Cumming, J. A. & Wooff, D. A. 2007. Dimension reduction via principal variables. Computational Statistics & Data Analysis 52(1): 550-565. Extends from GP McCabe's 1984 paper. Also check out the references therein. David On 27/10/2010 13:23, K F Pearce wrote: > Hello all, > > I'd appreciate it if anyone has any references on the following or can offer their views: > > I am very familiar with using PCA in the usual way i.e. using a set of p correlated variables we generate a set of p uncorrelated PCs where each PC is a linear combination of these variables....we can choose to retain the first m PCs....then, say, plot these PCs' scores against each other. > > Now my question is....I have discussed, informally, in the past that PCs can be used also to reduce a number of variables. Say we had p potential correlated variables....we could conduct a PCA on these variables....and then, from each of the subsequent *important* PCs, choose those variables with the largest (positive or negative) coefficients.....these variables are taken as being the 'important ones' and can be used in further analysis. > > Does this seem OK? > > If so, can this technique be used when the p variables are (i) potential independent variables or (ii) potential response variables. > > Many thanks for your views on this, > Kind Regards, > Kim > > You may leave the list at any time by sending the command > > SIGNOFF allstat > > to [log in to unmask], leaving the subject line blank. -- David Wooff, Director, Statistics and Mathematics Consultancy Unit, & Senior Lecturer in Statistics, University of Durham. Department of Mathematical Sciences, Science Laboratories, South Road, Durham, DH1 3LE, UK. email: [log in to unmask] Tel. 0191 334 3121, Fax 0191 334 3051. Web: http://maths.dur.ac.uk/stats/people/daw/daw.html You may leave the list at any time by sending the command SIGNOFF allstat to [log in to unmask], leaving the subject line blank.