When is it appropriate to use linear regression for a binary outcome?

I understand that it is conventional to use logistic regression for a binary outcome because it generates a linear list of outcomes which avoids the problem of generating estimates greater than 1 or 0. However my professor advised me to use a linear model to predict my outcome. She said that it is actually quite common in Economics to do this. Does anyone know why using OLS regression could be deemed appropriate?

  • 2
    $\begingroup$ Hi @Sam Finegold and welcome to the site. As you have correctly said, the standard approach if you have a binary outcome is logistic regression. If you use OLS for this kind of data, this is called Linear Probability Model. The coefficients express the change in probability that $Y=1$ for a unit change in $X$. But this approach has shortcomings, e.g. that the predictions are outside the range of 0, 1. And there is heteroskedasticity by design. $\endgroup$ – COOLSerdash May 31 '13 at 21:19
  • 1
    $\begingroup$ I frankly don't see what advantages Linear Probability Models have compared to logistic regression. I'd stick to logistic regression as this deals naturally with the shortcomings of an OLS with a binary outcome. $\endgroup$ – COOLSerdash May 31 '13 at 21:27