Bayesian inference for partial orders from random linear extensions: power relations from 12th Century Royal Acta

We give a new class of models for time series data in which actors are listed in order of precedence. We model the lists as a realisation of a queue in which queue-position is constrained by an underlying social hierarchy. We model the hierarchy as a partial order so that the lists are random linear extensions. We account for noise via a random queue-jumping process. We give a marginally consistent prior for the stochastic process of partial orders based on a latent variable representation for the partial order. This allows us to introduce a parameter controlling partial order depth and incorporate actor-covariates informing the position of actors in the hierarchy. We fit the model to witness lists from Royal Acta from England, Wales and Normandy in the eleventh and twelfth centuries. Witnesses are listed in order of social rank, with any bishops present listed as a group. Do changes in the order in which the bishops appear reflect changes in their personal authority? The underlying social order which constrains the positions of bishops within lists need not be a complete order and so we model the evolving social order as an evolving partial order. The status of an Anglo-Norman bishop was at the time partly determined by the length of time they had been in office. This enters our model as a time-dependent covariate. We fit the model, estimate partial orders and find evidence for changes in status over time. We interpret our results in terms of court politics. Simpler models, based on bucket orders and vertex-series-parallel orders, are rejected. We compare our results with a stochastic process extension of the Plackett-Luce model.