Dunn posthoc to KW?
A. Scott McCulloch
having trouble finding how to do what I need.
I'm doing a bunch of nonparametric tests - Kruskal-Wallis - and want to
be able to do posthoc follow-ups, e.g. Dunn.
I can do this in InStat, but InStat doesn't let me have a grouping
column, and with a limited number of columns, would require multiple
data files for this project - so I'd really like to do it in SPSS, but
just can't find any such option.
Is there a way to do posthoc tests when doing K Independent Samples
nonparametric testing in SPSS?
A. Scott McCulloch
<myl...@ascottmcculloch.com> said:
Just realized that perhaps I should have made the title of my post more
general, since if there are other posthoc tests that can be easily done
in this context, that would be fine.
I also realized that I need posthoc testing for nonparametric K related
samples also.
Thanks,
Scott
--
A. Scott McCulloch, MBA
Ph.D. Candidate
Institute of Transpersonal Psychology
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
smccu...@itp.edu
http://www.itp.edu
http://www.multipleangles.com
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
Bruce Weaver
> On 2005-12-11 12:00:44 -0400, A. Scott McCulloch
> <myl...@ascottmcculloch.com> said:
>
>> I'm very new to SPSS 11.0.4 on Mac OS X, using the GradPak -- and I'm
>> having trouble finding how to do what I need.
>>
>> I'm doing a bunch of nonparametric tests - Kruskal-Wallis - and want
>> to be able to do posthoc follow-ups, e.g. Dunn.
>>
>> I can do this in InStat, but InStat doesn't let me have a grouping
>> column, and with a limited number of columns, would require multiple
>> data files for this project - so I'd really like to do it in SPSS, but
>> just can't find any such option.
>>
>> Is there a way to do posthoc tests when doing K Independent Samples
>> nonparametric testing in SPSS?
>
>
> Just realized that perhaps I should have made the title of my post more
> general, since if there are other posthoc tests that can be easily done
> in this context, that would be fine.
>
> I also realized that I need posthoc testing for nonparametric K related
> samples also.
>
> Thanks,
> Scott
>
I think you need to give a little more info.
1. How large are your samples?
2. Are the contrasts simple (i.e., each contrast involving only two
groups/treatments) or complex?
3. Why are you using a rank-based test? (Parametric tests are more
robust than many of us were taught in the past, and ANOVA might not be
as bad as you think.)
--
Bruce Weaver
bwe...@lakeheadu.ca
www.angelfire.com/wv/bwhomedir
A. Scott McCulloch
Thanks for the reply, Bruce...
The sample sizes vary from 3 to 62. I've got data from three separate
assessments, two of which have 5 subscales each (and no meaningful
"total" score), one of which has 5 subscales and a "total" score - so,
16 "variables" in all.
So, let's see if I can describe it succinctly:
I've got 6 samples, which are really 2 sub-groups (employees and
executives) from each of three larger samples (3 separate
organizations).
org. 1 N = 37
employee group 1 N = 34
exec group 1 N = 3
org. 2 N = 62
employee group 2 N = 58
exec group 2 N = 4
org. 3 N = 23
employee group 3 N = 18
exec group 3 N = 4
I am testing:
- between sub-groups within each organzition (2 independent samples -
empl v. exec for each org.) for each assessment subscale ;
- between organizations as a whole (3 independent samples) for each
assessment subscale;
- between sub-groups across organizations (3 independent samples - empl
v. empl v. empl, and exec v. exec v. exec) for each assessment subscale;
- between assessment subsalces (5 related samples) for each assessment
for each sub-group
Because of the nature of two of the assessments, my advisor and I
decided that the use of conventional statistical analysis should be
done relatively conservatively. (Two of the assessments are measuring
unconscious patterns of attitude and identification through a
projective measure in such a way that it's not entirely clear that all
of the assumptions of conventional statistics apply.)
So, although I had initially intended to do parametric tests in all of
the "non-exec" groups related testing, after much debate, we decided it
would be better to take the more conservative approach across the
board. Since the data for those two assessments requires additional,
qualitative, analysis as well, that would "balance out" the
conservative statistical approach when it comes to the discussion of
the data (i.e. only the statistics part of this study is "conventional"
or "conservative").
What I'm now trying to acheive is to be able to do the nonparametric
tests *with* the posthoc testing - which I can't figure out how to do
in SPSS. But, since SPSS allows for grouping columns and filtering, as
well as running tests on multiiple variables at once, I was hoping
there was a way -- InStat can do what I want in terms of the tests, but
doesn't give exact P values, and would require that I make 18 separate
data files, and run each variable separately.
If there's not a way to do it in SPSS, we did leave open the
possibility of doing parametric testing where all involved Ns are large
enough (i.e. no exec groups), and then do the most conservative posthoc
tests available; nonparametric tests would still be used for all
comparisons involving one or more exec groups. Given the nature of the
data, there is not a clear "right way" or "wrong way" - just a
difference in how I explain what I did and what the implications are --
all of which will be required regardless.
That's about as clear as mud, eh? ;-)
Avikam Nuriel
I want to write simple prog that write data to file
and to add to each case index i.e 1 to case nomber one 2 to case nomber
two etc.
regatds,
Avikam Nuriel
The Hebrew University
of Jerusalem
Israel
Art Kendall
should work.
compute seqnum = $casesnum.
var labels seqnum 'sequential position in file . . . '.
save outfile= 'd:\project\with sequence.sav'.
If you wish to create an ASCII file try something like this.
WRITE OUTFILE= 'd:\project\with sequence.txt' TABLE /seqnum all.
Art
A...@DrKendall.org
Social Research Consultants
University Park, MD USA
(Inside the Washington, DC beltway.)