[R] Residuals of tobit{AER} and testing distribution normality

[David Winsemius]

On Mar 25, 2014, at 11:02 AM, Michal Kvasnička wrote:

> Hallo.
> 
> What are the residuals returned by resid() in this code?
> 
> library(AER)
> m <- tobit(y ~ x1 + x2, data=dat)
> rr <- resid(m2, type = "response")
> 
> Can I use them to test the normality of the random variable distribution
> this way?
> 

I understood tobit regression to not be assuming Normal distributions, but rather truncated Normal.

> qqnorm(rr)
> shapiro.test(rr)

That's not to say that the residuals are not supposed to be Normal. But most tests of normality are insensitive with small N and overly sensitive with large N, _and_ most methods are fairly robust to departures, so normality testing is either misguided or not necessary in most situations.

> 
> And is it correct to handle heteroskedasticity of the model this way?

Correct? (... and I see not descriptions of any "handling" so did you mean "assessment" which I addressed above.)  What are the real-world costs of an incorrect answer? are you doing anything for which estimates in the tail of your distributions would be critical?

> 
> coeftest(m2, vcov=sandwich)
> 
> Many thanks for any help or hint.
> 
> Best wishes,
> Michal Kvasnicka
> 
> 	[[alternative HTML version deleted]]
> 
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.

David Winsemius
Alameda, CA, USA