Incentive design in sequential statistical protocols

Many statistical protocols are deployed inside incentive systems: firms decide whether to run and submit trials, researchers decide which projects to pursue, and regulators or conferences commit to approval rules before observing outcomes. We study a sequential principal-agent model of hypothesis testing in which a principal commits to a history-dependent testing threshold and an agent chooses costly effort and submission. The principal observes only public testing outcomes, not the latent quality that determines social value. Even in this no-feedback environment, dynamic testing rules can create continuation incentives that encourage effort and selective submission. We illustrate this effect in a two-period construction and analyze timeout policies in an infinite-horizon discounted setting, showing how future access to testing can substitute for direct observation of the agent’s private information.