A Selective Review of Bayesian Variable Selection Procedures
Event Type:
Colloquium
Speaker:
Fletcher Christensen
Event Date:
Friday, August 28, 2015 -
12:30pm to 1:30pm
Location:
SMLC 356
Audience:
General PublicFaculty/StaffStudentsAlumni/Friends
Sponsor/s:
Stat group
Event Description:
A Bayesian approach to variable selection in regression problems is to
embed all possible submodels in a heirarchical mixture model and use MCMC
procedures to identify which submodels perform best. There is substantial
variability in how this approach performs, however, based on how priors are
chosen for the regression parameters for variables whose inclusion is being
evaluated. This talk will focus on three methods for addressing the problem
of prior se lection: (1) George & McCulloch's (1993, 1997) Stochastic
Search Variable Selection (SSVS), (2) the g-prior mixture approach of Liang
et al. (2008), and (3) the criterion-based approach of Bayarri et al.
(2012).
