Planned missingness in intensive longitudinal studies: Extensions and comparisons of multiform designs

Technological advances in data collection have made intensive longitudinal studies (ILSs) increasingly feasible. However, conducting such studies often leads to increased participant burden due to the high frequency of measurements, resulting in nonresponse and a consequent reduction in data quantity and quality. Previous studies have shown that planned missingness designs can help optimize data collection. In this paper, we extend two multiform designs commonly used in planned missingness research and propose a flexible alternative involving completely random sampling. These designs are better tailored to ILSs by varying item subset combinations across measurement occasions for each participant. Specifically, we examined (1) the anchor test design, where a core subset is administered to all participants across all occasions; (2) the matrix sampling design, where subset combinations rotate systematically; and (3) the random sampling design, where subset combinations are randomly assigned at each occasion. We conducted two simulation studies to evaluate the performance of these designs within the dynamic structural equation modeling (DSEM) framework across varying key model parameters, sample sizes, numbers of time points, and planned missingness proportions. An empirical example is also provided to demonstrate the potential of multiform designs in ILSs. Results indicate that these multiform designs can yield unbiased parameter point estimates with acceptable credible interval coverage rates, while maintaining sufficient statistical power. Therefore, the extended and proposed multiform designs enable researchers to reduce both the costs and participant burden while still obtaining adequate data to capture characteristics in the dynamic process.