Response Surface Analysis lavaan error when using binary outcome variable and specifying WLSMV estimator

[Dennis Jacobsen]

Dear all,

I am conducting a response surface analysis (RSA) with a binary outcome variable using the RSA package in R. In line with the recommendation in the RSA help file, I fit my binary outcome variable with a probit link function, where I define it as "ordered" and specify the lavaan estimator to "WLSMV". However, when doing so R produces the following error message:

Error in inherits(NVarCov, "try-error") :
  object 'NVarCov' not found
In addition: Error message:
In lav_model_vcov(lavmodel = lavmodel, lavsamplestats = lavsamplestats, :

  lavaan WARNING: unknown se type: robust

Unfortunately, I have been unable to find a way to solve this and using the RSA package is much more efficient than available alternatives. Therefore, I was hoping some of you would know what might be causing the error and could point me in the right direction toward a solution.

Please do let me know if you need any additional information. Thanks for your help!

Very best,
/Dennis Jacobsen (PhD student at Yale School of Management).

[Yves Rosseel]

Hello Dennis,

I would need more information. Are you using a current version of
lavaan? How did you specify your syntax and sem() call?

On 11/16/22 20:59, Dennis Jacobsen wrote:
> In lav_model_vcov(lavmodel = lavmodel, lavsamplestats = lavsamplestats, :
>   lavaan WARNING: unknown se type: robust

Normally, se = "robust" is internally changed to "robust.sem". What
happens if you explicitly add se = "robust.sem" to your sem() call?

Yves.

[Dennis Jacobsen]

Hi Yves, 

Thank you for your response. 

I am using the RSA function in which the estimator and standard error specifications get passed through to the sem function of the lavaan package. This makes troubleshooting a little harder, but after following your advice of changing the standard error specification the focal issue appears to have been resolved. However, I receive a new error message which indicates an underlying model identification issue. I will investigate this in more detail and try to resolve it by myself before asking further questions.  

Best, and thanks again, 

/Dennis.