ABSTRACT
One often wishes to follow tests of significance by calculation of a measure of the size of the effect. It is not enough to decide that an association is real; one wants to know how big it is. However, neither the value of a test statistic nor its descriptive level is satisfactory as an effect-size measure. Both indicate how likely the result is if the null hypothesis distribution is true, and so they are influenced by the size of the sample. For any alternative to the null hypothesis, X 2 and G 2 get larger and the descriptive level smaller as N increases. An adequate measure of the effect cannot vary in this way.
