Notes on “The Confounding Effect of Class Size on the Validity of Object-Oriented Metrics”

October 2, 2008

The Confounding Effect of Class Size on the Validity of Object-Oriented Metrics: Khaled El Emam, Saida Benlarbi, Nishith Goel, and Shesh N. Rai.

The gist: This is a complex paper. First, the authors review a number of published object-oriented complexity metrics and their relation to defects. Along the way, they jab at object-oriented programming in general. Second, the authors report the results of a statistical experiment, from which they conclude that code size is a confounding factor which destroys the external validity of all the previously mentioned metrics.

The gist, remix: Some real scientists take a look at a software engineering research niche, and kick its ass

The good: This is a great paper. I loved the quantitative pokes at OO. They lay out their scientific experiment carefully and clearly. I am persuaded.

The bad: I had difficulty following some of the advanced statistics. Simpler writing would have made it easier to follow logic of the paper. As is, it put me to sleep.

The ugly: This is a “Well, duh” paper. Bigger programs are more complex and have more defects. This is not a surprise. As a program solves more complicated problems, code size and code complexity increase together. The authors distinguish code size from other complexity metrics, but I see them as effectively equivalent entities at different levels of abstraction.

%d bloggers like this: