* for both defined and undefined control samples * using the least-squares approach currently used for bias estimation with `center()` * also demonstrate how to do this with `nnet::multinom()`