Skip to main content
Fig. 2 | BMC Bioinformatics

Fig. 2

From: Examining the role of unmeasured confounding in mediation analysis with genetic and genomic applications

Fig. 2

Using Umediation for one unmeasured confounder due to population stratification, the proportion of simulations where the results match for the mediated effect whether the unmeasured confounder U is included or excluded from the analysis is greater than 98% and the proportion of simulations where the results match for the direct effect is greater than 89% for an effect of confounding less than or equal to that of the observed second PC for genetic ancestry (i.e. γ U  = α U  = β U  ≤ 5 and U is normally distributed with mean 0 and variance 0.001). For a very strong effect (i.e. γ U  = α U  = β U  > 5 and both U1 and U2 are normally distributed with mean 0 and variance 0.001), then the unmeasured confounder changes the results of the mediation analysis significantly (i.e. the proportion of simulations where the results match for the direct effect whether the unmeasured confounder U is included or excluded from the model decreases to 39%). Therefore, the results of the mediation analysis would not change dramatically due to unmeasured confounding of population stratification as long as the unmeasured PC has an effect similar or less than the second measured PC of genetic ancestry. This becomes more extreme for 2 unmeasured confounders as seen in the right hand plot

Back to article page