The relationships identified as significantly heterogeneous were then examined to determine whether this could be explained by systematic differences across the individual assessment instruments. Due to the mixed nature of these data (i.e., multiple measures of a construct per sample), calculating the potential moderating effects using the Hedges and Olkin (1985) analog to ANOVA was not feasible, as this method requires independence of units. As a proxy to this technique, a set of independent effect sizes were calculated for each of the FFM and PD measures so that they could be visually compared to one another. For example, all samples in which the NEO PI-R was administered were meta-analyzed separately to determine this instrument's specific relationships with the DSM personality disorders. This same technique was repeated for all DSM and FFM measures that were utilized in at least three samples. In cases where samples reported multiple effect sizes for the same measure, the construct not currently being investigated were averaged together. For example, when the NEO PI-R was correlated with two unique PD instruments within the same sample, these values were averaged to arrive at an independent NEO PI-R effect size for the sample.
