[Smoking and mortality. Evaluation of statistical analyses of causality]

Present quantification of "death by smoking" is based on unreliable, biased generalizations of retrospectively collected data. Big samples cannot heal the bias. Not even causality can be derived. Even less possible is an adequate estimate of the degree of the risk of death caused by "smoking" by means of epidemiological methods. The first aim of the paper is a discussion of the difficulties to prove causality with statistical procedures. Defining four variants of excess mortality it is proved that these different definitions are mixed in the current problem discussion. Besides it is difficult, if not impossible, to find appropriate estimates of the degree of dependency considering the necessary complexity of the model.