CMStatistics 2015: Start Registration
View Submission - CMStatistics
B0740
Title: Higher-order adjustments of the signed scoring rule root statistic Authors:  Valentina Mameli - University of Udine (Italy)
Monica Musio - University of Cagliari (Italy)
Laura Ventura - University of Padova (Italy) [presenting]
Abstract: Proper scoring rules can be used as an alternative to the full likelihood, when the aim is to increase the robustness. Proper scoring rule inference is usually based on the first-order approximations to the distribution of the scoring rule estimator or of the scoring rule ratio test statistic. However, several examples illustrate the inaccuracy of first-order methods, even in models with a scalar parameter, when the sample size is small or moderate. Analytical higher-order asymptotic expansions for proper scoring rules, generalizing results for likelihood quantities but allowing for the failure of the information identity, have been previously discussed. However, the calculation of the quantities involved in the analytical adjustments of the signed and signed profile scoring rule root statistic is cumbersome, even for simple models. The aim is to discuss the alternative approach to higher-order adjustments, based on a parametric bootstrap.