Can binary variables be used as time-varying predictors in CRE?

[Sylvie Rod]

Hello, I have a question about using time-varying variables in a CRE (correlated random effects) model. Is it acceptable to include dichotomous (0/1) variables as time-varying predictors, or do they need to be continuous?

For example, suppose I observe marital status over five years, and an individual changes from single (0) at t1 to married (1) at t3. Can I treat this marital-status variable as a time-varying variable in a CRE model?

Thanks.

[Andrew Musau]

A variable is considered time-varying simply when its value changes over time, regardless of whether it is categorical or continuous. In Stata (and in panel-data methods generally), any variable that switches from 0 to 1 or 1 to 0 within units across periods is treated as time-varying.

In practice, some variables (such as marital status) may not change for most individuals over short panels (e.g., five years). In that sense, they may be empirically time-invariant for a large share of the sample, even though conceptually they could change. One may describe such a variable as "time-invariant" in this short-run context, but that is a linguistic or empirical convenience, not a theoretical claim that the variable never changes.