I have a database of applicants for a grant and their applications for each year. I want to know the effect of the grant on nr of publications. But first i want to perform a common trend analysis. i did this in STATA.

My dependent variable is nr_publ = number of publications.

My independent variable is grant = 0 for non-granted, 1 for granted

I created time dummies for each year in my pre-treatment period ( year of application - 5 until year of application + 2) and each year in my post-treatment period (year of application +2 until year of application +8. represented by tt_X = years before application, tt0 year of application and ttX year of application

Secondly, i created interaction terms of the time dummies and the receipt of grant (also dummy which is 1 when the researcher obtained a grant and 0 if he did not), by grant * timedummy, again grant_tt_x = x years before application, grant_tt0 = year of application and grant_ttX is x year after application

i thought the appropriate regression was:

xtreg nr_publ grant tt_4 tt_3 tt_2 tt_1 tt0 tt1 tt2 tt3 tt4 tt5 tt6 tt7 tt8 grant_tt_4 grant_tt_3 grant_tt_2 grant_tt_1 grant_tt0 grant_tt1 grant_tt2 grant_tt3 grant_tt4 grant_tt5 grant_tt6 grant_tt7 grant_tt8,robust fe

i left out the first pre-treatment year (tt_5, and the interaction term for this year grant_tt_5)

However, this gives me a negative estimates for all interaction terms which is unlikely. I assume that i did something wrong.

can somebody explain to me, what is the correct regression? or what i did wrong?

  • $\begingroup$ But are the estimates statistically significant? That's usually the measure by which this model for the test of parallel trends is judged. If the difference between groups in one pre-grant period is not statistically different from the difference in another pre-grant period, then that's an indication that trends are parallel. You can also perform a joint test on all pre-grant interactions. For more info: stats.stackexchange.com/a/160361. (Not seeing anything wrong with the code, if vars were created correctly. Did you check that # of obs. and # of groups in output is correct?) $\endgroup$ – AlexK May 10 '19 at 23:10
  • 1
    $\begingroup$ Also, plotting the means for each period in each group should help. $\endgroup$ – AlexK May 11 '19 at 0:07

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.