Our results show that when the causal variant is directly genotyped by the low density array, the use of imputed data is not essential (Figure 4A). However, in cases where the causal variant is not typed or imputed in the low density panel, the variance of PP4 is much higher (Figure 4B). In this situation, the resulting PP4 statistic tends to decrease even though considerable variability is observed. Inspection of simulation results in Figure 5 (bottom row for tagging SNP, leftmost graph for shared causal variant) shows that while PP4 tends to be lower than for its counterpart with complete genotype data (top row, leftmost graph), PP3 remains low. This indicates that more probability is given to PP0, PP1 and PP2, which can be interpreted as a loss of power rather than misleading inference in favour of distinct variants for both traits.
