Some observations on the definition and estimation of effect sizes

General considerations Many valuable comments have been made by contributors and while some of the key issues have been aired, I would like to suggest that prior to considering effect sizes it is important to pay attention to the correct specification of the statistical model being used. Thus, for standard regression or multilevel models the assumption of Normality is typically made and much of the literature on effect sizes, especially that which concentrates on standardised effects, assumes Normality. A prior transformation to Normality, for example using Normal scores, may often be needed, for both for the response and predictor distributions. Likewise, the existence of complexity in the form of interactions, or random coefficients in a multilevel model, should be explored and where such complexities exist a graphical presentation of effects will usually be especially helpful.