0
$\begingroup$

I have a longitudinal dataset, so for each company different year observations. The time period of the dataset is 1993 to 2008. I tested a FGLS model on the whole dataset.
Now I want to test the FGLS model on three separate models (1993-1999)-(2000-2002)-(2003-2008).
More specifically I want to compare the regression coefficients of the three models.

So my questions are:

  • What would be a appropriate statistical test to compare the coefficients?
  • What is the code for the test in stata?
Improve this question
$\endgroup$
1
  • $\begingroup$ If the three models use strictly the same IVs and DV you can compare their standard errors and p-values. You should specify what is the goal of your research to have a more accurate answer tough. I'm not clear why you would split the sample size like this. $\endgroup$
    – Robert Kubrick
    Commented May 12, 2012 at 14:44

1 Answer 1

Reset to default
1
$\begingroup$

Since the single time-coefficient model is nested in the multiple-time-coefficient model, you could fit both and do an F test. You will have 2df in the nominator (2=3-1). If it rejected, the single time coefficient assumption is implausible.

Improve this answer
$\endgroup$
2
  • $\begingroup$ this approach would force the same covariance structure onto all three periods. This would bring efficiency gains if the structure indeed stayed the same, but it would also imply efficiency losses if it actually changed from one period to another (it would suffice for the variance to go up, for instance). $\endgroup$
    – StasK
    Commented Oct 11, 2012 at 14:38
  • $\begingroup$ @StasK: I think that the suggested calculs of DF does actually allow for same-coefficient-different-covariance. On the other hand, I am pretty sure a naive run in Stata will indeed not allow for a different covariance along the periods. $\endgroup$
    – JohnRos
    Commented Oct 12, 2012 at 21:01

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.