Table 1. Percentage error (root mean square error) in the simulated odds ratio (OR) with confounder (simulation size = 10,000).
| Significance criterion (p < 0.05) | N = 200 | N = 500 | N = 1,000 | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| ORx | ORx,z | ORx ' | ORx,z ' | % indicating confounding effect | ORx | ORx,z | ORx ' | ORx,z ' | % indicating confounding effect | ORx | ORx,z | ORx ' | ORx,z ' | % indicating confounding effect | |
| Corr (X, Z) | |||||||||||||||
| 0.1 | 2.50 (0.16) | −0.46 (0.16) | 2.36 (0.17) | 1.69 (0.17) | 10.72 | 1.14 (0.10) | −1.01 (0.10) | 1.35 (0.10) | 0.68 (0.10) | 19.51 | 1.17 (0.07) | −0.44 (0.07) | 1.27 (0.07) | 0.63 (0.07) | 34.67 |
| 0.2 | 3.40 (0.17) | −3.15 (0.17) | 3.10 (0.17) | 1.61 (0.17) | 10.21 | 2.82 (0.11) | −1.69 (0.10) | 2.36 (0.11) | 0.98 (0.10) | 19.32 | 2.54 (0.08) | −0.82 (0.07) | 2.22 (0.08) | 0.93 (0.07) | 33.18 |
| 0.3 | 4.44 (0.18) | −5.21 (0.18) | 4.28 (0.18) | 2.03 (0.18) | 9.30 | 3.45 (0.11) | −3.43 (0.11) | 3.50 (0.11) | 1.43 (0.11) | 18.58 | 3.22 (0.08) | −1.97 (0.07) | 3.30 (0.08) | 1.32 (0.08) | 32.64 |
| 0.4 | 5.38 (0.18) | −8.10 (0.20) | 5.59 (0.18) | 2.74 (0.18) | 9.46 | 4.61 (0.12) | −5.00 (0.11) | 4.63 (0.12) | 1.80 (0.11) | 17.69 | 4.35 (0.09) | −2.85 (0.08) | 4.38 (0.09) | 1.75 (0.08) | 30.18 |
| 0.5 | 6.34 (0.19) | −10.79 (0.23) | 6.45 (0.19) | 2.82 (0.19) | 9.17 | 5.56 (0.12) | −6.93 (0.13) | 5.61 (0.12) | 2.09 (0.12) | 16.17 | 5.28 (0.09) | −4.18 (0.08) | 5.36 (0.10) | 1.97 (0.08) | 27.46 |
Data are simulated from the model Prob(Y = 1)/Prob(Y = 0) = exp(0.1)X + exp(0.1)Z.
Percentage error and root mean square error in correct decision are underlined.
ORx: Percentage of OR of X where Z was not adjusted and the criterion has indicated a confounding effect.
ORx,z: Percentage of OR of X where Z was adjusted and the criterion has indicated a confounding effect.
ORx': Percentage of OR of X where Z was not adjusted and the criterion has not indicated a confounding effect.
ORx,z': Percentage of OR of X where Z was adjusted and the criterion has not indicated a confounding effect.