Week 8

Lesa Hoffman - Professor, The University of Iowa

On the Strategies for Disaggregating Between-Person Relations across Individual Time Slopes from Within-Person Relations in Longitudinal Data

A primary motivation for conducting longitudinal research is to examine relationships between variables at both the between-person and within-person levels of analysis simultaneously. Given the prominent role of time in the longitudinal designs, however, these two types of relations are likely to be insufficient in characterizing the full spectrum of ways that predictors and outcomes that are each measured repeatedly over time can relate to each other. In this talk I will focus on the deleterious impact of ignoring between-person relationships involving individual differences in change over time on the accuracy of within-person fixed effects. Because there is a large and fragmented literature describing different models for capturing between-person and within-person relations in longitudinal data in the first place, I will review these different approaches: univariate multilevel models (using person-mean-centered or detrended predictors), path analysis (to examine cross-lag or mediation relationships), structural equation models (for "latent" change), as well as strategies for latent centering of predictors: multivariate multilevel models, multilevel structural equation models, and "dynamic" structural equation models. I will also report the results of a simulation to demonstrate the problems that arise from ignoring between-person relationships involving time slopes: biased within-person fixed effects at the same occasion, as well as bias in other types of within-person fixed effects (e.g., their change over time; cross-lag associations). I will conclude with recommendations for best practice in longitudinal modeling.