st: Bootstrap: Which standard errors to use?

classic Classic list List threaded Threaded
13 messages Options
Reply | Threaded
Open this post in threaded view
|

st: Bootstrap: Which standard errors to use?

Supnithadnaporn, Anupit
Dear all,

I am using bootstrap in my study and Stata reports 2 types of standard errors
of beta: (1) bootstrap std. err. right to the observed coef. and (2) se shown in
the second part of the table. They are quite different. How does Stata calculate
both of these SEs? Which one would be better to use?

Would anybody please explain or suggest?

Thank you
Anupit


. local vehicle age lnodo peuro pasia pother  
. local owner black other
. set seed 9999    
. bootstrap _b _se, reps(10000) saving("C:\data\logitBOOT2Abse.dta", replace): ///
  logit tresimp indlninc `owner' `vehicle', robust


Logistic regression                             Number of obs      =       465
                                                Replications       =      4899

------------------------------------------------------------------------------
             |   Observed   Bootstrap                         Normal-based
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
b            |
    indlninc |  -.2818828   .4831361    -0.58   0.560    -1.228812    .6650466
       black |   1.184949   .6092399     1.94   0.052    -.0091396    2.379037
       other |   .5112052     .73946     0.69   0.489    -.9381097     1.96052
         age |   .2291411    .091903     2.49   0.013     .0490146    .4092677
       lnodo |   .2668319   .5548409     0.48   0.631    -.8206363      1.3543
       peuro |   .3441804   .9544531     0.36   0.718    -1.526513    2.214874
       pasia |   .1141557   .7772255     0.15   0.883    -1.409178     1.63749
      pother |   .1774283   .7139723     0.25   0.804    -1.221932    1.576788
-------------+----------------------------------------------------------------
se           |
    indlninc |   .3751495   .1164555     3.22   0.001     .1469009    .6033982
       black |   .5377806   .1174539     4.58   0.000     .3075752    .7679861
       other |   .6130217   .1737154     3.53   0.000     .2725458    .9534977
         age |   .0683818   .0179438     3.81   0.000     .0332126    .1035509
       lnodo |   .3736896   .1584534     2.36   0.018     .0631266    .6842526
       peuro |   .8376832    .242609     3.45   0.001     .3621782    1.313188
       pasia |   .6548303   .1592494     4.11   0.000     .3427073    .9669534
      pother |   .7830163   .2246008     3.49   0.000     .3428068    1.223226



*
*   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/
Reply | Threaded
Open this post in threaded view
|

Re: st: Bootstrap: Which standard errors to use?

Maarten buis
--- "Supnithadnaporn, Anupit" <[hidden email]> wrote:
> I am using bootstrap in my study and Stata reports 2 types of
> standard errors of beta: (1) bootstrap std. err. right to the
> observed coef. and (2) se shown in the second part of the table.
> They are quite different.
<snip>
> . local vehicle age lnodo peuro pasia pother  
> . local owner black other
> . set seed 9999    
> . bootstrap _b _se, reps(10000) saving("C:\data\logitBOOT2Abse.dta",
> replace): ///
>   logit tresimp indlninc `owner' `vehicle', robust

The standard error in the second part is the standard error of the
standard error. Remember that the standard error is also an estimate,
so you can have sampling uncertainty around that too. The standard
error in the second part of your table tells you about the uncertainty
about the estimate of the standard error. So, it can be interesting too
look at it, but the standard error you are after is in the first part
of the table. Normally you would not ask for the second part of your
table (and exclude the -_se- from the -bootstrap- statement).

Hope this helps,
Maarten

-----------------------------------------
Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands

visiting address:
Buitenveldertselaan 3 (Metropolitan), room N515

+31 20 5986715

http://home.fsw.vu.nl/m.buis/
-----------------------------------------


     
*
*   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/
Reply | Threaded
Open this post in threaded view
|

Re: st: Bootstrap: Which standard errors to use?

Supnithadnaporn, Anupit
Thank you Marrten for your explanation. This helps a lot.

Anupit

----- "Maarten buis" <[hidden email]> wrote:

> --- "Supnithadnaporn, Anupit" <[hidden email]> wrote:
> > I am using bootstrap in my study and Stata reports 2 types of
> > standard errors of beta: (1) bootstrap std. err. right to the
> > observed coef. and (2) se shown in the second part of the table.
> > They are quite different.
> <snip>
> > . local vehicle age lnodo peuro pasia pother  
> > . local owner black other
> > . set seed 9999    
> > . bootstrap _b _se, reps(10000)
> saving("C:\data\logitBOOT2Abse.dta",
> > replace): ///
> >   logit tresimp indlninc `owner' `vehicle', robust
>
> The standard error in the second part is the standard error of the
> standard error. Remember that the standard error is also an estimate,
> so you can have sampling uncertainty around that too. The standard
> error in the second part of your table tells you about the
> uncertainty
> about the estimate of the standard error. So, it can be interesting
> too
> look at it, but the standard error you are after is in the first part
> of the table. Normally you would not ask for the second part of your
> table (and exclude the -_se- from the -bootstrap- statement).
>
> Hope this helps,
> Maarten
>
> -----------------------------------------
> Maarten L. Buis
> Department of Social Research Methodology
> Vrije Universiteit Amsterdam
> Boelelaan 1081
> 1081 HV Amsterdam
> The Netherlands
>
> visiting address:
> Buitenveldertselaan 3 (Metropolitan), room N515
>
> +31 20 5986715
>
> http://home.fsw.vu.nl/m.buis/
> -----------------------------------------
>
>
>      
> *
> *   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/

--
Anupit Supnithadnaporn
PhD Candidate
School of Public Policy
Georgia Institute of Technology
D.M.Smith Building TPAC Room 018
685 Cherry Street
Atlanta GA 30332

*
*   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/
Reply | Threaded
Open this post in threaded view
|

Re: st: Bootstrap: Which standard errors to use?

Stas Kolenikov
In reply to this post by Maarten buis
The second part of the table actually shows that the Hessian-based
standard errors of logit are biased down by some 10-20%, for whatever
reason. The average standard error for indlninc variable is 0.375,
while the bootstrap measure of variability is 0.483. The first one is
purely model-based, while the second one is more robust to various
model misspecifications. I'd be curious as to whether -logit ...,
robust- standard errors would be close to the bootstrap ones, in your
case.

On 12/8/08, Maarten buis <[hidden email]> wrote:

> --- "Supnithadnaporn, Anupit" <[hidden email]> wrote:
>  > I am using bootstrap in my study and Stata reports 2 types of
>  > standard errors of beta: (1) bootstrap std. err. right to the
>  > observed coef. and (2) se shown in the second part of the table.
>  > They are quite different.
>
> <snip>
>
> > . local vehicle age lnodo peuro pasia pother
>  > . local owner black other
>  > . set seed 9999
>  > . bootstrap _b _se, reps(10000) saving("C:\data\logitBOOT2Abse.dta",
>  > replace): ///
>  >   logit tresimp indlninc `owner' `vehicle', robust
>
>
> The standard error in the second part is the standard error of the
>  standard error. Remember that the standard error is also an estimate,
>  so you can have sampling uncertainty around that too. The standard
>  error in the second part of your table tells you about the uncertainty
>  about the estimate of the standard error. So, it can be interesting too
>  look at it, but the standard error you are after is in the first part
>  of the table. Normally you would not ask for the second part of your
>  table (and exclude the -_se- from the -bootstrap- statement).
>
>  Hope this helps,
>  Maarten
>
>  -----------------------------------------
>  Maarten L. Buis
>  Department of Social Research Methodology
>  Vrije Universiteit Amsterdam
>  Boelelaan 1081
>  1081 HV Amsterdam
>  The Netherlands
>
>  visiting address:
>  Buitenveldertselaan 3 (Metropolitan), room N515
>
>  +31 20 5986715
>
>  http://home.fsw.vu.nl/m.buis/
>  -----------------------------------------
>
>
>
>
>  *
>  *   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/
Reply | Threaded
Open this post in threaded view
|

Re: st: Bootstrap: Which standard errors to use?

Supnithadnaporn, Anupit
In reply to this post by Supnithadnaporn, Anupit
----- "Stas Kolenikov" <[hidden email]> wrote:

> I'd be curious as to whether -logit ...,
> robust- standard errors would be close to the bootstrap ones, in your
> case.

I guess you mean the result from running -logit, robust- one time, right?
Would these SE seem to make sense to you at all?

Thank you,
Anupit

Logistic regression                               Number of obs   =        465
                                                  Wald chi2(16)   =      44.37
                                                  Prob > chi2     =     0.0002
Log pseudolikelihood = -73.949457                 Pseudo R2       =     0.2407

------------------------------------------------------------------------------
             |               Robust
     tresimp |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
    indlninc |  -.2818828   .3751495    -0.75   0.452    -1.017162    .4533968
       black |   1.184949   .5377806     2.20   0.028      .130918    2.238979
       other |   .5112052   .6130217     0.83   0.404    -.6902953    1.712706
         age |   .2291411   .0683818     3.35   0.001     .0951153    .3631669
       lnodo |   .2668319   .3736896     0.71   0.475    -.4655863    .9992501
       peuro |   .3441804   .8376832     0.41   0.681    -1.297648    1.986009
       pasia |   .1141557   .6548303     0.17   0.862    -1.169288      1.3976
      pother |   .1774283   .7830163     0.23   0.821    -1.357255    1.712112
       displ |  -.0444871   .3161949    -0.14   0.888    -.6642178    .5752435
      indefi |   .3926529   .6165853     0.64   0.524     -.815832    1.601138
       indfi |  -.5801308   .6734173    -0.86   0.389    -1.900004    .7397428
      indmfi |  -1.667607   1.093033    -1.53   0.127    -3.809912    .4746974
        egr1 |  -1.307301   .5173223    -2.53   0.012    -2.321235   -.2933684
        tac1 |  -.1766159   .7599367    -0.23   0.816    -1.666064    1.312833
         car |  -1.189119   .8430763    -1.41   0.158    -2.841518    .4632803
         van |  -1.120382   .9862894    -1.14   0.256    -3.053474    .8127092
       _cons |  -3.180481   6.637172    -0.48   0.632     -16.1891    9.828137
------------------------------------------------------------------------------


*
*   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/
Reply | Threaded
Open this post in threaded view
|

Re: st: Bootstrap: Which standard errors to use?

Stas Kolenikov
Those are exactly the reported standard errors in your second panel.
That's interesting; I am used to thinking that they are close to the
bootstrap standard errors. How about the -oim- standard errors without
the -robust- option?

On 12/8/08, Supnithadnaporn, Anupit <[hidden email]> wrote:

> ----- "Stas Kolenikov" <[hidden email]> wrote:
>
>  > I'd be curious as to whether -logit ...,
>  > robust- standard errors would be close to the bootstrap ones, in your
>  > case.
>
>
> I guess you mean the result from running -logit, robust- one time, right?
>  Would these SE seem to make sense to you at all?
>
>  Thank you,
>  Anupit
>
>
>  Logistic regression                               Number of obs   =        465
>
>                                                   Wald chi2(16)   =      44.37
>                                                   Prob > chi2     =     0.0002
>  Log pseudolikelihood = -73.949457                 Pseudo R2       =     0.2407
>
>  ------------------------------------------------------------------------------
>              |               Robust
>      tresimp |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
>  -------------+----------------------------------------------------------------
>     indlninc |  -.2818828   .3751495    -0.75   0.452    -1.017162    .4533968
>        black |   1.184949   .5377806     2.20   0.028      .130918    2.238979
>        other |   .5112052   .6130217     0.83   0.404    -.6902953    1.712706
>          age |   .2291411   .0683818     3.35   0.001     .0951153    .3631669
>        lnodo |   .2668319   .3736896     0.71   0.475    -.4655863    .9992501
>        peuro |   .3441804   .8376832     0.41   0.681    -1.297648    1.986009
>        pasia |   .1141557   .6548303     0.17   0.862    -1.169288      1.3976
>       pother |   .1774283   .7830163     0.23   0.821    -1.357255    1.712112
>        displ |  -.0444871   .3161949    -0.14   0.888    -.6642178    .5752435
>       indefi |   .3926529   .6165853     0.64   0.524     -.815832    1.601138
>        indfi |  -.5801308   .6734173    -0.86   0.389    -1.900004    .7397428
>       indmfi |  -1.667607   1.093033    -1.53   0.127    -3.809912    .4746974
>         egr1 |  -1.307301   .5173223    -2.53   0.012    -2.321235   -.2933684
>         tac1 |  -.1766159   .7599367    -0.23   0.816    -1.666064    1.312833
>          car |  -1.189119   .8430763    -1.41   0.158    -2.841518    .4632803
>          van |  -1.120382   .9862894    -1.14   0.256    -3.053474    .8127092
>        _cons |  -3.180481   6.637172    -0.48   0.632     -16.1891    9.828137
>  ------------------------------------------------------------------------------
>
>
>
>  *
>  *   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/
Reply | Threaded
Open this post in threaded view
|

Re: st: Bootstrap: Which standard errors to use?

Supnithadnaporn, Anupit
----- "Stas Kolenikov" <[hidden email]> wrote:
> How about the -oim- standard errors
> without
> the -robust- option?
>

As you requested. What would -oim- stand for?

Thank you,

Anupit



Logistic regression                               Number of obs   =        465
                                                  LR chi2(16)     =      46.89
                                                  Prob > chi2     =     0.0001
Log likelihood = -73.949457                       Pseudo R2       =     0.2407

------------------------------------------------------------------------------
     tresimp |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
    indlninc |  -.2818828   .3439042    -0.82   0.412    -.9559228    .3921571
       black |   1.184949   .5615598     2.11   0.035     .0843117    2.285586
       other |   .5112052   .6401804     0.80   0.425    -.7435253    1.765936
         age |   .2291411   .0710728     3.22   0.001      .089841    .3684412
       lnodo |   .2668319   .3439181     0.78   0.438    -.4072352     .940899
       peuro |   .3441804    .896089     0.38   0.701    -1.412122    2.100483
       pasia |   .1141557   .6634897     0.17   0.863     -1.18626    1.414572
      pother |   .1774283   .8487587     0.21   0.834    -1.486108    1.840965
       displ |  -.0444871   .2766297    -0.16   0.872    -.5866715    .4976972
      indefi |   .3926529   .6296414     0.62   0.533    -.8414216    1.626727
       indfi |  -.5801308   .6346339    -0.91   0.361     -1.82399    .6637289
      indmfi |  -1.667607   1.121463    -1.49   0.137    -3.865634    .5304192
        egr1 |  -1.307301   .5190509    -2.52   0.012    -2.324623   -.2899803
        tac1 |  -.1766159   .7663102    -0.23   0.818    -1.678556    1.325324
         car |  -1.189119   .8525678    -1.39   0.163    -2.860121    .4818834
         van |  -1.120382   .9176824    -1.22   0.222    -2.919007    .6782421
       _cons |  -3.180481   5.519478    -0.58   0.564    -13.99846    7.637497
------------------------------------------------------------------------------


*
*   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/
Reply | Threaded
Open this post in threaded view
|

Re: st: Bootstrap: Which standard errors to use?

Martin Weiss-5
"Observed Information Matrix"

HTH
Martin
_______________________
----- Original Message -----
From: "Supnithadnaporn, Anupit" <[hidden email]>
To: <[hidden email]>
Sent: Monday, December 08, 2008 9:01 PM
Subject: Re: st: Bootstrap: Which standard errors to use?


> ----- "Stas Kolenikov" <[hidden email]> wrote:
>> How about the -oim- standard errors
>> without
>> the -robust- option?
>>
>
> As you requested. What would -oim- stand for?
>
> Thank you,
>
> Anupit
>
>
>
> Logistic regression                               Number of obs   =
> 465
>                                                  LR chi2(16)     =
> 46.89
>                                                  Prob > chi2     =
> 0.0001
> Log likelihood = -73.949457                       Pseudo R2       =
> 0.2407
>
> ------------------------------------------------------------------------------
>     tresimp |      Coef.   Std. Err.      z    P>|z|     [95% Conf.
> Interval]
> -------------+----------------------------------------------------------------
>    indlninc |  -.2818828   .3439042    -0.82   0.412    -.9559228
> .3921571
>       black |   1.184949   .5615598     2.11   0.035     .0843117
> 2.285586
>       other |   .5112052   .6401804     0.80   0.425    -.7435253
> 1.765936
>         age |   .2291411   .0710728     3.22   0.001      .089841
> .3684412
>       lnodo |   .2668319   .3439181     0.78   0.438    -.4072352
> .940899
>       peuro |   .3441804    .896089     0.38   0.701    -1.412122
> 2.100483
>       pasia |   .1141557   .6634897     0.17   0.863     -1.18626
> 1.414572
>      pother |   .1774283   .8487587     0.21   0.834    -1.486108
> 1.840965
>       displ |  -.0444871   .2766297    -0.16   0.872    -.5866715
> .4976972
>      indefi |   .3926529   .6296414     0.62   0.533    -.8414216
> 1.626727
>       indfi |  -.5801308   .6346339    -0.91   0.361     -1.82399
> .6637289
>      indmfi |  -1.667607   1.121463    -1.49   0.137    -3.865634
> .5304192
>        egr1 |  -1.307301   .5190509    -2.52
> 0.012    -2.324623   -.2899803
>        tac1 |  -.1766159   .7663102    -0.23   0.818    -1.678556
> 1.325324
>         car |  -1.189119   .8525678    -1.39   0.163    -2.860121
> .4818834
>         van |  -1.120382   .9176824    -1.22   0.222    -2.919007
> .6782421
>       _cons |  -3.180481   5.519478    -0.58   0.564    -13.99846
> 7.637497
> ------------------------------------------------------------------------------
>
>
> *
> *   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/
Reply | Threaded
Open this post in threaded view
|

Re: st: Bootstrap: Which standard errors to use?

Antoine Terracol
In reply to this post by Stas Kolenikov
Stas Kolenikov wrote:
> Those are exactly the reported standard errors in your second panel.


Which, if I followed the thread correctly, should not come as a surprise
since Anupit's original -bootstrap- command called -logit, robust-

Antoine
*
*   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/
Reply | Threaded
Open this post in threaded view
|

Re: st: Bootstrap: Which standard errors to use?

Stas Kolenikov
On 12/8/08, Antoine Terracol <[hidden email]> wrote:
> > Those are exactly the reported standard errors in your second panel.
>  Which, if I followed the thread correctly, should not come as a surprise
> since Anupit's original -bootstrap- command called -logit, robust-

Right, I did not really pay much attention up there :)).

Well the -robust- standard errors are in fact closer to -oim- standard
errors than to the bootstrap standard errors. It is difficult to come
up with a meaningful suggestion in this situation as to which standard
errors are better. A (former) econometrician inside me would like to
remind that modeling the 0/1 decision to buy something (which this
application seem to be related to based on the variable names at
least) treated as the imperfect observation of the underlying
continuous propensity to buy is subject to the scale indeterminacy, so
that the identified combinations of parameters are "slope"/"standard
deviation of the error term" rather than "slope" as it is the case
with linear regression. Biostatisticians would rightfully raise a brow
here -- "What is he talking about? This is a GLM with a canonical
link... and the scale parameter here is 1". Well this is a matter of
interpretation! If you want an economics interpretation, then you
would need to make sure you control that sigma in the denominator to
really talk about betas being on the same scale (and only then the
bootstrap will make sense) -- which unfortunately cannot be
guaranteed.

Another aspect is the numeric stability of the logistic regression
estimates. For some bootstrap samples, the logit estimates are not
defined -- say if you sampled all zeroes, or as many ones as you have
regressors in the model so that the outcome of 1 can be perfectly
predicted with coefficient values at infinity. In some likelihood, the
samples that are "close", in some sense, to those extreme outcomes may
also produce "large" estimates of coefficients. Are those sensible
outcomes for the bootstrap? Probably not; hence the bootstrap
procedure might need to be modified to control the relative
proportions of 0s and 1s. In the simplest way, you do some sort of
stratified bootstrap: resample separately as many zero outcomes as
there were in the original sample, and as many ones as there were
originally. Is that a better bootstrap scheme? At least it takes care
of that infinite estimates issue. In Stata, you can do this by simply
adding -strata(response_variable)- to your bootstrap options.
Stratification usually brings down variances, and I would expect in
this case that the standard errors will now be much closer to the
-oim- and -robust- ones.

--
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/
Reply | Threaded
Open this post in threaded view
|

Re: st: Bootstrap: Which standard errors to use?

Supnithadnaporn, Anupit
I really appreciate Stas for your elaborative explanation.
My data and model are simpler than your speculation.

Pr(FailEmissionTest) = f(OwnerCharacteristics,VehicleCharacteristics)

Owner characteristics = {Income, Race}
Vehicle characteristics = {Age,Mileage,Type(car,van,truck),ProductionCountry,
                          OtherEmissionControlTechnologies}

tresimp = First test result (1=Fail, 0=Pass)
indlninc = Ln of Individual household income

Vehicle emission test is legally required in some places in order to control
for pollution. However, this might cause more burden to the poor because
they tend to own old vehicles (I control for vehicle age as well).

In my data, only around 5.37% fail the first emission test.


As you suggested, ...

Thank you
Anupit

. bootstrap _b _se, reps(10000) strata(tresimp) saving("C:\ANUPIT\e1outACIlogitBOOT2Abse
> S.dta", replace): ///
> logit tresimp indlninc `owner' `vehicle', robust
(running logit on estimation sample)


Logistic regression

Number of strata   =         2                  Number of obs      =       465
                                                Replications       =      5070

------------------------------------------------------------------------------
             |   Observed   Bootstrap                         Normal-based
             |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
b            |
    indlninc |  -.2818828    .491087    -0.57   0.566    -1.244396      .68063
       black |   1.184949   .6017671     1.97   0.049     .0055069     2.36439
       other |   .5112052   .7492401     0.68   0.495    -.9572784    1.979689
         age |   .2291411    .088497     2.59   0.010     .0556902    .4025921
       lnodo |   .2668319   .5443578     0.49   0.624    -.8000898    1.333754
       peuro |   .3441804   .9920844     0.35   0.729    -1.600269     2.28863
       pasia |   .1141557   .7469322     0.15   0.879    -1.349805    1.578116
      pother |   .1774283   .7320229     0.24   0.808     -1.25731    1.612167
       displ |  -.0444871   .3925379    -0.11   0.910    -.8138473     .724873
      indefi |   .3926529    .808736     0.49   0.627    -1.192441    1.977746
       indfi |  -.5801308   .8902797    -0.65   0.515    -2.325047    1.164785
      indmfi |  -1.667607   .7216526    -2.31   0.021     -3.08202   -.2531942
        egr1 |  -1.307301    .609217    -2.15   0.032    -2.501345   -.1132581
        tac1 |  -.1766159   .9885491    -0.18   0.858    -2.114137    1.760905
         car |  -1.189119   1.641974    -0.72   0.469    -4.407329    2.029092
         van |  -1.120382   1.712288    -0.65   0.513    -4.476406    2.235641
       _cons |  -3.180481   9.098116    -0.35   0.727    -21.01246     14.6515
-------------+----------------------------------------------------------------
se           |
    indlninc |   .3751495   .1107681     3.39   0.001      .158048    .5922511
       black |   .5377806   .1017104     5.29   0.000     .3384319    .7371293
       other |   .6130217   .1557758     3.94   0.000     .3077068    .9183367
         age |   .0683818   .0153066     4.47   0.000     .0383814    .0983822
       lnodo |   .3736896   .1499676     2.49   0.013     .0797586    .6676207
       peuro |   .8376832   .2292782     3.65   0.000     .3883061     1.28706
       pasia |   .6548303   .1297987     5.04   0.000     .4004296    .9092311
      pother |   .7830163   .2208942     3.54   0.000     .3500715    1.215961
       displ |   .3161949    .082542     3.83   0.000     .1544155    .4779743
      indefi |   .6165853   .1236292     4.99   0.000     .3742765     .858894
       indfi |   .6734173   .1310668     5.14   0.000     .4165312    .9303034
      indmfi |   1.093033   .2006469     5.45   0.000      .699772    1.486293
        egr1 |   .5173223   .0856415     6.04   0.000     .3494679    .6851766
        tac1 |   .7599367   .1663593     4.57   0.000     .4338784    1.085995
         car |   .8430763   .3607768     2.34   0.019     .1359668    1.550186
         van |   .9862894   .3964712     2.49   0.013     .2092202    1.763359
       _cons |   6.637172   1.948067     3.41   0.001     2.819031    10.45531
------------------------------------------------------------------------------
Note: One or more parameters could not be estimated in 4930 bootstrap replicates;
      standard error estimates include only complete replications.


    Variable |       Obs        Mean    Std. Dev.       Min        Max
-------------+--------------------------------------------------------
         age |       465    6.926882    3.313216          3         20
       lnodo |       465    11.41426    .5641127   8.209309    12.9082
       black |       465     .255914    .4368437          0          1
       other |       465     .144086    .3515552          0          1
      indefi |       465    .2129032    .4098007          0          1
-------------+--------------------------------------------------------
       indfi |       465    .2236559    .4171428          0          1
      indmfi |       465    .2903226    .4544001          0          1
         car |       465    .7225806    .4482074          0          1
         van |       465    .1849462    .3886721          0          1
       peuro |       465    .0666667    .2497125          0          1
-------------+--------------------------------------------------------
       pasia |       465    .2430108    .4293635          0          1
      pother |       465    .1204301    .3258143          0          1
    indlninc |       465    10.82838    .5955187   8.517183   11.51293
     tresimp |       465    .0537634    .2257932          0          1
        egr1 |       465    .7591398    .4280662          0          1
-------------+--------------------------------------------------------
        tac1 |       465    .0688172    .2534157          0          1
       displ |       465    2.936825     1.05423          1        5.9

*
*   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/
Reply | Threaded
Open this post in threaded view
|

Re: st: Bootstrap: Which standard errors to use?

Stas Kolenikov
I see. Well now I cannot think of much else to ask for :)). I thought
that the results with "stratified" bootstrap would work towards
producing smaller standard errors, but they did not seem to. They are
about the same as before.

One can argue that there is still an underlying quality of the
vehicle... and there is some room for measurement error with that
thing.

5% of 465 observations is 25... which is dangerously close to the 17
d.f.s you have. So perfect separation is a clear possibility for some
bootstrap samples.

On 12/8/08, Supnithadnaporn, Anupit <[hidden email]> wrote:

> I really appreciate Stas for your elaborative explanation.
>  My data and model are simpler than your speculation.
>
>  Pr(FailEmissionTest) = f(OwnerCharacteristics,VehicleCharacteristics)
>
>  Owner characteristics = {Income, Race}
>  Vehicle characteristics = {Age,Mileage,Type(car,van,truck),ProductionCountry,
>                           OtherEmissionControlTechnologies}
>
>  tresimp = First test result (1=Fail, 0=Pass)
>  indlninc = Ln of Individual household income
>
>  Vehicle emission test is legally required in some places in order to control
>  for pollution. However, this might cause more burden to the poor because
>  they tend to own old vehicles (I control for vehicle age as well).
>
>  In my data, only around 5.37% fail the first emission test.

--
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/
Reply | Threaded
Open this post in threaded view
|

Re: st: Bootstrap: Which standard errors to use?

Supnithadnaporn, Anupit
> One can argue that there is still an underlying quality of the
> vehicle... and there is some room for measurement error with that
> thing.
>
Absolutely agree on this point.
Thank you so much for your suggestions.

Anupit
*
*   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/