While we stated that logistic regression violates some of the assumptions of linear regression, it still requires the following:
(a) independence of observations,
(b) little to no multicollinearity among predictors, and
(c) a linear relationship between the predictors and response.
However, a key subtlety is that the linearity is based on the logit of the response, not the raw 0/1 values. Thus, this assumption can only really be validated after fitting the model by examining the relationship between each predictor and the fitted logit values.