To study the effect of missingness in the phenotypic data on the power to detect GVs, we conducted eight simulation studies in which we studied two types of missingness in five different genotype-phenotype models. The effect of missingness completely at random (MCAR) was studied by simulating data in which each of the 20 simulated phenotypes had 10% missingness distributed randomly across individuals. With 2000 subjects and 20 phenotypes, this results in ∼4000 missing values (i.e., 10% of the total of 40000 observations). In addition, we studied the effect of blockwise missingness; 400 randomly selected subjects in each simulated file had valid data only for the first 10 of 20 phenotypes (e.g., comparable to the situation that data of two samples are combined: in sample 1 (N = 1600), a full 20-item questionnaire is administered, while in sample 2 (N = 400), the abbreviated version of 10 items is administered). This results again in 4000 missing values, i.e., the amount of missingness is the same across the two missingness scenarios, but the distribution is different.
