Historically, analysis in the team context has been the focus of inspection research. Hence, since meetings increase interval, it was important to determine exactly how meetings contribute to inspections and whether superior alternatives exist.

To help answer these questions, I examined approaches for inspecting software, both with and without meetings (In the meetingless inspections, the formal meeting is replaced with additional individual analysis). I hypothesized that inspection methods that eliminate meetings are at least as cost-effective as methods that rely heavily on them, and probably more so. I expected to see this result because I believed the benefit of additional individual analysis to be greater than that of holding inspection meetings.

To evaluate these hypotheses, I designed and conducted a controlled experiment with 21 graduate students in computer science (Spring 1995), and with 27 professional software developers (Fall 1996). I discovered that the meetingless inspections found more defects than those with meetings. I also found that the meetingless method performed the best solely because it found more new faults in the second phase (normally the second phase is a meeting). Finally, I found no convincing evidence that inspections with meetings found specific faults more often than the meetingless inspections did.

Although meetings are thought to be essential to successful inspections, these results suggest that they may not find more defects than individuals working alone. Still, meetings do suppress false positives and allow groups to make decisions and coordinate activities. Therefore, inspection meetings may be beneficial, but they can not be justified on the basis of defect discovery alone. These results are consistent with later studies by Johnson and Tjahjono and Lau et al.

In fact, Philip Johnson and I have shown that these results are, at the level of hypothesis testing, identical to those of Johnson and Tjahjono (using a simplified form of statistical meta-analysis). This joint work is novel because it shows a method for and gives an example of how multiple studies can be compared, even though they were designed and conducted independently, had different subjects, and different experimental tasks.

In a separate effort, undertaken with Perpich et al, I conducted an industrial case study of about 3000 inspections in which meetingless inspections performed as well as inspections with meetings.

The most complete descriptions of this work appear in Porter and Johnson.