Hello,

Answering straight away to your question. Yes, it is possible. But you have to take care several things first. What is your research question? Your sample size? Why do you want to reduce to three factors those ten items?

If you have answered all of these, one thing to make into account is that if you want to separate 10 items into 3 factors it is highly possible that you will end up with a factor that has two items. This is problematic because factors with two items tend to be very unstable. It is preferable have factors with 3 items or more.

I recommend reading this paper: Costello & Osborne, (2005), which gives some good advice on what to take into account when performing EFA. 

http://pareonline.net/pdf/v10n7.pdf

Regards, 

In the literature, EFA is used for systematic data reduction where numerous factors emerge in research where there is multiple choices or simply yes/no questions. In other words, EFA reduces the dimensionality of the original space and to give an interpretation to the new space, spanned by a reduced number of new dimensions which are supposed to underlie the old ones. Theoretically, EFA uncovers the interconnectedness and correlation among factors in a matrix.

First of all it is not clear what do you mean by multiple choice items. If you are capturing your participants' responses on these items with Likert type scale, then conducting EFA won't be a problem. It is also not clear why you want to classify the items into three distinct groups. If you have theoretical basis for three groups then you can directly conduct CFA and find out whether the items result in three distinct factors or not by assessing  fit statistics and standardized loadings of the items on intended factor. 

I think for multiple choice items and based on the purpose of your classification the only best option for you is cluster analysis. Unless there is a strong theoretical evidence for the three distinctive groups, then allow the data to speak for  itself. Although you can still restrict yourself to the three classes depending on your research question. Several programs perform cluster analysis including spss. There are some specifically design for that but expensive (e.g., latent Gold). Mplus is also an excellent program for latent class analysis. You can also see this master thesis https://helda.helsinki.fi/handle/10138/144279 as well as

Pastor, D. A., Barron, K. E., Miller, B., & Davis, S. L. (2007). A latent profile analysis of college students’ achievement goal orientation. Contemporary Educational Psychology, 32(1), 8-47.

for a guide and other references.

For Arjun Shrestha's question, it was my impression that Usman Ladan has a "Pick from the following" type question, with 10 choices.  That gives him ten items that are coded either 1 (the person selected that one) or 0 (the person did not select it).  Leaving aside the thorny "Can you perform factor analysis on dichotomous items" problem, there are two very different answers to this question depending on whether the participants could only pick one from the list or could pick as many as they wished. 

   If participants could only pick one of the ten, then there is a zero-sum problem.  Picking one item means ALL the other items will not be picked.  Factor analysis cannot function under this situation, as there is no shared variance.  Your best approach would be a simple classification through percents.  I am not sure if cluster analysis would give you much more information than just looking at the percents.

   If participants could pick as many as they like, then factor analysis could give you information on which items participants selected in common, hence those items could be argued as "going together."  However, using FA to do this task invites criticism from all the "Thou shalt not conduct factor analysis on non-ordinal/non-interval data" crowd, so use at your peril.

Many thanks for your answers-appreciated. They really helped in clarifying many issues relating to my query. Julia Smith has got it right. The question contains 10 response categories numbered 1-10 for the participants to pick one. Initially I intend to conduct EFA at the first and CFA at the second stage of the analyses. Since EFA is not a feasible option and coupled with the previous research that has supported the three distinct groups I would now directly conduct CFA. However, I need some literature for guidance on how to conduct CFA using AMOS or LISREL since SPSS can't  perform such analysis.

Hi Usman,

In my opinion you can perform EFA with SPSS : Analyse Dimension Redution factor Analysis variables (treated as quantitatives) Extract Fixed number of factors 3.

But you can make a better analyziz through SEM.

Have a nice time

Helena 

Please refer using LISREL For Structural Equation Modeling by E. Kevin Kelloway, Sage  Publications.

I think it is possible, but you should consider  the sample size, the reliability of each factor items.

Can exploratory factor analysis be performed on a multiple choice question?

You can use SPSS base-module to perform EFA as shown in the following YouTube link.  It is a main link - from there you can select the relevant video clips.

https://www.youtube.com/results?search_query=how+to+perform+EFA+in+SPSS

Hi

1. Yes, it is possible.

In order to develop the analysis, please refer to:

Landau, S. and Everitt, B. S.(2004). A Handbook of Statistical analyses using SPSS.

Howitt, D. and Cramer, D.(2008). Introduction to SPSS.

Regards,

Zuhair