Suppose I have 3 response variables (Y1, Y2, Y3, which are vectors of count data). Assume I also have multiple co-variates that are both categorical (e.g. X1, X2, X3) and numerical (e.g. Z1, Z2, Z3). The goal is to, for instance, predict the response variables as a function of the co-variates, and also assess for any interaction between the categorical co-variates. By considering only a single response Y1, I can use e.g. a Binomial or Poisson distribution to model the data. But I want to model the whole data set as a multivariate distribution. Does anyone know of an "R" library for this problem? Any suggestions?
Hi Jimmy, you may want to look into ordination statistics, in my understanding it will allow you to look at relationships between covariates and response variables as well as among covariates. It is normally used with ecological community data but can be used for count data also. Look into Hellinger transformations to normalize your data before running analysis (see below paper for example). I am not an expert on this method but hopefully this helps!
summary of ordination stats: http://ordination.okstate.edu/overview.htm
R package: http://cran.r-project.org/web/packages/vegan/vegan.pdf
example: Denton, R.D., and S.C. Richter. 2013.Amphibian communities in natural and constructed ridge top wetlands with implications for wetland construction. Journal of Wildlife Management 77:886-889.
- Badges
- Science topic
- Similar topics
- Biological Science
- Physiology
- Physiological Processes
- Stress