Jie Ma

Many policies require citizens to actively and voluntarily participate in their implementation in order to produce successful outcomes, a situation referred to as citizen coproduction. Though citizen coproduction is necessary in many policy areas, it could create difficulties for democratic accountability because citizens and politicians share responsibility for policy failures and the attribution of blame may be difficult. We analyze the problem of democratic accountability with citizen coproduction in the context of a model in which politicians, who vary with respect to their competence at selecting policies, choose whether or not to implement policies whose success depends on the coproduction choices of citizens. For citizens, the problem is one of attribution: when citizen coproduction is low, it may be hard to select good incumbents because policy failures are frequent even in light of good policy choices. Furthermore, incumbent decisions anticipate these effects, which sometimes implies that incumbents are less likely to choose reforms when they are more likely to be successful.

Institutions that scrutinize wrongdoing are expected to bring more of it to light. We show that whether they do depends on what an investigation reveals, not on whether one occurs. We study a disclosure game in which an agent privately learns a damaging fact and chooses whether to reveal it, and in which silence may trigger an investigation that reports a verifiable finding. Absent investigation, everyone who knowingly conceals is worse than he is, so disclosure is his only way to separate from them. An investigation that establishes exactly what the agent knew lets him be judged as himself rather than alongside those concealers, and disclosure falls. An investigation that reports only that something was concealed does the opposite: being caught is no longer a shelter but a stigma, since it again groups him with concealers who are uniformly worse, and disclosure can rise above its no-investigation level. The design problem of scrutiny is therefore not how much to investigate, but how finely to report what is found.

Why do biased advisers sometimes improve collective decisions? I model a sequence of biased Senders who inherit accumulated knowledge, choose whether to run a costly experiment, and pass information down until it reaches a Receiver who decides for all. Bias cuts both ways: biased Senders are reluctant to relay evidence, but because they are hard to persuade, they have the strongest incentive to produce counterevidence against the prevailing consensus. I characterize the biases sustaining an all-honest equilibrium and show the set is nested along the chain—as deliberation lengthens, only more extreme Senders still experiment. Larger groups deliberate less effectively for the same reason. I then compare two organizational forms: sequential talk, in which each Sender observes what her predecessors pass down and speaks once before handing off, and group talk, in which multiple Senders share a common pool of knowledge and contribute to it in parallel. Sequential talk degrades faster along the chain than group talk. With the committee fixed, adding members shrinks every Sender's incentive to produce; but once participation is endogenous, surplus entrants simply become rubber stamps and neither structure collapses. The optimal deliberative structure therefore depends on the distribution of biases in the organization.