In article <[EMAIL PROTECTED]>, Andrew Walsh <[EMAIL PROTECTED]> wrote: >hi all,
>i would like to examine the factor structure of a 20 item 'true/false' >scale i'm currently using. the authors have indicated that this measure >has a single factor solution, however my intuition says otherwise. >could you direct me to a few literary sources that specifically >addresses factor analysis with a dichotomous variable? >thanks in advance, >~aw I cannot give offhand references, but I can point out what to look for to some extent. The first question is, what is the MODEL? One of the approaches is to assume that the results are obtained from a multivariate normal distribution by using a threshhold for each variable. This would indicate that one should use the tetrachoric correlations. Assuming that this is true, the estimators would be consistent and asymptotically normal; however, this is extremely sensitive to deviations from normality, whereas factor analysis is not that sensitive, and with normalization on the factor loading matrix only, essentially insensitive. The other is what is called a latent variables model. This does not take the discrete results as discretization of continuous variables, and tries to explain them directly. The methods used are not related to the linear models used in factor analysis. -- This address is for information only. I do not claim that these views are those of the Statistics Department or of Purdue University. Herman Rubin, Department of Statistics, Purdue University [EMAIL PROTECTED] Phone: (765)494-6054 FAX: (765)494-0558 . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
