BenBarnard / covTestR

Tests for Covariance Matrices
https://www.bearstatistics.com/covTestR
2 stars 0 forks source link

Inconsistency between differences packages for BoxM test associated to df #12

Open Flaiba opened 1 year ago

Flaiba commented 1 year ago

Hello, I want to do a test to compare matrix variance-covariance between two treatments. Initially, I decided to use a Box M test employing the package rstatix and biotools with the following syntaxis

(As example mode I used the iris data)

library(biotools) boxM(iris[,1:4],iris$Species) library(rstatix ) box_m(iris[, -5], iris[, 5])

But when I used library(covTestR), employing the script homogeneityCovariances(iris, group = Species) or

irisSpecies <- unique(iris$Species) iris_ls <- lapply(irisSpecies, function(x){as.matrix(iris[iris$Species == x, 1:4])}) names(iris_ls) <- irisSpecies

BoxesM(iris_ls)

I found a different result compared to other packages. For the first 2 tests, I found significant differences, but not for the last one. The differences between tests are associated with the degree of freedom. This discrepancy makes me doubt about using tests other than Box M, whether they are reliable or not. My concern arises because the Box M test is very sensitive to non-compliance with multivariate non-normality, and for my data, such a thing happens (My data do not accomplish multivariate normality). So I am interested in applying another more suitable test to compare the matrices, however, I would not know which one to apply from the others available in the 'covTestR' package and additionally if these other tests are performing the test reliably. Thanks a lot

image