02-13-2013 11:44 AM
I'm currently conducting a discriminant analysis on four predefined groups. my dataset includes 58 independent variables, so im trying to narrow them down a bit. When running the analysis i get a structure matrix with the discriminant loadings. However, i'm unsure as to what criteria i should use for selecting predictor variables. According to Hair et al. (2010) 0.4 should be the cut-off, but this cut-off point is very subjective. I heard that another option could be to take the square root of the mean of the discriminant loadings and apply that as the cut-off criteria. Has anyone got an answer for this situation. If so, sources would be very much appreciated!
02-14-2013 07:54 AM
Thx for the reply. I might have been a little unclear, but i have already carried out the analysis, running the standard enter method. However, i need a source for how i should interpret the resulting discriminant loadings (from the structure matrix). Hair et al. writes that variables with loadings below 0,400 should be ignored for interpretation purposes. as quoted "I heard that another option could be to take the square root of the mean of the discriminant loadings and apply that as the cut-off criteria".