Results of the ARIMA intervention time-series analyses
| Collisions with injuries Quebec City* | Collisions with material damages† | Collisions with injuries for the rest of the province* | ||||
| β | p Value | β | p Value | β | p Value | |
| AR1 | – | – | 0.47 | 0.01 | – | – |
| SAR1 | – | – | −0.22 | 0.03 | – | – |
| SAR2 | – | – | −0.44 | 0.01 | – | – |
| SAR3 | – | – | −0.69 | 0.01 | – | – |
| SMA1 | 0.55 | 0.01 | – | – | 0.32 | 0.01 |
| Intervention | 8.75 | 0.02 | −63.58 | 0.05 | 41.11 | 0.48 |
| Trend | – | – | 1.12 | 0.01 | – | – |
| Constant | – | – | 825.83 | 0.01 | – | – |
| Model parameters | ||||||
| (p,d,q) s(p,d,q)12 | (0,0,0) (0,1,1) | (1,0,0) (3,0,0) | (0,0,0) (0,1,1) | |||
| Test for autocorrelation | ||||||
| Q‡ | 0.44 (lag 1), p>0.51 | 0.03 (lag 1), p>0.86 | 1.95 (lag 1), p>0.16 | |||
| Q‡ | 13.75 (lag 12), p>0.32 | 10.29 (lag 12), p>0.59 | 10.96 (lag 12) p>0.53 | |||
The interaction term (time×intervention) was not statistically significant and therefore pulled out of the three models.
↵* Autoregressive and moving-average parameters were significant in the estimation stage but were no longer significant when the independent variables were included in the models. They were therefore discarded from the final models.
↵† When a linear trend was included in the model, 11 dummy variables for each month were added in the model to control for seasonal variations in the distribution of collisions (December=reference category).
↵‡ Box–Ljung test for autocorrelated residuals.