Help using bootstrap command

[Alfonso Sánchez-Peñalver]

Hi everyone,

I am starting to try and figure out how to bootstrap my ml estimations, and I am having trouble so I was hoping you could give some pointers. Basically I am estimating a conditional logit model to try and make it work. Here is the code to setup the data, and the program I have created to do the estimation.

version 15
clear all
set more off

// Path to testing directory
local tdir "/Users/alfonso/Dropbox/My Documents/Programming/Stata/Testing/nelogit"
// Load the data for estimation
use "`tdir'/estimationdata"

// Have to check why income is missing and if order makes a difference. For now
// we drop missing observations of both.
drop if missing(Income, Order)
global rpv "Cal Sug Sod Fat Price Serv Brand"
global fpv "c1 c2 bmi1 bmi2 inc1 inc2 new1 new2 labin1 labin2"

// Common mata external value
mata nelogit_J = 3                                                                // Number of alternatives

capture porgram drop bsnelogit_fp
program def bsnelogit_fp, eclass
    version 15
    
    // Setup to use our estimator.
    global nelogit_method = 1                                                    // ml evaluator to be conditional logit
    quiet regress choice $fpv $rpv, noconst                                        // Initial values from regression
    tempname b0
    mat `b0' = e(b)
    // Mata external values
    mata: nelogit_X = st_data(.,"$fpv $rpv")                                    // independent variables
    mata: nelogit_Y = st_data(.,"choice")                                        // dependent variables
    
    // Estimation of fixed parameters
    ml model d2 nelogit_d2 (mean:choice = $fpv $rpv, noconst), maximize            ///
        missing init(`b0', copy) search(off) nolog
end

When I test the program I get the right estimation

. // Test the program
. bsnelogit_fp

. ml display

                                                Number of obs     =      9,282
                                                Wald chi2(17)     =     454.02
Log likelihood = -3119.7094                     Prob > chi2       =     0.0000

------------------------------------------------------------------------------
      choice |      Coef.   Std. Err.      z    P>|z|     [95% Conf. Interval]
-------------+----------------------------------------------------------------
          c1 |   1.555249   .2332806     6.67   0.000     1.098028    2.012471
          c2 |   1.384334   .2361873     5.86   0.000     .9214152    1.847252
        bmi1 |  -.0081936   .0049027    -1.67   0.095    -.0178028    .0014156
        bmi2 |  -.0059896   .0048895    -1.22   0.221    -.0155729    .0035936
        inc1 |    .000045    .000092     0.49   0.624    -.0001352    .0002252
        inc2 |   .0000352   .0000921     0.38   0.702    -.0001453    .0002157
        new1 |  -.1135318   .1045485    -1.09   0.278    -.3184432    .0913795
        new2 |  -.0033199   .1046828    -0.03   0.975    -.2084944    .2018545
      labin1 |   .2851946   .1047451     2.72   0.006     .0798979    .4904913
      labin2 |   .3654949   .1048108     3.49   0.000     .1600696    .5709202
         Cal |  -.0005064   .0008785    -0.58   0.564    -.0022283    .0012155
         Sug |   -.034054   .0072128    -4.72   0.000    -.0481909   -.0199171
         Sod |   .0106329    .001784     5.96   0.000     .0071364    .0141295
         Fat |   .0330286   .0131235     2.52   0.012     .0073071    .0587501
       Price |  -.2055178     .03213    -6.40   0.000    -.2684916   -.1425441
        Serv |   .0040012    .000982     4.07   0.000     .0020765    .0059259
       Brand |   -.228362     .06646    -3.44   0.001    -.3586212   -.0981027
------------------------------------------------------------------------------

However, when I try to bootstrap my estimation, it gets stuck and doesn't produce a single dot, and I have to hit the break button to stop it.

. bootstrap _b, reps(50) seed(1234): bsnelogit_fp
(running bsnelogit_fp on estimation sample)

Bootstrap replications (50)
----+--- 1 ---+--- 2 ---+--- 3 ---+--- 4 ---+--- 5

Any suggestions as to what I may be doing wrong? Thanks!!!!

[Donghui Wang]

Hi Alfonso,
I came across this post you wrote last year. Were you able to solve this problem? I got the same problem when try to bootstrap after the ml command.

[FernandoRios]

Hi Donghui
The most important question. What type of ML estimation are you trying to model?
can you provide the code?
I suspect that the problem has little to do with bootstrap, and more with what is fed through ML.
Fernando

[Alfonso Sánchez-Peñalver]

I actually found the problem. The program I was developing is for a multinomial logit. Since the data is in long form, I had to use the cluster() and idcluster() options appropriately.

[Donghui Wang]

Hi Fernando,
You are absolutely right. I was able to bootstrap the CI after correcting my ml program. Thanks!