A key challenge in designing social transfer programs is determining who should receive benefits. In developing countries, large informal sectors make income and employment hard to observe, and governments often rely on alternative targeting approaches. As countries develop and administrative capacity expands, income-based means-testing with third-party reporting is likely to become more widespread. Yet as countries make this transition toward means-testing, and when much of individuals' income still remains unverifiable, enforcing an income threshold may look very different than it would under full information. We study this question in the context of Brazil's Bolsa Família, a means-tested program in which eligibility is based on self-reported income, supplemented by annual audits in which the federal government cross-checks beneficiaries' declarations against formal administrative records. Exploiting quasi-random variation in whether a beneficiary's formal employment is visible to the government, we estimate the effect of enforcement on formal labor outcomes, and income and household composition reporting. We find that beneficiaries whose formal employment becomes visible to the auditor are substantially more likely to update their records and have more chances to have their benefits cancelled, confirming that audits are effective at detecting ineligible recipients. We then examine whether beneficiaries adjust along less observable reporting margins or whether enforcement induces exit from formal employment. To trace how responses evolve as verification capacity expands, we then leverage subsequent audit cycles that progressively incorporated additional data sources into the audit process.

How do public policies diffuse across governments? We study the adoption of a simple, low-cost, and relatively uncontroversial tax policy by municipal governments in Brazil. We revisit a field experiment that randomized 1,818 mayors to an information treatment where they learned about the policy and its effectiveness. We test whether the information treatment has an effect on the policy adoption of other municipalities. We find robust evidence of quantitatively large spillovers. The aggregate impact of geographic spillovers on adoption is comparable in magnitude to the direct treatment effect. Having a treated neighbor influences municipalities that were themselves treated, suggesting that these policy spillovers are not driven merely by the spread of the originally-provided information. About 40% of the spillover effect is mediated by adoption of the policy by neighboring municipalities. Considering all municipalities nationwide, geographic proximity plays the largest role in generating policy diffusion. One we zoom into a smaller geography such as a state, ideological/political similarity plays the dominant role in directing diffusion. Economic or demographic similarly do not appear to drive diffusion. Using unique survey data on mayors' communication networks, we show that communication patterns show the strongest role for geographic proximity, with political and economic similarity playing a statistically significant but smaller role.

We examine the role of politician preferences and beliefs about tax policy, drawing on a unique survey of more than 700 local government leaders (mayors) in Brazil. Mayors rate raising tax revenue as an important policy priority and express strong beliefs that a range of tax interventions have the potential to improve revenue collection. Mayors are generally confident in their beliefs and often appear overoptimistic relative to existing empirical evidence. Mayoral political preferences as well as beliefs about the effectiveness of tax interventions are predictive of mayors seeking out further knowledge by attending an information session on raising tax revenue.

Many transfer programs around the world face budgetary, logistical, or institutional limitations that prevent all means-eligible individuals from accessing benefits. These capacity constraints are often viewed as suboptimal as they hinder transfers from reaching eligible recipients. However, such constraints may also serve as a targeting mechanism when individuals' true ability or income is unobserved by the government. Intuitively, these limitations introduce uncertainty into program participation, reducing the incentive for less disadvantaged individuals to become eligible, and thereby leaving more resources available to those most in need. This paper develops a model that incorporates uncertainty surrounding program benefit receipt and analyze the welfare effects of an intervention that reduces the probability of receipt while increasing the benefit amount. I show that, under risk aversion, changes in the probability of benefit receipt and benefit size are not equivalent in terms of expected utility and, therefore, targeting. I calibrate the model using Bolsa Família parameters to illustrate this point. Finally, I show that such an intervention improves welfare only when the gains from lower social costs are strong enough to compensate for individuals' welfare loss from greater income risk.