Empirical likelihood provides likelihood-based confidence regions and tests without requiring the user to know a parametric family generating the data. The computation of the empirical likelihood ratio function for the mean reduces to a convex optimization. This paper reformulates empirical likelihood optimization as the minimization of a self-concordant convex function. Under self-concordance there is a mathematical guarantee of convergence for Newton iterations with backtracking. Art Owen is a professor of Statistics at Stanford University. He holds a Bachelor of Mathematics from the University of Waterloo, and a PhD in Statistics from Stanford University. His research interests focus on measuring uncertainty from data with minimal assumptions. This includes the method of empirical likelihood, which provides likelihood inferences without assuming parametric model forms. He has also worked on randomized quasi-Monte Carlo which attains close to an error rate n -3/2 for integration of smooth multidimensional functions while allowing sample driven error estimates.

The award-winning paper will be presented by Professor Owen at the Annual Meeting of the Statistical Society of Canada to be held in Toronto, Ontario, May 25-28.