SCHEDULE

ABSTRACTS

Secrets of a Consulting Statistician's Art
J. Sarkar
Indiana University - Purdue University at Indianapolis

What is a consulting statistician? I will answer this question by first demonstrating the thinking process a consulting statistician goes through in solving a problem, by means of a very simple example. Then I will highlight some of the essential characteristics of a good statistical consultant in terms of personal disposition, educational training and career objectives. I will share some of my own consulting experience and what I have learned from them. The talk should be easy to understand by graduate students in statistics.

New Classes of Multivariate Survival Distributions and their Application
Mark Carpenter
Auburn University

In this paper, we proffer new classes of multivariate survival distributions with potentially widespread applications in survival and reliability modeling. These classes include families that with marginal distributions that are either three-parameter gamma or Weibull, among others. In some cases, the distribution is absolutely continuous, but in others the distribution places a positive probability mass on a set of measure zero (similar to the multivariate exponential motivated by Marshall and Olkin). The emphasis in this presentation will be on the real world motivation and application of the results.

Statistical collaboration in medical research
Theda A. Foster
Department of Biostatistics
Tulane University Health Sciences Center

Collaboration in medical research offers the statistician the opportunity to use both theoretical and applied training in statistics. Medical research in basic or clinical sciences poses biological hypotheses related to disease processes that are to be addressed in a particular research study. As a member of the research team, the statistician translates the biological hypothesis into a testable statistical hypothesis. This involves close collaboration with the study's principal investigator in planning the study, developing the experimental design, and quantifying both the variability and the reliability of data to be collected. This planning lays the foundation for valid statistical analyses. Finally, the statistician collaborates with the investigator in interpreting results of statistical analyses in the context of the biological question. Important skills needed by the statistician include the ability to listen to both expressed and unexpressed ideas of the biologist. Additionally, communication skills verbal, written and presentation promote success of the collaboration.

Developing treatment prescriptions for precision agriculture
Renee Wang and Kevin McCarter
Department of Experimental Statistics
Louisiana State University

Precision agriculture equipment allows variable-rate treatment (VRT) applications that can be adjusted according to the changing characteristics and treatment requirements of a field. This is in contrast to broadcast treatment applications, where the same rate is applied across the entire field. If a field has a high level of spatial variability, the broadcast approach can result in an inefficient and ineffective use of the treatment being applied. Since spatial variability of field characteristics increases with the size of the field, variable-rate treatment application becomes increasingly applicable as the size of the field increases, such as in large commercial farming operations. Correct use of VRT application can result in a reduction of input costs, as well as a reduction in the negative environmental impact due to the use of the treatment, and therefore has the potential to increase a producer's profit while helping them become better stewards of the environment. Researchers at the LSU Ag Center are becoming increasingly called upon to deliver variable-rate treatment prescriptions to the Louisiana producers they serve, and they have sought our help in developing tools that will enable them to construct treatment prescriptions in an efficient and timely manner. Developing a treatment prescription is a complicated, time-consuming process involving the use of GIS software, performing statistical analysis, and processing of the results of that analysis to produce the final treatment prescription. In this talk we discuss the process of developing a treatment prescription and provide software specifications for that part of the process which can be automated.

Multivariate repeated measures analysis of mangrove forest litter dynamics along the Shark river estuary in the Florida coastal everglades
Denise Moore
Department of Experimental Statistics
Louisiana state University

The measure of litterfall in mangrove forests is a useful tool in describing ecosystem function and testing hypotheses about productivity under different environmental settings. By using a repeated measures analysis, this study sought to evaluate the spatial (sites) and the temporal (seasonal and inter-annual) patterns in litterfall production (g m-2 day-1) resulting from interannual and long-term changes in freshwater discharge. From January 2001 to December 2004 mangrove litter dynamics were monitored in three sites along the salinity gradient of the Shark River. Litter was separated into three components, leaves, reproductive parts and wood. Since several components of litter were to be analyzed, a multivariate approach was necessary. Though the multivariate analysis can be performed as a general linear model in SAS (GLM procedure), the use of covariance structures is not permitted. Therefore, an alternative mixed model was performed in SAS (Mixed procedure) which utilizes a multivariate repeated measures covariance structure. Results from both analyses are presented and discussed.

A revisit to the common mean problem: comparing the maximum likelihood estimator with the Graybill-Deal estimator
Nabendu Pal
Department of Mathematics
University of Louisiana at Lafayette

For estimating the common mean of two normal populations with unknown and possibly unequal variances the well-known Graybill-Deal estimator (GDE) has been a motivating factor for research over the last five decades. Surprisingly the literature does not have much to show when it comes to the maximum likelihood estimator (MLE) and its properties compared to those of the GDE. The purpose of this note is to shed some light on the structure of the MLE, and compare it with the GDE. While studying the asymptotic variance of the GDE, we provide an upgraded set of bounds for its variance.A massive simulation study has been carried out with very high level of accuracy to compare the variances of the above two estimators results of which are quite interesting.