For every [person], the world is as fresh as it was at the first day, and as full of untold novelties for him who has the eyes to see them.
— Thomas Henry Huxley, A Liberal Education (1868)
Q: Statistics, eh? So what's that, you go to baseball games? What do you do in the off season?
A: At $6 a hot dog, no way. And it's not about baseball anyway. (Not that there's anything wrong with that.) It's about uncertainty, my friend. Uncertainty.
Q: Well if you've got such a handle on this uncertainty thing, why are you writing this web page instead of sitting on an island somewhere, rich and happy?
A: You mean with the baseball and football players? Uh, not a bad way to go, I admit, but where's the fulfillment, the challenge, the grappling with big questions and big ideas, the struggle to make the world a better place? You won't find that on a beach sipping mai tai's.
Q: Riiight. Got it. So where do you get your fulfillment?
A: Understanding new ideas, solving problems, writing and design, drawing insights from data.
Q: Sounds like fun, I suppose, if that's what gets you going. But can you get more specific here?
A: See those tabs at the top of the page? Click on those for more details.
Q: Will do, chief. But for now, just give me the quick tour.
A: OK, a quick one; I've got a meeting. The motivation for much of my research is the challenge of accurately assessing uncertainties in complex scientific problems. I'm particularly interested in what is called the nonparametric case, where the object of inference is infinite-dimensional — a function.
Q: What makes that especially interesting?
A: There's a lot of room in infinite-dimensional space. Care is needed if the assumptions and inferences are going to be well-tuned to the objects of interest. It's easy to understate the uncertainties; it's easy to overstate them as well.
Q: Where does that lead you?
A: Several directions, really. As a problem in mathematical statistics, we now know a great deal about the performance of nonparametric estimators, but many open questions remain about nonparametric inference: confidence sets and related ideas. I'm actively working on these questions. These issues also crop up frequently in applications, especially in this age of large and high-dimensional data sets. One common approach is for practitioners to reconstruct — estimate — an object of interest from data and then read off from this estimate a variety of inferences. But this is rarely a good approach. First, the estimate by itself does not tell us about the level of uncertainty. Second, even when pointwise uncertainties are computed for the reconstruction, we can usually do better by making inferences about quantities that directly relate to the questions being asked.
Q: What kind of applications?
A: I do a lot of work in astronomy and cosmology, neuroscience, and more recently have begun working on problems in evolutionary biology. In each case, I've built a collaboration with working scientists, which is, by itself, a rewarding enterprise.
Q: And does this come out in your teaching?
A: As much as possible. I try to devise realistic and layered examples in all my classes, whether drawing experiments from the literature or exploring applicaitons from my research.
Q: What else drives your teaching?
A: I pay substantial attention to incentives for learning. I also experiment freely with new pedagogical ideas and techniques.
Q: Incentives, eh? Such as paying the kids to study?
A: (Smiling) Not quite, no, though there's some evidence that works. Look, the current system is filled with bright, capable people and produces good work at all levels. But the incentives for learning are seriously messed up. Consider failure, for example. Failure is an important part of the learning process, but students have no real opportunity to fail in a constructive way without potentially substantial consequences. Homework should be a mechanism for meaningful practice with the ideas in a course, but you're caught between two hard places. Put too much weight on it, and it becomes more important to students to finish it correctly than to understand it. Put too little weight on it, and the other demands on a student's time dominate practice.
Q: What have you done about that?
A: A variety of different approaches with various levels of success. My favorite is the Mastery Exam system I developed. More on that under Teaching.
Q: And what do you do outside work?
A: Lots of projects with the kid. We're building robots and doing card tricks lately. Astronomical observing. I study and raise insects. Reading and writing, blogging too, though I've been quiet lately. Other things at shorter or longer time scales, something of a random walk.
Q: Not much on the blog recently. Maybe you've really given it up but not admitted it to yourself.
A: Hmmm. Don't think so. I keep hoping to get back to it more regularly, but there's been too much going on. I'll keep my options open.
Q: What about sleep?
A: Not so much. That's not soo good.
Q: Nice talking to you. I'll take a look around.
A: And you as well. Welcome, and enjoy.