Measurement error can arise not only from a marker in LD, but also from the numerical coding of the genotype. If, for example, the true effect of a causal variant is dominant, but it is coded as additive and the linear model is otherwise correct, then the miscoding could also lead to a spurious interaction term. Furthermore, use of imputed rather than directly measured genotypes also creates measurement error, particularly for causal variants with strong effects because imputation is usually performed assuming no association with the outcome. Finally, exchanging the roles of gene and environment reveals that measurement error in the exposure could also create a spurious interaction even if the genotype is accurately measured.