Statements versus predicates in spectral bug localization

This paper investigates the relationship between the use of predicate-based and statement-based program spectra for bug localization. Branch and path spectra are also considered. Although statement and predicate spectra can be based on the same raw data, the way the data is aggregated results in different information being lost. We propose a simple and cheap modification to the statement-based approach which retains strictly more information. This allows us to compare statement and predicate "metrics" (functions used to rank the statements, predicates or paths). We show that improved bug localization performance is possible using single-bug models and benchmarks.