The crash analysis used an interrupted time series approach following the procedure described by (Box and Jenkins, 1976). Blood alcohol concentrations (BACs) of drivers involved in fatal crashes provide the best objective measure of alcohol involvement, but fatal crashes were not appropriate here because of the small number of such crashes that occurred in the study jurisdiction (three in the entire year 1998). Instead, two surrogate measures of Tipsy Taxi program effects were used in this analysis, nighttime crashes and injury crashes1. Quarterly counts of such crashes were used in the analysis.
In order to help rule out the effect of other factors unrelated to the program, two comparison jurisdictions with similar socio-economic and DUI enforcement systems were used, Gunnison County (including Crested Butte) and San Miguel County (Including Telluride). A step function intervention at the start of the program (December 1984) was used, and the comparison counties were used as explanatory series. To meet the stationarity requirements of the Box-Jenkins procedure, logarithmic transformations of the time series were used.
The analysis of nighttime crashes found a small, but statistically insignificant (t = -0.48), reduction in nighttime crashes of about 4% after the program began. This lack of any meaningful reduction is apparent from Figure 4 which shows the number of nighttime crashes that occurred in each quarter during the period 1976-1998. The symbols represent the raw data, and the legend indicates the two modeled series that were fitted to the data. The difference between the two modeled series is too small to be seen on the graph.
The analysis of injury crashes gave quite different results, a highly significant (t = -2.61) reduction of 15%. The reduction is clearly visible on the graph (Figure 5). A separate time series analysis of the comparison series of injury crashes indicated no reduction at any intervention point near the start of Tipsy Taxi (mean = 35 injury crashes per quarter).
There were too few fatal crashes in Pitkin County and the comparison counties for formal time series analysis. Nevertheless, we plotted the annual number of such crashes versus year for the period to see if any differences were suggested between the two groups. The results are shown in Figure 6, and do suggest a reduction in the number of Pitkin County fatal crashes after the intervention, and little or no reduction in the number of fatal crashes in the comparison counties.
We then performed a before-and-after analysis of the ratio of Pitkin Countyís fatal crashes to the comparison countiesís fatal crashes to see if there was any significant change in the ratio after the intervention. We found that the ratio decreased from 0.78 to 0.60 (30%), but that, as expected, the decrease was not statistically significant (p=0.29). We also performed an ANOVA analysis of fatal crashes as a function of county (Pitkin and comparison) and period (before and after), finding no significant interaction effect (p=0.89). There were, however, significant main effects for both county (p=0.01) and period (p=0.02). The mean in Pitkin County was less than the mean in the comparison counties (4.6 per year and 6.8 per year, respectively), and the mean after Tipsy Taxi was less than the mean before Tipsy Taxi (4.7 per year and 6.6 per, respectively).
Because Pitkin County has relatively few crashes, reductions which approach statistical significance are unlikely to be found except with interventions with dramatic results. The fact that nighttime, injury and fatal crashes all have declined coincident with the implementation of the Tipsy Taxi program and that injury crashes declined significantly gives credence to the proposition that this ride service program has served to help reduce alcohol-related crashes.
1Nighttime and injury crashes are often used as proxy measures of alcohol-related crashes in studies of this nature involving small jurisdictions.