Measuring Social Attitudes With a Best-Worst Multidimensional Unfolding Model: Separating Proximity From Statement...

This study proposes a Best-Worst multidimensional unfolding model for measuring social attitudes by separating two sources of comparative proximity judgments: multidimensional semantic proximity between respondents and statements and statement-level distance-independent selectability. The model locates respondents and statements in a common multidimensional attitude space and represents choices of the statements closest to and farthest from a respondent's opinion through utility differences based on Euclidean distance. A statement-specific parameter captures a tendency to be selected as closest more often and as farthest less often after respondent-statement distance is controlled. All parameters are estimated jointly in a Bayesian framework. Simulation studies evaluated parameter recovery under practical Best-Worst block designs and examined two forms of model misspecification. Under the correctly specified model, statement configurations and distance-independent selectability were recovered accurately across the examined conditions, whereas recovery of respondent locations depended more strongly on dimensionality and the amount of comparison information. Omitting distance-independent selectability systematically displaced statement locations and distorted interstatement and respondent-statement distances. When the fitted dimensionality was lower than the true dimensionality, recovery of both the spatial structure and distance-independent selectability deteriorated. These results show that multidimensional proximity and statement selectability should be modeled jointly when Best-Worst judgments are used to interpret the semantic structure of social attitudes.