Starting from a given object representing a fitted model
(within a certain set of model classes) whose (non-)linear predictor
includes some ordered factor(s) among the explanatory variables, a new
model is constructed and fitted where each named factor is replaced by a
single numeric score, suitably chosen so that the new variable produces a
fit comparable with the standard methodology based on a set of polynomial
contrasts. Two variants of the present approach have been developed, one in
each of the next references: Azzalini (2023)