附註 
Source: Dissertation Abstracts International, Volume: 4812, Section: B, page: 3612 

Thesis (Ph.D.)University of Toronto (Canada), 1987 

Estimation of the correlation of measurements among family members began with the work of Sir Francis Galton. He proposed the interclass correlation to assess the degree to which sons' heights are passed down genetically from their fathers. Later, Karl Pearson formulated the intraclass correlation to measure the degree to which sons of the same father are alike in height. Sir Ronald Fisher improved upon Pearson's estimator of the intraclass correlation through the analysis of variance (ANOVA) procedure 

A clarification is given of the relationships among the different methods of point and interval estimation of the intraclass correlation coefficient in the unbalanced oneway randomeffects model. On the basis of an extensive Monte Carlo simulation done to compare the various methods, a version of the point estimator originally developed by Fisher for the balanced case and the interval estimator based on this point estimator and its asymptotic variance are both recommended. The maximum likelihood method and Wald's method are both shown by simulation to provide unacceptable interval estimators for small values of the intraclass correlation parameter 

An estimator based on Fisher's approach is developed for the interclass correlation coefficient using arbitrary weights. The asymptotic variance of this estimator is derived. A Monte Carlo simulation is done to evaluate different weighting schemes. By a separate Monte Carlo simulation, these estimators are found not to differ greatly from the maximum likelihood estimator in terms of asymptotic efficiency 

Using the method of weighted sumsofsquares, estimators are developed for the generalization of intraclass and interclass correlations for a random vector measured on the members of an arbitrary number of classes within independent groups. With the derivation of the asymptotic variances of these estimators of the intraclass and interclass correlation matrices, a largesample alternative to interval estimation by the method of maximum likelihood is proposed and illustrated with a wellknown example 

School code: 0779 
Host Item 
Dissertation Abstracts International 4812B

主題 
Statistics


0463

Alt Author 
University of Toronto (Canada)

