Natural experiments such as studies comparing differentially exposed siblings (65) or exposed adoptees with natural offspring have the potential to reduce biases and confounding found in more traditional observational studies, although measurement error remains an issue. Mendelian randomization is one such natural experiment, with the additional advantage of being free from alcohol assessment problems. Causal inference is enhanced by studying genetic markers that are specifically associated with the exposure of interest, but not with other socially-patterned risk factors because genotypes are randomized at conception, and clearly they can be measured objectively (66).
