Advisor: Peter Hoff
Maximum likelihood estimation is one of the most popular methods for performing statistical inference. Unfortunately, it requires the full and accurate specification of the data generating model in order to achieve maximum efficiency. When model specification is in doubt, the Sandwich estimator of variance can be used as a consistent estimator of the variance of the estimate of the parameter in question. However the Sandwich is a notoriously bad estimator at small sample sizes. In this talk we present a pivot quantity that provides more accurate confidence interval coverage at small sample sizes than the corresponding Sandwich-based confidence intervals, while equalling the asymptotic efficiency of the Sandwich.