Estimates structural equation models by case-based least
squares: the latent scores of every observation are treated as free
variables of a constrained optimization problem, so that arbitrary
nonlinear model equations, bounds and constraints on latent variables
and inequality constraints on parameters become possible. Model
equations are specified as plain text (e.g. "y == a*exp(b*eta)").
Gradients are obtained by automatic differentiation via 'RTMB', and
the constrained problem is solved with 'nloptr' (SLSQP or augmented
Lagrangian). Missing data are handled case-wise. The methodology is
described in Oldenburg (2024)