A nonparametric distance model for unidimensional unfolding (Q4037073)

[For the entire collection see Zbl 0754.00011.] Unidimensional unfolding means to determine, from the set of individual rankings (on options), a \(J\)(oint) scale (i.e. unidimensional) on which individuals as well as options are ordered. This aim is placed here in the context of social choice theory: how to find out a ``fair'' group consensus ordering? Distance measures for rankings are recalled, especially Feigin and Cohen's probabilistic model [see \textit{P. D. Feigin} and \textit{A. Cohen}, J. R. Stat. Soc., Ser. B 40, 203-213 (1978; Zbl 0391.62035)] based on the number of inversions between rankings. Such models are strongly unimodal distance ones if there is a modal ordering \(u_ m\) such as the probability \(P(u)\) is nonincreasing as \(u\) moves farther from \(u_ m\) along a certain type of path. It is proved that if the data follow such a model, the maximum likelihood and the minimum-number-of-inversions criterion yield the same mean/modal/median ordering which can be viewed as the group consensus ordering. Moreover, the so-called social preference function is then unimodal on the \(J\) scale. The unfolding procedure is presented on two sets of data.