While I'm no advocate of TDD, I think they trivialized the arguments for it. They just measured "productivity" (number of tasks completed) and "quality" (numbe

While I'm no advocate of TDD, I think they trivialized the arguments for it. They just measured "productivity" (number of tasks completed) and "quality" (number of tests that passed). The primary argument I've heard for TDD is modifying an existing system. You have a lot of tests that all successfully pass, so you can modify the system without fear of breaking it without knowing you did. Another reason is that the system becomes self documenting, making it easier to understand for people in the future. This study didn't really attempt to tackle those. Also, the tasks the programmers were assigned were tested for "quality" and completeness automatically. So it makes me think the tasks were of the type that favor being tested automatically, and thus biased towards benefiting from TDD.