On the independence of software inspectors

This paper investigates the proposition: are the results (defects found) of individual software inspectors during a software inspection process (preparation phase) statistically independent or dependent? The rationale for exploring this proposition is that many capture -recapture models assume statistical independence in their formulations. Recently these models have been applied to estimating the number of undiscovered defects undetected by an inspection process. Using large amounts of inspection data from a previous experiment, and subsequently simulating a capture-recapture time-series from this data, the paper explores this proposition. The correlation within the time-series is then estimated using non-parametric techniques. The paper concludes that, for the data used, no evidence exists to support the statement that the results from individual inspectors are dependent. (C) 2002 Elsevier Science Inc. All rights reserved.