Nonlinear SEM with latent variables restricted to the unit interval

Modeling
1

Dear All,
I’m new to stan and wonder if someone has experience with the following kind of model (or if anyone sees reasons that estimating such models is doomed and impossible anyway).
The model I’m interested in are SEM models with nonlinear relations between latent variables and values of latent variables restricted to the unit interval in order to make fuzzy logic interpretation possible.
Any hints are welcome!
Reinhard

2

The blavaan R package is probably capable of doing what you’re looking for. Bayesian Latent Variable Analysis • blavaan

3

Thanks for your answer, however, as far as I understand blavaan uses the lavaan model syntax which is restricted to linear models, and even in linear models I don’t know if blavaan can restrict latent variables to an interval.
Reinhard

4

The nice thing about Gaussian latents is that sensible covariance is fairly straightforward. If it were me I would stick to unconstrained / Gaussian latents and transform to the space needed as an intermediate step before the measurement layer. But my fuzzy logic understanding is insufficient to even be considered fuzzy…

1 Like