Time Aggregation and Missing Time Frames in Causal Research With Panel Data

A common goal in psychological and behavioral research is the study of causal mechanisms underlying a particular process of interest. These research questions are commonly investigated using panel data obtained at relatively large time intervals of months or years and with measurements referring to the past week, month, or year. In this article, we investigate to what extend panel data can be used to study a causal mechanism when the causal process is assumed to play out at a (much) faster timescale than that at which the panel data were obtained. Specifically, we used simulations to study the impact of time aggregation (i.e., aggregating scores over multiple occasions) and systematic(under)sampling (i.e., when parts of the ongoing process are not covered with the measurements) on the ability of popular dynamic models to approximate the effects of a time-varying exposure on a time-varying outcome in three different scenarios. The results show that time aggregation and systematic (under)sampling can lead to severe over- and underestimation of the causal effect, implying that wrong (and potentially harmful) conclusions can be draw. We discuss the implications of these findings for applied researchers.