On software reliability and code coverage

Failure and module coverage data were collected during the system testing phase of a commercial hardware-software system called Product S. The Goel-Okumoto and Musa-Okumoto models were used to predict the Mean Test Cases to Failure (MTCTF) for these data. It was observed that the errors in predictions of MTCTF correlate with a sharp rise in code coverage. The sharp rise occurs when testers test code which was not previously tested. This observation suggests that software reliability prediction models may be able to improve the accuracy of prediction by accounting for code coverage.