On the Measurement of the Influence of “Broad Categories” on Correlation