The authors present a counter proof for the “Goldilocks conjecture,” which states that there is an optimum size for software components. According to the Goldilocks conjecture, components that are larger or smaller than this optimum size will have a higher defect density.
The authors show that plotting the defect density against the component size is an incorrect statistical practice, and that the Goldilocks conjecture is a mathematical artifact of plotting a variable against its own inverse. The authors then present empirical evidence, using three independent systems, to show that the conjecture does not hold for object-oriented software, and hence that the process of class decomposition is valid for reducing defects.
Even if you disagree with the authors’ results, the paper presents a very detailed literature survey of what different authors consider the optimum component size in various languages and systems. The paper is definitely worth reading for anyone interested in coding best practices and standards, even if they only read the introduction.