Comparisons of models for BaP and smoking versus lung cancer
| Model | Knots | Model terms | Deviance | AIC | Dispersion |
| Smoking only | – | 5 | 82.9 | 92.9 | 2.6 |
| Multiplicative | |||||
| Groups | – | 11 | 24.6 | 46.6 | 0.9 |
| Exponential, be = 17(11,24) | – | 6 | 57.8 | 69.8 | 1.8 |
| Linear, bl = 28(16,42) | – | 6 | 54.5 | 66.5 | 1.7 |
| 2 Knots by maximum likelihood | 8 | 34.0 | 50.0 | 1.1 | |
| Power curve, p = 0.15(0.09–0.22), bp = 4(3,4) | – | 7 | 37.9 | 51.9 | 1.2 |
| Additive | |||||
| Groups | – | 11 | 23.0 | 45.0 | 0.8 |
| Exponential, be = 26(15,27) | – | 6 | 55.5 | 67.5 | 1.7 |
| Linear, bl = 35(22,51) | – | 6 | 50.2 | 62.2 | 1.5 |
| 2 Knots by maximum likelihood | 8 | 33.2 | 49.2 | 1.1 | |
| Power curve* |
AIC, Akaike’s information criterion.
*The power curve model was not fit with risks additive to smoking risks due to technical difficulties.
Coefficients b and their standard errors are in units of 10−4. Comparing down columns: low deviance indicates better fit; differences in excess of 3.84 are significant (p<0.05).
Coefficients are not directly comparable between linear/exponential additive/multiplicative models.