To analyse all the study subjects without having to remove one of each kin pair, we considered implementing a mixed-regression association analysis in which phenotypic correlation was proportional to K [38], [39], while age and sex were fitted as fixed effects. The motivation for this model is that any polygenic component of disease causation should be shared among individuals according to their kinships. However, in fitting the mixed model the maximum likelihood estimate of the component of variance attributable to kinship was zero. In fact, among the 19 pairs of apparent first-degree relatives, the correlation in neuroticism was −0.36, which leads to an upper 97.5% confidence limit for the heritability of neuroticism of 0.22. As a consequence of this negative correlation in our study, there is no inflation of test statistics if we ignore kinship and analyse all the study subjects using standard linear regression. After adjusting for age and sex, the EPQ-N score was close normal (Gaussian) and no transformation was necessary.
