CFA error message

WWU

Hello,

when conducting the confirmatory factor analysis in JASP I get the following error message: "The model is not admissible: lavaan WARNING: covariance matrix of latent variables is not positive definite; use lavInspect(fit, "cov.lv") to investigate."

What can I do about the problem?

Thank you!

EJ

Dear WWU,

This warning suggests that the model is overparameterized. I would try a simpler model first, and see whether that works...

Cheers,

E.J.

WWU

Dear EJ,

thanks for your fast reply. I actually cannot use another model. Short explanation: I have made a survey that contains 2 scenarios that were randomly assigned to the participants. So I split the data set between the two scenarios. For the first scenario I don't get an error message, but for the second scenario I do, even though both used the same model.

Is there no other option than using a simpler model? I think it's possible to use the "structural equation modelling" for conducting the CFA, isn't it?

Best regards

WWU

EJ

Hi WWU,

Ah, I see. Strange. Well we do run SEM to do the CFA but I'll pass this on to Erik-Jan (with a k) who might have bright ideas.

Cheers,

EJ

WWU

Hey EJ,

thanks a lot! I hope we can find a solution.

Best regards

Lea

evankesteren

Hi Lea,

unfortunately, I cannot solve the problem without more information. Could you send me an email at ej.vankesteren@jasp-stats.org with a more detailed description of your model and your data? In general, this issue could be due to any number of problems, such as small sample size or assumption violations (e.g., the residuals are not normally distributed).

Erik-Jan

majan

Hi everyone,

I apologize for hijacking this thread, but on the topic of CFA.. please help :)

I have a data sample and want to perform factor analysis. First I want to use randomly selected 50% of the sample for a principal component analysis, and then the other 50% for a confirmatory analysis, but I'm not sure how I can split the sample and then use the other half later on. Could you please help?

Thanks!

Kind regards

EJ

You could generate a column with random binary numbers. Then you could use that to filter one subset or the other.

E.J.