SEM model group analysis not concave

[George Yiallouros]

Hello there !

I am trying to use Stata and check my SEM model for measurement in-variance among two different groups. The issue that i am facing is that my model doesn't concave when i apply NO constrains to it. However as soon as i apply constrains to the measurement intercepts and thereafter to other parameters (e.g. measurement coefficients, structural coefficients) the model concaves. Essentially I would like to assess the model without any constrains and then check for goodness of fit when applying the constrains, however I can't get it to run..What is your idea about that ? I sense that the variability in the measurement intercepts is not "helping" stata to run the model !??


Thanks in advance for your response.
George

[Phil Bromiley]

You didn't get a quick response. You'll increase your chances of a useful answer by following the FAQ on asking questions - provide Stata code in code delimiters, readable Stata output and sample data using dataex. Without the code, we're just guessing what you really did. Furthermore, being able to replicate a problem is often essential to solving it, particularly if the problem reflects an interaction of your model and the data (as not concave problems probably do).

I would have thought that normally adding constraints helps models estimate. There are some options for maximizing that you might try.

[Christopher Bratt]

Apart from supporting the recommendation by Phil Bromley, let me add this: a SEM model (meaning a model with latent variables, not simply path analysis) will never converge if you do not add constraints. It will not be identified.

One indicator per factor commonly needs to have its loading fixed so the factor has a scale (assuming you have at least three indicators for the latent). For invariance testing it will be preferable to standardise the factor instead (variance = 1), so all factor loadings can be tested for equality across groups, in addition to indicator intercepts or thresholds.

If this answer solved your problem, then I would strongly recommend reading about SEM in general and tests of measurement invariance specifically before applying any of it. There are frequent misconceptions/limited understanding of SEM, even in recent literature published by StataCorp that discusses/presents SEM.