 | The 131st Annual Meeting (November 15-19, 2003) of APHA |
|
| The 131st Annual Meeting (November 15-19, 2003) of APHA |
J. Jackson Barnette, PhD, College of Public Health, University of Iowa, E 220 D General Hospital, Iowa City, IA 52242, 319 384 5489, jack-barnette@uiowa.edu
In situations where we are interested in comparing means of groups in experimental and quasi-experimental designs using analysis of variance procedures, the standardized effect size has a role in sample size determination and in aggregation of reporting in meta-analysis. There are several variations of the standardized effect size, but the one being used in this discussion is the (range of means) / (root MSE). Often researchers use hypothetical values for the standardized effect size, such as we will be looking for an effect size of 0.5, which Cohen labeled as a moderate effect size. Also, average effect sizes are often used in meta-analysis without regard to the design characteristics including number and size of samples.
The primary purpose of this presentation is to demonstrate, though use of a Monte Carlo generated data sets, that there is very strong bias in the standardized effect size and that this bias must be considered in several procedures used to compare group means. For, example, in a situation where there are a large number of groups with small sample sizes, the observed probability of getting a standardized effect size of .5 or larger is 0.9984. In the case where there are four groups with 25 subjects per group, the observed probability of getting an effect size of .5 or greater is 0.2937. Thus, observed values of standardized effect size could have very different interpretations depending on design characteristics.
Learning Objectives:
Keywords: Biostatistics, Methodology
Presenting author's disclosure statement:
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.