Design, sample size planning, and preprocessing of intensive longitudinal data studies

Plenty of research focuses nowadays on understanding how psychological processes dynamically evolve within persons. To capture these processes, data are collected through intensive longitudinal designs (ILD) and modeled with time series approaches. A popular approach is Vector Autoregressive (VAR(1)) modeling of order 1, in which observations at the current measurement occasion are modeled as a function of the observations at the previous occasion. To interpret VAR(1) parameters using hypothesis tests, it is crucial to perform rigorous sample size planning to ensure sufficient statistical power. Sample size planning methods are largely lacking and existing methods for cross-sectional data are invalid as they ignore the serial dependence in the data. This project aims to develop simulation-based techniques to perform sample size planning for ILD, focusing on VAR(1) and its multilevel extensions. We will first study how to optimally select the number of measurement occasions for a single individual. Next, we will compare event- and time-contingent sampling schemes in terms of power. Third, when simultaneously studying multiple persons, we will investigate the optimal trade-off between the number of persons and measurement occasions. Fourth, we will inspect if an alternative statistical quality measure, predictive accuracy, can adequately inform about sample size planning. Finally, we will disseminate the new techniques by building freely available software.