Detecting Latent Ideology in Expert Text: Evidence From Academic Papers in Economics

Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1804–1809, October 25-29, 2014, Doha, Qatar. c⃝2014 Association for Computational Linguistics Detecting Latent Ideology in Expert Text: Evidence From Academic Papers in Economics Zubin Jelveh1, Bruce Kogut2, and Suresh Naidu3 1Dept. of Computer Science & Engineering, New York University 2Columbia Business School and Dept. of Sociology, Columbia University 3Dept. of Economics and SIPA, Columbia University zj292@nyu.edu, bruce.kogut@columbia.edu, sn2430@columbia.edu Abstract Previous work on extracting ideology from text has focused on domains where expression of political views is expected, but it’s unclear if current technology can work in domains where displays of ide- ology are considered inappropriate. We present a supervised ensemble n-gram model for ideology extraction with topic adjustments and apply it to one such do- main: research papers written by academic economists. We show economists’ polit- ical leanings can be correctly predicted, that our predictions generalize to new do- mains, and that they correlate with public policy-relevant research findings. We also present evidence that unsupervised models can under-perform in domains where ide- ological expression is discouraged. 1 Introduction Recent advances in text mining demonstrate that political ideology can be predicted from text – often with great accuracy. Standard experimen- tal settings in this literature are ones where ide- ology is explicit, such as speeches by American politicians or editorials by Israeli and Palestinian authors. An open question is whether ideology can be detected in arenas where it is strongly dis- couraged. A further consideration for applied re- searchers is whether these tools can offer insight into questions of import for policymakers. To ad- dress both of these issues, we examine one such domain that is both policy-relevant and where ide- ology is not overtly expressed: research papers written by academic economists. Why economics? Economic ideas are important for shaping policy by influencing the public debate and setting the range of expert opinion on various policy options (Rodrik, 2014). Economics also views itself as a science (Chetty, 2013) carefully applying rigorous methodologies and using insti- tutionalized safe-guards such as peer review. The field’s most prominent research organization ex- plicitly prohibits researchers from making policy recommendations in papers that it releases (Na- tional Bureau of Economic Research, 2010). De- spite these measures, economics’ close proximity to public policy decisions have led many to see it as being driven by ideology (A.S., 2010). Does this view of partisan economics have any empiri- cal basis? To answer the question of whether economics is politicized or neutral, we present a supervised ensemble n-gram model of ideology extraction with topic adjustments.1 Our methodology is most closely related to Taddy (2013) and Gentzkow and Shapiro (2010), the latter of which used χ2 tests to find phrases most associated with ideology as proxied by the language of U.S. Congresspersons. We improve on this methodology by accounting for ideological word choice within topics and in- corporating an ensemble approach that increases predictive accuracy. We also motivate the need to adjust for topics even if doing …

Content truncated to 500 words