Created in collaboration with Danny Tobin, Tabinda Siddiqi, Cornelius von Lenthe, Bahjat Mansour, and Parthu Kalva
Abstract What are the political beliefs of China’s unelected officials on politically-sensitive topics? To what degree can responses from surveying the Chinese public be used to provide insight on the beliefs of officials? In this study, we apply machine learning logistic, random forest, and boosting classification models to answer these questions by predicting the responses of officials to censored questions concerning western political systems and property ownership. After distributing a survey to the Chinese public and later giving a modified version of the survey to officials, researchers have learned more about the contours of ideology in the Chinese public and the extent to which officials’ beliefs align or diverge in the aggregate from the citizens they serve. However, nine questions of the original citizen survey were censored when given to officials, leaving private beliefs of CCP officials on democracy, individual ownership of land, the use of force in Taiwan, and the one child policy unknown. We seek to understand the level of consensus on some of these questions by first validating a predictive model on two questions in the civilian dataset where the answer in the official set is known and then by applying the best model to the two censored questions of interest in the official dataset where a response is not known. If successful, this method could provide insights on opportunities for policy engagement on topics where there is disagreement or shifting beliefs within the Chinese Communist Party.