| Science Forum Index » Statistics - Math Forum » Variation of Friedmans Rank Test... |
|
Page 1 of 1 |
|
| Author |
Message |
| Brew... |
 Posted: Mon Nov 02, 2009 1:05 pm |
|
|
|
Guest
|
I would like two compare two ranked lists, however some items from
list one are not contained in list two and vice versa. Also both lists
are of the same length. Is there some variation of the Friedman test
that I can perhaps use? |
|
|
| Back to top |
|
|
|
| Ray Koopman... |
| Posted: Mon Nov 02, 2009 8:54 pm |
|
|
|
Guest
|
On Nov 2, 3:05 pm, Brew <atb... at (no spam) gmail.com> wrote:
[quote]I would like two compare two ranked lists, however some items from
list one are not contained in list two and vice versa. Also both lists
are of the same length. Is there some variation of the Friedman test
that I can perhaps use?
[/quote]
If these are subjective rankings given by people then you should be
aware that there can be context effects. For instance, the ordering
of objects A,B,C,D when they are ranked along with E and F can differ
from their ordering when they are ranked along with G and H.
With regard to Rich's suggestion, be aware that the Spearman rank
correlation (or, when the there are more than two rankings, the
average Spearman correlation) can be expressed in terms of Kendall's
coefficient of concordance W, which in turn can be expressed as a
function of Friedman's Q, so the three approaches are equivalent. |
|
|
| Back to top |
|
|
|
| Rich Ulrich... |
| Posted: Mon Nov 02, 2009 11:59 pm |
|
|
|
Guest
|
On Mon, 2 Nov 2009 15:05:40 -0800 (PST), Brew <atbrew at (no spam) gmail.com>
wrote:
[quote]I would like two compare two ranked lists, however some items from
list one are not contained in list two and vice versa. Also both lists
are of the same length. Is there some variation of the Friedman test
that I can perhaps use?
[/quote]
No.
About the only thing that you can do with two separate
lists of ranks is look at the (rank) correlation of the items
that are included in both.
--
Rich Ulrich |
|
|
| Back to top |
|
|
|
| Brew... |
| Posted: Tue Nov 03, 2009 10:58 pm |
|
|
|
Guest
|
Hi,
Thanks for you're responses, what I am trying to do is come up with
some metric to compare two tag clouds built on a ranked list. All the
terms that fall below the top 50 are not to be considered, (I think
this is the best approach to take maybe I am wrong) So now I have two
ordering of terms in the top 50.
I am worried that by just ignoring the terms that dont exists, a
penalty of some form isn't introduced. I have hand rolled a metric but
I think I should probably be using some pre-packaged measure of
correlation between the clouds.
Its also been suggested to look at the Wilcoxon Signed Ranks Test, and
I assume that I should make some variation for missing terms.
Thanks a Million,
Brew |
|
|
| Back to top |
|
|
|
|
