Many important policy reforms have two key features. First, all units are exposed to the reform at the same time. Second, the reform changes units' incentives to take up an already available treatment. How can we leverage the reform to identify treatment effects in the absence of an unexposed comparison group? While there is no cross-sectional variation in policy exposure, units differ in how their treatment take-up responds to the reform. To leverage this variation for identification, I extend the local average treatment effect (LATE) framework to a panel setting where a unit's latent compliance type, defined by comparing potential treatment choices with and without the policy in each period, can change over time. I establish that there are two routes to point identify the LATE under parallel trends. The first approach restricts the mechanisms governing selection into treatment, ruling out specific transitions in compliance types over time. The second restricts heterogeneity in treatment effects across latent compliance types. To accommodate richer selection mechanisms and unobserved effect heterogeneity, I develop a marginal treatment effects approach that delivers sharp bounds on causal parameters of interest under economically motivated shape restrictions. Applying the framework to Medicare Part D, I estimate that prescription-drug coverage improved medication access and reduced labor supply among individuals induced to obtain it by the reform.

Economists often estimate causal effects of policies on multiple outcomes and summarize them into scalar measures of cost-effectiveness or welfare, such as the Marginal Value of Public Funds (MVPF). In many settings, microdata underlying these estimates are unavailable, leaving researchers with only published estimates and their standard errors. We develop tools for valid inference on functions of causal effects, such as the MVPF, when the correlation structure is unknown. Our approach is to construct worst-case confidence intervals, leveraging experimental designs to tighten them, and to assess robustness using breakdown analyses. We illustrate our method with MVPFs for eight policies.

The effect of a treatment may depend on the intensity with which it is administered. We study identification of ordered treatment effects with a binary instrument, focusing on the effect of moving from the treatment’s minimum to maximum intensity. With arbitrary heterogeneity across units, standard IV assumptions (Angrist and Imbens, 1995) do not constrain this parameter, even among compliers. We consider a range of additional assumptions and show how they can deliver sharp, informative bounds. We illustrate our approach with two applications, involving the effect of (1) health insurance on emergency department usage, and (2) attendance in an after-school program on student learning.

Tax benefits tied to children form a central component of the social safety net in the United States. To participate in these programs, taxpayers must claim a child on their tax return. We study the claiming of children on tax returns by drawing on health insurance information returns to establish the presence of children in the United States. We estimate that the vast majority of insured children (approximately 95 percent) and a significant majority (between 88 and 97 percent) of all U.S. children are claimed on tax returns. Unclaimed children are disproportionately concentrated in lower income households.

Tax administrations around the world are investing heavily in digitization, creating new opportunities for enforcement. Can low-cost enforcement tools built atop this digital infrastructure increase tax capacity? We study an Indian intervention implementing automated cross-checks to target fraudulent input tax credits. Using administrative data, we show that the policy reduced fraudulent credits and increased tax payments, resulting in a 5 percent reduction in overall evasion at minimal cost. We find little adverse impact on legitimate activity. This type of policy is readily implementable in many developing countries that already have digital filing but are yet to exploit its full potential.

We study the role of job transitions and firm pay policies in the Black-White earnings gap in the US. We use administrative data for the universe of employer-employee matches from 2005-2019 to analyze worker mobility in a general but tractable framework, which allows for firm effects that depend on workers’ job history. Using differences in average pay between origin and destination firms as the treatment intensity of a job move, we analyze transitions up and down the job ladder and estimate race-specific passthrough rates of average firm pay into a mover’s own earnings. First, we find race-specific asymmetry around the direction of the move, whereby losses experienced in downward transitions are meaningfully larger than gains from upward transitions with a similar treatment intensity. For a \$1 earnings increase in transitions up the job ladder, earnings passthroughs in transitions down the job ladder impose an earnings loss of \$1.25 among White workers and \$1.50 among Black workers. Second, we uncover career setbacks as a novel pathway in the evolution of racial earnings gaps. In transitions down the job ladder, Black workers lose an additional \$0.24 for every \$1 decrease in White workers’ earnings, a finding which prevails across sex and age. This “racial penalty” is not driven by differential pay, as it is completely absent when Black and White workers move between the same firm pairs. Instead, the penalty is due to differential sorting following career setbacks, so that Black workers regain employment in “worse” jobs, with strong evidence for racial differences in access to short-run liquidity as a mechanism. Overall, our findings offer a robust and computationally simple framework for modeling earnings determination processes and have implications for safety-net policies in the American labor market.