Estimating Infrastructure Condition from a Biased Sample

Abstract

In this paper, we address the problem of making inferences about a population of infrastructure facilities from a subset that is a biased sample. We consider the case in which the sample is biased toward facilities in worse condition or requiring more expensive repair. Two methods are developed that incorporate a model of the process through which the sample is selected. One of the methods is based on well-known truncated distributions, whereas the other assumes that the bias operates continuously. The methods are applied to a class of facilities under the Federal Aviation Administration’s jurisdiction known as “unstaffed facilities.” These consist of structures housing radars, navigation aids, radio beacons, and other ground-based equipment, and no previous system-wide evaluation has been attempted for these facilities. We present and discuss the estimates obtained from both the methods, and examine their goodness-of-fit with the sample. Given the premise that bias exists, the continuous bias model proved more suitable. However, the continuous bias model did not surpass the truncation models in terms of goodness-of-fit.