Does It Matter Which Subtests We Use to Measure General Cognitive Ability? Group Differences Across Test Composites

ABSTRACT We demonstrate that overreliance on the common convention in our field that virtually any composite of three or more specific cognitive tests will result in a robust measure of general cognitive ability (GCA) can lead to inconsistent inferences about the nature of subgroup differences in GCA. Using data from Project Talent and NLSY97, we examined how different composites of cognitively loaded tests can vary in terms of subgroup differences. After forming all possible three‐test composites in each dataset, we found that while these composites generally had very similar correlations with GCA, they varied much more widely in their degrees of gender and race/ethnic mean score differences. We discuss considerations that can aid in assembling test composites that better capture true subgroup differences.