Scaling Policy Positions From Coded Units of Political Texts∗

Applying a coding scheme to discrete text units has long been the most common method for estimating substantive quantities of interest about the authors of these texts, whether for political, social, economic, or other substantive reasons. In political analysis, researchers typically build scales of policy positions from the relative frequencies of text units coded as left versus right policy categories. In this paper we reexamine the theoretical and linguistic basis for such scales, proposing a new alternative based on the logarithm of odds-ratios that is consistent with this underlying political and linguistic mechanism. We contrast this scale to previous approaches using text units coded into political categories from the discipline’s longest-running content analysis dataset, that of the Comparative Manifesto Project (CMP). We demonstrate that the logit scale avoids widely acknowledged flaws in previous approaches and validate it through comparison to independent expert surveys of policy positions. Applied to existing CMP data, without requiring any estimation or inferential procedures, we show how to unlock more policy dimensions, for more years, than have ever been provided before, and we make this new dataset available along with estimates of uncertainty for each measure. Finally, we draw some lessons for the future design of coding schemes for political texts.