JOURNAL
OF EXPERIMENT.tL
Experiments ALEX
SOCIAL
PSYCHOLOGY
1,
on the Alteration BAVELAS,
55-70
of Group
ALBERT H. HASTORF, AND W. RICHARD KITE Stanford
(1965)
ALAN
Structure’ E. GROSS,
University
A fundamental problem in social psychology is the relationship between an individual’s behavior and how that behavior is perceived and evaluated by others. For example, there has been persistent interest in the process which generates a status hierarchy in small face-to-face groups such that some members are perceived as Yeaders” and others not. Most of the research on this question has attempted to chart the course of a naturally emerging structure, sociometrically define the leader, and then attempt to define those aspects of his behavior that led to his being perceived as the leader. However, the behavior of the leader is normally so complex that it has been exceedingly difficult to isolate the behaviors that significantly influence the perceptions of the other group members. Bales (1950) has explored this approach most thoroughly, with one of the most persistent findings being that the people seen as leaders talk a great deal. This report will describe a series of studies in which an attempt is made to alter experimentally the verbal behavior of an individual in a group discussion by the use of an operant conditioning procedure. Our primary concern was to develop a workable procedure for increasing one group member’s verbal output, to define some of the variables which appear crucial to this procedure, and to explore the other group members’ percept,ions of this change in behavior on such dimensions as quality of ideas and leadership.” Previous research with operant conditioning techniques in group situations has most commonly made use of confederates. Pepinsky, Hemphill, and Shevitz (1958) demonstrated that “accepting” or “rejecting” reactions on the part of confederates influenced the number of leadership attempts made by a naive subject’. By making a straightforward application of a standard verbal conditioning procedure to a group situation, Bachrach, Candland, and Gibson (1961) have shown that the verbal ‘This research was supported in part by National Science Foundation grant 24137. *Some of these issues are discussed in a preliminary report of this research ( Hastorf, 1964) 55
56
BAVELAS,
HASTORF,
GROSS,
AND
KITE
output of a naive group member can be increased by the headnods, “umm humms,” and agreements of two confederates. It should be noted that in both of these studies, the “group” aspects of the experimental situations were severely attenuated by the use of confederates. In such an experimental set-up it, is impossible to obtain data on other group members’ perceptions of the ‘(target” subject whose behavior was being altered. Oakes, Droge, and August (1960) demonstrated that verbal behavior can be either increased or decreased by the use of lights as reinforcers or punishers. Aiken (1964) has described a similar procedure in which lights as reinforcers or punishers were used for subjects in a group situation where each subject is provided with private feedback on his performance. This procedure has the significant advantage of bringing about a change in the verbal behavi’or of a subject in the presence of other subjects who are unaware of the exact nature of the reinforcements given the “target” person. The studies reported below are directed toward answering the following questions. When lights are used as signals or reinforcers in a group situation, h’ow much change in verbal behavior can be obtained? Must the reinforcing lights be directly contingent on talking or will a random pattern of lights also increase verbal output in a group discussion atmosphere? If an increase in the verbal output of a group member is obtained, will that person maintain his new verbal level in a following session where no lights are expected? Finally, how do the other group members evaluate the contributions of the “target” person? Do they increase his status on such dimensions as quality of ideas and leadership? EXPERIMENT
I
Method Subjects. Seventy-two male students from industrial psychology and industrial engineering classes at Stanford University were recruited “to participate in group discussions of case problems.” The 8s were divided into eighteen four-man groups, half of which were assigned to the Experimental condition and half to the Control condition. Group members were not well acquainted with each other prior to the experimental session. Apparatus. Each of four positions at a discussion table was equipped with a reflector box which contained two small lights, one green and one red. These boxes were flared toward each participant so that only he could see the lights facing his position. The lights were controlled from an observation room which was separated from the discussion room by one-way glass. Clocks and counters were used to record talking time and frequency for each S. Whenever an S talked, or whenever a red or green light was turned on, an Esterline-Angus pen recorder was activated, thereby providing a sequential event record.
ALTERATION
OF
GROUP
STRUCTURE
57
Procedure. The Ss were told that the discussions would be observed and recorded from behind the one-way glass. The E explained that he was interested in the study of group discussion techniques from an educational viewpoint. It was further explained that several different human relations problems would be discussed so that the dynamics of the group discussion process could be analyzed. Following these brief and purposely vague introductory remarks, the four Ss read the first case problem and were instructed to begin a l&minute discussion period during which they “should discuss the pertinent facts which will affect a decision.” This initial discussion was intended to provide an operant level or baseline measurement of verbal activity. An observer operated the clocks and counters, which provided, respectively, a record of cumulative talking time and a record of the total number of times each S talked. This recording procedure was also followed in the two subsequent discussions. At the end of each discussion period, E re-entered the discussion room and administered a short sociometric questionnaire. The Ss were required to rank all group members, including themselves, on four key items: amount of participation, quality of ideas, effectiveness in guiding the discussion, and general leadership ability. After reading the second case problem, the groups were given further oral instructions. The E stated that in contrast to the usual nonfeedback procedure, group discussions might be more effective “if the participants are given an occasional sign that they are doing the kinds of things that will help the group arrive at intelligent solutions while at the same time yielding the maximum educational benefit to the group.” The Ss were then told that some discussion groups work on their own and some groups “are provided with feedback information as to how they are doing as the discussion proceeds.” At this point the experimental groups were told that they would be receiving feedback. The E directed attention to the small red and green lights which had previously been dismissed as extraneous equipment that “we won’t be using now,” and told the group that these lights would serve as the source of the feedback information. Control groups were told that they would receive no feedback. Both Experimental and Control groups received vague descriptions of the criteria that were to be used in evaluating their discussion:
Many psychologists have studied group discussion of problems such as this one. Most of these investigators have found that maximum benefit is gained from such discussions when the group proceeds in an orderly way through various stages of development. For instance, one psychologist has cautioned against proceeding into the problem-solving stage of discussion too rapidly before there has been enough orientation. Other research has given us clues as to the value of cooperation, suggestion, conciliation, and other forms of group behavior during certain stages of the discussion process. The case you have just read has been thoroughly analyzed in terms of how it can be discussed most effectively. Although, of course, there is no single correct solution to the problem, we have developed a definite set of principles, such as those I have just mentioned, which enable us to know whether or not you are following the best course; that is, using the best techniques in contributing to the discussion for the benefit of the group. Note that although this will be a group discussion, it will be your individual contributions to the discussion that will be judged.
58
IMVELAS,
MASTORE‘,
GROSS,
AND
KITE
At t,his point the Control groups were reminded that, they ~vould not rrreivr an?. feedback as to how they were being evaluated. Hower~r, thr Experimental groul~ were told t,hat they would rcreivc feedback which would bc continpcnt upon the’ vaguely defined quality of their contributions: Thus, whenever you make a contribution to the discussion which is helpful 01’ functional in facilita.ting the group process, your green light will go on like thi:: (green lights turned on). Are all your green lights on? Fine. Whenever you hf,hnvc’ in a way which will eventually hamper or hinder the group process, your rrd light will go on like this (red lights turned on). Are all your red lights on? Goocl. It is conceivable that even remaining silent when you might have been clarifying :I point that had been made earlier is a dysfunctional or hindering type of behavior. This would rate a red light indicating you should have said something at that point. On the other hand, silence might be good when talking would serve to confuse a good point that had already been made. Since it is often impossible for us to determine the effect that a single statement or thought will have on the group discussion, a feedback light might be referring to the cumulative effect of two or three successive contributions to the discussion. Of course much of the time neither of your lights will be on, indicating that your behavior has been neither helping nor hindering the group, or that we simply can’t validly analyze what has been going on in the group at that time. Note that the discussion table is constructed so that each participant can see only the lights directly in front of him. During the discussion the fact that your lights are either on or off should not be mentioned. This would, of course, tend to disrupt the natural discussion atmosphere. Group members in both conditions were told the discussion would last for 20 minutes and that they should “try to bring in the various possible facts that can be considered relevant to the problem in the case.” At the end of the first discussion period, the Ss were rank-ordered on the basis of objective behavior measured by the clocks and counters and on the basis of the perceptions of the group members measured by the responses to the four sociometric items. There was usually a close correspondence between amount of talking and average ranking on the four sociometric items. In the few cases in which these measures were inconsistent or contradicted each other, the ranking adopted was made on the basis of total talking times. The third or fourth ranked man in this hierarchy was designated as the target person (TP). One of the less talkative men was selected so that here would be “room” to effect a relatively large change in verbal behavior. An exceedingly quiet man was not selected as the TP because it was felt that if he had been extremely quiet it would be difficult to alter his behavior. The man who was ranked first was designated M-l; the other two men were labeled M-2 and M-3, respectively. The experimental manipulation consisted of flashing Z’P’s green light whenever he made declarative statements or stated an opinion, and flashing the others’ red lights if they (M-l, M-2, M-3) engaged in these same behaviors. Occasionally, TP received a red light for remaining silent, and other group members received green lights for interacting with TP, especially for agreeing with him. No definite criteria or set of rules were followed for administering lights or for controlling the number of lights distributed. The light operator’s task was to increase the TP’s verbal output during the second discussion and to decrease or inhibit talking by the other group members. The operator was to select for reinforcement
ALTERATION
OF GROUP
STRUCTURE
59
those statements by the TP which would intuitively appear to result in increased sociometric status. After the second discussion, the Ss again completed the sociometry questionnaire and then read the third case problem. The Experimental groups were told that the lights would “not be operating-just as in the first case discussion; so don’t pay any attention to the equipment.” At the end of this final lo-minute discussion, the last sociometric questionnaire was administered, followed by a postsession questionnaire. Each S was asked to rank the three case problems from most to least liked and to indicate whether he felt that he had talked more, less, or about the same as usual during each of three discussions. Experimental Ss responded to items which asked how much attention was paid to the lights and whether the lights were perceived as helping or hindering the discussion. After this questionnaire was completed, the purpose and design of the experiment was fully explained to the Ss, and questions were answered.
Results The experimental procedure is clearly effective in altering both the distribution of verbal outputs and the sociometric &ructure of the group. In all nine Experimental groups, TP’s talking time and frequency of talking increased during the second discussion when lights were being used. Furthermore, this change was strongly reflected in the sociometric votes of the other group members: all nine Experimental TPs received higher average rankings after the second discussion than after the first discussion (Table 1). Frequency of talking data, which are correlated with total talking time (r = .91) are omitted from Table 1. Since the rankings for guidance, best ideas, participation, and leadership turned out to be highly correlated, t,he sociometric data is reported as the mean ranking of these four items. The TPs’ self-rankings are excluded from these averages. In eight of nine cases TPs’ ratings drop somewhat following the discussion of the third case, but in only one group do the ratings the TP receives drop back below the baseline level of the first period. Increased sociometric ratings and length of talking time in the second period for Experimental TPs are significantly greater than the slight increase shown for Control group TPs. The drop-off from the second to third discussions is significant for the objective talking measures and the sociometric data. Despite the drop-off, TPs’ level of output and sociometric ratings remain significantly higher than they were after the first discussion. The TPs’ perception of their own behavior as reflected in their selfratings followed a similar pattern. Their self-ratings also rose after the second period and then dropped off somewhat when the lights were not used in the third period. The TPs’ self-ratings are presented in Table 2. Although mean self-ratings for each discussion period are slightly higher
TABLE SOCIOMETRIC
RANK
10 OUTPUT
AND VERBAL
Mean rankings received by 7’P from other group members (1 to 4)
OF TARGET
Time talked by TP expressed :LSa percentage of total group talking time
Discussion period Experiment Control (N = 9 groups)
PERSON
Discussion period
1
2
3
1
*,
3
3.05
2.81
2.80
17.3
20.2
19.5
Exp. I (N = 9 groups)
3.23 1.70 lvs.2&3,p< .Ol 2vs.3,~ < .05
2.30
15.7 37.0 1 vs. 2 & 3, p < .Ol 2 vs. 3, p < .Ol
26.!)
Exp. II (N = 7 groups)
3.18 2.13 1 vs. 2 & 3, p < .02
2.36
17.4 31.1 lvs.2,p<.O5 lvs.3,p= .02
29.0
Exp. III (N = 7 groups)
3.12
2.80
2.75
19.8
20.4
20.9
Exp. IV (N = 7 groups)
3.24
2.95
3.11
20.9
22.2
18.6
Exp. V (N = 7 groups)
3.08
2.66
2.82
19.3
24.3
22.2
5 Only p values of less than 0.10 are indicated. All p values are two-tailed. Differences for experimental groups between first and second discussions were compared with corresponding differences for control groups by the Mann-Whitney Test. Significance levels for differences between the second and third discussions and first and third discussions were computed by the sign test. TABLE MEAN
SELF-RANKINGS
2 OF TARGET
PERSON
Discussion period Experiment Control I
II
1
2.67 2.91 lvs.2,p 2~3.3,~ 2.79 1 vs.3,p
III IV V
2
3
2.31 1.36
2.39 2.05
1.6-L
1.64
2.64 2.14 2.07
2.79 2.50 2.39
< .02 < .05 < .05
2.93 3.04 2.68 60
Others
TP
Others
TP
Others
Others TP
TP
Others
Others TP
TP
9
21
7
21
7
21
21 7
7
27
27 9
N
1.85
2.14
2.17
2.17
2.14
1.86 2.29
2.43
2.15
2.30 2.56
2.56
Case 1
2.29
2.14
2.17
1.50
2.05
2.67 2.14
1.86
2.07
1.81 1.00
1.56
Case 2
a Did not receive feedback. b One group did not complete postsession questionnaires.
V
IV
III
II
I
Control
Experiment
3
.oo
+0.14
1.71 1.86
-0.22
-0.50
-0.04
+0.19 -0.29
-0.43
+0.04
-0.04 -0.33
-0.22
Period 1
1.67
2.33
1.81
1.48 1.57
1.71
1.78
1.89 2.44
1.89
Case 3
DATA
-0.29
-0.29
-0.05
+0.33
+0.07
-0.48 -0.29
$0.43
-0.37
+0.07 +0.67
+0.33
Period 2
-0.10
-0.14
-0.05
.oo
-0.04
+0.05 +0.29
+0.57
+0.07
-0.04 +0.22
-0.11
Period 3
Estimate of talking (1 = more, 0 = same, - 1 = less)
QUESTIONNAIRE
TABLE
Case preference (1 = most liked, 3 = least liked)
POSTSESSION
4
-0
-0
4b
3
7 1
4
8
-(I 8
-(I
Help
7
-a
-0
2
12
4 4
3
9
-a 0
-a
No diff.
1
10
-(I
-a
0
6
10 2
0
10
-0
-0
Hinder
Influence of lights (Frequencies)
3 3 si
s 0 g $ z
Ei * :! !2
i?
62
BAVELAS,
HASTORF,
GROSS,
AND
KITE
than ratings received from others, the magnitude of change between discussions is very similar to changes in ratings made by others. Data from the postsession questionnaire are presented in Table 3. Theac~ data indicate: (1) a strong liking for the second case lnoblem among Experimental TPs, while no such consistent preference exists for any one case among the other group members; (2) TPs were aware that they talked a great deal during the second discussion: 6 of 9 felt that t,hey talked “more than usual” during this period while all but one of the other group members felt that they had talked “less than usual” or “about the same”; (3) 8 of 9 TPs responded that the feedback lights had “helped” them during the second discussion while the ot,hcrs were split bet,ween feelings of being “helped” and being “hindered.” EXPERIMENTS
II
AND
III
In Experiment I red and green lights were distributed to the Ss on the basis of the Es’ intuitive judgments as to what behaviors should be encouraged or discouraged. It was thought t,hat such a procedure would be most effective in bringing about the desired changes in the behavior of the 5’s. When the effect,iveness of this procedure was demonstrated, the question arose as to how crucial were the Es’ choices of which behaviors to reinforce in producing the observed effect. Did the TP’s increase in sociometric status result from an increase in certain categories of verbal output, or was it, simply the result of his talking more in general? Experiments II and III represent an attempt to answer this question by eliminating the Es’ judgments from the administering of the lights. Fift,y-six undergraduate Xs were assigned to 14 four-man groups. Seven groups were run in each experiment.
Experiment
II
Procedure. Experiment II was an exact replication of Esperiment I in terms of the general procedure followed. The only modification was in the manner in which the red and green lights were administered to the Ss during their second case discussion. All “feedback” lights were administered automatically by a preprogrammed event-controlling unit which was activated by the same switches used to record the Ss verbal output. In this way it was possible to make the lights contingent upon the Ss’ talking without regard to its content. The program unit was connected to a 25position stepping switch which moved to the next position with the recording of each discrete utterance by any of the Ss. Each S had a fixed sequence of 25 events programmed for him: he could receive a green light, a red light, or no lights each time he talked. TP received a “leadership encouraging” schedule of 15 green lights with the remaining 10 positions blank. A separate timing device was connected to the TP’s circuit which delivered a red light to him for every 45 seconds of continuous silence. The other three Ss received schedules consisting of 7 red lights, 2 green identical “followership encouraging” lights, and 16 blanks. These two schedules were intended to approximate the
;ZLTERATION
OF GROUP
63
STRUCTURE
schedules administered in Experiment I in terms of both absolute number of lights and ratios of red to green. Instructions regarding the onset of the “feedback lights” prior to the second discussion were the same as in Experiment I. Subjects also filled out the same sociometric questionnaire after each of the three discussions and the same postexperimental questionnaire as were administered in Experiment I.
Results. Although it was the intention of the Es to provide Xs with schedules of red and green lights which at least approximated those received by their counterparts in Experiment I, this objective was not achieved. The actual mean numbers of lights received by the Ss in both experiments are shown in Table 4. It is obvious that all Ss in Experiment II received considerably fewer lights of both kinds than did the comparable Ss in Experiment I. NUMBER Experiment Green
TABLE -l OF LIGHTS RECEIVED
1’P
M-l
M-.2
M-S
38.1
15.2
15.1
13.8
7.6
7.8
9.6
9.3
15.4
1.7
1.7
0.4
9.7
6.0
3.3
I Red Green IL Red Green
5 .3 38
15
16
14
7
8
10
9
-
-
-
-
-
7 .!)
7 .9
5.1
6.4
7.7
5.6
III Red Green
14.0
IV Red
5.0
Green V Red
-
This deficit appears to have been mainly due to the fact that the programmed schedules were simply too sparse, particularly in terms of green lights. There is also, however, a methodological problem inherent in the design of this study. In order for the TP to receive green lights, it was necessary for him to talk. If the programmed schedule of green lights was insufficient to produce a sizeable increase in his talking over time, then this placed an obvious limitation on the number of green lights he would receive over the course of the entire discussion.
64
BAVELAS,
HASTORF,
tiROSS,
Ah’D
IWl’k:
The major results of Experiment II are presented in Table 1. When compared with TPs in the Control groups, it can be seen that even making the green lights contingent upon sheer talking produces an increase in both the sociometric rankings received from others and verbal output. These increases are not, however, as great as those obtained in Experiment I. From the first to the second discussion the TP shows a significant increase in sociometric rankings received from others and a significant increase in talking time. These gains in sociometry and talking time are maint,ained by the TP to a significant extent t,hroughout the t,hird discussion. Unfortunately, the fact that fewer lights were given to the Ss in Experiment II than to the Ss in Experiment I makes it impossible to give an unequivocal answer to the question of how much the Es’ judgments contributed to the over-all effect. It does appear safe to conclude, however, that such judgments as were being made in Experiment I were at least not indispensible in producing significant changes in both verbal behavior and sociometric rankings.
Experiment
III
Procedure. As a second method of eliminating the Es' judgments from the administration of lights, seven groups of 8s received red and green lights on a timecontingency basis. A leadership schedule was derived by averaging the number of red and green lights received by TPs in Experiment I within each successive &minute interval and then distributing this number randomly over an equal period of time. In a like manner a different schedule was derived for M-l, M-2, and M-3 on the basis of the average number of lights received by their counterparts in Experiment I. These schedules were administered manually by a single E who viewed a large clock with a sweep-second hand in conjunction with the four schedules written out on large sheets of cardboard. This procedure allowed Ss to receive the same number and ratio of lights received in Experiment I, but without regard to whether or not they were talking at the time. In all other respects, the procedure was the same as that followed in Experiment 1.
Results. Table 4 shows the numbers of lights received by all Ss in Experiment III. It can be seen that these numbers correspond to the averages of Experiment I. The changes in sociometric status and verbal output that resulted from this procedure were no greater, and in some cases were smaller, than those obtained in the control groups (see Table 1). EXPERIMENTS
IV AND V
The experiments described thus far have all involved an attempt to alter the verbal behavior of all four group members. This procedure
ALTERATION
OF GROUP
STRUCTURE
65
consisted of essentially two operations: (1) an attempt to increase the TP’s output by rewarding his talking and punishing him for being silent, and (2) an attempt to decrease the other members’ output by punishing their talking and rewarding their silence. Although these two operations can be conceptualized separately, they did not function independently in the experiments reported above. The behavior change of the TP in Experiments I and II may have been due to one or the other of these two techniques, or to an interaction of the two. In order to determine the independent effects of these two operations, two additional experiments were conducted. Experiment IV provided only the TP with feedback, while Experiment V provided only the three nontarget Xs with feedback. Experime,nt
IV
Procedure. To provide only the TP with feedback during the second discussion, it was necessary to instruct the other three members of the group in such a way that they would not anticipate any feedback while at the same time instructing the TP in the usual manner. To achieve this, two sets of written instructions were prepared and were passed out to the group just prior to the second discussion. The instructions that the TP read were a written version of the standard instructions given to all Ss in the experimental condition of Experiment I. The other three Ss were given a written version of the control instructions of Experiment I. Both sets of instructions were carefully prepared so as to appear identical in terms of location of paragraph indentations, margins, and other typographical aspects. No Ss in any of the groups reported awareness of this difference in the instructions. A single E administered the red and green lights to the TP in the same manner that they were given to the TPs in Experiment I. No explicit attempt was made to replicate the mean numbers of each kind of light given in Experiment I.
Results. The sociometric and verbal output measures for Experiment IV show virtually no changes over the three discussions and do not differ from the results obtained in the control condition (Table 1). The mean number of lights received by the TP are reported in Table 4. In comparison with the figures for Experiment I, the number of green lights received by the TPs is much smaller. This was due mainly to the fact that the TPs did not markedly increase their verbal output in the second discussion. Experiment
V
Procedure. This experiment was essentially the complement of Experiment IV. Instead of encouraging the TP to talk more, the procedure of Experiment V consisted of discouraging the other three members of the group and withholding all feedback from the TP. Written instructions were again used to instruct the Ss differentially with regard to the administration of feedback in the second discussion. The TP received the control version while the other three were given the experimental version.
66
RAVELAS,
HASTORF,
GROSS,
Ah-D
KITE
The E who administered the fecdbnck lights to the three nontarget 8s followed the same general rules that were observed in Experiment I in attempting to dccrease their verbal output in the second discussion.
Results. The TPs in Experiment V increased slightly in ranking and verbal output from then first to the second (Table 1). None of these increases were significant, however. in Table 4, the nontarget ‘3s received slightly fewer red lights half as many green lights as did the nontarget Ss in Experiment
tiociometric discussion As sllo~l and about I.
DISCUSSION
The results of Experiment I clearly demonstrate that the procedure used is an effective method of changing the verbal output of selected group members in a desired direction. Furthermore, changes in the sociometric structure of the group are highly correlated with verbal out,put changes (T = 34). It had been anticipated that a crucial element necessary for the success of this manipulation was the manner in which the experimenters determined the appropriate times to rcinforce group members. Therefore, the results of Experiment II were somwhat surprising in that the programmed machine produced not only an increase in the TP’s verbal output, but also a significant rise in his sociometric st.atus. One possible explanation for the similarity in the results of Experiments I and II is that the same general class of behaviors was being reinforced by both the experimenters and the programmed machine. Although there are no data to confirm this notion, it was the opinion of the experimenters that all discussions were relatively homogeneous in content and highly task-oriented, with very few irrelevant or disruptive stat’ements being made by any of the participants. This being the case, it seems reasonable to assume that the majority of the TPs’ statements that were reinforced in both Experiments I and II were task relevant in nature, and, therefore, that an increase in such verbal output would result in higher sociomet.ric rankings for the TPs. Some indirect support for this contention is provided by Oakes (1962), who used a reinforcement technique similar to the one described in this report in an attempt to determine which of the twelve Bales’ categories of response are most susceptible t,o reinforcement. His findings show that only one of these categories, “giving opinion, evaluation, during analysis, expresses feeling, wish,” could be increased significantly a group discussion. It should be noted that more t,han 50% of the total responses were coded in this one category. These findings appear to coincide with our contention that discussion content was bot.11 relevant to the case discussion and homogeneous across groups. Therefore, it is not surprising that the csperimenter function of selecting statements
ALTERATION
OF GROUP
STRUCTURE
67
to “reinforce” in Experiment I is not critical in the alteration of group structure. At the same time, the results of Experiment III indicate that lights must be contingent upon verbal behavior to be effective. Receiving encouraging lights at predetermined intervals did not prove sufficient to significantly alter the initial group structure. The technique which was used successfully to modify group structure can be considered as two separate operations. One operation consisted of positively reinforcing or encouraging t,he target person to step up his verbal output. A complementary operation was employed simultaneously to depress the verbal output of the other group members except when they complied or agreed with the TP. It could be argued that one or the other of these two operations alone might account for most of the behavioral and sociometric changes. Experiments IV and V were designed to test the independent effects ‘of each of these two component techniques. The results suggest that both operations are necessary to produce modification of the group behavior. The ineffectiveness of either operation used separately indicates that without some encouragement a quiet group member will not spontaneously increase his output when other members are artificially depressed; and conversely it is not enough to encourage a quiet individual to participate more unless “room” is provided for his increased verbal output. It is also possible that when both techniques are employed, agreeing and complying behaviors of other group members may increase and provide additional social reinforcement for the TP. The previously mentioned high correlation between sociometric rankings and total talking time deserves further consideration. Given a situation in which four strangers are brought together, allowed to interact for a brief period, and then asked to evaluate one another on characteristics such as “best ideas,” “guidance,” and “leadership,” one might expect that sheer amount of talking would be a salient factor in determining these evaluations. It would be misleading, however, to conclude that such a high correlation between talking and sociometry always obtains in group situations, or that talking, regardless of its quality or appropriateness, always leads to the perception of good ideas and leadership ability. Such a hypothesis would have to be tested under a wider variety of situations than the present experimental design affords. For instance, other group members may rate a talkative man highly on “best ideas” if he is perceived as talking a great deal because he had earned encouragement from expert evaluators. On the other hand, others may not positively evaluate a talkative man’s ideas if they are aware that experimenters are manipulating rather than evaluating his behavior. These considerations have been explored by Hastorf et al. (1964). The results of Experiments I and II show that the increase in the
68
BAVELAS,
HASTORF,
GROSS,
AND
KITE
TP’s verbal output from the first to the second discussion is partly carried over to the third period when the lights are not used. The question of how long the effect lasts and whether or not it will generalize to other similar situations is unanswered. A series of experiments to test the perseveration and generalization of these effects is in progress. If TP gains in sociometry and talks more, some one person or combination of other participants must lose sociometric votes and talk less. Because of this “degrees of freedom” restriction, it is of some interest to explore the dynamics of the change situation. For instance, one might ask which group member(s) loses when TP gains. It would be reasonable to predict that each man loses sociometric votes and talking time in proportion to what he has to lose, i.e., M-1, the man who was originally ranked the highest, would lose the most, and M-W and M-S should lose proportionally less. An analysis of ratings received by M-l, M-d, and M-S (Table 5) shows that in Experiment I, M-l’s losses alone account for TP’s gains. In Experiment II, although M-1 suffers the greatest loss, M-d also talks somewhat less and receives lower ratings on the questionnaires. It might also be asked if the increase in the TPs’ sociometric rank from the first to the second discussion was largely due to the other members assigning lower ranks to themselves. This would result in an artificial elevation of the TPs’ rank. The data were analyzed with all self-rankings eliminated and each subject’s ratings of the other group members reranked from 1 to 3. This analysis revealed substantially the same pattern of results as those presented in Table 1. Two opposing hypotheses may be entertained regarding the TPs’ affective response to the experimental manipulation. It could be predicted that participants who are rewarded for talking will gain more satisfaction from the discussion and will generalize some of this affect to the case problem, or alternatively that the TP will feel uncomfortable in the unfamiliar role of a high participator. The postsession questionnaire results clearly support the former prediction in that TPs enjoyed participating in the second discussion and indicated strong preference for the second case problem. During the first discussion, before any lights were administered, high participators showed no more preference for the case than did low participators. In summary, the experiments described above provide a workable technique for the alteration of verbal output and sociometric structure in a group situation and define some of the necessary conditions to obtain such changes. Two problems emerge that are of significance to the understanding of
ALTERATION
OF GROUP
TABLE SOCIOMETRIC
RANK
AND
VERBAL
Experiment
TP
Sociometric ranka Control Verbal output (‘%) Sociometric rank
OUTPUT DISCUSSION
69
STRUCTURE
5 CHANGES PERIOD
FROM
DISCUSSION
PERIOD
1 TO
2
M-l
M-2
M-S
f0.24
-0.12
-0.35
1-0.25
+2.9
-2.3
-3.9
+3.3
+1.53
-0.72
-0.61
-0.31
-5.0
-0.7
-0.56
-0.05
-7.1
f3.6
I Verbal output (%) Sociometric rank
+21.3 +1.05
-15.7 -0.36
II Verbal output (%)
+13.7
-10.3
Sociometric rank
+0.32
+0.06
-0.32
+0.10
Verbal output (%)
+0.6
-2.3
-0.2
+2.1
Sociometric rank
+0.29
-0.45
so.11
-0.08
Verbal output (%)
f1.3
-3.5
$1.7
+0.6
Sociometric rank
$0.42
-0.14
-0.38
+0.08
Verbal output
$5.0
-4.8
-1.3
+1.2
III
IV
V (%)
0 Total positive and negative sociometric changes are not necessarily equal because self-rankings have been excluded from the averages which are presented in Table 1.
social interaction. The first relates t,c the perseveration and the generalization of behavior change. What are the conditions under which behavior change would perseverate and generalize to other conditions? We have obtained evidence of some perseveration to a session which immediately followed the acquisition session. An important variable in this respect is the way in which the TP himself views the situation. We would hypothesize that the more an individual perceives changes in his behavior as being self-caused and not the result of external forces (in Heider’s (1958) sense of the word), the more likely he will maintain some of this behavior change. The second general problem concerns the perception and evaluation of one person’s behavior change by others. In the experiments described above when a TP’s verbal output was markedly changed, the other group members attributed high quality to the output. It is our
hypothesis that obscrvcrs ~lio are able to sw tlic rcw-arding or punishing light’s would bc more likely to attribute the chnngc: in behavior to the influence of the lights and xould thus be 1~~s willing to nttributc SUCII qualities as leadership to the TP. REFERENCES AIHEN, E. operant Number
G. Interpersonal behavior changes perceived as accompanying the conditioning of verbal output in small groups. Tech~licnl 12~~0~1 II. Western Behavioral Sciences Institute, 1963. BACHR~CH, A. J., CANDLAND, D. K., AND GIBSOK, J. T. Group reinforcement of individual response experiments in verbal behavior. In Irwin A. Berg and Bernard M. Bass (Eds.), Co~lfornzity and deviation. Sew York: Harper, 1961. Pp. 258-285. BALES, R. F. Interaction process analysis: A method for the study of small groups. Cambridge, Mass.: Addison-Wesley Co., Inc., 1950. HASTORF, A. H. The “reinforcement” of individual actions in a group situation. In L. Krasner and L. P. Ullmann (Eds.), Research in behauior modificnfior>. Sew York: Holt, 1964. HASTORF, A. H., KITR, W. R., GROSS, A. E., AND WOLFE, LTN J. The perception and evaluation of behavior change. Unpublished manuscript. Stanford University, 1964. HEIDER, F. The psychology of interpersonal relations. -\iew York: Wiley, 1958. OAKES, W. F. Reinforcement of Bales’ categories in group discussion. Psychol. Rep.. 1962, 11, 427-435. OAKES, W. F., DROGE, A. E., AND AUGUST, B. Reinforcement effects on participation in group discussion. Psychol. Rep. 1960, 7, 503-514. PEPINSKY, P. N., HEMPHILL, J. Ii., AND SHEVITZ, R. N. Attempts to lead, group productivity, and morale under conditions of acceptance and rejection. J. Abnorm. sot. Psych& 1958, 57, 47-54.