Multiple Comparisons in SAS: Classical to Modern Methods

Duration: 3.0 days Course fee: $2,175 EPTO units: 4.2 CEUs: 1.8

Presented by Peter Westfall, Ph.D., professor of statistics at Texas Tech University and editor of The American Statistician

Students evaluate the rationale for multiple comparisons and multiple testing procedures (MCPs), the different types of MCPs, and when to use them. Students learn how to analyze data using SAS software and how to use state-of-the-art MCPs that enable nonnormal distributions, incorporate correlation structures and logical dependencies, and generally include the most powerful and flexible methods currently available.

The primary applications are taken from pharmaceutical science and biotechnology, including genomics, and additional applications are taken from business intelligence, agriculture, social science, epidemiology, and engineering. Statistical analyses include MCPs in studies with multiple outcome measures, multiple group comparisons, regression response surface comparisons, and functional data. Recent releases of SAS contain the most powerful MCPs currently available, enabling researchers to control both Type I and Type II errors at acceptable levels. In addition, new ODS graphics displays make the results of multiple comparisons easy to understand and present.

The course uses primarily the GLIMMIX and MULTTEST procedures in SAS/STAT, which subsume the MCP functionalities of the GLM and MIXED procedures, and provide additional powerful methods and graphical output. Students explore these procedures using a series of hands-on exercises. Upon completion of the course, participants will be able to use and display results from the best and most modern multiple comparisons in a wide variety of practical settings.

Learn how to

obtain and interpret simultaneous confidence intervals and hypothesis tests using standard methods such as Tukey's and Dunnett's
produce graphical output showing results of multiple comparison analyses
use PROC GLIMMIX and PROC MULTTEST for customized multiple comparisons applications with correlated and non-normal data
compute power for multiple comparisons methods
identify the most powerful multiple comparisons methods for your studies.

Who should attend

Statisticians, scientists, and managers who work for bioscience and biotechnology companies or government agencies; academic researchers in psychology, biology, education, social science, medicine, engineering, and business; editorial staff of scientific and technical journals; statistical support personnel who need to assist researchers in selecting appropriate statistical tools

Expand/Collapse All Expand All Collapse All Print version

Prerequisites

Before attending this course, you should be familiar with the basic structure and concepts of SAS (DATA step and SAS/STAT procedures such as PROC REG and PROC GLM). You should be familiar with the core concepts in descriptive statistics, probability, p-values and confidence intervals, Type I and Type II errors, multiple regression and correlation, and analysis of variance. Familiarity with multivariate analysis, mixed linear models, nonparametrics, analysis of binary data, and generalized linear models are helpful, but not assumed.

Course Contents

Multiple Comparisons and Multiple Tests Using SAS

review of hypothesis testing and confidence intervals
analysis of data from a study of a cold remedy
sources of multiplicity and effects of multiplicity
multiple comparisons and multiple tests: basic notions

Classical to Modern Multiple Comparisons Methods for Control of FWER in Normal Linear Models

Dunnett's method
Tukey's method
simulation-based methods

Power and Sample Size Determination for Multiple Comparisons Procedures

power calculation and sample size determination using individual power
more general power calculation

Step-Down Testing Methods

Holm's step-down testing method
free step-down testing method using correlations
logically constrained step-down method: Shaffer's Bonferroni-based method
step-down method using logical constraints and correlations: the extended Shaffer-Royen method

Step-Down Testing with Non-Normal, Heteroscedastic, and Correlated Data

PROC GLIMMIX solution for heteroscedasticity
multiple comparisons with repeated measures data
multiple comparisons with alternative data distributions

Nonparametric Resampling-Based Testing Using the MULTTEST Procedure

the MULTTEST procedure
bootstrap multiple testing
permutation multiple testing

Closed Testing Procedures: Hommel, Hochberg, Gatekeepers, and Fixed Sequences

the Simes test and Hommel's method
Hochberg's method
closed testing using the Fisher combination test
fixed sequence and gatekeeping procedures

Controlling the False Discovery Rate

the Benjamini-Hochberg false discovery rate controlling method
using PROC MULTTEST to adjust for multiplicity using the Benjamini-Hochberg method

Recommendations and Summary

summary
final words: Bonferroni
key references: books
key references: journal articles

Software

This course addresses SAS/STAT.

Course Materials

Students receive a hardcopy of the course notes and, in some courses, can choose to take home a copy of the course data.

Share Your Thoughts

Are there additional topics you'd like for this course to address? Would you like for this course to be offered at another training facility? Let us know by adding to our Interest List.

This page was created using SAS software.

Learning Center

Stay in Touch

Not sure what course to take?

Submit Your Course Requests

On-site Pricing