Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
| From | alfredo jimenez palmero <alfredojimenezpalmero@hotmail.com> |
| To | <statalist@hsphsun2.harvard.edu> |
| Subject | st: FW: Conditional or Nested Logit Model |
| Date | Mon, 6 Sep 2010 10:58:16 +0000 |
Dear Stata users, I want to run a model that could be either a Conditional Logit Model or a Nested Logit Model with two branches and two steps. Following the Stata user guide, I should run a hausman test to verify wheter there is a tree structure decision, or if all alternatives are possible substitute (testing the independence of irrelevant alternatives IIA). To do so, I have to run the full CLM, save the results under Hausman, and then run the estimation again, excluding the alternatives of one of the two branches. Finally I have to run Hausman test with the less option and I will get a p-value indicating if the IIA is supported (if the p-value is high) or not. I have done these before, sometimes confirming IIA, sometimes not. however this is the first time Stata offers me the following result: . hausman, less You used the old syntax of hausman. Click here to learn about the new syntax. ---- Coefficients ---- | (b) (B) (b-B) sqrt(diag(V_b-V_B)) | Consistent Efficient Difference S.E. -------------+---------------------------------------------------------------- Var 1 | -3.598664 1.359594 -4.958258 2.169517 Var 2 | -2.774566 -1.055487 -1.719079 1.812072 Var 3| -4.721951 .047857 -4.769808 3.427473 Var 4 | 3.548023 .3284012 3.219622 4.252454 Var 5 | .0101701 -.0059828 .0161529 .1143274 Var 6| 3.361641 1.280073 2.081568 .7336166 Var 7| 1.877154 .2127511 1.664403 1.475632 ------------------------------------------------------------------------------ b = consistent under Ho and Ha; obtained from clogit B = inconsistent under Ha, efficient under Ho; obtained from clogit Test: Ho: difference in coefficients not systematic chi2(7) = (b-B)'[(V_b-V_B)^(-1)](b-B) = -503.27 chi2<0 ==> model fitted on these data fails to meet the asymptotic assumptions of the Hausman test; see suest for a generalized test Can anyone explain to me what is wrong? Is this evidence enough to justify choosing NLM or CLM? Perhaps the problems is excluding this branch? Should I exclude the branch I use as a reference or the one I interact the individual attributes with? Thank you! Alfredo Jiménez Palmero University of Burgos * * 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/