Absorbing many fixed effects when using -heckman-?

[Jen Zhen]

Dear list members,

I efficiently estimated a Regression with many fixed effects using -areg-.

Now I'd like to explore whether the results are robust to accounting for selection into being observed, using -heckman-.

But I am facing the problem that given my computer and Stata version I cannot explicitly include all fixed effects (about 7'500 in total, estimated on a sample of about 25'000 observations). Nor can I easily absorb them as in -areg-. What is the best way around this problem?

I thought of demeaning all variables from the mean within each of the 7'500 Groups, but given that I have many variables in the Regression that seems quite complicated as well?

Then I thought of estimating the two steps of the Heckman procedure manually, but then I face the same Problem when estimating the first stage using -probit-, plus I would not be sure whether I would end up with the correct standard errors.

Any suggestions would be much welcome.

Thank you so much and kind regards,
JZ

[Carlo Lazzaro]

Jen:.
Some remarks about your model:
- you seem to have too many predictors (and groups) for your sample size. Hence, the first advice would be to be more parsimoniuos and reduce the number of predictors. Please consider that there should be 20 observations per predictor (Katz MH. Multivariable Analysis. Second Edtion. NY: Cambridge University Press, 2006: 81), even though 10 obs per predictor may sound wise enough;
- please provide the list with what you typed under -areg- and what Stata gave you back (as per FAQ). Thanks.

[Joao Santos Silva]

Dear Jen,

I believe the kind of trick used in -areg- does not work with the non-linear model you have in mind.

Moreover, I would like to echo and reinforce Carlo's concern: in general, the square of the number of parameters divided by the number of observations should be small (that is, it should go to zero as the sample size increases). That is certainly not what you have and so, as Carlo noted, you should reconsider the specification of your model.

Best regards,

Joao

[Freddie Schicko]

please excuse me for digging up this thread but I have the same question:

So is it possible (syntaxwise) to include many fixed effects in a heckman model? at the moment I get r(103) when I try to a specific kind of model (that I like very much in 1-step OLS but heckman framework is the correct setting, given a selection issue)