TDD Effects: Are We Measuring the Right Things?

Scientific studies about the impact of Test-Driven Development (TDD) start to appear since 2002, resulting in approximately 30 papers until now [6], [7], [3]. In general the two main evaluated hypothesis are the ones stated by Kent Beck[1]: that TDD produces code with less defects (external quality) and that it produces code that is simpler, less coupled and more cohesive (internal quality). Although studies may suggest good results in term of external quality, it does not conclude too much regarding internal attributes. Common difficulties, like controlling the experiments variables, are generally considered in the studies. But the few conclusions may be result of a bigger problem: we may have adopted wrong hypothesis or assumptions about the practice's benefits.