# Scaling of linear discriminant from lda in MASS

I was attempting to replicate in R an example my instructor used in class from SPSS. However, it looks like the lda$scaling value function from MASS reports "a matrix which transforms observations to discriminant functions, normalized so that within groups covariance matrix is spherical." Thus, my linear discriminants are different than those from class. While I am aware standardizing the data does not matter (per this discussion), I am not sure if this is what is happening in lda$scaling.

Do the values reported by lda$scaling follow the same process as described in the linked discussion, and thus does not matter in interpretation? How do I 'un-scale' the linear discriminants so they match output from SPSS? Here is a reproducible example: require(MASS) gp.Y1=c(1,1,1,1,1,1,1,1,1,1,2,2,2,2,2,2,2,2,2,2,3,3,3,3,3,3,3,3,3,3,4,4,4,4,4,4,4,4,4,4) cm.x1=c(99,97,94,95,93,94,97,96,99,95,66,67,66,68,62,62,62,63,70,65,67,64,60,66,66,66,69,66,69,70,64,67,61,66,64,62,65,63,60,70) ap.x2=c(61,61,61,67,65,65,62,62,68,68,85,85,85,80,84,86,87,84,81,89,69,68,64,62,70,65,62,70,63,66,70,61,64,67,69,68,62,61,67,66) cp.x3=c(88,82,87,75,80,74,89,77,81,74,89,83,81,79,83,72,76,82,87,84,74,74,71,80,89,80,85,88,84,71,73,82,82,80,84,89,72,89,86,72) df=as.data.frame(cbind(gp.Y1,cm.x1,ap.x2,cp.x3)) mylda=lda(gp.Y1~cm.x1+ap.x2+cp.x3,data=df) mylda$scaling

• – ttnphns Feb 10 '18 at 4:29