You have -robust- option in one of your runs, but not in the other.
This affects how the estimate of the VCE is computed, and I believe
this would also affect the running estimate of the Hessian. The
initial steps before iteration 0 are performed without regard to the
curvature of the likelihood being maximized, so the two procedures
start from the same point. But then they are using somewhat different
estimates of the second derivative matrix, I guess, and that's where
the discrepancies are coming out of. By the way, neither of the
estimation procedures converged. With 40 observations, there is no way
you get reasonable results for 28 parameters. No way. And you do see
that in some parameters shooting off to infinity.
On 2/9/06, Qiuqiong Huang <[email protected]> wrote:
Hi
I found that when I run mlogit on different machine, I got different
results. The data and variables used in the regression are exactly the
same. But one machine ran 13 iterations. another machine only ran 11
iterations and did not converge. And the results (magnitudes and standard
errors of coefficients are totally different too). On both machine, I
used Stata 8.2.
is there any way I can find out why i got different results? Below are
the two sets of results.
Thanks a lot.
Qiuqiong
***********************************************************************
***********************************************************************
***********************************************************************
Machine 1:
. xi: mlogit incentive2 DClined seri_mud avgDCnumgate_2 avghhplot wstress
age_vill edu_vill pthsch ptnonag
ptmig i.county, b(1) robust tr
i.county _Icounty_31-35 (naturally coded; _Icounty_31
omitted)
Iteration 0: log pseudolikelihood = -41.286362
0 0 0 0 0
0 0 0 0 0
0 0 0 0 -.9444616
0 0 0 0 0
0 0 0 0 0
0 0 0 0 -.1823216
Iteration 1: log pseudolikelihood = -23.942025
-.1535105 -1.862865 -.0126249 -.1058805 -.0903134
-.0083592 .0718334 .0156798 -.0209434 -.0644558
3.273693 .7684405 1.147216 -.7303947 1.996602
.7346775 .4799768 -.0084356 .3432856 .8750169
.0167751 -.1539001 .0164632 -.0378387 .0005579
-2.041441 -.3045206 -.4144813 -1.28777 -4.461722
Iteration 2: log pseudolikelihood = -20.07641
.......Omitted here
Iteration 10: log pseudolikelihood = -4.2511694
216.9474 -83.40591 -.0279035 6.761672 64.32783
22.57524 27.4512 8.661309 -10.73628 -3.635502
222.3915 41.19815 30.15323 -13.08641 -1635.229
200.3627 -3.597955 .0447571 41.53482 87.61121
15.90677 27.2541 8.738907 -8.765639 -.4245701
-83.01943 -71.07202 -171.0882 -218.1327 -1636.855
Iteration 11: log pseudolikelihood = -3.7138854
264.8277 -99.28156 -.0282477 7.7863 81.29004
27.57892 32.82099 10.73473 -13.09572 -4.473515
267.4728 44.75714 35.35282 -20.16417 -2002.28
246.2304 -4.792918 .0514819 51.2159 110.3305
19.22811 32.69608 10.69847 -10.85353 -.3073647
-110.3294 -89.29089 -215.4751 -269.2497 -1999.169
Iteration 12: log pseudolikelihood = -3.5259999
290.1185 -106.7525 -.0283225 9.239672 89.29666
30.11146 35.97972 11.79514 -14.29023 -4.862053
286.5799 45.57917 34.45153 -27.19174 -2195.181
270.1947 -5.378952 .0567416 56.18665 121.0073
21.10159 35.89417 11.73846 -11.91466 -.3277298
-120.984 -97.89167 -236.4222 -295.3328 -2193.674
Iteration 13: log pseudolikelihood = -3.510283
296.3671 -108.5992 -.0283358 9.596246 91.27622
30.73784 36.75933 12.05715 -14.58576 -4.958122
291.3211 45.79315 34.24537 -28.90659 -2242.855
276.1129 -5.525269 .0580142 57.41468 123.6471
21.56427 36.68407 11.99532 -12.17688 -.3327286
-123.6217 -100.0199 -241.603 -301.7788 -2241.715
Multinomial logistic regression Number of obs =
40
Wald chi2(24) =
.
Prob > chi2 =
.
Log pseudolikelihood = -3.5107012 Pseudo R2 =
0.9150
***********************************************************************
***********************************************************************
***********************************************************************
Machine 2:
. xi: mlogit incentive2 DClined seri_mud avgDCnumgate_2 avghhplot wstress
age_vill edu_vill pthsch ptnonag
ptmig i.county, b(1) tr
i.county _Icounty_31-35 (naturally coded; _Icounty_31
omitted)
Iteration 0: log likelihood = -41.286362
0 0 0 0 0
0 0 0 0 0
0 0 0 0 -.9444616
0 0 0 0 0
0 0 0 0 0
0 0 0 0 -.1823216
Iteration 1: log likelihood = -23.942025
-.1535105 -1.862865 -.0126249 -.1058805 -.0903134
-.0083592 .0718334 .0156798 -.0209434 -.0644558
3.273693 .7684405 1.147216 -.7303947 1.996602
.7346775 .4799768 -.0084356 .3432856 .8750169
.0167751 -.1539001 .0164632 -.0378387 .0005579
-2.041441 -.3045206 -.4144813 -1.28777 -4.461722
Iteration 2: log likelihood = -20.07641
.7064876 -2.962282 -.0142221 -.1936033 .08562
........Omitted.......
Iteration 10: log likelihood = -4.2511694
216.9474 -83.40591 -.0279035 6.761672 64.32783
22.57524 27.4512 8.661309 -10.73628 -3.635502
222.3915 41.19815 30.15323 -13.08641 -1635.229
200.3627 -3.597955 .0447571 41.53482 87.61121
15.90677 27.2541 8.738907 -8.765639 -.4245701
-83.01943 -71.07202 -171.0882 -218.1327 -1636.855
Iteration 11: log likelihood = .
312.7081 -115.1572 -.0285919 8.810928 98.25225
32.5826 38.19077 12.80816 -15.45516 -5.311528
312.554 48.31613 40.5524 -27.24193 -2369.331
292.098 -5.987881 .0582066 60.89697 133.0497
22.54945 38.13805 12.65803 -12.94142 -.1901592
-137.6394 -107.5098 -259.862 -320.3667 -2361.483
Multinomial logistic regression Number of obs =
40
LR chi2(28) =
.
Log likelihood = . Prob > chi2 =
.
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
--
Stas Kolenikov
http://stas.kolenikov.name
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
