This paper proposes Vuong-type tests to select between two partially-identifed mo- ment inequality models based on their Kullback-Leibler distances to the true data dis- tribution. The candidate models can be non-overlapping or overlapping. We treat the two cases separately and develop a testing procedure for each. Both testing procedures are shown to have correct asymptotic size in a uniform sense, despite the potential lack of point identifcation of the model parameters. The tests are shown to be consistent and have nontrivial power against n1=2-local alternatives. Monte Carlo simulation of an entry game is used to demonstrate the …nite-sample performance of the tests. The tests are relatively easy to implement.

Paper can be downloaded at: http://pantheon.yale.edu/~xs37/JMP_shi.pdf