Bookmark and Share

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.


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

st: failure estimating bootstrap on logistic model


From   Andrew Waxman <andwax@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   st: failure estimating bootstrap on logistic model
Date   Wed, 3 Mar 2010 20:07:24 -0500

Hi,
I am trying to estimate a logistic model to predict firm exit and want to
bootstrap my standard errors.  My dataset is large (225,000 obs) and
although there are some missing values, this shouldn't significantly
restrict the number of observations when bootstrap draws a sample.

However, the majority of reps seem to fail (note the "x"'es below) when I
estimate using a logistic model:

*. bootstrap, r(50): logistic exit lnfirmage lnL lnL2 skratio fempsh fown
 gown  exporter lnrKreppw lnTFPCDols yd2*
*> -yd11 provd2-provd26  if year!=1990  & tc==2 & outlier!=1*
*(running logistic on estimation sample)*
*
*
*Bootstrap replications (50)*
*----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5 *
*xxxxxx.xxxx.xxxxxxxx.xxxxxxxxxx.x.xxxxxxxxx.xx.xxx    50*
*
*
*note: yd2 dropped because of collinearity*
*note: yd3 dropped because of collinearity*
*note: yd4 dropped because of collinearity*
*note: yd5 dropped because of collinearity*
*note: yd6 dropped because of collinearity*
*note: yd7 dropped because of collinearity*
*note: yd8 dropped because of collinearity*
*note: yd11 dropped because of collinearity*
*
*
*Logistic regression                             Number of obs      =
25380*
*                                                Replications       =
  7*
*                                                Wald chi2(6)       =
  .*
*                                                Prob > chi2        =
  .*
*Log likelihood = -5468.5047                     Pseudo R2          =
 0.1066*
*
*
*
------------------------------------------------------------------------------
*
*             |   Observed   Bootstrap                         Normal-based*
*        exit | Odds Ratio   Std. Err.      z    P>|z|     [95% Conf.
Interval]*
*
-------------+----------------------------------------------------------------
*
*   lnfirmage |   .8826269   .0127857    -8.62   0.000     .8579197
 .9080457*
*         lnL |   .0688836   .0104223   -17.68   0.000     .0512068
 .0926627*
*        lnL2 |    1.23799   .0172181    15.35   0.000     1.204699
 1.272201*
*     skratio |   1.914752   .3221713     3.86   0.000     1.376865
 2.662769*
*      fempsh |   1.465115   .2135063     2.62   0.009     1.101105
1.94946*
*        fown |   1.246216   .1670346     1.64   0.101     .9583057
 1.620626*
*        gown |   .7605959   .0838717    -2.48   0.013     .6127609
 .9440975*
*    exporter |   1.568062   .1541614     4.58   0.000     1.293239
 1.901288*
*   lnrKreppw |   .9256719   .0146664    -4.87   0.000     .8973681
 .9548684*
*lnTFPCDolsss |   .9377086    .038908    -1.55   0.121     .8644688
 1.017153*
*         yd9 |   .6756658   .0590381    -4.49   0.000     .5693193
 .8018773*
*        yd10 |   2.429092   .1524321    14.14   0.000     2.147973
 2.747003*
*      provd2 |   1.357248   11.32525     0.04   0.971     1.07e-07
 1.72e+07*
*      provd3 |   1.018087    8.62425     0.00   0.998     6.27e-08
 1.65e+07*
*      provd4 |   .4307106   3.530132    -0.10   0.918     4.55e-08
4080270*
*      provd5 |   3.372531   26.91701     0.15   0.879     5.42e-07
 2.10e+07*
*      provd6 |   1.235142   10.24105     0.03   0.980     1.08e-07
 1.41e+07*
*      provd7 |   5.591253   48.60804     0.20   0.843     2.23e-07
 1.40e+08*
*      provd8 |   4.408462   37.58628     0.17   0.862     2.44e-07
 7.97e+07*
*      provd9 |   1.281082   10.76104     0.03   0.976     9.07e-08
 1.81e+07*
*     provd10 |      1.255   10.46518     0.03   0.978     1.00e-07
 1.57e+07*
*     provd11 |    1.54895   12.95652     0.05   0.958     1.17e-07
 2.04e+07*
*     provd12 |   1.453425   12.24842     0.04   0.965     9.75e-08
 2.17e+07*
*     provd13 |   1.401437   11.68034     0.04   0.968     1.13e-07
 1.74e+07*
*     provd14 |   2.839406   23.93431     0.12   0.901     1.90e-07
 4.25e+07*
*     provd15 |   1.465131   12.07134     0.05   0.963     1.42e-07
 1.51e+07*
*     provd16 |   15.04903   128.1326     0.32   0.750     8.51e-07
 2.66e+08*
*     provd17 |   1.609744   13.47251     0.06   0.955     1.21e-07
 2.14e+07*
*     provd18 |   1.460923   12.50193     0.04   0.965     7.59e-08
 2.81e+07*
*     provd19 |   1.959788   16.44873     0.08   0.936     1.41e-07
 2.73e+07*
*     provd20 |   1.334879   11.01907     0.03   0.972     1.26e-07
 1.42e+07*
*     provd21 |   2.486556   20.49772     0.11   0.912     2.39e-07
 2.58e+07*
*     provd22 |   5.840088   48.96986     0.21   0.833     4.26e-07
 8.01e+07*
*     provd23 |   1.392887   11.71483     0.04   0.969     9.66e-08
 2.01e+07*
*     provd24 |    6.23053   51.59921     0.22   0.825     5.56e-07
 6.98e+07*
*     provd25 |   1.453785   11.93851     0.05   0.964     1.49e-07
 1.42e+07*
*     provd26 |   2.422081   21.33134     0.10   0.920     7.72e-08
 7.60e+07*
*
------------------------------------------------------------------------------
*
*Note: one or more parameters could not be estimated in 43 bootstrap
replicates;*
*      standard error estimates include only complete replications.*



Yet when I perform the same procedure using OLS, none of the reps fail:
*
*
*. bootstrap, r(50): reg exit lnfirmage lnL lnL2 skratio fempsh fown  gown
 exporter lnrKreppw lnTFPCDols yd2-yd11*
*>  provd2-provd26  if year!=1990  & tc==2 & outlier!=1*
*(running regress on estimation sample)*
*
*
*Bootstrap replications (50)*
*----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5 *
*..................................................    50*
*
*
*Linear regression                               Number of obs      =
25380*
*                                                Replications       =
 50*
*                                                Wald chi2(38)      =
  .*
*                                                Prob > chi2        =
  .*
*                                                R-squared          =
 0.0503*
*                                                Adj R-squared      =
 0.0489*
*                                                Root MSE           =
 0.2409*
*
*
*
------------------------------------------------------------------------------
*
*             |   Observed   Bootstrap                         Normal-based*
*        exit |      Coef.   Std. Err.      z    P>|z|     [95% Conf.
Interval]*
*
-------------+----------------------------------------------------------------
*
*   lnfirmage |  -.0064203   .0019687    -3.26   0.001    -.0102788
-.0025617*
*         lnL |  -.1464816    .008121   -18.04   0.000    -.1623984
-.1305648*
*        lnL2 |   .0122006   .0008017    15.22   0.000     .0106292
.013772*
*     skratio |   .0374925   .0098978     3.79   0.000     .0180931
 .0568919*
*      fempsh |   .0197131   .0068764     2.87   0.004     .0062356
 .0331907*
*        fown |   .0089389   .0055844     1.60   0.109    -.0020063
 .0198841*
*        gown |  -.0090385   .0070631    -1.28   0.201    -.0228819
 .0048049*
*    exporter |   .0202212   .0038746     5.22   0.000     .0126271
 .0278153*
*   lnrKreppw |  -.0036477   .0007043    -5.18   0.000    -.0050281
-.0022673*
*lnTFPCDolsss |  -.0040976   .0022188    -1.85   0.065    -.0084463
 .0002511*
*         yd2 |  (dropped)*
*         yd3 |  (dropped)*
*         yd4 |  (dropped)*
*         yd5 |  (dropped)*
*         yd6 |  (dropped)*
*         yd7 |  (dropped)*
*         yd8 |          0   .0284675     0.00   1.000    -.0557954
 .0557954*
*         yd9 |  -.0170108   .0285136    -0.60   0.551    -.0728965
 .0388749*
*        yd10 |   .0576383   .0282818     2.04   0.042      .002207
 .1130696*
*        yd11 |  (dropped)*
*      provd2 |   .0118879   .0283366     0.42   0.675    -.0436508
 .0674265*
*      provd3 |  -.0022906    .032666    -0.07   0.944    -.0663148
 .0617335*
*      provd4 |  -.0153793   .0308718    -0.50   0.618     -.075887
 .0451285*
*      provd5 |    .077172   .0380442     2.03   0.043     .0026067
 .1517373*
*      provd6 |   .0077553     .03216     0.24   0.809    -.0552772
 .0707877*
*      provd7 |   .1416116   .0861636     1.64   0.100     -.027266
 .3104892*
*      provd8 |   .1190133   .0383966     3.10   0.002     .0437574
 .1942692*
*      provd9 |   .0116131   .0288091     0.40   0.687    -.0448516
 .0680779*
*     provd10 |   .0104755   .0273441     0.38   0.702     -.043118
 .0640689*
*     provd11 |   .0246304   .0285513     0.86   0.388    -.0313292
 .08059*
*     provd12 |   .0191271   .0283466     0.67   0.500    -.0364312
 .0746854*
*     provd13 |   .0158017   .0274104     0.58   0.564    -.0379216
.069525*
*     provd14 |   .0780028   .0331598     2.35   0.019     .0130108
 .1429948*
*     provd15 |   .0197735   .0307751     0.64   0.521    -.0405445
 .0800916*
*     provd16 |   .3027415   .0882585     3.43   0.001     .1297581
 .4757249*
*     provd17 |    .020612   .0284689     0.72   0.469     -.035186
 .0764099*
*     provd18 |   .0163567   .0359175     0.46   0.649    -.0540403
 .0867537*
*     provd19 |   .0308439   .0398664     0.77   0.439    -.0472929
 .1089807*
*     provd20 |   .0144719   .0313564     0.46   0.644    -.0469856
 .0759294*
*     provd21 |   .0602479   .0296055     2.04   0.042     .0022222
 .1182736*
*     provd22 |   .1165868   .0584628     1.99   0.046     .0020018
 .2311717*
*     provd23 |   .0159528   .0275981     0.58   0.563    -.0381384
.070044*
*     provd24 |   .1591771   .0483605     3.29   0.001     .0643922
 .2539621*
*     provd25 |   .0154854   .0416812     0.37   0.710    -.0662083
 .0971791*
*     provd26 |   .0420502   .0425296     0.99   0.323    -.0413064
 .1254067*
*       _cons |   .4176809   .0461915     9.04   0.000     .3271473
 .5082146*
*
------------------------------------------------------------------------------
*

Is this problem related to the non-linearity of the estimation method?  Is
there an obvious fix?


Very grateful for your help.

Andrew Waxman
World Bank, Research Dept.

*
*   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   |   Site index