Title: Leadership and the group-induced shift
Full Citation
Permanent Link: http://ufdc.ufl.edu/UF00098656/00001
 Material Information
Title: Leadership and the group-induced shift a field study of a complex decision-problem
Physical Description: x, 128 leaves : ill. ; 28 cm.
Language: English
Creator: Jewell, Linda Neal, 1941-
Publication Date: 1977
Copyright Date: 1977
Subject: Leadership   ( lcsh )
Group decision making   ( lcsh )
Management thesis Ph. D   ( lcsh )
Dissertations, Academic -- Management -- UF   ( lcsh )
Genre: bibliography   ( marcgt )
non-fiction   ( marcgt )
Thesis: Thesis--University of Florida.
Bibliography: Bibliography: leaves 118-127.
Statement of Responsibility: by Linda N. Jewell.
General Note: Typescript.
General Note: Vita.
 Record Information
Bibliographic ID: UF00098656
Volume ID: VID00001
Source Institution: University of Florida
Holding Location: University of Florida
Rights Management: All rights reserved by the source institution and holding location.
Resource Identifier: alephbibnum - 000207913
oclc - 04080350
notis - AAX4717


This item has the following downloads:

leadershipgroupi00jewe ( PDF )

Full Text







The author wishes to express her appreciation to the members of

her supervisory committee for their assistance in completing this

dissertation. Professor H. Joseph Reitz was an extremely capable

chairman whose active interest and assistance throughout the duration

of this project were invaluable. Professor Barry M. Schlenker

provided much needed guidance through the extensive body of literature

relevant to the topic of this paper and Professor Jerald W. Young

was particularly helpful in the area of data analysis. In addition,

all members of the committee deserve commendation for their comments

and suggestions on the style and method of presentation of the final


The author also wishes to thank Yvonne Robertson, who typed this

manuscript. Her skill as a typist and her knowledge of the require-

ments for placing this dissertation in completed form made the task

considerably easier.

Finally, appreciation is extended to the author's husband, Don,

without whose support this project would not have been possible.


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

LIST OF TABLES ........... ......... ................ v

LIST OF FIGURES............................ . ........... vii

ABSTRACT................... .... ........ .. .... ......... viii


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


Quality of Group Solutions to Problems ........... 5
Nature of Task................................. 5
Composition of the Group ....................... 11
Group Dynamics ................................. 15
Summary ....................................... 27
Acceptability of Group Solutions to Problems..... 27
Attitudes and Behavior......................... 29
Compliance-Conformity.................. ......... 30
Susceptibility to Influence.................... 33
Summary...................................... 35

THE RISKY SHIFT PHENOMENON........................ 36

IV THE PRESENT STUDY ................................. 40

Background .................. .............. 40
Leadership............... ..................... 41
The Boulanger and Fischer Study................ 42
Purpose of the Present Study................... 45
Method.................... .. ............... 46
Subjects .................. .................... 46
Identification of Emergent Leaders ............. 50
Group Composition.................. ........... 52
Instruments ..................... ............ 53
Experimental Manipulation...................... 59
Procedure......... ........................... 62
Hypotheses and Analyses ................ ...... 64
Alternative Hypotheses......................... 71


Results and Discussion .............................. 75
Main Hypotheses .................. ............. 75
Alternative Hypotheses................... ........... 81
Other Analyses................... ........... 88
Summary and Conclusions.............. ...... ...... .... 100


INSTRUMENTS.. .............................. .... ............. 110


EXPERIMENTAL INSTRUCTIONS................................... 114

REFERENCES......................... .............. ............. 118

BIOGRAPHICAL SKETCH ................. ......................... 128



Table 1 Zero-order correlation coefficients among
experimental variables............................ 74

Table 2 Pre-group DSP score means and standard
deviations in the four experimental
conditions .................................. 76

Table 3 Mean change scores from pre-group to group
administration of the DSP by experimental
condition............... .... ...... ............ 79

Table 4 Duncan New Multiple Range Test of differences
among the mean pre-group to group change
scores.................. ........ ............. 80

Table 5 Mean change scores from group to post-group
administrations of the DSP by experimental
condition.................. ............... 82

Table 6 Duncan New Multiple Range Test of differences
among the mean group to post-group change
scores...................... ......... ....... 83

Table 7 Partial correlation coefficients: group DSP
scores with each predictor variable, con-
trolling for all other variables................. 84

Table 8 Partial correlation coefficients: Post-group
DSP scores with each predictor variable, con-
trolling for all other variables. Experi-
mental conditions I, II, and IV.................. 86

Table 9 Partial correlation coefficients: Post-group
DSP scores with each predictor variable, con-
trolling for all other variables. Experi-
mental condition III ........................... 87

Table 10 Mean pre-group to post-group DSP change
scores for Internalizers and Externalizers
by quality of information...................... 90

Table 11 Mean pre-group to post-group DSP change
scores for Internalizers and Externalizers by
quality of information............. ... ....... 91

Table 12 Mean change scores by level of expressed
confidence in pre-group DSP solution........... 93

Table 13 Number of subjects in each experimental con-
dition reporting the six degrees of "Satis-
faction with group interaction"................ 94

Table 14 Mean group to post-group change scores for
subjects reporting the six degrees of "Satis-
faction with group interactions"............... 96

Table 15 Number of subjects in each experimental
condition reporting the six levels of "Con-
fidence in group decision"................... 97

Table 16 Mean group to post-group change scores for
subjects reporting the six levels of "Con-
fidence in group decision". ................... 98

Table 17 Mean semantic-differential scores for the
four experimental conditions................... 99



Figure 1 The four experimental conditions ............... 65

Figure 2 Expected score shifts for the four experimental
conditions. .................................. 69

Figure 3 Obtained mean score shifts for the four experi-
mental conditions over the three DSP ad-
ministrations.................. ................. 77

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



Linda N. Jewell

December 1977

Chairman: H. Joseph Reitz
Major Department: Management

The use of problem-solving or decision-making groups is a

common phenomenon in business organizations. The two major

assumptions underlying the use of such groups are that solutions so

produced will be of higher quality and greater acceptability to

those who must implement them than solutions produced by individuals.

Examination of the literature relevant to these assumptions suggests

that both are subject to qualifications and that further research

into both is required before the complete nature of those qualifications

is understood.

The research question addressed in the present investigation

was: What are the effects of deliberate attempts by certain key

individuals to influence problem-solving groups toward a particular

solution on (a) the quality of the solution and (b) the acceptance

of the group solution by the individual members? The same-task, Pre-

group (individual) solution--Group solution--Post-group (individual)

solution research paradigm of the Group-induced Shift research was

employed. Comparison of initial individual solutions with group

solutions provided information as to the relative quality of group

versus individual solutions. Comparison of group solutions with

subsequent final individual solutions allowed for an estimation of the

acceptability of the group solutions.

Subjects were 16 intact sub-sets of real, on-going task groups

in which the most influential member had been identified prior to

the experimental session. The Desert Survival Problem, a complex,

factual-decision problem with a correct solution, was the experi-

mental task. Four experimental Conditions were created by varying

(1) the quality of problem-relevant information interjected into

the group (Good-Poor) and (2) the status of the individual who held

the information (Leader-Non-leader).

Condition I: Leader-Good Information

Condition II: Non-leader-Good Information

Condition III: Leader-Poor Information

Condition IV: Non-leader-Poor Information

With respect to the quality of group solutions, it was expected

that Leaders in possession of problem-relevant information would

effectively influence their group solutions in the direction of that

information, regardless of its objective quality, whereas Non-leaders

would not. Specifically, in terms of mean improvement from Pre-group

to Group solution, it was hypothesized that

I (most improvement) > II and IV > III (deterioration)

Obtained results were:

I and II (most improvement) > IV > III (deterioration)

Thus, the effectiveness of influence attempts was affected by the

quality of the problem-relevant information as well as by the status

of the source.

With respect to the acceptability of the group solution, it

was hypothesized that group solutions in Conditions I, II and IV

would represent conformity (true change) and those in Condition III,

compliance (surface change). Therefore, the final solutions of

subjects in Conditions I, II, and IV were expected to closely resemble

the solutions of their groups, while those of subjects in Condition

III were expected to more closely resemble their own initial solutions.

As Condition II group solutions were expected to be poorer than initial

individual solutions, the following specific hypothesis about mean

improvement from Group to Post-group solutions was advanced:

III (improvement) > I, II, IV (no change)

Results fully supported this hypothesis.

Other results of the study suggest that both satisfaction with

group interaction and confidence in group solution can, at least in

established groups, exist independently of the objective quality of

the group's problem solution. Finally, substantial support for a

relationship between individual susceptibility to influence and scores

on the Locus of Control Scale was found. This relationship was

moderated by both source and quality of information.

Implications of findings for organizations and for further re-

search are discussed.


In a 1965 review of the state of group problem-solving research

at that time, Hoffman writes: "One of the foremost difficulties

(in evaluating this state) concerns the term 'problem solving' it-

self. Problem solving has been used with reference to tasks as varied

as judging the number of dots briefly displayed on a large card,

to producing answers to arithmetic reasoning problems, to solving

the complex problems faced by the management of large business

organizations" (p. 113). Davis (1973) has suggested that the fundamental

difficulty lies in distinguishing between problem solving and decision

making. Group decision tasks, according to Davis, emphasize the

selection of a response from a set of mutually exclusive and exhaustive

alternatives. Problem-solving tasks, on the other hand, emphasize the

processing of information to construct response alternatives, one or

more of which may or may not be "logically correct." In practice,

it is difficult to maintain this distinction. In organizations, the

construction of response alternatives to a particular problem is

usually for the purpose of selecting one of these alternatives upon

which to act; i.e., once the alternatives have been constructed they are

treated as if they are exhaustive and the process becomes, in Davis'

terms, a decision-making one. The term "decision-problem" (Davis, 1973,

p. 98), used in the title of this dissertation,would seem to be

appropriate in such situations. For convenience, however, the term

generally used in the body of the paper will be simply "problem" as

much of the research reviewed does not take place in organizational


The use of groups to confront problems and make decisions about

dealing with these problems is a fact of life in the modern American

business organization (Hoffman, 1965). Whatever the details of a

particular problem, the distinguishing feature of the task confronting

all such groups is that a single solution (answer, product, or action)

representing the group is required (Kelley & Thibaut, 1969, p. 12).

Maier early distinguished between two dimensions of a solution

to a problem--quality and acceptance (Maier, 1952). Quality refers

to the decision itself and reflects the ability to produce and process

information effectively. Acceptance "refers to the degree to which the

group that must execute the decision accepts it" (Maier, 1963, p. 253).

The assumptions underlying the increased use of problem-solving groups

in organizations are that the solutions generated by such groups will

have both higher quality and higher acceptance than those produced

by individuals (Reitz, 1976, p. 371). This paper will examine these

assumptions in some detail. In Section II a review of the literature

relevant to the issues is presented. First the issue of the relative

quality of group versus individual solutions to problems is considered.

Three variables are examined. First, the literature relevant to the

nature of the problem-solving task is reviewed. Second, the composition

of groups, with particular attention to the homogeneity-heterogeneity

dimension, is considered. Finally, the dynamics of group interaction

are examined. In the discussion of dynamics is found the literature

on cooperation-competition, status differences, and pressures toward

uniformity and the influence of these factors on communication processes

in groups. Power and group cohesiveness receive particular attention

with respect to their influence on the nature of the grouo product.

The question relevant to the second assumption underlying the

increased use of problem-solving groups is: Are group-generated

solutions to problems more acceptable than individual solutions? A

review of the literature relevant to this question also is presented

in Section II. This review leads into consideration of the conformity-

compliance distinction and from there to an examination of some of the

factors affecting individual susceptibility to influence.

From the general considerations of group processes presented in

Section II, the transition to the particular is accomplished in

Section III by means of a brief discussion of a particular line of

group problem-solving research--the Risky Shift or Group-induced Shift.

Out of this work comes the general research paradigm used in the

present study; i.e., pre-test, group test, post-test. The Group-

induced shift literature also provides the "benchmark" study for the

present investigation--a 1971 Risky-Shift study by Boulanger and

Fischer. That study investigated the disproportionate influence

in groups of "natural" or emergent leaders. The purpose of the present

study was to investigate the possible extension of the findings of

those authors to real groups in a complex problem-solving situation.

The present study, described in Section IV, was specifically

intended to increase understanding of the influence of emergent group

leaders upon (a) group-generated solutions to problems (quality) and

(b) subsequent individual solutions to the same problems (acceptance).

The problem-solving task was selected in line with conclusions from the

literature as to the type of task upon which groups may be expected to


perform better than individuals. The subject pool and experimental

setting allowed for assumptions as to heterogeneity of group membership,

cooperative organization of groups, and substantial group cohesiveness,

all factors found to facilitate group problem solving. In addition,

the availability of certain individual-difference information allowed

for some consideration of the relationship of these variables to

individual patterns of problem solution from pre- through group, to



Quality of Group Solutions to Problems

It has long been assumed that groups, by virtue of greater avail-

able resources, should be able to produce problem solutions of higher

quality than individuals. Like most such broad assumptions, this one,

while it will be seen to hold some validity, must be qualified. Major

reviews of the literature in the area show that the relative problem-

solving superiority of groups to individuals is neither clear-cut nor

simple (Kelley & Thibaut, 1969; Shaw, 1976). At the outset, the nature

of the task facing the individual or group has consistently proved to

be an important moderating variable.

Nature of the Task

The two most common experimental designs for investigating

individual versus group problem solving are (a) the same individuals

solve problems alone and similar problems in groups (within-subjects

design), and (b) one sample of individuals solves problems alone while

a different sample solves similar problems in groups (between-subjects

design). There are variations in the measures of performance which

include quality of solution, time to solution, number of problems

solved, and number of trials to solution. There is also considerable

variation in the kinds of tasks utilized in this research. Examination

of some of the literature will illustrate the critical role which the

nature of the task has played in determining the conclusions which have

been reached as to group versus individual problem-solving performance.

As discussed by Shaw (1976, p. 58), there are really two questions

to be answered when comparing individual and grouo performance: (a)

Is group performance superior to that of the average individual? and (b)

Is group performance superior to that of the most proficient individual?

The following review will look at studies in which (1) group performance

is superior to both (a) and (b) above, (2) group performance is better

than average but not as good as the best individual, and (3) group

performance is inferior to individual performance.

Group performance is superior to individual performance. An

early study by Marjorie Shaw (1932) has become a classic in the com-

parison of group and individual performance. Members of a class in

social psychology were divided into two sets of subjects. In the first

half of the experiment, members of one set worked in groups of four

while those in the second set worked as individuals. Problems were of

the puzzle-type, such as the well known Cannibal Problem. During the

second half of the experiment, roles of the subjects were reversed.

During this session, problems were of the type of picking the best

location for a school or the best routes for two school buses, given

certain information. Groups produced more correct solutions on both

types of problems, but they were often slower.

More recently, Faust (1959) found that four-man groups were able

to solve more anagram problems than were nominal four-man collections

(four individuals working separately and given credit as a "group" for

solution if any one individual achieved it). The task was sequential in

nature--each anagram had to be unscrambled before the next one was at-

tempted. Davis and Restle (1963) also found that groups solved more

problems than did individuals. Problems in this study were of a puzzle

type. Finally, Barlund (1959) found that four six-member real groups

reached significantly more correct decisions on a series of syllogism

tasks than did comparable nominal groups. Additionally, 27 of the 29

groups outscored the member identified on a pre-test as being best

at the task.

The foregoing review is, of necessity, sketchy, but the reader

with even passing familiarity with the tasks mentioned will quickly

perceive that they have two characteristics in common; they have

multiple parts, and they are susceptible to division of labor. Group

members can generate satisfactory solutions when individuals cannot

by putting together (pooling) information not uniformly held among

the members or by combining part solutions. The process may be thought

of as one in which the various group members "cover" for one another's

gaps in knowledge or problem-solving deficiencies. "Even if one member

of the group knows much more than anyone else, the limited unique

knowledge of lesser-informed individuals can serve to fill in some

gaps in knowledge" (Maier, 1967, p. 240). But it goes beyond mere

coverage of information gaps: "It has been shown that individuals

get into ruts in their thinking . . The individual is handicapped in

that he tends to persist in his approach and thus fails to find another

approach that might solve the problem in a simpler way. Individuals

in a group have the same failing but the approaches in which they are

persisting may be different . . Since group members do not have

identical approaches, each can contribute by knocking others out of ruts

of thinking" (Maier, 1967, p. 240).

Evidence as to whether group problem solving is anything more

than pooling; i.e., whether new responses not available in individual

repertoires is made possible by the group process is mixed (Banghart &

Spraker, 1963; Dunnette, et al., 1963; Lorge, et al., 1953; Permutter,


Group performance is above average individual performance. A

second group of studies has found group performance to be above the

average performance of individual subjects but generally below the

performance of the best individual. An early investigation by

Gurnee (1937) compared classroom groups and noninteracting individuals

on multiple-choice achievement test performance (between-subjects

design). The group performance scores were determined by majority

vote. Results showed that group scores were superior to the average

of the individual scores, but only approximated the scores of the best

individuals. Still earlier, Marston (1924) found a trained "judge"

(individual) to be more accurate than a "jury" (group) in reconstructing

a staged classroom incident. The difference in performance between the

juries and averages of non-trained individuals was not significant.

In later studies, Hudgins (1960) found that the arithmetic test

performance of groups of fourth graders approximated the level of the

best member (within-subjects design). Hall, et al., (1963) studied

group and individual predictions made after seeing the first half of

a movie as to what would happen in the last half. They found group

decisions to be better than average individual decisions made before

the discussion, but to be less accurate than the best individual


Studies finding group performance better than average individual

performance tend to be fraught with methodological difficulties (see

Shaw, 1976, pp. 58-61). Where results can be interpreted unambiguously,

the problems on which group performance was compared with average

individual performance tend both (a) to require few steps and (b)

to have judgements or solutions which can be verified by all persons in

possession of the original facts of the problem. The first

characteristic identifies such tasks as being neither so complex as

to require more information and skills than one person can be expected

to possess nor so complex that the group is likely to interfere with

individual solution. The second characteristic implies that upon

achieving the solution of such a problem, the individual can readily

convince other group members of its validity. Kelley and Thibaut

suggest that plurality support for the solution at the outset, or

attainment of the solution by a high-status group member are alternate

conditions under which a particular solution will be adopted (1969,

p. 65). The last condition will be central to the study described in

Section IV.

Group performance is inferior to individual performance. It has

been stated that groups produce higher quality solutions than individuals

to problems which have multiple parts and allow for division of labor

and pooling of the results of this labor. On problems requiring few

steps and having verifiable solutions it has been stated that groups

still tend to have the advantage over the average individual problem-

solver. Tasks which violate all of these conditions are those whose

solution ". .. requires thinking through a series of interrelated steps

or stages, analyzing a number of rules at each point, and always

keeping in mind conclusions reached at earlier points" (Kelley & Thibaut,

1969, p. 69-70). Such problems might be called multiple-stage problems.

They are not amenable to division of labor and the large number of

possible lines of reasoning any individual may follow in confronting

such a problem makes it difficult to demonstrate the correctness of a


given answer. On such tasks it appears that members of problem-

solving groups interfere with one another more than they assist one

another. Several studies are illustrative.

A 1958 study by Faucheux and Moscovici comparing individual and

group problem-solving performance reported mixed results. On one

series of tasks, group performance was superior. On another it was

no better than the average individual and inferior to nominal group

scores. Examination of the tasks involved showed the first set to

have the previously-described characteristics of tasks on which group

performance tends to be superior to individual performance. The

second set, Euler's figures, required following a consistent strategy

to solution. Similar results are reported by Lorge and Soloman (1959)

and Davis and Restle (1963). On all of these tasks, it appears that

group members created "noise" for one another; that is, "The verbal-

izations of several members who have started at different points and

are pursuing different lines of reasoning are mutually disruptive"

(Kelley & Thibaut, 1969, p. 70). The point is consistent with

Zajonc's (1966) suggestion that the presence of others facilitates

well-learned responses (performance) but interferes with less well-

established responses (learning). The responses necessary to cope

with the kinds of tasks described in this section tend to be ones

which individuals in general have the least opportunity to rehearse.

Additional data from these studies suggested that all members tended

to contribute to the discussion whether their ideas were helpful or

not (Davis & Restle) and that the difficulty of demonstrating the

correct solution seemed to inhibit the contributions of proficient

members (Lorge & Soloman).

The question to which this section of the present paper is

addressed is: Do groups produce higher quality solutions to problems

than individuals? A partial answer to that question has been developed.

Groups do tend to attain better success than any member acting alone

on certain types of problems. However, even if the cards are stacked

in favor of groups by limiting the task to one on which group performance

tends to be superior, additional considerations remain.

Composition of the Group

It is obvious that analysis of group problem solving cannot exist

independently of the individuals who compose the group. "There can be

no doubt that the kinds of individuals who make up a group constitute

a set of powerful determinants of group behavior" (Shaw, 1976, p. 194).

But the characteristics of the particular members of a group do not

tell the whole story. Data from a 1960 study, for example, show clear

differences in amount of problem-related activity, not only among members

of any given group, but also among groups (Hoffman & Smith). Thus,

"The particular pattern or combination of individuals making up a

group is also highly important" (Kretch, et al., 1962, p. 463). It is

this patterning or combination which is of primary concern here. The

argument as to whether there is "really" such a "thing" as a group

(opposed to a collection of individuals) will be sidestepped. The

issue of focus is a comparison of individual and group problem solving.

A group solution, as noted, is a product, answer, or action representing

the individuals in the group. To this extent, the group may be seen

to be "real" and individual differences may be conceptualized as

differences among patterns in group composition.

Given the large number of individual characteristics and within-

individual patterns of these characteristics in the world, there exists,

of course, an infinite number of group combinations. To investigate

the specific effects of even a small number of these combinations

is both impractical and of questionable utility, as any one combination

may be said to be a unique event. There does seem be be utility

however, in giving some attention to one global dimension of group

composition--heterogeneity versus homogeneity.

For purposes of this discussion, a heterogeneous group is com-

posed of individuals who have different levels or "amounts" of some

trait or characteristic. Homogeneous groups consist of members similar

on that trait or characteristic. Obviously, groups may be heterogeneous

with respect to some traits and homogeneous with respect to others. A

group of females chosen at random from a large introductory psychology

class, for example, will be homogeneous with respect to sex, but

relatively heterogeneous with respect to intellectual ability and

personality characteristics.

The literature offers a substantial number of studies utilizing

heterogeneity-homogeneity of group composition as the independent

variable of interest. Both the traits studied and the dependent vari-

ables examined vary widely. For present purposes, the dependent vari-

able of interest is problem-solving performance. A brief review of

the effects on this performance of heterogeneity versus homogeneity of

group composition with respect to ability, sex, and certain personality

traits is presented below.

Ability. The evidence with respect to heterogeneity versus

homogeneity on the trait of ability (measured by pre-tests on the same

or similar tasks or by scores on intelligence or aptitude tests) is

clear--superior performance results from superior ability (Laughlin

& Betz, 1975). When the over-all ability levels of two groups are

not comparable, groups with the highest proportion of high-ability

individuals, be they heterogeneous or homogeneous in composition,

perform better. But if the over-all ability level is comparable,

heterogeneous groups usually perform better (Goldman, 1965; Laughlin

& Branch, 1972). It is suggested that the explanation for this

tendency to superiority is the same as that for the superiority of

groups to individuals on certain kinds of tasks--there is likely to

be greater range in problem-solving approaches in heterogeneous


Sex. With respect to sex composition of groups and problem-

solving performance, the literature is largely silent. Such evidence

as is available suggests that mixed-sex groups perform more effectively

than same-sex groups (Hoffman & Maier, 1961). More studies have

investigated the effects of heterogeneous versus homogeneous group

composition on conformity in groups (Luchins & Luchins, 1955; Reitan

& Shaw, 1964; Tuddenham, et al., 1958). According to Shaw (1976),

the evidence favors the conclusion that both sexes conform more in

mixed-sex groups. No doubt this process seriously confounds investi-

gation of the influence of sex composition on performance. The role

of conformity in determining group products will be considered at

greater length in a later section of this paper.

Traits. The largest number of investigations into homogeneity

versus heterogeneity of groups utilize relative similarity on personality

traits. Taken as a group, the trait studies, like the ability studies,

show a slight tendency to favor heterogeneous groups when the criterion

is problem-solving performance. Hall (1975) found heterogeneous groups

(formed on the basis of scores on the Fundamental Interpersonal

Relations Orientation Behavior Scale) in a field study to be more

productive than homogeneous groups. Hoffman and Maier (1961) formed

heterogeneous and homogeneous groups on the basis of Guilford-

Zimmerman Temperament Survey profiles. Heterogeneous groups produced

a higher proportion of high quality solutions in three of the four

studies where problems had quality solutions. A later study

(Hoffman & Maier, 1966) confirmed these results. Finally, Sorenson

(1973) found clear superiority in both quality and quantity of

solutions on two intellective tasks for groups heterogeneous on the

trait of creative potential as opposed to groups homogeneous on this

trait. Creative potential was measured by scores on the Remote

Associates Test and a test of Social Differentiation.

The implications of the findings outlined above relevant to

heterogeneous or homogeneous groups with respect to problem solving

are clear so far as they go. Heterogeneous groups tend to outperform

homogeneous groups. The question, "Do groups outperform individuals?"

then, must take congnizance of the relevant traits of both individual

subjects and group composition. The problem is particularly acute

with respect to ability. A group is, by definition, more heterogeneous

than an individual. But the problem-solving superiority of

heterogeneous ability groups to homogeneous groups is predicated on

the basis that the over-all ability level of the two groups of subjects

is equivalent. Particular care must be taken, therefore, in evaluating

the relative superiority of problem solving performance of groups and

individuals. It may be that within-subjects designs deserve more

attention in such research.

With respect to the question which opened this section--Do

groups produce higher quality solutions than individuals?--, there

are now two partial answers available. (1) They do tend to perform

better on certain types of tasks, but (2) this better performance on

these tasks is dependent to some unknown extent on a heterogeneous

mix of individual characteristics.

At this point it might be tempting to conclude that if one

brings together a mixed group of bright individuals and presents

them with a multiple-part problem with specifiable criteria for a

good solution, the solution which the group develops will certainly

be superior to that which any one member might have developed alone.

Unfortunately, the mere presence of sufficient complementary in-

formation and skills is no guarantee that they will be used in making

a decision. Certain group pressures can operate to interfere with

or block the effective use of group resources.

Group Dynamics

As the term is used in the present paper, a problem-solving

group is a collection of individuals interacting on a face-to-face

basis. Group solutions are the result of group discussion as to the

problem issues, alternatives, and probable outcomes of proposed

solutions. Communication processes, therefore, are the means by which

a group develops a solution to a problem. If group resources are to

be utilized effectively, it is imperative that all members holding

information relevant to the problem have the opportunity to communicate

it. A first condition necessary for varied viewpoints to be heard

and considered is that the group be cooperatively organized.

Cooperation Competition. Hoffman writes: "Another contributor

to ineffective problem solving is the failure to organize or plan the

attack on the problem" (1965, p. 100). To "organize or plan an attack"

it is necessary that group members be agreed on a common goal. Such

agreement forms the base for cooperative organizations: ". . in a co-

operative situation group goals are homogeneous (i.e., members hold

the same goal for the group) and in a competitive situation, group

goals are heterogeneous (i.e,, group members hold differing goals for

the group)" (Shaw, 1976, p. 324).

The effects of cooperative versus competitive group organization

upon group processes have been extensively studied by Deutsch (1949

a & b). Several findings relevant to communication processes have

consistently emerged from these investigations. Compared with

competitively organized groups, those organized cooperatively showed

greater diversity in amount of contributions per member, attentive-

ness to fellow members, mutual comprehension of communication,

common appraisals of communication, and friendliness during discussion

(1949 a). Deutsch makes the link between these findings and group

productivity explicit: "To the extent that the results have any

generality, greater group or organizational productivity may be

expected when the members subunits are cooperative rather than

competitive in their interrelationships. The communication of ideas,

coordination of efforts, friendliness and pride in one's group which

are basic to group harmony and effectiveness appear to be disrupted

when members see themselves to be competing for mutually exclusive

goals .. The implications for committees, conferences, and small

groups in general appear fairly obvious" (1949 b, p. 230).

I /

The discussion above has been directed toward the point that

cooperative group organization facilitates communication necessary to

the effective use of group problem-solving resources. In "real-life,"

however, there are few, if any, purely cooperative or competitive

situations. Most everyday situations involve a complex set of goals

and subgoals. In a problem-solving group, members have needs and

goals relative to their relationships with other group members as

well as a stake in the common group goal. The requirements of these

two sets of goals may conflict. A member may withhold a valuable
insight into the problem, for example, because his need to appear

competent in the eyes of other group members requires that he not

risk the possibility that his contribution will be "put down."

Kelley and Thibaut state the implication of this "mixed motive"

(Schelling, 1960) situation for communication as follows: "Taken

together, the common and conflicting interest components create

dilemmas for the group member over being open versus secretive,

being honest versus deceitful, being trusting versus distrustful"

(1969, p. 38).

Status. The nature and strength of needs which can put the

member of a problem-solving group into a mixed motive situation

may be expected to vary considerably from one individual to the

next. However at least one characteristic of many groups--the

presence of status differences among the members--has been found

to have such consistent effects upon communication patterns that

it might be considered an almost universal "mixed motive generator."

A person's position is his place in the social system. His

status is the evaluation accorded that position (Shaw, 1976, p. 245).

Status is defined by the perceiver. Many positions carry almost

universally agreed-upon high status--heads of state, scientists,

and very wealth persons, to give some examples (Hodge, et al., 1964).

In other cases, a position viewed by some as having high status

may be viewed by others in an opposite manner. The head of a drug

ring, for example, may have high status in the eyes of the street

dealer while, at the same time, he is considered to be the "dregs

of humanity" by the population in general.

Status may be both ascribed and achieved (Shaw, 1976, p. 245).

Ascribed status is accorded the individual through no fault or merit

of his own (e.g., age, kinship). Achieved status is based on

individual achievement or failure. When the organizational problem-

solving group is formed, some members may have ascribed status. The

boss' nephew is a familiar example. Other status differences will

be based on past achievements such as the attainment of a high

position in the company. As group members interact, differential

achievement within the group itself may reinforce initial status

difference patterns or create new ones. The individual with the

highest position in the organization may, in fact, make the greatest

contribution to the group. Or the boss' nephew may be incompetent.

At this point, however, the focus is on the effects of status

differences upon communication, whatever the history of these


The first statement to be made about the influence of status

differences upon the pattern and content of communications is that,

in general, more communications are directed toward high-status

group memberss. A number of classic studies in the area have

demonstrated this communication pattern both in the laboratory and

in the field (see Thibaut, 1950; Back, et al., 1950). Additional

findings of these and other studies (see also Kelley, 1951; Worchel,

1957) reveal the content of communications directed toward high-

status individuals to be more positive than those directed toward

low-status individuals. Finally, studies have demonstrated that

the participation and influence of members in problem-solving

groups are directly related to the relative status of the members

(Harvey, 1953; Sherif, et al., 1955). "These effects are

primarily due to the human tendency to inflate expectations of per-

formance by high-status individuals and to overevaluate the

performance of high-status individuals" (Reitz, 1976, p. 380).

The implication of the findings outlined above for group

problem solving are considerable. Status differences among members

can affect information sharing, processing, and evaluation and,

through its effects on these variables, considerably affect the

nature of the group product. Of particular impact may be the un-

willingness of low-status membersto criticize suggestions of high-

status members (Janis, 1972). The disproportionate influence of

certain group members upon the nature of the group solution to a

problem was the specific focus of the research project to be

described in Section IV of this paper. The point to be emphasized

at this stage is that the kinds of communication processes necessary

to the efficient use of group problem-solving resources may be

considerably disturbed by the presence of status differences

within the group.

To recapitulate, it has been stated that groups achieve solutions

to problems by means of communication. Communication processes

are facilitated by cooperative group organization. But even within

cooperative groups distortions in communication are likely to occur

as the result of the presence of status differences. The next

section will address yet another barrier to the efficient utilization

of information in groups.

Pressures toward uniformity in groups. Even in group situations

where members might feel no status constraints against certain kinds

of communications, the free exchange of ideas, evaluation of suggestions,

and consideration of outcomes and alternatives relative to a problem

may be considerably hampered by other forces. Chief among these

forces is a pressure toward consensus or unanimity. Part of this

pressure stems from the nature of the task which the group is con-

fronting; i.e., a group solution is required. As Maier notes, this

requirement tends to set up a situation in which "Reaching agreement

in a group is often confused with finding the right answer . ."

(1967, p. 241). Support for this statement is found in a description

by Hoffman of the behavior of subjects in his own investigations of

group problem solving: "We have often had subjects who violently

opposed the majority's solution announce their capitulation with 'I

thought we were all supposed to agree'" (1965, p. 101).

Pressures toward unanimity stemming from definition of the task

as one of reaching an agreed-upon solution may be expected to exert more

influence upon the nature of the final solution than the initial elicita-

tion of ideas about the problem (although the latter effects will certainly

occur as disagreements about some points reduce the time available for

raising others). From this perspective, the situation is a bargaining

one. Disagreements will be settled on the basis of the balance of

power and the relative successes of influence attempts within the


Influence is defined by Collins and Raven as ". .. a change in

cognition, attitude, behavior, or emotion of P (person) which can

be attributed to 0 (other)" (1969, p. 160). The ability to in-

fluence stems from power: "If 0 is capable of influencing P with

respect to certain state, we say that 0 has power over P with

respect to that state" (Cartwright & Zander, 1968, p. 216). The

factors affecting individual susceptibility to influence attempts

will be discussed in IIB. Attention will be turned here to a brief

consideration of the sources of power for members of problem-solving

groups. French and Raven have distinguished five types of power:

reward, coercive, legitimate, referent, and expert (Cartwright &

Zander, 1968, p. 263).

The basis for reward power is the ability to administer positive

reinforcement. All group members possess this power to a certain

degree in that they are in a position to verbally reinforce state-

ments of other members (see Crowne & Strickland, 1961). Additionally,

in real-life groups, one or more members may hold specific reward

power over others in that they may be "the boss" outside the group

(or may be perceived to have the ear of a superior not in the group).

Finally, in both experimental and real-life groups, one member may

hold the power to distribute rewards for participating in the group

or rewards attached to the outcome of the solution.

Coercive power is based on the ability to punish or withhold

positive reinforcement. The remarks of the proceeding paragraph

are also applicable here.

Legitimate power stems from some sort of code or standard--

cultural values, social structure, designation--by virtue of which

it is agreed that an individual has power. Reward and coercive

power are often closely associated with legitimate power.

Referent power is based on identification. If P is so highly

attracted to 0 that he wishes to be like, or have a close association

with 0, then 0 has referent power. A unique feature of referent

power is that its effect may operate totally independently of the

possessor--one individual may influence the behavior of another

without being aware of doing so.

Expert power is called "informational power" by Deutsch and

Gerard (1955). It will vary with the extent of the skills and

knowledge attributed to the source. French and Raven (1959) note

that expert power exerts influence primarily on cognitive structures.

The various power bases briefly described above are seldom

independent. Possession of one is often associated with possession

or acquisition of others. In the problem-solving group, the

actual balance of power relationships is apt to be extremely complex.

It has been stated that the perceived requirement for an

agreed-upon solution puts many problem-solving groups into a

bargaining-type situation and that the nature of the final solution

will depend to some extent upon the distribution of power within

the group and the outcome of attempts on the parts of members to

influence one another in the direction of particular arguments or

proposed solutions. The significance of this process for the quality

of the end product is discussed by Maier: "In reaching consensus

or agreement, some members of a group must change . . If persons with

the most (objectively) constructive views are induced to change, the

end-product suffers; whereas if persons with the least constructive

point of view change, the end-product is up-graded" (1967, p. 243).

With this point the present discussion moves still closer to the

empirical investigation to be described in Section IV. There are

still, however, a number of issues to be addressed before turning

to that investigation.

As pointed out earlier, part of the pressure toward unanimity

in problem-solving groups stems from the perceived nature of the

task and this pressure tends to have its greatest effect upon the

nature of the solution. It is conceivable that such pressure can

lead to "knock down, drag out" fights, "hung juries" and other such

abortions of the problem-solving process in groups composed of

members who do not like one another and/or hold differing perceptions

of the group's goal. Such groups may be said to be characterized by

low cohesiveness.

Definitions of group cohesiveness vary (see Shaw, 1976, pp. 197-

198). A simple one is proposed by Cartwright: "....group cohesiveness

refers to the degree to which members of a group desire to remain

in the group"(1968, p. 91). Desire to remain in a group is commonly

stated to be a function of the attractiveness of the group. This

attraction will depend on such factors as the attractiveness of

individual members and attractiveness of group goals. Groups with

high status, groups whose members have similar backgrounds and

similar attitudes, successful groups, and groups with successful

members, for examples, have been found to have greater cohesiveness

than groups without these characteristics (Reitz, 1976, p. 325).

Other factors increasing cohesiveness receive less attention, but

may be equally important. "Cohesiveness would include not only

the attraction that group holds for its members but also any other

force operating on the individual to stay in the group" (Kiesler &

Kiesler, 1970, p. 65). Chief among these other forces may be the

alternatives to membership in a particular group.

Thibaut and Kelley (1959) suggest that individuals evaluate

the rewards and costs associated with group membership in terms of

certain internal standards. The Comparison Level (CL) is a standard

against which the attractiveness of a group is evaluated. The

Comparison Level of Alternatives (CL-alt) is a standard against which

the individual decides whether to stay in a group or leave it.

Theoretically, some or all members of a group may have a strong desire

to remain in the group, not because it is particularly attractive

in an absolute sense, but because it is more attractive than the

available alternatives.

Whatever its source, group cohesiveness is a powerful force for

unanimity within groups. "Perhaps the most widely reported

characteristic of cohesive groups is the greater tendency of individual

members to influence and be influenced" (Collins & Raven, 1969, p. 123).

Of themselves, these attempts and successes at influence serve basic

group maintenance functions (Shaw, 1976, p. 198). The relationship of

pressures to unanimity arising from a desire to keep the group

together and effectiveness of group problem solving is complex. On

the one hand, some level of cohesiveness is necessary for a group to

tackle a problem at all. "Whether a problem-solving orientation is

taken or not depends first on the readiness of the group members to

attempt to keep the group intact" (Kelley & Thibaut, 1969, p. 3).

On the other hand, a very high level of group cohesiveness may lead

to the replacement of independent creative thinking by what Janis

(1972) has called "Groupthink." Groupthink refers to a "deteriori-

zation of mental efficiency, reality testing, and moral judgement that

results from in-group pressures" (Janis, 1972, p. 9).

Not all highly cohesive groups are afflicted by Groupthink

defects in problem solving, but a high level of cohesiveness is a

necessary condition for the phenomenon to occur. It appears, there-

fore, that for purposes of effective group problem solving, some

level of group cohesiveness sufficient to allow for solving the problem,

but less than that which puts concern for group maintenance and

image above concern for the work at hand, is desirable.

Given that some level of group cohesiveness is required for

effective group problem solving, the question becomes: What effect

do the pressures toward uniformity resulting from cohesiveness have

on problem-solving processes? Remembering that the means by which

a group reaches a solution to a problem is communication, brief

attention will be given to the effects of cohesiveness upon group


A first effect of cohesiveness on interaction is on amount of

verbalization. Studies suggest that there is greater communication

in cohesive groups (Lott & Lott, 1961; Moran, 1966). Additional

evidence strongly suggests that verbal exchanges in cohesive groups

are more positive in tone than are those in non- or low-cohesive

groups (Back, 1951; French, 1941; Shaw & Shaw, 1962).

Lewin (1947) emphasized the need for fact-finding and objective

appraisal of alternatives in making decisions. It might be expected

that the increased communication and general positive tone of

cohesive groups would facilitate the free expression of ideas necessary

to obtaining the facts and evaluating the alternatives. To a certain

extent, evidence supports this expectation. Members of a highly

cohesive group feel freer to deviate from majority opinion than

members of other groups (Kiesler & Corbin, 1965). Persistent

deviancy, however, will first result in increased communication

toward the deviant and finally in rejection of the deviant by the

group (Schacter, 1951). The extent to which communication patterns

in cohesive groups will detract from elicitation of ideas relevant

to the problem, therefore, will depend to a considerable extent on

how quickly group opinion becomes solidified. Until it does, the

standard of "deviancy" is loose. Once the group achieves a high

degree of like-mindedness, however, information important to the

best solution of the problem is likely to be rejected if it is in-

consistent with the consensus. Maier has analyzed this process in

terms of valence of solutions (1967). His research shows that the

first solution which reaches a certain value (in his system of

calculating solution valences) is adopted by the group 85% of the

time (Hoffman & Maier, 1964). "Higher quality solutions introduced

after the critical value for one of the solutions has been reached

have little chance of achieving real consideration" (Maier, 1967,

p. 241). Thus, the quality of the end product of group problem-

solving processes in cohesive groups will, through the kinds of

communication patterns peculiar to such groups, vary considerably.


It will be useful to return once again to the question to

which the foregoing review of the literature has been addressed: Do

groups produce higher quality solutions to problems than individuals?

The answer can now be stated as follows: The quality of a group

solution to a problem depends upon (1) the nature of the problem,

(2) the composition of the group, and (3) the nature of the communi-

cation processes by which the problem solution is sought. To the

extent that (1) the problem has multiple parts and a demonstrable

solution, (2) the group contains a heterogeneous mix of competent

individuals, and (3) status differences and pressures toward uni-

formity do not seriously distort or prematurely truncate fact-finding

and discussion of the issues, it may be expected that the greater

resources available to a group will be utilized to produce a

solution of higher quality than would be produced by any one member

acting alone. Attention will now be turned to the second assumption

underlying the use of organizational problem-solving groups.

Acceptability of Group Solutions to Problems

The gist of the "acceptability argument" is as follows: "Many

problems require solutions that depend upon the support of others

to be effective. Insofar as group problem solving permits partici-

pation and influence, it follows that more individuals accept solu-

tions when a group solves a problem than when one person solves it .. . A

low quality solution that has good acceptance can be more effective

than a higher quality solution that lacks acceptance" (Maier, 1967,

p. 240).

Before proceeding to examine this agrument, it must be pointed

out that it contains an implied qualifying clause; that is, this

argument assumes ". .. .. that the outcomes of all members (of the

problem-solving group) will in some manner be determined by that

(group) product" (Kelley & Thibaut, 1969, p. 12). In real-life

groups, the extent to which this assumption is true can vary sub-

stantially. Certain groups make decisions which affect them only

indirectly (e.g., a legislative committee which allocates funds to

a University system). Other problem-solving groups are made up of

representatives of those who must implement the solution (e.g.,

union-management bargaining teams). Finally, there are problem-

solving groups which contain all individuals directly affected

by the solution (e.g., group of employees who set their own

production goals for a given period of time).

Organizations make use of all three types of problem-solving

groups described above although the second type is probably the

most common. It is not always clear which type of group is the

referent when claims are made as to the greater acceptability of

group-generated to individual-generated solutions. If the group is

of the first type described, the situation would appear to be little

different for those who must implement the decision from one in

which the decision was made by an individual. (The primary difference

would be the feelings of increased confidence in the validity of the

decision which might result from the knowledge that more than one

person made it.) Greatest acceptability gains would be expected when

all persons involved participated in the problem-solving process with

the representative problem-solving group falling somewhere in between.

A 1971 field study by Powell and Schlacter of varying degrees of

participation in solving work scheduling problems supports these

expectations. To avoid confusion, then, the present discussion will

be limited to the relatively "pure" type of problem-solving group

in which all members will be directly affected by the solution


Attitudes and Behavior

The acceptability argument for using groups to solve problems

really has two parts. First, it is argued that participation in

developing and planning the solution to a problem heightens under-

standing of the solution and commitment to it (Thibaut & Kelley,

1959). Second, it is argued that the increased understanding and

commitment will be associated with behaviors conducive to more

effective implementation of the solution. This is, of course, an

argument based upon the concept of attitude. Attitudes are said to

have three components--affective, cognitive, and behavioral (Reitz,

1976, pp. 257-258). The acceptability argument is predicated on the

assumption that the first two components of one's attitudes toward

the solution to a problem will be more favorable when the solution

is arrived at by a group than by an individual and that this favorable

orientation will lead to behaviors which facilitate implementation

of the solution.

The literature generally supports the first of the above two

contentions. Expressed satisfaction with solutions appears to

increase when individuals participate in solution generation as

opposed to having the solution imposed upon them by one individual

(Bragg & Andrews, 1973; Carey, 1972; Coch & French, 1948; Ivancevich,

1972; Powell & Schlachter, 1971; White & Ruh, 1973). The extent to

which these more positive attitudes lead to behaviors which facilitate

implementation of the solutions is another matter. Reviews of the

appropriate literature reveal no one-to-one correspondence between

expressed attitude and subsequent behavior (Brigham, 1971; Kiesler,

et al., 1969; Wicker, 1969).

Kelman (1958) has suggested that before one can make any pre-

dictions about the way attitude changes will be reflected in be-

havior, one must know something about the nature and depth of the

change. "It is not enough to know that there has been some measure-

able change in attitude: usually we would also want to know what

kind of change it is. Is it a superficial change, on a verbal level,

which disappears after a short lapse of time? Or is it a more

lasting change"(p. 52). The distinction between superficial and

lasting changes is often represented as a distinction between com-

pliance and conformity.


Both conformity and compliance refer to changes in behavior

resulting from real or imagined group pressure. Compliance is

"...overt behavior which becomes more like that which the group wishes

its members to show" (Kiesler & Kiesler, 1970, p. 3). Conformity

refers to private acceptance--a "real" change in the direction of

group attitudes and beliefs (Kiesler & Kiesler, 1970, p. 3). The

only way to determine the difference between the two is to observe

what happens when the group is not present. If an individual continues

to express the same attitudes or beliefs or to engage in behaviors

which support these attitudes or beliefs, conformity is assumed. The

argument for using groups to solve problems which is under discussion

here is, then, based on the assumption that the unanimity represented

by the group's solution is conformity, not compliance.

There are basically only two strategies for determining whether

the belief in a group-produced solution implied by agreement with it

represents an individual's "true opinion" as to the problem under

consideration or whether it is a "surface" response to group pressure.

One method involves giving each member of a group the opportunity

to express his private opinion as to the best solution to the problem

after the group has made its decision. The correspondence between

group solutions and post-group-discussion private solutions is taken

as a measure of conformity. Results generally suggest substantial

positive correlations between individual and group solutions

(see Muehleman, et al., 1976; Timmons, 1939).

The second method of investigating whether participation in

generating a problem solution increases the chance that the solution

will be implemented is to actually observe the behaviors of those who

participated in the solution. The classic study of this nature was

carried out by Lewin and his associates during World War II (Lewin,

1947). The problem to be solved was to maintain nutritional standards

of the American population in the face of a shortage of meat for

civilians. The solution advocated by the experts was increased con-

sumption of certain low-consumption meat items still in abundant

supply (heart, brains, etc.) and of milk, milk products, orange juice,

and cod-liver oil. Lewin employed a number of traditional methods

(lectures, appeals to partiotism, etc.) to persuade housewives to buy

more of the items named. An additional method involved housewives

discussing the problem among themselves with experts available to

answer questions. To test the effectiveness of the various methods,

researchers followed up on the actual food-buying behavior of the

participants. Results clearly indicated the superiority of the group

discussion method; i.e., members of discussion groups increased

consumption of the items named significantly more than did other


Other investigations into the changes in behavior resulting

from group problem solving have yielded mixed results. In the Powell

and Schlacter study (1971), there was no change in absenteeism rates

following group determination of work schedules. In present terms,

absenteeism was the problem and employee development of work schedules

was the solution. Employees in participant groups expressed greater

satisfaction with their work schedules than did employees in groups

where changes were passed down by management, but they did not change

their behaviors. On the other hand, the increased satisfaction with

method of problem solving found by Bragg and Andrews (1973) was

associated with both a decrease in absenteeism and an increase in


Evidence as to whether participation in solving problems and

making decisions facilitates implementation of the solutions may be

described as "conclusively inconclusive." If following through on

a problem solution may be regarded as conformity and agreeing to a

solution (evidencing satisfaction with it) but not following through

as compliance, one can say only that both appear to be elicited in

problem-solving groups. As the unit of analysis of interest to

organizations with respect to this question is the individual, it is

appropriate to consider briefly some of the individual factors which

determine whether an individual's subscription to a group solution

to a problem represents conformity or compliance.

Susceptibility to Influence

The most powerful variable affecting the ability of others to

influence an individual's "true belief" about the solution to a

problem (i.e., to elicit conformity) appears to be what Kelley and

Thibaut refer to as "stability of attribution" (1969, p. 7). In

common terms, it is the extent to which the individual feels con-

fident of the correctness of his own analysis of the problem and

his own view of the best solution. Attributional stability has been

found to be low with little soical support (Asch, 1951), poor or

ambiguous information (Asch, 1951), problems too difficult for the

person's capabilities (Coffin, 1941; Patel & Gordon, 1960), and

other experiences engendering low self-confidence (Boomer, 1959;

Kelley & Lamb, 1957; Hochbaum, 1954).

Obviously, in a group situation, confidence in one's own opinions

and abilities is not an all-or-none phenomenon, but a relative one.

In addition to making inferences about his own competence, an individual

makes inferences about the competence of the group (Ettinger, et al.,

1971). Whether an individual is "truly" persuaded by information

provided by other members of the group will partially depend upon

his perception of his own competence relative to that of other

members. This perception will be based upon the reputation which

others bring to or establish within the group (credibility), on the

compellingness of the arguments presented, and on the confidence

which others appear to have in their own opinions (Kiesler &

Kiesler, 1970).

There also appear to be certain personality variables which

affect general confidence in one's own abilities. Evidence

suggests, for example, that the individual described by Rotter (1966)

as an "internalizer" (one who believes that what happens to him is

primarily the result of his own actions) is less susceptible to

certain kinds of influence than is is the "externalizer" (one who

believes that what happens to him is largely beyond his control)

(Minton, 1972; Ritchie & Phares, 1969; Sherman, 1973). Generally,

however, the search for sovereign attributes of a "conforming

personality" has not been especially fruitful (Hollander & Willis,


Assuming that an individual member of a problem-solving group

is convinced that the solution toward which the group is moving is

less than correct, what might make him go along with that decision;

i.e., to comply? Kiesler and Kiesler have suggested a number of

reasons (1970, p. 41-45). One is an interest in presenting a united

front. "For the group to be perceived as accomplishing something,

it often must also be perceived as not having internal disagreement"

(p. 41). Another is a need to be liked, accepted, or, at least, not

rejected. A related reason is the necessity for maintaining the


relationships within the group because of the expectation of future

interaction as a group. It is also possible that one may desire to

continue participating in the group's activities (even though there

is no requirement to do so) and sees "not rocking the boat" as a

price to be paid for staying in the group (Hollander & Willis, 1967).


Compliance can serve a wide variety of personal needs and per-

ceived instrumentalities. Unless the consequences of a decision are

perceived to be very personally important, it is not unlikely that

the individual who sees the group consensus moving away from his

own private opinion as to the best solution to a problem will go along

with the group solution (Kelman, 1958). What happens after that

cannot be predicted at the individual level. Both Festinger (1957)

and Lewin (1947) make cases for the idea that counter-attitudinal

behavior (in this case, ascribing to a problem solution one does not

really believe in) tends to produce other behaviors consistent with

that behavior (in this instance, implementing the solution). On the

other hand, Brehm (1966) has suggested that an individual who complies

to group pressure but does not privately accept the group's action

may attempt to deal with this perceived threat to his freedom of

action by behaving subsequently in a manner contrary to the group.

There is, then, no simple answer to the question with which this

section began: Are group produced solutions to problems more

acceptable than individual solutions?


Section II of the present paper has given rather extensive

attention to the literature relevant to the many factors which

must be considered when examining the social situation called

here "Group Problem Solving." Because organizations are moving

increasingly toward the use of groups as opposed to individuals

to solve their problems, the framework for that presentation

consisted of an examination of the two primary assumptions under-

lying this shift--increased quality and acceptability of group-

generated solutions. As stated, the study described in the next

section was designed for the purpose of increasing understanding

of those two assumptions. Before the specifics of that investi-

gation can be presented, however, it is necessary to move out of the

general literature addressed in section II and into a particular

line of group problem-solving research--the Risky Shift.

Almost 15 years ago, a graduate student at MIT completed a

Master's Thesis which has stimulated a large and still-growing body

of related research. Reports of over 60 published studies and as

many unpublished ones on the subject of the "Risky Shift" phenomenon

(Stoner, 1961) appear in The Psychological Abstracts between 1972

and 1975. Stoner's initial conclusion that groups make riskier

decisions than individuals ran directly counter to "conventional

wisdom" as well as to then-prevailing views among social psycho-

logists that group decisions will be more conservative than individual

decisions or will represent averaging of individual decisions (Barlund,

1959; Farnsworth and Behner, 1931; Festinger, 1954; Krech, Crutchfield

& Ballachey, 1962). Subsequent research with Stoner's paradigm has

replicated his findings in a variety of settings (Bateson, 1966;

Jamieson, 1968; Marquis, 1962; Wallach, et al., 1962) and in a

variety of other countries (Bell & Jamieson, 1970; Lamm & Kogan,

1970; Rim, 1963; Vidmar, 1970).

As the body of related research built, it became apparent that

the Risky Shift phenomenon was broader than originally conceptualized.

A number of researchers have shown that items similar to those on

Stoner's original Choice Dilemmas Questionnaire (CDO) can be written

so as to elicit a shift toward caution within the same paradigm

(NordhVy,1962; Pruitt & Teger, 1967; Stoner, 1966). Others have

demonstrated that Stoner's paradigm can be used to change expressed

attitudes--a change which has been called "polarization" or

"extremization" of opinions (Doise, 1969; Gouge & Fraser, 1972;

Moscovici & Zavalloni, 1969; Myers & Bishop, 1970). Shifts have also

been demonstrated to occur in situations where positive and negative

consequences may actually result (Bem, et al., 1965; Blank, 1968;

Pruitt & Teger, 1969). In accordance with these and other findings,

Pruitt (1971) has suggested that a more appropriate name for this

line of research might be "group-induced shift" (p. 340).

As Pruitt's term implies, choice-shift research involves group

processes. The typical experimental design involves administering

the CDO (or variant) to subjects individually without their knowing

they will have further exposure to the task. Following some experi-

mental treatment, in which subjects meet as groups with the same

task, a post-test measure is taken on an individual basis, group

consensus basis, or both. Results have been remarkably consistent.

Whatever the experimental treatment--group discussion to consensus,

free discussion without consensus, observation of discussion,

discussion with or without revealing initial choices of members--

the scores of both groups and individuals tend to shift after the

experimental treatment. The shifts vary, of course, in magnitude.

Irrelevant group discussion and an irrelevant activity interposed

between the two administrations of the choice instrument have

failed to demonstrate a shift (Clark & Williams, 1970; Lamm & Kogan,


Despite consistent research findings, the practical significance of

the choice-shift may be open to question. A review of the literature

revealed the average shift per CDO item to be from six chances in ten

to five in ten (Cartwright, 1971). Its heuristic value suffers no

such limitations. As Cartwright points out: "A major reason for the

1The CDO asks subjects to serve as advisors to hypothetical
persons who are described as having to choose between two alternatives.
One alternative is attractive but less likely to succeed than the other
which is safe but not so attractive. Subjects give their advice on a
scale of odds; the recommendation is in terms of the minimum odds of
success they would require before suggesting that the less safe
alternative be chosen.

great interest in the risky shift is that it has stimulated

theoretical controversy" (1971, p. 361). Pruitt enlarges upon the

point: "If, as it now appears, there are many dimensions on which

groups differ from the average of the individuals who compose them

and if, in addition, social psychology can develop an efficient

theory about these differences, it will be a real accomplishment"

(1971, p. 340). The theoretical issues are still far from settled

(Myers & Lamm, 1976). At the heart of the controversy is the

question of the group processes involved. It has been firmly

established that group discussion is essential for the occurance of

shifts. How this process works is still under investigation.

Many researchers have advanced theoretical explanations for

the group-induced shift. The present research was not directed

toward explaining the phenomenon so only a listing will be provided

here for purposes of illustrating the diversity of the approaches:

Leadership Theory (Marquis, 1962); Relevant Arguments Theory

(Nordh0y, 1962); Diffusion of Responsibility Theory (Wallach, et al.,

1964); Social Comparison Theory (Brown, 1965); Familiarization

Theory (Bateson, 1966); Pluralistic Ignorance Theory (Levinger &

Schneider, 1969); Release Theory (Pruitt, 1969); and Committment

Theory (Moscovici & Zavalloni, 1969). The interested reader is

referred to the references noted and to the December, 1971 issue

of the Journal of Personality and Social Psychology for details of

these alternative explanations for the group-induced shift.



The study proposed here makes use of the group-induced shift

phenomenon. The focus, however, is not on risk-taking, but on

problem solving. One conclusion of the review of the relevant

literature presented in Section II was that solutions generated by

groups may or may not be superior to those made by individuals

depending upon the nature of the task, the composition of the

group, and the dynamnicsof the group interaction. The group-induced

shift research suggests that, whether they are better or not, group

solutions to certain kinds of problems are different from the

solutions made by the same persons acting as individuals. It also

suggests that, under certain conditions, individual solutions will be

different following group discussion from what they were before. The

present study investigated the possible extension of these findings

from the typical opinion problems of the group-induced shift re-

search to a complex, factual problem which has a correct (optimal)

solution. The group-induced shift paradigm was chosen for its

usefulness in exploring the two assumptions underlying organizational

use of problem-solvino groups; it allows for both a comparison of

individual and group decisions (quality) and for an estimation of

the extent to which the group solution represents individual con-

formity or compliance (acceptance).


The specific focus of the reported investigation was the

possible disproportionate influence of one member of a group upon

the solution to a problem developed by that group. The usual label

attached to this person is "Leader"; i.e., that person who has

relatively greater influence potential in a relationship

(Gold, 1958).

As discussed in a previous section, power arises from a

variety of sources. For present purposes, it is sufficient to

dichotomize these bases into legitimate and social (including reward,

coercive, referent, and expert); the two kinds of leaders

associated with these bases will be called "formal" and "emergent"


The effects of formal leaders upon group processes have been

studied fairly extensively. Hoffman provides a capsule summary of

this research: "The presence of authority relations in a group

seems to change the character of the discussion .. the group has to

spend considerable time either supporting or rejecting his views

rather than seeking for alternatives . Status differences affect

the behavior of those who are high in authority as well as those

with little power. Placing a person in a leadership position

generally influences his actions in the group" (1965, pp. 108-109).

With regard to the latter point, Meadow & Zander (1965) found the

high-status person in a group to be more involved with the group's

goals, to be more concerned that the group be correct, and to

perceive that he has greater influence over the group than other


The effect of formal authority relations within a group on the

nature of the group solution to a complex problem appears to be

fairly clear. Unless the leader takes steps to separate the

"discussion leading" function from the functions of contributing

and evaluating ideas, the group solution will be more like the

initial preferred solution of the leader than of the other members

(Torrance, 1955; French & Synder, 1959). The critical "steps" to

be taken include developing the "solution mindedness" (Maier, 1967)

of the group by encouraging the free expression of ideas, insisting

that minority viewpoints are heard, and discouraging the premature

evaluation of ideas (Maier & Salem, 1952; Hoffman, 1965; Janis,


Many problem-solving groups are formally leaderless in the

sense that no one person has been appointed or elected leader or

chairperson. In most such groups, a dominant individual emerges

who captures more than his share of the limelight (Hollander &

Julian, 1965). This individual is usually referred to as the

"emergent" leader. The literature on problem solving in groups

is largely silent on the subject of whether this type of leader

has the same kind of effects on a group-generated solution to a

problem as a formal leader. Suggestions are found, however, in a

1971 Risky-shift study by Boulanger and Fischer.

The Boulanger and Fischer Study

Boulanger and Fischer placed stooges instructed to adopt

either risky or conservative stands in 12 four-man discussion groups.

Stooges were empirically selected on the basis of performance in pre-

experimental sessions. The usual group-shift paradigm was

employed in these sessions (except that no post-test was administered),

but only risk-oriented CDO items were used (see Nordhty, 1962).

Groups were required to discuss these items until a unanimous

decision was reached on each. Two observers behind one-way mirrors

recorded the numberof suggestions made by each subject in the group

discussions. Following group decisions on the risk-oriented items,

subjects were individually asked a series of questions about the

group discussions. On the basis of answers to these questions and

observer records, leaders were identified.

The same subjects used in the earlier sessions were used in

Boulanger and Fischer's experimental sessions. Subjects identified

as leaders all agreed to serve as stooges and were briefed individually

on their roles. Six control groups in which identified leaders were

given no special instructions (but spoke individually with the

experimenter for the same length of time as other leaders) brought

the total number of groups in the study to 18. The experimental

sessions were identical to earlier sessions except that the five

conservative items from CDO were used and individual post-test measures

were also taken. Data were total scores over caution items. Results

showed a significant group shift on these caution-oriented items in the

risky direction for the risky-leader groups and significant shifts

in the cautious direction for the caution-leader and control groups.

The cautious shift held up for scores on individual post-tests while

subjects in the risky-leader groups reverted to original pre-group-

discussion decisions when responding to the questionnaire items

again individually.

Boulanger and Fischer interpret the results of their study

as support for the leadership explanation of the Risky-Shift. As

noted earlier, the present study is not concerned with explaining

the Risky Shift phenomenon. The finding from the Boulanger and

Fischer study of primary interest here is the differential effect

on individual post-test performance of the three sets of leader

instructions. The finding that risky-leader groups made significantly

more risky decisions on caution-oriented items than other groups,

but individual members of these groups reverted back to less risky

pre-test solutions when given the chance to do so (whereas other group

members did not) lends substantial support to the suggestion that

"unanimous" group solutions do not necessarily reflect the true

opinions of the individual group members.

The problemsin Boulanger and Fischer's study were conceived

for experimental purposes. They are heavily value-laden and their

correct solution is a matter of opinion. The extent to which the

findings of investigations employing the CDO, or variants thereof,

generalize to problems of a type of more interest to organizations

is unknown. The possible conceptual link between the two situations

has been suggested by Vinokur.

If for a given choice dilemma there is a pre-
ponderance of persuasive arguments for one
alternative, then individual and group shifts
in risk level are predicted to occur in the
direction of the alternative favored by the
arguments if group shift toward greater
risk (or conservatism) following persuasive
arguments is substituted for group movement
toward better problem solving, then the analogy
between the risky shift and group problem
solving becomes apparent. It is for future
research to determine whether the psycho-

logical mechanisms underlying these
phenomena are merely analogous or
in fact identical. (1971, p. 86)

A survey of the literature since 1971 indicates that Vinokur's

suggestion has been largely ignored.

Purpose of the Present Study

The present study investigated Vinokur's suggested analogy

between the Risky-shift and group problem solving. The methodology

was similar to that of Boulanger and Fischer, but was extended

to include investigation of influence exerted by non-leaders. In

addition, the investigation was moved from the laboratory to a

controlled field setting in which "real" rather than ad hoc

groups served as subjects. The specific research question

addressed was: What are the effects of deliberate attempts by

certain key individuals to influence their problem-solving groups

in a pre-determined direction on (a) the group solution (quality)

and (b) the post-group-discussion private solutions of individual

members (acceptance)? In its approach to this question, the present

study attempted to pull together a number of lines of investigation

and to fill certain gaps in the existing literature. Specifically,


a) employed the group-induced shift paradigm in a more

realistic problem-solving situation.

b) moved the group-induced shift paradigm into a field

setting employing real groups.

c) investigated the effects of emergent leaders, as com-

pared with formal leaders, on a group-produced problem


d) compared private solutions of subjects after group

discussion with decisions before. This comparison

is almost totally lacking in the literature, even

in the formal leader studies.



Subjects were students in two MBA (Master of Business Adminis-

tration) courses at a large urban university. These two courses

typically have an enrollment of approximately 40 students each.

Classes meettwo nights a week for two hours each session. Most

of the students are employed full time and most are considerably

older than the typical university day student. Thus, while subjects

were students, they were also members of organizations--and there-

fore members of the relevant population to which results of this

study might be generalized.

The MBA classes which made up the subject pool are taught on a

work group basis. The first two or three class meetings of the term

are devoted to a discussion of how the course will be conducted and

to introductory lecture material. At the end of these meetings,

students are divided into groups of varying size by the professor.

For the remainder of the term, group members gather material on

assigned topics outside of class and work as groups on this material

in class. The professor is available at all times as a resource.

Two professors have taught one section of these classes in the
manner outlined above for six years. Both report observing that a

Students of only one of these professors participated in the
study. This professor holds a Ph.D. in psychology.

leader will consistently emerge in these groups somewhere between

the second and third week of interaction and, with rare exceptions,

that person will continue to occupy the leader position for the

remainder of the term.

The existence of the particular type of classroom setting

described above provided a unique opportunity for responding to the

oft-cited need for experiments with "traditioned" groups (Lorge,

et al., 1958; Cartwright, 1973; Dion, et al., 1970). There are

two primary considerations behind calls for taking research into

group processes outside the lab. One lies in the group development

literature (see, for example, Bennis & Shepard, 1965) and relates

to the readiness of a group to confront a problem. "...groups are

sometimes handicapped...by their initial lack of organization. If

they are studied over a longer time period (and hence, given the

time to organize) or if allowed to attain organization by other

means, they will often make a better showing. This point also

argues for the desirability of studying existing groups, with relevant

prior experience in addition to the ad hoc groups generally employed"

(Kelley & Thibaut, 1969, p. 76).

The second primary reason for studying real groups has to do

with the question of the group processes in these groups as compared

with ad hoc groups. "Do the results obtained from 'artificial' groups

functioning in a laboratory have any relevance for natural groups

functioning 'out there' in the 'real world?' The argument of the

critics is that laboratory groups cannot hope to re-create the

richness of groups in the natural situations. Therefore, a principle

that operates in such ad hoc groups cannot be expected to operate

in ongoing groups exposed to the complex pattern of variables that

exists in the larger society. Furthermore, it is asserted that

members of such (ad hoc) groups are probably not motivated,that the

whole situation is regarded as a game..." (Shaw, 1976, p. 386).

The setting used in the present study was a compromise between

reality and the control afforded by the laboratory. Subjects were

students, but, as noted, they were also a subset of the relevant

"real world" population. The groups had "history, tradition, and

norms" (Hoffman, 1965, p. 126) to a degree substantially greater

than that of the typical laboratory group if less than that of a

very long term group. Similarly, while control in this situation

was less than in a laboratory, it was considerably greater than

would be possible in the typical organization setting. There are,

of course, certain demand characteristics associated with a class-

room. There is no reason to conclude, however, that these were any

more detrimental to this investigation than the demand characteristics

of the laboratory are to the laboratory experiment.

Given the nature of the present investigation, two variables

whose control was of particular concern were group cohesiveness and

the associated cooperation-competition dimension of group organization.

Group cohesiveness has been discussed at some length in a previous

section of the present paper (see pp.23 to 27). It was concluded

that a certain level of cohesiveness, or motivation to maintain the

group, is required at the outset if a group is to take on a problem-

solving task. As Kelley and Thibaut note: "This motivation to maintain

the group appears to rest primarily on the members' beliefs that

common interests are large and important in relation to individual

ones. In short, members perceive a sufficient basis of correspondence

to act in unison" (1969, p. 31).

In the present situation, the motive to keep the group intact

could be assumed to be operative. Individuals unhappy with the

class or with their groups dropped the course or changed groups

very early in the term. The groups as they existed at the time of

data collection had as goals the assimilation of course material

and the achievement of a good grade in the course. Given the

latter goal, it was also reasonable to assume that the groups were

organized cooperatively. A substantial part of the grade in the

courses used in the study is based on group performance. With respect

to this component of the overall evaluation, each member receives

the grade assigned to his group. Thus there is, in Thibaut &

Kelley's terms "perfect correspondence" of outcomes. It is, in this

sense, a pure cooperation situation. In reality, of course, the

situation may be somewhat of a mixed motive one. Within groups,

members may vie for acclaim, prestige, or prominence. The critical

point, however, is that external rewards are distributed exactly

evenly. The requirements that "common interests are large and

important relative to individual ones" was undoubtedly met to a

considerably greater degree than would be possible in the typical

organizational setting and motivation to perform well was almost

certainly considerably higher than in the typical laboratory setting.

The importance of the assumption that all members of experimental

groups be about equally concerned that the group product be superior

is illustrated by reference to the Cartwright and Zander analysis'

of power and influence in groups. As these authors note: "Most

theorists assume that influence should be viewed as a relationship

between two social entities such as individuals, roles, groups, or

nations" (1968, p. 215). The entity exerting the influence is

designated the agent and the entity over which the influence is

attempted, the target. In the present case, the agent is an

individual and the target is the group. The object of the influence

attempt is to bring about a particular group action. Unless it can

be assumed that members of the group have a common goal, the analysis

is meaningless.

In summary, the field setting chosen for this investigation

allowed for satisfactory control of certain important variables known

to influence group processes while also allowing for the investigation

of certain aspects of these processes in "traditioned" groups.

A total of 77 subjects in 16 groups participated in the study.

Group size ranged from four to six members. Subject age ranged from

22 to 44 with a median of 27. With respect to employment, 95% of the

subjects were employed full-time, 3% were employed part-time, and the

remaining 2% were full-time students. Thirteen, or 17% of the subjects

were female and two, or 12.5% of the identified group leaders were females.

Data were collected over three academic quarters.

Identification of Emergent Leaders

There were two sources of information as to the emergent leaders

of groups in the present study--professor observations and peer

nominations. Professor nominations based on observations were obtained

approximately half way into each term of data collection. A list of

the two individuals in each group felt by the professor to have had

the greatest general influence on group interaction to that point was

given to the experimenter at that time. Approximately one week later,

the professor used a few minutes of class time to ask each student

to rank the members of his or her group from "most generally

influential" to "least generally influential." Students were

specifically instructed to include themselves in the rankings and to

put themselves at the top if appropriate.

Gibb (1969) has stated: "There is good evidence that members

of a group can identify reliably those individuals who exert most

influence upon them and that leaders defined this way are closely

correlated with leaders identified by external observers and by

other criteria" (p. 211). A review of that evidence by Hollander

(1964) supports Gibb's contention both as to the reliability and

validity of peer nominations. In the present study, the various

rankings achieved by each group member were summed and that individual

within each group with the lowest total (i.e., highest ranking) was

designated "emergent leader." Ties or near-ties occurred in five

groups. In those cases, the names of the two individuals were written

on slips of paper and a colleague of the experimenter chose "the


The Professor nominations were used to check peer nominations: i.e.,

all individuals selected as emergent leaders were named by both peers

and professors. In two cases this required an adjustment in procedure.

In those instances, there was a tie in peer nominations, but only one

of the individuals had been identified by the professor as a leader.

To maintain the professor check on leader selection,that individual

was selected as the leader for purposes of the study.

Group Composition

As groups grow, the distribution of participation among the

members becomes severely skewed (Bales & Borgatta, 1955; Stephan &

Mishler, 1952). To increase the opportunities for all group members

to participate in group discussion as well as to be consistent, for

purposes of comparison, with the bulk of the small group literature,

group size in the current study was limited to four to six members.

This presented no problem at the lower end as groups in the classes

used are never smaller than four. However, they are frequently

larger than six--some have had as many as ten members. Therefore

experimental units had to be formed within the limits of on-going

work group size during the particular terms of data collection.

In addition to size, it was necessary for between-group

comparisons that the ability (as measured by initial DSP scores) com-

position of the experimental groups be roughly equivalent. Finally,

each group had to contain one, but only one, individual identified

as a leader.

The size, ability, and leadership requirements of group com-

position resulted in some few members being "left out" of some groups

as they existed in the classroom. In some cases, these were persons

whose initial DSP score was too far, in either direction, from

average performance. In others, they were simply individuals from


groups too large to be an experimental group. Finally, in those

groups which had more than one emergent leader, as defined by peer

and professor nominations, that individual not selected as the

emergent leader for purposes of the study fell into the "left over"

category. "Left over" group members from whatever category were

formed into "patch-up" groups for the experimental sessions. As

these groups consisted of members who did not have a history of

working together as an intact group, their scores were not used in

the analysis.

In summary, each experimental group consisted of all, or some

sub-set up to six, of the members of an on-going classroom work

group. Groups were roughly equivalent on ability to perform the

task as individuals; that is, they were formed so as to not differ

significantly on task pre-score means and standard deviations.

Finally, each group contained one, and only one, individual nominated

as the most influential in the group.


The Desert Survival Problem (DSP). The task chosen for use in

this investigation was the Desert Survival Problem (Lafferty, et al.,

1972). As the title implies, the Desert Survival Problem presents

subjects with a hypothetical situation in which they are the survivors

of an airplane crash in the desert. It is assumed that both the pilot

and the co-pilot were killed in the accident. A list of 15 items,

supposed to be objects salvaged from the wreckage, is provided. The

task is to rank these 15 items in order of importance for survival.

Time limits for accomplishing this ranking are at the discretion of the

experimenter or group leader and depend on the purpose for which the

the exercise is used. Suggested maximum times are ten minutes for an

individual ranking and 45 minutes for a group ranking and these time

limits were employed in the present study. The DSP score is determined

by calculating the absolute difference between subject ranking and

the correct ranking on each item and summing across all items. A perfect

score would be zero. The poorest possible score is 112.

The correct ranking of the DSP items is based on over 2,000 actual

cases in which individuals lost in the desert lived or died depending

upon the survival decisions they made. Alonzo W. Pond3 contributed the

expert rankings based on the real-life cases and on his own experiences

living for many years on deserts all over the world.

Scores for 2,620 individuals and 510 groups are reported in the

DSP manual. Of these 510 groups, group performance was better than the

average of the individual group members in 425 groups and poorer in 85.

Of the various categories of subjects for which data are available,

"mixed management" and college students are most relevant to subjects

in this study. The average individual score for "mixed management"

subjects is 61.11 and the average group score is 52.01 (156 groups).

For college students, the average individual score is 69.41 and the

average group is 66.67 (28 groups). The best group score ever reported

is 10.0 (junior high students).

The choice of the DSP for the group task in the present in-

vestigation was based on several considerations. (1) The DSP falls

into the class of problems most often delegated to organizational

3Former Chief of the Desert Branch of the Arctic, Desert, Tropic
Information Center of the Air Force University at Maxwell Air Force Base.

decision-making groups. As Hoffman notes, one of the forces behind

organizational requirements for group problem solving is that ". .. the

information needed for most management decisions must come from a

variety of sources whose functional interdependencerequires its

simultaneous consideration and evaluation by all concerned . ."

(1969, p. 100). That is, it is assumed that various members of the

decision-making body have different, but compatible, pieces of

information, or different amounts of the same information, and the

exchange of these pieces of information is essential for problem

solution. In Thibaut & Kelley's terms, this type of task has

conjunctive information distribution requirements (1965, p. 14) and

it is, therefore, also the type of task on which groups may be

expected to generate solutions superior to those of individuals. A

statement from the DSP manual makes its inclusion into this classifi-

cation of problems specific: ". . the Desert Survival Problem is

really a test of how well a team can gather and utilize the various

bits and pieces of information that each has on the subject"

(Lafferty, et al., 1972, p. 3).

(2) The DSP is based on "real life" and informal evidence from

individuals who have used it (including the author) suggests that it

is highly effective in stimulating subject interest and involvement.

Confidence in measures of performance on any experimental task is

increased if it can be assumed that subjects were interested in the

task and that they were trying to perform well. In the present case,

these assumptions are critical. The usual situation in the classrooms

used in this investigation is that students are very interested and

very involved with the activities in their groups. The present

investigation sought to accomplish an experimental manipulation in a

natural situation. Therefore it was essential that the task be

sufficiently interesting to maintain the usual level of activity.

(3) The DSP has a correct (optimal) solution. The use of raters

to evaluate the quality of a group decision is a standard practice in

group decision research. Unfortunately, even when interrater

reliability is quite high (e.g., .90 .95), there is still

10" to 20% disagreement. The DSP makes it unnecessary to introduce

such variance into the evaluation of group decisions. One score is

obtained and that score may be directly compared to other scores in

the same or other studies using the DSP without concern for the

standards used to determine the evaluation. Additionally, as discussed

in a previous section, verifiability of problem solutions also tends

to give groups an edge over individuals.

(4) The DSP is typically used in a manner which compares

individual scores before group discussion with group scores after

discussion. While the usual intent behind the DSP is to train groups

in effective decision making, not to demonstrate group shifts, the

data are still appropriate for comparison purposes. Such data are

especially useful in the present case as indicators of expected score

changes from individual to group administrations.

In summary, the DSP was selected as the task in this investigation

because it is conjunctive in nature, interesting to subjects, yields

an objective score, and provides normative data. There were three

administrations: Pre (individual), Group, and Post (individual). A

copy of the DSP is to be found in the Appendix.

Group Interaction Questionnaire (GIQ): In the present discussion

the point has been made repeatedly that one rationale for the use of

problem-solving groups in organizations is the supposed increased

acceptability of the problem solutions. In research in this area,

acceptance has been most often measured by subjects' responses to a

five- or six-point Likert-type question: "How satisfied are you with

the group's solution?" However, the correlations between such

satisfaction ratings and actual quality-of-solution scores tend to be

uniformly low (Hoffman, 1965), suggesting that these ratings may be

more influenced by freedom in the problem-solving process than by the

perceived quality of the solution.

The dimension of satisfaction with group solution of primary

interest in the study reported here was the extent to which subjects

accepted or rejected that solution in a subsequent private expression

of opinion as to the correct solution (the third administration).

However, in the interests of shedding additional light on that measure,

as well as for purposes of consistency with the literature, it was

felt that the more traditional measure should be included. In

response to the question raised by Hoffman, the usual single Likert-

type item was replaced by two such items--one dealing with satisfaction

with group interaction and one with confidence in the group's solution

to the problem. Together, these two items constitute the Group

Interaction Questionnaire, a copy of which is to be found in the


Semantic Differential (S-D): Likert-type measures of satisfaction,

such as those described above, have been criticized for both limited

reliability and the underlying unidimensional nature of the attitude

construct which is implied (Scott, 1967). "Factor analytic studies

of morale inventories have shown, if nothing else, that there are

discernable aspects of the work surroundings to which individuals

differentially respond"(Scott, 1967, p. 8). One approach to this

differential responding is the semantic differential technique

(Osgood, 1952). This technique uses bipolar adjectives set against

concepts; a subject differentiates the meaning of a concept by

placing a check mark at the point between the adjectives (7-point

scale) which best represents his reaction to the thing signified by

the concept.

Scott and his colleagues have factor-analyzed semantic

differential data from a large number of organizational settings

(Scott, 1967, private communication) for a variety of job-related

concepts. One of these concepts, "Me at Work," has consistently

yielded five primary factors: General Affective Tone (I), General

Vigor (II), General Emotionality (III), Personal Worth (IV), and

Personal Committment to Organizational Demands and Factors (V). The

"Me at Work" concept (re-labeled "How I Felt During the Group

Discussion") was included in the present study for purposes of

gaining information as to actual interest in the task (assumed, in

advanced, to be substantial) and reactions to the group interaction

processes. An established instrument, rather than a study-specific

one was used to facilitate scoring and interpretation. Scott's

subjects were similar enough to those used in the present study

(in that they were generally white-collar members of organizations)


to engender satisfactory confidence in the use of this questionnaire.

A copy of the "How I Felt During the Group Discussion" semantic

differential is to be found in the Appendix. It, together with the

GIQ, was administered once after the group discussion and solution

and before the final individual administration of the DSP.

In summary, dependent variable data available for each subject

in the present investigation were as follows:

a) Pre-group-discussion DSP score

b) Group DSP score

c) Post-group-discussion DSP score

d) Difference scores relative to a, b, and c above (b-a;
b-c; a-c)

e) Two scores from the GIQ

f) Five factor scores on the S-D

In addition to the nature of the experimental treatment (see

next section), three additional pieces of independent-variable type

information were available for each subject: Final course grade;

Score on Rotter's Locus of Control (I-E) Scale; and Level of

Confidence in initial (Pre-group) decision (6-point Likert-type

item ranging from zero--Very Uncertain--through five--Very Confident).

Experimental Manipulation

A basic purpose of the current study was to follow up the

Boulanger and Fischer finding that an emergent leader in an inter-

acting problem-solving group will influence the decision of that

group in the direction of his own predisposition. In that study,

the predisposition was an assigned risk or caution orientation. The

instructions given to the leaders were general "stance" instructions.

The present task is very different from the Choice Dilemma Questionnaire

used by Boulanger and Fischer. CDQ decisions are largely based on

values; the DSP solution depends on knowledge of facts about the

environment. The predisposition in the latter case will be in the

nature of a good solution to the problem or a poor one relative to

these facts. General "stance" instructions are inadequate in this

case. A good score on the DSP results from making the fundamental

decision to stay at the crash site and wait for rescue. A "stance"

instruction might be to "argue for a ranking of items based on the

decision to stay at the crash site." It is to be expected, however,

that individuals will vary widely on how this instruction is carried

out; that is, they will have very different ideas of what a ranking

based on the decision to stay at the crash site should be. The same

may be expected for a ranking based on the decision to leave the

crash site and attempt to walk to safety--a poor decision.

The information generated in the group discussion has been found

to be a major factor in group-induced shifts (Silverthorne, 1971).

To bring some of the potentially wide variation in information

resulting from general instructions under control, "dummy" rankings

were substituted for the original rankings of key individuals when

original rankings were returned to all subjects prior to the group

discussion (see Procedure). Instruction sheets were attached to all

individual rankings and, in the case of the dummy rankings, this

instruction sheet explained the nature and purpose of the substituted

ranking. (See Appendix for copies of all experimental instructions.)

This instruction sheet was substituted for Boulanger and Fischer's

pre-experimental-session instructions, as it was not to be expected

that working subjects would be able to get to class early for personal

instruction. A pilot study found this information manipulation to

be satisfactory. The professor reported that no key subjects raised

questions or appeared outwardly puzzled or distressed by the


As the dummy rankings used in this study constitute information

relevant to the task, it was necessary to provide a mechanism for

separating information effects from leadership effects. The question

is: Does the injection of information relative to a predisposition

into the group influence the group solution to a problem in the direction

of that predisposition regardless of who holds the information? In

half of the experimental groups, therefore, the dummy rankings were

given to individuals not selected as emergent leaders. Individuals

to whom dummy rankings were given will hereafter be referred to as

"key subjects." It should be kept in mind that one-half of the key

subjects were identified leaders and one-half were not.

Dummy DSP rankings were constructed so as to yield either good

or poor numerical scores equidistant from the pre-test average of the

80 individuals for whom pre-test data were available in the pilot

study. These scores (Good--58; Poor--84) were used as guidelines in

assigning experimental groups to either a "Good Information" (Good

Info) or "Poor Information (Poor Info) condition. Experimental

groups were assigned randomly to the "Leader" or "Non-Leader"

conditions, but within each of these conditions, the pre-test scores

of the key individuals were considered in assigning the groups to

either the Good Info or Poor Info condition. Key individuals who

were also Leaders received dummy rankings compatible with their own


initial pre-test scores; i.e., if they had a good score on the pre-

test, their group was assigned to the Good Info condition. If their

score on the pre-test was poor, their group was assigned to the

Poor Info condition. Key individuals who were not leaders were

selected so as to allow for a similar matching of original pre-

disposition and dummy rankings.

The procedure described above was felt to be necessary as the

entire thrust of this study is directed toward the real-life

analogy of a situation in which the emergent leader of a group is

committed prior to group discussion to a particular solution to a

problem. There is a considerable body of literature which suggests

that public behavior which is inconsistent with private beliefs

(in this case arguing for a ranking of DSP items dissimilar from one's

own ranking) has implications for both subsequent behavior and

subsequent expressions of opinion on the subject (Carlsmith, et al.,

1966; Helmreich & Collins, 1968; Hoyt, Henley, & Collins, 1972).

The extent to which leaders and/or non-leaders could influence group

decisions in a situation where they were arguing positions contrary

to their own beliefs is an interesting research question in its own

right, but in the present study counter-attitudinal arguments were



The procedure followed in the present study can be summarized as


1. The DSP was individually administered to all students in

the classes from which subjects were to be drawn at the beginning of

the academic term. (This was done by the professor in conjunction

with some other tests which are routinely administered each term at

that time.)

2. Professor and peer nominations for group leaders were

collected approximately one-half to three-quarters of the way into

the term (data collected by the professor).

3. Experimental groups were formed on paper (see Group

Composition and Experimental Manipulation). This was accomplished

by the experimenter who then made up packets of experimental

materials to be distributed by the professor at the time of the

experiment. The distribution process was arranged so as to keep the

professor "blind" as to the identity of the key individuals and the

experimental condition of each group.4

4. The experimental session was conducted. It was presented

as one of the usual class activities by the professor to maintain

the naturalness of the situation. The session took place approxi-

mately three weeks before the end of each term of data collection

and consisted of the following steps:

a. Pre-determined groups were formed in the classroom.

b. Original or dummy rankings were returned to subjects.

c. Subjects discussed the DSP in groups and arrived at a
group solution.

The professor was, however, aware of the general purpose and
hypotheses of the study as the use of his classes was made conditional
upon his being satisfied as to the adequate design of the study, its
value, and its usefulness as a learning tool for his students.

d. Subjects filled out the GIQ and S-D individually.

e. Subjects took the DSP again as individuals (groups
broken up).

f. Professor de-briefed subjects and lead discussion of
group problem-solving.

Figure 1 is useful as a picture of the experimental situation.

Hypotheses and Analyses

Data were collected at three times during this study. For

convenience, these points will be referred to as Pre-group, Group,

and Post-group. Experimental groups were formed so as to be equivalent

(in terms of initial DSP scores) at the Pre-group data point. The

primary focus of this study, then, was on changes from Pre-group

to Group (as an indicator of quality of group solution) and changes

from Group to Post-group (as an indicator of acceptance of solution).

With respect to changes from Pre-group to Group scores, the

suggestions of Vinokur (1971), Janis (1972), and Boulanger and

Fischer (1971) lead to the expectation that the emergent leader of a

group will exert greater influence on the group's solution to a

problem than will other members whether his/her preferred solution is

objectively better or worse than those of the other members. In the

present study's experimental Condition I, the identified emergent

leaders of the groups were given information which made "their" pre-

ferred solutions objectively better than those of other group

members. Therefore, it was expected that these groups would show the

greatest improvement from Pre-group to Group of all Conditions; that is,

good information in the hands of influential group members should

supplement the usual individual-to-group improvement on the DSP.

Quality of


Information Received By

Leader Non-Leader

Condition I Condition II

Condition III Condition IV

Figure 1

The four experimental conditions.

Experimental Condition III was the opposite of Condition I. In

these groups, emergent leaders were given information which made

"their" pre-group solutions objectively poorer than those of other

group members. Therefore, it was expected that these groups would

do less well on the Group administration of the DSP than on the

Pre-group administration; that is, the influence of the emergent

leaders in Condition III was expected to override the usual individual-

to-group improvement on the DSP.

In the remaining experimental Conditions (II and IV), the

emergent leaders of the groups were given no special information.

In these groups, leaders had their original rankings in front of them

during the group discussion of the DSP. The good or poor information

in these groups was given to non-leaders selected as described in

the "Method" section. As non-leaders were not expected to exert

disproportionate influence in the group, the expectation was that the

groups in these two Conditions would improve from Pre-group to

Group decision, as is typical on the DSP, but not as much as would

groups in Condition I.

Expectations as to changes in performance from Group to Post-

group administrations of the DSP in this study were based on the idea

presented earlier that subscription to the decision reached by a

group represents conformity in some cases and compliance in others.

Under the usual (non-manipulated) group problem-solving conditions,

these variables may be expected to be randomly distributed. Some

members are truly convinced of the validity of the group solution

while others have gone along with solutions which deviate (to differing

degrees and in differing directions for the various individuals) from

what they privately believe to be the best solution. In such

a situation, the expected value of the score of subsequent private

expressions of opinion is close to the group score (see Davis, 1973;

Muehleman, et al., 1976; Timmons, 1939). In the present study,

experimental Conditions II and IV were normal in that experimental

information was interjected into these groups by members whose in-

fluence was not typically greater or less than that of other group

members. In these Conditions it was expected that, consistent with

the literature, Post-group DSP scores would generally resemble Group


In Condition I, the identified emergent leaders argued for an

objectively good solution to the problem. As stated previously,

improvement from Pre-group to Group decision is typical on the DSP.

Boulanger and Fischer (1971) found that when the emergent leader of a

group argued in favor of the position which is the typical end

result of non-manipulated group discussion on the problem (in their

case, a cautious position on the caution-oriented CDQ items), sub-

sequent private expressions of opinion did not vary much from the

Group decisions. As the present study was designed to investigate

the extension of those authors' findings to a complex decision-making

problem, the expectation was (by analogy) that Post-group scores of

subjects in Condition I would be very close to Group scores (reflecting

conformity in the group decision process). In a similar manner,

Boulanger and Fischer found that, in groups where the leader had argued

for a position contrary to the typical end result of non-manipulated

group discussions (in their case, a risky position on the caution-

oriented CDQ items), subsequent private decisions reverted back in

the direction of Pre-group-discussion opinions. In the present

study, leaders in Condition III argued for a substantially poorer

solution to the problem than those initially made by the members of

their groups. Thus it was expected that, on the Post-group decision,

members of Condition III groups would revert back toward their Pre-

group-discussion scores (reflecting compliance in the group decision


Figure 2 illustrates the shifts which are expected to occur from

individual Pre-group solutions to Group solutions to individual

Post-group solutions for each of the four experimental Conditions.

Specifically, it was hypothesized that:

HI: Mean change socres from Pre-group to Group DSP solutions

will be positive in Conditions I, II, and IV (i.e., scores will

be better on the Group solution) and negative in Condition III

(i.e., scores will be poorer on the Group solution). Pre-group

to Group change score means for the four experimental Conditions

will order themselves as follows:

I (most improvement) > II and IV > III (deterioration)

H2: Mean change scores from Group to Post-group DSP solutions

will be positive in Condition III (scores will be better on the

Post-group solution). No change from Group to Post-group will

occur in Conditions I, II, and IV. Group to Post-group change

score means for the four experimental Conditions will order them-

selves as follows:

III (improvement) > I, II, and IV (no change)

Pre-group Group Post-group

Poor (high)





Good (low)

Figure 2

Expected score shifts for the four experimental conditions: I (Leader-
Good Info); II (Non-Leader-Poor Info): III (Leader-Poor Info); IV
(Non-Leader-Poor Info).

Raw data for testing Hypothesis 1 were algebraic differences

between each subject's initial DSP score and the score of the DSP

solution produced by that subject's group. For example, a subject

with a Pre-group score of 70 in a group whose Group score was 54,

would receive a change score of +16 (the plus sign indicating that

the subject improved from Pre-group to Group score). If his or her

Pre-group score had been 50, the change score would be -4 (the minus

sign indicating that the subject's Pre-group score had been better

than the Group score). The mean change score for each experimental

Condition was determined by dividing the algebraic sum of all

subject change scores in that Condition by the number of subjects.5

Duncan's New Multiple Range Test (Kirk, 1968, p. 93) was used

to analyze the significance of the differences between the mean

change scores in the four experimental Conditions. Hypothesis 1

predicts no difference between Conditions II and IV as there was no

basis for such a prediction. However, so as not to lose any infor-

mation about possible obtained differences between these Conditions,

all pair-wise mean comparisons were examined. Such an analysis is,

in fact, a combination of a priori, or planned comparisons (I vs.

II and IV vs. III) and a posteriori, or after-the-fact comparisons

(II vs. IV). Spearate multiple comparison procedures exist for the

two kinds of comparisons. It is also possible to use both a priori

5Consistent with the Boulanger and Fischer methodology, data
from key subjects (n = 16) were not used in any of the analyses.

and a posteriori procedures in the same experiment. However, it was

judged unnecessary to do so in the present case as the Duncan a

posteriori procedure provides the required information with only

a slight loss of power as compared with a priori tests.6

Raw data for testing Hypothesis 2 were algebraic differences

between a subject's Group score and his individual Post-group score.

Calculation of these figures and analysis of the data were accomplished

as described above. In both cases, where the direction of a difference

was hypothesized in advance, a one-tailed test of significance was

used. Other comparisons were subjected to a two-tailed test.

Alternative Hypotheses

The focus of this investigation was on score changes over three

trials on the same task. Hypotheses 1 and 2 state differences in

the patterns of these score changes which were predicted among the

four experimental Conditions. They are, then, hypotheses about the

relative positioning of the Conditions at the three times of data

collection (see Figure 2). There is, however, another approach which

could have been taken.

There exists in the group problem-solving literature a con-

siderable body of empirical work regarding the mathematics of the

relationship between individual scores on experimental tasks and group

6(Dietrich, 1977, private communication). No over-all F-test was
performed as such a test is not required when interest is in specific
predictions, not in whether or not anything happened in the experiment
(Kirk, 1968, p. 73). T-tests were not used as the comparisons to be made
were not orthogonal. The various other comparison procedures available
(LSD, Scheffe, etc.) were, for various reasons, less appropriate than
Duncan in the present case. The interested reader is referred to Kirk
for a complete discussion of the advantages, disadvantages, and appro-
priate usage of the most common multiple comparison procedures.

scores on the same or similar tasks. In the case of the Choice

Dilemma Questionnaire, such analyses have involved subject-by-subject

and item-by-item examinations of range, distribution, item inter-

correlations, shifts from pre-test to group test, and so on (see

for example, Burnstein, et al., 1971). This approach has served to

to direct attention to the prediction of group scores through the

utilization of various pieces of information about group members'

initial scores (see Davis, 1973). Much of this attention has been

focused on determination of the "best" model for predicting group

scores. A major issue has already been mentioned: Do group decisions

represent anything more than some combination of individual decisions?

The present study lends itself without much difficulty to the

alternative approach described above. For those who prefer it, or are

simply interested in making a comparison with the first approach, a

set of hypotheses addressed specifically to the issue of score pre-

diction has been developed. There is nothing new in these hypotheses.

They represent only another perspective on the same investigation.

From that perspective, there are five relevant pieces of numerical

information about each group in the present context which are avail-

able for predicting the group's score: (1) the score resulting from

the DSP ranking which the identified emergent leader has in front of

him/her during the group discussion (Leader Score), (2) the score of

the DSP which is manipulated; that is, that resulting from the ranking

which the key subject has in front of him/her during the group dis-

cussion (Information Score); (3) the best Pre-group score in the

group (Best Score); (4) the most extreme Pre-group score in the group

(Extreme Score); and (5) the average of the scores of the individual

group members on the Pre-group administration of the DSP (Average


I -

There are also several relevant pieces of numerical information

available for predicting individual Post-group scores: (1) through

(5) above, plus Group score, and the Pre-group score of the individual

subject (Pre Score). The theoretical considerations outlined in

connection with the first set of Hypotheses lead to the following

alternative hypotheses relative to the best predictors of Group and

individual Post-group DSP scores.

H3: The best predictor of the Group score will be the score

of the solution for which the identified emergent leader of

the group argues, whether or not he/she is the key subject in

the group.

H4: The best predictor of individual Post-group scores in

Conditions I, II and IV will be the Group score.

H5: The best predictor of individual Post-group scores in

Condition III will be individual Pre-group scores.

Correlation analysis was used to test Hypotheses 3 through 5.

Because the nature of the experimental manipulations created substantial

intercorrelations among some of the variables involved (see Table 1),

partial correlation coefficients were calculated (Nie, et al., 1976).

This procedure allows for an estimate of the relationship between two

variables when other correlated variables have been partialled out


7Blalock (1963) has argued that this method of dealing with the
problem of multicollinearity is inadequate due to the sensitivity of
partial correlation coefficients to sampling error. In the present
case, however, a means of estimating the accuracy of obtained coefficients
is provided by results from a completely different approach to the same

Table 1

correlation coefficients

among experimental variables.

3roup Score

Pre Score


Best Score

Extreme Score

average Score

_eader Score

Group Pre Information Best Extreme Average Leader

Group Pre Information
Score Score Score

-.17 .75










.39 .16 .17

.03 -.19


(N = 61)









Results and Discussion

Pre-group DSP score means and standard deviations for each of the

four groups within the four experimental Conditions are shown in

Table 2. T-tests (two-tailed with pooled error terms) of group mean

differences both within and between experimental Conditions revealed

no significant differences. As these tests were not orthogonal, they

were biased in favor of finding significant differences when, in

fact, none existed (Kirk, 1968). Given that no differences were found

with this liberal procedure, more conservative tests were not made.

The F-max test of group variance differences within and between

Conditions also revealed no significant differences. Therefore,

groups within the four experimental Conditions were considered to be

equivalent on Pre-group DSP scores for purposes of tests of the

stated Hypotheses.

Main Hypotheses

Over-all pattern of the results. Figure 3 is a graph of obtained

experimental data. A comparison with Figure 2 which shows the _re-

dicted pattern of results reveals strong correspondence between the

two so far as patterning is concerned. The departure of obtained

from predicted lies in the mean Group score for Condition II. It will

be recalled that the experimental information was given to non-leaders

in Conditions II and IV. That procedure was followed in order to

separate information effects, if any, from leadership effects, if any.

Obtained results clearly indicate that both effects were obtained, but

that information effects were assymetrical. Non-leaders with poor

Table 2

Pre-group DSP score means and standard deviations
in the four experimental conditions.


Condition I (Leader-Good Info)

Group 1
Group 2
Group 3
Group 4

Condition II (Non-leader-Good Info)

Group 1
Group 2
Group 3
Group 4

Condition III (Leader-Poor Info)

Group 1
Group 2
Group 3
Group 4

Condition IV (Non-leader-Poor Info)

Group 1
Group 2
Group 3
Group 4






































Obtained mean score shifts
three DSP administrations:
III (Leader Poor Info): IV

Figure 3
for the four experimental conditions over the
I (Leader Good Info); II (Non-leader Good Info);
(Non-leader Poor Info).

information (Condition IV) influenced their groups to make poorer

Group scores than groups with non-leaders with good information

(Condition II), but this detrimental influence was markedly less than

that exerted by leaders with poor information (Condition III). In the

good information Conditions (I and II), however, non-leaders were

about as successful as leaders in exerting a positive influence on

Group scores.

Pre-group to Group changes. From a preliminary consideration

of the obtained mean Group DSP scores, attention is now turned to the

mean change scores from Pre-group to Group DSP administrations which

are shown in Table 3. As predicted by Hypothesis 1, the mean change

scores are positive in Conditions I, II, and IV and negative in

Condition III. All means are different from zero at or beyond the .05

level of significance (one-tailed tests).

Results of the Duncan Test of the significance of the differences

between the means shown in Table 3 are presented in Table 4. As

predicted, the mean change in Condition I was significantly greater

than the mean change scores in Conditions III and IV (one-tailed tests).

Also, as predicted, mean change scores in Condition III were

significantly lower than those in all other Conditions (one-tailed tests).

The prediction that the mean change score in Condition I would be

greater than that in Condition II (which had not been predicted to

differ from Condition IV) was not supported (one-tailed test). The

mean Pre-group to Group change score in Condition 11 was significantly

different from that in Condition IV (two-tailed test) and not signifi-

cantly different from that in Condition I (one-tailed test). Thus,

Table 3

Mean change scores from pre-group to group
administration of the DSP by experimental condition.

*p < .05 (t-test of Ho: ij = 0; one-tailed-test)

Condition I (Leader-Good Info) 25.67*

Condition II (Non-leader-Good Info) 19.25*

Condition III (Leader-Poor Info) 4.07*

Condition IV (Non-leader-Poor Info) 3.63*

Table 4

Duncan New Multiple Range Test of differences
among the mean pre-group to group change

*p < .05 (test of Ho: pa = ib; one-tailed test)
**p < .05 (test of Ho: -a = ub; two-tailed test)

X3 Xq X2 X1
x3 4 2 x1
(-4.07) (3.63) (19.25) (25.67)

Condition III: X3 = -4.07
(Leader-Poor Info) 7.7* 23.32* 29.74*

Condition IV: X4 = 3.63
(Non-leader-Poor Info) 15.62** 22.04*

Condition II: X2 = 19.25
(Non-leader-Good Info) 6.42

Condition I: X1 = 25.67
(Leader-Good Info)

while the predicted mean ordering among the Conditions of Hypothesis 1

was obtained, the location of the "greater than" signs was shifted.

Actual results were:

I and II (most improvement) > IV > III (deterioration)

Group to Post-group changes. Obtained mean change scores from

Group to Post-group DSP administrations for the four experimental

Conditions are shown in Table 5. As predicted by Hypothesis 2, the

mean for Condition III is positive. It is different from zero at

the .05 level of significance (one-tailed test). Mean change scores

for Conditions I, II, and IV range from slightly positive to moderately

negative. As expected, none of these means is significantly different

from zero (two-tailed tests).

Results of the Duncan Test of the significance of the differences

between the means shown in Table 5 are presented in Table 6. The

prediction that the mean Group to Post-group change score in

Condition III would be significantly greater than those in Conditions

I, II, and IV, which, in turn, would not be significantly different

from one another, was fully supported.

Alternative Hypotheses

Partial correlation coefficients between Group DSP scores and the

various predictors are presented in Table 7. Hypothesis 3 predicted

that the best predictor of the Group score would be the Leader Score.

The highest coefficient actually obtained was between Information Score

and Group Score, but this coefficient is not significantly different

from that obtained between Group Score and Leader Score (test after

Blalock, 1972).

Table 5

Mean change scores from group to post-group
administrations of the DSP by experimental condition.

*p < .05 (t-test of Ho: p = o; one-tailed test)

Condition I (Leader-Good Info) 1.53

Condition II (Non-leader-Good Info) 2.63

Condition III (Leader-Poor Info) 6.21*

Condition IV (Non-leader-Poor Info) .75

Table 6

Duncan New Multiple Range Test of differences
among the mean group to post-group change

X2 X1 X4 X3
(-2.63) (-1.53) (.75) (6.21)

Condition II: = 2.63
(Non-leader-Good Info) 1.10 3.38 8.84*

Condition I: X1 = 1.53
(Leader-Good Info) 2.28 7.74*

Condition IV: X4 = .75
(Non-leader-Poor Info) 5.46*

Condition III: X3 = 6.21
(Leader-Poor Info)

* p < .05 (test of Ho: pa = Pb; one-tailed test)

Table 7

Partial correlation coefficients: Group DSP scores with
each predictor variable, controlling for all other
variables (n=16).

*P < .05 (test of Ho: y = 0)

**p < .01 (test of Ho: y = 0)

Predictor Variable

Leader Information Best Extreme Average
Score Score Score Score Score

Dependent Variable

Group Score .58* .62** -.35 .17 -.21

Partial correlation coefficients between individual Post-group

DSP scores and various predictors for Conditions I, II, and IV are

shown in Table 8.8 It can be seen that, consistent with Hypothesis 4,

the highest coefficient obtained was between Post-group and Group


Obtained partial correlation coefficients between individual

Post-group DSP scores and the various predictors for Condition III

are presented in Table 9. Despite the sharply reduced degrees of

freedom when this Condition is analyzed alone, the Hypothesis 5

prediction that Pre Score would be the best predictor of Post-group

Score in this Condition is supported.

In general, results of the analyses of Hypotheses 1 and 2 and

results of the analyses of Alternative Hypotheses 3, 4, and 5 are

consistent. Combining the data from all Conditions, as was done

for the correlational analysis of Hypothesis 3, revealed Leader Score

and Information Score to be the two most powerful predictors of

Group Score. The Condition by Condition analysis of Hypothesis 1

suggests that this finding results from a strong, but assymetrical,

information effect (obtained in addition to a leadership effect).

Results of the analysis of Hypothesis 2 are supported by the

correlational analyses of Hypotheses 4 and 5: Subjects in Condition

Hypothesis 4 was based on the expectation that the forces acting
in these three Conditions would be the same at this stage, regardless of
the nature of the information which had been interjected into the groups.
Therefore, these Conditions were combined for this analysis.

9Therefore, with respect to Blalock's caution, it is noted that this
correspondence tends to increase confidence in the obtained partial cor-
relation coefficients. Without such a cross-check, Blalock's argument
would suggest that the obtained coefficients could not be interpreted.

Table 8

Partial correlation coefficients: Post-group DSP scores
with each predictor variable, controlling for all other
variables. Experimental conditions I, II and IV (n = 47)

* p < .01 (test of Ho: y = 0)

Predictor Variable

Group Leader Information Best Extreme Average Pre
Score Score Score Score Score Score Score

Dependent Variable

Post-group Score .53* .20 .21 .19 .12 -.04 .26

Table 9

Partial correlation coefficients: Post-group DSP scores with
each predictor variable, controlling for all other variables.
Experimental condition III (n = 14)

Predictor Variables

Group Leader Information Best Extreme Average Pre
Score Score Score Score Score Score Score

.00 **

*p < .01 (test of Ho:

.00 .00 .00 .64*

y = 0)

**no variance on this variable



III tended to revert back toward their Pre-group scores on the Post-

group administration of the DSP and subjects in the other Conditions

had Post-group DSP scores which were very close to the scores which

their groups had achieved.

Other Analyses

In addition to analyses of the relationships predicted by the

hypotheses, a number of exploratory or "data snooping" analyses

utilizing the other information which was available were performed.

Locus of Control (I-E). Several analyses were carried out to

follow up suggestions in the literature as to differences between

individualswhom Rotter (1966) labeled "Internalizers" and those he

labeled "Externalizers" (see p. 34, this paper). 0 Ryckman, et al.,

(1971), for example, found that Internalizers are generally more

confident about their own performance than are Externalizers. No

difference between the two groups on Pre-group DSP solution confidence

expression was found in the present study by any method of analysis.

Sherman (1973) reported Externalizers to be more influenced by

persuasive messages than Internalizers. No such over-all relation-

ship was found in the present study. I-E scores did not correlate

significantly with any measures of change: Pre-group Group (r = .04),

Group Post-group (r = .08), or Pre-group Post-group (r = .07).

However, closer examination of these data yielded results of consider-

able interest. These results are described below.

10In all cases, Internalizers were arbitrarily defined as individuals
scoring below the mean I-E score for all subjects in the study and
Externalizers as individuals scoring above the mean.

Ritchie and Phares (1969) found that Externalizers changed more

when the influence attempt came from a high-prestige source than

from a low-prestige source. Internalizers, on the other hand, were

more responsive to the content of the communication and changed

equal amounts whatever the source. Further, Kelley and Thibaut

(1969) report that Internalizers more generally take an information

orientation toward tasks, are more critical of the information put

forward, and are less attentive to the hedonistic aspects of the

group interaction than are Externalizers. Combining the Ritchie

and Phares and the Kelley and Thibaut reports and formulating them

in terms of the present study, leads to the following expectations

about the change performance of those who scored high on the I-E scale

(Externalizers) versus those who did not (Internalizers).

1. Change scores for Externalizers will be higher in the two

high-prestige (Leader) Conditions (I and III) than in the two low-

prestige (Non-leader) Conditions (II and IV).

2. Change scores for Internalizers will be higher in the two

Good Information Conditions (I and II), regardless of the source of

the information, than in the Poor Information Conditions (III and IV).

A rough test of these expectations was carried out using change

scores from Pre-group to Post-group (selected as the two data points

at which individuals should have had the freest opportunity to express

their real opinions). Results of the analysis are presented in

Tables 10 and 11. Despite the crude definition of the two groups,

results are consistent with the expectations and are significant


Table 10

Mean pre-group to post-group DSP change scores for Internalizers and
Externalizers by source of information.


Externalizers 15.1

Internalizers 17.1

*P < .025 (t-test of Ho: Pa




= "b; two-tailed test)

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

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