I am currently working on my master thesis and I am in the middle of my data cleaning up process. I am studying the effects of certain mergers during a sample period of 20 years. I use time fixed effects and firm fixed effects, hence my regression specification is as follows : y= a+ posttreat+ e.

I am aware that observations require to be in the pre-treatment and post-treatment group when the equation is y= a+ post+ treat+ post*treat + e. However, I am wondering if this is also the case when using a regression specification which takes fixed effects into consideration.

Since I am studying firms over a period of 20 years and only limited firms cover the entire period I am wondering if it is necessary if they are included for the entire sample period (especially given the fact that some might act as treatment group or observe time fixed effects). If I would have to drop all firms which are not included in the entire sample period, it might drastically decrease the number of observations in my sample. Is it logical to set a certain threshold, for example that a firm should have at least 5 observations (years) to be included in the sample?

Can somebody please advice me on this topic? Thanks a lot in advance!


DiD can be implemented using repeated cross-sections, so there is no need to limit your data to a balanced panel.

You should still have a post dummy in the FE specification, since it is time varying. The reason treat is omitted in the FE model is because it is time-invariant.


Your Answer

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

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