Partial-correlation adjustment can destroy the construct being measured

Evidence

Scores on his Intrinsic scale have been found to be highly correlated (typically 0.50 or higher) with scores on Orthodoxy and Fundamentalism scales (Altemeyer & Hunsberger, 1992; Batson et al., 1993). This finding has led some researchers to adopt a strategy of using partial correlations to assess the relationship between the Intrinsic scale and prejudice, adjusting for the relation with Orthodoxy or Fundamentalism. It is not clear that this strategy is wise, however, for two reasons. First, the correlations of the Intrinsic scale with Orthodoxy and Fundamentalism are so high that the adjusted Intrinsic scores may no longer measure the same construct. — , Rudman, Laurie A - On the Nature of…, 2575

Categorization

• XREF: A methodological caution that generalizes far beyond religiosity research - connects to the broader critique of statistical control in observational social science found in work on collider bias and Lord’s paradox.