Using Internet Searches in Russia for Measurement and Prediction

• In Event: Big Data and Machine Learning

Abstract

In this paper, I describe and employ novel search engine data from leading Russian site Yandex to measure public sentiment and behavior. This data covers a large variety of search terms at weekly and monthly intervals at the subnational level in Russia from 2014 to 2017. I use machine learning to build predictive models of support for Putin’s regime, voting behavior, general political attitudes, and other estimands. I validate these measures using publicly-available polling data, official statistics, and new MRP estimates of region-level attitudes. Finally, I discuss possible substantive uses for and limitations of such data in the study of comparative politics.

Author

• Noah Buckley, Columbia University