Overview
James G. Scott is a statistician at the University of Texas at Austin, known for foundational work on Bayesian multiplicity correction and sparse signal detection. With James Berger he established that Beta(1,1) priors on variable-inclusion probabilities automatically correct for multiplicity in large model spaces (Scott-Berger 2006, 2010).
Key Contributions / Features
- Scott-Berger multiplicity correction: placing Beta(1,1) on inclusion probability q induces shrinkage of posterior inclusion odds when many predictors are tested; avoids the need for an explicit Bonferroni-type correction (Hahn, Carvalho, Scott 2012)
- Co-developed sparse factor analytic probit: q_s ~ Beta(1,1) on loading inclusion in each factor column s
- Horseshoe prior for sparse normal means (Scott-Carvalho); co-development of sparse Bayesian regression methods
Related