Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | John Antonakis <john.antonakis@unil.ch> |
To | statalist@hsphsun2.harvard.edu |
Subject | Re: st: Same results OLS and Random Effects |
Date | Sat, 19 Jun 2010 21:07:44 +0200 |
loneway y zidTo better grasp the implication of ICC = 0, one formula to calculate the ICC is (after estimating a basic ANOVA):
(MSb - MSw) / ((MSb + (k-1)*MSw)) Note: k = average group size; MS=mean square; b=between; w=within. The rho in xtreg is calculated in a slightly different way.Also, after running xtreg, what do you get when you run -xttest0-? Next, after estimating the random-effects estimator ideally with cluster-robust SE, what does -xtoverid- give you?. If the test is significant, you must use fixed-effects (in the context of xtreg, fe or with OLS and explicit dummies; and it would be best to also use a cluster correction for the standard errors, because even if you don't have between-group variation nor random effects, within cluster-disturbances may still be correlated).
HTH, John. ____________________________________________________Prof. John Antonakis, Associate Dean Faculty of Business and Economics
Department of Organizational Behavior University of Lausanne Internef #618 CH-1015 Lausanne-Dorigny Switzerland Tel ++41 (0)21 692-3438 Fax ++41 (0)21 692-3305 Faculty page: http://www.hec.unil.ch/people/jantonakis Personal page: http://www.hec.unil.ch/jantonakis ____________________________________________________ On 19.06.2010 19:18, Pelkowski, Jodi wrote:
Nature of problem: In running xtreg with random effects, sigma_u and rho are reports as “0”. I obtain the exact same coefficients and standard errors when OLS is run. (So basically Stata is defaulting to OLS). There is no problem using fixed effects. When the same regression is estimated using the Limdep regression package, sigma_u and rho are both >0 and the coefficients and standard errors differ from OLS. Why is Stata defaulting to OLS. Whom do I believe Stata or Limdep? Note: I checked to see if there was anything strange with the regressors or the id variable. There isn’t. For example if I switch the dependent variable with a regressor, Stata works fine. The original dependent variable takes on integer values between 1 and 255 with lots of variation. The sample size is 931 and there are 108 groups. Below is a simplified version of regression that was run: . xtset zid panel variable: zid (unbalanced) . xtreg y x stars, re Random-effects GLS regression Number of obs = 931 Group variable: zid Number of groups = 108 R-sq: within = 0.0193 Obs per group: min = 1 between = 0.1555 avg = 8.6 overall = 0.0534 max = 37 Random effects u_i ~ Gaussian Wald chi2(2) = 52.36 corr(u_i, X) = 0 (assumed) Prob > chi2 = 0.0000 ------------------------------------------------------------------------------ y | Coef. Std. Err. z P>|z| [95% Conf. Interval] -------------+---------------------------------------------------------------- x | -.5014859 .2454555 -2.04 0.041 -.9825698 -.020402 stars | -14.90484 2.827657 -5.27 0.000 -20.44694 -9.36273 _cons | 207.6687 17.41683 11.92 0.000 173.5324 241.8051 -------------+---------------------------------------------------------------- sigma_u | 0 sigma_e | 73.292219 rho | 0 (fraction of variance due to u_i) . sum y x stars zid Variable | Obs Mean Std. Dev. Min Max -------------+-------------------------------------------------------- y | 931 120.6498 74.36455 1 255 x | 931 78.77133 10.84436 44.01 106.93 stars | 931 3.18797 .9413473 1 5 zid | 931 57.77336 32.22844 1 114 * * 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/
* * 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/