Multivariate analysis for proportional data
On 25/12/14 04:05, Rich Shepard wrote:
On Wed, 24 Dec 2014, Mayara Amescua wrote:
I have a data set owing some response variables with proportional values and independent variables of mixed effect. How it is possible performing a multivariate analysis of this type once the MCMCglmm package doesn't allow for binomial family?
Mayara, If your data are proportions you should look carefully at compositional data analysis (CoDA). There are seveal R packages for this, including 'compositions', 'robCompositions' (robust compositions), and 'zCompositions'. For documention and lerning there are the 'Lecture Notes on Compositoinsl Data Anayis' by Pawlowsky-Glahn, V., Egozcue, J. J., and Tolosana-Delgado, R. (available as a PDF but I don't recall the URL off-hand); 'Analyzing Compositional Data with R' by van den Boogaart, K.G. and Tolosana-Delgado, R.; and 'A Concise Guide to Compositional Data Analysis' by John Aitchison, also available as a PDF.
I am of course prejudiced, but I think that in respect of compositional data analysis one would be well-advised to have a read of "Colours and Cocktails: Compositional Data Analysis. 2013 Lancaster Lecture" by J. L. Sealy and A. H. Welsh, Aust. & N. Z. J. Statist., vol. 56, 2014, pp. 145 -- 169. cheers, Rolf Turner
Rolf Turner Technical Editor ANZJS