Persons as Contexts: Evaluating Between-Person and Within-Person Effects in Longitudinal Analysis

TLDR

Time‑varying predictors are widely used in developmental multilevel models, yet omitting them can bias effect estimates. This didactic article explains why and how to treat persons as contexts in longitudinal analysis. The authors distinguish constant between‑person from time‑specific within‑person variation by modeling persons as contexts and provide accompanying syntax in an electronic appendix.

Abstract

Relationships among multiple variables over time are of interest in many developmental areas and are frequently examined using time-varying predictors in multilevel models. Yet an incomplete specification of time-varying predictors will usually result in biased model effects. Specifically, the impact of constant, between-person sources of variation must be differentiated from the impact of time-specific, within-person sources of variation - that is, persons should be modeled as contexts. The current didactic article expands upon previous work to address why and how to model persons as contexts in longitudinal analysis. An electronic appendix of syntax for estimating these models is also provided.