Academic

Keiran Healy, of Crooked Timber and Duke Sociology fame, discusses his production possibility frontier on the Set Up:

Doing good social-scientific research involves bringing together a variety of different skills. There's a lot of writing and rewriting, with all that goes along with that. There is data to manage, clean, and analyze. There's code to be written and maintained. You're learning from and contributing to some field, so there's a whole apparatus of citation and referencing for that. And, ideally, what you're doing should be clear and reproducible both for your own sake, when you come back to it later, and the sake of collaborators, reviewers, and colleagues. How do you do all of that well? Available models prioritize different things. Many useful tricks and tools aren't taught formally at all. For me, the core tension is this. On the one hand, there are strong payoffs to having things organized simply, reliably, and effectively. Good software can help tremendously with this. On the other hand, though, it's obvious that there isn't just one best way (or one platform, toolchain, or whatever) to do it. Moreover, the people who do great work are often the ones who just shut up and play their guitar, so to speak. So it can be tricky to figure out when stopping to think about "the setup" is helpful, and when it's just an invitation to waste your increasingly precious time installing something that's likely to break something else in an effort to distract yourself. In practice I am only weakly able to manage this problem.

There's more here, and I thank the Harvard Social Science Statistics Blog for the pointer.