But since the individual effects are so simple, can’t we just do this specific model without the fancy infrastructure? Same as with what’s happening here: Algebraic sovler problems (Laplace approximation in Stan) ?
Looking at eq. 24 here: https://arxiv.org/pdf/1206.5754.pdf
The things we need to consider are the prior covariance (which is diagonal) and then the Hessian of the likelihood with respect to the parameters we’re Laplacing out. This Hessian is still diagonal, right? So it’s just math with diagonal matrices?