no default or option (outside of explicit model syntax) to covary observed and latent covariates

[vsavalei]

Suppose the model is, compmodel <- "C1 ~ C2 + Y7", where C1 and C2 are latent variables, and Y7 is an observed variable. When such a model is fit using sem(), lavaan specifies a zero correlation between the predictors C2 and Y7. This is not the case if both predictors are observed, or if both predictors are latent. Questions:

1. Should the default be to covary all predictors, whether latent (auto.cov.lv.x=TRUE by default) or observed (always true), or will this wreak havoc with some other common SEM?
2. Should there be an option to easily specify these covariances? Currently, orthogonal/orthogonal.x/orthogonal.y options only apply to latent variables.

[yrosseel]

I added an 'auto.cov.x' option. If set to TRUE, it does what you suggest: all exogenous variables (observed and latent) are then allowed to covary.

But of course, setting this to TRUE per default breaks a few packages. In order not to complicate the release of 0.7-1 even more, the option is now still set to FALSE. I will keep this issue open until we can set it to TRUE.