Correlating testability with fault detection
M. Ray Mercer, B. Underwood · International Test Conference · 1984
This work refines a statistical approach to the evaluation of testability algorithms, by: 1) using a more sensitive measure of correlation between testability measure and ease of fault detection, 2) adopting the proportion of test patterns that detected a fault as the measure of actual ease of detection, and 3) using a multiple correlation/regression approach that determines the optimal weighting of controllability and observability. Results indicate a moderate correlation between SCOAP testability and fault detection proportion (approx .. 5), higher than previously found in such approaches but still low enough to indicate substantial room for improvement in the prediction provided by testability measures. In addition, the sum of controllability and observability predicts the results of fault detection almost as well as the optimal linear combination of the two.