Analyzing and Interpreting Test Results - Software Testing: Testing Across the Entire Software Development Life Cycle - page 197

The Rayleigh curve is a well-documented mathematical formula in a series
of engineering formulas called the Weibul distribution. Although purely abstract
in nature, the Weibul distribution seems to predict surprisingly well the behavior
of a number of physical phenomena ranging from the fl ow of rivers to the bounce
of ball bearings to the discovery rate of software defects. In the early 1990s, en-
gineering researchers applied the Rayleigh curve to software development project
defect discovery rates and found to their delight that the Rayleigh curve was more
predictive than any other method attempted to date. [48] Prediction accuracies
within

10% were seen in several large research projects over a 3-5-year de-
fect discovery comparison. Because the Rayleigh curve is pure mathematics and
the actual defect discovery curve is purely behavioral (developer processes and
skills, tester processes and skills, development and testing environments, tools,
and so forth), then the key to successfully using the Rayleigh curve on your project
is to also apply a healthy dose of judgment, experience, and intelligence.
Here is one approach found effective in applying the Rayleigh curve to your testing
results by comparing it to the defect discovery curve plotted during your most recent proj-
ect. For the purposes of demonstration, consider Figure 12.15 to be your prior project.

/

Figure 12.15

Prior development project defect tracking log

The prior project curve has a familiar contour although it is not as smooth as
the idealized curves in previous examples. The main peak of 749 defect discoveries
occurs during week 10 of the project. There are a couple of secondary peaks that
appear at week 6 and week 15. These peaks are expected because the testing team
doubles their effort twice: at the end of the Preliminary construction phase and again
when all of the software components are integrated for the fi rst time during the Final
construction phase. Because these peaks are so pronounced and because they appear
very regularly across several prior projects, the testing team is convinced that their
testing has maximum effectiveness.
To compare the prior project defect discovery curve with the Rayleigh curve,
fi rst place an imaginary gunsight on the prior project defect discovery curve at the
discovery peak as in Figure 12.16.