Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Block bootstrapping in quantile regression


From   Jorge Eduardo Pérez Pérez <perez.jorge@ur.edu.co>
To   "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu>
Subject   Re: st: Block bootstrapping in quantile regression
Date   Wed, 18 Nov 2009 23:37:38 -0500

This works perfectly. Thank you !
_______________________
Jorge Eduardo Pérez Pérez



On Tue, Nov 17, 2009 at 7:15 AM, Austin Nichols <austinnichols@gmail.com> wrote:
> Then try:
>
> clear all
> program iqr, eclass
> version 10.1
>  if replay() {
>  syntax [anything] [, Level(real 95) ]
>  eret di, level(`level')
>  }
>  else {
>  qui {
> syntax [varlist] [, Q1(real .25) Q2(real 0.75) Level(real 95) *]
> tab i, gen(_d)
> drop _d1
> gettoken depv vl:varlist
> tempname b1 b2 beta
> qreg `depv' `vl' _d*, quantile(`q1') `options'
> matrix `b1'=e(b)
> tempvar e1 e2
> g `e1'=e(sample)
> qreg `depv' `vl' _d*, quantile(`q2') `options'
> g `e2'=e(sample)&`e1'
> count if `e2'
> loc N=r(N)
> matrix `b2'=e(b)
> matrix `beta'=`b2'-`b1'
> ereturn post `beta', dep(`depv') e(`e2') obs(`N')
> eret local cmd="iqr"
> drop _d*
>  }
> eret di, level(`level')
>  }
> end
> sysuse nlsw88
> egen c=group(ind occ), label
> drop if mi(c)
> bs, rep(20) cl(c) id(i) seed(4):iqr wage age grade
>
>
> 2009/11/16 Jorge Eduardo Pérez Pérez <perez.jorge@ur.edu.co>:
>> Thanks for the prompt response. Although the program works for the
>> case you posted, it doesn't work for my particular application,
>> because I have to include cluster-specific dummies (fixed effects)
>>
>> sysuse nlsw88, clear
>> egen c=group(ind occ), label
>> drop if mi(c)
>> xi: bs, rep(20) cl(c) seed(4):iqr wage age grade i.c
>> Bootstrap replications (20)
>> ----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5
>> xxxxxxxxxxxxxxxxxxxx
>> insufficient observations to compute bootstrap standard errors
>> no results will be saved
>> r(2000);
>>
>> I tried generating the dummy set before instead of using xi in the
>> estimation command, and It didn't work either
>>
>> xi i.c
>> bs, rep(20) cl(c) seed(4):iqr wage age grade _I*
>>
>> Finally, i thought it was because some incompatibilty between the
>> cluster() option in bootstrap and including the dummy set, but it is
>> not the case either
>>
>> bs, rep(20) seed(4):iqr wage age grade _I*
>>
>> Yet using the command without bootstrapping works fine
>>
>>  xi: iqr wage age grade i.c
>>
>> I looked at the trace of the command but it is not particularly
>> helpful. Could this be due to insufficient iterations for each
>> quantile regression in each step of the bootstrap? I tried adding a
>> wlsiter(100) option, but  it didn't work either.
>>
>> Thank you !
>> _______________________
>> Jorge Eduardo Pérez Pérez
>> Joven Investigador
>> Facultad de Economía
>> Universidad del Rosario
>> Calle 14 #4-69 Bogotá, Colombia.
>> Tel: (571) 2970200 ext 7836
>>
>>
>>
>> On Mon, Nov 16, 2009 at 9:13 PM, Austin Nichols <austinnichols@gmail.com> wrote:
>>> clear all
>>> program iqr, eclass
>>> version 10.1
>>>  if replay() {
>>>  syntax [anything] [, Level(real 95) ]
>>>  eret di, level(`level')
>>>  }
>>>  else {
>>>  qui {
>>> syntax [varlist] [, Q1(real .25) Q2(real 0.75) Level(real 95) *]
>>> gettoken depv vl:varlist
>>> tempname b1 b2 beta
>>> qreg `depv' `vl', quantile(`q1') `options'
>>> matrix `b1'=e(b)
>>> tempvar e1 e2
>>> g `e1'=e(sample)
>>> qreg `depv' `vl', quantile(`q2') `options'
>>> g `e2'=e(sample)&`e1'
>>> count if `e2'
>>> loc N=r(N)
>>> matrix `b2'=e(b)
>>> matrix `beta'=`b2'-`b1'
>>> ereturn post `beta', dep(`depv') e(`e2') obs(`N')
>>> eret local cmd="iqr"
>>>  }
>>> eret di, level(`level')
>>>  }
>>> end
>>> sysuse nlsw88
>>> egen c=group(ind occ), label
>>> drop if mi(c)
>>> bs, rep(20) cl(c) seed(4):iqr wage age grade
>>>
>>>
>>> 2009/11/16 Jorge Eduardo Pérez Pérez <perez.jorge@ur.edu.co>:
>>>> Hi everybody
>>>>
>>>> I'm trying to estimate an interquantile range regression with
>>>> block-bootstrapped standard errors. I've a dataset of cities and
>>>> months and i'm trying to estimate a differences in differences model,
>>>> so i need the bootstraped s.e's to take into account the
>>>> within-cluster correlation. Computation of the regressions using iqreg
>>>> is VERY slow, so I tried something like
>>>>
>>>> program myqreg, eclass
>>>> version 11
>>>> syntax varlist  [, `options' Q1(real .25) Q2(real 0.75)]
>>>> gettoken depv vl : varlist
>>>> tempname b1 b2 beta
>>>> qui qreg `depv' `vl', quantile(`q1')
>>>> matrix `b1'=e(b)
>>>> qui qreg `depv' `vl', quantile(`q2')
>>>> matrix `b2'=e(b)
>>>> matrix `beta'=`b2'-`b1'
>>>> ereturn post `beta'
>>>> end
>>>>
>>>> glo seed=4020
>>>> bootstrap _b , rep(200) cluster(city) idcluster(icity) seed($seed):
>>>> myqreg lwage _Iciudad* _Iañomes* treatment
>>>>
>>>> However, it doesn't seem to work. Every bootstrap replication fails.
>>>>
>>>> Any suggestions? Thank you !
>
> *
> *   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index