Table 6.
Results of Search for Optimal Training Set Composition
| Response biases (relative to
training set) |
||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Model | RMSE | Non-Japanese faces | Japanese faces | American style | Japanese style | Display rule manipulation | Happy | Sad | Afraid | Angry | Surprised | Disgusted |
| Baseline Japanese | .0102 | 100% | 100% | 100%. | 100% | None | 0.2279 | 0.1394 | 0.0970 | 0.1124 | 0.2548 | 0.1187 |
| Baseline Americat | .0923 | 100% | 100% | 100%. | 100%. | None | 0.1317 | 0.2267 | 0.1788 | 0.1741 | 0.1279 | 0.1609 |
| Best Japanese | .0035 | 60% | 100% | 100%. | 100% | Anger 75% | 0.1419 | 0.2115 | 0.1247 | 0.2161 | 0.1605 | 0.1453 |
| Best American | .0206 | 100% | 90% | 100% | 60%. | None | 0.1863 | 0.1768 | 0.1805 | 0.1781 | 0.1347 | 0.1347 |
Note. Display rule manipulation corresponds to lower frequency of exposure to one or more emotion categories in the training set.