Testing Equality of Coefficients across Two Regressions

[james xu]

Dear all,
I am trying to conduct a statistical test whether the coefficients from two second-stage regressions are significantly different from each other.
My two regressions looks like:

ivreghdfe DISP  log_ME  (h_diffs =test lngdp epu_state) if log_ME>=r(p50), a(cusip st year) cluster(st_year)
ivreghdfe DISP  log_ME  (h_diffs =test lngdp epu_state) if log_ME<r(p50), a(cusip st year) cluster(st_year)

I was trying to use the following steps:

ivreghdfe DISP  log_ME  (h_diffs =test lngdp epu_state) if log_ME>=r(p50), a(cusip st year) cluster(st_year)
estimates store cequalzero
ivreghdfe DISP  log_ME  (h_diffs =test lngdp epu_state) if log_ME<r(p50), a(cusip st year) cluster(st_year)
estimates store cequalone
suest cequalzero cequalone

From the last step, I got the error message saying that

cequalzero was estimated with cluster(st_year).
re-estimate without the cluster() option, and
specify the cluster() option with suest.

Can anyone inform me how to correct my code? Thank you so much!

[Joseph Coveney]

Can anyone inform me how to correct my code?

Well, given the error message, maybe try this (in red) for starters.

ivreghdfe DISP log_ME (h_diffs =test lngdp epu_state) if log_ME>=r(p50), a(cusip st year) // cluster(st_year)
estimates store cequalzero
ivreghdfe DISP log_ME (h_diffs =test lngdp epu_state) if log_ME<r(p50), a(cusip st year) // cluster(st_year)
estimates store cequalone
suest cequalzero cequalone, vce(cluster st_year))

[james xu]

Well, given the error message, maybe try this (in red) for starters.

ivreghdfe DISP log_ME (h_diffs =test lngdp epu_state) if log_ME>=r(p50), a(cusip st year) // cluster(st_year)
estimates store cequalzero
ivreghdfe DISP log_ME (h_diffs =test lngdp epu_state) if log_ME<r(p50), a(cusip st year) // cluster(st_year)
estimates store cequalone
suest cequalzero cequalone, vce(cluster st_year))

Hello Joseph,
Thank you for your advice. From my understanding, you only clustering the SE for the suest command in the last line, correct? However, if I do so, I get a new error message saying

unable to generate scores for model cequalzero
suest requires that predict allow the score option

What could be wrong with this error message? Thank you

[Joseph Coveney]

. . . you only clustering the SE for the suest command in the last line, correct?

Yes. That is what the error message and the help file instruct.

I get a new error message . . . What could be wrong with this error message?

It seems that the user-written command ivreghdfe doesn't make the scores available as required and as stated in the help file for suest. But I am completely unfamiliar with ivreghdfe and you might want to take this up with the command's author.

[Andrew Musau]

You can look at this: https://www.statalist.org/forums/for...ferent-samples. But with ivreghdfe (from https://github.com/sergiocorreia/ivreghdfe), you'd need to interact the fixed effects as well.

[james xu]

You can look at this: https://www.statalist.org/forums/for...ferent-samples. But with ivreghdfe (from https://github.com/sergiocorreia/ivreghdfe), you'd need to interact the fixed effects as well.

Hi Andrew,
Thanks for your response. May I ask what you mean by

interact the fixed effects

[Andrew Musau]

May I ask what you mean by

interact the fixed effects

gen group= log_ME>=r(p50)
ivreghdfe ..., a(0.group#(cusip st year) 1.group#(cusip  st year)) ...