Skip to main content

Table 2 VIF and Type-I Error Control in the presence of a random effect

From: Quantitative gene set analysis generalized for repeated measures, confounder adjustment, and continuous covariates

    VIF Estimate Type-I Error
Sample size Replicates True VIF QuSAGE Q-Gen Qusage Q-Gen
5 2 20.024 10.02 18.08 .129 .049
5 5 20.024 10.38 18.47 .114 .052
5 10 20.024 10.44 19.46 .105 .054
5 2 6.435 3.63 5.99 .093 .055
5 5 6.435 3.69 5.99 .077 .055
5 10 6.435 3.70 6.28 .061 .048
15 2 20.024 10.37 19.44 .147 .052
15 5 20.024 10.47 19.13 .138 .052
15 10 20.024 10.49 19.75 .137 .051
15 2 6.435 3.69 6.27 .124 .057
15 5 6.435 3.72 6.18 .121 .054
15 10 6.435 3.71 6.38 .114 .053
  1. VIF and type-I error estimates under a longitudinal simulation in the presence of a subject specific random effect. Under the current simulation setting, the VIF estimates under QuSAGE are drastically underestimated and lead to inflated type-I error rates. Error rates in bold indicate they are within the margin of error (0.05 ± 0.00427) for the simulation study. The type-I error rates for Q-Gen are consistantly in control with mild error inflation when there are fewer replicates