RE: st: svy bootstrap

["Michael Palmer" <Michael.Palmer@anu.edu.au>]

I don't know how to do this. I have spent the day reading Stata manuals and various other references. 
Obviously the below do file is wrong (error message 'invalid syntax'). I gather that I must write a separate program for the weights and then incorporate it in the below twopart program. I am at a loss, any help would be greatly appreciated.  


capture program drop twopart
program twopart, rclass
tempname sm1 sm0 m_1 m_2 m_3 m_4 
tempvar pr_y yhat ehat smear smear_ins sm1 smear_noins sm2 yhat_duan yhat_pwd1 yhat_pwd0 y_pwd1 y_pwd0 
syntax varlist [pweight] [=exp] [if] 
syntax newvarlist [pweight] [=exp] [if]
logit inpat_public $xlist1 if age > 5 `weight' `if' `exp'
predict `pr_y', pr
reg ipexp_public_log $xlist1 if age > 5 & inpat_public==1 `weight' `if' `exp'
predict `yhat', xb
predict `ehat', residuals

egen `smear_ins' = mean(exp(`ehat')) if insurance==1  
summ `smear_ins' `typlist' `if' `exp'
scalar `sm1' = r(max)
egen `smear_noins' = mean(exp(`ehat')) if insurance==0 
summ `smear_noins' `typlist' `if' `exp'
scalar `sm0' = r(max)
gen `yhat_duan' = exp(`yhat')*`sm1' if insurance==1 
replace `yhat_duan' = exp(`yhat')*`sm0' if insurance==0

gen `yhat_pwd1' = `yhat_duan' if pwd==1 
mean `yhat_pwd1' `typlist' `if' `exp'
scalar `m_1' = _b[`yhat_pwd1']
gen `yhat_pwd0' = `yhat_duan' if pwd==0 
mean `yhat_pwd0' `typlist' `if' `exp'
scalar `m_2' = _b[`yhat_pwd0']
gen `y_pwd1' = `pr_y' if pwd==1 
mean `y_pwd1' `typlist' `if' `exp'
scalar `m_3' = _b[`y_pwd1']
gen `y_pwd0' = `pr_y' if pwd==0 
mean `y_pwd0' `typlist' `if' `exp'
scalar `m_4' = _b[`y_pwd0']

return scalar prod2 = `m_1'*`m_3' - `m_2'*`m_4'
end

set more off
twopart
return list
bsweights bsw, reps(100) n(-1)
bs4rw (prod2=r(prod2)), rw(bsw*): twopart [pw=psindwt]

Best regards,  


Michael Palmer
PhD Candidate 
National Centre for Epidemiology and Population Health 
The Australian National University 
Ph. 6125 0538
M. 0437 867 940

-----Original Message-----
From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Stas Kolenikov
Sent: Monday, 7 March 2011 11:26 PM
To: statalist@hsphsun2.harvard.edu
Subject: Re: st: svy bootstrap

Take a look at the [R] bootstrap entry and their running examples.
Yours will run along the similar lines, and have [pw] where applicable. Remember, unlike the -bootstrap- that actually changes the data in memory, the survey bootstrap only substitutes different sets of weight variables.

On Mon, Mar 7, 2011 at 2:24 AM, Michael Palmer <Michael.Palmer@anu.edu.au> wrote:
> Thank you for your response. Your explanation makes sense and I'll be sure to cite your paper. But I'm not exactly sure how I might rewrite the program so that it accepts pw. Do you happen to have any example programs that you could send me or better yet offer any advice on how to rewrite my program?
>
> Best regards,
>
> Michael Palmer
> PhD Candidate
> National Centre for Epidemiology and Population Health The Australian 
> National University Ph. 6125 0538 M. 0437 867 940
>
> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu 
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of Stas 
> Kolenikov
> Sent: Monday, 7 March 2011 3:24 PM
> To: statalist@hsphsun2.harvard.edu
> Subject: Re: st: svy bootstrap
>
> The command -mean- had difficulties understanding what the variable name was. It was expecting a variable name, but -r(prod2)- was not a variable. Of course it is not a factor variable or a time series lagged variable, but that's the best error message Stata could send to you, given that nobody would put anything but variable names after -mean-.
>
> To make your program work with -bs4rw- or -svy, vce(bootstrap)-, you 
> need to rewrite it so that it accepts [pw] or [iw] and runs all 
> estimation commands with those weights... and produces something 
> reasonable in the end. Then you can run -bs4rw (prod2 = r(prod2) ), 
> rw( bsw* ) : twopart [pw=original weight]- to obtain your bootstrap 
> standard errors. (I would not trust 50 replicates that much, and it 
> should be pretty easy to increase this number; it does not look like 
> you have any extraordinary difficult computations within your
> program.)
>
> I see that you found my -bsweights- command useful, and I hope you will cite my SJ paper documenting this command.
>
> On Sun, Mar 6, 2011 at 10:20 PM, Michael Palmer <Michael.Palmer@anu.edu.au> wrote:
>> Dear Statalist,
>>
>> I am trying to bootstrap standard errors in a two-part model using 
>> complex survey data. The first model is a logit and the second a log 
>> linear model of expenditures. From the below do file I get the 
>> following error message. Any help is greatly appreciated.
>>
>>  bsweights bsw, reps(40) n(-1)
>>  svyset [pweight=psindwt], strata(domain) psu(commcode) 
>> vce(bootstrap)
>> bsrweight(bsw*)
>> . svy: mean r(prod2)
>> (running mean on estimation sample)
>> factor variables and time-series operators not allowed an error 
>> occurred when bootstrap executed mean r(101);
>>
>
> --
> Stas Kolenikov, also found at http://stas.kolenikov.name Small print: 
> I use this email account for mailing lists only.
>
> *
> *   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/
>



--
Stas Kolenikov, also found at http://stas.kolenikov.name Small print: I use this email account for mailing lists only.

*
*   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/