What makes inspections work?

Software inspections are considered a cheap and effective way to detect and remove defects from software. Still, many people believe that the process can be improved and have thus proposed alternative methods. We reviewed many of these proposals and detected a disturbing pattern: too often competing methods are based on conflicting arguments. The existence of these competing views indicates a serious problem: we have yet to identify the fundamental drivers of inspection costs and benefits. We first built a taxonomy of the potential drivers of inspection costs and benefits. Then we conducted a family of experiments to evaluate their effects. We chose effort and interval, measured as calendar time to completion, as our primary measures of cost. Defects discovered per thousand lines of code served as our primary measure of inspection benefits.