Identifying assumptions and research dynamics

A representative researcher pursuing a question has repeated opportunities for empirical research. To process findings, she must impose an “identifying assumption”, which ensures that repeated observation would provide a definitive answer to her question. Research designs vary in quality and are implemented only when the assumption is plausible enough according to a KL-divergence-based criterion, and then beliefs are Bayes-updated as if the assumption were perfectly valid. We study the dynamics of this learning process and its induced long-run beliefs. The rate of research cannot uniformly accelerate over time. We characterize environments in which it is stationary. Long-run beliefs can exhibit history-dependence. We apply the model to stylized examples of empirical methodologies: experiments, causal-inference techniques, and (in an extension) “structural” identification methods such as “calibration” and “Heckman selection.”

Joint with Andrew Ellis