I am running a regression where the dependent variable is income gain. In about sixty percent of the cases the intervention fails and there is no income gain. Will having such a high proportion of values at one level distort the regression or its interpretation? I would think it would reduce variation although there are about 13 thousand cases.