(Incorrect) Negative Eigenvalues in JASP EFA Scree plot?
Hi Folks, I can't find a discussion on this but apologies if I missed it in my search. I am new to JASP, and normally teach using SPSS, but we are making use of JASP as an alternative during lockdown teaching. We will likely continue to use it for some purposes going forward, and there is much to like. There are some nice features I want to benefit from in my teaching (though also some gaps...discussion for another day.)
In any case, I ran EFA in JASP as a parallel to a lab exercise I normally do in SPSS, based on Thurstone's classic box problem. I could figure out most of the JASP options, and the factor loadings in the pattern matrix are pretty close to what I normally get so that is encouraging. I expect some small differences due to differences in numerical methods.
However, the eigenvalues have me baffled. The JASP output does not include the classic table of eigenvalues and only prints the proportions of variance for the factors. However, in the scree plot we can see the eigenvalues roughly in the plot...and it seems that many of them fall into the negative range. This is not correct. Negative eigenvalues indicate that the matrix is not positive-definite and normally would crash the EFA. In any case I know from my own example data (constructed for the exercise) that all the eigenvalues are positive.
Since the analysis itself runs, and the loadings are consistent, I wonder if there is an issue in the way the parallel analysis is implemented or reported in the JASP output? The eigenvalue line should definitely not drop below 0.
Thoughts?
Ken.
Comments
Hi Ken,
Hmmm. I'll attend our EFA expert to this. Seems important, but I'm relatively sure there is a simple explanation. To get to the bottom of this, we might need a data file for which this happens. It is probably best to post the issue on our GitHub page (for details see https://jasp-stats.org/2018/03/29/request-feature-report-bug-jasp/) as this communication with the team much more effective.
On a different note, you can also use the GitHub page for feature requests -- we're happy to help out with functionality that you believe is important but still missing.
Cheers,
E.J.
Thanks E.J! I'll check out the instructions for raising the issue via GiHub as well, and glad you are open to feature requests also. As a teacher I very much appreciate the possibility of a dialog about what is useful for teaching, as well as just for the analyses themselves :)
Cheers,
Ken