Table 3. Summary of method performance in three data-generating scenarios.
Table shows means across all data sets for: root mean squared error (RMSE), coverage (Cvg), true selection rate for main effects (TSR), false selection rate for main effects (FSR), true selection rate for interactions (TSRint), and false selection rate for interactions (FSRint). Top-performing methods will have low RSME, coverage near the nominal level (0.95), high TSR and low FSR. For each measure and exposure-response scenario, results from the top-performing method(s) are listed in bold.
| Method | RMSE | Cvg | TSR | FSR | TSRint | FSRint |
|---|---|---|---|---|---|---|
| h1(x): linear with multiplicative interactions | ||||||
| NPBr | 1.02 | 0.73 | 0.85 | 0.35 | – | – |
| NPB | 0.54 | 0.95 | 0.92 | 0.10 | 0.59 | 0.02 |
| UPR | 2.01 | 0.56 | 0.25 | 0.26 | – | – |
| SPR | 1.59 | 0.54 | 0.63 | 0.53 | – | – |
| BKMR | 0.55 | 0.96 | 1.00 | 0.39 | – | – |
| LM | 1.01 | 0.73 | 0.84 | 0.29 | – | – |
| LM-int | 0.73 | 0.95 | 0.68 | 0.04 | 0.32 | 0.04 |
| h2(x): nonlinear with multiplicative interactions | ||||||
| NPBr | 0.77 | 0.80 | 0.79 | 0.22 | – | – |
| NPB | 0.69 | 0.86 | 0.78 | 0.16 | 0.25 | 0.01 |
| UPR | 1.42 | 0.56 | 0.27 | 0.24 | – | – |
| SPR | 1.27 | 0.58 | 0.68 | 0.58 | – | – |
| BKMR | 0.59 | 0.92 | 0.96 | 0.48 | – | – |
| LM | 0.78 | 0.81 | 0.78 | 0.17 | – | – |
| LM-int | 0.89 | 0.91 | 0.54 | 0.08 | 0.20 | 0.07 |
| h3(x): constant function of fixed profiles | ||||||
| NPBr | 1.11 | 0.66 | 0.66 | 0.11 | – | – |
| NPB | 1.02 | 0.75 | 0.68 | 0.13 | 0.06 | 0.02 |
| UPR | 1.41 | 0.55 | 0.27 | 0.25 | – | – |
| SPR | 1.38 | 0.54 | 0.68 | 0.59 | – | – |
| BKMR | 0.69 | 0.91 | 0.97 | 0.64 | – | – |
| LM | 1.13 | 0.70 | 0.69 | 0.14 | – | – |
| LM-int | 0.99 | 0.91 | 0.56 | 0.14 | 0.12 | 0.11 |