ANOVA Assumptions for Percentages

#1
I am trying to analyze land use data (e.g. an given area is made up of 14% residential land, 36% agricultural land, and 50% forested land). So the data from landuse is the independent factors while I have several categorical fixed factors I am interested in seeing the relationships with.

When performing an ANOVA I cannot get this data to pass a Levene's Test (even with an arcsine transformation). Can I not perform an ANOVA on percentage data. Am I better off trying to correlate these categorical data with the percentage data?

Any suggestions are greatly appreciated.

Rich
 

terzi

TS Contributor
#2
I've heard of some issues regarding Levene's test. I'd suggest you to use a graphic review instead (scatter plot of fitted versus residuals).

Regarding transformations, I'm aware that the one that works the best for stabilizing variance is the inverse, x^-1. Still, if you are having problems with assumptions and you want to avoid the complications that come with transformed data, you can try some non-parametric alternatives, like Kruskal Wallis or Friedman Test.