Table 1.
Comparing power and type I error across methods using simulated data
| Gene |
TNN
(97 SNPs) |
LEPR
(79 SNPs) |
FLT3
(87 SNPs) |
TCIRG1
(139 SNPs) |
GSN
(131 SNPs) |
|
|---|---|---|---|---|---|---|
| Type I error | LSS | 0.010 | 0.045 | 0.020 | 0.015 | 0.020 |
| t-Statistic | 0.075 | 0.100 | 0.085 | 0.100 | 0.100 | |
| Power | LSS | 1.000 | 0.855 | 0.025 | 0.030 | 0.100 |
| t-Statistic | 1.000 | 0.995 | 0.140 | 0.110 | 0.035 | |
This table shows the power of detection for each of the 5 considered genes when considering 200 simulated Q1 (null) and 200 SBP phenotypes. The type I error of LSS appeared to be well controlled below 0.05 (row 1), whereas the t-statistic shows slightly inflated type I error rates (row 2). Both LSS and the t-test performed well when analyzing TNN and LEPR (rows 3 and 4). Both methods showed smaller power in the analysis of FLT3, TCIRG1, and GSN (rows 3 and 4). Neither method showed uniquely better performance across the 5 genes studied