I'm not sure whether factor analysis will work with binary variables. A more appropriate technique might be Cochran's Q. This will look for differences in the scores of binary variables. It might enable you to establish a partial ordering on your questions rather than a factor grouping.
Mehmet is correct. You'll probably need to forgo SPSS and use either STATA or R. I'm not as familiar with R, but in STATA you compute the pairwise tetrachoric correlations and that pairwise matrix can then be used to conduct factor analysis or PCA, whichever suits your purpose best.
One of the assumptions in factor analysis is to have items/variales measure in a continous scale {interval or ratio}. However, all the available EFA were run on orinal scales {Likert type}. in exceptional cases, binary variables {dichotomous} can be handled as a special case.
The current "fashion" is to use tetrachoric instead of person's correlations when you have binary data, but even a brief internet search will show that each have limitations, and that they can give different results. If possible, I would recommend running both and comparing the results.
Also if you can produce a matrix of tetrachoric correlations in SPSS (I think you might need a "macro" to do that), then you could use that matrix as the input to the Factor Analysis command (i.e., it does not require raw data).
You can still working on SPSS if you want. You just need to use the TETRA-COM Macro for SPSS. This Macro computes a tetrachoric correlations matrix and then you are able to performed an EFA via SPSS FACTOR module using ULS extraction method.
I have a follow-up question. I have coded about 6000 scenes from different television shows for the presence/absence of ten different humor types. Now I want to see in what way these humor types co-occur (within a scene). I have read the article by Lorenzo-Seva and Ferrando (2012) about tetrachoric correlation, but the problem is that tetrachoric correlation is specifically intended for dichotomies that are obtained by dichotomizing truly continuous variables (e.g., a scale scored as agree-disagree, which is a continuum in reality), whereas I have truly dichotomous scores. I am a bit stuck on what sort of analysis I should use. Do you have any suggestions? Thank you in advance!
I think you can apply an EFA on the tetrachoric-correlations matrix (using the software "FACTOR"). You could also explore the "Gifi model" options (e.g., HOMALS) for optimal scaling (available in R o SPSS).
I calculated the tetrachoric correlations matrix using the MACRO provided by Lorenzo-Seva and Ferrando (2012). However, I noticed that all negative correlations are not significant, despite some of them being rather high, and that all positive correlations are significant, despite some of them being close to zero. So, something is wrong, anybody got an idea what went wrong? Thank you in advance.
Dear Amber van der Wal , if you want, send me a partial copy (few cases and variables) of your raw data to make a fast check data codes and correlations to [email protected]. But let me recommend you to use MPLUS (in my opinion: the best to make all kind of factor analysis) and specify variables as ordinal
Dear Rodrigo Ferrer, thank you so much for your offer! In the meantime, I have downloaded FACTOR and using that program, I get the same correlations but with the right significance levels this time. I will double check with MPLUS. Just as a warning to people planning to use the SPSS macro to calculate tetrachoric correlations, have a close look at the significance levels, as they may be incorrect.