jamovi

We actually already provide a oblique rotation by default (oblimin) and also calculate the number of factors through parallel analysis by default. So I guess we are doing a good job

Ravi wrote:Ok, I now added 'maximum likelihood' and 'principal axis factoring' as possible extraction methods (next to the 'minimum residual' method). Should be available in the next version. Is there another extraction method that you think should be added?

Thats really, really cool !! I tried the new version 0.9.5.8. and was really thrilled to see principal axis factoring - as option to choose. So i tried it - but still get Initial Eigenvalues below Zero - in Spss the values of the Eigenvalues add up to the number of factors (items) you analyse - so either (and thats quite possible) there is something I dont understand (and theres lots of that) or maybe there is a bug in the analysis provided. Maybe you could check this.

Reason I write is - we would really love to tell students - use jamovi instead of spss - but when we can not reproduce (or understand) the results, which are different from what we teach students, we will have to wait - when it comes to prinicipal axis factoring - and it would be so cool - if this works out, cause spss does not do parallel analyses (except if you bring in a few sheets of syntax code) - and jamovi would be a really good alternativ for students.

Hi Paul,

maybe SPSS uses the principal components analysis eigenvalues to conduct the parallel analysis? I double checked it in R using the psych package and it's also giving me negative eigenvalues there. Also, here's an example of STATA producing negative eigenvalues: https://stats.idre.ucla.edu/stata/faq/h ... -in-stata/. So I assume there's not an issue with negative eigenvalues, but I am curious to know about the discrepancies between SPSS and jamovi/R/STATA.

Ravi

This paper explains why in EFA you can have negative eigenvalues, while in PCA you can't: https://pdxscholar.library.pdx.edu/cgi/ ... health_fac

we would really love to tell students - use jamovi instead of spss - but when we can not reproduce (or understand) the results

if you can relax the assumption that spss is correct, or "most correct", or "the gold standard", you'll probably have an easier time of this. but i appreciate that can be challenging (especially with colleagues).

have you heard the story about the lavaan sem package? people were always complaining that it didn't provide *exactly* the same results as the commercial packages, so yves rosseel ended up reverse engineering the other packages, figuring out *exactly* what they did (which wasn't quite correct), and then adding options to mimic the (not quite correct) output from other packages.



cheers

jonathon

Hi all
Will you implement some estimators for orderer categorical variables (polychoric correlation matrices)? Maybe I'm asking too much, but a large part (if not the most) of social science students mainly use EFA with ordinal data (Likert scales).
Congrats for the project.