Life course researchers often collect longitudinal data by assessing repeatedly, over long time spans, the same individuals. Such data are inherently dependent, and thus cannot be analyzed with standard classical models, like ordinary least squares regression, because statistical inference about the estimated parameters would be incorrect. In this chapter we present two related families of statistical models for longitudinal data: linear mixed-effects models and structural equation models. Both classes of models allow analyzing quantitative longitudinal data and explicitly define parameters related to both stability and change processes. The models also allow studying interactions between individual and contextual characteristics, both of which may be stable in time or vary across time. Advantages and recent extensions of these models include, among others, (a) statistical advances to cope with incomplete data, without the need to impute incomplete data, nor to limit analyses to complete cases; (b) multivariate specifications, to study how multiple dimensions of one’s life may change in parallel or even exert reciprocal influences; (c) multiple group analyses, to compare groups of known membership; (d) latent class analyses, to uncover previously unknown group membership according to specific statistical features. We will discuss similarities, advantages, and drawbacks of both families of models and illustrate them by analyzing public health data from the Swiss Household Panel.
Copyright 2015, The Author(s).