separating within- and between-person effects
Definition
Separating within- and between-person effects refers to the statistical decomposition of variance in intensive longitudinal data into components attributable to stable individual differences across persons and to fluctuations occurring within a single person over time. The distinction matters because within-person and between-person correlations between the same pair of variables can diverge substantially, as illustrated by cases where faster typing within an author predicts more errors, while faster-typing authors on average make fewer errors, or where exercising raises heart rate within a person while frequent exercisers tend to have lower resting heart rates. A common shortcut for achieving this separation is to use person-wise sample means as proxies for true person-wise means, centering the data within persons to isolate within-person effects while treating relationships among those means as estimates of between-person effects. However, observed correlations between person-wise means are a mathematical function of both the true between-person correlation and the within-person correlations, meaning that spurious between-person correlations can emerge from within-person dynamics alone, with the bias being most severe when the number of time points is low and between-person variance is small relative to within-person variance.
Sources: Haslbeck & Epskamp (2024)
Related Terms
- longitudinal data analysis (1 shared article)
- multi-level modeling (1 shared article)
- vector autoregressive models (1 shared article)
Applications
Separating Within- and Between-person Effects and Multilevel Vector Autoregressive Models
Multilevel vector autoregressive (VAR) models, implemented in packages such as mlVAR, use person-wise sample means to simultaneously estimate within-person temporal and contemporaneous effects and between-person network relationships among stable means. Because person-wise sample means contain variance from within-person deviations, the between-person networks estimated through this stepwise approach can reflect within-person correlations rather than true population between-person correlations, a bias that carries over directly to the estimated between-person network structure.
Sources: Haslbeck & Epskamp (2024)
Separating Within- and Between-person Effects and Dynamic Structural Equation Modeling
Dynamic Structural Equation Modeling (DSEM) jointly estimates within- and between-person parameters in a single step rather than using person-wise sample means as proxies, thereby avoiding the contamination of between-person estimates by within-person variance. This single-step approach is recommended specifically when the number of time points per person is low or when between-person variance is small, conditions under which the bias introduced by the stepwise person-means approach is most pronounced.
Sources: Haslbeck & Epskamp (2024)
