Variance-component estimation from human sibship data

The large-sample relative efficiencies of the analysis-of-variance (ANOVA) estimators of variance components and the intraclass correlation coefficient rho are investigated for the unbalanced single classification in the context of family studies. The efficiency of an analysis based on the method of unweighted group means is also investigated. From a Monte Carlo study which generates the group sizes from typical family-size distributions it is found that the relative efficiency of the ANOVA estimators of the between-group variance component exceeds 95% for values of .2 less than or equal to rho less than or equal to .4, but can fall below 60% for values of rho that are very close to zero. For the estimation of the between-group variance component the method of unweighted means tends to be preferable to the ANOVA method only if rho greater than .5.