Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.

# Re: st: Bootstrapping, Robust and Weight options in regress

 From Danny Dan To statalist@hsphsun2.harvard.edu Subject Re: st: Bootstrapping, Robust and Weight options in regress Date Thu, 16 Feb 2012 23:55:04 -0600

```Dear Stas,

could guide me further in solving the issue.........

I am actually using Coarsened Exact Matching (CEM using -cem- program)
in order to match my data. In CEM weights are generated after the
matching and I am using those weights in my regression analysis, where
I want to use bootstrap for standard errors. Do you think here I can
safely use -bootstrap- in the regression equation or still there will
be problem with sample variability accountability, and therefore, I
will have to write my own program? If the later is true then would it
be possible to guide me further into writing the program?

Thank you,

Dan

On Thu, Feb 16, 2012 at 9:38 PM, Stas Kolenikov <skolenik@gmail.com> wrote:
> First, in its simplest form (as implemented in -bootstrap- command),
> the bootstrap method assumes i.i.d. data. Weights of whatever flavor
> mean that data are not i.i.d. (heteroskedastic with aweights, sampled
> with differential probabilities with pweights), and you need to modify
>
> Second, if you get your weights from a matching procedure (or any
> other input into the regression is obtained via some sort of
> estimation-prediction procedure), you have to bootstrap the whole
> process rather than its final stage, the regression. In Stata terms,
> you need to write your own little -program- that (i) accepts
> [pweights] as an input, (ii) does matching, (iii) produces weights,
> and (iv) feeds them into regression. Otherwise, your standard errors
> will be too small, and won't account for sampling variability in the
> intermediate statistics (such as, in your case, weights).
>
> Third, if things are done right, the bootstrap and the robust standard
> errors are asymptotically equivalent. Conceptually, you might be able
> to get some sort of second order improvements if you bootstrap the
> t-statistic and then refer the actual t-statistic value to your
> bootstrap distribution. But that's pretty convoluted, and it does not
> seem like you are interested in this.
>
> On Thu, Feb 16, 2012 at 10:14 PM, Danny Dan <danny2011dan@gmail.com> wrote:
>> Dear Friends,
>>
>> (1) I am trying to use both weights and vce(bootstrap) option in my
>> regression analysis as following:
>>
>> regress Y X (weight=wt), vce(bootstrap)
>>
>> The weights are generated using a Matching method, however, I cannot
>> do so as I am getting the following error:
>>
>> "Weights not allowed r(101);"
>>
>> I have tried using aweight, pweight, fweight and other weight options
>> available in STATA for regress and also sometimes getting the error
>> "may not use non-integer frequency weights r(401);".
>>
>> Therefore, nothing is working out. How can I use bootstrap option and
>> weight together in my regression?
>>
>> (2) Also is there anyway I can use both robust and bootstrap options
>> together with and without the weight option?
>
>
> --
> 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/
```