Hello,
I am trying to model a weibull model with sample selection in stata 8.2.
My problem is that Stata produces unsatisfactory output. I have the
feeling that rounding in the maximization routines contribute to these
results but I am not sure. I provide the output from my do-file.
sysuse cancer, clear
(Patient Survival in Drug Trial)
.
. gen drug2 = drug == 2
. gen drug3 = drug == 3
.
.
. /*
> WEIBULL MODEL WITH SAMPLE SELECTION FROM RIGHT TRUNCATED SPELL DATA
(OUTFLOW SAMPLE)
>
> The likelihood for this model is given by
>
> (1) L = f(t) / F(t) (control for a
overrepresentation of short spells relative to long spells)
>
> with weibull density function (see Maximum Likelihood Estimation with
Stata (2nd edition), 2004: page 221)
>
> f(t; e, g) = (e/g) *(t/e)^(g-1) * exp(-(t/e)^g
>
> and failure function
>
> F(t; e, g) = 1 - exp(-t/e)^g
>
> with parameters (gamma) g and (eta) e=exp(Xb)
> */
.
. capture program drop myweibull_sampleselection_lf
.
. program myweibull_sampleselection_lf
1. version 8.1
2. args lnf leta lgam
3. tempvar p M R
4. quietly {
5. gen double `p' = exp(`lgam')
6. gen double `M' = ($ML_y1*exp(-`leta'))^`p'
7. gen double `R' = ln($ML_y1)-`leta'
8. replace `lnf' = -`M' + $ML_y2 * (`lgam' -
`leta' + (`p'-1) * `R') - ln(1-(exp(-`M')))
9. }
10. end
.
.
. ml model lf myweibull_sampleselection_lf (lneta: studytime died =
drug2 drug3 age)/ lngamma
.
. ml maximize
initial: log likelihood = -742.8257
alternative: log likelihood = -355.05562
rescale: log likelihood = -192.23867
rescale eq: log likelihood = 452.20041
Iteration 0: log likelihood = 452.20041 (not concave)
Iteration 1: log likelihood = 452.31598 (not concave)
Iteration 2: log likelihood = 452.35131 (not concave)
Iteration 3: log likelihood = 452.35485
Iteration 4: log likelihood = 452.36837 (not concave)
Iteration 5: log likelihood = 452.36972 (not concave)
Iteration 6: log likelihood = 452.36999 (not concave)
numerical derivatives are approximate
flat or discontinuous region encountered
numerical derivatives are approximate
nearby values are missing
Iteration 7: log likelihood = 452.37005 (not concave)
numerical derivatives are approximate
flat or discontinuous region encountered
numerical derivatives are approximate
flat or discontinuous region encountered
Iteration 8: log likelihood = 452.37005 (not concave)
numerical derivatives are approximate
flat or discontinuous region encountered
numerical derivatives are approximate
flat or discontinuous region encountered
I would really appreciate anyone's ideas about this issue.
Best regards,
Jasper Kok
Ph-D student
Maastricht University
Netherlands
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
