Back to Annual Meeting Page		American Public Health Association 133rd Annual Meeting & Exposition December 10-14, 2005 Philadelphia, PA

Abstract #107742

J. Jackson Barnette, PhD, Biostatistics, School of Public Health, University of Alabama at Birmingham, RPHB 130, Birmingham, AL 35294-0222, (205) 975 7742, jbarnette@ms.soph.uab.edu

Recently, there has been a call, particularly in behavioral research journals, for providing post hoc effect sizes and/or measures of strength of association as indicators of the practical significance to accompany statistical significance. These measures are often used in meta-analysis. ANOVA-type measures are in the form of the standardized effect size (attributed to Cohen), eta-squared, omega-squared, and the intraclass correlation. Very recently, there has been a call for reporting confidence intervals around these measures.

There are several issues related to these practices. One relates to the conversion of any of these measures to another of them needed to conduct meta-analysis based on a common metric, such as the conversion of the standardized effect size to eta-squared or the reverse. When there are two groups, this is straightforward and exact. However, when there are more than two groups, often the case in practice, the equations presented by several authors are not accurate. Actual relationships, which will be illustrated, among the standardized effect size and the other measures are not exact or easily determined when there are more than two groups.

The basis for effect size/measures of association confidence intervals is based on the use of non-central statistical distributions and many practitioners are not familiar with their use. The need for doing this and some examples using SAS macros will be presented. However, there is a major problem found in these confidence intervals in that they are usually so large they provide very limited information for judging the accuracy of the estimates.

Learning Objectives: At the conclusion of the session, the participant in this session will be able to

• Describe the nature of the relationships among ANOVA-type effect sizes and measures of association when there are two groups and when there are more than two groups
• Describe the nature of effect size/association measures confidence intervals based on central and non-central statistical distributions.
• Compute, using a SAS macro, confidence intervals for the standardized effect size and eta-squared.
• Evaluate the confidence intervals produced around the standardized effect size and eta-squared.

Keywords: Statistics, Outcome Measures

Presenting author's disclosure statement:

I wish to disclose that I have NO financial interests or other relationship with the manufactures of commercial products, suppliers of commercial services or commercial supporters.