Interval estimation based on the profile likelihood: Strong Lagrangian theory with applications to discrimination

Strong Lagrangian theory is used to illuminate the properties of the profile log likelihood. General conditions are given which lead to a simplification of the computations required to plot this function or calculate interval estimates based upon it. The results presented are rather general with a variety of possible applications. In particular, it is shown that they provide a simple solution to the important practical problem of obtaining an interval estimate for the posterior log odds ratio in the two population discrimination problem. The results are applied to the multivariate normal unequal covariance matrix case and are illustrated by an application.