CFA: JASP doesn't show p-values and standard errors of factor loadings

leonhard

Hey there!

I did a CFA analysis that showed good model fit. Than I wanted to see, if there is a second-order factor. After running this analysis model fit was stil good. But I can't see any standard errors or p-values for the factor loadings (only the estimates). I am wondering, what that means. In another forum I read that statistic programms don't show p-values and standard errors, if assumptions are violated. This could be the case here (because the Mardias Test that I used for the data im my efa was siginificant).

So

1. Why can't I see the standard errors and p-values of my factor loadings?
2. If it is because I assumptions are not met, do I have to ignore the model, dispite showing good fit?

Thank you!

Leonhard

patc3

No it's not because the assumptions are not met. Can you show a screenshot of your analysis and your output?

patc3

OK I could reproduce your situation with a dataset that ships with JASP in the data library (14- SEM: Grade point average):

I did a little bit of digging by running the same analysis directly in R:

setwd("c:/Program Files (x86)/JASP/resources/Data Sets/Data Library/14. SEM/")
df<-read.csv("Grand Point Average.csv")
library(lavaan)

"
f1 =~ NA*gpa1+gpa2+gpa3
f2 =~ NA*gpa4+gpa5+gpa6
f =~ NA*f1+f2
f1~~1*f1
f2~~1*f2
f~~1*f
" |> cfa(df)

What JASP doesn't show you is this warning:

Warning message:
In lav_model_vcov(lavmodel = lavmodel, lavsamplestats = lavsamplestats,  :
  lavaan WARNING:
    Could not compute standard errors! The information matrix could
    not be inverted. This may be a symptom that the model is not
    identified.

JASP doesn't show this warning (rightfully or not) because it's "just" a warning, not an error. And lavaan says the model converged:

> analysis@optim$converged
[1] TRUE

lavaan suggests the model may not be identified (i.e. trying to estimate too many parameters for the number of variables included). Basically your model might be too complex for the number of variables--but I did try to add more items in each factor, and was still getting this warning (and no test for the factor loadings).

Looking at my factor loadings, the correlations between my items seem to be quite low. Also I was getting another warning saying I had negative latent variances, so perhaps the scores don't vary much across people. These are the kind of things you can look for in your own analysis (particularly the size of the factor loadings). Also you need to make sure you have enough items in each factor (like 3 at a bare minimum).