|Saturday, February 22|
|PS3 Poster Session III & Continental Breakfast||
Sat, Feb 22, 7:30 AM - 9:00 AM
An Empirical Comparison of the Accuracy and Precision of Effect Size Indices for Artificially Dichotomized Variables: A simulation Study (302817)Yi-Hsin Chen, University of South Florida
Harold Holmes, University of South Florida
Jeffrey D. Kromrey, University of South Florida
Yong Li, University of South Florida
George MacDonald, University of South Florida
Patrice Rasmussen, University of South Florida
*Patricia Rodriguez de Gil, University of South Florida
Jeanine Romano, University of South Florida
Keywords: Effect size, Dichotomized variables, simulation study.
Reject and fail to reject decisions obtained in hypothesis testing are usually reported along measures of effect size that communicate information on the strength of the relationship between variables. Seven effect size indices for artificially dichotomized response variables have been proposed in the literature (standardized proportion difference, phi coefficient, arcsine transformation of proportions, Hasselblad and Hedges (1995) log odds ratio transformation, Cox (1970) log odds transformation, probit transformation, and biserial phi-coefficient), which estimate the standardized mean difference effect size that would have been realized before dichotomizing the response variable. A simulation study was conducted to investigate the accuracy and precision of these effect size indices. The factors investigated included overall sample size (n1 + n2 = 30, 60, 120, 240), ratio of sample sizes in the two groups (1:1, 1:2, 1:4), population effect size (0, .2, .5, .8), continuous score cut point for dichotomization (.10, .25, .40, .50, .70), and population variance ratio (1:1, 1:2, 1:4). Results are provided in terms of recommendations for estimating and using these effect size indices.