Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | Alan Acock <acock@mac.com> |
To | statalist@hsphsun2.harvard.edu |
Subject | st: tests with xtmixed and small number of groups |
Date | Sun, 19 Feb 2012 14:03:20 -0800 |
Maas & Hox, Sufficient Sample Sizes for Multilevel Modeling, Methodology (2005); Vol. 1: 86-92, DOI 10.1027/1614-1881.1.3.86 conclude: "Since the estimates of the regression coefficients are unbiased, even if the sample is as small as 10 groups of five units, bootstrapping or other simulation-based methods may also be useful to assess the sampling variablity, provided we are interested only in the regression coefficients." Stata's xt commands report z-tests which obviously assume a large number of groups.I've seen people use number of groups - 1 degrees of freedom and compare these z-tests scores to the t distribution for groups- 1 degrees of freedom. How can we get unbiased standard errors and t-tests with a small number of groups? Is there some other better way? I have 14 schools with roughly 100 students from each school Alan Acock * * 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/