Table 1. Power to detect simulated correlation QTL across a wide range of scenarios (n = 1000 for all simulations).

Simulation parameters
(1) Correlation QTL effect*	x	x	x	x	x	x	x	x
(2) Genotype predicts the mean of gene one expression levels† (e.g., through an eQTL or proportion QTL)		x	x		x
(3) Genotype predicts the mean of gene two expression levels†			x
(4) Genotype predicts the variance of gene one expression levels‡ (e.g., through a varQTL)				x	x
(5) A variable that is random with respect to genotype predicts the mean of gene one expression levels†(e.g., through technical, environmental, or cell type heterogeneity effects)						x
(6) A variable that is random with respect to genotype predicts the mean of gene two expression levels†						x
(7) A variable that is random with respect to genotype predicts the variance of gene one expression levels‡(e.g., through technical, environmental, or cell type heterogeneity effects)							x	x
(8) A variable that is random with respect to genotype predicts the variance of gene two expression levels‡								x
Results
Power (proportion of true positives detected, Bonferroni-corrected p<0.05)	0.5091	0.4934	0.5055	0.0716	0.0679	0.4998	0.0884	0.0533
False positive rate (proportion of true negatives detected, nominal p<0.05)	0.0519	0.0529	0.0494	0.0532	0.0507	0.0527	0.0529	0.0522

Simulated effect sizes were as follows: correlation QTL^* = 0.3, mean effects^† = 1, variance effects ^‡ = 10. When Bonferroni-corrected p-values were used, the false positive rate was 0 across all simulation scenarios. Abbreviations: eQTL = expression QTL and vQTL = variance QTL.