Inna Khousnoullina <inna.khousnoullina@uni-konstanz.de> asks
about a nested anova model
. anova f1 sex_int / sex|sex_int /
where the F test of sex_int uses sex|sex_int as the error term.
This gives sex_int a p-value of .0712 (with 1 numerator and 2
denominator degrees of freedom). (The test of sex|sex_int against
residual error gives a p-value of .0423, with 2 and 3524 dfs)
Inna also ran
. anova f1 sex_int sex|sex_int
In this case the F test of sex_int uses residual as the error
term. This gives sex_int a p-value of .0000 (with 1 numerator
and 3524 denominator degrees of freedom).
Inna asks:
> which output should I take for calculating intraclass
> correlation coefficient? In the first case my major effect
> (interviewer gender') is not significant. I would prefer the
> second one. By are there some rules in this case?
The general rule is that you can not ignore the nested term when
it is significant. I believe the general guideline is that you
could not go to the second model unless the first model test of
sex|sex_int had a p-value of .25 or higher.
For an example and discussion of pooling nested terms in a nested
anova, see Example 3, pages 80-81 of "[R] anova postestimation"
in the Version 9 Stata manual.
Ken Higbee khigbee@stata.com
StataCorp 1-800-STATAPC
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/