Estimating true scores in kinship data

Estimated true scores are occasionally used to provide an estimate of scores which would be obtained if measurement error could be removed from fallible measures. When one knows only a test's relibility and an observed score for each subject, true scores are estimated by regressing observed scores toward the group mean. When relatives' observed scores are also available, estimates of true scores can be improved by using the additional information. Observed scores of subjects and relatives can be used as predictors of subjects' true scores in a multiple regression. The resulting formula is presented for the case in which only one relative's score per subject is known. Estimation of true scores for the general case of N relatives per subject is also discussed.