Re: Rank correlation

This is a "non-technical" recipe-type of reply, beacuse your level of
knowledge of the subject matter seems to be rather ...

Please, before or at least immediately after "doing" the statistics
mentioned, read about them and try to understand them or at least try to
find a qualified somebody to explain things to you!

First, you can test if the "people" tend to agree about the rank-order of
preference for the topics with Kendall's coefficient of concordance (W).

Your data is already arranged properly for that - in the SPSS dialogue
(Analyze -> Nonparametric Tests -> Several Related Samples) just put in all
the variables (i.e., the five of them, one for each topic).

If the Exact button apears in the dialog (which means that you have the Exct
module in your SPSS installation), press it and set the option from
Asymptotic to Exact. Otherwise, you'll have to settle for the asymtotic p.

If W is significantly different from 0 (i.e., your p is below 0.05), it
might make sense to reorder the topics by average rank if they are not
already in such order (so first compute those averages if you haven't yet;
and drag columns left/right in the Data Editor to achieve the desired order)
and then test the significance of this order using Page's test. SPSS doesn't
have it "canned", but it's really easy to compute it with a spread***, or
even by hand. You even don't have to read a textbook on nonparametric
statistics (such as Siegel & Castellan, or Conover, though it wouldn't hurt
you), let alone dig up the original reference -- nowadays you just "consult"
Wikipedia:

http://en.wikipedia.org/wiki/Page's_trend_test

All you'll need is the critical value of the chi-squared distribution for 1
degree of freedom. I'll tell you it's 1.96 sqared, which makes ... (again,
if you care to find out why so, it won't hurt you, but if you don't care, so
be it).

BTW, your dataset (10 cases by 5 vars) is roughly on the boundaries that the
rules of thumb state for validity of the chi-squared approximation of the
distribution of the test statistic. But judging by your post, you won't
investigate that too deeply, so "just do it".

Lastly, I suggest you to forget about the "pairwise ranking" stuff. If the
Page test "yields" p<0.05, you have all the information for a useful
conclusion ("people agree that the topics can be ordered by preference as
follows: <order of topics by mean rank>").

If not, but W is significant (which is unlikely to happen), just judge by
mean ranks and conclude something like (> meaning "is preferred" over)

Topic 1 > Topics 2,3,4 > Topic 5.

If you want to do "advanced things" (which I doubt, but nevertheless), read
Legendre, P. (2005). Species Associations: The Kendall Coefficient of
Concordance Revisited. Journal of Agricultural, Biological, and
Environmental Statistics, Volume 10, Number 2, Pages 226-245. Then download
his software (trivial to use, DOS-type executable, prepare data in ASCII
file -- evrything explaind on the website) and run the permutation test for
W and the a-posteriori tests.

Regards,

Gaj Vidmar
University of Ljubljana, Faculty of Medicine, Institute of Biomedical
Informatics

"mpb" <mpb.vu2@xxxxxxxxx> wrote in message
news:1172654703.938565.161550@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
Hi,

I have conducted a ranking survey on 10 people to rank 5 topics 1 to
5, 1 is the most important and 5 is the least according to thier view;
i got followin data-


People topic1 topic2 topic3 topic4 topic5

people1 1 2 4 3 5
p2 2 1 3 5 4
p3 1 3 4 5 2
p4 2 5 1 3 4

and so on to

p10 4 2 3 5 1

how can i analyze these ranking using SPSS? how can i test
significance and compute pairwise ranking?

regards,

mpb


.