Should I use Mann-Whitney?

#1
I have two independent, unpaired samples. Both distributions are non-Gaussian. One is plant data (n=95), the other is insect data (n=48). Both sets of data were sampled from the same location. Both sets of data are scores (i.e. nothing was measured). The scores are assigned to each species of plant and insect and are based on fidelity to undisturbed ecosystems. I did not create the scores, they were taken from the literature and are based on habitat data and expert opinion.

I need to determine what, if any, relationship exists between the two sets of scores.

Would Mann-Whitney be the appropriate test?
 

gianmarco

TS Contributor
#2
Hi, given the non-normal distribution of data, non parametric statistics are the appropriate choice. But, since you are seeking to assess the degree of correlation between the two set of measurements, you should use Spearman correlation. It measure the degree of monotonic correlation between two variables.

For further details, you could search for Spearman correlation in this same forum. I remember to have replied to several questions about it; so earlier posts can be useful as well.

Hope this helps
regards
gm
 
#3
I thought about Spearman, since it can be used with ordinal variables (which is what the scores I mentioned are), but I guess I didn't realize it could be used on unequal pairs (i.e. a correlation between n=95 and n=48).

Thanks for the help! I appreciate it.
 
Last edited:

gianmarco

TS Contributor
#4
Sorry, while replying I forgot the important point of unequal samples size!!
Obviously, you cannot calculate correlation of any kind with different sample size.

I am unsure about what test you can perform in this situation. I was thinking about Mann-Whitney, that will test if a significative difference exist between the mean ranks of your samples. But, this will not provide any kind
of correlation, just an hypothesis test about the difference between the scores. I doubt that this is what you need...

May be somone else in the forum will provide further help

regards
gm
 
#6
so you have scores for multiple plant species and multiple insect species?
Yes, there are 95 species of plant and 48 species of insect, all have a score. Many scores are repeated, but only because multiple species have the same fidelity rank.

Also, "relationship" in my first post was probably a poor choice of words. I do not necessarily need to find a correlation. What I am basically looking for is to determine if the central tendency of the plant scores are different from the central tendency of the insect scores. Conventional wisdom (and a lot of anecdotal data in the literature) is that if the plants have a higher fidelity rank the insects will by default.

If the data were different I would just use students-t to see if there was a significant difference between the means. Since I can't do that I'm trying to see what else could work.

My first thought was Mann-Whitney.

Basically, my hypothesis is that there should be no significant difference between the scores of the plants and insects (i.e. if insects of a certain fidelty score prefer habitats of plants with a certain fidelity score then the differences between those two scores for the two assemblages should be negligble).

Sorry for all the editing of this post - its late here and I had trouble organizing my thoughts.
 
Last edited:

bugman

Super Moderator
#8
Yeah Grungy Goose I'll second that.

I think you are ok with the unequal sample size, but you might need to confirm this. I don't often look at rank data, so I'm a little rusty.