The overwhelming majority of party policy analyses assume an Euclidean space adequately reducible to a single dimension. This spatial metaphor was adopted in Downsian models before its empirical utility was tested and later work has shown several problems with measuring policy in terms of few-dimensional spaces. The current paper argues theoretically as well as gives two empirical presentations of how the difference between political parties can be more adequately estimated on the basis of pairwise measures of programmatic overlap. This measure can be calculated on the basis of existing non-aggregated manifesto content analysis data and through adapting computer content analysis to estimate the difference between two manifestos directly. The paper provides an example of both approaches and shows how the measure, aggregated for sets of parties, can be implemented in the analysis of party interaction, helping to explain some of the current "puzzles" of coalition formation.
