Problems with using odds ratios as effect sizes in binary logistic regression and alternative approaches
Possibility of heteroskedasticity is a good reason not to dichotomize a continuous variable for use as outcome in logistic regression.
Should you perform logistic regression on a dichotomized continuous variable when you have access to the continuous variable? I'm not sure.
Using binary regression software to model ordinal data as a multivariate GLM
Using glmer() to perform Rasch analysis