insufficient observations in bootstrap user written program

[Guillaume Geri]

Hi StataList
I use Stata 14.1 on OSX.
I'm running a cost-effectiveness analysis and in order to get factors associated with costs, I run a twopart model using the twopm Stata command. It perfectly works.

The problem is I need to use both multiple imputations and bootstrapping (to get correct evaluations of CI).
That's why I've written the code below

/ evaluation first part to get OR & CI in appropriate format

***************program two-pm******************************
logistic charge $var_indep
estout, c(b ci) drop(_cons) label eform
matrix matrix_logistic = r(coefs)

capture program drop boot_tpm_cond
capture program define boot_tpm_cond, eclass
// use second part to get conditional costs
glm    totalcharge_ccr    $var_indep if totalcharge_ccr>0, ///
        link(log) family(poisson)
        
margins , dydx(*) post
end

capture program drop boot_tpm
capture program define boot_tpm, eclass
// use stata twopm command to get unconditional costs
twopm    totalcharge_ccr    $var_indep, ///
        firstpart(logit)    ///
        secondpart(glm, link(log) family(poisson))
margins, dydx(*) nose post
end
***************program two-pm******************************




***************program bootstrap******************************
capture program drop bootstrapperso
program define bootstrapperso, eclass
// premier bootstrap : condtional margins
bootstrap _b, reps($rep) seed(2) bca ties nodrop: boot_tpm_cond
local n_reps_cond = e(N_reps)
matrix b_cond = e(b)
matrix ci_cond = e(ci_bca)

// deuxieme bootstrap : unconditional margins
bootstrap _b, reps($rep) seed(2) bca ties nodrop: boot_tpm
local n_reps_uncond = e(N_reps)
matrix b_uncond = e(b)
matrix ci_uncond = e(ci_bca)
end
***************program bootstrap******************************



***************mi estimate sur l'ensemble******************************
mi estimate,noisily cmdok: bootstrapperso
***************mi estimate sur l'ensemble******************************

But when I run the program bootstrapperso, the following message appears:

. bootstrapperso
(running boot_tpm_cond on estimation sample)

Jackknife replications (2085)
----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5 
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn    50
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn   100
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn   150
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn   200
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn   250
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn   300
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn   350
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn   400
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn   450
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn   500
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn   550
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn   600
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn   650
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn   700
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn   750
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn   800
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn   850
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn   900
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn   950
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  1000
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  1050
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  1100
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  1150
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  1200
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  1250
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  1300
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  1350
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  1400
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  1450
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  1500
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  1550
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  1600
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  1650
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  1700
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  1750
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  1800
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  1850
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  1900
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  1950
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  2000
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn  2050
nnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnnn
insufficient observations to compute jackknife standard
errors
no results will be saved
r(2000);

I've read several posts treating such a similar error message (like this one http://www.statalist.org/forums/search?r=42907686&p=12) but adding the no drop option did not change anything. Of course, I've not been able to run the mi estimate command.

Thank you for you help

[Maarten Buis]

You added the option bca, so it will also Jackknife your model. In order to do so, your program needs to allow the if condition. You can do so with syntax and marksample.

Having said that, it would be useless. The reason is that afterwards you intend to use mi on it, and mi will only look at the point-estimates and their standard error, and these are unaffected by the bca option.

You seem to have problems with the normal approximation for the sampling distribution of your estimates. If those deviations are severe enough for you to choose bootstrap, then you probably cannot use mi. So you probably have to make a choice for one or the other.

[Guillaume Geri]

Thank you very much Marteen for your reply.
You are totally right, the cost variable has a very right-skewed distribution.
The reason why I tried to use the bca option was with the bc option, my code worked well but some confidence intervals were missing. Could it be because I ran my model with too few replications (to check if it ran)?

[Guillaume Geri]

To be accurate, there was no bc option, but I retrieved CI bc estimation.