It would be nice if it would be true, the graph looks great and props for supplying the data. Correlation does not equal causation though. Are there any scientifics researches on TDD or any programming styles?
The IBM research is interesting, though it seems as anecdotal as the OP, it also has the 50% reduction in faults, perhaps if there's a few more cases like this a metastudy could yield some conclusive evidence.
The problem with this study and many others that report on TDD's benefits is that they don't compare the results with a test group that writes unit tests after the fact. So, it's not at all clear what they're validating: the benefits of unit tests (which I think are well accepted) or the benefits of writing tests before code. I suspect it's primarily the former.
The research in this paper does not result in any conclusive evidence that says TDD is better, it merely showed that TDD encourages developers to work longer until a higher code quality is attained. If I read it correctly.
It's going to be impossible to quantify "better" anyways. It depends on what you're optimizing for. This research is incredibly valuable in helping us understand the results of TDD, particularly if we're optimizing purely for quality:)