Hausman xtpoisson xtreg

[John Friend]

Hello,
I am trying to perform a Hausman to compare my xtreg and xtpoisson regressions. However, after performing the test I receive "no coefficients in common; specify equations(matchlist)
for problems with different equation names."

That's my Code:

xtpoisson probpend mietdiffwi $beschaeftigung $control, fe
est sto pois_haus

xtreg probpend mietdiffwi $beschaeftigung $control, fe
est sto fe_haus

hausman pois_haus fe_haus


Is there anything I could change?

Many thanks!

[Andrew Musau]

Why would you want to compare FE and FE Poisson? Depending on the nature of your outcome, you should be able to decide which of these estimators is appropriate. For hausman, as the equations from the commands do not match, you want:

hausman pois_haus fe_haus, equations(1:1)

[John Friend]

Hi Andrew, thanks for your answer. I was just thinking of a post estimation comparison of the two estimators. Since I have count data I`ll stick to FE Poisson.

[Jeff Wooldridge]

It's not really an issue of having a count variable or not. xtreg estimates a linear model and xtpoisson estimates an exponential model. Those are very different functional forms and it makes no sense to compare the coefficients from them. One would hope the signs are the same, and one can compute partial effects for the xtpoisson that can be compared with the linear model coefficients (but that's not easy).

With a nonnegative outcome like a count, c(i)*exp(x(i,t)*b) is usually the better functional form, so it makes sense to go with it. But you can use the linear model to compare directions of effects.

[John Friend]

Thank you, Professor Wooldridge!