Abstract not found

this paper is to present a heuristic derivation of the variance-covariance structures of the binomial proportions obtained from designs involving one or two random components. The random components may be nested, or occur in an analysis of covariance models. For a design with only one random component, the models can reduce to an over-dispersion model (Williams, 1982) or an equal correlation structure model (Zeger and Liang, 1986; Prentice, 1988). Unlike the GEE approach, the present paper uses the multivariate over-dispersed binomial instead of the multivariate binary outcomes. This paper derives a tractable variance-covariance form that characterizes the underlying covariance structure of the over-dispersed binomial outcomes explicitly. The variance-covariance component parameters are estimated directly. The derivation is similar to that of the random effects modeling (e.g., Miller and Landis, 1991) but the parameters are modeled in a multiplicative rather than additive formulation. 3 Two-Way Layout Models