Composite scores (e.g., mean or sum of survey items) are widely used as outcomes or predictors in psychological and social science research despite methodological concerns regarding measurement error. Despite extensive study of measurement error in path models, relatively little attention has been paid to this methodological issue in latent growth modeling (LGM), particularly when predictors vary over time. Time-varying predictors allow for modeling occasion-specific influences beyond underlying developmental trajectories, yet they are frequently operationalized using composite scores that implicitly assume perfect reliability. To address this gap, the present study investigates the consequences of ignoring measurement error in composite time-varying predictors within the LGM framework. Notably, this is the first study to evaluate the single-indicator (SI) modeling approach as a method for correcting measurement error in time-varying predictors. We compared the traditional LGM incorporating composite predictors with the LGM that incorporates the SI factor approach to account for measurement error in time-varying predictors using the Early Childhood Longitudinal Study, Kindergarten Class of 1998–1999 (ECLS-K) dataset and a Monte Carlo simulation. The Monte Carlo simulation results revealed that ignoring measurement error in time-varying predictors attenuated occasion-specific effects by up to 30%. These findings underscore the necessity of correcting for measurement error using SI factor modeling to ensure the validity of developmental inferences in LGM.
Kejin Lee (Wed,) studied this question.