Non-random sampling in human genetics: estimation of familial correlations, model testing, and interpretation

Abstract

By choice or necessity, human geneticists and genetic epidemiologists often design studies that involve non-random sampling of clusters of individuals, and yet address hypotheses appropriate to the population as a whole. Failure to adjust for the non-randomness of data often leads to biased parameter estimates and misspecification of predictive models that cause familial resemblance of traits. We develop an approach to adjust for common forms of non-randomness in the context of estimating familial correlation with minimal distributional assumptions and discuss its implications in connection with adjustments for concomitant variables.