|
Activity Number:
|
481
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Wednesday, August 9, 2006 : 2:00 PM to 3:50 PM
|
|
Sponsor:
|
General Methodology
|
| Abstract - #307321 |
|
Title:
|
Confidence Interval Coverage for Four Effect Sizes for Predictor Variables in a Multiple Linear Regression Model
|
|
Author(s):
|
Todd Bodner*+
|
|
Companies:
|
Portland State University
|
|
Address:
|
P.O. Box 751 PSY, Portland, OR, 97207,
|
|
Keywords:
|
multiple linear regression ; effect sizes ; confidence intervals ; coverage probabilities
|
|
Abstract:
|
Bodner and McCune (2006) compared four effect sizes for multiple regression predictors and reviewed their confidence interval (CI) procedures. The present study evaluates the coverage probability of those CI procedures using a simulation experiment. Varied were the degree of association among the variables and the sample size. Results indicate that CI procedures for unstandardized and standardized partial regression coefficients exhibit excellent coverage probabilities across the conditions studied. However, CI procedures for the semi-partial correlation and change-in-R-squared statistic exhibit varying and often poor coverage probabilities. Although Bodner and McCune (2006) favor the latter two effect sizes for effect size description, these results suggest that CIs based on these statistics should be used for the heuristic purpose of quantifying precision rather than formal inference.
|
