The universal LRT provides finite-sample valid hypothesis tests and confidence sets in any setting for which we can compute the likelihood ratio. We present the first in-depth exploration of the size, power, and relationships between several universal LRT variants. We illustrate the benefits of the universal LRT in a test of a non-convex doughnut-shaped null hypothesis.
Robin Dunn, Aaditya Ramdas, Sivaraman Balakrishnan, Larry Wasserman
Shape-constrained density estimation poses a middle ground between fully nonparametric and fully parametric density estimation. Log-concavity is a common choice of shape constraint. Using universal inference, we develop the first test for log-concavity that is provably valid. Validity holds in finite samples.
Robin Dunn, Aditya Gangrade, Larry Wasserman, Aaditya Ramdas
We extend conformal prediction methods to a two-layer hierarchical setting. Conformal prediction constructs valid prediction sets in finite samples, even if the predictive model is incorrect.
Robin Dunn, Larry Wasserman, Aaditya Ramdas
February 2022
Journal of the American Statistical Association
We construct the first risk score for end-stage knee OA, using an end-stage definition that depends on symptomatic and radiographic criteria. We note particular implications for clinical trial patient selection.
Robin Dunn, Joel Greenhouse, David James, David Ohlssen, Peter Mesenbrink