Why Multicollinearity Matters: A Reexamination of Relations Between Self-Efficacy, Self-Concept, and Achievement.
Authors:
- Herbert W. MarshHMHerbert W. Marsh
- Martin DowsonMDMartin Dowson
- James PietschJPJames Pietsch
- Richard WalkerRWRichard Walker
- 1 September 2004
- journal article
- Published by American Psychological Association (APA) in Journal of Educational Psychology
- Vol. 96 (3) , 518-522
Abstract
Multicollinearity is a well-known general problem, but it also seriously threatens valid interpretations in structural equation models. Illustrating this problem, J. Pietsch, R. Walker, and E. Chapman (2003) found paths leading to achievement were apparently much larger for self-efficacy (.55) than self-concept (-.05), suggesting - erroneously, as the authors' reanalysis shows - that self-efficacy was a better predictor of achievement. However, because standard errors for these two paths were so huge (.25) thanks to the extremely high correlation between self-concept and self-efficacy (r = .93), interpretations were problematic. In a model comparison approach to this multicollinearity problem, constraining these two paths to be equal provided a better, more parsimonious fit to the data and also substantially reduced the standard errors (from .25 to .03)Keywords
This publication has 5 references indexed in Scilit:
- The relationship among self-concept, self-efficacy, and performance in mathematics during secondary school.Journal of Educational Psychology, 2003
- Introducing LISREL2000
- Causal ordering of academic self-concept and achievement: Reanalysis of a pioneering study and...Educational Psychologist, 1999
- Organization of children's academic self-perceptions: Reanalysis and counter-interpretations of confirmatory factor analysis results.Journal of Educational Psychology, 1997
- Confirmatory Factor Analyses of Multitrait-Multimethod Data: Many Problems and a Few SolutionsApplied Psychological Measurement, 1989