Standardization issue
Opened this issue · 3 comments
brandmaier commented
Onyx currently standardizes covariances with respect to total variance in variables; however, users expect a correlation of the residual covariance (in line with lavaan output). See these equivalent models that yield quite different estimates of the residual correlation. The right model's result is identical to the lavaan output from standardizedSolution()
@timovonoertzen, let's discuss how to proceed.
brandmaier commented
here is a useful reference: https://tdjorgensen.github.io/SEM-in-Ed-compendium/ch6.html
brandmaier commented
My suggestion is that the backend should provide standardized estimates via the ParameterSet class. What do you think, @timovonoertzen ?
timovonoertzen commented
Mmh… could be done. So far, the Estimate class is minimal and tries to avoid redundant information, and we should avoid that it is computed a unnecessary steps during the optimization. But it would be doable to introduce a method in the class that delivers derived values.
Von: Andreas Brandmaier ***@***.***>
Gesendet: Mittwoch, 8. Mai 2024 11:40
An: brandmaier/onyx ***@***.***>
Cc: timovonoertzen ***@***.***>; Mention ***@***.***>
Betreff: Re: [brandmaier/onyx] Standardization issue (Issue #12)
My suggestion is that the backend should provide standardized estimates via the ParameterSet class. What do you think, @timovonoertzen <https://github.com/timovonoertzen> ?
—
Reply to this email directly, view it on GitHub <#12 (comment)> , or unsubscribe <https://github.com/notifications/unsubscribe-auth/AP2O65BZH6ATCAKCOSQHBTLZBHXIFAVCNFSM6AAAAABCRIFJIKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCMBQGE3TSMJZGM> .
You are receiving this because you were mentioned.Message ID: ***@***.***>