Duality in balance optimization subset selection

In this paper, we investigate a specific optimization problem that arises in the context of Balance Optimization Subset Selection (BOSS), which is an optimization framework for causal inference. Most BOSS problems can be formulated as mixed integer linear programs. By relaxing the integrality constraints so that fractional contributions of control units are permitted, a linear program (LP) is obtained. Properties of this LP and its dual are investigated and a sensitivity analysis is conducted to characterize how the objective value changes as the covariate values are perturbed.