I'm somewhat new to R, so I'm guessing this might be a basic question. In any case, I have some interval-censored data that I'm trying to fit a parametric model to. I've looked at the complementary log-log plot of the survival curves and have ruled out a Weibull model; the only other model my class has discussed — at least insofar as assessing the appropriateness of the model — is the log-logistic, so I figure I should try that as well. However, I'm not sure how to plot the log of the survival odds against $\log (t)$. Could anyone offer any advice on how to do this?

Beyond that, I've already run survreg on my data and hit a bit of a bump: When I tried to create the survival object using type interval2, I received a message saying that I had "invalid survival times for this distribution". When I changed the type to "interval", it worked fine. What are the reasons this might happen?

Thank you!


1 Answer 1


I'm 90% sure your problems with survreg is that it does not allow you declare having the left side of your interval to be $\leq 0$ (emphasis on equal) or the right side of your interval to be $\infty$.

If you're not married to the AFT model fit by survreg, I may biased-ly point you toward the package icenReg. This fits both semi-parametric and non-parametric Cox-PH and proportional odds models. And it allows you to have the left side of your interval be equal 0 and the right equal $\infty$.

In fact, for your problem of comparing fits, you can use the function diag_baseline. It will fit several choices of parametric baseline survival functions against the equivalent semi-parametric model of choice (i.e. proportional odds or proportional hazards).

But it does not fit a AFT model at this time.


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.