The simulation of individual prediction in an 'alone' - and 'group' - condition

30 subjects had to predict a stimulus persons reactions to the itmes of a value test. they were given 5 answers of the s-person among which each of the five assumed dimensions of the unnormed test was represented. then 10 additional answershad to be predicted. after each prediction the true answer of the s-person was announced (by the experimenter) or could be found in an envelope. the same 30 subjects were run with the same value test but 2 different s-persons (one constructed, the other a modified empirical person) first in an "alone condition" where the prediction was given without interaction and then in a "group condition" where the 5 members first fixed an individual prediction each and then discussed these and derived a group prediction item by item. in the group condition the amount of individual interaction seemed to have no influence on the immediate outcome of the group decision (almost without exception - 56 out of 60 cases - the median of the individual predictions was chosen). it therefore was supposed that the interaction might have some influence on the group members individual solution process for the subsequent trials. to prove this assumption some models to simulate the individual prediction process were constructed. in the alone condition a variety of programs which assumed that for the first trials "projection" will take place (i.e. predict one's private opinions for the s-persons) and that then a learning process, which leads to a preference for the most often heard category of the s-persons answers, will take over, were the models with the best fit (119 maximal correct predictions out of 300) . these models did very bad in the group condition. in this situation models which assumed the summation of reaction tendencies and the realization of the biggest one were much more appropriate. the most important summands of the reaction potential were the "content structure" of the test itself (the face validity of the items presumably) and some preferences for not too extreme but yet information-conveying response categories (maximal success achieved was 136 correct predictions out of 300). the prior answers of the s-person were important only for the first trials and then ones own and the partner's priors predictions became predominately important. also the projection mechanism or one's private opinion to the items was much less important in the group than in the alone condition. the dissimilarities between the solution process in the alone-condition and the group-condition are exhibited by the simulation models which best describe each of the two situations. to obtain more perfect fit in both situations the consideration of personality variables and information on the test items and of verbal content of interaction in the group situation seems quite inevitable. only the second of the three kinds of information necessary was used to a small extent in our models, the primary goal of the investigation being to exploit the redundancies in both situations considered.;