Estimating Dynamic Treatment Effects from Project STAR (Working Paper 36)

This paper considers the analysis of data from randomized trials which offer a sequence of interventions and suffer from a variety of problems in implementation. In experiments that provide treatment in multiple periods (T>1), subjects have up to 2^{T}-1 counterfactual outcomes to be estimated to determine the full sequence of causal effects from the study. Traditional program evaluation and non-experimental estimators are unable to recover parameters of interest to policy makers in this setting, particularly if there is non-ignorable attrition. We examine these issues in the context of Tennessee's highly influential randomized class size study, Project STAR. We demonstrate how a researcher can estimate the full sequence of dynamic treatment effects using a sequential difference in difference strategy that accounts for attrition due to observables using inverse probability weighting M-estimators. These estimates allow us to recover the structural parameters of the small class effects in the underlying education production function and construct dynamic average treatment effects. We present a complete and different picture of the effectiveness of reduced class size and find that accounting for both attrition due to observables and selection due to unobservable is crucial and necessary with data from Project STAR

Parallel nonconvex generalized Benders decomposition for natural gas production network planning under uncertainty

A scenario-based two-stage stochastic programming model for gas production network planning under uncertainty is usually a large-scale nonconvex mixed-integer nonlinear programme (MINLP), which can be efficiently solved to global optimality with nonconvex generalized Benders decomposition (NGBD). This paper is concerned with the parallelization of NGBD to exploit multiple available computing resources. Three parallelization strategies are proposed, namely, naive scenario parallelization, adaptive scenario parallelization, and adaptive scenario and bounding parallelization. Case study of two industrial natural gas production network planning problems shows that, while the NGBD without parallelization is already faster than a state-of-the-art global optimization solver by an order of magnitude, the parallelization can improve the efficiency by several times on computers with multicore processors. The adaptive scenario and bounding parallelization achieves the best overall performance among the three proposed parallelization strategies.