Statistical Issues with the Analysis of Panel Data
The primary data analyzed in this book are collected from a series of panel studies. Since there are some statistical issues with this kind of analysis, I include a brief discussion of those problems and issues in this appendix, but this section is not essential to understand the analysis presented in the empirical chapters. Specifically, there are various concerns with how the dependent variables are measured. In panel studies, there are a number of approaches to analyzing effects over time. Two of the most common groups of approaches are “change score” models and “regression adjustment” or “static score” models (Judd and Kenney 1984; Finkel 1995). Each of these models improves upon standard cross-sectional approaches by incorporating the dynamic components, but in so doing they can suffer from a different set of statistical problems.
The “change score” method uses simple differences between the level of the dependent variable at the various time points. In a two-wave study, the change score is simply the difference between the dependent variable at the first and second measurements. This method regresses the change in Y on the change in X. Variations in this general model include using either X from a single time point rather than the change in X. This model is also known as the “unconditional change score” or the method of first differences (Finkel 1995, pg. 5). The key characteristic of this model is that the dependent variable is the difference between the two time points and it is not included as an independent variable. Essentially, this model is more concerned with predicting or explaining change, rather than predicting or explaining the absolute level of the dependent variable at the second time point. A particular advantage of this type of analysis of panel data, as opposed to the types discussed below, is that it controls for problems