V.  RESULTS AND DISCUSSION

Connecticut Alcohol-Related Fatality Analysis

I. ARIMA Analyses of All Connecticut Alcohol-Related Fatalities

The random effects ARIMA model (000) (000) results indicated that there was a significant decrease in Connecticut’s alcohol-related fatality trend following the beginning of the campaign and including an 18-month period from July 2003 through December 2004 compared to the trend from January 2000 through June 2003 (p=.042).  As indicated in Table 35, the estimated reduction in the number of alcohol-related fatalities determined by the ARIMA analysis was 2.055 lives each month for the 18 months following the beginning of the campaign for a total of an estimated 37 lives saved.

The estimated reduction each month based on the ARIMA model for all alcohol-related fatalities matches the actual mean monthly decrease in Connecticut’s alcohol-related fatalities from 12.8 from January 2000 through June 2003 to 10.7 from July 2003 through December 2004. However, the analysis predicted that had there been no campaign, alcohol-related fatalities in Connecticut would likely have increased beyond the number that actually occurred during the period.

Table 35. Connecticut Alcohol-Related Fatality Trend ARIMA Results: Parameter Estimates
for Alcohol-Related Fatalities           

Another ARIMA analysis of the Connecticut alcohol-related fatality trend used the alcohol-related fatality totals for each month from contiguous counties as a covariate to help account for noise and the effects of drinking and driving trends in bordering counties as well as seasonal and economic variations.  The use of the covariate helped to clarify the effect of the campaign on Connecticut’s alcohol-related fatality trend.  The total number of alcohol-related fatalities each month from contiguous counties in New York (Duchess, Nassau, Putnam, Suffolk, and Westchester), Rhode Island (Berkshire, Hampden, and Worcester), and Massachusetts (Kent, Providence, and Washington) was used to construct the covariate trend for Connecticut’s alcohol-related fatality trend.

The random effects model (000) (000) was used for the ARIMA because the inclusion of the alcohol-related fatalities from contiguous counties as a covariate left no significant autocorrelations and no significant partial autocorrelations.  This model left a significant “sudden and sustained” effect on alcohol-related fatalities coincident with the beginning of the campaign and continuing through the end of 2004 (p=.01).

The results indicated that the campaign saved an estimated 2.604 lives each month, which is more than the previous estimate of about 2.055 lives each month without the covariate.  This estimate is a better prediction of the number of lives saved because it controls for more extraneous influences than the ARIMA model that did not include a covariate.  Thus, the total estimated lives saved increased from about 37 to about 47 with the addition of a covariate to the ARIMA model. 
           
Table 36 shows the estimated reductions in alcohol-related fatalities.  The significant value for the covariate indicates that the number of alcohol-related fatalities in contiguous counties is related to the number of alcohol-related fatalities in Connecticut.

Table 36. Connecticut Alcohol-Related Fatality Trend ARIMA Results Including a Covariate: Parameter Estimates for Alcohol-Related Fatalities Using Alcohol-Related Fatalities from Contiguous Counties as a Covariate

Figure 2 shows graphically the significant reduction in the predicted alcohol-related fatality trend in Connecticut after contiguous county data were used to remove any noise, seasonal, region wide trends, and economic variations that may have obscured the effect of the campaign on the trend.

Figure 2.  Connecticut Predicted Alcohol-Related Fatalities 2000-2004 After Contiguous
County Data and Modeling Applied to Remove Noise and the Effects of Region Wide Efforts to Combat Drinking and Driving as Well as Seasonal and Economic Variations