Advocating a New Default for Time Series Modeling: Asymmetric Temporal Dynamics using Threshold Dynamics Parameters
Year: 2024
Type: preprint
Abstract: The availability of intensive longitudinal data has rapidly increased, and with it our ability to understand how behavior evolves. Statistical models to analyze such data often include dynamics parameters that capture dependencies within or between variables over time. However, these dynamics parameters are typically thought to operate independently of the current value of the time series. Here, we argue that by doing so important information on psychological processes is overlooked. We therefore propose the DYNamics of ASymmetric TIme series (or DYNASTI) approach in which dynamics parameters are allowed to differ, depending on whether they operate above, or below, the time series mean of a given person or unit of analysis. For instance, the effect of exercise on health outcomes may vary depending on whether a person exercises more, or less, than average. Or, the inertia of sleep may vary depending on whether a person sleeps more, or less, than their average. Through extensive simulations, we show that DYNASTI implementations of two commonly used time series models (DSEM and RI-CLPM) are able to adequately recover both symmetric and asymmetric temporal dynamics above and below the time series mean. Importantly, we also show that assuming symmetric dynamics (as in the vast majority of the literature) when processes are in fact asymmetric leads to incorrect conclusions about these dynamics. We further illustrate the workings of DYNASTI implementations and how they can lead to new insights in empirical examples regarding negative affect, smoking urges, and anxiety. To implement, and extend, DYNASTI models we provide an annotated, open source code base built in RStan. Together, these results advocate DYNASTI implementations as a new default for time series modeling.
Language: English
Subfield: Experimental and Cognitive Psychology
Field: Psychology
Domain: Social Sciences
Open Access status: gold
OpenAlex ID: W4401754070