EFA for Dichotomous Variables

[E. Bernardo]

Can we do exploratory factor analysis for dichotomous variables in SPSS?

I would like to hear your thoughts about this.

Thank you.

Eins

[Bruce Weaver]

I have not read these particular notes, but have found Phil Ender's notes on other topics to be good.  From a very quick glance, it looks like he discusses the approach of using a matrix of tetrachoric correlations as input for the factor analysis.

  http://www.philender.com/courses/multivariate/notes2/morefa.html

See also:

http://www-01.ibm.com/support/docview.wss?uid=swg21475247
http://www.ncbi.nlm.nih.gov/pubmed/22477439

HTH.

E. Bernardo wrote

Can we do exploratory factor analysis for dichotomous variables in SPSS?
I would like to hear your thoughts about this.

Thank you.
Eins

[Kirill Orlov]

Eins,
Two most well-known factor-analytic approaches to truly dichotomous variables are
1) Factor analysis of tetrachoric correlations.
2) Latent trait analysis aka IRT (Item response theory) analysis.
(1) could be called factor analysis via probit link, and (2) could be called factor analysis via logit link.
SPSS Statistics has no in-built commands to do either, unfortunately. But there likely exist macros or extensions written by users.

11.02.2014 11:52, E. Bernardo пишет:

Can we do exploratory factor analysis for dichotomous variables in SPSS?

I would like to hear your thoughts about this.

Thank you.

Eins

[Ryan]

It is certainly possible to fit the standard dichotomous Rasch and 1PL IRT models in SPSS without the use of extensions. In fact, one could fit a multilevel 1PL IRT model in SPSS if one so desired.

Ryan

On Tue, Feb 11, 2014 at 7:47 AM, Kirill Orlov <[hidden email]> wrote:

Eins,
Two most well-known factor-analytic approaches to truly dichotomous variables are
1) Factor analysis of tetrachoric correlations.
2) Latent trait analysis aka IRT (Item response theory) analysis.
(1) could be called factor analysis via probit link, and (2) could be called factor analysis via logit link.
SPSS Statistics has no in-built commands to do either, unfortunately. But there likely exist macros or extensions written by users.

11.02.2014 11:52, E. Bernardo пишет:

Can we do exploratory factor analysis for dichotomous variables in SPSS?

I would like to hear your thoughts about this.

Thank you.

Eins

[E. Bernardo]

To all those who shared their thoughts about EFA for Dichotomous variables, thank you.

Eins

11.02.2014 11:52, E. Bernardo пишет:

Can we do exploratory factor analysis for dichotomous variables in SPSS?

I would like to hear your thoughts about this.

Thank you.

Eins

[Art Kendall]

What is the goal of the exploration?

Are you creating summative scale scores?
Are you just checking the scoring key of pre-existing scales?

Would you be using any scores in individual assessment or would you be using the scores in some analysis?

If I recall correctly there is are python or R way to compute tetrachoric correlations and python or R ways to do IRT. (the procedures suggested by Kiril.)

Also use CATPCA.  (however without access to the documentation at this time I have the impression that CATPCA deals only with the total variance of the items rather than the common variance

Finally, just run FACTOR.
compare and contrast the scoring keys suggested by each method. 
If there are not meaningful differences in the solutions report the results from the PAF and mention that the other methods were tried and gave similar results.


On 2/11/2014 3:00 AM, E. Bernardo [via SPSSX Discussion] wrote:

Can we do exploratory factor analysis for dichotomous variables in SPSS?

I would like to hear your thoughts about this.

Thank you.

Eins

---

If you reply to this email, your message will be added to the discussion below:

http://spssx-discussion.1045642.n5.nabble.com/EFA-for-Dichotomous-Variables-tp5724439.html

To start a new topic under SPSSX Discussion, email [hidden email]
To unsubscribe from SPSSX Discussion, click here.
NAML

[Kirill Orlov]

Art, with dichotomous data CATPCA will give the same results as will PCA by FACTOR. There is no way to quantify a *dichotomous* scale. Except for quantifying it the multiple nominal way, which will turn the analysis into multiple correspondence analysis, not factor analysis.

You are right saying that CATPCA is PCA, not common EFA. But you always can do quantification by PCA under the m-components goal, and then use FACTOR to do real factor analysis with these transformed data. My own experience tells me that this is very efficient: preliminary nonlinear quantification of the scale often makes final factors more interpretable.

[Art Kendall]

good point.
On 3/3/2014 12:14 PM, Kirill Orlov wrote:

Art, with dichotomous data CATPCA will give the same results as will PCA by FACTOR. There is no way to quantify a *dichotomous* scale. Except for quantifying it the multiple nominal way, which will turn the analysis into multiple correspondence analysis, not factor analysis.

You are right saying that CATPCA is PCA, not common EFA. But you always can do quantification by PCA under the m-components goal, and then use FACTOR to do real factor analysis with these transformed data. My own experience tells me that this is very efficient: preliminary nonlinear quantification of the scale often makes final factors more interpretable.

===================== To manage your subscription to SPSSX-L, send a message to [hidden email] (not to SPSSX-L), with no body text except the command. To leave the list, send the command SIGNOFF SPSSX-L For a list of commands to manage subscriptions, send the command INFO REFCARD