Post by SFCPost by RyanPost by SFCThanks for the link.
However, I would like to fit models similar to those in item response
theory, with estimates of thresholds and slopes (for two-parameter
logistic models). I need to do factor analysis instead of principal
component analysis. Sorry for not making this clear.
--
Shu Fai
fromhttp://listserv.uga.edu/cgi-bin/wa?A2=ind0804&L=spssx-l&P=R26044> When all variables are dichotomous, the result of CATPCA is equal to the result of standard PCA.
(A line fitted on 2 points will always be a straight line, so the transformations are linear, no matter the scaling level.)
Anita van der Kooij
Data Theory Group
Leiden University
so in SPSS one can use FACTOR of CATPCA they will give the same results.
It is conventional to factor dichotomous data for creation of summative
scales.
Art Kendall
Social Research Consultants
In SPSS, is there any simple way to factor analyze binary data from
true/false or yes/no items? Preferably without using R.
I have other options to factor analyze binary data myself (e.g.,
Mplus, LISREL, R). However, I am interested to know whether this can
be done in SPSS. Some of my friends do not have access to those
options.
--
Shu Fai- Hide quoted text -
- Show quoted text -
I suggest you consider using another statistical software program to
fit item response theory (IRT) models. You can use specialized
software programs such as Winsteps (for Rasch models) or PARSCALE (for
various IRT models). It is also possible, though a bit more difficult,
to fit such models in other general software programs such as SAS.
Ryan
Thanks. I use BILOG-MG and Mplus for the analysis. However, I am
preparing something for others who have access only to SPSS, to do a
factor analysis on a specific scale with dichotomous responses. All I
need is simply the thresholds and slopes, which will be further
manipulated by some syntax commands.
I thought about using the programmability extension in SPSS, such that
they can use R without actually knowing how to use R. However, this
requires them to install both the extension and R. I am not sure
whether it is easy for them, but this is a possible option.
If someone has already created a script or macro specifically for one
factor IRT model, it would be great.
--
Shu Fai- Hide quoted text -
- Show quoted text -
It seems like you are using the terms factor analysis and item
response theory interchangeably. At any rate, Art has explained that
there are several types of factor analysis in SPSS. If that's what you
want to do, then you have several options.
I have never tried to fit an IRT model in SPSS or via R. Thinking
aloud for the moment, if I only had access to SPSS [ignoring R] and I
was trying to fit a 1-parameter IRT model (i.e. Rasch model) for
dichotomous items, I would probably fit a generalized linear model.
First, I would set up the data set in vertical format as follows:
Person_ID Item_ID Response
1 1 0
1 2 1
1 3 1
2 1 1
2 2 0
2 3 1
3 1 0
3 2 0
3 3 1
.
.
.
N
Then I would use the following code to fit the model:
GENLIN Response (REFERENCE=LAST) BY Item_ID (ORDER=ASCENDING)
/DISTRIBUTION=BINOMIAL LINK=LOGIT
/MODEL Item_ID INTERCEPT=NO
/REPEATED SUBJECT=Person_ID SORT=YES CORRTYPE=EXCHANGEABLE
ADJUSTCORR=YES COVB=ROBUST
/PRINT CPS DESCRIPTIVES MODELINFO FIT SUMMARY SOLUTION.
This model can be thought of as a repeated measures binary logistic
regression model.
If I were using SAS, I would abandon this model in place of a
generalized linear mixed model, with a person random intercept
specification. That too has limitations given the usual normality
assumption of the random effect, but it is more in line with how I
understand Rasch models. In an ideal world, it would probably be best
to fit a Rasch model in Winsteps.
I realize this is not what you're asking for, but thought I'd share
my thoughts on how to fit a 1-parameter IRT model in SPSS. Also, there
might very well be a better approach to fit a 1-parameter IRT model in
SPSS, but this is my best guess at the moment.
Ryan