st: AW: bca' option in bootstrap causing Stata to crash?

["Martin Weiss" <martin.weiss1@gmx.de>]

<> 

What does your user-written "bs_secondstage" do? You can use the -profiler-
to show which -program-s it calls along the way. The key to the mystery
could well be there, and there is no way for the list to know. Stata does
not crash for an easy replication of your task:


*************
sysuse auto, clear
drop if mi(rep78 )
bs r(mean), seed(1992) bca cl(rep78) strata(foreign) reps(1000) : su price
*************



HTH
Martin


-----Ursprüngliche Nachricht-----
Von: owner-statalist@hsphsun2.harvard.edu
[mailto:owner-statalist@hsphsun2.harvard.edu] Im Auftrag von
Edmond.Ng@lshtm.ac.uk
Gesendet: Freitag, 28. Mai 2010 14:50
An: statalist@hsphsun2.harvard.edu
Betreff: st: bca' option in bootstrap causing Stata to crash?

Dear Statalisters, 
 
I am running some user-written code for something known as 2-stage bootstrap
algorithm on simulated clustered data where individuals are nested within
clusters. At first stage the code samples whole clusters (done by Stata
command 'bootstrap') and samples individuals within the chosen clusters at
the second stage (using a user-written code). I ask Stata to provide
bias-corrected and accelerated (BCa) confidence interval for each simulation
for assessing conf. interval coverage. 
 
However, Stata crashes on some of the simulated datasets. I have looked at
the datasets that caused the crashes but could not find anything obviously
odd. Of course, it is possible that some simulated datasets are peculiar
enough to lead to invalid or inadmissible results that could have caused
Stata to crash. 
 
Though I notice when I take the 'bca' option off from the boostrap command
line (see below) Stata no longer crashes. 
 
> bootstrap r(ct) r(cco) r(et) r(eco) r(dc) r(de) r(inb), cl(cluster) ///
 strata(treat) reps(1000) bca nodots nowarn ///
 saving(bsampling, replace): ///
 bs_secondstage cost qaly treat
 
I have two questions: 
 
1) Are there any known bug related to the 'bca' option in current version of
bootstrap? I am using Stata version 11. I found some correspondence on 'bca'
related problems in the Statalist archives, for example, the one titled
"problem with -bca- option with bootstrapping a user-written program", but
it was dated year 2005. I presume those known bugs would have been fixed by
now. 
 
2) My 2nd question is whether there is any clever trick to stop Stata from
crashing on the problematic datasets (in a forvalues loop) , skip current
one, and continue onto the next dataset. I have tried the 'capture' command
hoping to prevent Stata from crashing but to no avail. It just kept closing
after issuing the same uninformative warning message "Stata has encountered
a problem and needs to close..."  Are you aware of other tricks like that of
'try()' in R ? 

Someone suggested to me to 'set trace on' to see where the crash occurs.
Very useful suggestion. However, what was even stranger was then when I
applied the same code (with same seed) on the same problematic datasets
repeatedly Stata crashed at different points! 
 
Thank you for any suggestion/solution in advance. 
 
Best wishes,
 Edmond 
 
Health Services Research Unit, 
London School Hygiene & Tropical Medicine, 
London , UK 


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/