Episode-contingent experience-sampling designs for accurate estimates of autoregressive dynamics

Affect dynamics are often studied by means of first-order autoregressive (AR) modeling applied to intensive longitudinal data. A key target in these studies is the AR parameter, which is often tied conceptually to regulatory behavior in the affective process. The data is typically gathered using experience sampling methods, which are designed to pick up on fluctuations in affective variables as they evolve over time in naturalistic settings. In this manuscript, we compare classical time-contingent sampling designs to episode-contingent sampling designs, which initiate sampling when an emotional episode has been signalled. We define emotional episodes as periods where an affective process strays relatively far away from its mean. Compared to time-contingent designs, episode-contingent designs leverage on increased affective variability, which can have beneficial implications for the precision of the ordinary least squares AR effect estimator. Using an extensive simulation study, we attempt to delineate which characteristics of an episode-contingent design are important to consider, and how these characteristics are related to estimation benefits. We then turn to an empirical illustration, showing how an episode-contingent design can be implemented in practice. We also show that various patterns we expect based on the theoretical parts of the manuscript are recovered in the data. We conclude that episode-contingent designs can have marked benefits for the precision of the AR effect estimator, and discuss a number of challenges when it comes to implementing episode-contingent designs in practice.