LISTSERV at the University of Georgia
Menubar Imagemap
Home Browse Manage Request Manuals Register
Previous (more recent) messageNext (less recent) messagePrevious (more recent) in topicNext (less recent) in topicPrevious (more recent) by same authorNext (less recent) by same authorPrevious page (September 2001, week 1)Back to main SAS-L pageJoin or leave SAS-L (or change settings)ReplyPost a new messageSearchProportional fontNon-proportional font
Date:         Tue, 4 Sep 2001 19:21:41 GMT
Reply-To:     Steve Gregorich <gregorich@PSG.UCSF.EDU>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         Steve Gregorich <gregorich@PSG.UCSF.EDU>
Organization: UCSF/Medicine
Subject:      Re: Factor analysis and dichotomous variables
Content-Type: Text/Plain; charset=US-ASCII

Gijs, IMHO you should largely ignore Kim and Meuller's discussion of factor analysis of dichotomous manifest variables in factor analysis. The world has changed considerably since their book was published. What may have once been acceptable practice is no longer. In general, I direct you to publications of Bengt Muthen who has pioneered defensible approaches in this area.

Also, I read the abstract of the Haley article found at the link below. If the abstract is a guide, the article would not seem to offer much. Haley et al liken a factor analysis performed on phi coefficients to a linear dummy-variable regression. That is incorrect. The model is better described as a set of linear regression models, each with a binary outcome.

There are many well-known problems with Pearson correlations computed from binary data (i.e., phi coefficients). You should avoid using them. Binary items are defensible for factor analysis if you can first defend estimation of inter-item correlations via tetrachoric correlations. Again, Bengt Muthen's research is the main source of information about related methods.

Steve Gregorich

> Gijs, In addition to the references provided by Laura Copeland, you may >ant to check the following websites: >APPENDIX III-C : Discussion of Factor Analysis with Dichotomous >Variables ,Haley RW, Kurt TL. Self-reported Exposure to Neurotoxic >Chemical Combinations in the Gulf War: A Cross-sectional Epidemiologic >Study. JAMA 1997; 277:231-237 >http://www.swmed.edu/home_pages/epidemi/gws/app_3c.htm

>Course notes from Quantitative Research in Public Administration, G. >David Garson, Instructor, North Carolina State University E-mail >David_Garson@ncsu.edu; Tel. 919-515-3067; Fax 919-515-7333 >http://www2.chass.ncsu.edu/garson/pa765/factor.htm >Note that Garson also references the Kim and Mueller text cited by Laura >Copeland. It should be noted that S. C. Rowat in a study, Chemical >Susceptibility, Injury, And Reactivity: Mechanisms and Measurement, >identifies 19 Factor Analysis Study Criteria and Their Compliance by 4 >Studies in the Literature. One of these criteria is "Avoid dichotomous >(yes/no) variables unless addressed with special techniques." References >cited by Rowat include: > Comrey AL. Common methodological problems in factor analytic studies. J >consult clin psychol. 46:648-659 (1978). and >Streiner DL. Figuring out factors: the use and misuse of factor analysis. >Can J Psychiatry. >39(3):135-40 (1994).. >The URL for the Rowat study is: >http://www.rowatworks.com/Science/CSIR/C-SIR_Table-2.html#Anchor-Table-14210 >Dick March >South Florida Water Management District


Back to: Top of message | Previous page | Main SAS-L page