2

I have individual panel-level data and am using conditional binomial fe logit regressions to estimate the effect of minimum wage on Employment.

1) When I try to include year dummies most of them are excluded due to multicollinearity. I have tried evertying - dropping variables, changing the values of the year dummies - but nothing seems to work. Any advise?

2) When I try to include state dummies all of them are excluded due to no within-group variation. This makes sense, as the survey I use to get the data does not follow movers. Is there still a way to control for state effects?

user40908
  • 53
  • 5
  • 2
    State effects are implicitly controlled for by using fixed effects. If your time dummies are auto-dropped, its probably because they are indeed perfect linear combinations of other things you've got. If so, you're good. – generic_user Mar 15 '14 at 20:43
  • 2
    You might want to cluster on state though, rather than ID. – generic_user Mar 15 '14 at 20:44
  • Would it be possible to cluster on both ID and state? The problem with the time dummies is that all but three of them are auto-dropped. Should I still include those three? – user40908 Mar 15 '14 at 23:48
  • Unless there are ID's that span states, then you can cluster on the nesting variable. I think there is a paper by Colin Cameron at UC Davis that explains this. Re: dummies: are they collinear? – generic_user Mar 16 '14 at 00:14
  • I have solved the state issue, but after two days of trying to figure out how to solve the multicolinearity issue with my year dummies I am on the verge of giving up. The model runs perfectly well until I incorporate the year dummies, at which point 3 or 4 (depending on which specification I run) of the 10 year dummies are dropped due to collinearity. Does anyone know what the potential problem could be? I would be so incredibly grateful if anyone could help me with this! – user40908 Mar 17 '14 at 23:05

0 Answers0