Tell me more ×
Cross Validated is a question and answer site for statisticians, data analysts, data miners and data visualization experts. It's 100% free, no registration required.
up vote 2 down vote favorite
1

I have a dataset with 43,422 observations and a left-censored (at 0) dependent variable. Of the $n$ observations, 42,536 are left-censored and 886 are not. I plan on analyzing this with a Tobit model and discovered there are three different implementations in R (AER, VGAM, and censReg). Without thinking through a model, I decided to try regressing against a few of the other variables in the dataset to get a feel for presentation. The model is nonsensical, but that should not be the point. Much to my surprise, I get wildly different answers for the same model for the AER and VGAM implementations of Tobit, and censReg never returns. So I tried again in Stata, which at least is returning reasonable responses (correctness is not verified).

My question is, what could cause the AER, VGAM, and censReg implementations to behave this way? Is it possible my model is too censored?

share|improve this question
2  
This is probably a bad combination of identification issues (too many 0s) and numerics. Why don't you stick with Stata if it works? – StasK Mar 2 '12 at 17:43
I may have to. That will really crimp my use of Sweave, though. – James Howard Mar 2 '12 at 18:05
1  
You can use Stata file to output anything in any format if you need to. But I agree, not having a fully functional analogue of Sweave for Stata is a sizable limitation. – StasK Mar 2 '12 at 19:21
An interesting idea, I post here to see if others find it valuable. I realized ultimately, I only care if the result is non-0, so I switched to logit and probit. – James Howard Mar 3 '12 at 15:57

Know someone who can answer? Share a link to this question via email, Google+, Twitter, or Facebook.

Your Answer

 
discard

By posting your answer, you agree to the privacy policy and terms of service.

Browse other questions tagged or ask your own question.