Dear List
I have one of my colleague is doing something that looks a bit odd to me. I haver talked to him on numerous occasion about it but he thinks what he does is perfectly valid.
He has a large dataset (about 1000 records and 20-30 attitude scales). He first does a factor analysis to reduce the number of dimension from 20-30 to 5-8 factors. The odd part is that he then use MDS (SAS doesn't have the limit that SPSS and others have on MDS) to create a 2 dimensional map off those 6-8 factors where each set of coordinates represent one person.
I don't think this can possibly work as the factors are orthogonal. I can't see how MDS would take 6-8 independent dimensions and reduce them further to 2 dimensions. MDS will spit out some coordinates anyway but when I looked at the relationship between the original factor scores and the MDS coordinates it is pretty weak. A few factors only would correlate with the MDS dimensions.
Has anybody got some views on that? Who is right?
(None of us are statistician)
Cheers
Eric
|