Statalist


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

st: AW: Bootstrap and p-values


From   "Martin Weiss" <martin.weiss1@gmx.de>
To   <statalist@hsphsun2.harvard.edu>
Subject   st: AW: Bootstrap and p-values
Date   Sat, 19 Sep 2009 18:55:41 +0200

<> 

" Also it is not clear to me if the reported bias is a quantity that 
adds to the standard error or  must be subtracted from the standard 
error in calculating the variance."

Not sure what you mean. The bias is the difference btw the coefficient
estimate of your chosen estimator and the mean of the bootstrap estimates.
The standard error is the standard deviation of those bootstrap estimates,
as seen in this code:


*************
sysuse auto, clear

//get bs estimates
bs, reps(200) saving(myfile, /* 
*/ replace) seed(1020): /* 
*/ reg pr we le he

//store coeffs in -local-s
foreach var of varlist  weight length headroom{
loc `var'coeff=_b[`var']
}

//get saved file of bs estimates
u myfile, clear

//calculate bias and bs standard error
foreach var in weight length headroom{
qui su _b_`var'
di in red "Coeff `var', Bias (Mean of Bootstraps-Original): "  /* 
*/ r(mean)-``var'coeff' _n  /* 
*/ "Bootstrap se: " r(sd) _n
}

//compare to official results
estat boot, all
*************



HTH
Martin


-----Ursprüngliche Nachricht-----
Von: owner-statalist@hsphsun2.harvard.edu
[mailto:owner-statalist@hsphsun2.harvard.edu] Im Auftrag von Giulio Rizzoli
Gesendet: Samstag, 19. September 2009 17:17
An: statalist@hsphsun2.harvard.edu
Betreff: st: Bootstrap and p-values 

This mail didn't go through twice
I try the third time after reading the FAQ instructions.
Dear all:
Many medical variables are correlated, so I'm using the bootstrap 
program to verify the most
frequently extracted variables within a backward and forward algorithm.
In the example below Nyha4 is by far the most extracted variable and 
therefore the most significant statistically.
I usually publish tables of the selected variables with their 
confidence limits.
I imagine that the bias corrected confidence limits provided by 
bootstrap are the most correct ones.
Nonetheless until now I have always published the normal confidence 
limits, because the p-value can be easily controlled from reviewers, 
using the formula to calculate the variance under the normality assumption:
        (lnucl-lnlcl)/(2*1.96))^2 .  // lnucl=log of upper C. L. 
lnlcl=log of lower C. L.

I would like to know there is a method to calculate the p_value 
starting from the BC confidence limits.
Also it is not clear to me if the reported bias is a quantity that 
adds to the standard error or  must be subtracted from the standard 
error in calculating the variance.
The BC confidence limits are narrower. So i imagine that bias should 
be subtracted from standard error because bias relates to the part of 
measurement error that is casual.
Am I correct?
If it is so it is important to have a p-value for this estimate but 
the calculation with the usual zeta= _b/_se assumes a normality 
distribution. Is it valid in this setting ?

It woul'd be wonderful if someone has a routine to calculate the 
different p-values starting from the ereturn list of the bootstrap 
saved values or could show how to calculate BC related p-values.

. bootstrap "sw, pr(.1) pe(.05): stcox fsex AOI sonoreint age nyha4 
CPRENF CPRESPF CPIABP" _b,reps(1000) dots

command:      sw , pr(.1) pe(.05) : stcox fsex AOI sonoreint age 
nyha4 CPRENF CPRESPF CPIABP
statistics:   b_sesso    = _b[fsex]
               b_age      = _b[age]
               b_nyha4    = _b[nyha4]
............................................................................
....
 >....
Bootstrap statistics                              Number of obs    =
327
                                                   Replications     =
1000
----------------------------------------------------------------------------
--
Variable     |  Reps  Observed      Bias  Std. Err. [95% Conf. Interval]
-------------+--------------------------------------------------------------
--
      b_fsex  |   597 -.3829885 -.1357204   .138327  -.6546561  -.1113208
(N)
              |                                      -.8652664  -.3303381
(P)
              |                                      -.4775621  -.2981049
(BC)
              |                                      -1.328869   .6971875
(BC)
        b_age |   550  .0824691  .0245982  .0309384    .021697   .1432412
(N)
              |                                       .0681392   .1776492
(P)
              |                                      -.1053623   .1156664
(BC)
      b_nyha4 |   845  1.043275  .2214171  .3910648   .2757012   1.810849
(N)
              |                                        .652275   2.103901
(P)
              |                                       .5398257   1.679764
(BC)
----------------------------------------------------------------------------
--
Note:  N   = normal
        P   = percentile
        BC  = 
bias-corrected 








*
*   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