Hi,
Firstly, sorry I didn't make myself clear. The post was referring to tests
where you could carry out multiple comparision tests using nonparametric
tests the same way as the oneway with the scheffe option would do (report
the significance of the pairwise difference in means between the various
groups).
My problem was this.
I have a variable class which represents 4 classes of people who buy shares.
They are male exec directors, female execdirectors, male non execdirectors,
female non execdirectors. The corresponding var is class.
I also have a variable caar. Which is the sum of returns they have made by
trading shares over some no of days.
I want to see which group on average does better.
The stats are as follows
. robvar caar, by(class)
| Summary of caar
class | Mean Std. Dev. Freq.
------------+------------------------------------
1 | .02007266 .10025548 6564
2 | .02274502 .109724 3341
3 | .00473056 .12638999 108
4 | .02476845 .08051064 48
------------+------------------------------------
Total | .0208178 .10372577 10061
W0 = 3.4910828 df(3, 10057) Pr > F = .01497735
So clearly, the variance is not homogenous across groups. Also, I was
worried about the small size of the last group and therefore the concern
with using the ANOVA. When I add up the returns over 60days there seems to
be no problem (ignoring the bartlett's reported in the anova results) as
the results show
. robvar caar, by(class)
| Summary of caar
class | Mean Std. Dev. Freq.
------------+------------------------------------
1 | .04196137 .19154713 6564
2 | .04502643 .20246685 3341
3 | .03568936 .21235151 108
4 | .0182026 .19047529 48
------------+------------------------------------
Total | .04279852 .19545032 10061
W0 = 1.3983703 df(3, 10057) Pr > F = .24121868
And I can use the oneway with the scheffe option to give me
oneway caar class,scheffe
Analysis of Variance
Source SS df MS F Prob > F
------------------------------------------------------------------------
Between groups .055679831 3 .018559944 0.49 0.6922
Within groups 384.244654 10057 .038206687
------------------------------------------------------------------------
Total 384.300334 10060 .038200828
Bartlett's test for equal variances: chi2(3) = 15.3442 Prob>chi2 = 0.002
Comparison of caar by class
(Scheffe)
Row Mean-|
Col Mean | 1 2 3
---------+---------------------------------
2 | .003065
| 0.909
|
3 | -.006272 -.009337
| 0.991 0.971
|
4 | -.023759 -.026824 -.017487
| 0.872 0.828 0.966
Carrying out the kwallis test tells me
.kwallis caar,by(class)
Test: Equality of populations (Kruskal-Wallis test)
+--------------------------+
| class | Obs | Rank Sum |
|-------+------+-----------|
| 1 | 6564 | 3.31e+07 |
| 2 | 3341 | 1.68e+07 |
| 3 | 108 | 532072.00 |
| 4 | 48 | 222811.00 |
+--------------------------+
chi-squared = 1.092 with 3 d.f.
probability = 0.7791
chi-squared with ties = 1.092 with 3 d.f.
probability = 0.7791
which confirms the result of the levene's test above.
Are there options to make kwallis report results by pairs of groups?
Thanks
rajesh
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of n j cox
Sent: 09 January 2007 18:30
To: [email protected]
Subject: Re: RE: st: RE: RE: -oneway- and unequal variances - thanks
A short answer to the question posed here is Yes,
many such techniques exist, and one has already been
mentioned more than once in this thread, Kruskal-Wallis
(in Stata -kwallis-).
Your library should have texts on nonparametrics at
a variety of levels, and the Stata manual has many
references. Also, some others such as the Friedman
and Jonckheere-Terpstra test have been implemented
by users. Use -findit- to find them.
That said, there is a discernible bias on the part
of StataCorp that such tests are not nearly as useful
or important as their proponents make them out to be.
Be that as it may, once you get past a few traditional
basics such as Mann-Whitney and K-W, there are many,
many such tests and it's rather hard to identify
which are really deserving of implementation.
A longer answer would need to be based on what your
"similar problem" is. I rather suspect that your real
question is much more specific than the one you asked,
as this is the only way to explain the utterly incorrect
report that there are no Stata commands in this area.
Nick
[email protected]
Rajesh Tharyan
I was facing a similar problem and was following the discussion. Are there
are non parametric tests to compare multiple groups, some user ado perhaps
that you know of? I read a previous post on this list and the answer was
no. I use stata 8.
Regards
Rajesh
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Michael Mitchell
Sent: 09 January 2007 04:02
To: [email protected]
Subject: Re: st: RE: RE: -oneway- and unequal variances
Greetings
I would suggest seeing
http://www.ats.ucla.edu/stat/stata/library/homvar.htm
which has a detailed discussion of this topic, a program you can use
to see the impact for your situation, and additional references.
*
* 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/
*
* 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/
