A multivariate comparison of several models of cognitive balance theory



Journal Title

Journal ISSN

Volume Title



This investigation was concerned with examining Ss' perception of sentiment relations in three-person experimentally manipulated groups. It was assumed that such perception obeys certain Gestaltlike principles and that models of cognitive balance can, to varying degrees, account for the dynamics involved. Seventy-two undergraduate female volunteers in groups of three ad-libbed discussions of a hypothetical vacation. Each S was instructed to role play positive, negative, or mixed (like one, dislike the other) sentiments toward the other two members of her group. In a deception paradigm (D) each S thought herself to be the only role player in the group. In a no-deception paradigm (ND) each S knew that everyone in the group was role playing. Role play instructions were coordinated so that each of four triad configurations (+++, ++-, +--,---) was represented by a group and so that each of the component dyad relationships was reciprocal. The design was counterbalanced by running each of the four groups three times using both the D and ND paradigms (total 24 groups). Subjects' ratings of group sentiments were categorized according to the independent variables suggested by the cognitive balance models proposed by Heider, Newcomb, and Taylor. The three models, each in two variations (I and II), were compared using multiple regression analysis on the dependent variables: 'pleasantness' (PL), 'subjective tension' (ST), 'tension release' (TR), and 'changes in sentiment ratings over time' (Dn). Results indicate (a) initial sentiment ratings were significantly associated with role play instructions (r = .70 to.82, p's < .01), (b) deception conditions were more sensitive to subtle independent variable effects. The ND condition generated the same general data pattern as the D condition, but the latter was better able to produce extreme scores and certain interaction effects, and (c) the Newcomb II and Taylor II models best predicted Dn and ST while the Newcomb I model best predicted PL scores. Overall, the most powerful model appeared to be the 'continuous degree of balance' variation of Taylor's model (Taylor II). This model accounted for a mean 28 percent of the dependent variable variance. Each of the models tested relies on the analysis of a certain conceptual set of triad information. These sets can be seen as the interrelation of specific 'agreement' (A) and 'positivity' (P) variables. When the global effects of these two variables were assessed it was found that they accounted for a mean of 38 percent of the dependent variable variance and that most of this variance was accounted for by the P component. The more positive the relations in a group, (a) the less likely they were to change (r = -.86, p < .01), (b) the higher the PL rating (r = .76, p < .01), and (c) the lower the ST ratings (r = -.30, p < .01). While the three balance models relying on only 'partial information sets' did surprizingly well in ordering the results, they were differentially appropriate to the various dependent variables. It was concluded that, for the type of situation studied, the tendency toward positivity appears to best account for the early developmental dynamics of the group.



Interpersonal relations, Communication--Psychological aspects