Discussion on: ‘Simple or complex statistical models: Non-traditional regression models with intuitive interpretations’ by Gillian Z. Heller

1 Comment by Jan Beyersmann My comment focuses on the time-to-event part of Professor Heller's paper but will also lean on the binary outcome part for discussion of the proposed concept of 'simplicity'. For survival data, key points made by Professor Heller include (in my words): 1. Hazards are not easy to interpret. 2. Parametric models are underused. 3. Accelerated failure time (AFT) modelling (and, perhaps, generalizations of it) provides an attractive interpretation for survival times T ∈ (0, ∞). I contend that there is general agreement on these points in survival analysis, although it does not necessarily translate into statistical practice. Hence, reading Professor Heller's fresh take on the subject, rightfully putting a spotlight on the advances made in distributional regression, was a pleasure. The aim of my comment is to argue that not one approach solves it all and why hazards have been and arguably will continue to be central to time-to-event analyses: they are the identifiable quantities and work in very general event history scenarios beyond the T ∈ (0, ∞) situation subject to (random) right-censoring.