This paper studies the self-sufficiency project (SSP), a controlled randomized experiment concerning welfare, by estimating a model of endogenous skill accumulation, multidimensional job opportunities, and time-varying opportunity costs of labour market time. Methods for estimating dynamic programming models with unobserved heterogeneity are extended to account for unexpected policy interventions and endogenous sample selection and initial conditions. Parameters are identified and consistently estimated by imposing optimal responses to the exact form of the SSP earnings supplement and the experimental program, which induces exogenous variation between treatment groups and within groups as treatment progresses. The estimated model tracks primary outcomes well in and out of sample, except for underestimating trends in the sample of new welfare applicants. Predictions from counterfactual experiments run counter to non-structural results reported elsewhere, and they suggest that details of the SSP's design are critical for interpretation of results. The separate SSP Plus treatment may have longer lasting and more generalized impacts than the in-sample impacts suggest.

Using data from professional baseball, basketball, and hockey, we estimate the parameters of a sequential game model of best-of-n championship series controlling for measured and unmeasured differences in team strength and bootstrapping the maximum-likelihood estimates to improve their small sample properties. We find negligible strategic effects in all three sports: teams play as well as possible in each game regardless of the game's importance in the series. We also estimate negligible unobserved heterogeneity after controlling for regular season records and past appearance in the championship series: Teams are estimated to be exactly as strong as they appear on paper.

We estimate an agency model using the payroll records of a copper mine that paid a production bonus to teams of workers. We estimate the cost of incomplete information due to insurance and incentives considerations and the inefficiency caused by the simple form of the incentive contract itself. At the estimated parameters the cost of worker risk aversion (insurance) is of similar magnitude to moral hazard (incentives). Overall, incomplete information accounted for one-half of the bonus system's inefficiency relative to potential full information profits. The other half is attributed to the bonus system's inefficient generation of incentives and insurance relative to the optimal incentive contract.