Nonnormality in SEM

#1
Hello, I have a data set consisting of 700 participants with 90 items. Multivariate normality seems to be violated. There are three steps offered to handle non normality. One of them is using WLS but requires higher samples. second one is to turn the scores into normal scores and get asymptotic covariance and follow, but this also does not work. The output shows actually the same wording as what I wrote into syntax. That is, LISREL does not calculate the result. The third one is using item parceling which is not so much offered by statisticians?

What do you suggest me to do?
 

CB

Super Moderator
#2
When in this situation in the past I've used fit statistics that are scaled for non-normality (e.g., Satorra-Bentler scaled "robust" fit statistics), and used bootstrapping to obtain confidence intervals for parameter estimates to deal with the potential impacts of non-normality on confidence interval coverage.
 
#3
Hello, thank you for your reply. May I ask you one more question? I wanted to get Satorra Bentler Chi square with calculating asymtotic covariance matrix. I can construct the acm file from output options and write syntax accordingly. But It works for nearly 40 items, that is for the whole model, it does nor converge the model, instead in the output, it just gives what I wrote in the syntax. Dis you have any problem related to running acm file with this kind of sample size and item number?
 

CB

Super Moderator
#4
I am not sure what the problem might be. Presumably the output file provides some error/warning messages? Are you sure your model is identified?