I’m studying about beetle infestation in wood. So, I randomly selected trees in the forest to observe the presence of beetles and some properties of those tress (10 properties such as diameter, wood density, C/N ratio, etc.). I would like to know which factor(s) (properties of wood/tree) relate to the infestation selection by those beetles. Someone suggested me to use multivariate analysis in R but I’ve never used it before. Which statistic should I use for my study because there are many statistics in the group of multivariate analysis such as PCA, discriminant analysis and cluster analysis. If possible, could you please give me an example or step for the analysis in R?
Thank you in advance.
Multivariate analysis is the correct method, such as ANOVA, MANOVA, or GLM (a type of regression with multiple outcomes). You must analyse ALL of your data together so as to reduce a large number of variables to a smaller number of factors for data modeling. But remember you only have one replicate i.e. a forest. But the results are still relevant, especially if the trees are widely spaced. You could first derive your ANOVA table and then do post-hoc tests to observe for differences between the tree property variables. Then consider PCA, canonical correlation analysis (CCA), or cluster analysis – these are all good and will complement your main analysis. Discriminant functional analysis is like ANOVA, but in reverse, so I don't think you need to do that imo :)
How do you measure the dependent variable - infestation of bees? Is it on a binary scale (Yes-1 and No-0) as likely? That would make it ordinal. The independent variables, the diameter, wood density,etc. are all ratio variables. The solution may lie on a multiple regression of the dependent variable on the independent variables. The statistically significant variables will be the relevant ones. The non-significant variables are not relevant.
first try dimension reduction for your covariates to derive possible components that factor together to avoid collinarity in your analysis,
if presence is coded binary a (first) generalized mixed model could be:
model1 = glmer(BeetleYesNo~factor1+factor2+factor3+(1|treeidentity),data=URdata, family="binomial")
drop me a note if you need more help
Best
Xaver
- Badges
- Science method
- Similar topics
- Biological Science
- Ecology