## Avoid partial replicate designs, pleeeze! [🇷 for BE/BA]

Hi again,

it is rather experimental.
I do not have a level of understanding that allows me to use V=ZGZt+R,
so I will be constructing V directly from the data listing without the intermediary Z and G. I do not wish to get bogged down by the various aspects of covariance matrices; if CSH does not pan out well in SAS then it is an implenetation or convergence issue only, it is not because CSH is irrelevant. CSH is for all practical purposes the covariance matrix that makes sense, which means we have:
A variance for T (combined within and between)
A variance for within-R
A variance for between-R

Hessian, positive definite, etc.. all this comes down to the details of SAS's or WinNonlin's optimiser. I am playing around with optim in R (BFGS, L-BFGS-B bounded, and Nelder-Mead) and with one of my own making.

Pass or fail!
ElMaestro