I performed an LDA for 50 elements vis-a-vis land cover type on R. The final plot showed 13 elements which contributed the most to the separability of the 5 land cover types. However, when I created a correlation matrix for the 13 elements, some of these elements have strong positive correlations despite their LDA arrows showing up in opposite directions. Is this actually possible and can this be explained? Or the problem may actually be my code?
What I did:
I performed an LDA on R. The output are two biplots: one with the land covers, and one with 13 elements (from the original 50 elements). I then ran the cor function for the 13 elements to confirm the placement of their loadings in the biplot (i.e. if elements A and B are strongly positively correlated in the correlation matrix, then their arrows should be in the same direction). However, the opposite has occurred. Elements that are positively correlated (r= 0.7 to 0.8) sometimes end up with arrows in the opposite direction. What should I do?