Table 2.
Final Boosted Regression Trees and Main Effects Linear Models
| Mean Cost (Dollars)† | ||||||||
|---|---|---|---|---|---|---|---|---|
| Percent of Variance Explained† | Predicted | |||||||
| Year | Plan Type | Nodes/Tree* | Boosted Trees | Main Effects | Gain‡ | Observed | Boosted Trees | Main Effects |
| 2001 | Indemnity | 9 | 49.7 | 47.2 | 2.5 | 4,005 | 3,986 | 4,007 |
| PPO | 9 | 52.1 | 47.6 | 4.5 | 3,141 | 3,136 | 3,157 | |
| POS | 6 | 49.8 | 42.5 | 7.3 | 2,787 | 2,783 | 2,805 | |
| 2002 | Indemnity | 8 | 17.7 | 16.6 | 1.1 | 4,141 | 4,184 | 4,216 |
| PPO | 12 | 15.2 | 14.2 | 1.0 | 3,898 | 3,885 | 3,910 | |
| POS | 2 | 15.2 | 14.8 | 0.4 | 2,953 | 2,982 | 2,994 | |
Number of terminal nodes per tree in final boosted regression trees model.
Based on test samples.
Gain=percent of variance explained by tree boosting minus percent explained by main effects.
POS, point-of-service; PPO, preferred provider organization.