By Lubinski, David; Humphreys, Lloyd G.
Psychological Bulletin, Vol 107(3), May 1990, 385-393.
Abstract
The traditional methodology for assessing moderator variables (hierarchical multiple regression analysis) is examined. Possible drawbacks of this technique for corroborating psychological theories (J. R. Busemeyer and L. E. Jones; see record 1983-24711-001) are illustrated empirically on the basis of an analysis of 400,000 subjects. This article tested a well-known (and currently popular) substantive hypothesis: A synergistic relation exists between mathematical ability and spatial visualization in the prediction and development of sophisticated levels of advanced mathematics. Using the traditional methodology, this hypothesis was confirmed; however, on further analysis, using a more systematic approach, it was demonstrated that this finding was spurious. Suggestions are offered for modifying the traditional methodology used for assessing moderator effects (for both applied and theoretical purposes). These amount to ways for minimizing Type I and Type II errors. (PsycINFO Database Record (c) 2010 APA, all rights reserved)
http://www.vanderbilt.edu/Peabody/SMPY/SpuriousModeratorEffects.pdf
www.psychspace.com心理学空间网