Problem with ommitted interaction variables for fixed effects panel data

[Layla Mayer]

Dear all,

I´m having problems estimating fixed-effects models including interaction-variables for a two-wave-panel.

I´m using xtreg varlist, fe in STATA 12 and would like to include interaction-variables in my model. The interactio term includes a timevaring dummy-variable (giving information wether respondants expirienced childbirth or not) and a time-constant factor variable. I created the interaction-variable using i.var1#i.var2

STATA estimates effects for only a few of the categories of the interaction-variable and states that the other categories are "omitted because of collinearity".

Can someone explain to me, why STATA ommits some of the categories of the interaction-variable in this fixed effects model and how I can solve this problem so I can estimate the moderator effects of time-constant variable on the effect of childbirth?

Thanks alot!

[Carlo Lazzaro]

Layla:
welcome to this forum.
What you're experiencing is exactly what to expect from the -fe- estimator: as reported in any decent panel data econometrics textbook, it gets rid of any time-invariant predictor.
Besides, what Stata tells is again in line with the expectations: if predictor are collinear with panel fixed effect, they will be omitted.
As usual, Stata (not STATA, please) is not to blame: it did exactly what you ask it to do.
The main issue with your data, as it is often the case, rests on the fact that you want to estimate coefficients for time-invariant predictor, but it's simply unfeasible with the -fe- estimator.
I would investigate., via -hausman-, whether -re- specification (which allows time-invariant predictor coefficents to be estimated) fits your data better then the -fe- one.
As an aside, as reported by the FAQ, please note the the best way to get helpful replies is to post what you typed and what Stata gave you back using CODE delimiters. Thanks.