Upper bounds on the minimum coverage probability of model averaged tail area confidence intervals in regression

Abstract

Frequentist model averaging has been proposed as a method for incorporating "model uncertainty" into confidence interval construction. Such proposals have been of particular interest in the environmental and ecological statistics communities. A promising method of this type is the model averaged tail area (MATA) confidence interval put forward by Turek and Fletcher, 2012. The performance of this interval depends greatly on the data-based model weights on which it is based. A computationally convenient formula for the coverage probability of this interval is provided by Kabaila, Welsh and Abeysekera, 2016, in the simple scenario of two nested linear regression models. We consider the more complicated scenario that there are many (32,768 in the example considered) linear regression models obtained as follows. For each of a specified set of components of the regression parameter vector, we either set the component to zero or let it vary freely. We provide an easily-computed upper bound on the minimum coverage probability of the MATA confidence interval. This upper bound provides evidence against the use of a model weight based on the Bayesian Information Criterion (BIC).