I work in a lab where group members sometimes individually carry out an experiment of the following form:

  • pairs are bred at some definite time
  • the pairs are grouped by in experimental treatments (i.e. 25 pairs in condition A, 25 pairs in B)
  • the time to a certain event occuring is measured

Usually we would analyse this data in the lazy/standard way - i.e. we would use linear models to look for differences between group means and then ANOVA to test if the difference in group means are significant between treatments.

I have recently realised that as our data is interval-censored (i.e. we don't really record event time-points, but windows of time in which the event occured), and have begun using interval-censored survival models.

What I can't give is a coherent explanation of why the former approach is a poor one. What is actually better about survival analyses than linear models for analysing survival data?


1 Answer 1


In brief:

  1. survival times tend to be skewed and heteroskedastic; survival models have that built in. You can do that with some GLMs (like say a gamma), but survival models tend to have more readily available choices. Even with no censoring at all I sometimes use survival models (e.g. for Weibull regression) when most people might think of a GLM.

  2. You don't expect to see location shifts in survival.

  3. If you can't observe the full survival time every time, you'll have some right censoring; again, you don't have to use a survival model with censoring, but it's got it built in.

  4. You're making fuller use of the available timing information. (If the bins are narrow that maybe won't matter such a lot.)


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.