[R] tobit regression model

[Achim Zeileis]

On Thu, 10 Mar 2011, Jen wrote:

> Hi,
> I'm trying to fit a tobit regression model to some data. When fitting the
> exact same data in Stata, I have no problems at all, however R won't
> converge. Its not a maxiters thing, since I've tried increasing this
> already. I need to be able to fit the model in R since there are users of
> the code that don't have a Stata license.
>
> The code is:
>
> require(AER)
> left = 3.218476
> x = c(0,6,12,18)
> y = c(10.065819,7.803843,5.164786,3.218476)
> mod<-tobit(y~x, left = left, right = Inf)
>
> This gives back the following warning:
> Warning in survreg.fit(X, Y, weights, offset, init = init, controlvals =
> control,  :
>  Ran out of iterations and did not converge
>
> Has anyone come across this problem before or know a way to fix the 
> problem?

For such a dataset, it's unlikely that the optimization may not be 
straightforward. I'm not sure what exactly Stata does and what survreg() 
does for choosing starting values and for declaring convergence. But in 
case of survreg() you can check the docs/code, and I would expect that 
also Stata documents this sufficiently well. This may well be the source 
of the differences in this small example.

What I did was simply providing my own starting values. In this case, the 
OLS regression coefficients should be rather close to the tobit result. 
Thus I did

m1 <- lm(y ~ x)
m2 <- tobit(y ~ x, left = 3.218476, init = coef(m1))

for which convergence is declared.

If anyone wants to use survreg() directly instead of the tobit() 
interface:

y2 <- Surv(y, y > 3.218476, type = "left")
m3 <- survreg(y2 ~ x, dist = "gaussian", init = coef(m1))

which is exactly what tobit() does internally...

hth,
Z

> Thanks in advance,
> Jennifer
>
> --
> View this message in context: http://r.789695.n4.nabble.com/tobit-regression-model-tp3345789p3345789.html
> Sent from the R help mailing list archive at Nabble.com.
>
> ______________________________________________
> 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.
>