Geometric understanding of likelihood ratio statistics

Abstract

It is well known that twice a log-likelihood ratio statistic follows asymptotically a chi-square distribution. The result is usually understood and proved via Taylor's expansions of likelihood functions and by assuming asymptotic normality of maximum likelihood estimators (MLEs). We obtain more general results by using a different approach the Wilks type of results hold as long as likelihood contour sets are fan-shaped. The classical Wilks theorem corresponds to the situations in which the likelihood contour sets are ellipsoidal. This provides a geometric understanding and a useful extension of the likelihood ratio theory. As a result, even if the MLEs are not asymptotically normal, the likelihood ratio statistics can still be asymptotically chi-square distributed. Our technical arguments are simple and easily understood.