Applying OLS regression to dichotomous dependent variables is problematic for several reasons. A major reason is that it violates some important assumptions: The errors cannot be normally distributed, and they cannot have a constant variance. In addition, such a linear model does not confine the level of the variable to a valid, discrete category. That is, predicted values based on such a model would fall outside of the logical range of zero and one.
Last Modified: 02/14/2019