🚀 go-pugleaf

Instead of running a classical factor analysis on a set of dichotomous data
(with the inherent problems arising from the linear model assumption) one
might consider a Boolean factor analysis.  Here the scores and loadings are
binary and a variable may have a loading of one for several factors.  Fit is
estimated by comparing observed binary responses with those estimated by
dot multiplying factor scores and loadings and making a tally of positive
and negative discrepancies.

In the BMDP package, program 8M performs Boolean factor analysis.  It works
on a set of dichotomous variables (or allows the user to treat polytomous
variables as dichotomous by specifying:  nonresponses = 1,2;  responses =
4,5).  It also handles missing data and allows the user to set values to
unknown.

I will send more information on the program, including formulae and
computational methods to anyone interested.  Just let me know your postal
address.
----------
Bill Sanders   BMDP Statistical Software, Inc.      Phone: 310-479-7799
1440 Sepulveda Bl, Los Angeles, CA 90025            Fax:   310-312-0161
email:  bmdp@vm.usc.edu  or  bmdp@uscvm.bitnet

===============================================================================

BMDP Statistical Software, Inc.          Telephone:(310) 479-7799
1440 Sepulveda Boulevard                 FAX:      (310) 312-0161
Suite 316                                Telex    4972934 BMDP UI
Los Angeles, CA 90025