Properties of Effective Metrics for Coverage-Based Statistical Fault Localization

Abstract

In this paper, we investigate several coverage-based statistical fault localization metrics that have performed well in recent comparisons of many metrics, in order to better understand the properties of effective metrics. We first algebraically and probabilistically analyze the metrics to identify their key elements. Then we report on an empirical study we conducted to assess the relative importance of those elements. The results suggest that the most effective metrics contain a product of two terms: one that estimates the failure-causing effect of a program element (possibly with confounding bias) and one that weights the first term based on the evidence for the existence of faults in other program elements.