EN
Applying logistic regression to small-sized data sets very often leads to the problem of complete separation. Generally speaking, separation is caused by a linear combination of covariates that perfectly separates successes (events) from failures (non-events). In such cases, results obtained by maximum likelihood method should not be trusted, since at least one parameter estimate diverges to infinity. A systematic review of the literature resulted in two theoretically sound procedures which always arrive at finite estimates, i.e. those of H. Heinze and S. Schemper (2002) and also R. Rousseeuw and C. Christmann (2003). The main goal of the paper is to compare them.