Multiple independent variables in an xtreg with fe

[Tom Prosser]

Dear forum,
I'm trying to analyse the influence of occupation on ideology using the British Election Study; this is a large panel study of the British electorate. Specifically, I'm looking at the influence of transitioning into a job in the charity sector (for the first time) on economic ideology (left-right) and have created independent variables that record whether respondents are 1,2, or 3 years pre or post their transition (so six variables in total). I'm using xtreg on Stata17 and have included pre-transition variables in my model (see below) as I want to look for a self-selection effect. There's also a year dummy in my model and I use fixed effects and robust standard errors.

xtreg econideology 3yrpre 2yrpre 1yrpre 1yrpost 2yrpost 3yrpost i.year, fe robust

My question - given I'm using fixed effects, should I be including the six pre and post transition variables in my model? Would it be better to include them individually?

I'd be immensely grateful for any advice,
Tom

[Jeff Wooldridge]

Tom: This looks like a kind of staggered intervention to me, and I would estimate a staggered diff-in-diffs where the treatment is changing occupation. You then need to define a variable, say "cohort," which is the year the person changes occupations. I'm assuming the population includes people not in the charity sector for at least the first year of data. Then

jwdid econideology, ivar(id) tvar(year) gvar(cohort) never
estat event
estat plot
estat simple

This will give estimates by (cohort,year) pair and also an event-study plot. The last command gives a single weighted average estimate, but estat event is more like what you typed above. You can include controls if you want to.