Cleveland Chapter
of the American Statistical Association

About the ASA

Regular Meetings

Conferences

K-12 Programs

Officers

Membership Info

Links

Chapter News

May 9, 2008

Applied Longitudinal Analysis:
Garrett Fitzmaurice

The goal of this seminar is to provide an introduction to statistical methods for analyzing longitudinal data. The main emphasis is on the practical rather than the theoretical aspects of longitudinal analysis. The course begins with a review of established methods for analyzing longitudinal data when the response of interest is continuous. A general introduction to linear mixed effects models for continuous responses is presented. When the response of interest is categorical (e.g., binary or count data), a number of extensions of generalized linear models to longitudinal data have been proposed. We present a broad overview of two main types of models: "marginal models" and "generalized linear mixed models". While both classes of models account for the within-subject correlation among the repeated measures, they differ in approach. Moreover, these two classes of models have regression coefficients with quite distinct interpretations and address somewhat different questions regarding longitudinal change in the response. In this course we highlight the main distinctions between these two types of models and discuss the types of scientific questions addressed by each.

May 21, 2007

Regression Modeling Strategies:
Frank E. Harrell, Jr.

The first part of this course presents the following elements of multivariable predictive modeling for a single response variable: using regression splines to relax linearity assumptions, perils of variable selection and overfitting, where to spend degrees of freedom, shrinkage, imputation of missing data, data reduction, and interaction surfaces. Then a default overall modeling strategy will be described. This is followed by methods for graphically understanding models (e.g., using nomograms) and using re-sampling to estimate a model's likely performance on new data. Then the freely available S-Plus Design library will be overviewed. Design facilitates most of the steps of the modeling process. Two of the following three case studies will be presented: an interactive exploration of the survival status of Titanic passengers, an interactive case study in developing a survival time model for critically ill patients, and a case study in Cox regression. The methods in this course will apply to almost any regression model, including ordinary least squares, logistic regression models, and survival models.

May 15, 2006

Causal Inference in Experiments and Observational Studies:
Donald B. Rubin

This course will present the Rubin Causal Model perspective for statistical inference for causal effects through potential outcomes. There are three parts to the course.
The first part establishes the primitives that form the foundation: units, treatments, potential outcomes, the stability assumption, and the assignment mechanism. 
The second part presents inference based solely on the assignment mechanism; this perspective generalizes Fisher's (1925) and Neyman's (1923) randomization-based approaches, and emphasizes the central role of the propensity score (Rosenbaum and Rubin, 1983), thereby creating a bridge between experiments and observational studies.  
The third part presents inference based on predictive models for the distribution of the missing potential outcomes, formally, Bayesian posterior predictive inference (Rubin, 1978).   In practice, the predictive approach is ideal for creating statistical procedures, whereas the assignment-based approach of Fisher is ideal for traditional confirmatory inference, and the assignment-based approach of Neyman is ideal for evaluating procedures.  
For best practice, being facile with all three approaches is important.  

April 1, 2005

Survival Analysis: Statistical Methods for Censored and Truncated Data:
Melvin L. Moeschberger

In this tutorial session, basic elements of survival analyses will be presented. The quantities used to summarize event time data will be defined. The speaker will focus on nonparametric estimation of these quantities including the Kaplan-Meier estimator. Weighted log rank tests and their use in comparing the survival experience in two or more populations will be discussed. Focus will be on regression techniques for censored data based on the Cox regression model. Testing and model building using fixed and time dependent covariates through a series of examples will be examined. The material covered in the session will be taken from the new second edition of the speaker's book. Methods will be illustrated using medical data and SAS statements will be discussed to perform various analyses.

April 26, 2004

Categorical Data Analysis: Alan Agresti

This tutorial surveys methods for correlated categorical data, which occur with repeated measurement and other forms of clustering. The main focus is on two types of models. One type models marginal distributions, e.g., with generalized estimating equation (GEE) methodology. The other type uses random effects to describe subject-specific conditional distributions. Emphasis is on logit models for binary responses, but with some discussion of ordinal responses. Examples use SAS, mainly PROC GENMOD and NLMIXED.

May 19, 2003

An Introduction to Bayesian Modeling: James Albert

This seminar presents an overview of Bayesian modeling, emphasizing the rationale behind Bayesian thinking, rather than the technical aspects of modeling. The basic elements of Bayesian analysis, including choice of a prior, likelihood, and summarization of the posterior distribution will be illustrated in the simple setting of learning about a population proportion. Bayesian models will be surveyed for fitting continuous-response data, including regression and ANOVA models. Generalized linear models, such as logistic regression and item-response models will also be discussed. Examples of the use of Bayesian software will be provided. In all sessions, there will be a comparison of classical and Bayesian methods, with guidelines on the appropriateness of each method.

May 6, 2002

Data Mining: Where Do We Go From Here?: Dick De Veaux

The sheer volume and complexity of data collected or available to most organizations have propelled data mining to the forefront of making profitable and effective use of data. In this course, we'll take a brief tour of the current state of data mining algorithms. Using several case studies, we will examine how exploratory data modeling can be used to narrow the search for a predictive model and to increase the chances of producing useful and meaningful results.

May 24, 2001

Robust Engineering Using Taguchi Methods: Shin Taguchi

This seminar will address the following topics:

• Why robust design?
• How to apply Robust Design upstream
• How to define robustness
• Identify ways to measure robustness
• Determine applicability of Robust Design to your product of process
• How to determine the Ideal Function of a system
• How to translate customer intent and perceived result into engineering terms (signal and output response)
• How to identify noise and control factors of the system
• Define and differentiate between noise and control factors
• Four strategies to address the effects of noise on the system
• How to select test levels for noise and control factors
• How to lay out, conduct, analyze, evaluate, and confirm tests and results
• Calculate Signal-to-Noise Ratio and use it effectively for data analysis
• Identify control factors that show a high Signal-to-Noise Ratio
• Evaluate and improve robustness using the Signal-to-Noise Ratio
• Determine optimum control factor levels while maintaining or reducing cost

May 15, 2000

From Shewart Monitoring to Box-Jenkins Adjustments: J. Stuart Hunter
To hit a production target with least variability requires both attentive monitoring and careful adjustment. Fully automatic manufacturing processes are indeed possible, but the role of human attendants is far from vanishing. This seminar employs models descriptive of the industrial environment appropriate to the Shewhart monitoring chart and to the new Box-Jenkins adjustment chart. In application both charts are informative graphics designed for hands-on real-time use on the production floor. The two charts contribute to the full practice of SPC, "Statistical Process Control".

The Box-Jenkins model for a production process is that of a wandering mean about a fixed target. A simple statistic for such autocorrelated data is the EWMA, the Exponentially Weighted Moving Average. The EWMA can be employed as a smoother of a noisy wandering time series, and as an estimate of the current level of the series. When an estimated process level departs importantly from target an adjustment is usually ordered. The Box-Jenkins Manual Adjustment Chart provides both the estimated process level and the required adjustment. Minimum mean square error about target is the primary objective. In operation the Box-Jenkins Manual Adjustment Chart procedure is identical to that of automatic integral feedback control.

May 10, 1999

INTRODUCTION TO MIXED MODELS FOR THE PRACTICING STATISTICIAN - With Analysis Using PROC MIXED of the SAS® System: George Milliken
Mixed Models are needed to provide the analysis of data from most designed experiments. The presentation starts with an introduction to the terminology used in the discussion of mixed models including definitions of random effects, fixed effects, random effects models, fixed effects models and mixed effects models. The one-way analysis of variance model with unequal variances is used as the starting point of the discussion of the mixed models. The syntax of PROC MIXED of the SAS® system is described and the remaining examples are analyzed with the purposes of (1) providing an understanding of the features of PROC MIXED and (2) of developing an understanding of the characteristics of various designs and their analyses. The split-plot and strip-plot designs are used in many settings, including experiments that are conducted over two or more processing steps. Examples of industrial applications are used to demonstrate these designs and their analyses. Repeated measures designs are the most complex in that there is a correlation structure among the measurements made on the same experimental unit. PROC MIXED enables the analysts to model the covariance structure of the repeated measures and allows for the comparison of the fits in order to select an appropriate structure. An example from the pharmaceutical industry is used to demonstrate the procedures for selecting an appropriate model for a repeated measures data set. The concepts of narrow, intermediate, and broad inference spaces are described and their implications are demonstrated using an example from a multiple location trial, an experiment where the same design is conducted at several locations. Finally, the power of PROC MIXED is used to provide the analyses of an unconnected design and of a crossover design. Each participant will be provided with a set of notes that consists of copies of the transparencies used during the presentation. The goal of the seminar is to provide that participants with tools that will enable them to identify when a mixed model is needed to describe a given data set and how to accomplish the analysis using PROC MIXED of the SAS® system.

May 4, 1998

Re-Inventing Regression thru Graphics: R. Dennis Cook and Sanford Weisberg
Regression is the study of the change in a response variable as one or more predictors are varied. It is used to judge the effectiveness of a treatment, to form prediction equations, and for many other purposes. We present a new context for regression that requires few scope-limiting assumptions, and a corresponding collection of new methodological tools. Many of these tools use simple graphs, along with a well-developed theory, to discover information about the dependence of a response on the predictors. All the methods flow from a few key ideas concerning dimension reduction, understanding the role of the distribution of the predictors, and thinking about and using graphs in regression analysis.

Regression graphics have developed rapidly over the past six years, and many new developments are in progress. The topics were selected to be immediately useful in applications, to set the stage for future study in the area and show its promise.

Regression analysis is one of the fundamental tools for the practicing statistician. The traditional role of graphics in regression, at least with many predictors, has been peripheral, used mostly to judge model adequacy. Regression graphics moves graphs to the center of analysis. The new theory will be presented at a very general and intuitive level. We want to encourage to participants to use this approach in their own work and teaching.

The methodology described is very general, and can be used in almost any regression problem. in the course of the workshop, we will work examples with both continuous and binary responses.

All the methodology discussed will be illustrated with a computer package (R-code) that can be used for all the new methods described, and many standard methods for linear regression, nonlinear regression and generalized linear models. A copy of the most recent version of R-code, which runs on the Mac, PC, or Unix, will be made available to all workshop participants for use in their own work.

May 19, 1997

Artificial Neural Networks: A Practical Introduction: Richard D. De Veaux. and Lyle H. Ungar
Artificial neural networks are being used with increasing frequency for prediction and classification in high dimensional problems. Due to a variety of reasons, not the least being that they were originally inspired by attempts to model the human brain, they have received enormous publicity. This has led to a variety of claims by users of artificial neural networks and a great deal of misinformation about what these models can and cannot do well. Often, statisticians have been put into the position of having to defend their own practices against these claims.

This course provides a tutorial overview of artificial neural networks, focusing primarily on the most commonly used network, the backpropagation network. We explain, from a statistician's vantage point, why neural networks might be attractive and how they compare to other statistical estimation techniques. We will discuss the practical implementation issues surrounding the use of neural networks in applications. Modern statistical methods can be used on many of the same problems that neural networks are used for. We compare the use of neural networks to these statistical methods and discuss the relative advantages and trade offs between the use of these different tools.

May 20, 1996

Statistical Intervals: A User's Guide: Gerald J. Hahn and William Q. Meeker

June 1995

Analysis of Messy Data: Dallas E. Johnson, Kansas State University

May 1994

Interfaces Between Statistical Process Control and Engineering Process Control: John F. MacGregor

May 1993

What Do You Do When Standard Designs Don't Fit?: Christopher J. Nachtsheim, University of Minnesota

May 1992

Regression Modeling: Separating Signal from Noise in Data: Richard F. Gunst

May 1991

Design of Experiments: Douglas C. Montgomery

May 1990

Response Surface Techniques and Mixture Experimentation: John A. Cornell

May 1989

Statistical Methods for Quality and Productivity Improvement: George E. P. Box and R. Daniel Meyer, The Lubrizol Corporation

May 1988

Control Charting - Beyond Shewhart: J. Stuart Hunter

May 1987

Regaining the Competitive Edge: Howard E. Butler, Harold S. Haller, Robert V. Hogg, J. Stuart Hunter, R. Daniel Meyer, and Robert N. Rodriguez