Pooled cross section (not panel) data with binary dependent

[Nadhila Adani]

Dear Statalisters,

I am working on a pooled cross section data (different individuals in two time periods. Before and after the implementation of a program). The aim is to observe the impact of a program. If the dependent is not binary, then difference in difference method is the way to do it. But what if the dependent variable is binary ( 1 and 0 ).
I tried using probit and logit and thought they both result in bias estimation.
I tried using xtlogit ....., fe (assuming household fixed effect by typing xtset hhid beforehand) and wondered if this is the right estimation. And how do I interpret the coefficients in xtlogit?

Your help is very much appreciated.

[Stephen Jenkins]

You can't use "fixed effects" methods because you don't have a panel. Before proceeding, I strongly recommend that you read up on "differences in differences" methods, starting e.g. with Angrist and Pischke's Mostly Harmless Econometrics book, or their recent Mastering Metrics book. See also Austin Nichols's very useful review at http://www.stata-journal.com/article...article=st0136. You'll see from the reading that economists tend not to get hung up on the binary dependent variable aspect when doing diff-in-diff -- they typically employ linear probability models. But one could use logit or probit nonetheless.

Welcome to Stata Forum, but please respect Forum etiquette and re-register so that you use your real name (firstname familyname) -- the FAQ explains why, and how to do this. (It's easy: hit the Contact Us link at bottom right of the page.) Also, I see that you have posted your message many many times. Be aware that the Forum server often takes time before a message appears (one often sees a "working" message). Do not repost! Be patient. If you wish to check whether a message has been posted -- it virtually always has -- open a new browser window and check the Forum pages. Thank you.

[Anton Ivanov]

What is wrong with using binary DV with logit? Keep in mind that the interpretation of the results is based on the odds ration, not simply on the coefficients (may be that is why you though of "biased estimation"?).

As for difference in difference, keep in mind that there has to be theory that explains why IV are related to differences.