[Time series analysis of the influence of prevention using the example of the occurrence of accidents in Zurich before and after Tempo 50]

Abstract

In the evaluation of the effect of preventive measures data often arise at fixed time intervals and it is likely that successive observations are stochastically dependent. By means of the example of the "possible reduction of traffic injuries after the introduction of a speed limitation (50 km/h) in Zurich" it is shown how time series methods can help to assess the efficiency of such measures. For the series before the intervention a statistical model is identified which describes the probability structure of the series. This model is used to forecast the course of the series one would expect when no intervention takes place. The forecast and the actual series are then plotted in the same figure and the difference is analysed with a test. In addition to that the magnitude of the effect and its standard error are estimated (intervention analysis). The series "numbers of accidents with injuries" showed after the intervention a yearly reduction of approximately 250 accidents (14%) with a corresponding standard error of 30. The reduction is markedly less pronounced in the series "number of seriously injured persons" than in the series "number of slightly injured persons".