Pairwise Comparisons as a Scale Development Tool for Composite Measures

Abstract

Abstract Composite scales are widely used for measuring aggregate social science concepts. These often consist of linear indices obtained as the weighted sum of a set of relevant indicators. However, selecting coefficients (or weights) that reflect the substantive importance of each indicator towards the concept of interest is a difficult task. We propose a method for the generation of linear indices for aggregate concepts based on pairwise comparisons. Specifically, we ask a group of subject-matter experts to perform a series of pairwise comparisons, with respect to the concept of interest, between profiles displaying different combinations of indicators. This allows us to estimate coefficients for each indicator that provide a linear approximation to how experts make the pairwise evaluations. As we show, the method makes it straightforward to assess intercoder reliability, while being a more accessible task than directly asking experts for coefficients. We demonstrate our method with an application to the concept of ‘productive ageing’, including a cross-cultural comparison of weighting schemes derived from a group of Italian and a group of South Korean experts on this concept.