Feature request: "Effects-coding" method of factor scaling in SEM

[HeikoBreitsohl]

• Enhancement: In addition to "factor loadings" and "residual variance", add a third option "effects coding" to the options for factor scaling in the SEM module.
• Purpose: While it is not as well known, effects coding can be very useful for factor scaling, particularly with multiple-group analyses (see Little, Slegers, & Card, 2006). Briefly, effects coding entails constraining factor loadings to average to unity, and indicator intercepts to average to zero. Some of the advantages are: 1) you are able to make weaker invariance assumptions; 2) factor scaling uses all indicators (weighted by factor loading) instead of a single indicator; and 3) factors are scaled like the indicators, which sometimes helps with Interpretation. My guess is that it would be relatively easy to implement, but extremely helpful, because applying it manually is rather cumbersome. For example, in Mplus, you have to do it in the "model constraint" command, which can get a bit messy with many latent factors, and which makes some output options unavailable. I am unaware that any software package offers effects coding as a built-in option.

Keep up the great work!

Little, T. D., Slegers, D. W., & Card, N. A. (2006). A non-arbitrary method of identifying and scaling latent variables in SEM and MACS models. Structural Equation Modeling, 13(1), 59-72.

[vankesteren]

This is implemented in the CFA function in SEM since JASP version 0.10!