Abstract: This paper aims to provide an analysis of the statistical effect of selection on the basis of survival (survivorship bias) useful for estimation and forecasting purposes. The central point in the paper is an analysis of why the standard survivorship bias corrections suggested in Finance and embodied, for instance, in the CRISP database of mutual funds, are undubtedly useful but do not solve the survivorship bias problem. Different survival models imply, in general, different estimates and, in particular, different forecasts for the same dataset even when the dataset has been “completed” by adding all “dead” entities. Hence, the simple use of “survivorship free” datasets does not avoid survivorship bias.