Statistics at the University of Minnesota
Glen Meeden, with Martha Coventry
School of Statistics, University of Minnesota
The University of Minnesota sits on the banks of the Mississippi River, not all that far from where the river begins. A land-grant university, it was founded in 1851 to serve the "sons and daughters" of the state and further Minnesota's progress as it made the most of its natural resources—trees, minerals, rich soil, and the Mississippi itself.
In the 1930s and '40s, Minnesota was not so different from most other universities when it came to statistics. Many departments, like educational psychology, economics, mathematics, public health, and business, engaged in some statistics teaching and research. The University's Graduate School even offered a statistics PhD if a student could cobble together the right courses to get the plan approved. But there was not a school or department dedicated to the field.
As the 1940s gave way to the 1950s, faculty members who worked in statistics began lobbying for a department of their own. In 1958, the College of Liberal Arts (CLA) finally established the Department of Statistics. Minnesota's department was one of several that were founded around that time and went on to become leaders in the field, like Berkeley (1955), Harvard (1957), and Rutgers (1959).
The department hired some strong faculty members during the 1960s and early 1970s, like Ingram Olkin and Richard Savage. Three who spent a significant amount of time in the department were Milton Sobel; Professor Emeritus Bernard Lindgren who came from the Minnesota mathematics department—then and still a departmental ally; and Robert Buehler. Somesh Das Gupta joined in 1967, William Sudderth in 1969, Donald Berry in 1970, and Morris Eaton in 1971.
The chairs of the still-small department were often recruited from within the University, rotating in on loan from other disciplines. Palmer Johnson, the first chair, was in educational psychology and Leonid Hurwicz, future Nobel Prize winner, came from economics. Bernard Lindgren became chair in 1963.
In 1969, the School of Agricultural and the Agricultural Extension Service in St. Paul pressured the administration for their own statistics department to serve their consulting needs, and they got it. The first three hires were Frank Martin, Dennis Cook, and Kinley Larntz. Stephen Fienberg came on as chair and stayed at Minnesota until 1980.
The administration decided that the statistical presence in St. Paul should be officially allied with the still relatively new Department of Statistics. The solution was to disband the Department of Statistics and establish the School of Statistics, consisting of the Department of Theoretical Statistics in Minneapolis and the Department of Applied Statistics in St. Paul.
Dividing the new School of Statistics by discipline and geography worked reasonably well. The arrangement was probably most bothersome for graduate students who, at least three days a week, would have to travel to both locations. The free shuttle buses between them ran fairly often, but the ride was long. Students' memories of those days always include racing for the bus or waiting in the snow for the next one to arrive.
The professors in the applied department divided their workload 50/50 between teaching and consulting. That meant teaching one course per quarter and then being available to any faculty at the University for consulting help the rest of the time. This close relationship of the applied department with other fields was beneficial not only to their clients, but also to members of the department. For example, Dennis Cook's development of Cook's distance was stimulated by a consulting problem brought to him from the Animal Science Department.
"It was a wonderful set up," says Christopher "Kit" Bingham, professor emeritus. "Over the years, I got involved in all sorts of great research with medical doctors, economists, musicians, food scientists, soil scientists, conservation biologists, and veterinary biologists."
Although funding for consulting from CLA and the Agricultural Experiment Station slowly dried up, consulting remains a need at the University. Currently, the Statistical Consulting Service, led by Aaron Rendahl, a recent PhD graduate of the School, offers services in St. Paul and around the University, but for a fee. Many of the School's current faculty members continue to be involved in long-term consulting projects. For example, Birgit Grund has worked for many years on a large scale AIDS research project funded through the University's School of Public Health.
Consulting is also alive and well among graduate students, and the experience is a requirement for all second-year PhD students, who must complete a consulting project during their second summer.
When the University administration was setting up the applied and theoretical departments, it realized that the School would need a director to coordinate its activities. In 1971, it chose Seymour Geisser to fill this position. Geisser came from the University of Buffalo and was making his name as a brilliant and independent thinker. A Renaissance man of prodigious curiosity, Geisser dropped Latin into a conversation, enjoyed reading history, and became an expert on DNA statistical analysis.
"Seymour was an able administrator and a great scholar," says Glen Meeden, who succeeded Geisser as chair. "He quickly got the School up to speed and kept it there for 30 years."
Geisser ran up against the constraints of a public university budget, which waxed and waned with fluctuating state support, and he would push the liberal arts deans as hard as he could to get what he needed for the School. Despite financial limitations, he hired an impressive array of faculty members and he drove them to excel. One young professor, upon returning from a sabbatical, stopped in to see him. Geisser asked, "Well, how'd it go?" The professor replied, "I worked hard and kept busy." Geisser paused a second, then asked, "But did you accomplish anything worthwhile?"
After nearly 30 years of travelling between Minneapolis and St. Paul, many faculty members began hoping for a united School of Statistics. They felt graduate students were spending far too much time shuttling between locations, most undergraduate classes were in Minneapolis, and bringing the two faculties together could result in improved class size distribution and course offerings. Perhaps, they said, coming together could also lead to closer collaboration among faculty.
During the late 1990s, the University undertook badly needed renovations. One building that benefitted was the Art Deco-style Ford Hall on the Minneapolis campus mall and the administration gave the School of Statistics the third floor and part of the fourth floor. When the School moved into its new home, the administration formally disbanded the theoretical and applied departments. All faculty were now just part of the School of Statistics, and under one roof.
Geisser was a leading Bayesian who emphasized the prediction of observables rather than estimation of unobservable parameters. His standard question to the speaker at the end of a School of Statistics seminar was, "Well, what does this have to say about prediction?"
Geisser had come to Minnesota when Bayesian decision theory was out-of fashion in the United States and the national statistics community was not very welcoming of the approach.
There were Bayesians on the faculty when Geisser arrived, including Robert Buehler, William Sudderth, Donald Berry, and Morris Eaton. Over the years, Geisser hired additional Bayesians David Lane, Luke Tierney, Kathryn Chaloner, Jim Dickey, and Glen Meeden.
But although Geisser was a committed Bayesian, he further added to an early group of non-Bayesians, which included Dennis Cook, Kinely Larntz, Sanford Weisberg, Frank Martin, and Kit Bingham. New hires David Hinkley, Gary Oehlert, Douglas Hawkins, and Charles Geyer were frequentists who flourished at the School.
For more than 30 years, Cook and Weisberg have been leaders in the theory and application of regression analysis and graphics. Hawkins has been at the front of diagnostics and quality control. Geyer is an expert in Markov chain Monte Carlo methods. Recently he developed the R packages "Aster," which is meant for biologists wishing to do life history analysis.
The interaction between and reconciliation of frequentist and Bayesian approaches to statistical inference have been important themes in some of the faculty's research. Robert Buehler's foundational work, Meeden's Bayesian approach to finite population inference, the formal Bayes/decision theory work of Galin Jones and Morris Eaton, and Bill Sudderth's finitely additive approach to Bayesian inference have all contributed to a better understanding of this foundational relationship.
In 1972, the state of Minnesota had the largest educational computing network in the world, stretching from Moorhead in the far northwest to Rochester in the southeast. Called the MERIT system, it transmitted 10 characters per second on teletype machines, which were used as command-line interfaces with the mainframes. The University's computer lab in Ford Hall had five or six teletype machines that tapped into MERIT and the School of Statistics began to integrate computing into its courses.
Soon the University bought its own central computers and the School faculty wrote several innovative, interactive packages. Kit Bingham, for example, created an early interactive program called MATTER. Gary Oehlert joined the department in 1984 and soon after, he and Bingham began to develop MacAnova, a free open source statistics package.
Faculty members Luke Tierney, Dennis Cook, and Sandy Weisberg were interested in developing statistical software packages, like Lisp-Stat and Arc. Tierney's knowledge of computing was much deeper than most statisticians and quite unusual for then, and now. His expertise was crucial in developing the School's strength in computing.
Recently, statistical software development has turned more and more to R packages hosted on CRAN rather than stand-alone software. The School of Statistics faculty and students have contributed at least 16 CRAN packages in the past few years. Charles Geyer has been a leader in teaching students and faculty members the intricacies of R and how to write software packages. A few years ago when there was a problem with the computers and Geyer was not around, one of the graduate students said, "How can the School expect us to get any work done when Charlie is not here to make sure everything is working?"
From the beginning, the School offered courses for its own graduate students and for students from other departments. For years, it has taught introductory courses for undergraduates, some of who then become statistics majors. Like most U.S. statistics departments, Minnesota has never had a large number of undergraduate majors. It tended to average only 40 or so each year. Recently, however, the number has grown to around 80, in part because of an influx of international students.
Since most of the University's students live off campus the School has a web server with R web installed. This allows students with Internet access to use R over the web. In addition, the School has developed an on-line web-based system for grading homework. It allows each student to get his or her own version of a particular problem, which is then automatically graded. Although the system can only handle routine problems, it is a way to provide timely student feedback in an era of declining resources.
The School has hired more than half its faculty members in the last dozen years. The following are some brief highlights of their research:
Peihua Qiu's work has been in two main areas. The first is in jump regression analysis, which concerns regression modeling when the regression function has jumps or discontinuities. The second is in quality control where Qiu, along with Doug Hawkins and others, have developed changepoint methods for quality control problems.
Tiefeng Jiang's research focuses on two things: searching for similarities among two or multiple protein or DNA structures in three-dimensional spaces, and exploring random matrices and their relationship to statistics, mathematics, and statistical physics.
Galin Jones is interested in Markov chain Monte Carlo methodology with an emphasis on studying convergence rates of Markov chains. He is also interested in Markov chains in decision theory, hierarchical models and applications of statistical methodology in biological, agricultural, and environmental settings.
Some of Snigdhansu Chatterjee's work deals with climate change statistics and would be nearly unimaginable 20 years ago. He works with a large team consisting of climate scientists, hydrologists, and computer scientists.
For large and complicated problems arising from massive data sets, Lan Wang has developed lack-of-fit tests that can be used for model checking and some novel model selection methods. One important area she has worked in is quantile regression.
Xiaotong Shen and Hui Zou have been working on supervised, semisupervised, and unsupervised learning problems using regularization, where data with and/or without labels have been processed through classification, clustering, and classification/clustering models with prediction as a goal.
Yuhong Yang has been working on theories and applications of model selection and combining statistical procedures, especially when a large number of models are involved. He has also studied optimal data splitting ratio in cross validation and has contributed to understanding the possibilities and limitations of statistical learning.
Adam Rothman is the School's only assistant professor and joined the faculty in the fall of 2010. His interests are in theory, methodology, and computational algorithms for statistical problems involving high-dimensional data. Specific topics include multivariate analysis, machine learning, and efficient optimization.
Over the past 35 years, the School has maintained its international reputation as a center for teaching and research. This is evidenced not only by faculty publications and the School's PhD graduates, but also by the professional recognition afforded the faculty. Stephen Fienberg was coordinating editor and applications editor of the Journal of the American Statistical Association, David Hinkley served as editor of the Annals of Statistics, and Morris Eaton was co-editor of the Annals of Statistics. Nearly all of the current and former faculty of the School are or were fellows of the American Statistical Association or the Institute of Mathematical Statistics.
There are a variety of noteworthy faculty activities that have added to Minnesota's reputation for excellence and innovation. Certainly one is the publication of advanced level books. These include a multivariate analysis book by Morris Eaton; a prediction book by Seymour Geisser; a number of regression and graphics books written jointly and separately by Dennis Cook and Sandy Weisberg; a book on quality control by Douglas Hawkins and David Olwell; a finite population sampling book by Malay Ghosh and Glen Meeden; a book on gambling theory by Ashok Maitra and William Sudderth; an experimental design book by Gary Oehlert; and a book on jump regression by Peihua Qiu.
At the School's 25-year celebration in 1996, past students were the only speakers. Now the School is commemorating its 40th year. It has again invited former students to come and speak about their work. With the strong nucleus of younger faculty, the School should remain a vibrant center for the practice and development of statistics, with the expectation that future celebrations will feature former students who will bring perspective to the past and insights into the future of the field.