Interval models for comparative probability on finite sets (Q1819810)

Let \(\prec\), is less probable than, be a comparative probability relation on the algebra \({\mathcal A}\) of all subsets of a finite set S. This paper considers a number of interval representations for \(\prec\), each of which is based on monotonic functions \(P_*\) and \(P^*\) on \({\mathcal A}\) with \(P_*\leq P^*\), \(P_*(\emptyset)=P^*(\emptyset)=0\), and \(P_*(S)=P^*(S)=1\). The basic interval model has \(A\prec B\Leftrightarrow P^*(A)<P_*(B)\), for all A and B in \({\mathcal A}\). Stronger models require a probability measure P such that \(P_*\leq P\leq P^*\) and propose additional restrictions on the functions such as superadditivity of \(P_*\) and subadditivity of \(P^*\). We also consider semiorder representations for which it is never true that \(P_*(A)<P_*(B)\leq P^*(B)<P^*(A).\) The paper first explains the nature of our interval order and semiorder models for comparative probability, then comments on connections to upper and lower probabilities in the mathematics and statistics literature. Conditions on \(\prec\) that are necessary and sufficient for most of the interval models are identified. Most of the axioms used for this purpose are familiar from previous work on comparative probability.