cancel
Showing results for 
Show  only  | Search instead for 
Did you mean: 
Choose Language Hide Translation Bar
bsam12
Level I

Starting values for Structural Equation Models

Hello,

 

I am really enjoying the JMP Pro 16 SEM Platform - thank you to the development team for all the hard work and effort you have put into it, and the community for raising excellent suggestions for additions (e.g., "Assess Measurement Model").

 

Is there a way to specify starting values for parameters in the SEM platform? I am trying to implement the Comprehensive CFA Marker Technique suggested by Williams et al. 2010 to assess common methods variance (see:https://journals.sagepub.com/doi/abs/10.1177/1094428110366036). It seems starting values would be helpful as JMP seems to be having trouble (in some cases) estimating some model parameters once the additional latent marker variable and items are added to the model.

 

Also, if there is an easy way to automate the output suggested by Williams et al. 2010 in a future release of JMP Pro, that would be amazing!

 

Thank you!

1 REPLY 1
LauraCS
Staff

Re: Starting values for Structural Equation Models

@bsam12,

Great to hear you're enjoying the SEM platform! Currently, users aren't able to specify starting values as we've invested lots of effort in algorithms that, we hope, eliminate that burden from users. With that said, there could very well be data characteristics that lead to non-convergence and perhaps enabling user-specified starting values could help. I'll make note of your request (thank you!).

 

With regard to the specific model(s) you're fitting, I see that Williams et al. (2010) suggest a general model with some potential restrictions (e.g., omitting the covariance of the marker latent variable with the substantive latent variables, fixing the loadings of the marker latent variable prior to linking it with the substantive indicators, etc.). I'm curious if there are enough restrictions in the models that are giving you trouble to ensure identification of the model. Indeed, Figure 1 in the article is very similar to a bifactor model and, without important restrictions, it's possible the model is simply not identified. Would you be able to share your data and scripts with our tech support team? You can send them to support@jmp.com and we'd be happy to take a look at the specifics.

 

Best,

~Laura

Laura C-S