Configural Frequency Analysis (CFA) is a method for cell‐wise testing in contingency tables whether some model is contradicted. Goals of CFA include the identification of groups of cases that occur at higher (or lower) rates than expected from some model. These groups are then interpreted from a typological perspective. Thus far, CFA has chiefly used statistics that are relatives of the Pearson X‐squared. This paper proposes using a broader range of concepts of nonindependence. Specifically, it proposes using Goodman's (1991) three measures of nonindependence, the unweighted interaction, λ, the relative difference, ?, and the weighted interaction, λ. In a simulation such characteristics of these measures as marginal‐dependence are illustrated. The use of these measures in CFA is illustrated using k‐sample CFA designs. Suicide data are analyzed using the three measures.
Notes
Preparation of this article was supported in part by NIA Grant #5T32 AG00110–07 to Alexander von Eye and Grant No. RO1 A GO 9984 to Michael J. Rovine.