So far we have focused on the frailty model as a way of dealing with possible heterogeneity due to unobserved covariates. This is the main interpretation of frailty in the application to univariate time-to-event data. As discussed in the last chapter, this results in selection over time; for example, this is shown as leveling-off or crossing-over effects in population hazards. The concept of frailty introduced by Vaupel et al. (1979) to biostatistics and by Lancaster (1979) to the econometric literature originates from this kind of models.