, 99 leaves : ; 28 cm.
Language: English
Creator: Schauer, Charles Arnold, 1949-
Publication Date: 1979
Copyright Date: 1979
Subject: Neuropsychology   ( lcsh )
Reading disability   ( lcsh )
Psychology thesis Ph. D   ( lcsh )
Dissertations, Academic -- Psychology -- UF   ( lcsh )
Genre: bibliography   ( marcgt )
non-fiction   ( marcgt )
Thesis: Thesis--University of Florida.
Bibliography: Bibliography: leaves 91-98.
Additional Physical Form: Also available on World Wide Web
Statement of Responsibility: by Charles A. Schauer.
General Note: Typescript.
General Note: Vita.
 Record Information
Bibliographic ID: UF00097462
Volume ID: VID00001
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
Resource Identifier: alephbibnum - 000097484
oclc - 06572385
notis - AAL2924


This item has the following downloads:

developmentalrel00scharich ( PDF )

Table of Contents
    Title Page
        Page i
        Page ii
        Page iii
    Table of Contents
        Page iv
    List of Tables
        Page v
        Page vi
        Page vii
        Page viii
        Page ix
        Page x
        Page 1
        Page 2
        Page 3
        Page 4
        Page 5
        Page 6
        Page 7
        Page 8
        Page 9
        Page 10
        Page 11
        Page 12
        Page 13
        Page 14
        Page 15
        Page 16
        Page 17
        Page 18
        Page 19
        Page 20
        Page 21
        Page 22
        Page 23
        Page 24
        Page 25
        Page 26
        Page 27
        Page 28
        Page 29
        Page 30
        Page 31
        Page 32
        Page 33
        Page 34
        Page 35
        Page 36
        Page 37
        Page 38
        Page 39
        Page 40
        Page 41
        Page 42
        Page 43
        Page 44
        Page 45
        Page 46
        Page 47
        Page 48
        Page 49
        Page 50
        Page 51
        Page 52
        Page 53
        Page 54
        Page 55
        Page 56
        Page 57
        Page 58
        Page 59
        Page 60
        Page 61
        Page 62
        Page 63
        Page 64
        Page 65
        Page 66
        Page 67
        Page 68
        Page 69
        Page 70
        Page 71
        Page 72
        Page 73
        Page 74
        Page 75
        Page 76
        Page 77
        Page 78
        Page 79
        Page 80
        Page 81
        Page 82
        Page 83
        Page 84
        Page 85
        Page 86
        Page 87
        Page 88
        Page 89
        Page 90
        Page 91
        Page 92
        Page 93
        Page 94
        Page 95
        Page 96
        Page 97
        Page 98
    Biographical sketch
        Page 99
        Page 100
        Page 101
        Page 102
Full Text








Finding appropriate words to express the depth of my

gratitude has proven difficult. Paul Satz, Jacqueline

Goldman, Eileen Fennell, Carol Van Hartesveldt, and Randy

Carter have been superlative committee members. They have

allowed professional growth at my own rate, yet have

always been there for stimulation, encouragement, techni-

cal assistance, and to foster new ideas. Individually and

as a committee they represent the highest professional

integrity combined with tremendous human warmth. Their

broad contributions to my development are greatly appreci-

ated. I especially thank Paul Satz for all he did as

Chairman, for setting the tone of the committee, and for

his support and guidance on the dissertation. His dedi-

cation to and concern for students are unparalleled. I

feel honored to have been one of them and saddened to be

his final University of Florida doctoral candidate.

So many others have made significant contributions,

but only a few can be acknowledged here. Molly represents

so much to me and has helped tremendously in my personal

growth. Ellie and Mike have been such good friends. The

computer analyses could not have been done without Robin.

Andy helped maintain my calm and perspective during the

grueling final months. And throughout the years of study,

my family has provided constant love, support, and

encouragement. Thank you all.








Overview 1
Nonstatistical Classification Systems 9
Statistical Classification Systems 23
Summary and Rationale 33


Subjects 38
Procedure 39
Test Measures 39
Statistical Analysis 41


Cluster Analytic Solution 46
Primary External Validation 55
Influence of Cluster Types 61
Influence of Age 63
Secondary External Validation 64
Intelligence 64
Parental Reading Levels 67
Neurological Examination 68
Socioeconomic Status 71


Internal Validation 74
Primary External Validation 75
Secondary Extenal Validation 83





Table Page

1 Mean Factor Scores By
Neuropsychological Subgroups. . . . . ... 48

2 Correlations Between Achievement
Measures. . . . . . . . . . ... 56

3 Mean WRAT Scores (Probe 3)
By Neuropsychological Subgroups . . ... . 57

4 Mean PPVT IQ Scores, Maternal
and Paternal Reading Levels by
Neuropsychological Subgroups. . . . . ... 65

5 Neurological Ratings By
Neuropsychological Subgroups. . . . . ... 70

6 Socioeconomic Status By
Neuropsychological Subgroups. . . . . .. 72


Figure Page

1 Subgroup Factor Scores for Clustering
Variables and Standard Scores for
Achievement Variables . . . . . ... .49

2 Configuration of Neuropsychological
Subgroup Means for Clustering Variables . . 54

3 Configuration of Neuropsychological
Subgroup Means for WRAT Scores at
Probe 3 . . . . . . . . ... .. . 59

4 Configuration of Neuropsychological
Subgroup Means for IQ Scores (Probe 3)
and Maternal Reading Level. . . . . .. 66

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



Charles A. Schauer

December 1979

Chairman: Paul Satz
Major Department: Psychology

Current discussions of specific reading disabilities

increasingly mention the heterogeneity of this population

and the need for ways to delineate meaningful subgroups.

This study attempted to respond to this need by developing

a multivariate classification system based on neuropsy-

chological subskills which was then related to achievement

measures. A unique characteristic of this study is its

longitudinal approach. Futhermore, the classification

system is based on the performance of children from all

skill levels in order to verify that specific deficits are

unique to single categories of achievers.

Data were collected from 211 boys first during their

kindergarten year of school. They were then re-evaluated

twice on the same measures at subsequent 3-year periods;

the entire study spanned 6 years. The basic test measures

at each probe included the Developmental Test of Visual-

Motor Integration. The Recognition-Discrimination Test,

the Embedded Figures Test, Peabody Picture Vocabulary Test

raw score, WPPSI/WISC Similarities, the Verbal Fluency

Test, and the Dichotic Listening Test (total recall.).

From an independent factor analysis of the Satz Predictive

Battery these measure were found to be factor-pure and

strong contributors to particular factors. They are also

untroubled by floor and ceiling effects. Consequently,

they were combined into spatial-perceptual and conceptual-

linguistic factor scores at each of the three probes.

These neuropsychological factor scores were then submitted

to a cluster analytic procedure which resulted in 12

distinctive patterns of performance. Internal and

external validity were verified through a series of

statistical procedures.

Results confirmed a direct relationship between

neuropsychological subgroups and achievement. Subgroups

with the best neuropsychological performance were also the

best achievers, just as those most deficient neuropsycho-

logically were also the lowest achievers. Examining the


impact of each factor type, however, revealed much more

complex relationships. One subgroup with a specific

perceptual deficit had very low achievement. Another

subgroup deficient only in verbal skills had low reading

and spelling scores and average arithmetic skills. At the

opposite end of the continuum, subgroups with relative

strengths in either perceptual or verbal skills had very

superior achievement scores. For all levels of skills,

however, the impact on achievement is greatest when

neuropsychological skills are equivalent.

Clear developmental changes were evident in three

subgroups. For two subgroups the changes were regressive.

The third subgroup showed consistent improvement in

neuropsychological subskills. For these subgroups, the

eventual level of performance, rather than the develop-

mental patterns, seemed most clearly related to

achievement competency.

A trend was apparent for a relationship between

perceptual skills and arithmetic achievement. From other

analyses, it was determined that the subgroups which were

lowest on neuropsychological skills were also rated as

having lower socioeconomic backgrounds and as having more

frequent positive neurological findings. In addition, the

mothers of these children were also the poorest readers.

No significant relationship was found between fathers'

reading scores and the subgroups.

The results of this study underscore the benefits

from examining children within the context of a carefully

generated classification system. Evidence contrary to

unitary deficit models was discussed, as were implications

for future research.




Consider the following two quotes:

More than fifteen thousand articles on the teaching of
reading have appeared in professional journals in the
last forty years; failure in reading is the largest
single cause of school failure during the grade school
years. With so much attention devoted to the teaching
of reading, it seems anomalous that the problem of the
nonreader remains with us. (Smith & Carrigan, 1959,

Despite over 30 years of research on specific reading
and writing disabilities in children with normal
intelligence, the results are contrary as indicators
of either causes of the disabilities or remedial
programs that can help persons to overcome them.
(Valtin, 1979, p. 204)

Taken at face value, these statements present a dim

view of the previous 60-year accomplishments in the areas

of reading and, more specifically, reading disabilities,

despite highly competent professionals having spent

innumerable hours, manpower, and enormous sums of money in

clinics, tutorial, and research programs. Admittedly, the

prevailing opinion in not one of complete futility, but any


optimism (Benton, 1978) comes at the expense of restricting

expectations to progress of a less wholistic nature. The

fact remains that we do not have specific and effective

remediation methods for failing readers. Nor do we have a

satisfactory understanding of how reading disability

relates to the overall process of reading.

It is legitimate to ask whether research should

continue for reasons other than a "thirst for knowledge."

From a practical point of view, making such a decision

must include considerations of prevalence and consequences

of reading failure. Due largely to differences in

diagnostic criterion and age categories being studied,

estimates of the number of children whose reading skills

are significantly below expectation for their

chronological and mental age have varied from 10 to 30

percent of the population (Applebee, 1971; Kline, 1972;

Benton, 1975; Gibson & Levin, 1975; Rutter, Tizard, Yule,

Graham, & Whitmore, 1976). Awareness of the impact on

this staggering number of children is heightened by an

understanding that theirs is more than a simple inability

to process written symbols. Quite often other areas of

academic achievement are also deficient, so that there is

a high frequency of referral to child clinics (Meerloo,

1962; Mendelson, Johnson, & Stewart, 1971) if not complete

school dropout (Silberman, 1964). Additional followup

studies demonstrate a relationship between learning

disabilities and severe behavioral or emotional disorders

(Robins, 1966; Rutter et al., 1976; Watt & Lubensky, 1976)

and adult criminal behavior. Wright (1974) went so far

as to postulate a theory in which reading failure is the

single most significant factor in those forms of

delinquency described as being antisocially aggressive.

The hope remains that such significant distress and loss

of human productivity could be lessened if adequate

remediation techniques were available; to that end

research continues.

In examining the reasons for the limited progress in

this area, it might prove advantageous to briefly look at

the lines of research in reading; as elaborated by Gibson

and Levin (1975), there are essentially two. For

approximately the first quarter of the twentieth century,

experimental and educational psychologists attempted to

understand the process of reading, raising many of the

basic problems still confronting us today. Around 1920

the focus of research changed dramatically by attempting

to compare the value of one method of teaching reading to

another. What resulted was a mass of confusion with

advocates of each curriculum attacking the other. It only

took 40 years -- until about 1960 -- for us to realize

that this was fruitless and that the concern should be

with the process of reading rather than the outcome of a

method. With the "swing of the pendulum" back to this

earlier trend, the integration of information from

perceptual learning and cognitive and linguistic

development would seem to increase our chances of

understanding what has gone wrong for those children who

do not develop adequate reading skills.

In what could be described as a subtrend of reading

research, the literature has fairly mushroomed on those

children who fail to learn to read effectively despite

adequate intellectual, physical, and social motivational

development. The terminology for this group includes such

words as dyslexia, specific reading retardation, specific

learning disability, unexpected reading failure, and

specific developmental dyslexia, to choose just a few.

Recent reviews (Satz, 1977; Rutter, 1978; Eisenberg, 1978;

Valtin, 1979) have examined the issues involved in the

various definitions. What is of importance to this

discussion is that all definitions have proven

inadequate. Fundamental disputes about the nature of the

reading problems have resulted in circular and vague

definitions, definitions by exclusion, and a looseness in

the use of terms. Ross (]976, p. 11) stated the problem

like this:

Stripped of those clauses which specified what a
learning disability is not, this definition is

circular, for it states, in absence, that a learning
disability is an inability to learn. It is a reflection
on the rudimentary state of knowledge in this field that
every definition in current use has its focus on what the
condition is not, leaving what it is unspecified and thus
ambiguous. Furthermore, when defined in this manner
"learning disability" is a heterogeneous category;
progress in this field demands further refinement of the
definition and an identification of subcategories.

The interest of professionals from education to

medicine is a mixed blessing as the role of variables of

interest to a single professional discipline is emphasized

and restricted conceptual models are constructed

(Applebee, 1971). As a direct consequence of definitional

inconsistencies, different criteria are used in sample

selection, uncertainty exists as to what, if any,

comparison group should be used (Valtin, 1979), and

inferences across studies become tenuous. The result has

been further chaos.

Recently there has been discussion of whether or not

concepts such as developmental dyslexia are even valid. A

recent study by Taylor, Satz, and Friel (1979) cuts to

the heart of this issue. Carefully satisfying

definitional requirements, comparisons were made between

dyslexic and nondyslexic disabled readers, and both of

these with normal readers. Results indicate that the two

categories of poor readers could not be distinguished

along any of several dimensions, including severity and

outcome of reading disturbance, frequency of reversal

errors, parental reading and spelling competencies, math

skills, neuro-behavioral performance, or personality

traits. The only significant differentiation was between

the total group of disabled readers and the group of

normal readers, thereby challenging the dissociation of

dyslexics from other forms of reading problems. The

authors attribute a large portion of the current confusion

in the area to use of such presumed concepts as dyslexia.

What is needed is the willingness among researchers

to retreat from historical trends so they can develop an

objective framework in which to classify and differentiate

the numerous types of learning deficiency. The focus,

however, should remain on matters of first-order relevance

(i.e., an objective and descriptive classification of

learning disorders) which can later be investigated on an

explanatory level (i.e., second-order relevance). A major

obstacle during the initial phase is the tendency to

confound classification with etiology in which the

"causal" event is loosely inferred from the minor signs

(often behavior) or is based on observable inner events

which may or may not exist (Satz, 1977). A descriptive

classification schema would have the significant advantage

of avoiding the use of hypothetical constructs. It would

also operationally specify those variables which must be

present to define the disorder. Such achievement-oriented

behaviors are measurable and allow one to specify the type

of special handicaps (e.g. reading, writing, arithmetic,

perception, speech, etc.) under consideration.

By searching for naturally occurring subtypes of

learning disorders, one has more or less, rejected the

unitary deficit hypothesis. Such a position identifies a

single type of reading difficulty having a single radical

cause (Wiener & Cromer, 1967; Applebee, 1971). Examples

include Smith and Carrigan's (1959) synaptic transmission

theory of reading disability, Delacato's (1959) theory of

central neurological organization, Cruickshank's (1972)

perceptual deficit hypothesis, and Vellutino's (1978)

verbal mediation hypothesis. Examinations of the

empirical and theoretical bases for such positions have

proven them untenable, adding further impetus to the

search for subtypes of learning disabilities (Benton,

1978; Doehring, 1978; Fletcher & Satz, 1979).

Because reading is a highly complex activity which

utilizes a great number of skills (Maliphant et al., 1974;

Gibson & Levin, 1975) and exhibits a strong relationship to

other academic subjects (Rutter, 1978), the researcher is

free to choose the level at which he will differentiate the

syndromes (Boder, 1973; Rutter, 1978). One approach is to

examine the actual pattern of reading deficit in isolation

or as it relates to other academic skills. Although quite

difficult, attempts have been made to classify retarded

spellers (Sweeney & Rourke, 1978), spelling and reading

errors (Ingram et al., 1970; Naidoo, 1972; Nelson &

Warrington, 1976), or reading, spelling, and math

performance (Rourke, 1978). The second approach is to

focus on those skills which have been shown to be important

in a child's academic development and which are felt to be

related to reading acquisition. Examples would be

differentiation based on some type of cognitive or motor

skills, disabilities in oral language, perceptual capacity,

or short-term memory. Such a search for patterns of

performance (and not just deficits) which may be associated

with particular levels of reading proficiency is not meant

to provide etiological statements, but rather is a

quantitative attempt to describe patterns of skills at a

more microcosmic level since any one of several clusters

of deficiencies can limit the development of reading

skills (Mattis, French, & Rapin, 1975).

Once the assumption of heterogeneity of learning

disabilities has been adopted and the dimension of

differentiation specified, the researcher must decide on

the technique for determining the "correct" number of

subgroups and for assigning individual children to the

cluster types. For the purposes of this paper, an

important distinction is whether the classification schema

is derived through the use of high speed computers and

multivariate statistical procedures or via visual

inspection of the data. Studies using the latter technique

will be reviewed first and more briefly because of their

weaker methodology.

Nonstatistical Classification Systems

An important study historically is that by Monroe

(1932) because of the attempt to decrease the complexity

and variability of reading disability cases by identifying

subgroups on the basis of the nature and method of-their

being referred. One group was referred exclusively

because of deficient reading. The second group had a

variety of behavior and environmental problems, in

addition to the poor reading, and were often referred by

social agencies or the juvenile court. The final group

was composed of children with borderline or deficient

intelligence. In order to be able to make comparisons

between groups, Monroe developed a reading index by

comparing each child's composite reading grade with his

average chronological, mental, and arithmetic grade. The

three groups of reading deficiency cases were found to

have highly similar distributions of scores and

eventually had to be placed together in a single

distribution. Since subgroups based on type of referral

did not prove useful, Monroe examined the frequency of

ten patterns of errors (reversals, addition or omission

of sounds, substitution, repetition, etc.) for the

combined reading deficiency group in comparison to

controls. From this, she discussed the various defects

which could give rise to reading disability, the pattern

of errors resulting, and the remediation methods

recommended for children with each type of error pattern.

In a series of articles based on the Isle of Wight

Studies of an entire population (Rutter & Yule, 1975;

Rutter et al., 1976; Yule & Rutter, 1976), a broad

classification system is first described, depending on

whether the child fails to ever acquire reading skills or

if he somehow loses the skill after it seemingly was well-

established. Little is known of this latter group because

its relative infrequency discourages extensive research.

Those children who never learned to read were further

subdivided into specific reading retardation (disabled

independent of IQ) and general reading backwardness (low

achievement in relation to age regardless of IQ).

Multiple regression techniques were used to assign

children to the categories, but the distinction rests on

the congruence between the child's age, intelligence, and

reading attainment. Even though this classification

system is reported to be valid and as having educational

implications, it effectively does little more than

establish criteria for identifying a child as being

deficient in reading.

Critchley (1964) and Rabinovitch (1968) advocate a

common classification system based on proposed etiology.

Children are categorized as having secondary reading

retardation when the reading disability is associated with

an encephalopathic event and primary reading retardation

when there is a family history of learning disability

without clear evidence of brain damage. The group of

children with secondary reading retardation is shown to

have a better prognosis. The distinction is essentially

the basis for diagnosing specific developmental dyslexia

and, as such, is subject to the difficulties of definition

by excluding other possible pathologies.

From an examination of the literature, Bannatyne

(1971) developed a hierarchical classification of the

causes and types of all language and reading disabilities.

The first breakdown creates the major groupings of

dyslexia, aphasia, autism, emotional disturbance, low

intelligence, and "Others." The dyslexics are then further

divided into these causal groups: (1) primary emotional

communicative causes related principally to maternal

pathology; (2) minimal neurological dysfunction

characterized by perceptual or cognitive deficits; (3)

social, cultural, or educational deprivation; and (4)

genetic dyslexia. The categories are not mutually

exclusive, with characteristics being classified instead of


Another theoretical classification of the different

types of poor readers might be carried out by differ-

entiating individual reading performances (Vernon, 1977).

The first group would be those children who cannot read at

all. In addition to a measure of general linguistic

development, these children would be studied to evaluate

whether they can analyze or memorize word structures

visually or auditorily. The second group, who can read

only a few simple words and appear incapable of compre-

hending phonic teaching, should be studied for dif-

ficulties in assigning visual and symbolic material,

especially in relation to hemispheric functioning and

integration. Next, those children who can read simple

regular words but do not understand how to manipulate

irregular grapheme-phoneme correspondences would be tested

for deficiencies in conceptual reasoning. And finally,

for those readers who can read single words but cannot

group words syntactically into phrases, Vernon recommends

investigation of conceptual deficiencies and visual

sampling ability. As should be evident, no direct

assessment has been made of the validity for either the

basis of the classification or the varying methods for

studying the groups.

Interested in the developmental persistence of

neurological signs, Silver and Hagin (1960, 1964) examined

24 dyslexics with a battery of neurological and psycho-

logical tests on two occasions with an interval of 10 to

12 years. When originally tested, the children ranged

from 8 years, 6 months to 14 year of age, with intel-

ligence quotients from 81 to 123. All of the children had

been referred primarily for behavior disorder. In the

first study, three groups of specific reading disability

were identified: (1) a developmental group; (2) an organic

group with evidence of structural organic defect; and (3)

a very small group showing no perceptual deficits ot

signs. At follow-up, 15 of this sample were deemed to be

adequate readers, and they tended to come from the

"developmental" group and to be less severely retarded as

children. The "organic" group showed less improvement

than the others, and their specific perceptual deficits

and lack of clear cerebral dominance tended to persist.

No explanation was given as to the derivation of the three


Zigmond (1978) was also concerned with the course of

reading disability, but from an educator's point of view.

She suggested that there is a sizeable number of "problem

readers" who will achieve competency in reading if only

teaching is improved. This means that there will remain a

small minority of learners for whom attempts at achieving

reading competence will be unsuccessful. Zigmond would

only define this latter group as "dyslexic." Challenging

researchers to sort these two groups on medical,

psychological, and educational characteristics, she

concluded that individualized instruction seems to be the

most promising remedial approach.

Other discussions of reading disability subgroups

have used performance on single, objective tests to

determine subgroup membership. For Kinsbourne and

Warrington (1963), the differentiation was between Group

1, in which the score on the nonverbal section of the WISC

(or WAIS) exceeded that of the verbal section by at least

20 points, and Group 2, in which the converse was true.

There were six patients in Group 1 and seven in Group 2.

Their ages ranged from 8 to 14 years, with the exception

being a 31 year-old patient in Group 1. Additional

evaluations indicated that Group 1 patients were

characterized by delays in the acquisition of speech and

in clinically apparent difficulties in verbal compre-

hension and expression, symptoms analogous to aphasia in

adults. Group 2 children demonstrated poor performance on

tests of finger order and differentiation, significant

retardation in right-left orientation and mechanical

arithmetic, as well as constructional difficulty; this

grouping of symptoms is comparable to the Gerstmann

syndrome in adults. The authors concluded that among

backward readers and writers there exist two groups with

developmental defects reminiscent of adult acquired

cerebral syndromes and called them the "language retard-

ation" (Group 1) and the "Gerstmann" (Group 2) groups.

Smith (1970) also used intellectual abilities to

delineate between subgroups of educationally handicapped

children. Having tested 300 Anglo, male school children

with severe reading retardation despite at least average

Verbal or Performance Intelligence, Smith ascertained

patterns of specific abilities and weaknesses for each

individual by regrouping WISC subtest scores into

categories suggested by previous studies. Three patterns

of intellectual functioning were delineated in this group

of children, with no similar patterns emerging from 74

control subjects. Pattern I individuals had strength in

Spatial Ability and Spatial Organization, earned lower

scores in Symbol Manipulation than in Spatial Organi-

zation, and were deficient in Sequencing Ability. Pattern

II had deficits in Spatial Organization and/or Perceptual

Organization and frequently had deficits in Visual-Motor

Coordination. Pattern III had characteristics of both

Patterns I and II. The proportion of Pattern I

individuals tended to increase as the age of the children

increased. In Pattern II individuals, the opposite

tendency was apparent: 40 percent of the group at age 6,

but only 8.7 percent by age 11. The incidence of Pattern

III increased until age 13, then decreased rapidly.

Similarities were drawn between Pattern I and Bannatyne's

(1971) genetic dyslexia and Boder's (1973) dysphonetic

dyslexia. The variety of deficits in spatial ability for

Pattern II is also comparable to that in Bannatyne's minor

neurological dysfunction dyslexia.

The preceding notion of performance deficits changing

with age (Smith, 1970) is compatible with recent

formulations advanced by Satz and associates (Satz &

Sparrow, 1970; Satz, Rardin, & Ross, 1971; Satz & Van

Nostrand, 1973). This latter position postulates a

differential relation between cognitive variables based on

age. Briefly, at earlier ages learning disabled children

are hypothesized to be delayed in visual-perceptual

skills and at later ages, during failing reading, more

delayed in cognitive-linguistic skills. Although the

theory highlights the importance of both types of skills

at all ages, it posits two subgroups of learning disabled

children, with the primary deficit being perceptual for a

younger group and linguistic for the older group. Support

for this theory has been variable (Fennell, 1978;

Fletcher, 1978), with concurrent measures being more

favorable than longitudinal evaluations. The main problem

is that there may not be a complete transition from the

first to the second subgroup, creating the likelihood of

at least a third subgroup deficient in both areas.

Nevertheless, theories such as this reinforce the position

that age-dependent concepts either have relevance in the

context of subgroups or must be considered in delineating

those subgroups. The necessity of including developmental

or longitudinal parameters has been reiterated (Money,

1962; Rice & Mattsson, 1966; Satz & Sparrow, 1970,

Sontag, 1971) but remains largely ignored (Doehring,

1978; Satz, 1978).

A series of studies by Rourke and associates (Rourke

& Finlayson, 1978; Rourke & Strang, 1978) examined.

patterns of academic performance in relation to

neuropsychological variables. Forty-five 9- to 14-year

old children with learning disabilities whose WISC Full

Scale IQs fell within the range of 86 to 114 were divided

into three groups. Group 1 was composed of children who

were uniformly deficient in reading, spelling, and

arithmetic; children in Group 2 were relatively adept at

arithmetic as compared to reading and spelling; Group 3

was composed of children whose reading and spelling

performances were average or above, but whose arithmetic

was relatively deficient. Before discussing results, it

should be noted that the composition of each group is

obscured by overlapping criteria. For example, there is

no statistically significant difference in arithmetic

performance between Groups 2 and 3, despite the level

being described as "adept" for Group 2 and "deficient" for

Group 3. The authors skirt this issue by arguing for

patterns of performance instead of levels of performance.

Combining the results from both studies, the per-

formances of the children in the three groups were

compared on a total of 27 dependent neuropsychological

measures. The performances of Groups 1 and 2 were

superior to that of Group 3 on measures of visual-

perceptual, visual-spatial, some psychomotor, and tactile-

perceptual skills. There were no statistically

significant differences between the groups on "motor"

measures. Group 3 performed at a superior level to.that

of Groups 1 and 2 on measures of verbal and auditory-

perceptual abilities. These studies highlighted the

number of adaptive deficiencies which should render

children such as in Group 3 the focus of more serious

concern, especially because their pattern of

deficits is analogous to that seen in the Gerstmann

syndrome (Kinsbourne & Warrington, 1963). Furthermore, the

conclusion was made that subjects in Group 3 performed very

much as would be expected were they to have a relatively

dysfunctional right cerebral hemisphere, and subjects in

Groups 1 and 2 performed in a fashion similar to that

expected were they to have a relatively dysfunctional left

cerebral hemisphere.

From their work in a learning disability clinic, Cole

and Kraft (1964) observed patients falling into groups

with differing neuropsychological defects. In a sample

of 36 children without behavior disorders or mental

retardation, five groups emerged based on various

combinations of these characteristics: dyslexia, abnormal

speech on examination and/or history of retarded speech

development, and visuo-spatial constructional

abnormality. The five groups were defined as (1)

dyslexia with general language defect, (2) dyslexia with

visuo-spatial defect, (3) dyslexia without general

language or visuo-spatial defect but with abnormalities

of synthesis, (4) dyslexia with mixed defects, and (5)

specific learning disability without dyslexia. One

problem with this classification system is the small size

of some groups (as low as 4 for Group 2) presenting

serious questions of reliability. Nevertheless, beyond

the definitional criteria, intergroup variations exist

for the ratio of males to females, IQ distribution,

handedness, and right-left disorientation. Basic proper-

ties of all groups include family history of learning

disability, sinistrality or ambilaterality, and incidence

of abnormal neurological signs. Cole and Kraft concluded

that the similarities are great enough to define the

global syndrome of limited cerebral dysfunction of

childhood, a heredo-familial abnormality of cerebral


Currently, one of the most popular approaches is the

differentiation made by Mattis, French, and Rapin (1975)


and Denckla (1972, 1977). After an initial attempt to find

a difference between brain-damaged and non-brain damaged

dyslexics failed, Mattis et al. (1975) determined in a post-

hoc observation that 90 percent of the 82 dyslexic

(combined brain-damaged and non-brain-damaged) children

could be isolated into three independent dyslexic

syndromes. The children were aged 8 to 18 years, had a

Verbal or Performance IQ exceeding 80, and were being

evaluated at a neurological clinic for a learning or

behavior disorder. Denckla's (1972) subjects were also

seen in a neurology clinic, but no additional descriptive

data are provided. Of her last 190 private patients,

about 30 percent were seen as having an easily

recognizable dramatic cluster of signs. No explanation

was given as to the derivation of the clusters.

Although not modeled after the work of Cole and Kraft

(1964), the basic dyslexic syndromes of Denckla (1972) and

Mattis et al. (1975) are remarkedly similar. These

syndromes include: (1) a language disorder syndrome with

defects in both the understanding and expression of oral

language; (2) a dyscoordination syndrome with defects in

speech articulation and design copying and normal

understanding of oral language; and (3) a visuospatial-

perceptual disorder syndrome with defective visuo-

perceptive and visuoconstructive capacity and intact oral

language abilities. To this basic classification system

can be added the dysphonemic sequencing disorder syndrome

characterized by phonemic substitutions and missequencings

despite normal naming, comprehension, and articulation.

This syndrome was described by Denckla (1977) and has been

found in a cross-validation study of Mattis' work

(Erenberg, Mattis, & French, 1976). There is also a

verbal memorization disorder syndrome (Denckla, 1977) with

noteworthy impairment in sentence repetition and verbal

paired-associates learning in older children, and which

may be a milder and continuing form of the language

disorder syndrome of younger dyslexics.

A second frequently discussed classification system is

that described by Johnson and Myklebust (1967), Boder

(1970, 1971, 1973), and Tallal (1976). If the overall

differentiation between general and specific dyslexia is

overlooked, the criteria for the distinction used by

Ingram et al. (1970) is also consistent with this

system. Ingram et al. (1970) did a retrospective study

of 82 "highly pre-selected patients" who meet the common

definition of dyslexia. Boder (1973) had 107 children

between 8 and 16 years of age in her sample. From an

analysis of how the child reads and spells rather than at

what grade level, 100 of the 107 children exhibited one of

the three reading-spelling patterns. Dysphonetic or

auditory dyslexia reflects a primary deficit in symbol-

sound (grapheme-phoneme) integration resulting in an

inability to develop phonetic word analysis-synthesis

skills. Dyseidetic or visual dyslexia is marked by an

inability to perceive letters and whole words as

configurations, or visual gestalts. Finally, there is a

mixed dyslexia group with both deficits. Indirect

evidence supporting the validity of the dysphonetic and

dyseidetic clinical subgroups has appeared from

electrophysiological recordings of event-related

potentials (Fried, Tanquay, Boder, Doubleday, &

Greensite, 1979), but, in large part, use of such

dependent measures could not differentiate between

dyslexic subgroups. Children with audiophonological

errors are more frequently encountered than those with

visuospatial errors. In general, too, the outlook for

improvement in reading level appears to be better for the

child with audiophonological difficulties. None of the

three patterns have been found among normal readers.

The last two classification systems are good examples

of the differing approaches, one based on the analysis of

reading performance per se (Boder, 1973) and the other on

the occurrence of concomitant disabilities (Mattis et al.,

1975). It is possible that systematic relationships

exist, for example, between Boder's visuospatial type of

defective reading and the visuospatial-perceptual disorder

syndrome of Mattis, or between Boder's audiophonological

type and Denckla's dysphonemic sequencing disorder

syndrome. Investigative work for such similarities could

jointly consider the two approaches to classification by

analyzing reading performance and concomitant disabilities

(Benton, 1978) but this overlooks basic and pernicious

methodological limitations. In every classification study

reviewed up to this point, the rationale for the particu-

lar subcategories comes exclusively from the developer's

theoretical system. Such a priori grounding inherently

imposes biases sympathetic to the particular theoretical

position, etiological inferences, or investigator's goals

(Myklebust, 1968; Blom & Jones, 1970). Too often, then,

the classifications are treated as fact rather than an

hypothesis to be tested, subjecting the categories to

serious questions of validity. Furthermore, the method of

assigning children to categories has been based solely on

visual inspection of quite complex data sets. At a time

when rigorous and systematic statistical procedures are

available, such a practice is an unnecessary source of

error variance and demonstrates poor experimental design.

Statistical Classification Systems

In 1959, Smith and Carrigan published a study far

ahead of its time, as indicated by the early use of

advanced statistical techniques and the extent to which

their fellow researchers ignored their work. Thirty-two

retarded readers in the third through sixth grades were

given a battery of tests providing 18 scores on measures

from digit span to flicker fusion. Using a cluster

analysis technique, five subgroups were derived for

additional study. Group 1 was low in both cognitive-

associational skills and in perceptual-metabolic skills.

Group 2 was evenly balanced with neither high nor low

skills except for cognitive-perceptual skills, which were

low. Group 3 did not present a clear pattern and Groups 4

and 5 appeared to be superior to the other groups. Using

a variety of measures including height, weight, blood

pressure, bone age hemaglobin, protein-bound iodine, etc.,

an attempt was made to see if the physical characteristics

of the five subgroups conformed to theoretical expecta-

tions. The results were largely negative. Next, a

complex analysis was presented of the profiles of the five

subgroups in terms of the effects of varying amounts of

the 'two chemical substances acetylcholine and cholin-

esterase. The authors concluded that both the absolute

amount of the chemical present and the relative balance of

the two chemicals is important. Another part of the study

found subgroup differences along an anxiety dimension.

The study has a number of serious defects, including

a small sample size, lack of replication, lack of normal

controls, and use of fairly unreliable instruments.

Furthermore, the authors failed to describe the "cluster

analysis" technique, so one can only guess how it may have

been carried out at a time when the use of computers was

limited. Yet it is still interesting that a series of

subgroups did emerge from the analysis, subgroups which

showed a number of significant differences on a series of

measures unrelated to those by which they were originally


Naidoo (1972) also used a cluster analysis to examine

the possibility of subtypes within a clinic sample of 94

dyslexic boys, aged 8 to 13 years. She compared the boys

on a variety of data from the psychological and

neurological examinations, perinatal and developmental

histories, and family history information on reading,

spelling, speech difficulties, and left-handedness. Five

groups were identified using a single linkage cluster

analysis, the groups consisting of 27, 5, 3, 3, and 2

boys. Naidoo concluded that these were not clear

clusters, but instead the abilities appeared to exist

along a continuum. Nevertheless, she proposed four

groups, but almost one-third of the children could not be

included in any of these. Although one large group could

be characterized by linguistic problems, and another

smaller group by visual perception and memory problems,

the considerable variability within all groups made a

simple classification of all poor readers impossible.

These negative results, however, must be re-evaluated

with our current knowledge of cluster analysis. Since the

time of this study, it has been pointed out that use of

the single linkage method gives rise to a property called

chaining, which refers to the tendency of the method to

join together relatively distinct clusters if a small

number of intermediate points are present between the

clusters (Everitt, 1974; Blashfield, 1976). It appears,

therefore, that Naidoo mistakenly used a method which

looks for optimally connected clusters or outliers

(children who are not very similar to other children in

the data set) when she actually wanted one which would

discover homogeneous, compact clusters.

Another multivariate statistical classification

procedure which has been used within the multiple-syndrome

paradigm of reading disabilities is the Q-technique of

factor analysis. In such a procedure, the factor is

defined by the children who have high loadings on that

particular factor. It is an "inverted" method which

groups together children who show similar patterns of

test scores.

The first researchers to use the 0-technique on

children with reading problems were Doehring and Hoshko

(1977). They used two groups from somewhat different

populations. The first group had purer reading problems

and consisted of 34 children (31 boys) ranging in age from

8 years, 8 months to 17 years, 4 months, in IQ from 71 to

125, and in word reading level from grade 0.2 to 9.2 (as

measured by the Slosson Oral Reading Test). The second

group had a mixture of problems, with learning disorders,

language disorders, and mental retardation; for no

children was the primary diagnosis reading disability.

This group consisted of 31 children (21 boys) ranging in

age from 8 years, 2 months to 12 years, 5 months, in IQ

from 79 to 105, and in word reading level from grade 0.6

to 5.9. The results from 31 tests of reading-related

skills were submitted to a Q-technique factor analysis.

For both groups of children, three subgroups resulted.

Overall, eight children could not be classified, and three

had high loadings on more than one factor.

Subgroup IR (the R indicating membership in the group

with reading problems) was characterized by good perform-

ance on all of the visual matching and on three of the

seven auditory-visual matching tests, and poor performance

on the oral reading tests involving words and syllables.

In general, Performance IQs were higher than Verbal IQs.

The subgroup 2R profile revealed good performance on

visual scanning tests involving numbers and letters, very

poor performance on auditory-visual letter matching, and

relatively poor performance on two other auditory-visual

tests and on the four oral reading tests involving words.

The children seemed unable to associate printed and

spoken letters rapidly. In contrast, the children of

subgroup 3R were characterized by very good visual and

auditory-visual matching of single letters, and differed

from subgroup 1R in terms of the poor visual and very poor

oral word, sentence, and syllable reading skills. Using

independent criterion, most of the children in subgroup 1R

were classified as distractible and all but one were in

regular classes; most of the children in subgroup 2R were

also in regular classes; and most of the children in

subgroup 3R were in special classes, had repeated a grade,

and had some kind of family reading problem.

Subgroups for the children with mixed problems showed

remarkable similarity to those with reading problems.

The profile of subgroup 1M was very similar to the

profile of subgroup 2R but without as marked a deficit in

auditory-visual matching. The profile of subgroup 2M

was similar in almost all respects to subgroup 3R, with

the only marked discrepancies occurring on the auditory-

visual matching tasks. Subgroup 3M, however, was

completely different; both visual and auditory-visual

letter matching skills were low, two of eight visual

scanning tests were very high, and oral reading of words

and syllables was relatively good. This profile might

reflect a certain type of visual perception deficit.

When the results of all children were analyzed together,

subgroup IR retained its separate identity, subgroup 2R

and subgroup 1M formed a second subgroup, subgroup 3R and

half of subgroup 2M formed a third subgroup, and subgroup

3M disappeared. Doehring and Hoshko valiantly attempted

to relate their statistically derived subgroups to those

of other investigators, but his proved difficult due to

the complexity of their data.

Petrauskas and Rourke (1979) also sought to classify

subgroups of retarded readers using a Q-technique factor

analysis. For this purpose, 133 poor readers and 27

normal readers between the ages of 84 and 107 months were

selected. All children were right-handed with WISC Full

Scale IQ scores between 80 and 120 and had been screened

for evidence of sensory deficits, primary emotional distur-

bance, and linguistic and cultural deprivation. The total

sample was randomly subdivided into two equal subsamples.

A total of 44 measures were originally collected on each

child, although this number was reduced by a "rational

grouping procedure" so as to better reflect six skill

areas. An analysis of the data from 20 neuropsychological

tests resulted in six factors (or subtypes) from each of

the two subsamples examined separately and combined. Only

four of the subtypes were reliable enough for further

consideration. Additionally, just 119 of the 160 subjects

exhibited adequate loading on a single factor.

The test performance of subtype 1 is characterized by

the following: relatively well developed visual-spatial

and eye-hand coordination skills; average or near-average

performances on measures of tactile-kinesthetic abilities,

abstract reasoning, and nonverbal concept formation; near

average performances on word definitions; mildly impaired

word blending, immediate memory for digits, and store of

general information; moderately to severely impaired

verbal fluency and sentence memory. Subtype 2 is

characterized by: average or near-average kinesthetic,

psychomotor, visual-spatial constructional and word-

definition abilities, and nonverbal problem-solving and

abstract reasoning skills within a context that provides

immediate positive and negative feedback; borderline to

moderately impaired immediate memory for digits and other

"sequencing" skills, store of general information, sound

blending, verbal fluency, and concept formation when

substantial verbal coding is required and/or when no

positive and negative feedback is provided; and moderately

to severely impaired finger recognition, immediate visual-

spatial memory, and memory for sentences. The test

performances of subtype 3 is summarized by the following:

average or near-average finger recognition (left hand),

kinesthetic, visuo-spatial constructional, vocabulary and

sound-blending abilities, and nonverbal concept formation

within a context of immediate positive and negative

feedback; borderline to mildly impaired finger recognition

(right hand), immediate memory for digits, eye-hand

coordination under speeded conditions, store of general

information, and nonverbal abstraction and the shifting of

set without the benefit of positive and negative feedback;

mildly to moderately impaired verbal fluency, sentence

memory, and immediate visual-spatial memory; and

moderately to severely impaired concept formation which

involves substantial verbal coding. The performance of

subtype 4 was largely average due to the fact that'it was

the normal readers who loaded on this factor. Considering

this subtype, it seems likely that there are different

types of normal readers since all normal readers did not

have the same constellation of abilities.

Interpreting the above results is, to say the least,

laborious. While it is advantageous to have unbiased

computerized statistical programs available to demarcate

and assign individuals to clusters, the experiment must be

so designed that our understanding has increased

afterwards. While hoping to accomplish this, Doehring and

Hoshko (1977) and Petrauskas and Rourke (1979) made this

unlikely by using, respectively, 31 and 20 test scores,

far too many to determine adequately meaningful patterns

consistent with particular theoretical positions.

The final study to be reviewed here was recently

completed by Darby (1978). Working with a relatively

unselected sample of 236 boys in the fifth grade, from

within a single county and reflecting all levels of

achievement, nine general academic subgroups were

identified by submitting the boys' WRAT reading, spelling,

and arithmetic scores to a cluster analytic procedure.

These initial subgroups were then compared on a variety of

intellectual, neuropsychological, socioeconomic status,

and neurological variables. As expected, there was a

general trend for the low-achievement subgroups to-

perform more poorly on all variables. The two lowest

subgroups had an extremely high proportion of boys rated

"affected" on neurological examination. However, the

mean IQ for all achievement subgroups exceeded 90. One

notable finding was the absence of a purely reading

disabled subgroup, even though there was one deficient

only in arithmetic.

By virtue of their markedly deficient achievement

scores, the lowest two subgroups were identified as a

learning disabled population (N=89). A second cluster

analysis using two language and two perceptual-motor

variables generated four unique subtypes of learning

disabilities. The subtypes consisted of: (1) children who

showed neither language nor perceptual-motor deficits; (2)

children with deficits in both areas; (3) children with

generally average scores except in verbal fluency where

their performance was significantly impaired; and (4)

children with substantial deficits only on perceptual-

motor variables. These differences were present despite

no significant differences between subtypes on

achievement, socioeconomic status, neurological ratings,

or personality variables. This investigation supported

the use of cluster analytic techniques in the identifi-

cation of meaningful subtypes of low-achieving children.

Summary and Rationale

An attempt has been made in this paper to address

some of the critical issues in the area of reading

disability. Central to the discussion have been criti-

cisms of definitions, subject populations, methodologies,

and statistical procedures. The difficulties in defini-

tions have largely stemmed from their being vague,

circular, and exclusionary, often identifying reading

disabled children by what characteristics were not

present. Consequently, there is little consistency from

study to study as to the children that should be

investigated. Highly variable subject populations have

been used, sometimes with a very narrow age range and

other times, too large an age range. For example, one

study included subjects from 8 to 31 years of age. In

only a few studies have developmental parameters been

included, an essential consideration for a population that

changes as rapidly as children. In the midst of this

quagmire, samples have been further restricted on the

basis of intelligence or type of presenting complaint

(academic, behavioral, psychological, medical, or some

combination of these). It is little wonder that the

concept of specific reading disability is increasingly

criticized as being of inconsequential utility (Taylor,

Satz, & Friel, 1979).

More recently the trend is away from treating reading

disability as a homogeneous entity and towards viewing it

as a set of diverse syndromes. During the search for

individual patterns of deficits, new problems have

emerged. Much too often, delineations of subgroups have

been totally theoretical and without supporting data,

treating the existence of subgroups as fact instead of

something to be tested. Several of the preceding studies

included too few subjects, especially in the search for

subgroups; one in particular had glaring reliability

problems when a subgroup with two members was identified.

Completely overlooked have been subgroups of normal

readers, describing how they relate to subgroups of

children with deficient reading skills (Doehring, 1978;

Knights, 1978). Without concomitant evaluations of normal

readers and contrasts of their subgroup strengths and

weaknesses, one can never be sure that particular deficits

are idiosyncratic to reading failure. Methodologically

the studies have been plagued by retrospective designs,

the use of unspecified or small numbers of clustering

variables, and reliance upon visual inspection of complex

data arrays in separating the subgroups.

The final criticism centers around the use of

inadequate or inappropriate statistics. In some cases,

statistical procedures caused the results of the analyses

to be more complicated and uninterpretable than the

original data. An additional unacceptable condition, but

one which appears in the literature, is the delineation of

subgroups which either overlap one another or do not

classify all the children being studied. While offering

hope, the concern with subgroups of reading disability has

provided limited clarifications.

The current research attempts to circumvent these

problems in a search for naturally occurring subgroups

within a data set. The Darby (1978) study was a good

first attempt in the rigorous application of cluster

analysis to the area of reading disability, but it was

limited by the use of only four neuropsychological

variables in the clustering process and to comparisons at

a single age level. This experiment, while also

developing a classification system based on neuro-

psychological variables, departs in several significant

ways from what has previously been done. First, develop-

mental considerations are built into the classification

system from the beginning by including variables from 3

probes spanning a 6-year period. Second, a broader range

of subskills will be tapped by using performance on 7

neuropsychological variables at each of the 3 probes. And

third, the classification system will be based on children

from all reading levels.

Subsequent to the derivation of the classification

system from neuropsychological variables, its external

validity will be primarily assessed through subgroup

comparisons of achievement scores. Such an approach will

make it possible to examine the development of neuro-

psychological skills and their relationship to reading.

From between- and within-subgroup comparisons, it will be

possible to gain a better understanding of the complete

reading process: what patterns across time result in

superior reading skills, as well as what types of inter-

ference come from differing patterns of deficient


Choosing the appropriate statistical procedure for

classifying the children is extremely critical. In a

project such as this, the choice is essentially between

cluster analysis and Q-technique factor analysis.

Methodological constraints of these procedures will be

discussed in the next chapter. For now it is sufficient

to say that cluster analysis is more powerful (Sokal &

Sneath, 1963; Sneath & Sokal, 1973; Everitt, 1974;

Blashfield & Aldenderfer, 1978) and will, therefore, be

used in this study.

To summarize the objectives of this project, these

questions will be answered:

(1) Will definable clusters (subgroups) emerge on the
basis of factor-pure neuropsychological variables?

(2) Are the clusters related to external achievement
criteria such as reading, spelling, and arithmetic; and,
if so, does this relationship vary by factor type (per-
ceptual vs. verbal) and/or by age?

(3) Are the clusters related to external nonachievement
measures such as intelligence, parental reading level,
neurological status, and socioeconomic status?




The subjects for the present investigation are a

subsample of those children who participated in the

longitudinal study of Satz and associates (1973, 1976,

1978). In 1970, the entire population of white males

beginning kindergarten in Alachua County, Florida, were

given an extensive battery of cognitive, perceptual, and

motor tests. A cross-validation sample of white males

entering kindergarten in 1971 also joined the project

(total N of 678). After the Kindergarten examinations

(Probe 1), all children were subsequently tested at

roughly 3-year intervals, at the end of the second grade

(Probe 2) and the end of the fifth grade (Probe 3) on

identical measures to those at the first probe, in

addition to appropriate standardized achievement

measures. The present investigation is conducted within

the context of this large on-going project.

Based on their achievement performance at Probe 2, 80

boys were identified as disabled readers (reading scores

more than one standard deviation below the mean). Primary


and alternate control subjects were also identified at

that time, the criterion being one-to-one matching on age

and at least average proficiency in reading. This

procedure provided identical subjects based on sex, age,

race, geographical location, and source of referral. These

subjects were checked for completeness of neuropsychologi-

cal data at Probe 1 and re-evaluated at Probe 3 if they

still resided in Alachua County. Such a follow- back,

follow-up design made the data from 211 boys available for

this study. The mean age of these children at Probe 1 was

65.6 months (SD = 4.37), at Probe 2 was 93.2 months (SD

4.08), and Probe 3 was 130.0 months (SD = 3.9).

The original rationale for including only white

children was to increase the cultural and educational

homogeneity of the sample. Consequently, error variance

attributable to differences along these dimensions was

automatically reduced. The additional restriction of

testing only males made it more likely that the problems

of poor readers would be present, since the male-to-female

ratio is approximately 4 to 1 (Rutter, 1978).


Test Measures

A factor analysis of the original neuropsychological

test variables (Fletcher & Satz, 1979) identified those

measures which load most highly on the various factors.

The 7 test measures not restricted by floor or ceiling

effects were used in the search for naturally occurring

subgroups. Included were the following tests: (1) the

Recognition-Discrimination Test (RD), total correct at

time limit (Small, 1968); (2) the Embedded Figures Test

(EF), total correct at time limit; (3) the Developmental

Test of Visual Motor Integration (VMI), age- equivalent in

months (Beery & Buktenica, 1967); (4) the WPPSI or WISC

Similarities subtest (SIM), scaled score (Wechsler, 1949;

1967); (5) the Peabody Picture Vocabulary Test (PPVT), raw

score (Dunn, 1965); (6) a Verbal Fluency Test (VF), total

score (Spreen & Benton, 1965); and (7) a Dichotic Listen-

ing Test (DL), total words recalled. Raw scores were used

for PPVT because this eliminates the age-correction which

is built into the test. More detailed description of each

test measure is available in an earlier publication (Satz

& Friel, 1973). Inclusion of more variables than were in

the Darby (1978) study stems from a deliberate attempt to

represent more completely the children's subskills which

may be related to academic achievement.

Classification systems based on statistical

procedures such as cluster analysis have the distinct

advantage of deriving the cluster arrays in a systematic

and impartial manner. Nevertheless, human judgment is

introduced in the choice of clustering variables and

method, as well as in determining when an optimal number

of clusters has been obtained. Even though the highly

structured nature of the Florida longitudinal data renders

these choices less critical, this paper gave special

consideration to the questions of external validity,

meaningfulness, and utility of the cluster solutions.

The primary validity of derived subgroups was

examined using separate measures, particularly the.Wide

Range Achievement Test (WRAT) which was administered at

Probe 3 (Jastak & Jastak, 1965). The WRAT has subtest

scores in reading, spelling, and arithmetic, and has

gained acceptance as an economical and reasonably accurate

estimate of a child's level of school achievement (Rourke

& Finlayson, 1978). These variables were in the form of a

discrepancy score, derived by comparing the child's

chronological age with the age-equivalent score obtained

on each subtest. Additional comparisons were made using

data on intelligence, parental reading level, neurological

ratings, and socioeconomic status (SES).

Statistical Analyses

Difficulties inherent to the Q-technique factor

analysis make it unsatisfactory for imposing order on a

large data set. These problems are discussed more

completely in Fleiss and Zubin (1969) and Everitt (1974),

but can be briefly reviewed: (1) The Q-technique's use of

the correlation coefficient as a measure of similarity

between individuals can be questioned; (2) If the data are

not linear, the procedure is an idle exercise; (3)

Classification is most difficult when an individual has

high factor loading on more than one factor; and (4) Given

a set of p variables on each individual, the number of

subgroups can never be more than p-l, one less than the

number of variables.

These difficulties are not present in cluster

analysis, which, in fact, has several advantages. First,

once a classification system has been developed, it can be

subjected to a relocation procedure, a test for optimal

solution. Next, cluster analysis prevents a child from

being placed into more than one subgroup. Third, cluster

analysis is able to classify every subject in the sample,

even if a cluster with a single member should exist. And

finally, it is possible to use a procedure such as factor

analysis to reduce the number of variables before using

the cluster method. Flexibility like this permits a

complex data base with a reduction to fewer variables for

easier interpretation. The previously described studies

by Doehring and Hoshko (1977) and Petrauskas and Rourke

(1979) illustrate the need for this option. It was,

however, unavailable because their analyses used the 0-

technique, which is itself a factor analysis.

Factor analysis of the original neuropsychological

battery (Morris & Satz, 1978) identified two factors at

Probes 1 and 2 and three factors at Probe 3. Factor I

(sensorimotor-perceptual) remained stable across probes

and included these tests: Recognition Discrimination

Test, Embedded Figures Test, and Developmental Test of

Visual Motor Integration. At Probes 1 and 2, Factor II

(general conceptual-linguistic) tasks included: The WPPSI

or WISC Similarities subtest, the Peabody Picture

Vocabulary Test, Verbal Fluency Test, and Dichotic

Listening Test. This general conceptual-linguistic factor

split at Probe 3 into a receptive conceptual-linguistic

factor (consisting of WISC Similarities subtest and the

Peabody Picture Vocabulary Test) and an expressive

conceptual-linguistic factor (consisting of the Verbal

Fluency Test and Dichotic Listening Test).

The search for discrete, naturally occurring

subgroups of neuropsychological skills were

accomplished through a cluster analytic procedure

employing a CLUSTAN 1C Program (Wishart, 1975). Rather

than submitting 21 variables for each subject to the

clustering procedure (RD, EF, VMI, SIM, PPVT, VF, and DL

at each of three probes) and creating severe

interpretation problems as was discussed earlier, the data

was transformed to seven factor scores:

Probe 1: (1) Sensorimotor-perceptual (P1) consisting of
RD, EF, and VMI;
(2) General conceptual-linguistic (VI) con-
sisting of SIM, PPVT, VF, and DL;
Probe 2: (3) Sensorimotor-perceptual (P2) consisting of
RD, EF, and VMI;
(4) General conceptual-linguistic (V2) con-
sisting of SIM, PPVT, VF, and DL;
Probe 3: (5) Sensorimotor-perceptual (P3) consisting of
RD, EF, and VMI;
(6) Receptive conceptual-linguistic (V3-Rec) con-
sisting of SIM and PPVT;
(7) Expressive conceptual-linguistic (V3-Expr)
consisting of VF and DL.

With the neuropsychological subskills in this form, a

hierarchical agglomerative average-linkage method employ-

ing a squared euclidean distance similarity coefficient

was used to search for subgroups. This particular

combination of similarity measure and fusion technique has

proven most effective in deriving classification systems

from other subsets of the Florida longitudinal data

(Darby, 1978; Satz, Morris & Darby, 1979). A constraint of

cluster analysis is its lack of a generally accepted test

statistic for determining when the number of clusters

accurately reflects the underlying structure of the data.

In its absence, the process of combining data points was

traced through the means of the clustering variables and

the magnitude of the similarity coefficient at each

stage. A solution was sought which did not fuse markedly

dissimilar subgroups (or individuals), retained subgroups

with unique and interesting profiles, and avoided early

regression towards the overall sample mean. It was

essential to maintain the configural nature of the data

and not have purely scalar clusters in which the subskills

fell into a ranking from proficiency to deficiency.

The validity of these subgroups was then evaluated by

multivariate and univariate analyses of the clustering and

validity variables. These analyses were conducted using

the General Linear Models Procedure of the Statistical

Analysis Systems (SAS) program (Barr, Goodnight, Sail, &

Helwig, 1976). Individual means were compared using post

hoc Duncan's Multi-Range Tests (Winer, 1971). Because of

the nature of the data, chi-square tests for independence

were utilized for the teacher's rating of socioeconomic

status and the neurological ratings.



The organization of the Results section will follow

the objectives of this project as stated at the end.of

Chapter I.

Cluster Analytic Solution

(1) Will definable clusters (subgroups) emerge on

the basis of factor-pure neuropsychological variables?

Using guidelines identified in the previous chapter,

it was determined that a 12-cluster solution yielded the

most distinctive pattern of subgroups. Continuation of

the clustering process beyond this point would have fused

subgroups which are statistically different on two of the

seven clustering variables. Subjecting the initial array

to an evaluation of optimal cluster membership improved

the subgroup clarity by placing various children in more

similar clusters, thus increasing within-subgroup

homogeneity without a complete reorganization. Further

confirmation of the 12-cluster solution came from

submitting the data to additional hierarchical procedures.


Cluster membership using different procedures was

similar, the primary difference being in the degree to

which subgroups were separated from each other. Table 1

presents the final 12-cluster solution with the means for

each clustering variable. As additional validation,

examination of the subgroup means suggests that the

solution makes sense heuristically given what is known

about the data and the external criteria. This will be

elaborated upon in subsequent sections. More extensive

descriptive statistics can be found in the Appendix.

The upper half of Figure 1 presents a visual

representation of each subgroup. The achievement data in

the lower half of the figure will be used for later

contrasts. Before describing the clusters, attention

should be given to characteristics of the figure since an

understanding of it is critical to the current discussion.

The horizontal line represents the overall sample mean

for all tests, which is 0.00 since factor scores were

used as clustering variables. Units along the ordinate

correspond to positive and negative deviations from that

mean. Perceptual factor scores appear first as three

solid lines within each cluster, followed by the four

verbal factors as dotted lines. And finally, within each

factor sequence (perceptual or verbal), the data are

presented sequentially, i.e. Probe 1, Probe 2, and then

Probe 3.


ri Q

x 00

A o
> o














-4 1



ao r-i CNI o ao am o
Z N N C" C4 r-- 1-4 ,-4

0 C0

0 0


o o
0I r-

' C 0 L
i-4 i- C14

c m Ln o r- oo cm o

r- CN1

r- "












^ (
re -



cnO -4


co 0 R\i N (D a) 0 O0 (o0 '14 Ri
'O 1 co Ir c\i (J In co -
---O O O O O




,- oD Ca 0 a) wD t o4
I() ") \i 0 (I n (I) -
0 0 0 0 0 ( 0 -

j- U

I ri




___________ _____________ o

0 )-.-u

-- n


-U o



sU------------l in

00-- 0-0-0-0-0-

s~DI- e
j ____C



00 00 0 0 0- ^t

'aqiii 'uwnaqt '

(IJ 0 i It- (D1 Q) (") c r i
Lf) \) u>) i C\j 0 R\ Ir c) -
0 0 0 0 0 0 0 -u

soIqDIJDA buijalsnlo

From an examination of Figure 1 it can be seen that

there are two general approaches for depicting subgroup

patterns. The first approach involves comparing the

clustering variables, as a whole, to the overall mean.

For example, there are some clusters with above-average

performance (6, 7, 10, and 12), others with consistently

deficient performance (3, 8, and 9), and a third group

with scores both above and below the mean (1, 2, 4,'5, and

11). The second and more precise approach is to describe

the relative elevation for factor-type and year-probe.

The latter approach will be used for individual

descriptions of each cluster.

Subgroup 1 is distinguished by generally above-

average performance on all variables. Both factor types

(perceptual and verbal) are fairly equivalent across

probes. The only deviation from this very consistent

performance is a split between receptive and expressive

language skills at Probe 3. If a single language measure

had been used, the clear superiority of receptive language

skills would have been masked. Subgroup 2, in contrast,

is characterized by generally below average performance

which becomes lower with time. Perceptual skills began

just below average while verbal skills started at an above-

average level. Both factor types, however, were

deficient by Probe 3. The pattern for subgroup 3 has both

perceptual and verbal factors below average across probes.

The fourth cluster is characterized by a split as a

function of factor type. Verbal factors are consistently

well below average while the perceptual factor began just

below average and improved to its more adequate level

within the first three years of the study.

Subgroup 5, like subgroup 4 but at a slightly higher

level, improved on perceptual skills early with no further

relative gains. Its verbal skills, on the other hand,

began at a very deficient level and became better year by

year. Subgroup 6 demonstrates above average skills,

especially in the perceptual area; no age effect is

apparent. Similarly, subgroup 7 is very superior on all

measures. Because of improvements through the years,

verbal skills are much higher than perceptual ones by

Probe 3. Perceptual skills varied with time, but showed

no specific trend. Subgroup 8's pattern is very similar to

that of subgroup 2, but at a lower level. Subgroup 8

performed very poorly in both perceptual and verbal areas,

and then regressed during each re-examination. Subgroup 9

is like the preceding cluster in terms of overall

deficient performance for both factors; there are,

however, no age differences.

Like subgroups 6 and 7, subgroup In is distinguished

by above-average performance at all probes for both

factors. While verbal skills are at a high and variable

level, perceptual skills deteriorated over time to a level

just above average. In contrast to subgroup 4, subgroup

11 demonstrates a specific perceptual deficit: perceptual

skills are consistently very low while verbal skills are

better and just below average. Subgroup 12 performed at

an above-average level on all variables. There is a

suggestion that verbal skills are slightly higher than

perceptual, although this is obscured by the random

variation across probes.

The number of subjects in each subgroup is presented

in Table 1 and in the parentheses adjacent to the cluster

numbers in Figure 1. The overall sample size is

apparently large enough to allow an adequate number of

children in each cluster, the range extending from 9 to

29. The smallest subgroups are those with either the best

or worst overall performance, and the largest subgroup is

the one with the most average performance; this pattern

is just as is expected in a normal distribution. It is

very important to note that the cluster analytic procedure

was able to classify every child in the sample. This is

in marked contrast to other classification techniques

which generally account for only 40 to 70 percent of

their subjects.

It was next necessary to determine the extent to

which subgroups differed from one another. A multivariate

analysis of variance (MANOVA) on the seven clustering

variables revealed an overall significant effect for

subgroup (Hotelling-Lawley Trace = 15.89, F approximation

(77,1339) = 39.48, p<0.0001. This finding justified the

application of univariate analyses for each variable,

which yielded significant effects for subgroups on factors

PI, F(11,199) = 31.38, p<0.0001, on P2, F(11,199) =

49.51, p<0.0001, on P3, F(11,199) = 42.65, p<0.0001, on

Vl, F(11,199) = 49.53, p<0.0001, on V2, F(11,199) =

31.71, p<0.0001, on V3-Rec, F(11,199) = 52.91, p<0.0001,

and on V3-Expr, F(11,199) = 41.20, p<0.0001. Results of

significance tests (Duncan's procedure, p<0.05) for

differences between subgroup means on each clustering

variable are presented in Table 1.

Central to considerations of internal validation is

the extent of inter-subgroup heterogeneity on each

clustering variable. Figure 2 presents a visual

comparison of the subgroup means. An examination of

this figure reveals robust differences between

subgroups despite similarities between adjacent

means. Furthermore, comparisons between factors

indicate that relative subgroup rankings are very

different, thereby emphasizing the importance of each

clustering variable in descriptions of the subgroup


,1 r- 2

CL __ i :L

r- 00

T m

41 ,--1

'- U
4J btO
0 0

o o

X m
o u)
0 O


) 0

En 4-4


QJ 0
0 C

C ud

C) C

S 4



& I"
ro ^i
>, -] _

En 54



4 4

cQ cn

t .



00 -1

inr) IIna

> p F

patterns. From the preceding discussion of subgroup

patterns and supporting statistics, it is apparent that

the clusters are definable and internally valid.

Primary External Validation

(2) Are the clusters related to external achievement

criteria such as reading, spelling, and arithmetic, and,

if so, does this relationship vary by factor type

(perceptual vs. verbal) and/or by age?

Achievement scores are available for all children on

the Iota Word recognition Test at Probe 2 and Reading,

Spelling, and Arithmetic subtests of the Wide Range

Achivement Test at Probe 3. In addition, Gates-McGinitie

Vocabulary and Comprehension scores were collected for 154

of the children at Probe 2. All achievement scores are

highly correlated, but this is especially true of

language measures (Table 2). In light of this and so as

not to be repetitive, subgroup comparisons will only be

presented for the WRAT scores. This measure was

selected because of its sampling from more diverse

academic areas and because its scores should be more

stable, having been obtained at an older age.

Mean scores for the WRAT subtests are reported by

subgroup in Table 3. An examination of the subtest means



^ bO







-H -
u m

V) cd

C >

1 0
- .,-


c0 U

Table 3

Mean WRAT Scores (Probe 3) by Neuropsychological Subgroups

Cluster Mean SD

1.83 DE

-13.57 F

- 7.18 EF

- 6.95 EF

8.68 CD

30.37 AB

41.70 A

-20.56 FG

-30.50 G

19.89 BC

-15.19 FG

10.48 CD















Mean SD

- 9.76 CDE

-23.14 FG

-19.64 EFG

-16.95 DEF

- 5.42 BCD

9.89 A

23.50 A

-27.11 FG

-32.83 G

7.44 AB

-21.44 EFG

- 4.56 BC















Mean SD

-10.55 CD 9.94

-17.71 DE 7.73

-16.95 DE 5.60

-10.95 CD 9.28

- 5.63 BC 12.82

0.05 AB 8.78

4.70 A 17.02

-17.78 DE 6.82

-21.17 E 7.40

- 2.78 ABC 15.59

-19.63 E 8.79

- 8.76 C 8.41

-10.91 11.88

* Means followed by the same letter are not significantly
different within variables (Duncan's Multi-Range Test,
Alpha level = 0.05).














for the total sample reveals overall age-expected

performance in reading. However, in spelling and

arithmetic the sample means are approximately 11 months

below that expected from the children's chronological age.

In determining whether the clusters are related to

external achievement criteria, a MANOVA on the WRAT

subtests yielded an overall significant effect for

subgroup (Hotelling-Lawley Trace = 1.05, F approximation

(33,587) = 6.23, p<0.0001) justifying the application of

univariate statistics. Individual analyses reveal

significant effects for subgroups on reading, F(ll,199) =

15.49, p<0.0001, on spelling, F(11,199) = 11.41, p<0.0001,

and on arithmetic, F(11,199) = 10.14, p< 0.0001. Means for

each variable, with additional results from Duncan's

procedures (p<0.05), appear in Table 3 and Figure 3.

From an examination of Figure 3, it is apparent that

the classification system developed from neuropsycho-

logical factor scores also separates subgroups on academic

achievement variables. Even though adjacent subgroups are

not statistically significant, there is a clear

delineation for clusters which are several steps removed.

A comparison of Figures 2 and 3 suggests that the relative

ranking of subgroups on achievement measures is most

similar to that present with perceptual clustering

variables (PI, P2, and P3). To illustrate this point,


Wide Range Achievement Test



Subgroups within the same box do not differ significantly

Figure 3. Configuration of Neuropsychological Subgroup Means
for WRAT Scores at Probe 3

subgroups 6 and 7 are ranked highest, 1, 4, and 12 are

generally intermediate, and 8, 9, and 11 are ranked lowest

for both perceptual and achievement variables; an

altogether different pattern is present with the verbal


Figure 1 will be used to examine more closely the

relationship between achievement and neuropsychological

variables. For uniform comparisons, the achievement data

are presented in the form of standard scores which vary

above and below the various subtest means of 0.00. The

overall pattern is that high neuropsychological skills

(both perceptual and verbal) are related to high

achievement (e.g., subgroup 7), average neuropsychological

skills to average achievement, (e.g., subgroup 1), and low

neuropsychological skills to low achievement (e.g.,

subgroup 9). However, a more complex relationship is

apparent with comparisons of various combinations of

clusters. Take, for example, subgroup 4 characterized by

low verbal and higher perceptual skills and subgroup 11

with its low perceptual and higher verbal skills. Both of

these clusters have low (and statistically equivalent)

reading and spelling scores. Only in arithmetic did

subgroup 4 attain average scores. Subgroups 6 and 10

present a comparable pattern with higher relative

performance in one of the neuropsychological areas.

Subgroup 6 has higher perceptual than verbal skills while

subgroup 10 has higher verbal than perceptual skills.

Nevertheless, their corresponding academic proficiencies are

statistically equal. These comparisons suggest that

equivalent academic achievement can result from different

combinations of neuropsychological subskills. As

additional confirmation of this point, these comparisons

should be made: subgroups 1 and 5, subgroups 5 and 10, and

subgroups 5 and 12. Each comparison is characterized by

statistically equivalent achievement but with different

combinations of neuropsychological subskills.

Influence of Factor Types

The purpose of this section is to examine the

differential contribution of factor types on achievement.

Comparisons between individual subgroups will be used in

an attempt to partition out the effect of language and

nonlanguage measures on achievement competency. Stated

another way, what is the relationship between academic

achievement and neuropsychological factor type?

The impact of perceptual subskills is evident in

those combinations of subgroups with different perceptual,

equivalent verbal, and unequal achievement scores.

Individual comparisons of subgroups 6 and 12 and subgroups

5 and 11 best illustrate this point. Significant

differences between these pairings for all achievement

scores are present when similar differences occur only in

perceptual areas. The effect is comparable for subgroups

1 and 11 and subgroups 3 and 4 except that the differences

are most apparent on arithmetic tasks. While not

causative, the relationship between perceptual skills and

academic achievement (especially arithmetic) is apparent.

A similar comparison of verbal subskills and achieve-

ment was made using these comparisons: subgroups 1

and 10, subgroups 5 and 7, subgroups 4 and 10, and

subgroups 4 and 12. Such comparisons are characterized by

equivalent perceptual skills and unequal verbal and

achievement scores, thereby highlighting the relationship

between these latter two variables. Although the

achievement scores are not significantly different, the

same trend is apparent in subgroups 1 and 4, subgroups 1

and 12, and subgroups 6 and 7.

By comparing subgroup 11 with both 8 and 9, it

appears that very low perceptual skills counteract any

expected improvement related to better verbal subskills.

Subgroup 11 has perceptual skills equally low to those of

subgroups 8 and 9. However, subgroup 11's average verbal

skills were not sufficient for more adequate achievement


In conclusion, it appears that high perceptual skills

can lead to high achievement and that low perceptual

skills can lead to low achievement. An equivalent direct

relationship exists between verbal skills and achievement.

Furthermore, one factor type may moderate extreme levels

of the other factor type, but the extent to which they

interact depends on characteristics of each particular


Influence of Age

Subgroup 2 is unique in that its perceptual and

verbal scores decrease consistently over time and by

Probe 3 are no better than those of subgroup 3. Concur-

rently, these clusters have equivalent achievement scores.

It appears, therefore, that subgroup 2's early profi-

ciency in both perceptual and verbal areas was inadequate

to prevent the eventual poorer achievement. A similar

trend of decreasing subskill performance is apparent for

subgroup 8 while subgroup 5 reveals an increasing pattern of

improvement in neuropsychological subskills over time. In each

of these subgroups there are clearly developmental changes.

But rather than these patterns of developmental changes

being related to achievement, the relationship is stronger

between achievement and the eventual level of

neuropsychological subskills. In concrete terms, this

means that for subgroup 2 the impact on achievement of

decreasing perceptual and verbal skills is less important

than the eventual low level of functioning on the factors.

Secondary External Validation

(3) Are the clusters related to external nonachieve-

ment criteria such as IQ, parental reading level, neuro-

logical status, and socioeconomic status?


An analysis of variance of Peabody Picture Vocabulary

Test intelligence scores yielded a significant effect for

subgroup, F(11,199) = 25.11, p<0.0001. The results of

individual comparisons of subgroup means (Duncan's

procedure, p<0.05) are presented in Table 4 and Figure 4.

The total sample mean (104) closely approximates the stan-

dardization sample mean. As might be expected, subgroup 7

has the highest level of neuropsychological skills in addi-

tion to the highest IQ (139), while subgroup 9 is lowest

on both variables (IQ = 82). For some clusters, however,

there is an interaction between IQ and neuropsychological

performance. Subgroups 3 and 8 have average IQs of 96 and

91, respectively, in combination with very low

neuropsychological skills. In contrast, subgroup 6's

performance on neuropsychological and achievement measures

was extremely good compared to its IQ of 114.

A pattern similar to the achievement data is evident

in the IQ scores: in a ranking of the subgroups, adjacent

Table 4

Mean PPVT IQ Scores, Maternal and Paternal Reading
Levels by Neuropsychological Subgroups

Mean SD

1 113.66 B

96.57 DEF

95.50 EF

97.80 DEF

103.16 CD

113.79 B

138.90 A

90.89 FG

81.75 G

121.22 B

100.63 CDE

106.20 C


Reading Level*
Mean SD

10.61 12.38 A













12.41 A

11.81 AB

13.25 A

10.81 ABC

13.13 A

13.13 A

8.71 C

8.92 C

12.56 AB

9.76 BC

12.36 A













Reading Level
Mean SD

12.43 0.65

12.30. 0.89

13.32 0.85

10.87 0.94

11.28 1.01

14.66 0.90

13.05 1.11

12.14 1.26

10.43 1.29

15.48 1.33

11.68 0.96

12.22 0.74

* Means followed by the same
different within variables.

letter are not significantly















10 6 1

Reading Level*

4 6 7

2 1 12

10 3



9 8

Subgroups within the same box do not differ significantly

Figure 4. Configuration of Neuropsychological Subgroup Means
for IQ Scores (Probe 3) and Maternal Reading Level

means tend not to be significantly different. Comparisons

of Figure 4 with Figure 2 suggest similar relative

rankings on the IQ scores and clustering variables. The

relationship is strongest with V3-Rec, a finding

consistent with the opinion that the PPVT measures

receptive language abilities. Relative Q1 rankings are

also markedly similar to those of achievement measures

(Figure 3) except for subgroups 1 and 11, whose rankings

on all achievement measures are below that of their IQ


Parental Reading Levels

Reading grade equivalents were obtained on 141

mothers and 102 fathers of the children used in this

study. Because of differences on these measures, the

levels of parental socioeconomic status and education

level were taken into consideration through covariate

analyses of the data.

The effect of mothers' education level was

significant, F(1,127) = 72.63, p<0.001, as was the effect

of SES, F(1,127) = 11.52, p<0.0001. An analysis of

covariance of the mothers' reading grade equivalents

yielded a significant subgroup effect, F(11,127) = 2.72,

p<0.003. Subgroup means for maternal reading levels

(corrected for covariates) appear in Table 4. The results

of pairwise comparisons (Least Squares Means, p,0.05)

between subgroups appear in Figure 4. Although not nearly

as clear as other contrasts, there are significant differ-

ences between clusters. After adjustments for the covariates,

the mothers of children in subgroup 4 exhibited the highest

reading skills (grade equivalent = 13.25). Neuropsychologi-

cally, the children of subgroup 4 are characterized by a

specific verbal deficit. The mothers of subgroups 6 and 7

are also good readers (grade equivalents of 13.13) while the

mothers of subgroups 8, 9, and 11 are the worst readers

(8.71, 8.92, and 9.76, respectively). It should be noted that

the children in subgroups 6 and 7 demonstrate consistently

high neuropsychological skills while these measures are

very low for subgroups 8 and 9. Subgroup 11 has a

a specific perceptual deficit.

While there was a significant covariate effect for

the fathers' education level F(1,88) = 84.11, p<0.001, the

effect for SES was nonsignificant, F(1,88) = 0.29, p<0.59.

The analysis of covariance of the fathers' reading grade

equivalents failed to yield a significant subgroup effect,

F(11,88) = 1.76, p<0.071. The subgroup means for paternal

reading level also appear in Table 4.

Neurological Examination

During the fourth year of the longitudinal study, 143

of the children in this study received a neurological exam

consisting of the following: 1) a general examination

assessing cranial nerves, motor responses, sensation,

reflexes, and cerebellar functioning; 2) special exam to

evaluate fine and gross motor functioning, right-left

discrimination and eye tracking; and 3) an evaluation of

gross body anomalies or stigmata of head, eyes, ears,

mouth and feet. Neurologists from the University of

Florida conducted each examination without concurrent data

on the child and made an assignment to one of three

categories (affected, borderline-equivocal, or normal) on

the basis of overall clinical judgment and component

numerical scores.

The data on neurological ratings for each subgroup

are presented in Table 5. Overall, a high proportion (46

percent) of the children were rated as being "affected"

neurologically. Furthermore, the absence of data on 68

children results in some subgroups being under-

represented, such as subgroup 10 with 5 members. Because

of these observations, the neurological data must be

viewed with some caution. However, a significant

relationship between the distribution of neurological

status by subgroup was confirmed using the chi-square test
for independence, X = 52.75, p<0.0002. From an inspec-

tion of cell frequencies in Table 5, it is evident that

subgroups with the most deficient performance on

neuropsychological clustering variables (subgroups 8 and 9)

also have the highest frequency of positive neurological

CN 4 C) O O C 0o oC 0o r- o %-
- V}- n L n V' Nc C c % CI -It

4 0

1 O

0 3

S e

S 0

0 4


'-0 0
1-4 C14

un N C v- NCM

CmY Cc Co -4
-4 co c r- r--

co 0 0o 0

0C 0 C

m c 1 >


o ,-i 0C 0C

oC 03 C 03





Lr 0

H, o

E- Z
F >

03 '-

ca' m C



S 0 1 'C N oo N ,0 C3 c \o N r- 03 -o
- -zt

*r-4i3 0C C, .o 03 o' L N -4 C -.- -j CD'o co,0


,.u ,--.- ,--i
CM N- CCI co Lr) kM r-. 03 a' 0, r-- CN P,
b, -40CM

cY) CY

N1 L
J- N

findings (100 and 83 precent). Conversely the highest per-

forming subgroups (numbers 6, 7, 10, and 12) have fewer

children so rated (3, 0, 0, and 3 percent, respectively).

Socioeconomic Status

The measures of SES were obtained at Probe 3 on a

scale of (1) low or (2) average or above. The teachers

rated each child based on enrollment in the free lunch

program. From this gross differentiation of cultural

background only 16 percent of the children were designated

as having low SES. Based on the frequency distribution in

Table 6, the relationship between subgroups and this
variable was determined to be significant, X = 61.69,

p<0.0001. More particularly, those subgroups with the

lowest neuropsychological performance consistently had a

greater representation of low SES ratings (subgroups 8 and

9). Such ratings were infrequent, if at all present, for

clusters with average and above scores on all clustering

variables (e.g., subgroups 6 and 7).

Socioeconomic Status By

Table 6

Neuropsychological Subgroups











































% of
Freq Subgroup

0 0

4 20

8 36

2 10

1 5

1 5

0 0

5 56

9 75

1 11

1 6

2 8

34 16

Average or Above
% of
Freq Subgroup

29 100

16 80

14 64

18 90

18 95

18 95

10 100

4 44

3 25

8 89

15 94

23 92

176 84



The results of this project demonstrated the utility

of a cluster analytic approach to longitudinal neurp-

psychological data. Previous applications of cluster

methods to longitudinal data have been infrequent (Rice

& Mattsson, 1966), and cluster analysis has never been

used in a longitudinal study of learning disabilities. The

study was strengthened, in part, by its use of an

unselected group of failing achievers who were then

matched with twice as many same-aged children selected

only on the basis of their being at least average

achievers. The children were not included because of
clinic referral and were homnogeneous with regard to sex,

race, educational experience, and geographic location.

And finally, the study was unique in its inclusion of

repeated measures on the same children over a six-year

period, multiple measurements of neuropsychological

functioning, extensive neurological data, and family

history variables.

Internal Validation

The first and major question addressed by this pro-

ject was: Would definable subgroups emerge from longi-

tudinal neuropsychological data? The issues of whether

the subgroups are definable and, similarly, meaningful are

extremely important because cluster analysis cannot take

these factors into consideration. In fact, without proper

safeguards, cluster analysis will identify subgroups from

data in which there is no underlying structure. Cluster

analysis begins with every subject as an independent

cluster and then proceeds to combine those which are most

similar. If left unchecked, cluster analysis will

eventually organize all subjects into a single large

cluster. Of critical importance, then, is determining

when to stop the clustering process. When such

characteristics are acknowledged and appropriate safe-

guards taken, cluster analysis becomes a precise and

powerful statistical procedure.

Data analyses were used to confirm the internal

validity of the subgroups. Individual subgroups were

adequately compact (homogeneous) and statistically

different from other subgroups. Descriptions of some

subgroups proved easy because they were either high

(subgroup 7), intermediate (1 and 12), or low (3 and 9) on

all clustering variables. For the majority of the

subgroups, however, the pattern of neuropsychological

development was more complex. It became necessary to

consider the relative elevation of factor types in

addition to the overall level of the subskills. For

example, some subgroups demonstrated exclusively high

perceptual (subgroup 6) or verbal (10) performance

competency, while for others one of the performance

factors was low. Subgroup 4 was low only in verbal skills

and subgroup 11 was low in perceptual skills. For still

other subgroups it was necessary to consider changes with

time; two subgroups (2 and 8) consistently regressed on

both factor-types during the 6 years of the study while

subgroup 5 showed the pattern of improving subskills. The

results, then, confirmed that the subgroups were definable

and meaningful. Furthermore, descriptions of a child's

neuropsychological functioning must optimally contain

statements of the overall level of the skills, relative

elevations of factor-types, and whether any developmental

changes exist.

Primary External Validation

The study's second question involved whether the

neuropsychological clusters were related to achievement

variables. This was of particular concern because the

issue of external validity is virtually never addressed in

subgroup research regardless of the method of subtype

division. As was stated in the first chapter of this

paper, too many researchers have been satisfied to simply

describe the categories of poor readers which they believe

to exist based on clinical experience. When no data are

presented to justify the derivation of the categories, it

is not surprising that verification of subgroups'

meaningfulness is also lacking.

The results showed a robust relationship between

neuropsychological competency and reading achievement,

just as was expected (Darby, 1978; Rourke & Finlayson,

1978; Satz et al., 1978). For some clusters it was a

simple direct relationship of high neuropsychological

performance to high achievement (subgroup 7), inter-

mediate to intermediate (subgroup 1), or low to low

(subgroups 3 and 9). For other subgroups, however, the

relationship between achievement and factor type had to be

considered. The unique design of this investigation with

its separation of variables into perceptual and verbal

factor scores made it possible to study the relationship

in more detail.

Comparisons between individual subgroups, as was

elaborated in the previous chapter, helped clarify the

influences of individual neuropsychological factors upon

achievement. Independent of the method of examination,

the importance of language skills in academic achievement,

and especially reading, is certain. There are clearly

subgroups (4, 10, and 12) for which verbal skills are most


closely related to reading level. While this relationship

is well recognized for the learning disabled (Cole &

Kraft, 1964; Denckla, 1972; Benton, 1975; Yule & Rutter,

1976; Darby, 1978), similar conclusions can now be

extended to average and superior readers. This latter

point should not be minimized.

For other subgroups, however, the relationship

between perceptual performance skills and reading seemed

greater. Subgroup 11 is a case in point because it is

characterized by average verbal skills and a specific

perceptual deficit. Academically, this subgroup is

deficient and at a level consistent with its perceptual

skills. Apparently, then, verbal strengths in this

cluster group were insufficient to compensate for their

academic achievement failure. The comparison between

subgroups 6 and 10 also illustrated this point. Subgroup

6 is characterized by higher perceptual than verbal

skills; the reverse is true for subgroup 10. Furthermore,

their respective IQs are 114 and 121. If language skills

are the crucial factor, subgroup 10 should have a clear

advantage. Yet it is subgroup 6 which is slightly, though

nonsignificantly, better academically. The present

results provide additional evidence against the exclusive

role of language measures in the development of reading

competence (cf., Vellutino et al., 1975; Vellutino, 1977).

For example, it was shown that delays in perceptual

scores were associated with reading failure and

that elevations in this factor type were associated with

reading success. Intact verbal skills, in the presence of

deficient perceptual performance did not protect children

in these subgroups from reading failure. These findings

are compatible with other clinical subtype studies which

have employed less precise multivariate methods (Cole &

Kraft, 1964; Denckla, 1972; Boder, 1973; Mattis et.al.,

1975). However, in most of these approaches, the

perceptual deficit subtype group has either been small

(Denckla, 1972; Mattis et al., 1975) or associated with

other cognitive linguistic deficits (mixed subtype) (Cole

& Kraft, 1964; Boder, 1973). However, in a recent study

employing multiple cluster analytic techniques at one age

level (Satz, Morris, & Darby, 1979) results revealed a

distinct and homogeneous perceptual deficit subtype that

was almost as large as two of the language deficient

subtype groups. Each of these finding, in summary,

illustrates the complexity of neuropsychological

components in reading. It is becoming increasingly clear

that combinations of both perceptual and verbal factors

influence a child's reading level. The current research

indicates that there are at least 12 combinations of these

neuropsychological skills and that each pattern is

independently related to achievement.

For some subgroups it was necessary to consider age

effects and their relationship to achievement. Subgroups

2 and 8 are especially unique in that they demonstrate a

consistent regression in neuropsychological functioning

across time. Subgroup 2 began with average perceptual and

slightly above-average verbal skills, while subgroup 8

began at deficient levels and then got worse. The

existence of such non-recovering subgroups is contrary to

lag theorists (Satz & Sparrow, 1970) who predict that

learning disabled children eventually catch up on earlier

developing skills such as visual-perceptual and cross-

modal sensory integration. These children are further

characterized by IQs within the average range and a higher

incidence of "affected" ratings on neurological

examinations: half of the children in subgroup 2 and fully

100 percent of subgroup 8. Even in the presence of

adequate intellectual functioning and early strength on

neuropsychological skills, these children proved destined

for deficient academic achievement. If subgroup 2 had had

a single evaluation at one of the earlier probes, it would

have been inadequately represented and possibly classified

as unexpected reading failure (Symmes & Rapoport, 1972).

An intriguing and very important research project would be

to identify children of subgroups 2 and 8 at an early age

so that attempts could be made to prevent their academic

demise. The detection of subgroup 8 might prove easier

because of the high frequency of neurological problems.

Subgroup 2, however, began as a very average group of

children. For that reason, little attention was probably

called to them, making their delayed development less

apparent until later years when intervention is much more


The pattern of performance of subgroup 5 was unique

and interesting because both factor types improved over

time. Perceptual skills seemed to have largely reached

their eventual level by Probe 2, whereas expressive verbal

skills were still improving at Probe 3. It is a pattern

such as this that lag theorists have expected to find with

developmental dyslexia (Satz & Sparrow, 1970). While

evaluations of such theories have proven unsuccessful

(Fennell, 1978), this is perhaps because the research

concentrated on larger, more heterogeneous groups instead

of subgroups derived on the basis of multivariate

descriptive procedures.

Darby (1978), using cluster analytic techniques,

observed a relationship between depressed arithmetic

scores and perceptual-motor performance. A similar

relationship was in part demonstrated in the current

results. The arithmetic performances for subgroups 4 and

5 were somewhat higher than their reading and spelling

scores, just as their spatial-perceptual scores were

higher than their conceptual-linguistic scores. The

relationship for subgroups 7 and 11 was in the opposite

direction: poorer arithmetic relative to other achievement

areas when perceptual skills were also relatively

depressed. These results are comparable to those of

Rourke and associates (Rourke & Finlayson, 1978; Rourke &

Strang, 1978), even though they created their groups based

on achievement patterns and then examined external

neuropsychological patterns, the reverse of this study.

Since Rourke only examined children with learning

disabilities, there are no comparison groups for subgroups

5 and 7. However, subgroup 4 is similar academically and

neuropsychologically to Rourke's Group 2, and subgroup 11

is like his Group 3. Without more direct evidence as to

the functional integrity of the cerebral hemispheres, this

study cannot make statements as to right hemisphere

superiority for types of calculation abilities (Hecaen,

1962; Dimond & Beaumont, 1972). Nevertheless, the

relative strengths of subgroups 4 and 5 and relative

weaknesses of subgroups 7 and 11 are consistent with the

position (Rutter, 1978) that specific arithmetic

calculation skills are due to particular patterns of

visual-spatial organization and integration, skills

ordinarily thought to be subserved by the right cerebral

hemisphere (Reitan, 1966).

Comparisons between the current classification system

and those reviewed in the first chapter have proven

difficult. A primary reason for the paucity of compari-

sons stems from this study's inclusion of children from

all performance levels; all other classification systems

are based exclusively on disabled populations. Therefore,

the six subgroups with average or above performance cannot

be contrasted to any subgroups in the literature.

The second major obstacle to comparisons stems from

differences in data bases. Since some classification

systems were derived from clinical experience and no

validating data, comparisons were restricted to their

general statements describing each group. Differences

also occurred on the type of data upon which classi-

fication systems were based. As an example, Rourke and

Finlayson (1978) differentiated their groups on the basis

of achievement and then looked at differences in the

neuropsychological measures. In contrast, the current

classification system was based on neuropsychological

measures and validated with achievement scores. Since

opposite sequences were used, comparisons between studies

proved more difficult. And finally, even when other

classification systems were based on neuropsychological

measures, comparisons were limited if different tests were

used. To illustrate this point, Doehring and Hoshko

(1977) and Petrauskas and Rourke (1979) used measures of

conceptual flexibility, linguistic coding, temporal

sequencing, memory, auditory-visual matching, etc., all of

which had nothing comparable in this study. Because of

reasons such as these, comparisons to the results of other

studies have been restricted.

Secondary External Validation

A third major issue in the study was addressed to

external validation of the 12-cluster solution from

nonachievement measures. The measures used for these

comparisons included intelligence, maternal and paternal

reading levels, neurological ratings, and socioeconomic

status. The relationships were significant for all

measures with the exception of the fathers' reading

scores. The children from the subgroups with the lowest

neuropsychological scores had the lowest IQs, mothers who

were the poorest readers, were most frequently evaluated

as having neurological difficulties, and were of lower

SES. The opposite pattern was found for children with

high neuropsychological performance skills. As a whole,

these results provided additional support for the valid-

ity of the neuropsychological subgroups.

The high incidence of positive neurological findings

with deficient achievers has long been reported (Cole &

Kraft, 1964; Benton, 1975); this study added further con-

firmation. That proportionally more children in this

study received an "affected" rating is not surprising

given the high frequency of poor achievers (roughly one-

third of the original sample). Furthermore, the type of

examination conducted for this project included extensive

procedures designed to detect even very subtle deviations

in neurological status. The results are, therefore,

probably not equivalent to those of other studies..

The significant relationship between maternal reading

level and the neuropsychological subgroups is interesting

because it is consistent with the familial nature of

dyslexia. Even after the confounding effects of the

mothers' educational level and SES were removed through a

covariate analysis, it was still evident that children

with poor neuropsychological performance and low academic

achievement also had mothers who could not read well.

With this study's focus on the identification of sub-

groups, it was unable to explore the genetic versus

environment controversy more fully. Other recent studies,

however, have attempted to do this.

In a well-controlled study, Owen (1978) created 76

quartets of children, with each quartet consisting of a

learning disabled child, a matched academically successful

child, and one like-sex sibling for each. The results

showed that the parents and siblings of the learning

disabled children were more similar than the academically

successful children on a variety of achievement and neuro-

psychological tests. Although the results were compatible

with genetic and/or environmental modes of transmission,

additional evidence in favor of the genetic factor came

from the aggregation of neurological immaturities in the

siblings of disabled learners.

Taylor, Satz, & Friel (1979) also addressed this

issue by examining the reading and spelling percentiles of

the parents of normal and disabled readers. Since the

parents of probands were from lower SES and educational

backgrounds, an analysis of covariance was used. Results

showed that the parents of probands were significantly

lower in both academic areas, regardless of parental sex.

Of special note was the wide discrepancy between the

spelling skills of the two groups, suggesting that the

parents of the poor readers may have compensated to some

extent for their more severe reading problems as children.

Since compensations in spelling are more difficult,

spelling disability may be a more sensitive indicator of

adult language disability than reading. Again, the re-

sults gave credence to both social and genetic mechanisms

in the transmission of reading disabilities.

A crucial problem to all studies of familial factors has

been the failure to operationally identify subgroups of

the index children (Taylor et al., 1979). It is fool-

hardy to think that a single mode of transmission exists

in all types of readers. Owen (1978) attempted to iden-

tify categories on the basis of WISC IQ patterns but found

that the subgroups overlapped too much. Therein lies this

study's solution to this problem. The empirically-derived

classification system of this study may subsequently be

used as the framework for determining which subgroups have

specific genetic or cultural modes of transmission.

An unexpected finding was the absence of a signif-

icant relationship between subgroups and fathers' reading

levels. Conceivably, the relationship with the fathers

failed to reach significance because data were available

on just under half of the fathers. Consequently, some of

the subgroups were represented by as few as four fathers.

Table 4 shows that the subgroup grade equivalents for the

fathers ranged from 10.43 years to 15.48 years. For

mothers, however, the range was from 8.71 years to 13.25

years. It appears, therefore, that the fathers were not

represented by as many poor readers. Conceivably, the

fathers who failed to participate in the examination were

also the poorer readers and they were attempting to avoid

what was seen as a potentially embarrassing situation.

Such an absence of one extreme of the distribution would

be adequate to remove statistical significance.

Naturally, reading information would ideally have been

gathered on all parents, thereby strengthening the study.

Such an endeavor, however, is very difficult in this type

of longitudinal research.

Additional improvement in this study would have come

from use of more precise socioeconomic measures. Despite

differences on this variable, the use of a dichotomous

measure of SES has not been particularly useful at the

present stage of knowledge (Rutter, 1978). Also desirable

would be complete Wechsler intelligence scores and

measures of reading comprehension on each child, but such

ideal situations quickly fall prey to practicality in a

project of this scope.

If adequate funding were available, cluster analytic

investigation should be cross-validated on all children in

a geographical area, independent of sex or race.

Increasing the overall sample size would proportionally

increase the subgroup sample sizes. This would make it

possible to evaluate separate hypotheses for each homogene-

ous subgroup, such as specific neurological patterns,

patterns consistent with lag theorists, possible

differences in cognitive strategies, etc. If subgroups

were large enough, it would be interesting to perform

separate stepwise discriminant function analyses to

determine if different subgroups have different rankings

of predictor variables obtained at kindergarten. For each

subgroup, then, it could be determined if variables from

a particular factor-type were better predictors of

reading outcome. Another project might use Zigler's

approach to mental retardation (1969) and compare a

subgroup of deficient readers to an achievement-matched

group of younger average readers. Such an approach might

clarify whether the deficient children are simply delayed

on the tasks or if there is an actual difference in their


In summary, the present results lend further credence

to the position that reading is a highly complex activity

which is dependent upon many cognitive skills (Maliphant

et al., 1974; Benton, 1978; Rutter, 1978; Fletcher & Satz,

1979). Anyone who assumes that a unitary mechanism

(rate, deficit, or otherwise) accounts for a major

proportion of the variance in reading has simply been

misled. Consequently, it has not proved easy to

determine the relative importance of each skill in either

reading competence or reading difficulties. Twelve

independent patterns of neuropsychological performance

were identified by this investigation and related to 12

patterns of achievement. Through the use of factor

scores, credence was given to the position by Cruickshank

(1977) and Fletcher and Satz (1979) that both visual-

spatial and conceptual-linguistic processes interact with

developmental factors in reading proficiency. A high

frequency of neurodevelopmental subtype patterns was

associated with an increased risk of poor achievement.

The present results, while promising, still

demonstrate the complexity and ambiguities associated with

reading competency. Systematic studies using homogen-

eous, well-defined developmental subgroups seem to offer

the greatest hope for an understanding of the mechanisms

and processes of learning to read. From that may come

more effective preventive and remedial reading programs.


Factor Score Standard Deviations















for Neuropsychological Subgroups





































































































University of Florida Home Page
© 2004 - 2010 University of Florida George A. Smathers Libraries.
All rights reserved.

Acceptable Use, Copyright, and Disclaimer Statement
Last updated October 10, 2010 - - mvs