The use of discordant — one affected, one unaffected — sibling (sib) pairs in linkage study designs is currently in vogue in human genetics research because they are believed to afford greater statistical power to genome scans in search of markers linked to disease loci. In such study designs, discordant sib pairs are expected to share fewer alleles at disease-associated loci (or at loci linked to them) and to share alleles more frequently at unlinked loci. Lack of allele sharing is therefore the hallmark of linkage to the disease.

But there are problems associated with the use of discordant sib pairs, one of which is the focus of the study by Neale et al. — that of non-paternity. Their mathematical simulation shows that selecting for extreme phenotypic discordance between sib pairs can enrich for cases of non-paternity in a study sample, resulting in a sample that contains half, as well as full, sib pairs. Such a sampling bias can lead to false-positive reports of linkage because allele sharing is reduced in half sibs relative to full sibs.

In their study, Neale et al. first calculated the proportion of half sibs that are likely to be in a study sample that has been selected on the basis of extreme phenotypic discordance. They then plotted this proportion against the heritability of the trait under investigation. Strikingly, this analysis showed that the setting of extreme phenotypic thresholds for the selection of discordant sib pairs will probably result in nearly all the sibs in the sample being half sibs, rather than full sibs, when the trait being selected for is highly heritable. However, when lower selection thresholds are used and the trait is less heritable, smaller increases in the proportion of half sibs in a study sample can be expected.

Although, in practice, researchers rarely set such extreme phenotype selection thresholds — partly owing to the rarity of extremely discordant sib pairs —— Neale et al. still predict that there is a risk that half sibs could be present in a study sample selected from the upper 10% of a phenotype's distribution. Together with the fact that estimates of non-paternity in the general population range from 1–20%, this study flags the need for researchers to be aware of this risk and to carry out tests on genome-wide marker information to check for relationship errors in sib-pair samples.