How can I test for heteroscedasticity with a logit model. I use glm with family=binomial(link='logit')?



In general I don't think you need to worry about heteroskedasticity in a logit model, because your dependent variable is binary. So, your residuals are distributed in only two points on the x-axis when plotted against the fitted values of the model. It is highly unlikely that the variance of your "0" residuals is the same as the variance of your "1" residuals for any random sample for a binary variable.

In other words, the variance of your residuals is heteroskedastic by design. The variance of binomial data is determined by the mean. One number rules them all. Logistic regression is designed around this and therefore there is no assumption of equal variance. This is actually part of the impetus for using the non-linear logit method. You shouldn't need to test for or correct for heteroskedasticity; just be sure you know how to interpret the estimated effect size of the parameter estimate on the logit.

The assumptions are:

  1. linearity in log odds space
  2. independent errors
  3. multicollinearity among predictors isn't too high

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.