The Overestimation of Effect Sizes from F Values in Meta-Analysis: The Cause and a Solution
By Craig R. Hullett, Timothy R. Levine. 2003
54 citations recorded by the source corpus.
Because estimates of effect sizes are often either misreported or not reported at all, meta-analysts must use conversion formulas that allow them to estimate effect sizes from the information available in a research report. The focus of this paper is on formulas that convert F in ANOVA to eta-squared, d, or the correlation coefficient. This paper demonstrates that the traditional F to r formula can differentially, and in some cases radically, inflate estimates of effect size when extracting the results from studies with varying numbers of factors. Specifically, commonly used conversion formulas for converting F values to r actually yield a partial r statistic that can be substantially larger than the desired zero-order r depending upon the number and relevance of additional factors in ANOVA. An alternative method to calculate nonpartialled effect sizes according to the common formula (SS effect /SS total) is provided.
Subfields: Statistics and Probability
View Paper: https://www.semanticscholar.org/paper/73faf5e8f44c23dd00c7af1a045962373988c6bf
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Statistics and Probability
- B. Becker
- Ariel M. Aloe
- Nikos L. D. Chatzisarantis
- Martin S. Hagger
- Chris Stiff
- Chantelle Wood
- L. Fahrig
Citations in the corpus
- An Effect Size for Regression Predictors in Meta-Analysis (2012, background): “Still others have discussed the use of partial correlation values (Hullett & Levine, 2003; Levine & Hullett, 2002).”
- Do species life history traits explain population responses to roads? A meta-analysis (2012, background): “Note we could not include studies reporting partial R2 values (Hullett and Levine, 2003).”
- Ego depletion and the strength model of self-control: a meta-analysis. (2010, methodology): “Standard formulas were used to compute the effect size statistic from F ratios, t tests, or zero-order correlation coefficients (DeCoster, 2004; Hullett & Levine, 2003).”