Citizen Reporting of DUI- Extra Eyes to Identify Impaired Driving
Alcohol-Related Crash Results
The aim of this analysis was to determine if there was a significant decline in the number of monthly alcohol-related crashes in Montgomery County after Extra Eyes was implemented5.
Table 19 reports the actual number of alcohol-related crashes in the years 2000 through 2004 in each of the three counties.
We used ARIMA intervention analysis to examine the potential impact of Extra Eyes on crashes. ARIMA is the mathematical modeling of the dynamics within a time series to account for stochastic processes that produce time-related patterns in the series. The term ARIMA is a three-part acronym (AR, I, MA) that stands for the three types of dynamics that are accounted for by the model parameters: autoregressive (AR), integration (I), and moving-average (MA). An ARIMA process is the composite result made up of the sums of any auto-regressive and moving-average components, as well as any trend or drift (integration) that causes the series not to be stationary (i.e., not constant level).
In summary, ARIMA is a well-established analytic procedure used to determine whether an intervention at some point in time like Extra Eyes has an affect greater than would be expected if no intervention were introduced.
Crash data were aggregated into monthly time-series counts. Montgomery County was one time series, and the comparison counties (Prince George’s and Anne Arundel counties) were the others. We modeled/analyzed each of these two series separately, and then estimated parameters for the intervention effect for each, with the hypothesis that the intervention coefficient (pre-change/post-change) for the Montgomery County series would be significantly different from the intervention coefficient of the comparison counties’ series. The Extra Eyes intervention was initiated in November 2002. In each of these two time-series analyses, counts of non-alcohol-related crashes for Montgomery County and the two comparison counties were included in the model as a regressor series to partial out other within-site variance over time that would affect all crashes (e.g., seasonal/weather factors, economics, general levels of enforcement).
To dampen the effects of unobserved factors affecting all drivers (not just drinking drivers) we analyzed the ratio time series that was created by dividing the number of alcohol-related crashes by non-alcohol-related crashes. Additionally, similar ratio series from comparison counties were analyzed to capture the effects of any laws (statewide or local) or programs affecting these areas simultaneously. The monthly ratio series for the counties were analyzed in two ways: (1) individual models for each county with the intervention being the only covariate, and (2) one model with the ratio series for Montgomery County as the dependent variable and the ratio series for the comparing counties serving as covariates.
The monthly ratio series for Montgomery County is shown in Figure 6. The results presented in Table 20 indicate that there was no effect associated with the introduction of the Extra Eyes program, after controlling for autocorrelation. Nonsignificant results also were obtained when similar ratio series for Prince George’s and Anne Arundel counties were used as covariates in the model (Table 21).