Abstract
A readily implemented procedure is discussed for interval estimation of indexes of interrelationship between items from multiple-component measuring instruments as well as between items and total composite scores. The method is applicable with categorical (ordinal) observed variables, and can be widely used in the process of scale construction, revision, and development in social, behavioral, educational, and biomedical research. The approach can also be used for testing conventional or nontraditional hypotheses about these indexes. The procedure is illustrated with a numerical example.
ACKNOWLEDGMENTS
This research was supported by The College Entrance and Examination Board. We are grateful to C. Lewis, R. Martin, and M. Reckase for stimulating discussions on item-total correlation coefficients. We are indebted to the editor and two anonymous referees for a number of critical comments on an earlier draft, which have contributed significantly to its improvement.
Notes
1Due to the nonlinear functional nature of Fisher's z transformation, the resulting confidence interval for the interitem polychoric correlation coefficient obviously need not be symmetric.