Logistic regression and log-linear models in social research: how to make their application to complex tables more understandable