In this paper we deal with different ways of statistical modeling of real world accident data in order to quantify the effectiveness of a safety function or a safety configuration (i.e. a specific combination of safety functions) in vehicles. It is shown that the effectiveness can be estimated along the so-called relative risk, even if the effectiveness does depend on a confounding variable, which may be categorical or continuous. In a second step the quite usual and from a statistical point of view classical logistic regression modeling is investigated. Main emphasis is laid on the understanding of the model and the interpretation of the occurring parameters. It is shown that the effectiveness of the safety function also can be detected via such a logistic approach and that relevant confounding variables can and should be taken into account. The interpretation of the parameters related to the confounder and the quantification of the influence of the confounder is shown to be rather problematic. All theoretical results are illuminated by numerical data examples.