Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | Valerie Orozco <Valerie.Orozco@toulouse.inra.fr> |
To | "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu> |
Subject | st: RE : different results with mL : a seed problem? |
Date | Wed, 28 Sep 2011 18:34:33 +0000 |
Dear all, I have a question concerning my ml program. I obtain different results whether I run ml check and ml search or not or dependingon the initial seed given. In the 3 examples below, I show you that the results really differ (opposite sign for tau1 and tau2). It will be more reasonablethat Tau1 and Tau2 were positive and between 0 and 1 (but I don't want to impose it). Do you have some idea? How to justify that and what can I do ? Thank you very much for your help. Valérie CASE 1 : OK /* log likelihood -193.6561493493679*/ set seed A ml model d0 MyProg (mode=$ensx $ensz,nocons) /tau1 /tau2 ,technique(dfp) ml maximize , iterate(200) ltolerance(1e-3) ------------------------------------------------------------------------------ mode | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- eq1 | IndAir | 6.042543 1.198887 5.04 0.000 3.692767 8.392319 IndTrain | 5.06453 .6620514 7.65 0.000 3.766933 6.362127 IndBus | 4.096175 .615164 6.66 0.000 2.890476 5.301874 gc | -.0315903 .0081573 -3.87 0.000 -.0475784 -.0156023 ttme | -.1126117 .0141289 -7.97 0.000 -.1403039 -.0849196 air_inc | .0153227 .0093814 1.63 0.102 -.0030646 .03371 -------------+---------------------------------------------------------------- tau1 | _cons | .5860276 .1406654 4.17 0.000 .3103285 .8617267 -------------+---------------------------------------------------------------- tau2 | _cons | .3889992 .123669 3.15 0.002 .1466124 .631386 ------------------------------------------------------------------------------ CASE 2 : Pb with ml check and ml search when seed A /* log likelihood -208.2651378345047*/ set seed A ml model d0 MyProg (mode=$ensx $ensz,nocons) /tau1 /tau2 ,technique(dfp) ml check ml search ml maximize , iterate(200) ltolerance(1e-3) ------------------------------------------------------------------------------ 10 mode | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- eq1 | IndAir | 6.466593 2.519553 2.57 0.010 1.52836 11.40483 IndTrain | 1.720836 .7418628 2.32 0.020 .266812 3.174861 IndBus | .1836824 .7124329 0.26 0.797 -1.21266 1.580025 gc | -.069042 .0123095 -5.61 0.000 -.0931682 -.0449158 ttme | .0127155 .013757 0.92 0.355 -.0142478 .0396787 air_inc | .0366201 .0097708 3.75 0.000 .0174698 .0557705 -------------+---------------------------------------------------------------- tau1 | _cons | -.8089981 .3672655 -2.20 0.028 -1.528825 -.0891709 -------------+---------------------------------------------------------------- tau2 | _cons | -.430628 .2085702 -2.06 0.039 -.839418 -.0218379 ------------------------------------------------------------------------------ CASE 3 : OK with ml check and ml search when other seed /* log likelihood -193.6562582125274 */ set seed B ml model d0 MyProg (mode=$ensx $ensz,nocons) /tau1 /tau2 ,technique(dfp) ml check ml search ml maximize , iterate(200) ltolerance(1e-3) ------------------------------------------------------------------------------ mode | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- eq1 | IndAir | 6.040651 1.201523 5.03 0.000 3.685708 8.395594 IndTrain | 5.068941 .6620243 7.66 0.000 3.771397 6.366485 IndBus | 4.09906 .6151253 6.66 0.000 2.893437 5.304684 gc | -.0316954 .0081707 -3.88 0.000 -.0477096 -.0156812 ttme | -.1126556 .0141305 -7.97 0.000 -.1403508 -.0849604 air_inc | .0153522 .0093765 1.64 0.102 -.0030254 .0337297 -------------+---------------------------------------------------------------- tau1 | _cons | .5843276 .1401245 4.17 0.000 .3096886 .8589666 -------------+---------------------------------------------------------------- tau2 | _cons | .3879414 .1233578 3.14 0.002 .1461646 .6297182 ------------------------------------------------------------------------------ ------------------------------- Valérie OROZCO Toulouse School of Economics (INRA-GREMAQ) 21, allée de Brienne F-31000 Toulouse, France MF 219 +33 5 61 12 85 91 ------------------------------- * * 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/