tion. In the example, controlling the Diagnosis-Work History association for age, length of hospitalization, and sex, by combining these into a covariate set (possibly including quadratic and interactions as in the psychiatric status example of Table 6.5), recommends itself.
These considerations serve to underscore the flexibility and generality of the SC system. Information in any form may be represented as sets of variables, sets may be partialled from other sets for control or for the specification of research issues, and the strength and nature of the association between research factors determined. Subsets or single variables, partialled as necessary, represent the functional components whose relationships provide the analytic detail necessary for understanding associations. Issues of statistical power may be addressed in research planning. Attention to logical or causal structures and assessments of the magnitude and statistical significance of hypothesized relationships can thus come to the fore.
This chapter is a reworking and updating of the basic reference on set correlation ( J. Cohen, 1982b, reprinted as Appendix 4 in J. Cohen & P. Cohen, 1983). The new material includes estimators of the measures of association ("shrinkage" formulas) ( Cohen & Nee, 1984), power analysis ( J. Cohen, 1988b, chap. 10), robustness ( Cohen & Nee, 1989), examples from the SETCOR manual (J. Cohen, 1989), and properties of the measures of association ( van den Burg & Lewis, 1988). Conceptual and practical issues are emphasized and much of the technical detail is omitted.
I gratefully acknowledge my indebtedness to my wife, Patricia Cohen, for critical discussion and stimulating interaction.
Anderson, T. W. ( 1984). An introduction to multivariate statistical analysis ( 2nd ed.). New York: Wiley.
Cliff, N. & Krus, D. J. ( 1976). "Interpretation of canonical analysis: Rotated versus unrotated solutions". Psychometrika, 41, 35-42.
Cohen, J. ( 1968). "Multiple regression as a general data-analytic system". Psychological Bulletin, 70, 426-423.
Cohen, J. ( 1978). "Partialed products are interactions; partialed powers are curve components". Psychological Bulletin, 85, 858-866.
Cohen, J. ( 1982a). New-look multiple regression/correlation analysis and the analysis of variance/covariance. In G. Keren (Ed.), Statistical and methodological issues in psychology and social science research (pp. 41-69). Hillsdale, NJ: Lawrence Erlbaum Associates.
Cohen, J. ( 1982b). "Set correlation as a general multivariate data-analytic method". Multivariate Behavioral Research, 17, 301-341.