Measuring Democracy Using the Wisdom of the Crowds

Measuring democracy remains a core challenge in the social sciences. Leading indicators rely on the assessments of a small groups of experts. This paper identifies two key statistical limitations underlying these measurements. First, expert judgments are likely to be highly correlated, thus reducing the value of aggregation. Second, expert pools are often small and thus noisy. We model democracy as a latent, unobservable state and propose a new approach rooted in the logic of the "wisdom of crowds" to address these issues. We show that including a broader and more diverse set of observers—including non-experts—can improve inference. Our theoretical results quantify when such larger, decorrelated samples can outweigh the costs of noisier individual judgments, and thus provide more accurate estimates of the true state. We then develop an algorithm for implementation and apply it to data on democratic performance in the U.S., demonstrating significant empirical gains.