In many studies related to human cognition and behaviour, researchers measure people repeatedly over time to understand how cognition and behaviour are affected by external factors. Consider, for example, the question of whether social media use causes increased anxiety and stress in children. To answer such questions reliably, scientists need to carefully consider the role of time in their research. If measurements are taken too infrequently, or if participants are asked to summarise their experiences over a past period rather than reporting how they feel right now, important information about how stress and anxiety levels change over time can get lost.

In this paper, Jeroen Mulder and colleagues describe the effects of this loss of information (due to suboptimal timing in measurements) across several longitudinal research designs. Using computer simulations, they demonstrate that in these situations, popular statistical methods can produce dramatically wrong estimates of cause-and-effect relationships. Statistical methods can detect effects that do not actually exist, or miss relationships that are actually present. This study suggests methodological lines of research to address this issue, and underscores the critical importance of considering timing in the design of studies and subsequent data analysis.

This article helps applied researchers to make informed decisions about how to design their longitudinal study, specifically with regards to aligning their targeted causal effects with the temporal characteristics of the longitudinal process of interest.