Statistical Misreporting from Below in Communist Regimes, as Seen from Above

• In Event: Political Economy in China and Vietnam -- Domestic and International

Abstract

Authoritarian survival depends on whether regime leaders can collect reliable information from society. In authoritarian regimes with hierarchical and closed bureaucracies such as the one-party communist regimes of Soviet Union or China, however, even when they have significant capacity to collect information, government agents at lower levels have both the ability and the incentives to distort such information by misreporting internal government statistics. Even regime leaders do not know exactly which agent is lying and by how much, even when they may be aware of the problem at an aggregated scale. As a result, despite the penetration of highly sophisticated information collecting institutions and technologies, society remains largely ineligible to regime leaders.

Leveraging remote sensing data such as satellite images as credible and highly disaggregated proxies for patterns of misreporting and falsification by subnational governments, I show that the extent of misinformation facing authoritarian leaders is substantial. Specifically, using the case of Vietnam as an example, I use nighttime luminosity data to measure the exaggeration and under-reporting of subnational GDPs by provincial governments through a four-step process. First, I combine different sources of nightlight data to produce a consistent time-series dataset for the entire world that can be disaggregated to any geographic boundary. Second, I leverage incongruences in seasonal variation across patterns of consumption and manufacturing activities and across a number of matched countries to identify the sensitivity of these components of GDP statistics to changes in observed nightlight. Third, using a hierarchical model, I show that the relationship between official GDP numbers at subnational level in Vietnam and nightlight deviate from that in virtually any other country except for China. Finally, using a combination of forensics methods and expert surveys, I demonstrate that the degree to which the relationship between GDP numbers and nightlight diverges from the common pattern matches qualitative measures of statistical misreporting in accuracy, but far outperforms them in accessibility. I also outline how this method could be generalized to capture statistical misreporting across a wider range of policy areas and in different country contexts.

Author

• Minh Duc Trinh, Massachusetts Institute of Technology