Scalable methods for Bayesian selective inference (Q1657959)

It is well known that a selective posterior model along the conditional approach has two components: a truncated likelihood and a prior distribution on the parameters in the likelihood. The present work provides an optimization problem that approximates the otherwise intractable selective posterior and leads to scalable methods that give valid post-selective Bayesian inference. A randomization scheme is proposed under which the approximating optimization has separable constraints. The adjusted estimates empirically demonstrate better frequentist properties in comparison to the unadjusted estimates based on the usual posterior, when applied to a wide range of constrained convex data queries.