摘要：Test-retest studies for assessing stability and change are widely used in different domains and allow improved or additional individual estimates of interest to be obtained. However, if these estimates are to be validly interpreted the responses given at Time-2 must be free of retest effects, and the fulfilment of this assumption must be empirically checked. This article proposes a comprehensive item response theory-based approach for assessing retest effects at the individual level and test the assumption of local independence under repetition. The approach can be used with a wide array of unidimensional and multidimensional models, and is based on correlation-type and mean-square-type indices. Procedures for (a) establishing critical values for detection purposes and (b) interpreting the magnitude of the retest effects for the detected respondents are also proposed. Furthermore, the article discusses the consequences of not addressing retest effects in stability and change studies. The procedures were assessed with simulation and used in three empirical studies. In all cases they worked well and provided meaningful information.
其他摘要：Test-retest studies for assessing stability and change are widely used in different domains and allow improved or additional individual estimates of interest to be obtained. However, if these estimates are to be validly interpreted the responses given at Time-2 must be free of retest effects, and the fulfilment of this assumption must be empirically checked. This article proposes a comprehensive item response theory-based approach for assessing retest effects at the individual level and test the assumption of local independence under repetition. The approach can be used with a wide array of unidimensional and multidimensional models, and is based on correlation-type and mean-square-type indices. Procedures for (a) establishing critical values for detection purposes and (b) interpreting the magnitude of the retest effects for the detected respondents are also proposed. Furthermore, the article discusses the consequences of not addressing retest effects in stability and change studies. The procedures were assessed with simulation and used in three empirical studies. In all cases they worked well and provided meaningful information.
