I obtained my Ph.D. from the University of California at Los Angeles in 1976. Prior to coming to the University of Missouri-Columbia in the fall of 1982, I taught at the Unversity of Oregon. I spent the 1995-96 academic year on sabbatical at the Australian National University, North Carolina State University, and the National Institute of Statistical Sciences.

My general reseach interests lie in areas of theoretical and computational statistics. I have long been interested in topics in nonparametric curve estimation, smoothing splines, and semiparametric models. In addition to my theoretical work in these areas, in recent years I have been worked on a number of interesting applications including risk estimation and longitudinal data analysis. In addition, I became involved in several projects at NISS which are ongoing areas of research.

To support my work in computational statistics, I have been actively involved with my colleague Dongchu Sun in building and maintaining the departmental computing network of Sun workstations. Our expanding network has been vital in ongoing research in the department in nonparametric regression, Bayesian statistics, wavelets, isotonic inference, time series and reliability.

Finally, the department is delighted to have been selected to host a NSF/CBMS conference on Longitudinal Data Analysis featuring Nan Laird as the principal speaker, June 10-14, 1997. As the organizer, I hope you will visit our web site and consider attending!

Nonparametric Analysis of Covariance In an article to appear in JASA, Peter Hall, Catherine Huber and I derived a new method for assessing differences in treatment groups in the presence of a single covariate when the usual hypotheses are suspect. Bootstrapping contributes to the small sample performance, and simulation shows that the method approaches its asymptotic superiority to the classic test statistic for moderate sample sizes. Current work focuses on extending to multiple covariates and more complicated models.

Additive models in Longitudinal Data Analysis Motivated by an experiment monitoring milk production of cows over three summer months, doctoral student Sungwook Lee and I are studying studying additive models for longitudinal data. Moyeed and Diggle extended my 1988 JRSS B approach to estimation in partially linear models to the case of correlated data. We are currently studying other approaches to the problem and analyzing situations with more than one semiparametric term.

Spatial Interaction Models The term "spatial interaction model" is used by transportation researchers to describe models for forecasting travel behavior. For example, a probabilistic model might be constructed to predict the type and location of activities in a day for a person in with specific demographic characteristics living in one part of a city . Working with Eric Pas in the Engineering School at Duke University, Kenneth Vaughn in the Department of Transportation, and Dongchu Sun at MU, I have developed a new model for activity location choice. The new model is related a particular discrete log linear model, known as the gravity model, frequently used by transportation engineers, but we have been able to incorporate spatial relations in a continuous version. This new approach relies on a log additive density function with nonparametric terms. We have developed a new density estimation method to fit the model to data from a travel survey in Portland, Oregon. In the future, the model will be used to construct simulated travel patterns for driving a massive traffic computer simulation being developed at Los Alamos.

Ozone forecasting There is a long history of statistical methods in meteorology forecasting. In particular, the problem of forecasting daily maximum ozone has received considerable attention because of EPA standards mandated for cities. Working with Jerry Davis of the Meteorology Department at North Carolina State University, I have developed a quasi-likelihood generalized additive model for forecasting ozone in Houston. Our model is novel in that it is semiparametric and incorporates nonparametric smoothing. Due to the coastal meteorology of Houston, the nonparamatric model performs substantially better than conventional parametric ones. This application in itself is strictly applied, but problems we encounter under certain meteorological conditions have suggested an entirely new theoretical approach based on a wind-based coordinate system.

Click on icon to go back to the department homepage.