Decomposing the True Score Variance in Rated Responses to Divergent Thinking-Tasks for Assessing Creativity: A Multitrait-Multimethod Analysis.

Abstract

It is shown how the Correlated Traits Correlated Methods Minus One (CTC(M - 1)) Multitrait-Multimethod model for cross-classified data can be modified and applied to divergent thinking (DT)-task responses scored for miscellaneous aspects of creative quality by several raters. In contrast to previous Confirmatory Factor Analysis approaches to analyzing DT-tasks, this model explicitly takes the cross-classified data structure resulting from the employment of raters into account and decomposes the true score variance into target-specific, DT-task object-specific, rater-specific, and rater-target interaction-specific components. This enables the computation of meaningful measurement error-free relative variance-parameters such as trait-consistency, object-method specificity, rater specificity, rater-target interaction specificity, and model-implied intra-class correlations. In the empirical application with alternate uses tasks as DT-measures, the model is estimated using Bayesian statistics. The results are compared to the results yielded with a simplified version of the model, once estimated with Bayesian statistics and once estimated with the maximum likelihood method. The results show high trait-correlations and low consistency across DT-measures which indicates more heterogeneity across the DT-measurement instruments than across different creativity aspects. Substantive deliberations and further modifications, extensions, useful applications, and limitations of the model are discussed.