Table 4.
Mean association estimates in the population (‘no selection’) and among individuals with a CHD event (‘with selection’), and empirical power at a 5% level of significance for different magnitudes of confounding in the applied example (the β0 parameter is chosen such that the proportion of cases in the selected sample is about 20% for each value of βU and γU)
| No selection |
With selection |
|||
|---|---|---|---|---|
| βU, γU | β0 | Mean estimate | Mean estimate | Empirical power |
| 0 | –1.4 | 0.149 | 0.149 | 93.5% |
| +0.2 | –1.6 | 0.148 | 0.145 | 91.3% |
| +0.5 | –1.9 | 0.142 | 0.133 | 86.1% |
| +1 | –2.5 | 0.131 | 0.102 | 67.7% |
| +1.5 | –3.3 | 0.120 | 0.077 | 44.0% |
| +2 | –4.0 | 0.107 | 0.061 | 30.4% |