Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.

# Re: st: Bootstrap with xtregar fails

 From Nick Cox To statalist@hsphsun2.harvard.edu Subject Re: st: Bootstrap with xtregar fails Date Mon, 14 Jan 2013 18:55:54 +0000

```Bootstrapping of the standard kind pays no attention to the dependence
structure of panel data. Bootstrap plain and simple implies
independent observations. -xtregar- in contrast dos not.

Conversely, you would need to explain in a
paper|dissertation|thesis|book intended for any serious outlet (1)
what it is that you think you are doing when you combine -bootstrap-
and -xtregar- (2) why that makes sense.

I have no precise positive proposals for you, except that -xtregar-
provides a lot of output. If it's not enough, you could try plotting
data, predictions, residuals to see what extra insight they may
provide.

Nick

On Mon, Jan 14, 2013 at 6:40 PM, wasis dat <vasja.sivec@gmail.com> wrote:

> I am unable to bootstrap regression coefficients in a panel with
> autocorrelated residuals (command -xtregar). On the other hand I can
> bootstrap the coefficients when I assume that there is no
> autocorrelation in the residuals (command -xtreg):
>
>> bootstrap "xtreg y x, fe" _b[x],  reps(500)         (works)
>> bootstrap "xtregar y x, fe" _b[x],  reps(500)         (does not work...)
>
> command:      xtregar y x , fe
> statistic:    _bs_1      = _b[x]
>
> Bootstrap statistics                      Number of obs    =     27380
>                                                   Replications     =       100
>
> ------------------------------------------------------------------------------
> Variable     |  Reps  Observed      Bias  Std. Err. [95% Conf. Interval]
> -------------+----------------------------------------------------------------
>    _bs_1 |     0 -13.37994    .         .          .          .   (N)
>              |                                              .          .   (P)
>              |                                              .          .  (BC)
> ------------------------------------------------------------------------------
> Note:  N   = normal
>        P   = percentile
>        BC  = bias-corrected
>
> insufficient observations to compute bootstrap standard errors
> r(2000);
>
> I have tried different workaround methods but failed miserably. Has
> anybody encountered a similar problem? I would be eternally grateful
> for any kind suggestion.
*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/
```