The Use of Random Coefficient Modeling for Understanding and Predicting Job Performance Ratings
指出验证性因子分析在绩效评分方差分解中的局限，提出随机系数模型作为替代，并用实地数据发现评分者主效应方差是被评者主效应方差的两倍，且尽责性相关变量仅在评分者熟悉度高或评分人数多时正向预测绩效。
Earlier research using confirmatory factor analysis (CFA) suggests that most variance in job performance ratings is not attributable to ratee main effects. In this article, the authors point out several issues associated with CFA methodology and argue that random coefficient modeling (RCM) can be a useful alternative for estimating variances associated with ratee main effects, rater main effects, and the upper bound of Rater × Ratee interaction effects. Using an application of RCM on field data, the authors found that rater main effects variance was nearly two times as large as ratee main effects variance. They report meaningful contingencies of these findings by modeling rater familiarity with the ratee and the number of ratees rated by a rater. Finally, interactions revealed that Conscientiousness-related variables were positively related to job performance only when rater familiarity with the ratee was high or the number of ratees rated was high. The authors discuss how the RCM methodology can be used to assess the construct validity of job performance ratings and to test substantive hypotheses involving variance components, main effects, and interactions within nonindependent observations.