As a small part of a project, I have been asked to tailor a large panel dataset I have built into a form that is more comparable to one specific country.

So I have done the following:

Look at the ranges of my dependent variable, explanatory variables and controls for the country of interest.

Drop any firm-time observations that fall outside the ranges of any of these variables.

Run my regression on the truncated sample.

I was wondering whether there are any serious cons of this approach - particularly given that my method will include only a subset of the observations for some firms. Bear in mind that this is meant to provide a rough and ready indication rather than be a super robust exercise.

Many thanks for your help.


Your approach seems no good. Let us make it really simple: You have only the dependent variable and you want to compare the averages across countries. You keep the firms whose $y$ values are inside the range. By doing this, you are (sort of) equating $y$ values, which you aimed to compare. This is strange. In fact, this leads to endogenous sample-selection.

It seems, however, acceptable to keep only comparable firms (rather than firm-year) based on the exogenous variables.

| improve this answer | |
  • $\begingroup$ Fully agree. Apart from a few exceptions (e.g. dropping people with negative incomes), selection on the dependent variable usually leads to heavily biased results. $\endgroup$ – E. Sommer Jun 1 '18 at 8:09

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.