approximately 40% of the variance was accounted for. When maternal self- report was used for all indicators (i.e., criterion as well as predictor constructs), there was a strong fit of the model to the data and substantial criterion variance (approximately 40%) was again accounted for. This result was consistent with expectations.
Taken together, this series of analyses was interpreted as illustrating several major points: first, in terms of external validity--and we mean across dependent measures as well as across populations--constructs defined by multiagent and multimethod indicators are far more likely to continue to demonstrate predictive validity than constructs defined by single methods and agents; and second, at least with self-report data, monomethod operational definitions can provide models with good internal validity. This second point carries greater import than may be initially obvious. For example, although a mother's self-report may be distorted (see Dawes, 1985), her "life portrait" is likely to be a coherent one. A mother's view of her child's behavior is bound to influence the child, regardless of whether or not others share her view. In the clinical arena, the mother's perspective may indeed be what needs to be addressed. Her view may be of overriding importance for the family.
There is much work yet to be accomplished in this area, but for the first time since Campbell and Fiske introduced the multitrait-multimethod matrix, there appears to be a growing number of technical tools to help in understanding the impact of method variance on our work as social scientists.
Support for this project was provided by Grant No. HD 22679, NICHD Center for Research for Mothers and Children; Grant No. MH 38318, NIMH Mood, Anxiety and Personality Disorders Research Branch; and Grant No. MH 37940, NIMH Center for Studies of Antisocial and Violent Behavior.
The writers gratefully acknowledge the helpful comments by P. M. Bentler, R. Dawes, M. Stoolmiller, and J. Tanakaon previous drafts of this manuscript.
Bentler P. M. ( 1980). Multivariate analysis with latent variables: Causal modeling. Annual Review of Psychology, 31, 419-455.
Cairns R. B., & Green J. A. ( 1979). How to assess personality and social patterns: Observations or ratings? In R. B. Cairns (Ed.), The analysis of social interactions: Methods, issue, and illustrations. Hillsdale, NJ: Lawrence Erlbaum Associates.
Carnap R. ( 1956). The methodological character of theoretical concepts. In H. Feigl & M. Scriven