Which test should I use for multiple variables, multiple sites & non-parametric data?

#1
Hi All,

I've spent the past two weeks trying to figure out which test to use and have gotten nowhere so hoping someone here can help! I have been given data for 4 sites, where the percentages were recorded for a number of variables, including leaf litter cover, rhododendron cover, light availability, bare ground exposure, etc. For each site I have three measurements per variable.

The aim is to see whether there is a significant difference in habitat structure between the 4 sites, so I would like to do a test where multiple factors can be compared at once. I read that if it's percentage cover it is likely to need a non-parametric test so not sure if that's correct? I've come up with MANCOVA and Kruksall Wallis but don't know if I'm even heading in the right direction?

The other problem is that most stats packages I find seem to only compare up to three variables at once. Any suggestions greatly appreciated! :)
 
#2
Re: Which test should I use for multiple variables, multiple sites & non-parametric d

Hi All,

I've spent the past two weeks trying to figure out which test to use and have gotten nowhere so hoping someone here can help! I have been given data for 4 sites, where the percentages were recorded for a number of variables, including leaf litter cover, rhododendron cover, light availability, bare ground exposure, etc. For each site I have three measurements per variable.

The aim is to see whether there is a significant difference in habitat structure between the 4 sites, so I would like to do a test where multiple factors can be compared at once. I read that if it's percentage cover it is likely to need a non-parametric test so not sure if that's correct? I've come up with MANCOVA and Kruksall Wallis but don't know if I'm even heading in the right direction?

The other problem is that most stats packages I find seem to only compare up to three variables at once. Any suggestions greatly appreciated! :)
Hi, Liz. It seems the question requires some domain knowledge. Could you clarify what "habitat structure" entails? Are the IV's all independent of each other (i.e. does light availability affect bare ground exposure, for example)? Do the vars follow a normal distribution (this is where the non/parametric part may become applicable)?

As far as packages go, all major stats packages should be able to compare multiple vars, if you are setting up the problem correctly. If you want something sophisticated but not very user-friendly (i.e. programming, no real GUI), you can do any sort of analysis with R (http://www.r-project.org/).
 
#3
Re: Which test should I use for multiple variables, multiple sites & non-parametric d

Thanks for your speedy post! I've given it the term habitat structure, but really I just want to know whether there is a difference between the sites in terms of percentage cover for the variables. Your question about them being independent is something I'd overlooked (I haven't done stats for about 10 years!! I think all the factors are likely to be linked, as the main reason for comparing the sites is because of the difference in rhododendron treatments (Site A: Rhodo never present, site B: rhodo removed in 2008, site C: rhodo removed in 2012, site D: rhodo present and not removed). Therefore, the differences between the habitat structures are almost certainly linked with one another, I just need to show it statistically.

P.S. I am quite scared of using 'R', but if you have a suggestion for which stats test I should be using I will hopefully be able to use minitabe or PAST Stats, etc. :)
 

Karabiner

TS Contributor
#4
Re: Which test should I use for multiple variables, multiple sites & non-parametric d

What do you mean by 3 measurements per variable? The whole site was
measured at 3 different times? Three randomy selected areas from each site
were measured?

In any case, I doubt that ANOVA/MANOVA will give you reliable results,
due to the very small sample size. If you have 3 randomly selected areas
from each site, then Kruskal-Wallis tests seem appropriate; there will
be the need to consider possible problems with regard to multiple testing,
though.

With kind regards

K.
 
#5
Re: Which test should I use for multiple variables, multiple sites & non-parametric d

I mean three randomly selected areas. Within each samle measurement there were 4 pseudo-replicates (i.e. 12 individual measurements in total). So I suppose in theory I have a sample size of 12. The difficulty I have with the Kruskal-Wallis test is that it only seems to allow comparison of three variables at once. I was hoping for a test that would allow me to compare more than 3 variables. I think PCA might be relevant but not sure what free software is available to run this. Sorry for the bombardment of comments!