Adds option to perform EFA with ordinal variables and two other options (Mardia's tests; Parallel Analysis detailed output)

[francotisocco]

• Adds option to perform EFA based on the polychoric-tetrachoric correlation matrix, so as to allow for EFA with ordinal variables.
• If EFA is performed based on said correlation matrix, the KMO values, Bartlett's test and Parallel Analysis are computed/performed based on this matrix as opposed to the standard correlation matrix.
• Adds Mardia's tests of multivariate normality
• Adds a detailed output of the parallel analysis results.
• fixes Polychoric/tetrachoric factor analysis jasp-issues#903
• fixes [Feature Request]: Ordinal variables should be allowed in EFA jasp-issues#1624
• fixes Implement new unit tests for EFA, some small changes #96

.....

Hi! I modified the jaspFactor module to fit my analysis needs and then thought it could also be of use to others. I am not heavily experienced with Github so I am not sure whether I am conducting the pull request correctly - if anything, let me know!

Moreover, apologies in advance for posting so close to New Years.

[juliuspfadt]

Hi @francotisocco thank you very much for the PR and providing the fruits of your work for others. I apologise for the delay in looking into your PR :)
I started going over this and I am having trouble getting the poly matrix to run with a simple data set.
example_asrm.jasp.zip
It seems the same data set also does not work in R with the psych package. For example psych::polychoric gives the error JASP also displays. Do you have any idea why that happens?

[juliuspfadt]

Thanks again. This already looks really good. Some minor changes are in order. Also, do you feel comfortable making some unit-tests for the added features? Otherwise I can also do that.

[francotisocco]

Hi @francotisocco thank you very much for the PR and providing the fruits of your work for others. I apologise for the delay in looking into your PR :) I started going over this and I am having trouble getting the poly matrix to run with a simple data set. example_asrm.jasp.zip It seems the same data set also does not work in R with the psych package. For example psych::polychoric gives the error JASP also displays. Do you have any idea why that happens?

Hi Julius! Don't apologize at all, since I know the PR introduces many little things and nuances - It is probably me who should apologize :). Thank you deeply for your extensive reviews - I'll be responding to each suggestion within the next days.

Regarding what you mentioned about the dataset, I'll look into it around these days. It could be convergence issues (because it may happen with trying to estimate the poly matrix with few observations), although I wouldn't want to guess without taking a deeper look.

EDIT - UPDATE: I've been looking into it and it seems that the [example_asrm.jasp.zip] dataset has a combination that psych "does not like" while computing the poly mat - either via psych::polychoric or psych::mixedCor: a relatively "small" sample size + some response alternatives not present in certain variables while present in others (i.e., zero-cell entries when computing the matrix).

When it tries to correct for the presence of zero-cell entries in the presence of a "small sample size", the function apparently fails for "yet undetermined reasons" (this last phrase is a verbatim quote from the Psych documentation).

Further inspecting, it appears that the variable asrm_3 is missing the response category "5". If you remove it from the EFA query, the analysis runs properly - of course this is not 100% desirable since it's perfectly possible to have a dataset with missing response categories within variables "in real life".

Facing this, I could:
a) Include a clarification within the help file similar to what I wrote above.

b) Work on including a correction within the rscript so that if the function resolves in an error, the computation runs again without attempting to correct for continuity.

[juliuspfadt]

I think it would be fine to just display a more informative error message, no need to fix the function from psych. This message could just read: "Unfortunately, the estimation of the polychoric/tetrachoric correlation matrix failed. This might be due to a small sample size or variables not containing all response categories."
Obviously, adding some info to the help file is also good.

[juliuspfadt]

fixes jasp-stats/jasp-issues#1624

[juliuspfadt]

Hi @francotisocco it has been a while, and I just wanted to check in if there is anything I can still help you with for this PR?

[francotisocco]

Hi @JuliusPf! Sorry for the delay - i've had a surge of work and I'll probably be able to get back to these changes around April. I apologise! Regarding the unit-tests, I'm afraid I wouldn't know how to proceed to perform them. Of course, feel free to implement changes if you find a suitable moment; otherwise in a few weeks I'll get hands-on. Again, deep apologies for the delay and thank you for your work and follow-up!

PS: excellent you could also implement this within PCA :)

[juliuspfadt]

No need to apologise, you are doing us a favour! I can add the unit tests once you are done with the PR, no worries.

[francotisocco]

Apologies for the delay! I believe I've addressed all comments/reviews - by the way, thank you again for providing them since it helps me understand more about how the program works conjointly with R. If there were any further necessary modifications, don't hesitate in pointing them out.

[vandenman]

@francotisocco thanks for contributing!

[francotisocco]

@francotisocco thanks for contributing!

On the contrary! Thank you for working with what I could put together. A pleasure.