Hi Statalist,

I am running OLS for binary dependent variable after facing so many difficulties achieving convergence using Logit.

While doing so, I observed that OLS does not drop observations that perfectly predict outcomes as Logit would do.

My questions are : 1) How does OLS deals with observations that perfectly predict the outcome? These all 0 or 1 observations do not have any variation.

2) Should the OLS results with full sample and the OLS results after dropping observations that perfectly predict the should be the same?
(it is not the same for my study and wonder what is the reason)

I would appreciate it if you provide any helpful advice!