Statistics Interests

The department of Statistics at Chicago has a very nice logo, which I've borrowed above. You can head there to look at the courses offered by the department, the professors, and all that assorted goodness. The department server, like most of the computers in the department, is named after a famous statistician; here, it's Francis Galton. Galton was at the vanguard of research on experimental design, the control of parameters, which allowed statistical methods to become useful in applied fields such as sociology, economics, and psychology. You can read more about his work in The History of Statistics, ISBN 067440341X, by Chicago professor Stephen Stigler. An interesting side note is that Galton was heavily involved in the eugenics (racial superiority) movement of the era, one of the earliest applications of statistics.

So, what do I research?

One way to categorize the work that a Department of Statistics presents is to divide it into the following three categories.

• Probability, a mathematical discipline in which events are enumerated, measured, and manipulated via free parameters. Often, probability research is performed with little consideration to application or use, instead focusing on describing and mathematicing processes that may or may not exist in the physical world. Probability also includes approximation techniques, stochastic processes, and chains in space, which do have application. Finance and economics have seized upon stochastic (time) processes recently.
• Statistics theoretical, dealing with techniques to evaluate and determine the validity of formulations. This area includes the classical Neyman-Pearson theory, as well as hypothesis testing and Bayes-style tests. One contentious issue here is how to judge the validity of tests, through purely mathematical methods or via some intuitive concept.
• Statistics applied, including fields such as the design of experiments, computational techniques, utilizing techniques from the other categories in reality and industry, and selecting the appropriate information from the available pieces to form models.

My interests lie almost exclusively in the final area above, applied work. Not surprisingly, How does one formulate a good model? Given thousands of possible variables, and perhaps millions of data points, which ones work? My thesis topic is Bayesian Neural Networks and Variable Selection, and this link will get you a copy of the slides I used at my proposal talk in December 2003. My committee consists of advisor Rob McCulloch, Marc Coram, and Vanja Dukic.

Going Forward

At this point, it seems unlikely that I will take an academic position in a Department of Statistics when (When!) I complete my doctorate. It's not because I oppose research, rather that the modern academic position is rarely active enough in the field. In the past, major researchers also did applied research; for example, Fisher had an agricultural job. Even twenty years ago, one of the researchers of the powerful EM algorithm worked at the Educational Testing Service.

This point looms even larger today than in the past, because the concept of numerical analysis has penetrated almost every discipline. Economics quite frankly has become enthralled (in my opinion, too much) with the concept of statistics and econometrics. The work of Deming and Taguchi helps industrial processing; surveying influences advertising; I provided examples in banking and credit scoring; and experiments influence psychology and sociology. I even had a revealing talk with my neighbor recently about the use of T tests at divinity school, which startled me.

Finally, you might have noticed that I didn't describe exactly what a numerical probability means, because it's a point of some controversy. Is a probability set by completely objective trials and frequencies, or does subjective information or belief play a role? This is a philosophical debate, almost a theological one, and well within my interests. Let me summarize by saying that I am primarily Bayesian, in that I have a belief in common sense and subjectivity.

Resources

There was a good opinion piece in the Chicago Tribune on how numbers help us search for truth, available here. I also heard an interview on World News Now from sociology professor Joel Best, about the nefarious and pernicious use of numbers. In other words, to watch out for liars in the search for truth. Prof. Best also wrote an article for the Chronicle of Higher Education, which I present in formatted form on this page. I added some comments at the bottom of that page.

Last updated: September 2004. I'm not particularly exact about time normally, so why start now?