This paper discusses the use of mixture models to allow for a random effects
structure in the analysis of partially ranked data. Partially ranked data occur in
sample surveys, where respondents are invited to choose the preferred and second most preferred (or the most preferred and least preferred) out of a larger set of items. We adopt an paired comparison approach used by Dittrich et al (2007) which treats partially ranked data as originating from underlying Likert scales. The work is motivated by two questions on (post)materialism in consecutive sweeps of the British Household Panel Survey. As part of the survey, respondents were asked to choose the most preferred and the next most preferred out of a set of four items representing preferred priorities for government.