| I haven't read Bossavit's book. I have read McConnell's writings. Chapter 30 of "Making Software: What Really Works, and Why We Believe It" is by him, and edited by Andy Oram and Greg Wilson. The point of the book was to collect scientific evidence. I know Greg, and his high standards, so I strongly doubt that McConnell's evidence of weak. I'm immediately suspicious of the statement you quoted: "Not a single one of the references is to a replication, in the scientific sense of the term." Strict replication is not and never has been a requirement to good science. As a trivial example, how do you replicate a supernova observation? You can't. But you can make models and test the models against the observed data and against future supernovas, and you estimate the reasonableness of the model effectiveness. If that quote is indicative of Bossavit's views, then he is a poor judge of what's good evidence. Going on with more quotes you gave, "which we can safely expect to have wrought major changes in programming practice." That is a conjecture, not a statement of fact. It seems like he's saying that once upon a time there was 10x difference, but the Internet makes everything different now there isn't. However, without evidence to back it up, I don't know why I should believe it. What tests has Bossavit carried out to show that this is the case? To the contrary, L. Prechelt "An empirical comparison of C, C++, Java ..." was both carried out during the internet era, was done with good rigor, and shows a large spread in the overall time to complete the project. That's a small project, granted, but it's another data point in the overall trends. (If you don't like the small project size, another example, given by McConnell, is the large productivity difference between the Excel and Lotus 1-2-3 teams, which produced roughly comparable spreadsheet programs but with about an order of magnitude productivity difference across several different metrics.) Multiple independently determined data points which all trend in the same direction strengthen the likelihood that the ~10x hypothesis is a useful model. That's what scientific evidence looks like. McConnell even has a section in his chapter on "how meaningful it is to speak of an individual programmer's productivity", along with all of the caveats and cautions that everyone here is bringing up (some programmers produce negative lines of code, the best programmers tend to get the hardest problems, the effect of teams, etc.), and concludes "So while I see the value in measuring individual performance in research settings, I think it's difficult to find cases in which the measurement effort is justified on real projects." Edit: I see that Bossavit wrote in response to McConnell's chapter. McConnel's reply to Bossavit is at http://forums.construx.com/blogs/stevemcc/archive/2011/01/09... . |
Rather than try to defend someone else's words, though, I'll just direct you to the book. You're going into a lot of detail critiquing quotes from an end-of-chapter summary, which isn't fair to Bossavit's work. I hope you'll read the book, because I'd like to hear your critique of the full work.