Cleveland Chapter
of the ASA

About the ASA

Regular Meetings

Conferences

K-12 Programs

Officers

Membership Info

Links

Chapter News

October 29, 2007

Kernel Methods In A Regularization Framework For Nonparametric Model Building: Yoonkyung Lee, The Ohio State University

Regularization methods for model building and prediction are popular in statistics and machine learning. They may be viewed as the procedures that modify the maximum likelihood principle or the principle of empirical risk minimization. In particular, methods of regularization in reproducing kernel Hilbert spaces provide a unified framework for nonparametric statistical model building. Examples include smoothing splines and support vector machines.

In this workshop, kernel methods are explained with examples, and some issues of model selection and computation for the practical implementation of the methods are discussed. Applications to genomic data for building a medical diagnostic algorithm and selecting relevant biomarkers will be presented, as will marketing data for finding predictive demographic factors.

December 7, 2006

The Design of Industrial Screening Experiments: Angela Dean, The Ohio State University

Screening is the process of using designed experiments and statistical analyses to sift through a very large number of features, such as factors, genes or compounds, in order to discover the few features that influence a measured response. In current research, screening methods are actively being developed for the detection of factors which have a substantial effect on the average response or response variability in a complex system. In particular, the design and analysis of supersaturated and group screening experiments has been shown to be effective for this purpose and much research has recently been done in this area.

The workshop will discuss recent work on the construction of supersaturated designs as well as methods of analysis. Various types of two-stage screening experiments and their uses in searching for active factors will be discussed, and a description given of a recent group screening experiment that was run successfully at Jaguar Cars. A comparison between the methodology of supersaturated designs and two-stage group screening will be presented.

December 15, 2005

Hands-On Bayesian Data Analysis Using WinBUGS: William F. Guthrie, National Institute of Standards and Technology

This workshop is designed to provide statisticians, scientists, or engineers with the tools necessary to begin to use Bayesian inference in applied problems. Participants in the course will learn the basics of Bayesian modeling and inference using Markov chain Monte Carlo simulation with the open-source software package WinBUGS.
The workshop will introduce some of the theory underlying Bayesian analysis, but will primarily focus on Bayesian analysis of "real-world" scientific applications using examples from collaborative research with NIST scientists and engineers. Topics discussed will include Bayesian modeling, Markov chain Monte Carlo algorithms, convergence tests, model validation, and inference.

October 11, 2004

Using Propensity Score Methods Effectively: Thomas E. Love, Case Western Reserve University and MetroHealth Medical Center

This course is designed to provide a friendly, applied and practical survey of propensity score methods used for dealing with selection bias in observational studies of exposure effects. Propensity score methods are applicable whenever treatment or policy decisions are of interest, and numerous examples and illustrations will be presented from a variety of subject areas drawn from published articles in biostatistics, education and public health research and from the speaker's experiences working with industrial clients in insurance, market research and management consulting.

October 27, 2003

Analysis of Curves: Esteban Walker, University of Tennessee

Advances in technology have dramatically increased the amount and quality of data that are recorded in all areas of human endeavor. Thousands of measurements are available nowadays in situations where previously only a few measurements, at given points in time or space, were taken. These measurements allow the reconstruction of the whole profile or "signature". Basically, the profile becomes the unit of analysis. This seminar will discuss two problems with profiles: (1) how to determine if predetermined sets of curves are different, and (2) how to identify clusters in a set of curves. Examples from various fields will be presented. Instructions on the implementation of these techniques in S-Plus and SAS will be given.

October 14, 2002

A Review: Missing Data: Joseph Schafer, Pennsylvania State University

Statistical procedures for missing data have vastly improved in the last two decades, yet misconception and unsound practice still abound. In this seminar, we will

• frame the missing-data problem from a statistician's viewpoint
• introduce fundamental concepts regarding the distribution of missingness and missing at random (MAR)
• review older, ad hoc procedures including case deletion and single imputation, discussing their merits and weaknesses
• discuss the theory, implementation and use of maximum likelihood (ML) for incomplete data problems
• introduce the idea of multiple imputation (MI), discuss its properties, and demonstrate its use on a real data example
• discuss some of the latest developments in the missing-data literature that have appeared in the last five years, including weighted estimating equations and methods for handling missing values that are not MAR.

October 3, 2001

Experiments: Planning, Analysis, and Parameter Design Optimization : C. F. Jeff Wu, The University of Michigan

This seminar will be based on the book "Experiments: Planning, Analysis, and Parameter Design Optimization" by Jeff Wu and Mike Hamada (2000). Course notes will be made available to attendees. This book contains many new methods not found in existing textbooks, and covers more than 80 data sets and 200 exercises. The new tools covered include robust parameter design, use of minimum aberration criterion for optimal factor assignment, orthogonal arrays of economic run size, analysis strategies to exploit interactions, experiments for reliability improvement, and analysis of experiments with non-normal responses. Data from real experiments will be used to illustrate concepts. Time will be reserved for questions and discussion.

May 2, 2001

Permutation Tests: A Guide for Practicioners: Phillip Good, Information Research

This is a one-half day course on permutation methods. It is intended for practicing statisticians and others with interest in applying statistical methods. High school algebra is assumed but no higher-level mathematics is required. Some familiarity with computer simulation would also be helpful. Attendees will be given historical background on resampling methods and a formal introduction to these methods. Emphasis will be placed on the wide variety of applications of the techniques, the computer-intensive nature of implementation along with many examples and "real world" applications. The course is intended for those who use statistical methods in their work. This includes practitioners in medicine, business, engineering and the social sciences. It also will be useful to professors of statistics and those who do statistical research but may not be familiar with resampling methods and want to be updated on the latest advances in methodology and application. Dr. Good is the author of two popular texts on resampling methods. Resampling is a powerful technique which has only recently seen an explosion in applications due to enhancements in computational techniques that make these computer-intensive methods practical.

October 4, 2000

The Grammar of Graphics: Designing a System for Displaying Statistical Graphics on the Web: Leland Wilkinson, SPSS, Inc.

"The Grammar of Graphics" (GOG) is the title of a recent Springer-Verlag book that encapsulates a new theory of statistical graphics. GOG is based on an algebraic model. It contrasts with the prevailing view toward classifying and analyzing charts by their appearance - a view that one might call Pictures of Graphics (POG). In POG, there are pie charts, bar charts, line charts, and so on. Not only are most books and papers on graphics organized by chart type, but so also are most charting programs and production libraries.By contrast, GOG begins with a strong statement: there is no such thing as a pie chart. A pie chart is a stacked bar graphic measured on a proportional scale and transformed into polar coordinates. Significantly, the description of simple charts in POG (such as a pie) are more complex in GOG and seemingly complex charts in POG (such as scatterplot matrices) are simple to describe in GOG. This contrast between surface POG descriptions and deep GOG specifications exposes not only previously unappreciated subtleties in the structures of common charts but also the existence of charts not generally known. GOG is ideally suited for designing a system for interactive Web graphics. Examples will be shown using a Java production library called nViZn (formerly GPL).

November 3, 1999

Logistic Regression: Mike Kutner

Data are said to be binary when each outcome falls into one of two categories such as alive or dead, success or failure, true or false. Binary outcome data occur frequently in practice and are commonly analyzed using the logistic regression model. This workshop will emphasize the practical aspects of modeling binary outcome data using logistic regression, including checking the adequacy of the fitted models. Several examples from the health sciences area are presented to illustrate the techniques. Parallels between multiple linear regression modeling and multiple logistic regression modeling will be presented throughout the workshop. Therefore, attendees of this workshop are expected to be familiar with multiple linear regression modeling techniques.

September 18, 1998

Survival Analysis Extended to Recurrent Events Data with Repairable Products, Disease Recurrences and Other Applications: Wayne B. Nelson

Analysis of recurrent events data is an emerging area of survival analysis whose prominence is increasing due to its many and varied applications. Most life-data courses deal with a single endpoint, end of life or failure, which is modeled with a life distribution that must be estimated. In contrast, recurrence data are modeled with a stochastic process. This Workshop provides a simple nonparametric model and data plots and analyses including point estimates, confidence intervals, and comparisons for populations with recurrent events. This recent methodology is often more appropriate than parametric methods based on a nonhomogeneous Poisson process, which depends upon often unrealistic assumptions, such as independent increments.

October 5, 1996

Teaching Statistics to Non-Statisticians: A Panel Presentation

This year's chapter fall workshop will feature five of our chapter members who have taught statistics to nonstatisticians. They will describe the specific topics in their courses, what software they use, texts, audio-video materials, and demonstrate statistical devices such as the quincunx, black box, helicopter, and Deming funnel. They will focus on the details of what has worked and what hasn't worked.

The workshop presenters were:

• Susan Cowling, The Lubrizol Corporation
• Dennis Keller, RealWorld Quality Systems
• Shari Medendorp, The Cleveland Clinic Foundation
• Mukul Mehta, Quality Sciences
• Gary Skerl, ICI Paints
• Jeff Witmer, Oberlin College (discussant)