While attempting to validate a questionnaire, I have noticed that many authors (from different countries) use different softwares such as R Statistics or Mplus Statistical Package to perform exploratory factor analysis. Hence, will SPSS alone be sufficient in providing the analysis or do I need to use another software when dealing with dichotomous data?
Hey there Michael,
Thanks for your input. I have actually merged R Statistics to SPSS, but it is far too complicated for me. I guess I will stick to SPSS or maybe AMOS if needed.
Rabin
hi,
SPSS is suitable for doing EFA , however, AMOS will be usefull for CFA.
Regards
As others have mentioned, the software isn't terribly important. If the questionnaire contains items that you've developed, and you want to see how each of the items load onto latent factors, than EFA via SPSS will do the trick. EFA, however, is best used when you don't have a strong a priori theoretical framework. If you do, or if you're confirming items used somewhere else (like another paper) consider CFA.
You might also consider learning STATA since it can do all of the above analyses and is similar enough to SPSS that you should be able to pick it up easily. It can even do more complex analyses like SEM, HLM, etc.
If your data is dichotomous you should use WLSMV estimator which is available in Mplus. If you are attempting to validate questionnaire you should use CFA.
Yes, SPSS can be used for EFA quite successfully in many cases, however, SPSS Factor analysis procedure seems rather dated. I don't remember any essential change since version 8. There are many relatively new things in EFA, like GEOMIN or TARGET rotations, parallel analysis in determining number of factors, polychoric or tetrachoric correlation coefficients, which are not possible in SPSS directly. At least they require external SPSS macros or other additional software.
When you drive dichotomous data is best to work with polychoric correlation matrix (no available in SPSS). I recommend the free program you can download from: http://psico.fcep.urv.es/utilitats/factor/index.html
It is very easy to use and highly reliable.
If your data fit the Rasch model you can say that they are meaningfull.
If they do not fit the model they may be ..... rubbish
Dear guys,
Thank you so much for your input. Yes, I believe I need to use another software to validate my questionnaire.
Here's the latest update:
1) I have used STATA to perform a tetrachoric correlation of my data (I chose tetrachoric because it is similar to a previous study)
2) I managed to run it, but there is a warning about some parts of it being not positive-definite (hence converting it to positive semi definite)
3) I ran the factor analysis of the correlation matrix anyhow, but now I have this major problem of Heywood Case (at least it is not ultra Heywood)
What do I do?
Rabin
Guys, sorry for the inconvenience. I figured out why I had the problem above. I did not add the notable posdef command for the data analysis. It is all fine now. Thanks so much for your help.
- Badges
- Science topic
- Similar topics
- Indexes