Putting Text in Context: How to Estimate Better Left-Right Positions by Scaling Party Manifesto Data using Item Response Theory∗

For over three decades, party manifestos have formed the largest source of textual data for estimating party policy positions and emphases, resting on the pillars of two key assump-tions: that party policy positions can be measured on known dimensions by counting text units in predefined categories, and that more text in a given category indicates stronger emphasis. Here we revisit the inductive approach to estimating policy positions from party manifesto data, demonstrating that there is no single definition of left-right policy that fits well in all contexts, even though meaningful comparisons can be made by locating parties on a single dimension in each context. To estimate party positions, we apply a Bayesian, multi-level, Poisson-IRT measurement model to category counts from coded party mani-festos. By treating the categories as “items ” and policy positions as a latent variable, we are able to recover not only left-right estimates but also direct estimates of how each policy category relates to this dimension, without having to decide these relationships in advance based on political theory, exploratory analysis, or guesswork. Finally, the flexibility of our framework permits numerous extensions, designed to incorporate models of manifesto au-thorship, coding effects, and additional explanatory variables (including time and country effects) to improve estimates