Code coverage, what does it mean in terms of quality?

Unit code test coverage has long been known to be an important metric for testing software, and many development groups require 85% coverage to achieve quality targets. Assume we have a test, T/sub 1/ which has 100% code coverage and it detects a set of defects, D/sub 1/. The question, which is answered here, is "What percentage of the defects in D/sub 1/ will be detected if a random subset of the tests in T/sub 1/ are applied to the code, which has code coverage of X% of the code?" The purpose of this paper is to show the relation between code quality and code coverage. The relationship is derived via a model of code defect levels. A sampling technique is employed and modeled with the hypergeometric distribution while assuming uniform probability and a random distribution of defects in the code, which invokes the binomial distribution. The result of this analysis is a simple relation between defect level and quality of the code delivered after the unit code is tested. This model results in the rethinking of the use of unit code test metrics and the use of support tools.