that can be at most adjusted for the effect of other covariates such as gender or environmental effects, but neither gene–gene interactions nor gene-environment interactions can be tested. Model-based approaches overcome these limitations by directly modeling the effect of the genotype on the phenotype using regression and including extra terms that model explicitly the within family correlation. Examples are the variance component model proposed in [82] and implemented in the package QTDT, its recently proposed extension that uses the family structure to limit the genotyping effort to necessary family members [83], and generalized estimating equations (GEE) that model the variance covariance matrix of the observations by taking into account the family structure [84]. Hybrid approaches that combine family based tests from extended pedigrees with association tests in unrelated individuals can be very powerful and leverage the strengths of both approaches [85]. For example, Uda et al. [42] used a combination of variance components models in extended pedigrees of Sardinians and analysis in unrelated subjects to confirm associations of SNPs in BCL11A with fetal hemoglobin expression in β-thalassemia carriers and sickle cell anemia patients. In general, positive associations from both linkage and association studies strengthen the evidence for true positive findings
