Situational bandwidth and the criterion-related validity of assessment center ratings: is cross-exercise convergence always desirable?

This research examines the relationship between the construct and criterion-related validity of assessment centers (ACs) based on how convergence of dimension ratings across AC exercises affects their ability to predict managerial performance. According to traditional multitrait-multimethod perspective, a high degree of convergence represents more reliable measurement and has the potential for better validity. In contrast, the concept of situational bandwidth suggests that behavior assessed under a dissimilar set of circumstances should result in a more comprehensive assessment of a candidate's tendencies even though ratings are less likely to show high convergence. To test these opposing viewpoints, data from 3 operational ACs were obtained along with experts' evaluations of exercise characteristics and supervisors' ratings of candidates' managerial performance. Across the 3 samples, AC ratings taken from exercises with dissimilar demands had higher estimates of criterion-related validity than ratings taken from similar exercises, even though the same dimension-different exercise correlations were substantially higher between similar exercises. Composites of ratings high in convergence did not emerge as better predictors of managerial performance, and validity particularly suffered when derived from ratings that converged as a result of exercises with similar demands. Implications for AC design are discussed.