2

# Wilcoxon Signed-Ranks Test

Fri 13 Aug, 2010 11:05 pm
I have two related variables, female manta ray size (e.g. 3.5 m) and the number of males in her mating train (e.g. 14). Can I use a Wilcoxon Signed-Ranks Test to test if larger females have significantly more males in her mating train? I have samples for 12 females.
• Topic Stats
• Top Replies
Type: Question • Score: 2 • Views: 3,551 • Replies: 10
No top replies

Butrflynet

1
Sat 14 Aug, 2010 01:07 am
@test12345,
I have no idea, but hopefully this will get you started in the right direction if it is that test that is required:

http://en.wikipedia.org/wiki/Wilcoxon_signed-rank_test
0 Replies

Butrflynet

1
Sat 14 Aug, 2010 01:09 am
@test12345,
This link may help too. It gives you an opportunity to supply the two variables for the test to get a result.

http://faculty.vassar.edu/lowry/wilcoxon.html
0 Replies

test12345

1
Sat 14 Aug, 2010 02:03 am
I'm still somewhat confused because most Wilcoxon examples use before and after treatment examples, which means the variables are the same and taking the difference between the pairs makes sense. But if the variables are different, for example in my example of Body Size paired with Birthing rate, it seems like you can't just subtract "meters" from "# pups per year" to get the absolute difference. I guess they are both numbers so technically you could but seems strange for some reason. Anyone know if this is standard for using Wilcoxon Signed-Rank Test?
test12345

1
Sat 14 Aug, 2010 02:07 am
@test12345,
Ok, got my answer, did not read with enough detail. It applies to repeated measures, before and after, and MATCHED PAIRS. I think this is what I needed to know. Cheers

"Like the t-test for correlated samples, the Wilcoxon signed-ranks test applies to two-sample designs involving repeated measures, matched pairs, or "before" and "after" measures. "
laughoutlood

1
Sat 14 Aug, 2010 05:19 am
@test12345,
Test, all the best with your studies.
test12345

1
Sat 14 Aug, 2010 02:01 pm
@laughoutlood,
Ok, just to clarify, it seems I can't use Wilcoxon on my data because I'm not resampling the same individuals over time. I guess "Matched Pairs" means the same individual is measured twice, not that two different variables are measured on the same individual. If anyone can confirm this that would be appreciated.
engineer

1
Sat 14 Aug, 2010 02:39 pm
@test12345,
I think you are over-thinking it. A simple linear regression should tell you if the variables correlate.
JPB

1
Sat 14 Aug, 2010 02:44 pm
@engineer,
I had a whole post typed out on correlation and regression but then deleted it because she wants a significance test comparing size vs male train. She can do a linear regression against a null of b=0, but that doesn't exactly answer the question as she's stated it. She also only has 12 females so categorizing the females into buckets of small and large (even if she pushes mid-sized ones into one of those buckets) doesn't give her enough in each cell to do a chi-square. Maybe a Fisher's Exact test, but she's right. She doesn't want to use a Wilcoxan Rank Sum.
engineer

1
Sat 14 Aug, 2010 02:53 pm
@JPB,
I'd like to see the data set. If you've got enough spread in the independent variable, a linear regression might show significance. Post the data and let's see what you've got.
High Seas

1
Sun 15 Aug, 2010 07:53 am
@engineer,
Agreed. Wilcoxon is a non-parametric test and the original post doesn't specify any a priori assumption the underlying distribution is non-Gaussian; it might be. Post your small sample of 24 data points in pairs. Also read: http://www.graphpad.com/www/Book/Choose.htm
Quote:
Calculate linear regressions only if one of the variables (X) is likely to precede or cause the other variable (Y). Definitely choose linear regression if you manipulated the X variable. It makes a big difference which variable is called X and which is called Y, as linear regression calculations are not symmetrical with respect to X and Y. ....In contrast, linear correlation calculations are symmetrical with respect to X and Y. If you swap the labels X and Y, you will still get the same correlation coefficient.
0 Replies

### Related Topics

Amount of Time - Question by Randy Dandy
Statistics - Question by ekkline
Math of infinity - Discussion by dalehileman
Probability Question. - Discussion by babemomlover
Do I make the mistake? - Question by tetupioxi

1. Forums
2. » Wilcoxon Signed-Ranks Test