Re: st: Interpreting confidence interval of log of variance, xtlogit

[Maarten Buis <maartenlbuis@gmail.com>]

On Thu, Dec 6, 2012 at 3:17 PM, Therese Saltkjel wrote:
> I have performed a two level random intercept logit model; the xtlogit procedure in STATA (12.0), where individuals are nested within countries. The provided likelihood ratio test indicate that the rho is statistically significant, supporting the use of a multilevel model. However the log of variance, Insig2u in the output, is not significant as the confidence interval includes zero. I need help to interpret these results. In addition I need a general advice on what to report in a table; the log of variance or standard deviation of random intercept (sigma_u in the output)?

When testing you need to first think about what the null hypothesis
is, and whether that is a statement you want to test. If  you test
whether the log(variance) = 0, you are testing whether the variance =
1. Now it is up to you to decide whether that is meaningful.

Same is true with what you want to report. Remember that a statistical
model is just there to summarize the data. The stuff you report is
only there because it means something. So you need to decide which
story you want to tell, and which coefficients help you tell that
story. I would typically prefer the standard deviations, but others
prefer the variances and for good reasons. I don't know anyone who
prefers the log(variances), but I don't know everybody.

Hope this helps,
Maarten

---------------------------------
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/