Effort without Evidence

Abstract

Actions determine not only payoffs but what can be learned. I study communication across successive decision makers when hidden effort governs public evidence and a sender motivates her successor while underweighting effort cost. Common sender rankings can force every message into a Bayes-unavoidable nonidentifying region. If experimentation is costly, strong motivation can make persuasive encouragement incredible and produce inactivity after a confounded failure. If maximal effort is technologically saturated, the same bias can produce universal overexertion after a bad pooled history, which is equally uninformative. At such nodes collapse is monotone: once strong motivation eliminates the identifying action, strengthening it cannot restore information. Moderate motivation instead preserves causal distinctions. In a threshold application, stronger motivation raises measured success yet lowers welfare because society fails to learn that less effort would suffice. A Kullback--Leibler decomposition shows that certification directly adds no information after entry into a fixed recursively nonidentifying region. Whether it prevents entry depends on attribution--selection alignment: do certificate-induced posteriors preserve the identifying action? The corrective action payment changes sign across technologies---subsidize effort when agents do too little, but subsidize restraint when maximal effort is saturated. Motivation selects the costly action, not the informative one.