Supporting Parallel Trends (The Effect, Videos on Causality, Ep 54)

[@specfunctor]

Thanks for this great content!

[@siddharthkaushal5194]

great video, thank you! :)

[@schafer4935]

Thank you for the good video. I have a question. In your model Y=intercept+ß1*Time++ß2*Time * Group + E. We have to test wheter ß2 is different from zero. But how about a model in following set-up. Y = intercept + ß1 * Group + ß2 * Time + ß3 * Group * Time + E. Do I have to test (for parallel trend assumption), if ß1 differs from zero for the pre period? I am a bit confused because to my understanding ß1 depicts the effect of an observation being part of the treatment group and if it differs from zero, than there are (significant) differences and PTA doesn´t hold. Am I rigth? Very best regards!

[@NickHuntingtonKlein]

@@schafer4935 no, beta-1 would not be what you'd rest. In that model (assuming you've estimated it using only pre treatment data) you'd want to look at beta-3. This would test equality of (linear) prior trends.

[@samarth1906]

Great video! Just had a question as to why we're only adding the group fixed effects when testing the parallel trends assumption and not year fixed effects as well, thanks!

[@NickHuntingtonKlein]

You could add them (and if you imagined some sort of nonlinear shared trend combined with linear nonshared trends it might be a good idea) but in general in this test we want the time trend to show up in the time trend coefficients rather htan being absorbed by the time fixed effects.

[@samarth1906]

@@NickHuntingtonKlein Thanks! Could we also further relax parallel trends when testing for it by including covariates that vary across groups and time?

[@NickHuntingtonKlein]

@@samarth1906 Yes but be very careful how you do it, see my recent Substack post nickchk.substack.com/p/controls-in-difference-in-differences