I'm attempting to model some data using regression analysis that has different upper limits for different observations, i.e. there is no universal limit for the entire data set. However I do have observations that have been censored, but the amount of censoring will sometimes change between observations? As I understand it the censReg R package accepts one universal upper limit, is there another option either within this package or in another package/language that would allow me to model observations with different upper limits?

If there is no other option then would using a similar approach to zero-inflated models accurately model this behavior? And by that I mean build a model to estimate when an observation might be censored and include that in the final distribution of what I am trying to predict. That's the only approach I can think of that might help with this distribution behavior.


Yes, censReg only supports a scalar censoring point, i.e., constant over observations.

As an alternative you may use the crch package whose crch() function supports left and right limits that may be varying over observations. Additionally, it optionally allows to model heteroscedasticity conditional on regressors.

Furthermore, the survival package's survreg() function also supports arbitrary censoring patterns which have to be set up through a suitable Surv() object. It provides more response distributions than the two aforementioned packages.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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