Although Joseph as always gives good advice, it is also arguable that if
a structure of groups and of individuals within groups is key, then any
test like this is misdirected. It is difficult to advise without knowing
much more about the data.
Nick
n.j.cox@durham.ac.uk
Joseph Coveney
If you have ties in your response variable, then you'd be better off
using the
with-ties p-value. You can check for ties with the Stata command
-isid-.
sysuse census
kwallis medage, by(region)
isid medage
If you're interested in within-person comparisons, then there's
Friedman's test;
there are user-written commands, e.g., -friedman- and -skilmack-. For
between-groups testing where there are multiple observations on each
person,
there is van Elteren's test (two groups at a time), again, with a
user-written
command, -vanelteren-. Another user-written command that could be
helpful in
your case is -somersd-, which allows for a -cluster()- option. All of
these can
be downloaded from SSC (-help ssc- for how). You can try -permute- with
a
randomized-block ANOVA, too. As I recall, there's a worked example in
the
documentation for -permute-, which is an official Stata command.
"a statistical test for medians"--many nonparametric tests are tests of
location
only when the scale is the same between groups.
Mahboobeh Safaeian wrote:
I am using the Kruskal Wallis rank test to see if there are any
statistically
significant differences in 3 different groups. I have 2 questions
(unrelated):
1) How do I determine whether I use the p-values with ties or without
ties?
2) is there a statistical test for medians that can account for multiple
observations per person (kind of like a GEE)?
*
* 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/