Machine-learning applications to authoritarian selections: the case of China

Abstract

Elite selection in China has drawn significant attention given the importance of the country. Instead of relying on qualitative assessments from historical and personal insights, this study utilized machine-learning techniques to evaluate the promotion prospects of Chinese elites. By incorporating over 251 individual features of 18,179 officials from 1982 to 2020, I built up an ensemble model to calculate the promotion probabilities of the previous Politburo members of the Communist Party of China (CPC). Methodologically, this study finds that the machine-learning predictions yielded approximately 20% higher accuracy compared to the classical model, which employed the generalized linear model with theoretically identified variables. Moreover, this paper offers valuable insights into Chinese politics by highlighting that Xi Jinping’s selection of central officials has diverged from historical patterns, while his decisions on provincial promotions do not exhibit notable differences from those made by his predecessors.