Inference from sample splits when coefficients are not significant (Panel Data)


I’m new to this forum and have got a little stata problem.

Status Quo:

I use panel data and a fixed effects model. My independent variable is negative and statistically significant over the whole sample.

To test a theory I split the sample in two halves (A+B). Theory predicts that the independent variable will be more negative for sub-sample A than for sub-sample B. That’s why I executed the regressions for both sub-samples and compared the coefficients.


The coefficient is only statistically significant for sample A, but not for sample B. :mad:


  1. What does this result tell me about the theory (Nothing, I assume). :confused:
  2. I was wondering whether there is another way to test the theory. Is there something like a paired t-test for regression coefficients?




This sounds like an interaction - your hypothesis is that the effect of your independent variable is lower in sample A, which you can test with an interaction between the independent variable and the sample.
Thanks for you quick reply!.

How can I implement this in stata? Until now I used the "xreg dependentvariable independentvariable controllvariables, fe" command.