Factor analysis with dummy variables

[Alina Sheyman-3]

I'm trying to figure out if it makes sense to use factor analysis when
dealing exclusively with dummy variables. I'm worried that since I'll
be randomly assigning numeric values to such parameters as gender,
race, etc any sort of a correlation I find behind the variables will
be pretty meaningless. How correct is my logic?


>>> Error in line 4 of spssx-l.mailtpl: unknown formatting command <<<
 -> .................... <-

[Hector Maletta]

         Alina,
         Dichotomous variables can be regarded as interval-level variables
(because they have only one interval, which can be used as their unit of
measurement without loss of generality), and therefore their correlation
coefficients do make sense. They are in fact the same as the phi coefficient
of association.
         However, for factor analysis it is better to use categorical
principal component analysis (CATPCA in the Categories module of SPSS),
especially if you have multi-category variables which you have converted
into dummies. CATPCA accepts interval, ordinal or nominal scales without
restriction. In the case of ordinal or nominal variables, it also assigns
numerical values to the various categories.
         If you do not have access to the Categories module, and as an
imperfect solution to your plight, it is also possible (though not
recommendable) using the FACTOR command (in the Base module) with a set of
binary variables. In this case, any multi-category variable should be
converted into a set of dummies (leaving one category aside to avoid
redundancy).

         Hector

         -----Original Message-----
From: SPSSX(r) Discussion [mailto:[hidden email]] On Behalf Of
Alina Sheyman
Sent: 15 October 2007 12:52
To: [hidden email]
Subject: Factor analysis with dummy variables

         I'm trying to figure out if it makes sense to use factor analysis
when
         dealing exclusively with dummy variables. I'm worried that since
I'll
         be randomly assigning numeric values to such parameters as gender,
         race, etc any sort of a correlation I find behind the variables
will
         be pretty meaningless. How correct is my logic?


         >>> Error in line 4 of spssx-l.mailtpl: unknown formatting command
<<<
          -> .................... <-


>>> Error in line 4 of spssx-l.mailtpl: unknown formatting command <<<
 -> .................... <-

[David Greenberg]

Correlations do not depend on the scale on which a variable is measured. However, it is problematic to use most factor analysis routines to estimate models in which the observed variables are dichotomous, because the models assume that the distributions are continuous. There are, however, programs like LATENT GOLD that allow you to specify what type of variables you have. David Greenberg, Sociology Department, New York University

----- Original Message -----
From: Alina Sheyman <[hidden email]>
Date: Monday, October 15, 2007 12:04 pm
Subject: Factor analysis with dummy variables
To: [hidden email]

> I'm trying to figure out if it makes sense to use factor analysis when
> dealing exclusively with dummy variables. I'm worried that since I'll
> be randomly assigning numeric values to such parameters as gender,
> race, etc any sort of a correlation I find behind the variables will
> be pretty meaningless. How correct is my logic?
>
>
> >>> Error in line 4 of spssx-l.mailtpl: unknown formatting command <<<
>  -> .................... <-
>

>>> Error in line 4 of spssx-l.mailtpl: unknown formatting command <<<
 -> .................... <-