Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down at the end of May, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Significance Tests for Individual Random Effect Parameters/Variance Components in Hierarchical Linear Modeling


From   Maarten Buis <maartenlbuis@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Significance Tests for Individual Random Effect Parameters/Variance Components in Hierarchical Linear Modeling
Date   Fri, 8 Feb 2013 10:06:58 +0100

On Thu, Feb 7, 2013 at 10:18 PM, Anthony Fulginiti wrote:
> I was working on a manuscript a few months ago using hierarchical linear modeling to examine group differences in self-esteem over time.  During the manuscript preparation, I was reviewing many articles and books using hierarchical linear modeling and noticed that results for individual random effect parameters/variance components were usually accompanied by a chi-square/z statistic and a specific probability value.  However, in the STATA output, the random effect parameters are listed with an estimate, standard error and 95% confidence interval but no such chi-square/z statistic or specific probability value.
>
> Is there a way for STATA to produce such a statistic/probability value (via conducting significance tests) for the individual random effect parameters?  If so, how?

The problem with these chi-square/z-statistic is that they are test
statistics of a null hypothesis that is "on the boundary of the
parameter space". The null hypothesis is that the variance equals 0,
and a variance can only be larger than or equal to 0. Weird things
happen at such boundaries, and as a concequence the sampling
distribution of the chi-square under the null-hypothesis will not
correspond with a chi-square(1) distribution and the sampling
distribution of the z statistic under the null hypothesis will not
correspond to a normal(0,1) distribution. That is a good reason for
not reporting such statistics. See for example:

Gutierrez, R., S. Carter, and D. M. Drukker. 2001. On boundary-value
likelihood-ratio tests. Stata Technical Bulletin 60: 15–18.
<http://www.stata.com/products/stb/journals/stb60.pdf>

Hope this helps,
Maarten

By the way, in the Statalist FAQ you could (and should) have read that
Stata is spelled Stata and not STATA.

---------------------------------
Maarten L. Buis
WZB
Reichpietschufer 50
10785 Berlin
Germany

http://www.maartenbuis.nl
---------------------------------

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index