Dummy code & LAMBDA indices not adjusting when estimating latent interactions in SAM
E-M-McCormick opened this issue · 2 comments
This stems from a conversation I had with Yves at Working Group SEM in Twente - really appreciate the help!
A bit of background, we have a linear growth model with a observed distal outcome (z) where we want to include bilinear (int:slp) and quadratic (int:int & slp:slp) interactions of the growth factors. The main error we get is Error in LAMBDA[[b]][cbind(dummy.ov.idx, dummy.lv.idx)] <- 1 : subscript out of bounds which when we debugged (using debug(lavaan:::lav_sam_step1_local)) was coming because the indexes do not re-adjust when the dummy latent interaction column in LAMBDA is removed. When this column is removed the model tries to reference the distal outcome column as 4 instead of 3, which is then out of bounds. We tried to specify z as a latent variable explicitly fz but this did not solve the issue. See the attached zip file for a MRE with R code for generated data and model syntax. Happy to clarify further if anything is unclear.
Thanks for the reprex. This should work now with the github version. Let me know if you still get issues.