how to incorporate clustering in a chow test

[Yun Cheng]

Hello,

I'm doing subsample tests to compare two OLS regressions, one with above industry median level of reply (high_reply dummy == 1), the other with below industry median level of reply (high_reply dummy == 0). The two regressions are with industry and year FEs, and with standard errors clustered at the industry level

reg y x1 x2 x3 x4 x5 i.industry i.year if high_reply == 1, vce(cluster industry)

reg y x1 x2 x3 x4 x5 i.industry i.year if high_reply == 0, vce(cluster industry)

The -chowtest- command is from SSC, what I did is

chowtest y x1 x2 x3 x4 x5 i.industry i.year, group(high_reply)

I suppose sth is wrong there as the chowtest didn't consider clustering, but if I add vce(cluster industry) or sth similar, stata says option vce() not allowed. So any advice on how to consider clustering when using chowtest? THANKS

[Leonardo Guizzetti]

The Chow test is really just a fully specified test of interaction among subgroups. You can refer to the FAQ here for an explanation of what that means, and how to perform the Chow test in a more flexible way. It is straightforward to add clustering.

[Yun Cheng]

Thanks Leonardo, the website explains it very well.

[Carlo Lazzaro]

Yun:
as an aside to Leonardo's helpful suggestion, you may want to take a look at -suest-.