Model averaging quantiles for censored data

Abstract

Quantiles are of interest in food safety data dealing with a limit of detection. The limit of detection introduces a lot of uncertainty in the left tail of the underlying distribution, making quantile estimation for this part of the distribution difficult. Therefore we fit a model to the data and derive the model-based estimate for the quantile. Since the true distribution is unknown, model averaging is used to combine information from a set of models. In this paper we discuss two approaches to use model averaging for quantiles. The methods are applied to a data example and compared in a simulation study. The effect of an increasing percentage of censoring on the estimates is explored.