Analyzing Longitudinal Multirater Data with Individually Varying Time Intervals

Abstract

Numerous models have been proposed for the analysis of convergent validity in longitudinal multimethod designs. However, existing multimethod models are limited to measurement designs with equally spaced time intervals. We present a new multirater latent state-trait model with autoregressive effects (MR-LST-AR) for designs with structurally different raters and individually varying time intervals. The new model is illustrated using the German Family Panel pairfam. By means of stochastic differential equations, we show how key coefficients of convergent and discriminant validity can be examined as a function of time. We compare the results from continuous and discrete time analysis and provide code to fit the new model in ctsem. Finally, the advantages and limitations of the model are discussed, and practical recommendations are provided.