Testing group differences in paired comparisons data.

Abstract

The Bradley-Terry-Luce (Bradley & Terry, 1952; Luce, 1959) model for paired comparisons data is extended to the case in which there are several choice matrices, one for each of several groups. The extended model leads to tests of various null hypotheses: the hypothesis that there is a common stimulus dimension in all groups, the hypothesis that the scale value for a given stimulus in group g equals its scale value in a target group, and the hypothesis that the scale value for a given stimulus in group g equals its average scale value across all groups. Weighted least sqares and maximum likelihood methods for fitting the model are discussed