Making Cluster Analysis Inferential: From Discovery to Measurement

Cluster analysis is political science's primary tool for unsupervised discovery, yet the discipline has never required it to meet inferential standards. This paper introduces the Inferential Cluster Analysis (ICA) framework, which transforms cluster analysis from a descriptive tool into one capable of testing theoretical predictions, validating latent constructs, and supporting genuine inference. The framework builds three conditions into four steps : falsifiability, model fit, and robustness are enforced through theoretical pre-specification, multi-algorithmic estimation, model fitting and selection, and validation. An application tests a foundational assumption in acculturation research that has organized the field for decades. The binary model, which presumes a zero-sum relationship between heritage and American cultural attachments, fails to hold. In its place, four distinct orientations emerge: culture-affirming, assimilationist, bicultural, and demicultural. The framework thus both falsifies a longstanding assumption and validates a novel bidirectional model of acculturation.