Ray, Thanks for precising this..
-----Message d'origine----- De : [EMAIL PROTECTED] [mailto:[EMAIL PROTECTED] la part de Ray Koopman Envoy� : lundi 24 novembre 2003 09:16 � : [EMAIL PROTECTED] Objet : Re: [edstat] Factor analysis of count data [EMAIL PROTECTED] (Naji Nassar , MIReS) wrote in message news:<[EMAIL PROTECTED]>... > I'ld suggest a modelling approach. Suppose all our count data are > related to one Factor. > You can suppose that each observation (X1i to X8i) follow a Poisson > distribution (mean Lambda1i to Lambda8i) > The lambda are a liunear combination from the supposed Factor Score > Lambda1i to Lambda8i=(Alpha1 to ALpha8)*FSi In models like this it's usually the log of lambda, not lambda itself, that's taken to be the product of the factor scores and factor loadings. . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . ================================================================= . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
