Table 2.
Simulation results for scenario B with three active exposure and two null exposures. The table shows (from left to right) RMSE for the exposure-response function and 0.95 interval coverage for the exposure-response function, the average pointwise RMSE and pointwise 0.95 coverage for the weight functions for the three active exposures, the frequency of detecting a window in the three active exposures, the frequency of detecting a window in the two exposures not associated with the outcome, and the precision for window detection which is number of correct windows identified divided by total number of windows identified.
| Model | RMSE h | Coverage h | RMSE w(t) | Coverage w(t) | Pr(window∣active) | Pr(window∣null) | Precision |
|---|---|---|---|---|---|---|---|
| n = 100, noise: sd(ϵ) = 3.0 | |||||||
| BKMR | 2.087 | 0.758 | NA | NA | NA | NA | NA |
| BKMR-DLM | 1.940 | 0.945 | 0.667 | 0.936 | 0.000 | 0.000 | NaN |
| BKMR-DLM-quad | 1.399 | 0.955 | 0.539 | 0.910 | 0.298 | 0.007 | 0.984 |
| DLM | 2.047 | 0.869 | 0.935 | 0.835 | 0.652 | 0.465 | 0.678 |
| DLNM | 1.793 | 0.861 | 0.783 | 0.806 | 0.719 | 0.475 | 0.694 |
| n = 100, noise: sd(ϵ) = 7.5 | |||||||
| BKMR | 2.487 | 0.823 | NA | NA | NA | NA | NA |
| BKMR-DLM | 3.392 | 0.978 | 0.682 | 0.937 | 0.000 | 0.000 | NaN |
| BKMR-DLM-quad | 2.646 | 0.973 | 0.681 | 0.925 | 0.020 | 0.000 | 1.000 |
| DLM | 3.814 | 0.915 | 1.092 | 0.860 | 0.385 | 0.324 | 0.641 |
| DLNM | 3.496 | 0.894 | 1.012 | 0.814 | 0.430 | 0.332 | 0.661 |
| n = 100, noise: sd(ϵ) = 15.0 | |||||||
| BKMR | 2.948 | 0.898 | NA | NA | NA | NA | NA |
| BKMR-DLM | 5.973 | 0.990 | 0.683 | 0.937 | 0.000 | 0.000 | NaN |
| BKMR-DLM-quad | 4.480 | 0.989 | 0.727 | 0.926 | 0.000 | 0.000 | NaN |
| DLM | 7.130 | 0.928 | 1.201 | 0.842 | 0.303 | 0.280 | 0.619 |
| DLNM | 6.339 | 0.912 | 1.147 | 0.809 | 0.296 | 0.285 | 0.609 |
| n = 500, noise: sd(ϵ) = 3.0 | |||||||
| BKMR | 1.354 | 0.819 | NA | NA | NA | NA | NA |
| BKMR-DLM | 0.866 | 0.974 | 0.407 | 0.925 | 0.393 | 0.002 | 0.996 |
| BKMR-DLM-quad | 0.812 | 0.927 | 0.379 | 0.851 | 0.723 | 0.007 | 0.993 |
| DLM | 1.700 | 0.636 | 0.856 | 0.649 | 0.858 | 0.828 | 0.609 |
| DLNM | 1.122 | 0.903 | 0.653 | 0.832 | 0.857 | 0.790 | 0.619 |
| n = 500, noise: sd(ϵ) = 7.5 | |||||||
| BKMR | 2.213 | 0.706 | NA | NA | NA | NA | NA |
| BKMR-DLM | 2.261 | 0.983 | 0.661 | 0.936 | 0.000 | 0.000 | NaN |
| BKMR-DLM-quad | 1.543 | 0.963 | 0.526 | 0.914 | 0.330 | 0.002 | 0.995 |
| DLM | 2.300 | 0.823 | 0.923 | 0.837 | 0.693 | 0.485 | 0.682 |
| DLNM | 2.147 | 0.906 | 0.813 | 0.878 | 0.622 | 0.435 | 0.682 |
| n = 500, noise: sd(ϵ) = 15.0 | |||||||
| BKMR | 2.484 | 0.733 | NA | NA | NA | NA | NA |
| BKMR-DLM | 3.177 | 0.995 | 0.683 | 0.937 | 0.000 | 0.000 | NaN |
| BKMR-DLM-quad | 2.493 | 0.978 | 0.641 | 0.927 | 0.030 | 0.000 | 1.000 |
| DLM | 3.586 | 0.891 | 1.057 | 0.869 | 0.480 | 0.328 | 0.687 |
| DLNM | 3.695 | 0.912 | 0.965 | 0.874 | 0.423 | 0.298 | 0.681 |