Hi guys,
I am looking to look for any connection between an animals body condition and several numerical variables, some of which are not following a normal distribution. I am looking to run my data against: parasite load, pH, salinity, a measure for immune response, and maybe even more.
Does anyone have a suggestion for a statistical test that can handle more than three non-parametric variables?
Also, I am running my data in MiniTab 18 if anyone is interested.
Cheers
Joe
Hello Joe,
If the question concerns explanatory power of your IVs (parasite load, pH, salinity, etc.), and the body condition can be characterized as a categorical variable, then either binary logistic (if two levels of body condition) or multinomial logistic regression (more than two levels) could be used. Mintab 18 appears to include these methods (from vendor's web site).
Good luck with your work!
Thank you for the quick reply David.
I should have stated though that the body condition is a numeric variable calculated using morphometric measurements (length and weight).
Please tell us more about the body condition variable. Would it be fair to treat it as if it has interval scale properties? E.g., would it be fair to use means & SDs for descriptive purposes? If so, they I would just use OLS linear regression. If medians and IQRs are more appropriate for description, I might be inclined to use quantile regression. But a quick Google search suggests that Minitab may not have that option. HTH.
Bruce Weaver ,
It's a simple calculation using mass and length parameters for a approximation of the animals overall body. Each animal is measured and the calculation is applied for each og the individual animals.
Means and SD would be applicable as far as I can understand, as I have only a few outliers affecting the data (one of which I have omitted due to it being a measurement error).
MiniTab do not have that as far I can see. Could a multiple regression analysis work? What would be the pro's and con's compared with your recommendations?
Thanks :)
Given that means and SD are sensible for description, I would certainly estimate the OLS model, and then look at measures of influence such as Cook's Distance, DFITS and DFBETAS. A document I found below says that Minitab does not directly report the latter, but shows how you can calculate it (if you wish) for any suspect observations. HTH.
http://www2.hawaii.edu/~taylor/z632/MTBdiagnostics.pdf
PS- I forgot to mention in previous posts that variables are neither parametric nor non-parametric, statistical tests (or procedures) are. But even in that case, the distinction between parametric and non-parametric is rather slippery and not always very helpful. Take a look at the following links, for example. The second one is for a nice article by Ronán Michael Conroy in which he shows that the Wilcoxon-Mann-Whitney test (aka., the Mann-Whitney U test or the Wilcoxon rank sum test) does actually estimate a parameter. HTH. ;-)
- http://vassarstats.net/textbook/parametric.html
- https://www.stata-journal.com/article.html?article=st0253
- Badges
- Science topic
- Similar topics
- Biological Science
- Bioecology
- Systems Ecology
- Landscape Ecology
- Connectivity