The perception of contingency in depressed and nondepressed subjects

MISSING IMAGE

Material Information

Title:
The perception of contingency in depressed and nondepressed subjects
Physical Description:
viii, 178 leaves : ill. ; 28 cm.
Language:
English
Creator:
Miller, Ricky Earl, 1954-
Publication Date:

Subjects

Subjects / Keywords:
Depression, Mental   ( lcsh )
Affective disorders   ( lcsh )
Human information processing   ( lcsh )
Genre:
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )

Notes

Thesis:
Thesis (Ph. D.)--University of Florida, 1983.
Bibliography:
Includes bibliographical references (leaves 173-177).
Statement of Responsibility:
by Ricky Earl Miller.
General Note:
Typescript.
General Note:
Vita.

Record Information

Source Institution:
University of Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 000354661
notis - ABZ2800
oclc - 09875748
System ID:
AA00003841:00001

Full Text












THE PERCEPTION OF CONTINGENCY IN DEPRESSED
AND NONDEPRESSED SUBJECTS







BY

RICKY EARL MILLER


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



UNIVERSITY OF FLORIDA


1983
















ACKNOWLEDGEMENTS


I would like to take this opportunity to thank Dr.

Cynthia Belar, Dr. Wallace Mealiea, Dr. Michael Radelet,

and Dr. Jalie Tucker for the assistance they have provided

while serving as members of my supervisory committee. I

am especially grateful to my committee chairperson, Dr.

Suzanne Johnson, for her invaluable guidance and support

throughout the various stages of this research effort.

Appreciation is also extended to Dr. Nathan Perry,

for his assistance in securing laboratory space for this

research, and to Dr. William Froming and Mr. Terry Ladue,

for their assistance in implementing the screening pro-

cedures used to identify subjects for this research. I

would also like to thank Dr. Randolph Carter for his help-

ful advice regarding statistical issues.














ii

















TABLE OF CONTENTS

PAGE

ACKNOWLEDGEMENTS.......................................ii

ABSTRACT........ .......................................vi

CHAPTER

ONE INTRODUCTION....................................1

Review of Relevant Literature..................1
Introduction to the Concept of
Contingency and to the Contingency
Perception Process.........................4
The Perception of Contingency in
Nondepressed Subjects...................... 7
The Perception of Contingency in
Depressed Subjects .......................22
Discussion of Issues Relating to the
Assessment of Subjects' Contingency
Perception Tendencies .......................31
Introduction to the Current Research
Effort .......................................46

TWO METHOD......................................... 53

Subjects......................................... 53
Source of Subjects........................... 53
Subject Selection Procedure................. 54
Subject Selection Instruments Employed.........55
Beck Depression Inventory................... 55
Self-rating Depression Scale.................56
Experimental Design.............................58
Overview of Procedure........................58
Perception of Contingency Task..............58
Follow-up Questionnaire.....................62
Overview of Statistical Procedures............. 71

THREE RESULTS.......................................... 72

Description of the Final Sample................ 72
Perception of Contingency Measures .............75
Percentage of Trials in Each
Feedback Mode............................75


iii










Probability of the Bell Given a
Correct Versus an Incorrect Response.....86
Global Contingency Ratings ..................94
Interrelationships Among Subjects'
Contingency Judgements ...................96
Performance-Related Dependent Measures.......... 99
Ability Estimates............................99
Control Estimates...........................103
Recall of Positive Outcomes................ 103
Number of Correct Attempts................. 109

FOUR DISCUSSION.... ...............................117

Differences Between Depressed and
Nondepressed Subjects' Contingency
Perception Tendencies......................117
Review and Discussion of Major
Findings.............................. 117
Discussion of Possible Methodological
Confounds.............................. 122
Integration of the Current Findings
With Prior Research..................... 128
Review and Discussion of Other Findings.......134
Differences Between Depressed and
Nondepressed Subjects' Responses
On the Performance-Related Measures .....134
Other Findings Relating to the
Contingency Perception Process..........140
Other Findings on the Performance-
Related Measures.......... ...... ....... 141
Concluding Comments and Suggestions
For Future Research.........................142

APPENDICES

A SUBJECTS' INSTRUCTIONS FOR THE PERCEPTION
OF CONTINGENCY TASK..........................145

B STIMULUS WORDS, EQUAL PROBABILITY ASSOCIATES,
AND NONCONTINGENT FEEDBACK SCHEDULES
EMPLOYED IN THE PERCEPTION OF CONTINGENCY
TASKS.................... .. ........ 149

C FULL VERSIONS OF THE QUESTIONS COMPRISING
THE FOLLOW-UP QUESTIONNAIRE................. 154

D SUBJECTS' ANSWER SHEET FOR THE FOLLOW-UP
QUESTIONNAIRE...............................158

E COMPLETE RESULTS OBTAINED FROM THE REPEATED
MEASURES ANALYSES OF VARIANCE FOR ALL
DEPENDENT MEASURES.......................... 161









REFERENCE NOTES .................................. 172

REFERENCES................................. ......... 173

BIOGRAPHICAL SKETCH................................... .178
















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



THE PERCEPTION OF CONTINGENCY IN DEPRESSED
AND NONDEPRESSED SUBJECTS

By

Ricky Earl Miller

April 1983

Chairperson: Suzanne B. Johnson, Ph.D.
Major Department: Clinical Psychology


The current research is concerned with the process

by which subjects judge whether or not a causal relation-

ship exists between their actions and the occurrence of

subsequent outcomes-a judgemental process termed the per-

ception of contingency. In particular, this research

attempts to test the theoretical formulations of the

learned helplessness model of depression, which suggests

that depressed subjects should underestimate the degree

of contingency between their responses and subsequent

outcomes, relative to nondepressed subjects.

A new methodology for assessing subjects' contingency

perception tendencies was developed for this research in

order to correct specific difficulties identified in a









critical review of prior assessment procedures. In this

methodology, subjects were faced with an ambiguous word

association task with the understanding that they would be

receiving three different types of feedback regarding

their performance on the task: contingent, positive non-

contingent, and negative noncontingent. In fact, all feed-

back was noncontingently administered in accordance with

three predetermined reinforcement schedules. At the end

of each task, subjects were asked to estimate the degree

of contingency that was present in the task using several

different rating scales.

Subjects for this research were 40 depressed and 40

nondepressed undergraduate psychology students at the

University of Florida who had been identified on the basis

of their scores on two different depression inventories

administered several weeks apart. The obtained results

were in direct opposition to the predictions of the learned

helplessness model: Depressed subjects scored in the di-

rection of greater contingency perception, relative to the

nondepressed subjects, on all seven contingency perception

measures, obtaining significantly higher contingency rat-

ings on four of the seven measures.

These findings were seen to be consistent with more

traditional self-blame models of depression and with prior

attributional findings indicating that depressed subjects

assume more causal responsibility for events occurring in


vii









the world around them than do nondepressed subjects. A

number of additional findings relating more generally to

the perception of contingency process and to other differ-

ences between depressed and nondepressed subjects were

also obtained and discussed.


viii

















CHAPTER ONE
INTRODUCTION



An extremely important aspect of human psychological

functioning is the process by which subjects infer and

interpret causality in the world around them. Among the

most important of the various types of causal interpreta-

tions made by subjects are their judgements as to whether

or not a causal relationship exists between their actions

and the occurrence of subsequent outcomes: It is at least

partially on the basis of such judgements that subjects

accept or deny responsibility for events occurring in their

environments and assess their potential for exerting per-

sonal control over these events. This judgemental process

has been termed the perception of contingency.

Overall, the perception of contingency process has

received minimal research attention. However, there has

been an increased interest in this process during the past

few years, generated primarily by the formulations of a

theoretical model of depression which suggests that the

perception of contingency process may be centrally involved

in the etiology of at least some forms of depressive dis-

order. This theoretical formulation has been termed the


1









learned helplessness model of depression (Abramson, Selig-

man, & Teasdale, 1978; Seligman, 1975).

The learned helplessness (LH) model of depression

emerged out of a series of animal experiments concerned

with the effects of aversive stimulation (See Maier &

Seligman, 1976, for a review of this literature). These

studies found that subjects receiving noncontingent aver-

sive stimulation subsequently displayed disruptive emotion-

al responding, motivational deficits, and a proactive in-

terference with learning processes relative to controls

receiving equal amounts of escapable aversive stimulation.

Seligman (1975) hypothesized that these debilitating

characteristics, termed the "learned helplessness effect,"

might result from the animal's having learned that out-

comes are independent of its responding. He further sug-

gested parallels between the characteristics of learned

helplessness in his laboratory animals and the symptoms

of reactive depression in man, arguing that at least some

forms of depressive disorder may be associated with the

tendency to perceive a lack of contingency between one's

responses and subsequent outcomes.

Seligman and his associates (Abramson et al., 1978)

have subsequently revised the original LH model by adding

an attributional component to the original formulations.

Both the original and revised LH models have generated

great amounts of empirical research. Although these










studies have generally yielded supportive findings, re-

search in this area has been criticized for its failure

to directly address the fundamental components of the

model: "Surprisingly, after five years of intensive re-

search on human helplessness, the basic postulates of the

theory have not been adequately tested" (Alloy & Abramson,

1980, p. 60). In particular, it appears that it is the

central contention regarding the proposed differences in

the contingency perception tendencies of depressed and

nondepressed subjects that has not been adequately demon-

strated.

The current research focuses upon the perception of

contingency process and its relationship to depression.

The following review of relevant literature in this area

will first provide a brief introduction to the concept of

contingency and to the contingency perception process; a

review of prior research concerning the contingency per-

ception tendencies of depressed and nondepressed subjects

will then be presented. This literature review will be

followed by a discussion of conceptual and procedural

issues relating to the evaluation of subjects' contingency

perception tendencies, and, finally, by an introduction to

the current research effort.









Review of Relevant Literature



Introduction to the Concept of Contingency and to the
Contingency Perception Process

The concept of contingency has been employed in a

number of contemporary theories of learning (See Alloy &

Abramson, 1979, and Alloy & Seligman, 1979, for a discus-

sion of the development of the contingency concept in

learning theory). The concept of contingency that is

employed in the LH model is derived from an instrumental

learning framework, wherein it refers to the degree of

relationship between a subject's response and a subsequent

outcome. Although the current research is concerned pri-

marily with subjects' subjective contingency estimates

(and particularly with possible biases in these estimates),

it is important to understand how the objective degree of

contingency between a response and an outcome is concep-

tualized and quantified.

Seligman, Maier, and Solomon (1971) have suggested

that the degree of contingency between a response and a

subsequent outcome may be conceptualized in terms of the

conjoint variations in two response-outcome probabilities:

the conditional probability of the outcome given the pres-

ence of the response, p(O/R), and the conditional proba-

bility of the outcome given the absence of that response,

p(O/R). A response-outcome relationship is considered






5

to be fully contingent if, and only if, one of the condi-

tional probabilities is equal to 1.0 and the other is

equal to 0.0 (i.e., if Ip(O/R) p(0/ )l = 1.0). A re-

sponse-outcome relationship is considered to be totally

noncontingent if, and only if, the two conditional proba-

bilities are equivalent (i.e., if |p(O/R) p(O/R)I = 0.0).

Intermediate degrees of contingency are indicated when the

absolute value of the arithmetic difference between the

two probabilities falls between 0.0 and 1.0 (i.e., when

0.0 < p(0/R) p(0/R) < 1.0).

It is possible to quantify the objective degree of

contingency between a response and an outcome, using the

above system, when the two conditional probabilities are

known: The degree of contingency is equal to the absolute

value of the arithmetic difference between the two proba-

bilities (i.e., the degree of contingency = Ip(0/R) -

P(0/R)I ). For example, if p(0/R) = .80 and p(0/R) = .50,

then the degree of contingency is equal to .30. It is

important to recognize that response-outcome contingency

is independent of overall outcome frequency. For example,

there is an equal degree of contingency between a response

and an outcome where p(0/R) = .90 and p(O/R) = .70 as

there is where p(0/R) = .30 and p(0/R) = .10. Thus, there

may be just as much, or as little, contingency present

when a subject obtains a specific outcome on 90% of the

trials as there is when he receives it on 30%, depending

upon the specific conditional probabilities involved.









In spite of the fact that a number of research efforts

have examined the subjective contingency judgements of

depressed and nondepressed subjects (this research is re-

viewed below), it is only within the past two years that

a theoretical model of the contingency perception process

has been set forth. Crocker (1981) has proposed that the

process by which subjects make covariation judgements,

such as that employed in the perception of contingency,

is not a unitary one, but one that is composed of several

more or less clearly defined stages or steps. In step

one of her model, subjects must determine what data are

relevant to the covariation judgement, and then, in step

two, the relevant populations must be sampled. In step

three, subjects must assign values to the variables of in-

terest and classify them as either confirming or discon-

firming cases (i.e., as either indicative or contraindic-

ative of a contingent relationship). In step four, sub-

jects must recall the evidence and estimate the frequen-

cies of confirming and disconfirming feedback. Finally,

in step five, subjects must integrate the evidence to

obtain some overall conclusion regarding the degree of

covariation in the observed instances. Crocker also sug-

gests that biases and distortions may enter the contingen-

cy judgement process at any one, or all, of these judge-

mental steps.









Although Crocker's model remains essentially untested

at the present time, it is an intuitively appealing model

which does provide some conceptual framework for discussing

the perception of contingency process. Some of the re-

search findings reviewed below, as well as some of the find-

ings from the current research, will be discussed in rela-

tion to this model of the contingency judgement process.


The Perception of Contingency in Nondepressed Subjects

The first area of research to be reviewed here in-

volves a number of findings which suggest that subjects

are generally insensitive to the distinction between con-

tingent and noncontingent events (this area of research

is concerned with the perception of event-event relation-

ships, not specifically with response-outcome relation-

ships). Hake and Hyman (1953), for example, presented

subjects with either a randomly or probabilistically deter-

mined series of binary events, asking subjects to predict,

on a trial-by-trial basis, what the next event would be.

Many of the subjects were found to perceive the random

pattern as probabilistically determined. Similarly,

Smedslund (1963) had subjects examine a number of cards,

each of which indicated either the presence or absence of

a symptom and/or a disease. Subjects were found to erro-

neously deduce that there was a consistent relationship

between the symptoms and diseases when, in fact, there was

none. It appeared to Smedslund that subjects'erroneous










judgements resulted from their tendency to focus upon the

positive instances of co-occurrence of symptom and disease,

and by ignoring negative instances. Thus, these findings

indicate that subjects tend to perceive contingent rela-

tionships between events even when these events are noncon-

tingently related. This finding suggests that subjects

may also tend to erroneously perceive contingency between

their responses and subsequent outcomes under noncontingent

conditions.

A number of studies have yielded findings which may

be interpreted to indicate that subjects do tend to per-

ceive contingent relationships between their responses and

subsequent outcomes under noncontingent conditions. In

concept identification experiments, for example, subjects

have commonly been found not to recognize when positive

feedback is being received noncontingently; instead, they

have tended to report complex deterministic rules that

they feel govern their reinforcement (e.g., Levine, 1971).

Similarly, laboratory studies of "superstitious" responding

in humans (e.g., Bruner & Revusky, 1961; Wright, 1962)

have found that subjects typically develop complex patterns

of responding when faced with noncontingent feedback, pre-

sumably reflecting their beliefs that these outcomes are

determined by their responses. Thus, there are a number

of findings which suggest that subjects may be relatively

insensitive to the presence of noncontingent relationships,









displaying, instead, a bias toward the perception of con-

tingent relationships between their responses and subse-

quent outcomes.

In a related area of research, Langer (1975, 1978)

has noted that subjects frequently behave as if they can

control chance events (e.g., in gambling and in games of

chance). She has referred to this phenomenon as the

"illusion of control." The findings reviewed above may

be seen to represent such an illusion of control: The sub-

jects in these studies have behaved as if they could con-

trol the noncontingent outcomes they received. Langer

(1978) has discussed a number of general factors which may

help to explain why subjects make such judgemental errors:

"(1) persons are positively motivated to master their en-

vironment; (2) persons are motivated to avoid the negative

consequences of having no control; (3) persons may fail

to make the distinction between skill and chance elements

because of their simultaneous presence in most events; and

(4) it may be strategically efficacious for persons to

treat all events as controllable" (pp. 186-187).

Thus, Langer suggests that subjects may be broadly

biased toward the perception of contingency and control in

regard to the outcomes that they receive. However, she

has also presented a theory about the illusion of control

which suggests that specific environmental factors may

facilitate such biased perceptions: "The theory states










that those conditions that allow or encourage participants

to behave as if they are participating in a skill event,

induce an illusion of control" (pp. 200-201). In this re-

gard, it might be suggested that the findings reported

above, indicating that subjects are biased to perceive

contingency between their responses and subsequent out-

comes, may be at least partially reflective of demand

characteristics in the experimental settings which suggest

to subjects that they are participating in skill events.

Thus, it may be that subjects would not evidence such a

bias under conditions where contingency is not implied by

the characteristics of their environment.

Some support for the above suggestion has been ob-

tained by Peterson (1980) who was particularly concerned

with the above findings regarding subjects' apparent ten-

dency to misperceive contingency between noncontingent

events (these findings do not directly concern subjects'

perceptions of response-outcome relationships). He sug-

gested that the subjects participating in these experimen-

tal studies may have failed to consider the hypothesis

that events are randomly associated because of their as-

sumptions and expectations regarding such studies. In

order to test this suggestion, Peterson had subjects de-

scribe two series of binary events, one random and one

patterned. He found that subjects correctly identified

the random sequence only when the use of randomness had










been legitimized, either directly (through instructions)

or indirectly (through experience with a prior comparison

sequence). Thus, it appears that subjects' observed ten-

dency to perceive contingent relationships between unre-

lated events may be at least partially accounted for by

the demand characteristics of the experimental setting.

Similarly, it appears that these characteristics might also

partially account for the findings that subjects tend to

perceive contingency between their own responses and sub-

sequent outcomes.

Jenkins and Ward (1965) conducted a much more exten-

sive examination of the contingency perception process than

did the above studies. They presented subjects with a

series of differing contingency judgement problems under

a variety of instructional.and procedural conditions. For

each problem, subjects were presented with 60 trials of a

task in which their choice between two responses (pressing

either of two buttons) was followed by one of two possible

outcomes (illumination of one of two feedback lights) on

each trial. Subjects' responses were contingently related

to the outcomes to varying degrees in some problems (i.e.,

to varying degrees, pressing one button would result in a

particular outcome more frequently than would pressing the

other button), and noncontingently related in others (i.e.,

the outcomes were equally likely to occur given either of

the two responses). The problems differed not only in










the degree of contingency that was present, but also in the

overall frequency that the outcomes occurred (e.g., some

problems involved the same degree of contingency but dif-

fered in outcome frequency). Under one instructional con-

dition, subjects were directed to obtain a particular out-

come as often as possible (Score condition), whereas in

another they were directed to learn to produce either out-

come at will (Control condition). Subjects were also in-

volved in the tasks either as active participants (i.e.,

actually choosing which button to press on each trial) or

as spectators (i.e., observing others' responses on the

tasks and the resulting outcomes). Also, the instructions

for all subjects specifically indicated that events might

be unrelated to their responding during these problems.

At the end of each problem subjects were asked to

"make a judgement of the degree of control which had been

exerted over the outcomes by response choices" (Jenkins &

Ward, 1965, p. 6). The researchers explained that the term

"control" was substituted for that of "contingency" in

their instructions to subjects and in their rating scale

"because in the context of the task it seemed to be the

most natural way to communicate the technical meaning of

contingency with everyday language" (p. 2). (It will be

argued later that the terms "contingency" and "control"

do not convey the same meaning.) They found that subjects'

ratings on the judgement of control scale did not correspond









at all to the objective contingencies in the problems.

More specifically, they found that subjects' ratings on

this scale were highly correlated with the number of suc-

cessful trials they received on the tasks (i.e., with the

number of trials on which they actually received the out-

come they attempted to obtain). Jenkins and Ward concluded

from these findings that subjects lack a "statistical con-

cept of contingency" and that they "do not distinguish the

ability to manipulate outcomes from their ability to pre-

dict them" (p. 17).

Thus, while the first set of findings reviewed above

suggested that subjects may be broadly biased toward the

perception of contingency, and away from the perception of

noncontingency, Jenkins and Ward's findings suggest that

this type of bias may operate only under conditions in

which subjects receive large numbers of desired outcomes.

Further, these findings suggest that subjects may actually

tend to underestimate the degree of contingency between

their responses and subsequent outcomes under conditions

in which they they receive desired outcomes infrequently.

The discrepancy between these two sets of findings may be

related to the fact that the Jenkins and Ward procedure

legitimized the perception of noncontingency and/or to the

fact that the earlier studies did not examine conditions

in which subjects strived for, but were unable to obtain,

desired outcomes (most of the earlier research was concerned









either with event-event relationships or with response-

outcome relationships under positive reinforcement condi-

tions).

The findings of Jenkins and Ward (1965) have been

largely supported by another, more recent, examination of

the contingency perception process. Alloy and Abramson

(1979) conducted a series of experiments specifically for

the purpose of contrasting the contingency perception ten-

dencies of depressed and nondepressed subjects (the find-

ings relating to the depressed subjects will be reviewed

at a later point). In their procedures, subjects were

faced with one of several contingency judgement problems

which were similar to those employed by Jenkins and Ward:

Subjects were directed to assess the degree of relationship

between a single response choice (pressing versus not press-

ing a button) and the occurrence or nonoccurrence of a

single outcome (onset of a green light). The major depen-

dent measure consisted of subjects' ratings on a judgement

of control scale which was also similar to that employed

by Jenkins and Ward. Also as in the Jenkins and Ward pro-

cedure, subjects were specifically instructed that the out-

comes in the problems might be unrelated to their responses.

In Alloy and Abramson's (1979) Experiment I, subjects

were presented with problems in which the degree of con-

tingency between their responses and green light onset

was .25, .50, or .75. The problems were counterbalanced









so as to control for whether pressing or not pressing re-

sulted in the higher percentage of green light onset, and

the problems were selected in such a way that the overall

percentage of green light onset was negatively correlated

with the degree of contingency in the problems (e.g., the

response-outcome probabilities were such that the green

light occurred more often in the .25 condition than in

the .75 condition). The latter characteristic was included

to offset the relationship noted in the Jenkins and Ward

(1965) study between the frequency of the desired outcome

and subjects' judgements of contingency.

Alloy and Abramson (1979) found, in their first exper-

iment, that there was a close correspondence between sub-

jects' ratings on the judgement of control scales and the

objective contingencies present in the problems, and that

these ratings did not correlate with the frequency of

green light onset. They noted that these findings seemed

quite discrepant from those of Jenkins and Ward, and they

suggested that these differences might relate to differ-

ences in the operationalization of the contingency concept

in the two studies (viz., that the Jenkins and Ward proce-

dure examined the relationships between two responses and

two outcomes, whereas their own study examined the rela-

tionship between the presence or absence of a single re-

sponse and the occurrence or nonoccurrence of a single

outcome). However, it appears that the findings of the










two studies may not be as discrepant as they initially ap-

pear. It must be recalled that Jenkins and Ward employed

a variety of contingency judgement problems under several

different instructional and procedural conditions; thus,

the Alloy and Abramson findings may relate to only a specif-

ic subset of the Jenkins and Ward findings. When one ex-

amines only the one condition of the Jenkins and Ward study

that most closely parallels the Alloy and Abramson procedure

(the contingent problem, active subject, control instruction

condition), it is seen that these subjects were also quite

accurate in their contingency assessments (estimates of

35.5% and 61.2% for conditions where the objective degrees

of contingency were .30 and .60, respectively).

Thus, there appears to be some consistent evidence to

indicate that subjects may accurately perceive contingency

under certain types of conditions. An examination of the

conditions under which subjects' ratings did parallel ob-

jective contingencies in the two studies suggests that

there are two factors which may influence the accuracy of

subjects' ratings: the actual degree of contingency pres-

ent in the problem, and subjects' instructions regarding

their task in relation to the problems. Subjects' ratings

appear to be least likely to parallel objective contingen-

cies when they are faced with objectively noncontingent

outcomes and/or they are specifically directed to obtain

particular outcomes (as in Jenkins and Ward's Score









condition). Alloy and Abramson (1979) have commented on

this first factor, suggesting that "noncontingency is a

psychologically more difficult relationship to perceive

or understand than is contingency" (p. 474). The signif-

icance of the second factor will be discussed at a later

point.

In their Experiment II, Alloy and Abramson asked sub-

jects to judge the degree of contingency present in one

of two problems in which responses and outcomes were non-

contingently related. In one of the problems the green

light occurred noncontingently on 75% of the trials; it

occurred on only 25% of the trials in the second problem.

The results indicated that all subjects accurately detected

the noncontingency in the 25% reinforcement condition,

whereas the nondepressed subjects gave high ratings on the

judgement of control scale in the 75% reinforcement condi-

tion. Thus, it appears that the nondepressed subjects

displayed an illusion of control over the positive outcomes

in the 75% condition, but that they accurately detected

the noncontingency in the 25% condition. This pattern of

results is consistent with the overall pattern obtained

by Jenkins and Ward (1965): It suggests that nondepressed

subjects' contingency judgements are closely tied to rein-

forcement frequency.

The finding that nondepressed subjects'judgements of

control did not correspond to objective contingencies in










this experiment, but that they did in Experiment I, sup-

ports Alloy and Abramson's (1979) suggestion that noncon-

tingency may be a particularly difficult concept for sub-

jects to understand (at least for nondepressed subjects).

Alternately, however, it appears that nondepressed sub-

jects' judgement of control ratings in this second experi-

ment may not have been actually reflective of their contin-

gency perception tendencies. In particular, it appears

that these subjects may not have interpreted this task in

the manner intended by the experimenters. In this regard,

it may be noted that a number of subjects, particularly

nondepressed subjects, reported attempting complex hypoth-

eses on this task (e.g., pressing every other time, press-

ing early or late in the response period, etc.)--this was

not reported in regard to Experiment I. This finding

seems to indicate that these subjects did not interpret

their tasks in the manner intended by the experimenters,

since they should have been concerned only with the rela-

tionship between the two identified response choices (press

or not press) and the occurrence of the outcomes if they

had interpreted them correctly. This apparent misinter-

pretation of the task by the nondepressed subjects seems to

indicate that these subjects were not involved simply in

the assessment of the response-outcome relationships in

the problems, but, rather, that they were involved in ef-

forts to exert control over the outcomes in the task. The










suggestion that nondepressed subjects may have misinter-

preted their task in this experiment will be considered in

more detail at a later point.

In Experiments III and IV, Alloy and Abramson attempt-

ed to examine subjects' contingency judgements under con-

ditions which more closely paralleled real life, in that

they required subjects to make contingency judgements while

actively striving to obtain particular outcomes. In order

to accomplish this, subjects were assigned to either a

"win" or a "lose" condition. In the win condition, sub-

jects started the task with no money, but they were in-

structed that they could win money by making the green

light come on during the experiment; subjects in the lose

condition were given money at the outset of the experi-

ment, but they were told that they would lose money every

time the green light failed to come on. In Experiment III,

the occurrence of the green light was noncontingently re-

lated to subjects' responding and set to occur on 50% of

the trials. In Experiment IV, the onset of the green

light was contingently related to subjects' responding at

the .50 level of contingency. Thus, in these experiments,

Alloy and Abramson were not concerned with reinforcement

frequency, but, rather, with outcome valence: Although

subjects in the two conditions received essentially equal

amounts of positive versus negative feedback, the valence

of the outcomes differed as a result of the procedural dif-

ferences in the conditions.









The results for Experiments III and IV indicated that

nondepressed subjects gave significantly higher ratings on

the judgement of control scales in the win condition than

in the lose condition for both the contingent and noncon-

tingent problems. Their ratings accurately corresponded

with the objective degree of contingency only in the lose

condition of the noncontingent problem, where their ratings

accurately reflected the noncontingent nature of the feed-

back they received. Otherwise, they underestimated the

objective degree of contingency in the lose condition and

overestimated it in the win condition. These findings

seem quite consistent with the earlier findings indicat-

ing that nondepressed subjects' contingency perceptions

are closely tied to reinforcement frequency: Both sets of

findings indicate that nondepressed subjects tend to per-

ceive contingency in a self-serving manner, overestimating

contingency under positive conditions (high reinforcement

frequency and/or positively valenced feedback) and under-

estimating contingency under negative conditions (low rein-

forcement frequency and/or negatively valenced feedback).

It may be noted, however, that almost all of the subjects

in Experiment III reported attempting complex hypotheses

on the problems (the number of complex hypotheses attempted

in Experiment IV was not reported). As noted above, this

finding may be seen to indicate that these subjects may not

have interpreted their tasks in the manner intended by the









experimenters, and, therefore, that the obtained results

may not actually be reflective of subjects contingency

perception tendencies (Again, this suggestion will be dis-

cussed in more detail at a later point).

In summary, prior research has yielded two distin-

guishable proposals regarding the contingency perception

tendencies of nondepressed subjects. The early research

reviewed above suggested that these subjects might be

broadly biased toward the perception of contingent rela-

tionships and away from the perception of noncontingency.

However, it was noted that these findings were based upon

examinations of only limited aspects of the contingency

judgement process, and that they might also be reflective

of demand characteristics in the experimental settings

which encouraged subjects to interpret noncontingent events

as being contingently determined. More recent examinations

of the contingency judgement process, which have been more

comprehensive and have lacked the overt demand character-

istics referred to above, have indicated that while sub-

jects may accurately perceive contingency under some limit-

ed conditions, they typically perceive contingency in a

self-serving manner, tending to overestimate contingency

under positive conditions and underestimate it under nega-

tive conditions.

The suggestion that nondepressed subjects may display

a self-serving bias in their contingency perceptions is









consistent with a growing body of research findings which

indicate that nondepressed subjects frequently display

self-serving biases in their interpretations of events,

presumably relating to their efforts to protect and/or

enhance self-esteem. For example, a large body of research

has indicated that subjects typically make causal inter-

pretations in a self-serving manner, tending to attribute

positive outcomes to internal factors and negative out-

comes to external factors (See Miller & Ross, 1975, and

Bradley, 1978, for reviews of this research). Similarly,

nondepressed subjects have been found to underestimate the

amount of negative feedback they receive on performance

tasks (e.g., Nelson & Craighead, 1977), and to exaggerate

their level of social competence relative to the judge-

ments of other people (Lewinsohn, Mischel, Chaplain, &

Barton, 1980). Thus, there are a number of findings which

indicate that nondepressed subjects frequently employ self-

serving biases in their interpretations of events, and the

findings reviewed above suggest that such biases may also

be involved in their perception of response-outcome rela-

tionships.



The Perception of Contingency in Depressed Subjects

It was noted earlier that while the LH model has gen-

erated great amounts of empirical research, few studies

have attempted to directly test the fundamental prediction









of the model: that depressed subjects should underesti-

mate contingency relative to nondepressed subjects. Miller

and Seligman (1973) were the first to attempt to develop a

methodology for contrasting the contingency perception

tendencies of depressed and nondepressed subjects. They

developed an expectancy-change procedure which was derived

from a methodology employed in early research work concern-

ing Rotter's (1966) locus of control construct. Rotter

and his associates (e.g., Rotter, Liverant, & Crowne, 1961)

engaged subjects in tasks in which success appeared to be

determined either by chance (i.e., noncontingently) or

skill (i.e., contingently). They, then, had subjects ver-

balize their expectations for future success on a trial-

by-trial basis as they progressed through the task. They

found that subjects' verbalized expectancies for future

successes on the tasks were related to the interaction

between the outcomes they received and the skill-chance

nature of the task. In particular, they found that out-

comes of previous trials had a greater impact upon subjects'

expectancies for future successes when they believed the

outcomes were skill determined (contingent) than when they

believed the outcomes were chance determined (noncontin-

gent). Thus, Miller and Seligman reasoned that subjects

tending to perceive outcomes as noncontingently determined,

as depressed subjects are hypothesized to do by the LH

model, should exhibit smaller expectancy changes when









faced with ostensibly skill tasks than should those sub-

jects who are more prone to perceive contingency between

their responses and subsequent outcomes-no differences

are predicted for the expectancy changes of the two groups

in the chance condition, since all subjects should tend

to perceive noncontingency in this condition.

Miller and Seligman's (1973) results were supportive

of the LH model: Depressed subjects were found to display

smaller expectancy changes for skill, but not for chance,

tasks relative to the nondepressed subjects. These find-

ings have generally been supported in a number of addi-

tional studies which have employed a similar methodology

(e.g., Klein & Seligman, 1976; Miller, Seligman, & Kur-

lander, 1975). However, the expectancy-change methodol-

ogy has been seriously questioned on a number of grounds.

First, Dweck and Gilliard (1975) have found that trial-

by-trial expectancy ratings are quite intrusive, suggest-

ing that the findings obtained from such measures must be

interpreted with extreme caution. Second, Alloy and Selig-

man (1979) have suggested that these judgements may relate

more to subjects' expectational biases regarding their

potential for future control than to biases in their per-

ceptions regarding the contingency of feedback they have

actually received. Third, attributional theorists (e.g.,

Weiner, Nierenberg, & Goldstein, 1976) have suggested that

expectancy changes of the sort involved in these studies










are primarily reflective of subjects' judgements regarding

the stability of the perceived causal factors relating to

the outcomes. Thus, the smaller expectancy changes of the

depressed subjects may reflect differences in the extent

to which they attributed the outcomes to stable factors

rather than reflecting a tendency to underestimate contin-

gency. On the basis of the alternate explanations of the

expectancy-change results presented above, Seligman and

his associates (Abramson et al., 1978; Alloy & Abramson,

1979) have concluded that the expectancy-change procedure

cannot be assumed to accurately assess subjects' contin-

gency perception tendencies.

As already noted above, Alloy and Abramson (1979) em-

ployed a procedure similar to that of Jenkins and Ward

(1965) for contrasting the contingency perception tenden-

cies of depressed and nondepressed subjects. In their

Experiment I, it may be recalled that subjects were faced

with one of three contingency judgement problems in which

the objective degree of contingency was either .25, .50,

or .75. It was also noted that the judgement of control

ratings of both depressed and nondepressed subjects were

highly correlated with the objective degree of contingency

in the problems. Depressed and nondepressed subjects'

judgements were found not to differ at all in this exper-

iment. This finding is clearly in opposition to the basic

prediction of the LH model which suggests that depressed

subjects should underestimate contingency.










In Experiment II, Alloy and Abramson's subjects were

faced with problems in which their responses were unrelated

to the outcomes they received (i.e., they were objectively

noncontingent). In one problem, green light onset occurred

noncontingently on 25% of the trials; in the other condi-

tion it occurred noncontingently on 75% of the trials.

The depressed subjects' judgement of control ratings accu-

rately reflected the noncontingent nature of the feedback

under both conditions, whereas, it may be recalled, non-

depressed subjects' ratings indicated that there was a

significantly greater degree of contingency present in the

75% reinforcement condition. These findings can be inter-

preted to provide partial support for the LH model in that

they suggest that depressed subjects may be more sensitive

to the presence of noncontingent relationships, particu-

larly when receiving high-frequency noncontingent positive

feedback. However, this finding can also be interpreted

to indicate that depressed subjects are simply more accu-

rate in their contingency perceptions than are nondepressed

subjects. It may be noted, in this regard, that depressed

subjects' judgement of control ratings were uncorrelated

with the judged amount of reinforcement received, unlike

those of nondepressed subjects which were highly correlated.

This finding suggests that the depressed subjects' judge-

ments were not influenced by reinforcement frequency,

whereas nondepressed subjects' judgements were.









In their final two experiments, Alloy and Abramson

(1979) attempted to make their experimental task more like

those faced by subjects in the real world, in that they

required that subjects make contingency judgements while

attempting to obtain particular outcomes. As noted earlier,

subjects in the win condition gained money each time the

green light came on, whereas subjects in the lose condition

lost money each time the green light failed to come on.

In Experiment III, the occurrence of the green light was

unrelated to subjects' responding; in Experiment IV, the

onset of the light was contingently related to subjects'

responding at the .50 level of contingency. The results

for both experiments indicated that the depressed subjects'

judgements corresponded closely to the objective degree of

contingency in the problems, and that they did not differ

in the win and lose conditions. These findings support

the suggestion made above that depressed subjects may be

more accurate in their perceptions of contingency than are

nondepressed subjects.

Alloy and Abramson (1979) conclude on the basis of

their four experiments that "depressed people are 'sadder

but wiser' than nondepressed people" (p. 479) in their

judgements of contingency: They note that the depressed

subjects were quite accurate in their contingency judge-

ments throughout these experiments and that they did not

evidence the self-serving biases displayed by nondepressed










subjects. This pattern of results is not consistent with

the predictions of the LH model, although it is consistent

with a growing body of research findings which indicate

that depressed subjects do not display self-protective

biases commonly shown by nondepressed subjects, presumably

because the depressed subjects are less motivated to pro-

tect and/or maintain self-esteem. For example, several

studies have indicated that depressed subjects do not dis-

play the illusion of control over chance events commonly

observed among nondepressed subjects (Alloy & Abramson,

1982; Golin, Terrell, & Johnson, 1977; Golin, Terrell,

Weitz, & Drost, 1979). Depressed subjects have also been

found not to display self-serving attributional biases,

tending instead to attribute positive and negative outcomes

in a similar manner (e.g., Kuiper, 1978; Rizley, 1978).

Finally, depressed subjects have also been found to be

more accurate in their recall of performance feedback (Nel-

son & Craighead, 1977) and in their assessment of their

social competence (Lewinsohn, Mischel, Chaplain,& Barton,

1980) than are nondepressed subjects who tend to distort

these judgements in a positive direction. The degree of

consistency between the above findings and the pattern of

results reported by Alloy and Abramson (1979) provides

general support for the suggestion that depressed subjects

may be more accurate in their perceptions of contingency

than are nondepressed subjects.





-----------__________________________. ____










One additional study has attempted to contrast the

contingency perception tendencies of depressed and nonde-

pressed subjects. Miller (Note 1) exposed depressed and

nondepressed subjects to an ambiguous word association

task with the instructions that the feedback they received

regarding their performance would initially be contingent

upon their responding, but that at some point during the

course of the task they would begin to receive failure feed-

back on some trials regardless of their responses-i.e.,

the feedback would become noncontingent. In fact, feedback

was administered noncontingently for all trials of the task

in a predetermined sequence such that the frequency of neg-

ative outcomes was initially quite low, but gradually in-

creased as subjects progressed through the task. Subjects

were directed to identify the point in the task when they

perceived that they began to receive the noncontingent neg-

ative feedback by pressing a "stop" button on their control

panel. It was assumed that stopping earlier on the task

would be associated with a greater tendency to perceive

outcomes as being noncontingently determined.

Miller found that depressed subjects exhibited a mar-

ginally significant tendency to stop earlier on the task

(i.e., to perceive noncontingency earlier) than did the

nondepressed group. A high-depressed sub-group was found

to stop significantly earlier than did the nondepressed

subjects. Thus, Miller's findings do provide some support









for the basic prediction of the LH model (i.e., that de-

pressed subjects tend to underestimate contingency relative

to nondepressed subjects). However, there are methodolog-

ical issues which require that these findings be regarded

with caution. First, he found statistically significant

differences only between the nondepressed group and the

high-depressed sub-group which was identified on a post-hoc

basis. Second, there was a confound in Miller's procedure

between the perception of noncontingency and stopping

earlier on the task. This raises the possibility that

depressed subjects may have stopped earlier on the task,

not because they were more prone to perceive noncontingency,

but because they were less motivated and involved in the

task than were the nondepressed subjects. It may also be

noted that Miller's study examined subjects' contingency

perception tendencies only for negative outcomes-it may

also be important to examine subjects' contingency percep-

tion tendencies for positive events.

In summary, the above research has provided little,

if any, support for the LH model. Although the findings

from the expectancy-change research and from Miller's

methodology have supported the LH prediction that depressed

subjects should underestimate contingency relative to non-

depressed subjects, methodological considerations demand

that these findings be considered with caution. The find-

ings obtained from Alloy and Abramson's (1979) research have










provided no support for the LH model: They suggest that

depressed subjects are quite accurate in their contingency

judgements and that they do not evidence any consistent

tendency to underestimate contingency relative to the non-

depressed subjects.


Discussion of Issues Relating to the Assessment
of Subjects' Contingency Perception Tendencies


A critical evaluation of the research reviewed above

raises two major issues regarding the manner in which the

contingency perception process has been conceptualized and

assessed in these studies. First, it appears that most of

the research in this area has seriously confounded the

concepts of contingency and control in their experimental

procedures and assessment instruments, making it difficult

to determine if their findings actually reflect differences

in subjects' contingency perception tendencies or whether

they are more reflective of subjects' judgements regarding

their capacity to exert personal control over the outcomes

they received in these tasks. Second, it appears that

these studies have focused, almost exclusively, upon a lim-

ited aspect of the contingency perception process, and

largely ignored other, potentially more significant aspects

of this process. Each of these issues will be discussed

in more detail below.

It may be recalled that the term "control" was sub-

stituted for that of "contingency" in the experimental







32


instructions and assessment instruments employed in the

two major studies of subjects' contingency perception ten-

dencies (Alloy & Abramson, 1979, and Jenkins & Ward, 1965).

This substitution was based upon the assumption that the

two terms convey equivalent meanings. This assumption does

not appear to be correct. The concept of contingency was

defined earlier in terms of the degree of relationship be-

tween a subject's response and a subsequent outcome. In

contrast, the concept of control may best be defined in

terms of a subject's capacity to selectively influence out-

comes in a desired direction. Although there does seem to

be some degree of overlap between these two concepts, they

do not appear to be equivalent: The presence of a contin-

gent relationship between a response and an outcome appears

to be a necessary, but not sufficient, condition for the

exercise of control over that outcome. This conceptual

distinction suggests that while a subject's judgement of

his capacity to control an outcome may be at least partial-

ly reflective of his perception of the degree of contingen-

cy that is involved, it appears that other factors would

also influence this judgement.

The distinctions made above between the concepts of

contingency and control may be made clearer through the

use of an illustration. Assume, for example, that a fully

contingent relationship exists between response A and out-

come B-i.e., proper execution of A by the subject will










reliably result in the occurrence of outcome B. The pres-

ence of the contingent relationship between A and B indi-

cates that the subject has the potential to exercise full

control over the occurrence of outcome B, providing that

he is fully able to identify and properly execute response

A. If, however, the subject is unable, for some reason,

to identify and/or properly execute response A, then he

will be unable to exert control over the occurrence of B

in spite of the contingent relationship between A and B.

This illustration demonstrates at least one hypothetical

condition in which subjects' perceptions of contingency

and their judgements of control might be expected to be

incongruent: The subject who receives undesired feedback

in spite of a clearly contingent relationship may be ex-

pected to develop low judgements of control, but to main-

tain high perceptions of contingency.

The distinctions made in the foregoing example paral-

lel those made by Bandura (1977) between response-outcome

expectancies and feelings of self-efficacy. Subjects' re-

sponse-outcome expectancies are their beliefs that a par-

ticular outcome will follow a given response if it is exe-

cuted. In the above example, this corresponds with sub-

jects' perception of the degree of contingency between A

and B. Subjects' feelings of self-efficacy relate to their

beliefs that they will be able to identify and execute

the response necessary to obtain the desired outcome if










a contingent relationship exists. In the above example,

this corresponds with subjects' beliefs that they will be

able to identify and properly execute response A. Thus,

subjects' judgements of their potential for control over

an outcome may be seen to represent some combination of

their perception of the degree of contingency that is

present and their judgement as to their degree of efficacy

in identifying and executing the necessary controlling re-

sponse.

Reconsideration of the Jenkins and Ward (1965) and

Alloy and Abramson (1979) procedures in relation to the

distinctions made above suggests that the results obtained

in these studies may not be exclusively concerned with sub-

jects' contingency perception tendencies due to the con-

founding of the contingency and control concepts in sub-

jects' instructions for these tasks. More specifically,

it appears that the specific wording of subjects' instruc-

tions in these procedures may have encouraged subjects to

misinterpret their tasks in either of two ways: (1) that

they were to rate their actual, demonstrated degree of

efficacy in obtaining desired outcomes on these problems,

or (2) that they were to rate their perceived efficacy in

identifying and executing appropriate controlling responses

on these problems. Each of these suggestions will be con-

sidered in more detail below.









First, it would appear that the mere use of the term

"control" in the description of subjects' task would sug-

gest to subjects that their task is not so much one of

judging the degree of contingency that is present in the

problems as it is one of assessing their overall capacity

to obtain desired outcomes on these problems: To assess

one's degree on control over an outcome is to assess one's

capacity to selectively influence that outcome in desired

directions. Thus, given that subjects are directed to

rate "the degree of control which had been exerted over

the outcomes by response choices," as was the case in the

Jenkins and Ward procedure (p. 6), it does not seem partic-

ularly surprising that these ratings were highly correlated

with the actual frequency with which they obtained these

outcomes (i.e., with reinforcement frequency). Thus, the

finding that nondepressed subjects' "contingency" judge-

ments were highly correlated with reinforcement frequency

in Jenkins and Ward's study and in Experiment II of Alloy

and Abramson's study may be due to the fact that these

subjects interpreted the concept of "contingency" to be

equivalent to one's capacity to obtain desired outcomes,

as implied by the instructions they received for these

tasks.

It also appears that subjects' instructions in these

procedures were worded in such a way as to encourage sub-

jects to interpret their tasks as one of rating their









perceived efficacy in identifying and executing appropri-

ate controlling responses on tnese problems rather than

one of rating the degree of contingency that they perceived.

First, it may be seen that these instructions define the

concept in such a way that it would incorporate subjects'

efficacy judgements. For example, Jenkins and Ward's (1965)

instructions state that "complete control means that you

can produce the score light on any trial by your

choice of responses" (p. 6). Similarly, Alloy and Abram-

son's (1979) instructions state that "no control means that

you have found no way to make response choices so as to in-

fluence in any way the onset of the green light" (p. 451).

Clearly, the concept of "contingency" which is defined by

these instructions is one which concerns subjects' judge-

ments not only of the degree of contingency present in the

problems, but also of their efficacy in identifying and

executing appropriate controlling responses.

In addition to the general confounding of the contin-

gency and control conceptions, referred to above, the in-

structions in these studies also seem to directly imply at

several points that subjects do have the potential for

control over the outcomes in the task. For example, Alloy

and Abramson's instructions refer to their study as a

"problem-solving experiment" (p.451), and Jenkins and

Ward's instructions describe subjects' task as being one

of "finding a way to respond which will make the score









light appear as often as possible" (Score instructions),

and of "finding a way to control which of the outcomes .

will appear on any trial" (Control instructions) (p. 6).

Clearly, these instructions emphasize the identification

of appropriate controlling responses much more strongly

than they do the perception of response-outcome relation-

ships. In this manner, it appears that these instructions

imply to subjects that the outcomes within the tasks are

contingently related to some potential response on their

part, and, therefore, that their real task is to identify

the extent to which they are able to identify and execute

the appropriate controlling response. It appears that

this interpretation of the task would be particularly en-

couraged by the instructions for the conditions in which

subjects are specifically directed to obtain particular

outcomes, as in Jenkins and Ward's Score condition (de-

scribed above) or in Alloy and Abramson's win and lose

conditions (Experiments III and IV): "Now, in this prob-

lem-solving experiment it is your task to learn how to

turn on this green light" (p.464). Specific encouragement

to pursue a particular outcome seems tantamount to declar-

ing that a contingent response-outcome relationship does

exist, and, therefore, that subjects' task is to first

attempt to identify and execute the appropriate control-

ling responses and to then rate their degree of efficacy

in doing so.










There are specific indications in the results ob-

tained from the Alloy and Abramson (1979) study to indi-

cate that at least some subjects, under some circumstances,

did interpret their task as being more concerned with the

identification and execution of appropriate controlling

responses than with the perception of response-outcome

relationships. It may be recalled that the nondepressed

subjects in their Experiment II, and nearly all subjects

in the win and lose conditions of Experiment III, reported

attempting complex hypotheses during these tasks (Alloy &

Abramson did not report the number of subjects responding

in this manner in Experiment IV). These findings suggest

that these subjects did not interpret their task in the

manner intended by the experimenters, since they should

have been concerned only with the relationship between the

two identified response choices and the occurrence of the

outcomes if they had interpreted it as intended. These

findings also suggest that particular characteristics of

these subjects and/or these conditions may particularly

foster interpretation of the task in this manner.

In regard to the finding that all subjects reported

making complex hypotheses in Alloy and Abramson's Exper-

iment III (i.e., in the win and lose conditions), it was

noted earlier that the instructions for this experiment

seem to particularly encourage interpretation of the task

in terms of efficacy rather than contingency, because









of the fact that these instructions specified that subjects

were to pursue particular outcomes. Thus, it appears that

the self-serving bias evidenced by the nondepressed sub-

jects under the win and lose conditions might be much more

reflective of biases and distortions in subjects' judge-

ments regarding their efficacy in identifying and executing

appropriate controlling responses than they are of their

contingency perception tendencies. Although it might seem

that such distorted efficacy judgements could not occur

on this type of task because of the extremely limited and

simplistic response repertoire involved (either press or

do not press the button), it must be recalled that sub-

jects apparently did not perceive their response repertoire

in this manner. The fact that subjects engaged in complex

hypothesis testing on these tasks suggests that subjects

perceived that they had a wide range of response choices.

Thus, it appears that subjects' interpretations of their

response repertoire would provide them with sufficient

latitude to misperceive either that they were or were not

successful in identifying and executing some complex con-

trolling response.

It is interesting to note, in regard to the above dis-

cussion, that depressed subjects' ratings on the judge-

ment of control scales in Alloy and Abramson's (1979)

study did correspond with the actual response-outcome con-

tingencies present in the problems. There appear to be









at least two, possibly overlapping, explanations for this

discrepancy from the performance of the nondepressed sub-

jects. First, it may be that the depressed subjects, in

contrast to the nondepressed subjects, actually did inter-

pret their task in these experiments as one of contingency

perception, as intended by the experimenters. This sugges-

tion is at least partially supported by the finding that

depressed subjects attempted fewer complex hypotheses in

Alloy and Abramson's (1979) Experiment II than did did the

nondepressed subjects. Such differing interpretations of

their tasks by the two groups could reflect differences

in the salience of the two types of judgements in the two

groups: For example, depressed subjects may be particu-

larly concerned with assessing the degree of contingency

between their responses and subsequent outcomes, whereas

nondepressed subjects may be relatively more concerned

with evaluating their capacity to identify and execute con-

trolling responses. It also appears possible that the dif-

ferences in the performance of the two groups may reflect

real differences between the two groups, but in regard to

factors other than contingency perception. For example,

these findings may indicate that depressed subjects are

relatively accurate in their assessments of their efficacy

in identifying and executing controlling responses, whereas

nondepressed subjects may display self-serving biases in

these judgements.










The second major issue raised by the review of prior

research relating to the contingency perception tendencies

of depressed and nondepressed subjects concerns the neglect

of a potentially significant aspect of the contingency

judgement process. In almost all of the studies reviewed

above the contingency perception process has been reduced

to an essentially mathematical, information-processing

phenomenon. In the Alloy and Abramson (1979) procedure,

for example, subjects' task, as conceptualized by the ex-

perimenters, can be summarized as follows: (1) press the

button on about half of the trials and do not press it on

the remainder, (2) observe the relative frequency with

which the green light occurs given each of the above re-

sponses, (3) transform these observations into conditional

probability estimates, and (4) estimate the differential

probability of green light onset given either of the two

responses. Thus, subjects are given repeated opportuni-

ties to examine a single response-outcome relationship

while varying their behavioral response. Subjects have

only to appropriately sample the responses available to

them and to correctly process the available information in

order to discover the objective degree of contingency pres-

ent in the task.


The discussion of this second issue will ignore the con-
tingency-control confound discussed above in regard to
this research. Thus, it will be assumed, for purposes of
this discussion, that these prior studies have actually
examined the contingency judgement process.









The type of contingency judgement just described may

be contrasted with the type of contingency judgements sub-

jects are likely to encounter in their everyday lives.

These judgements are much more likely to concern ambiguous,

single-occurrence events where the objective degree of con-

tingency cannot be definitively determined through repeated

experimentation. For example, if a man is fired from his

job, or if he is involved in a serious automobile accident,

he does not have the opportunity to repeat the event mul-

tiple times, while varying his response, in order to de-

termine the degree of contingency involved. Nevertheless,

it appears that subjects do make contingency judgements

on the basis of such limited information, and it appears

that any differences between the contingency perception

tendencies of depressed and nondepressed subjects would

be most likely to be evidenced under such ambiguous condi-

tions.

The significance of the distinction made above be-

tween the two types of contingency judgements is made

clear when it is considered that subject differences on

each type of judgement may reflect different psychological

processes. Subject differences on the type of contingency

judgement examined by Alloy and Abramson (1979) would ap-

pear primarily indicative of differences in subjects' ca-

pacity to accurately process the information gathered dur-

ing their repeated examination of the response-outcome










relationship-to a considerably smaller degree it would

seem that these variations might also reflect underlying

biases in subjects' contingency perception tendencies. In

contrast, subjects' judgements of contingency for ambiguous

events would appear to be highly indicative of any such

underlying biases that subjects might have either toward

or against the perception of contingency between their re-

sponses and subsequent outcomes.

The distinctions made between these two types of con-

tingency judgements may also be seen to correspond, roughly,

to different stages in the contingency perception process

as presented by Crocker (1981). Judgements of ambiguous,

single-occurrence events may be seen to correspond to

stage three of her model, in which subjects classify spe-

cific instances as being either indicative or contraindic-

ative of a contingent relationship between a particular

response and outcome. The type of judgement assessed by

Alloy and Abramson's (1979) procedure appears to be con-

cerned exclusively with processes corresponding with Croc-

ker's stages four and five, dealing only with the process-

ing of contingency-judgement information and the derivation

of global contingency judgements on the basis of this in-

formation. Additionally, the information that is involved

here consists of the relatively clear-cut information pro-

vided by repeated exposures to a single response-outcome re-

lationship, and not that provided by subjects' own decisions










regarding the degree of contingency present in more ambig-

uous situations. Thus, it would appear that subjects'

judgements in the Alloy and Abramson procedure would not

reflect any biases subjects might potentially evidence at

stage three of Crocker's model, but be reflective only of

any biases they might display at stages four and five.

The above distinctions suggest that while Alloy and

Abramson's (1979) findings may provide considerable infor-

mation regarding the relative accuracy with which subjects

process information pertaining to contingency judgements,

these findings may actually provide relatively little in-

formation regarding subjects' basic tendencies to perceive

contingency between their responses and subsequent outcomes

under more ambiguous conditions. Thus, their major finding

may need to be reinterpreted to suggest that depressed sub-

jects are relatively accurate in their processing of con-

tingency judgement information, whereas nondepressed sub-

jects are prone to self-enhancing and self-protective biases

in their processing of such information. It remains possi-

ble, however, that depressed subjects might still be either

more or less prone to perceive contingency when actually

faced with ambiguous, single-occurrence events than are

nondepressed subjects.

The only study that has attempted to assess subjects'

contingency perception tendencies for ambiguous, single-

occurrence events is that of Miller (Note 1). It may









initially appear that Miller's procedure does not examine

subjects' contingency perceptions for single-occurrence

events, since it involves repeated trials as does the Alloy

and Abramson (1979) procedure. It must be recognized, how-

ever, that the repeated trials in Miller's procedure re-

quire that subjects make a series of contingency judgements

for a number of distinct response-outcome relationships,

since each trial involves different word association items.

Thus, while Miller's procedure does not assess subjects'

contingency judgements on a trial-by-trial basis, it may

be seen that subjects' contingency judgements in this pro-

cedure are reflective of such judgements, since they are

derived from their experiences with a series of ambiguous,

single-occurrence events. These types of judgements are

not assessed at all in the other contingency perception

studies.

In summary, almost all of the prior research regarding

subjects' contingency perception tendencies has been com-

plicated by serious conceptual and methodological diffi-

culties which make clear interpretation of their results

impossible. In particular, it appears that these studies

have confounded the concepts of contingency and control

in their procedures, making it difficult to determine if

their findings relate to subjects' contingency perception

tendencies or to their perceptions of their efficacy in

identifying and executing appropriate controlling responses.









It also appears that these studies have focused exclusively

upon a rather limited aspect of the contingency perception

process (involving the processing of clear-cut contingency-

judgement information), and ignored other, potentially more

significant, aspects of this process (e.g., subjects' judge-

ments of the degree of contingency present in ambiguous,

single-occurrence events).



Introduction to the Current Research Effort



The current research effort attempts to correct some

of the conceptual and procedural shortcomings noted in

the prior research in order to clarify the manner in which

depressed and nondepressed subjects perceive contingency

between their responses and subsequent outcomes. Specif-

ically, a new methodology for assessing subjects' contin-

gency perception tendencies is developed and applied in

the current research effort. This new methodology seeks to

eliminate the confound noted in prior studies between the

concepts of contingency and control, and to allow for ex-

amination, not only of subjects' processing of contingency

judgement information, but also of their tendencies to

classify ambiguous, single-occurrence outcomes as being

either contingent or noncontingent.

During the development of the current methodology,

efforts were made to simply modify Miller's (Note 1)









procedure, since his methodology was seen to involve sub-

jects' contingency perceptions for ambiguous, single-

occurrence events, and not to overtly confound subjects'

contingency and control judgements. Specifically, condi-

tions were added to Miller's procedure to allow for ex-

amination of subjects' contingency judgements for both

positive and negative events, and to associate the per-

ception of noncontingency with continuing longer on the

task. However, pilot testing showed these procedures to

be excessively difficult for subjects to understand, re-

sulting in extreme variability in their responses to the

task. Additionally, it was noted that the nature of the

stop-score procedure made it difficult to make comparisons

across subjects (since they could receive different ex-

periences on the task), and that it created difficulties

in data analysis by requiring the use on nonparametric

statistics.

In view of the difficulties encountered in the effort

to simply modify Miller's (Note 1) procedure, a new method-

ology for assessing subjects' contingency perception ten-

dencies was developed for the current research. Miller's

ambiguous word association task is also employed in this

methodology, but his stop-score dependent measure has

been dropped in favor of a more straightforward rating

scale format (similar to the assessment methods employed

by Alloy & Abramson, 1979, and Jenkins & Ward, 1965, but










with the confounds with the concept of control removed).

Specifically, in the current procedure subjects are faced

with the ambiguous word association task with the instruc-

tions that they will receive, in varying proportions,

three different types of feedback: contingent feedback,

noncontingent positive feedback, and noncontingent nega-

tive feedback. They are then exposed to varying amounts

of noncontingent positive and negative feedback on this

task and subsequently asked to rate the degree to which

they felt the outcomes they received actually corresponded

with the accuracy of their responses.

Several different contingency perception measures

are employed in the current research: These are considered

to evaluate subjects' contingency judgements at different

points in the contingency perception process (e.g., one

measure is considered to be highly reflective of subjects'

trial-by-trial judgements, whereas another is considered

to be more reflective of their final, integrated contin-

gency judgement following full cognitive processing of

all contingency-judgement information). It was hoped that

variations in subjects' contingency judgements among these

measures would help to differentiate any biases they might

display in their contingency judgements for ambiguous,

single-occurrence events versus biases in their process-

ing of contingency judgement information.










Several additional performance-related measures are

also included in the study. Separate scales were devel-

oped in order to assess subjects' judgements regarding

their ability to perform the word association task and

their capacity to exercise control over the occurrence of

the outcomes in the tasks. These measures were included

in order to further differentiate subjects' contingency

perceptions from their judgements of their efficacy in

performing appropriate controlling responses, and from

their overall judgements regarding their capacity to con-

trol outcomes. Another measure is included in the current

study in order to check for distortions in subjects' re-

call of the outcomes they received during the tasks. A

final measure is included to contrast subjects' approach

to the tasks (i.e., the number of times that they tried

to give correct versus incorrect answers during the task).

It is extremely difficult to establish specific hy-

potheses for the current research: The serious conceptual

and methodological difficulties, discussed above, in regard

to the prior research in this area result in there being

essentially no clear-cut findings upon which to base the

development of such hypotheses. Thus, no specific set

of hypotheses will be presented for the current research

effort. However, several possible predictions for this

research may be briefly considered. It is important to

recognize that these predictions must be developed in









relative terms, since the procedure concerns subjects'

judgements for amgiguous response-outcome relationships

for which there are no "accurate" judgements. Thus, it is

not possible to predict that either group will be more ac-

curate than the other in their judgements, but only that

one group will tend to perceive either greater or smaller

amounts of contingency relative to the other under partic-

ular reinforcement conditions.

One possible prediction for the current research

could be based upon the possibility that the conclusions

of the prior research in this area have been accurate in

spite of the conceptual and methodological difficulties

discussed above-i.e., that depressed subjects may be rel-

atively accurate in their contingency judgements, whereas

nondepressed subjects evidence self-serving biases. Al-

though it is not possible to directly address the accuracy

of subjects' judgements in the current research, such a

prediction would be supported if it were found that de-

pressed subjects' judgements varied only minimally with

changes in reinforcement frequency, whereas those of non-

depressives indicated greater amounts of contingency under

high-frequency positive feedback conditions and smaller

amounts of contingency under high-frequency negative feed-

back situations.

Alternately, it might be suggested that the predic-

tions of the LH model would be supported once the confound










between the contingency and control concepts was removed

and/or once subjects' contingency judgements for ambigu-

ous, single-occurrence events were assessed (as opposed to

the assessment of their capacity to process clear-cut con-

tingency judgement information). The predictions of the

LH model would be supported by the finding that depressed

subjects underestimated contingency, relative to the non-

depressed subjects, in all, or nearly all, reinforcement

conditions.

Finally, a number of other specific predictions might

be made for the current research. For example, it might

be predicted that depressed subjects would be more prone

to perceive contingency between their responses and sub-

sequent outcomes, not less prone as predicted by the LH

model. Such a prediction would be consistent with more

traditional self-blame models of depression (e.g., Beck,

1967), and with attributional findings indicating that de-

pressed subjects assume more responsibility for both pos-

itive and negative outcomes than do nondepressed subjects

(e.g., Kuiper, 1978; Rizley, 1978). It might also be pre-

dicted that depressed subjects might display a "self-

deprecating bias" (in contrast to nondepressed subjects'

self-serving bias), tending to perceive less contingency

for positive outcomes than for negative outcomes. Final-

ly, it might even be predicted that depressed and nonde-

pressed subjects will show no differences at all in their









contingency perception tendencies, but, possibly, that they

would show significant differences in their judgements of

their ability on the word association task and/or their

judgements of their capacity to control the outcomes they

received during these tasks. Such a prediction would be

consistent with the suggestions, made above, that the prior

research studies in this area may have been more concerned

with differences in subjects' efficacy and control judge-

ments rather than with differences in their contingency

perception tendencies.
















CHAPTER TWO
METHOD


Subjects


Source of Subjects

All subjects employed in the current research were

undergraduate psychology students at the University of

Florida. Subjects participated in the study in order to

partially fulfill psychology course requirements, and

they received no additional compensation for their partic-

ipation. The final sample of 80 subjects employed in the

study was drawn from a pool of 604 students who had pre-

viously been administered a depression inventory (described

below) for screening purposes. This pool of subjects was

comprised of two samples: (1) a primary sample, containing

396 subjects, which received the depression inventory dur-

ing the first week of the academic semester (approximately

2 months prior to follow-up testing), and (2) a secondary

sample, containing 208 subjects, which received the de-

pression inventory about 2 weeks prior to follow-up test-

ing.

The objections of Depue and Monroe (1978) to the use

of nonclinically depressed populations in examinations of










the LH phenomenon were considered. However, it appears,

as Seligman (19'8) has argued, that the use of nonclinical-

ly depressed college students as subjects in LH experiments

may be appropriate when it is recognized that the conclu-

sions from such research may be applicable only to mildly

depressed populations. It also appears that college stu-

dents may appropriately be used, as in the current research,

to screen new research conceptualizations thought to be

applicable to more severely depressed populations. It

must be recognized, however, that the conclusions from

such research may not be assumed to be applicable to clin-

ical populations without appropriate follow-up studies.



Subject Selection Procedure

Subjects meeting the preliminary criterion for admis-

sion to either the depressed or nondepressed groups on the

basis of their score on the screening instrument (admission

criteria are discussed below) were contacted and invited

to participate in a second part of the study for additional

class credit. Priority in scheduling went to subjects in

the primary sample in order to maximize the selection of

non-transiently depressed and nondepressed subjects. Sub-

jects who agreed to participate in the second part of the

study were seen individually by the experimenter, at which

time they were administered a second depression inventory

and then the experimental procedures. Only those subjects









who also satisfied a secondary admission criterion on this

second depression measure were retained in the final sample.

Testing was continued until each cell of a 2 (mood) X 2

(sex) X 2 (task order) design was filled with 10 subjects.

Thus, the final 80-subject sample was comprised of a 40-

subject depressed group and a 40-subject nondepressed group,

with each group being evenly balanced as to sex and task

order.


Subject Selection Instruments Employed


Beck Depression Inventory

The long form of the Beck Depression Inventory (BDI)

(Beck, 1967) was employed in this study as the primary

screening device for identifying depressed and nondepressed

subjects. Although Depue and Monroe (1978) have questioned

the appropriateness of this instrument for the assessment

of mild depression, Seligman (1978) has defended it use

in LH studies and has reported a number of research find-

ings supporting its use for such purposes. Additionally,

Bumberry, Oliver, and McClure (1978) have reported high

correlations in a college population between BDI scores

and clinical ratings of depression based upon psychiatric

interviews. The decision to employ the BDI as the primary

measure of depression in this study was based upon this

evidence. It was also noted that the BDI has been, by
far, the most frequently used measure of depression in









prior LH research. Thus, use of the BDI was considered to

enhance the comparability of the current findings with

prior research in this area.

Full BDI scores were calculated for all subjects in

accordance with the instructions of Beck (1967). Scores

on the scale range from 0 to 63, with higher scores indi-

cating greater severity of depression. The criterion for

admission to the depressed group was set at BDI 10. This

score is considered by Kovacs and Beck (1977) to represent

the low end of the mild range of depression, and it has

been recommended as the appropriate cut-off score for de-

tecting depression in normal and medical populations

(Schwab, Bialow, Clemmons, Martin, & Holzer, 1967). The

criterion for admission to the nondepressed group was set

at BDI 2.



Self-rating Depression Scale

The Self-rating Depression Scale (SDS) (Zung, 1965 &

1973) was employed in this study as a secondary measure of

depression. This instrument is comparable to the BDI in

that it is a brief, self-administered scale which samples

a wide range of symptom categories. This instrument was

chosen for use in this study in favor of the Multiple

Affect Adjective Checklist (MAACL) (Zuckerman & Lubin,

1965) which has been more commonly employed as a secondary

depression measure in prior LH research. The SDS was










chosen for use in the current research because it was seen

to provide a broader assessment of the depressive syndrome

than does the MAACL, which is designed to assess only tran-

sient depressive affect.

Scores on the SDS were calculated in accordance with

the instructions provided by Zung (1965 & 1973). Scores

on the scale range from 25 to 100, with higher scores in-

dicating greater severity of depression. Prior normative

studies (e.g., Zung, 1965, 1971, & 1972) have established

relatively good norms for subjects aged 20 to 50. However,

younger and older subjects have tended to obtain higher

scores than those within this range. Since subjects used

in the current research fell right at the borderline age

for which good normative data are available, it was subse-

quently decided to administer the SDS to a large sample

of college students in order to establish appropriate cut-

off scores for the current research. It was determined in

advance that appropriate cut-off points would be set at

the 30th percentile for the nondepressed group and at the

70th percentile for the depressed group. Following admin-

istration of the SDS to 208 students, the following were

established as the secondary admission criteria: SDS 5 38

for the nondepressed group, and SDS 48 for the depressed
1
group.


The SDS was administered along with the BDI to the 208
subjects comprising the secondary sample of the subject
pool used to select subjects for the current research.
(Subjects who were drawn from this sample to participate










Experimental Design


Overview of Procedure

Following completion of informed consent forms and

the SDS, subjects were administered three versions of a

perception of contingency task (PCT), completing a follow-

up questionnaire after each version of the PCT. The three

versions of the PCT were presented in either of two orders,

with half of the subjects in each group receiving them in

one order and the other half in the second order.



Perception of Contingency Task

Description of task. The PCT was adapted from Miller

(Note 1) who developed a perception of noncontingency task

using a word association procedure previously employed by

Wener and Rehm (1975) and Kuiper (1978). This procedure

involved a list of 100 stimulus words and their two most

popular associates. These words had been selected on the

basis of word association norms (Bilodeau & Howell, 1965;

Palermo & Jenkins, 1963; and Rapaport, Gill, & Schaefer,



in the experimental procedures were selected on the basis
of their BDI scores only-they were also required to satis-
fy the secondary admission criterion on a readministration
of the SDS in order to remain in the final sample.) The
cut-off score of SDS < 38, which was identified as the
30th percentile, actually included the lower 32.21% of the
sample due to the number of tie scores at SDS = 38. The
cut-off score of SDS > 48, which was identified as the
70th percentile, actually included the upper 31.25% of the
sample due to tie scores at SDS = 48.









1946) such that each stimulus word's two most popular asso-

ciates were known to have approximately equal probabilities

of selection. These word association items were employed

in the current research to create the three versions of the

PCT, with each version containing 32 of these items.

For each task, subjects were told that one of the

word association choices in each trial had been determined

to be the most popular associate to the stimulus word on

the basis of prior research, and that one part of their

task would be to decide which of the two choices they

thought was this most popular response. Following this

initial introduction to the task, subjects were given in-

formation regarding the feedback they would receive during

the task. The experimenter explained and demonstrated to

the subjects that the experimental apparatus could be set

in an "accurate feedback mode," such that pressing one of

the two buttons on their control panel (ostensibly the one

corresponding to the most popular response) would result

in the sounding of a bell, whereas pressing the other but-

ton (ostensibly the one corresponding to the less popular

response) would result in the sounding of a buzzer. The

experimenter then explained to the subjects that the ex-

perimental apparatus could also be set in two other feed-

back modes. He explained and demonstrated that pressing

either button would result in the sounding of the bell

when the device was set in the "positive feedback mode,"










whereas pressing either button would sound the buzzer when

the device was set in the "negative feedback mode." The

experimenter emphasized in his instructions that the accu-

racy of subjects' responses would determine the outcomes

they received only when the device was set in the accurate

feedback mode, whereas the accuracy of their responses

would make no difference when the device was set in either

the positive or negative feedback modes.

Subjects were told that their primary task in the ex-

periment was to discover how related the sounding of the

bell and buzzer were to the accuracy of their responses.

They were told that in order to do this they should con-

sider the feedback they received on each trial and try to

come to some judgement as to what mode the device was set

in. Subjects were advised that they should not necessarily

always give what they thought was the correct answer,

since it is as important to know how accurate the feedback

is when one gives a wrong answer as when one gives a cor-

rect one. Subjects were specifically advised that they

would receive a total of 32 trials in each task in order

to facilitate any effort they might make to equalize the

number of correct and incorrect attempts that they made.

The full instructions which were read to subjects by the

experimenter at the outset of the experimental task are

presented in Appendix A. The experimenter deviated slight-

ly from these instructions when necessary to assure subject

understanding of the task.









Apparatus and procedure. A visual display screen was

employed in the PCT, upon which slides containing the stim-

ulus words and their associates were displayed using a

standard slide projector. Below the display screen was

the subject's control panel, consisting of two buttons

labeled "1" and "2." The experimenter operated a master

control panel which was designed such that he could deter-

mine, on a trial-by-trial basis, whether the subject's next

response triggered the sounding of the bell or the buzzer.

The master control panel contained a 4-way toggle switch

which was labeled so as to encourage subjects' acceptance

of the experimental instructions. Two of the positions of

the switch were labeled "1" and "2," with the label "Accu-

rate Feedback Mode" positioned above them. The other two

positions of the switch were labeled "Positive Feedback

Mode" and "Negative Feedback Mode."

The experimenter sat to the side of, and slightly

behind subjects during administration of the PCT, such that

he could observe subjects' responses to the slides, but so

that subjects were not able to observe his manipulations

of the master control panel. After each response, the ex-

perimenter recorded the subject's response on a record form

and then preset the next feedback before advancing to the

next slide. This procedure provided a relatively uniform

4-5 second inter-trial interval during which the subject

could evaluate the accuracy of the feedback he received.










Feedback was administered noncontingently for all

trials of each task in a predetermined sequence. In Task

A, subjects received 16 positive outcomes (bell) and 16

negative outcomes (buzzer); in Task B, they received 24

positive outcomes and 8 negative outcomes; and in Task C

they received 8 positive outcomes and 24 negative outcomes.

The stimulus words, their equal probability associates,

and the noncontingent feedback schedules for each task

are presented in Appendix B.

All subjects were initially administered Task A of

the PCT. The order of the remaining two tasks was counter-

balanced such that half the subjects in each group complet-

ed Task B and then Task C (Order 1), whereas the remaining

subjects completed Task C and then Task B (Order 2). Full

counterbalancing of task order was considered unfeasible

because of the sample size limitations in the study. The

decision to have all subjects start with Task A was based

upon the assumption that this task would provide the most

effectively neutral experience for subjects, and that it

would, therefore, be least likely to result in any compli-

cating carry-over effects.



Follow-up Questionnaire.

General description. Subjects were administered an

identical follow-up questionnaire following each version

of the PCT. The full versions of these questions (See









Appendix C) were read to subjects by the experimenter fol-

lowing Task A; subjects completed these questionnaires on

their own following Tasks B and C, although they were pro-

vided with a written copy of the full versions of the ques-

tions and allowed to consult the experimenter if they did

not understand a question. Subjects recorded their own

responses to the follow-up questionnaire using a follow-up

questionnaire answer sheet (See Appendix D).

Eight of the 12 items on each questionnaire required

subjects to make a percentage estimate by marking at some

point along a 5 inch line-anchor points were provided at

the 0%, 50%, and 100% positions. These scales were scored

by first measuring the distance (in inches) from the 0%

position to the point marked by the subject. This dis-

tance was then converted into a percentage score by di-

viding it by 5 and then multiplying by 100.

The remaining four items of each questionnaire re-

quired subjects to estimate the number of trials on which

a particular event occurred. These four items consisted

of two sets of paired questions such that subjects' re-

sponses to the two parts of each pair should sum to equal

the total number of trials (32) in each task. The ques-

tions comprising the three follow-up questionnaires con-

tained all of the dependent measures for the experiment

as described below.










Description of dependent measures. The first six

items of each follow-up questionnaire were used to assess

subjects' contingency perceptions on that task. Question

#1 asked subjects to estimate the percentage of trials the

device was set in each of the three feedback modes (sub-

jects were specifically instructed that their three esti-

mates should total as close as possible to 100%). Sub-

jects' responses to the three parts of this question were

considered to reflect their tendencies to classify ambig-

uous, single-occurrence outcomes in each of the following

ways: (1) as contingent positive or negative events

(accurate feedback mode), (2) as noncontingent positive

events (positive feedback mode), or (3) as noncontingent

negative events (negative feedback mode). This measure

was considered to provide the most direct indication of

any biases subjects might evidence in their classification

of such ambiguous instances, and to be least reflective

of any biases subjects might evidence in the processing

of such information. In effect, this measure asked sub-

jects to simply recall their trial-by-trial classifica-

tions and to transform these into percentage estimates.

Each of the three percentages obtained in Question

#1 was considered to provide a measure of subjects' ten-

dencies to classify and recall events as being contin-

gently related to their responding. The accurate feedback

percent estimate was considered to provide the broadest










assessments of subjects' tendencies to classify and recall

outcomes as being contingently related to their responding:

the higher the percentage, the greater the tendency to

classify and recall events as being contingently determined.

The positive feedback percent and negative feedback percent

measures were considered to provide more specific assess-

ments of subjects' tendencies to classify and recall pos-

itive and negative outcomes, respectively, as being con-

tingently related to their responding: the lower the per-

centage, the greater the tendency to classify and recall

that category of outcomes as being contingently determined.

Question #2 asked subjects to estimate the frequency

with which they thought the desired outcome (bell) would

occur given either the presence or absence of a correct

response (subjects were specifically instructed that their

responses to the two parts of this question did not have

to total to 100%). Subjects' responses to part one of

this question were considered to reflect their subjective

estimates of the probability that the desired outcome

would occur given a correct response (p(0/R)); their re-

sponses to part two of this question were considered to

reflect their subjective estimates of the probability that

the desired outcome would occur given the absence of a

correct response (p(0/R)). It was assumed that greater

tendencies toward the perception of contingency would be

indicated by higher estimates to part one of the measure









(probability of the bell given a correct response) and low-

er estimates to part two (probability of the bell given an

incorrect response). A broader measure of subjects' con-

tingency perception tendencies was developed from these

estimates by calculating the absolute value of the differ-

ence between them: Greater differences between the two

estimates were seen to reflect greater tendencies toward

the perception of contingent relationships. This measure

may be seen to parallel the theoretical definition of con-

tingency provided earlier in this paper (i.e., the arith-

metic difference between p(0/R) and p(0/R)). The three

contingency measures derived from Question #2 were consid-

ered to reflect an intermediate stage of the contingency

judgement process: It was assumed that these estimates

reflected greater amounts of cognitive processing of sub-

jects' trial-by-trial judgements than did the feedback

percent estimates, but less than was involved in their

global contingency ratings (discussed below).

Question #3 provided the final measure of subjects'

contingency perception tendencies. This measure consisted

of a single question asking subjects to rate the overall

degree of correspondence between the sounding of the bell

and buzzer and the accuracy of their responses. This mea-

sure was assumed to provide the highest level assessment

of subjects' contingency perception tendencies, reflecting

the greatest amount of additional cognitive processing.









The remaining six items of the follow-up questionnaire

were used to form four performance-related dependent mea-

sures. Question #4 asked subjects to assess their efficacy

in performing the word association task. This measure was

included in the study in order to further differentiate

subjects' contingency perception tendencies and their judge-

ments regarding their efficacy in performing the appropri-

ate controlling response. It was subsequently recognized,

however, that the relationship between subjects' estimates

of their ability and their perceptions of contingency on

the tasks might be more direct. In particular, it was con-

sidered possible that highly confident subjects might be

more prone to perceive a lack of contingency on these tasks

than would less confident subjects. For example, a highly

confident subject who receives feedback which is incongru-

ent with his expectations might be expected to view this

feedback as noncontingent, whereas a less confident subject

who receives the same amount of incongruent feedback might

simply conclude that his expectations were incorrect. If

this type of relationship were to exist, then it would be

expected that subjects' judgements of contingency would

correlate negatively with their ability estimates on each

of the three tasks.

It was also recognized that the direction of any di-

rect causal relationship between subjects' ability esti-

mates and their perceptions of contingency might be reversed










from that described above. For example, the subject who

perceives that the positive feedback he is receiving is

contingent upon his responses may be influenced by this

feedback to make higher estimates of his ability than the

subject who concludes that the positive feedback is noncon-

tingent. If this type of relationship were to exist, it

would be predicted that subjects' estimates of their abil-

ity would be positively correlated on Task B, negatively

correlated on Task C, and uncorrelated or weakly correlat-

ed on Task A. Although the current procedure was not de-

signed to fully examine the nature of this relationship,

it was felt that some assessment of this relationship was

needed in order to rule out the possibility that subjects'

contingency estimates were mere reflections of their vary-

ing levels of confidence in their ability to perform the

word association task, and not true reflections of their

contingency perception tendencies.

Question #5 asked subjects to rate the degree of con-

trol they felt they had over the outcomes they received

in the various tasks. This measure was included to further

differentiate subjects' perceptions of contingency from

their judgements of control.

Question #6 asked subjects to simply recall the number

of positive and negative outcomes they received during

each task. Subjects were reminded, in conjunction with

this question, that they had completed a total of 32 trials









during each task-this was done in order to eliminate the

variability associated with some subjects remembering this

fact from prior instructions and some subjects assuming

that they had received either a larger or smaller number

of total trials. This item was included in order to check

for distortions in subjects' recall of the feedback they

received during the tasks.

Question #7 asked subjects to recall the number of

times during the task that they attempted to give correct

versus incorrect responses. Again, subjects were reminded,

in conjunction with this question, that they had completed

a total of 32 trials. This question was included in order

to contrast subject's approach to the task. Variations

here were assumed to correspond to sampling variations of

the sort referred to in step 1 of Crocker's (1981) model

of the contingency perception process. It was also recog-

nized, however, that variations in subjects' approach to

the task could directly influence their contingency judge-

ments by altering the amount of confirming versus discon-

firming feedback that they received. For example, a sub-

ject who always tried to give the correct response while

receiving the noncontingent negative feedback of Task C

would certainly be expected to give lower contingency

judgements than the subject who always tried to give incor-

rect responses on this task.

If variations in the number of positive versus negative










efforts made by subjects were to influence their contingen-

cy judgements in the manner described above, then it would

be expected that there would be a positive correlation be-

tween the number of positive efforts and judgements of

contingency on Task B, since increasing numbers of posi-

tive efforts should increase the amount of confirming feed-

back that is received (i.e., feedback that is consistent

with subjects' expectations, assuming that a contingent

relationship exists). On Task C, it would be expected

that such a relationship would result in negative corre-

lations between subjects' contingency judgements and the

number of positive efforts, since this approach would re-

sult in greater amounts of disconfirming feedback on this

task. Finally, no significant correlation would be expect-

ed on Task A, given such a relationship, since variations

in subjects' approach to this task should not affect the

amount of confirming versus disconfirming feedback received

(the amount of confirming versus disconfirming feedback

should be approximately equal regardless of subjects' ap-

proach). Thus, this measure was included to assess any

sampling biases subjects might evidence on these tasks, as

well as to examine the impact these differences might have

on subjects' contingency judgements. Again, the primary

concern is to assure that any observed differences on the

contingency perception measures are truly reflective of

contingency perception tendencies and not other factors,

such as subjects' approach to the task.










Overview of Statistical Procedures


All dependent variables were analyzed using a 2 (group)

X 2 (sex) X 2 (task order) X 3 (task) repeated measures

analysis of variance procedure. Follow-up analyses were

conducted using the least significant difference (LSD) pro-

cedure, for paired comparisons, and Duncan's new multiple

range test, for multiple comparisons. A small number of

additional group comparisons were conducted using Student's

t-test procedure. The interrelationships among selected

dependent variables were analyzed using Pearson's product-

moment correlation procedure. All tests of significance

were two-tailed and set at the .05 level of significance.

















CHAPTER THREE
RESULTS


Description of the Final Sample


Identification of the the 40 depressed subjects em-

ployed in the final sample required the testing of 58 sub-

jects who met the preliminary criterion for admission to

the depressed group. Sixteen to the 58 subjects were

dropped from the final sample for failure to satisfy the

secondary criterion for admission to the depressed group.

Two additional subjects were dropped from this group be-

cause of improper marking of responses on the follow-up
1
questionnaires.

Identification of the 40 nondepressed subjects em-

ployed in the final sample required the testing of 50 sub-

jects who met the preliminary criterion for admission to

the nondepressed group. Nine of the 50 subjects were

dropped from the final sample for failure to satisfy the

secondary criterion .for admission to the nondepressed

group. One additional subject was dropped from this group


These two subjects marked only the anchor points on all
three of the questionnaires, seemingly not recognizing
that they could mark intermediate points. The decision
to drop these subjects was made before their group mem-
bership was known.









because he found it extremely difficult to understand the

experimental instructions.'

The final 80-subject sample was drawn largely from

the primary sample of subjects that had received the BDI

during the first week of the academic semester. Specifi-

cally, 82.5% of the final sample was drawn from this group,

indicating that 66 of the 80 subjects satisfied the two

admission criteria across a time span of approximately two

months. The remaining 14 subjects were drawn from the

secondary sample of subjects where the time span between

their satisfaction of the two admission criteria was about

two weeks. The depressed group was drawn somewhat more

heavily from the secondary sample than was the nondepressed

group (22.5% for the depressed group, versus 12.5% for the

nondepressed group) due to the fact that a greater number

of subjects had to be tested in order to fill the depressed

group.

Depression inventory means for the final sample are

presented in Table 1 by group, group X sex, and group X

order. These figures reveal that the selection procedures

resulted in a marked separation between the depressed and

nondepressed groups. Analysis of the sub-group means us-

ing the t-test procedure revealed that the depressed

This subject was a foreign student whose understanding of
the English language was rather limited. Again, the deci-
sion to drop this subject from the final sample was made
before his group membership was known.










Table 1. Final sample depression inventory means and
standard deviations by group, group X sex, and group X
order.


BDI SDS
Grouping n Mean SD Mean SD

Depressed 40 14.925 4.725 55.125 6.231

Nondepressed 40 0.250 0.439 32.425 3.544



Depressed

Males 20 13.450 3.000 53.400 6.047

Females 20 16.400 5.679 56.850 6.072

Nondepressed

Males 20 0.200 0.410 33.100 3.478

Females 20 0.300 0.470 31.750 3.567



Depressed

Order 1 20 14.950 4.006 57.500 7.164

Order 2 20 14.900 5.457 52.750 4.064

Nondepressed

Order 1 20 0.300 0.470 32.800 3.592

Order 2 20 0.200 0.410 32.050 3.546









females in the sample obtained significantly higher BDI

scores than did depressed males (t (38) = 2.05, p .05),

and that the depressed females tended to obtain higher

scores than depressed males on the SDS (t (38) = 1.80,

p < .08). It was also discovered that depressed/Order 1

subjects obtained significantly higher SDS scores than did

depressed/Order 2 subjects (t (38) = 2.58, < .02).


Perception of Contingency Measures


Percentage of Trials in Each Feedback Mode

Subjects' raw score estimates of the percentage of

trials the device was set in each of the three feedback

modes were transformed prior to statistical analysis so

that the sum of each subject's three scores actually to-

taled 100%. This was accomplished by dividing each score

by the sum of the three scores and then multiplying by 100.

This transformation was undertaken in order to assure that

all subjects' scores had an equal influence in the deter-

mination of group means. The results of this transforma-

tion are reflected in Table 2 which presents a comparison

of mean feedback percentage estimates for transformed and

nontransformed data by group. These figures reveal that

this transformation resulted in only minimal changes in

the data.

Accurate feedback percent. Analysis of subjects'

transformed estimates of the percentage of trials the










Table 2. Comparison of mean feedback percentage estimates
for transformed and nontransformed data by group.


Accurate Positive Negative
Feedback Feedback Feedback
Group Percent Percent Percent Total

Depressed

Nontransformed 43.817 25.117 30.008 98.942

Transformed 44.585 25.332 30.083 100.000



Nondepressed

Nontransformed 34.483 29.500 34.008 97.991

Transformed 35.621 29.876 34.503 100.000










device was set in the accurate feedback mode (signifying

the perception of contingent feedback) yielded significant

main effects for group and task, and a marginally signif-

icant main effect for sex. Means and significance values

for these effects are presented in Table 3 (complete re-

sults of this analysis are presented in Table E-1 of Appen-

dix E).

The significant main effect for group indicates that

depressed subjects recalled classifying significantly more

of the outcomes they received as contingent events than

did the nondepressed subjects (p < .006). Further anal-

ysis of the significant main effect for task, using post-

hoc Duncan's comparisons, revealed that subjects recalled

classifying significantly fewer outcomes as contingent

events in Task C than in either Task A or B (P < .001);

their estimates for Tasks A and B did not significantly

differ. Examination of the marginally significant main

effect for sex indicates that males tended to perceive

greater amounts of contingent feedback than did females

( < .06).

Positive feedback percent. Analysis of subjects'

transformed estimates of the percentage of trials the de-

vice was set in the positive feedback mode (signifying

the perception of noncontingent positive feedback) yielded

significant main effects for group and task. Means and

significance values for these effects are presented in










Table 3. Means and significance values for significant
and marginally significant effects identified by analysis
of subjects' transformed accurate feedback percent estimates.


Effect Mean F df p

Group

Depressed 44.585
8.30 1,72 .005
Nondepressed 35.621



Sex

Males 43.111
3.74 1,72 .057
Females 37.095



Task

Task A 48.720

Task B 44.952 36.53 2,144 .001

Task C 26.637









Table 4 (complete results of this analysis are presented

in Table E-2 of Appendix E).

The significant main effect for group indicates that

depressed subjects recalled classifying significantly few-

er of the outcomes they received as noncontingent positive

events than did nondepressed subjects (p < .04). Further

analysis of the significant main effect for task, using

post-hoc Duncan's comparisons, revealed significant dif-

ferences between subjects' positive feedback estimates

for each of the three tasks (p < .05): Subjects reported

perceiving the greatest amount of noncontingent positive

feedback on Task B, an intermediate amount on Task A, and

the least amount on Task C.

Negative feedback percent. Analysis of subjects'

transformed estimates of the percentage of trials the de-

vice was set in the negative feedback mode (signifying the

perception of noncontingent negative feedback) yielded

significant main effects for group, sex, and task, and a

significant sex X task interaction. Means and signifi-

cance values for these effects are presented in Table 5

(complete results of this analysis are presented in Table

E-3 of Appendix E).

The significant main effect for group indicates that

depressed subjects recalled classifying significantly fewer

of the outcomes they received as noncontingent negative

events than did nondepressed subjects (2 < .04). The










Table 4. Means and significance values for significant
effects identified by analysis of subjects' transformed
positive feedback percent estimates.


Effect Mean F df

Group

Depressed 25.332
4.71 1,72 .033
Nondepressed 29.876



Task

Task A 23.642

Task B 41.281 44.41 2,144 .001

Task C 17.890










Table 5. Means and significance values for significant
effects identified by analysis of subjects' transformed
negative feedback percent estimates.


Effect Mean F df P


Group

Depressed

Nondepressed


Sex

Males

Females



Task

Task A


Task B


30.083

34.503


29.986

34.600


5.02


1,72


.035


1,72


.028


27.638


13.768 202.03


Task C



Sex X Task

Males

Task A

Task B

Task C


Females

Task A

Task B


Task C


2,144


.001


55.473


24.752

14.695

50.511


4.00


2,144


.021


30.525

12.841

60.434










significant main effects for sex and task must be inter-

preted in light of the significant sex X task interaction.

Further analysis of the significant sex X task inter-

action, using post-hoc LSD comparisons, revealed that fe-

males recalled classifying significantly more outcomes as

noncontingent negative events than did males on Task C

(P < .05), and that they tended in this direction on Task

A (2 < .10). In contrast, males reported a slightly higher

percentage of noncontingent feedback than did females on

Task B, although this difference did not approach signif-

icance. The significant sex X task interaction is depicted

graphically in Figure 1.

Post-hoc Duncan's comparisons contrasting the amount

of noncontingent negative feedback perceived on each task

yielded identical results for both males and females,

allowing for straightforward interpretation of the task

main effect. These comparisons revealed that subjects per-

ceived significantly different amounts of noncontingent

negative feedback on each of the three tasks (p < .001):

Subjects perceived the greatest amount of noncontingent

negative feedback on Task C, an intermediate amount on

Task A, and the least amount on Task B.



Comparison of Feedback Judgement Patterns by Group

The above analyses have individually contrasted de-

pressed and nondepressed subjects' percentage estimates

















* Males

o Females


Figure 1.


Subjects'
estimates
task.


negative feedback percent
as a function of sex and


U 0

(0



to S



(d


Q) >


0 XL

P 0)

E -
Hrl
0- t


TASK









for each of the three feedback modes. Figure 2 simultane-

ously presents the mean estimates for each group for all

three of the feedback modes, allowing for comparison of

the overall pattern obtained by each group. For each

group, the differences among subjects' estimates for each

of the three modes, across all tasks, were analyzed us-

ing a paired-comparisons t-test procedure. The results

for the depressed group indicated that these subjects

judged that the device was set significantly more often

in the accurate feedback mode than in either the positive

(t (39) = 4.95, p < .001) or negative feedback modes

(t (39) = 3.41, E < .01). The depressed subjects also

estimated that the device was set slightly less often in

the positive feedback mode than in the negative feedback

mode (t (39) = 2.63, p < .02).

In contrast to the pattern obtained by the depressed

subjects, there was no difference between nondepressed sub-

jects' accurate and negative feedback mode estimates

(t (39) = 0.43, ns), and they exhibited only a marginally
significant tendency to give higher accurate feedback mode

estimates than positive feedback mode estimates (t (39) =

1.87, p <.10). However, like the depressed subjects, non-

depressed subjects estimated that the device was set sig-

nificantly less often in the positive feedback mode than

in the negative feedback mode (t (39) = 2.26, p < .05).

















50

Depressed Group


088 Nondepressed Group
40



E- 0) 00000
T 00000c 00000
0 00000 0000C
0000 00000
30 00000 oooo
00000 000 0000C
o OOOOOC
0000 0000 0000C
a) a) 0000 OOOO O o0000 O C
S0000 00000000C
000 O000000 O0000C



0000i 2ssed an00000d nde esu0000C
o es f oooooc
40 0OC 0 O0000C
S00 o0000 0000C
-p
S0 0000 000C
Sc OOOOCe s 000i era OOe0C
i g r 2C C a i osseO a C
th devic C was setOC io
S000bC ack odeCs
000C 0000 O000C
000c 0000 000C
S10 oo oc00000 oooc
000oc 0000 000C
0000C OOOOC 00O00
0000C 000 C 000C
O000C 00OOC 000C
O000C 000 C 30000
0000C 000OC 0000
0000C 0000C 0000
0000C OOOOC 0000
OOOOC O000C0000


Accurate Positive Negative
Feedback Feedback Feedback
Mode Mode Mode




Figure 2. Comparison of depressed and nondepressed sub-
jects' estimates of the percentage of trials
the device was set in each of the three feed-
back modes.










Probability of the Bell Given a Correct Versus an Incorrect
Response

Probability of the bell given a correct response.

Analysis of subjects' estimates of the probability that

the bell would occur given a correct response (suggesting

the presence of a contingent relationship) yielded signif-

icant main effects for group and task. Means and signifi-

cance values for these effects are presented in Table 6

(complete results of this analysis are presented in Table

E-4 of Appendix E).

The main effect for group indicates that depressed

subjects reported significantly higher probabilities that

the bell would occur given a correct response than did

nondepressed subjects (p < .01). Further examination of

the significant main effect for task, using post-hoc Dun-

can's comparisons, revealed significant differences across

each of the three tasks (P < .05): Subjects gave the high-

est estimates that the bell would occur given a correct

response on Task B, intermediate estimates on Task A, and

the lowest estimates on Task C.

Probability of the bell given an incorrect response.

Analysis of subjects' estimates of the probability that

the bell would occur given an incorrect response ( sug-

gesting the absence of a contingent relationship) yield-

ed significant main effects for order and task, and









Table 6. Means and significance values for significant
effects identified by analysis of subjects' estimates of
the probability that the bell would occur given a correct
response.


Effect Mean F df

Group

Depressed 54.150
7.22 1,72 .009
Nondepressed 47.542



Task

Task A 60.075

Task B 64.963 145.74 2,144 .001

Task C 27.500









a marginally significant group X order interaction. Means

and significance values for these effects are presented in

Table 7 (complete results of this analysis are presented

in Table E-5 of Appendix E).

The significant main effect for order indicates that

Order 1 subjects reported significantly lower probability

estimates that the bell would occur given an incorrect re-

sponse than did Order 2 subjects (P < .003). Further anal-

ysis of the significant main effect for task, using post-

hoc Duncan's comparisons, revealed that subjects' estimates

were significantly higher in Task B than in either Task A

or C (p < .05); subjects' estimates were not significantly

different in Tasks A and C.

Further analysis of the marginally significant group

X order interaction using post-hoc LSD comparisons revealed

that while depressed/Order 1 subjects gave significantly

lower estimates than did depressed/Order 2 subjects (p <

.05), nondepressed/Order 1 and nondepressed/Order 2 sub-

jects did not differ significantly in their estimates.

This finding suggests that the main effect for order, dis-

cussed above, may relate primarily to differences among

the depressed subjects. The marginally significant group

X order interaction is graphically depicted in Figure 3.

In order to gain a better understanding of the order

effects identified on this measure, subjects' estimates

were plotted separately for each group by order and task









Table 7. Means and significance values for significant
and marginally significant effects identified by analysis
of subjects' estimates of the probability that the bell
would occur given the absence of a correct response.


Effect Mean F df P

Order

Order 1 32.208
10.01 1,72 .002
Order 2 41.000



Group X Order

Depressed

Order 1 28.533

Order 2 42.683
3.72 1,72 .058
Nondepressed

Order 1 35.883

Order 2 39.317



Task

Task A 35.588

Task B 42.475 4.61 2,144 .012

Task C 34.750























Order 1

o Order 2






p~


Depressed


Nondepressed


GROUP


Figure 3. Subjects' estimates of the probability
that the bell would occur given an in-
correct response as a function of group
and order.


-C 0




-4-

mno
o
-0) C
PCL




SH
U4-)
*f1W










(see Figure 4). This figure shows that while both depress-

ed and nondepressed Order 2 subjects made higher estimates

on Tasks B and C than did Order 1 subjects, the depressed/

Order 2 subjects made particularly high estimates on Task

C, and also on Task A, where no true order effect could

occur (since subjects in both orders received this task

first).

It may also be noted that while the main effect for

group did not approach significance (F (1,72) = 0.51, p <

.48), the depressed subjects did score in the direction

of greater contingency perception, relative to the nonde-

pressed subjects (means of 35.608, for the depressed group,

and 37.600, for the nondepressed group). The group X order

interaction, discussed above, indicates that this finding

was particularly evident among Order 1 subjects.

Differential between subjects' probability estimates.

Analysis of the difference between subjects' estimates

of the probability of the bell given a correct versus an

incorrect response yielded only a significant main effect

for task. Means and significance values for this effect

are presented in Table 8 (complete results of this analy-

sis are presented in Table E-6 of Appendix E).

Further analysis of the significant main effect for

task, using post-hoc Duncan's comparisons, revealed that

subjects obtained smaller differences between the two prob-

ability estimates discussed above on Task C, relative to























O 0 -P
O O oM





S-





+ O
bfl
C- +E-













o O 0


,. rl
0 0



























+ a
CC
















r I o
. a) 0














-P--
H m
E o o -P
(bD





O O 0 -,








S0suosoouO

T q JO Tqqo -


cf



cnt










C \ \7 CVC)


\SuOd9Gtj 4~GJJ0DI ue u^ nAT