0
$\begingroup$

When we have binary dependent and independent variableS (and binary Instrumental variable). in STATA which command gives better estimates either ivregress 2sls or ivrprobit ?. I am not estimating any margins, I just want to identify the causal relationship between my dependent and independent variable. I am little confused on which command to use.

$\endgroup$

1 Answer 1

0
$\begingroup$

Siri, when you have a binary outcome, you need to assume a mapping function (the probability distribution for outcomes between 0 and 1). This needs to be assumed - you cannot test what form this underlying mapping function takes.

A practical advice that suffices for most applied work: You should use and report a probit, logit, and linear probability model, the latter being the ivregress 2sls command. Ideally, if you are not somewhere at the tails, the three models give very similar estimates (not on the absolute level of the coefficient though, but the marginal effects), as well as similar significance levels. (If not, the choice of the "right" model becomes much more difficult, because there is no "right" model cause you do not know what the true underlying mapping function is.)

See also: How to choose between logit, probit or linear probability model?

https://blogs.worldbank.org/impactevaluations/whether-to-probit-or-to-probe-it-in-defense-of-the-linear-probability-model

Or "mostly harmless econometrics" for an introduction into the topic.

Good luck.

$\endgroup$

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.