longitudinal data analysis
Definition
Longitudinal data analysis refers to the statistical examination of repeated measurements collected from the same individuals over time, with the goal of separating within-person effects from between-person effects. In intensive longitudinal data, this separation is complicated by the fact that person-wise sample means, commonly used as proxies for stable individual traits, are partly determined by within-person variance. Observed correlations between those person-wise means are therefore a function of both the true population between-person correlation and the within-person correlations, meaning a spurious between-person relationship can appear even when none exists at that level. This bias worsens when the number of time points per person is low and between-person variance is small relative to within-person variance. Methods that jointly estimate within- and between-person effects in a single step are recommended to avoid this problem.
Sources: Haslbeck & Epskamp (2024)
Related Terms
- multi-level modeling (1 shared article)
- vector autoregressive models (1 shared article)
- separating within- and between-person effects (1 shared article)
Applications
Longitudinal Data Analysis and Multilevel Vector Autoregressive Modeling
Multilevel vector autoregressive models are a primary analytic framework applied to intensive longitudinal data, decomposing observed scores into stable person-specific means and time-varying deviations. A stepwise approach in which person-wise sample means are first extracted and then used to estimate between-person networks can induce bias in those between-person correlations when within-person dynamics are strong. Jointly estimating within- and between-person components in a single step removes this confound.
Sources: Haslbeck & Epskamp (2024)
Longitudinal Data Analysis and Within-person Versus Between-person Variance Decomposition
A central problem in longitudinal data analysis is that cross-sectional variance is partly a function of within-person variance, making person-mean-based estimates an unreliable guide to true between-person effects. The same two variables can show opposite correlation directions at the within-person and between-person levels. Analyzing only one level, or conflating the two, produces conclusions that reflect a blend of both sources of variation rather than either one cleanly.
Sources: Haslbeck & Epskamp (2024)
