Determinants of normative processes: comparison of two empirical methods of specification

摘要

This study focused on how an action determines impressions of the individuals participating in the action, a substantive area with problematic data in the form of constricted variances, multicollinearity, and excessive influence of a few extreme cases. Stepwise regression often is used to determine which variables influence an outcome, but can lead to mis-specifications with problematic data. Thus this study compared two methods of discovering determinants of impression formation: stepwise regressions and analyses of variance (ANOVA). About three-quarters of the specifications obtained with one method also were obtained with the other method. The shared specifications especially related to stability of impressions, the effect of behavior morality, and consistency between evaluation of behaviors and evaluations of the participants in the action. Unique specifications from ANOVA were easier to interpret than those from stepwise regressions because stepwise regressions brought in more complex interactions than did ANOVA. With both methods, results from sub-samples constituted approximate subsets of results from larger samples, indicating that key effects can be found in small studies. However, compared with ANOVA, stepwise regressions with sub-samples brought in more effects that were absent in results from the full sample. The advantages of ANOVA derived from dichotomizing exogenous variables, which ameliorated data problems.