Heterogeneity of coefficient on variable of interest over time

[Keegan Robertson]

I am working with an (unbalanced) panel where I have conducted a fracreg logit regression of my variable of interest with covariates on outcome of interest - can also be done with glm link(logit) fam(bin).

I have 58 time periods in my main sample, but some of my covariates are only available for the last 12 or so periods. If I run the reduced sample with increased covariates I get a different coefficient (lower magnitude and significance). I have a suspicion that within my sample there is heterogeneity of the coefficient of interest over time, and I am trying to find the best way to identify that (i.e. are there particular periods where the coefficient changes, possibly direction).

Any suggestions?

[Carlo Lazzaro]

Keegan:
if I'm getting yiou right, I would explore the chance of interacting the independent variables of interest with -timevar-.

[Keegan Robertson]

Thanks Carlo, I'll explore that. It was also suggested to me separately that I could use randcoef, does that make sense? I've tried looking at the documentation and can't seem to get my head around how it would work to identifying temporal heterogeneity, or pulling out particular years of interest.

[Carlo Lazzaro]

Keegan:
honestly, I am not familiar with the Stata comunity-contributed module -randcoef- (as you're kindly requested to mention it, for practical reasons explained in the FAQ. Thanks).
Sorry for that.