A Computer Implementation of the Reason Formulation Applied to Person Perception

Carlson, Ray H. (Ph.D., 1979)

Thesis directed by Associate Professor Peter G. Ossorio

An introductory discussion set forth two ways to approach the study of human behavior. One was identified as the Covering Law Model and the other as the Reason Formulation. The former emphasizes the establishment of general causal links between variables, whereas the latter emphasizes the role of reasons and intentions in explaining behavior. It was noted that the Covering Law Model has been the more dominant approach. A review of many philosophers of science, including William Dray and Theodore Mischel, was intended to give credence to the Reason Formulation as a legitimate and scientific approach to the explanation of human behavior. This research presents a rationale, justification, and implementation of the Reason Formulation in concrete form via computer modeling. This is derived from the larger conceptual framework of Descriptive Psychology developed by Peter G. Ossorio.

To demonstrate the use of the Reason Formulation a type of person perception experiment was designed. A computer program was devised to model observers who would make predictions about what actions an actor would choose across a range of situations. In order to model the observer's predictions data was gathered on each of four observers and five actors. The actors were asked to designate the actions that they would choose in 14 hypothetical situations. Subjects generated 20 appraisal terms via factor analysis which loaded significantly on 5 value perspectives: (1) ethical, (2) prudential, (3) hedonic, (4) intellectually esthetic, and (5) socially esthetic. Each observer rated each situational action with respect to each appraisal term. The net strength of each generic perspective for each action was computed. On the basis of least discrepancy, the computer models predicted the predictions of the observer.

Two models incorporating rules for the assessment and prediction of behavior by observers were implemented: (1) the Combinatorial Model, and (2) the Priority Model. The first gave equal weighting to each value perspective in assessment and prediction of behavior. The Priority Model gave equal weighting to all perspectives with the exception of doubling the weight of the most salient perspective. This was done to take a limited look at what difference varying weightings would make in modeling observer's predictions.

The results showed that both models correlated at .001 significance level with the predictions of observers. The Priority Model (1) produced consistently higher correlations across all observers, though insignificant for any one observer; and (2) is consistently better across all situations. The correlations between models and observers within situations varies widely. Both models also simulated Observer #1 significantly better than the others. All observers except #3 did better than the models at predicting the actual choices.

In discussion several issues were raised: (1) higher multiplicative factors in the Priority Model could be explored to discover its effect on goodness of fit; (2) there is the possibility that different sets of rules would yield better fit with different observers; and (3) more Individual Difference parameter data could show what kind of a person is modeled best by which model. In addition it appeared that those situations yielding highest correlations have a set of actions which are more dissimilar from each other. Finally, the poor performance of Observer #3 was speculated to be related to her poorer sense of judgment. Again, Individual Difference parameter data was seen as a potential contributor to more complete understanding of such results. Overall, the research showed the usefulness of the Reason Formulation in the instance of modeling person perception. [90 pp.]