I have two logit models: A standard binary one (y=0/1) and an ordered logit model (y = yes, definitely/yes, maybe/no, probably not/no, definitely not. Now, to get the marginal effects of the binary logit I use a Python package, where the default option for treating binary independent variables is to treat them as continuous: https://www.statsmodels.org/dev/generated/statsmodels.discrete.discrete_model.LogitResults.get_margeff.html
For the ordered logit model, I use an R package where the default is the opposite, to treat them as categorical: https://www.rdocumentation.org/packages/erer/versions/2.5/topics/ocME
I understand the difference, but I'm not sure which one is more suitable in a standard case. I have quite a few binary variables (more than 10) for my micro-oriented cross-sectional model.