Testing the difference in coefficients between two subsets

[Giovanni Coppola]

Good Evening everyone, I am asking your help in order to test whether the coefficient CSR that I am using as my main independent variable, differs within subgroups of the sample. I understood from previous posts that there are two ways to test the difference between coefficients: one it's about using <suest> command, another implies the creation of an interaction term within the regression and then a simple <test> command.


For my baseline regression I used <xtreg> and I saw that I cannot use <suest> since it is not supported for xtreg. So I decided to go for the creation of an interaction term. My baseline regression which tests whether CSR has an effect on cash is the following:


<xtreg> Cash_w CSR Size_w Leverage_w CAPEX_w SGA_w Dummy_2014 Dummy_2015 Dummy_2016 Dummy_2017 Dummy_2018 Dummy_2019, fe


and l guess it should be like that with the interaction term:


<xtreg> Cash_w CSR Size_w Leverage_w CAPEX_w SGA_w i.LitigationRisk##c.(CSR Size_w Leverage_w CAPEX_w SGA_w) Dummy_2014 Dummy_2015 Dummy_2016 Dummy_2017 Dummy_2018 Dummy_2019, fe

(Litigation risk it is a dummy variable which equals 1 for the subset of firms which are classified as having a higher litigation risk and 0 for firms characterized by a low litigation risk)


Now I would like to test if the difference between the CSR coefficients of two the subset groups with Litigation Risk = 1 and = 0 is significant. How should I use the <test> command, which is the right formulation for the two coefficients? Is the model with the interaction term correct?


Thanks a lot!

[Bruce Weaver]

Why do you need anything beyond the test on the coefficient for LitigationRisk#c.CSR?

[Giovanni Coppola]

Hi Bruce, so if I understood well, I just need to run the command <test> LitigationRisk#c.CSR to see the difference in the coefficients between the two litigation groups?

[lal mohan kumar]

Hi Giovanni Coppola
There are many experts in the platform who will respond but I have few comments and my job will be as an intermediary who can suggest to you some sources

1. If Dummy_2014 Dummy_2015 pertains to year dummies why don't you try i.year instead of using these year dummies.

2. Please check the below link for testing the significance of coefficients in two subsamples

https://www.statalist.org/forums/forum/general-stata-discussion/general/1449783-comparing-coefficients-of-two-equal-fe-regressions-from-two-subsamples?p=1450018#post1450018

[Fei Wang]

Hi Bruce, so if I understood well, I just need to run the command <test> LitigationRisk#c.CSR to see the difference in the coefficients between the two litigation groups?

Giovanni, you don't need to run more commands. Just look at the coefficient of LitigationRisk#c.CSR -- it means the difference in the effects of CSR on Cash_w between the subgroups -- if it's statistically significant (small p value), then the effects significantly differ.

[Giovanni Coppola]

Thanks Fei, I realized it after my reply to Bruce.