Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
Page 60 Chapter 8. Discussion of Results The Haddon Matrix was used initially to ensure that a comprehensive set of factors would be considered in studying the dramatic drop in traffic fatalities in the U.S. after 2007. The results showed that not all elements of the Matrixâvehicle, driver, and environmentâcontributed equally. In addition, factors that might be considered indirectly related to elements of the Matrix, i.e., the state of the economy, were most strongly associated with the decline. One of the complicating aspects of identifying and understanding the factors that influenced the number of traffic fatalities was that the number of fatalities is the product of risk and exposure, and changes in either or both can affect the number of fatalities (see Eq. 1 in Section 2). During the period from 2007 to 2012, both risk (the number of fatalities per VMT) and exposure (VMT) changed. This section discusses the relative importance of each in the reduction of fatalities. Figure 8-1 is a graph of the observed number of U.S. traffic fatalities from 2007 to 2012 plotted against two lines that represent: A. The number of fatalities that would have occurred if the level of exposure (VMT) had stayed constant at 2007 levels throughout the period and only risk varied. B. The number of fatalities that would have occurred if the level of risk had stayed constant at the 2007 level throughout the period and only exposure varied. Figure 8-1 Observed Traffic Fatalities versus Fatalities with Constant Exposure or Risk An examination of Figure 8-1 illustrates that most of the variation in traffic fatalities was due to changes in risk, although some portion can also be attributed to the change in exposure. Relatively speaking, fatalities due to changes in risk declined greatly in the first two years, accompanied by a moderate decrease due to reductions in traffic volume. From 2010 to 2012, fatalities continued to decline
Page 61 moderately due to further modest reductions in risk while the influence of traffic volume was relatively stable. Therefore, it is important to focus on the factors that influenced risk during this period in order to understand the reduction in fatalities. Table 8-1 identifies the expected association with traffic fatalities of the various parameters in the statistical models and describes the mechanisms through which they were hypothesized to affect the number of traffic fatalities. Most of the modeling results were consistent with these expectations. Some of the model results for the data used in the project were inconsistent with these hypothetical influence mechanisms. For example, it was expected that the wine consumption parameter would be positively associated with traffic fatalities (greater wine consumption associated with more traffic fatalities), but the modeling showed a weak negative association, such that greater per capita wine consumption in a state was associated with lower traffic fatalities. However, the effect size was small and the coefficient non- significant. Table 8-1 Explanatory factors and expected mechanisms of activity VariableÂ ExpectedÂ associationÂ withÂ trafficÂ fatalitiesÂ ExpectedÂ mechanismÂ TotalÂ VMTÂ PositiveÂ IncreaseÂ inÂ VMTÂ increasesÂ exposureÂ toÂ trafficÂ crashesÂ andÂ thereforeÂ fatalities.Â ProportionÂ ruralÂ VMTÂ PositiveÂ IncreasedÂ proportionÂ ofÂ ruralÂ VMTÂ increasesÂ proportionÂ ofÂ travelÂ onÂ riskierÂ roads,Â leadingÂ toÂ moreÂ fatalities.Â PumpÂ priceÂ NegativeÂ IncreasedÂ pumpÂ priceÂ raisesÂ costÂ ofÂ travel,Â reducingÂ totalÂ travelÂ andÂ discretionaryÂ travel,Â reducingÂ exposureÂ toÂ fatalÂ crashes.Â GDPÂ perÂ capÂ PositiveÂ GDPÂ perÂ capitaÂ reflectsÂ economicÂ activityÂ whichÂ inÂ turnÂ leadsÂ toÂ moreÂ travel,Â moreÂ exposureÂ toÂ crashes,Â andÂ moreÂ fatalities.Â MedianÂ IncomeÂ PositiveÂ IncreasedÂ medianÂ incomeÂ increasesÂ discretionaryÂ andÂ leisureÂ travel,Â resultingÂ inÂ moreÂ exposureÂ andÂ moreÂ fatalities.Â 16â24Â UnemploymentÂ NegativeÂ IncreasedÂ unemploymentÂ reducesÂ totalÂ travelÂ andÂ discretionary,Â leisureÂ travel,Â resultingÂ inÂ fewerÂ fatalities.Â CapitalÂ spend/mileÂ (lag)Â MixedÂ ImprovedÂ infrastructureÂ wouldÂ beÂ expectedÂ toÂ shiftÂ travelÂ toÂ higherÂ qualityÂ roads.Â ItÂ mayÂ alsoÂ induceÂ moreÂ travel,Â thusÂ moreÂ exposureÂ toÂ fatalities.Â SafetyÂ spend/mileÂ (lag)Â NegativeÂ IncreasedÂ trafficÂ enforcement,Â education,Â andÂ safetyÂ programsÂ wouldÂ reduceÂ riskyÂ drivingÂ andÂ reduceÂ fatalities.Â BeltÂ useÂ rateÂ NegativeÂ IncreasedÂ beltÂ useÂ providesÂ moreÂ protectionÂ toÂ vehicleÂ occupantsÂ andÂ reducesÂ theÂ probabilityÂ ofÂ fatalÂ injury,Â givenÂ aÂ crash.Â DUIÂ lawÂ ratingÂ NegativeÂ IncreasedÂ stringencyÂ ofÂ DUIÂ lawsÂ reducesÂ drunkÂ (risky)Â drivingÂ andÂ trafficÂ fatalities.Â
Page 62 VariableÂ ExpectedÂ associationÂ withÂ trafficÂ fatalitiesÂ ExpectedÂ mechanismÂ MotorcycleÂ helmetÂ lawÂ ratingÂ NegativeÂ IncreasedÂ stringencyÂ ofÂ motorcycleÂ helmetÂ lawsÂ providesÂ moreÂ protectionÂ toÂ motorcycleÂ ridersÂ andÂ reduceÂ theÂ probabilityÂ ofÂ fatalityÂ givenÂ aÂ crash.Â BeerÂ consumptionÂ PositiveÂ IncreasedÂ beerÂ consumptionÂ mayÂ increaseÂ drivingÂ whileÂ underÂ theÂ influenceÂ ofÂ alcohol,Â increaseÂ riskyÂ driving,Â andÂ increaseÂ trafficÂ fatalities.Â WineÂ consumptionÂ PositiveÂ IncreasedÂ wineÂ consumptionÂ mayÂ increaseÂ drivingÂ whileÂ underÂ theÂ influenceÂ ofÂ alcohol,Â increaseÂ riskyÂ driving,Â andÂ increaseÂ trafficÂ fatalities.Â PenetrationÂ ofÂ modelÂ yearÂ >1991Â NegativeÂ IncreasedÂ penetrationÂ ofÂ vehiclesÂ thatÂ provideÂ moreÂ occupantÂ protectionÂ andÂ moreÂ safetyÂ featuresÂ reducesÂ theÂ probabilityÂ ofÂ aÂ crash,Â andÂ reducesÂ theÂ probabilityÂ ofÂ fatalÂ injuryÂ givenÂ aÂ crash.Â 8.1 Variable influence estimated from the count models This discussion of results is based on the MNCS model (i.e., model not controlling for state) and the MCS model (model controlling for state) with VMT as exposure. In general, the MCS model provided better fit because the state fixed-effect in it is essentially a consistent adjustment that differentiates one state from another. However, the state fixed-effect may be correlated with other parameters in the model and may tend to obscure the influence of those other factors. Therefore, it is instructive to review the results from models that include and exclude state fixed-effects. The variables were grouped into the categories of Economic, Safety, Capital, Regulatory, and Vehicle Safety factors as shown in to determine the role each grouping played in determining fatalities. Table 8-2 Grouping of Variables GroupingÂ EconomicÂ SafetyÂ ExpendituresÂ RoadwayÂ CapitalÂ ExpendituresÂ RegulatoryÂ VehicleÂ SafetyÂ Va ria bl es Â ï· RuralÂ VMT ï· GDPÂ perÂ Capita Percent ï· Unemployment %Â forÂ 16â24Â year olds ï· Beer consumption ï· MedianÂ Income ï· SafetyÂ Expenditures perÂ highwayÂ mile (lawÂ enforcement, education,Â safetyâ relatedÂ capital investments,Â HSIP obligations) ï· Capital Expenditures perÂ highway mileÂ (excluding thoseÂ relatedÂ to safety) ï· Ratings: o DUI o Helmet o Safety Belts ï· PercentÂ of theÂ vehicle fleetÂ withÂ a 1991Â or newer modelÂ year
Page 63 The influence of the variables in these groupings can is provided in Table 8-3. The proportion of the decease is based on an average of the predicted and observed fatalities for the years 2008 through 2012. Table 8-3 Proportion of Predicted and Observed Reduction of Fatalities accounted for by groupings of Count Model Variables Â MCSÂ modelÂ MNCSÂ modelÂ VariableÂ GroupingÂ ReductionÂ inÂ PredictedÂ FatalitiesÂ AccountedÂ byÂ theÂ VariableÂ GroupÂ ReductionÂ inÂ ObservedÂ FatalitiesÂ AccountedÂ byÂ theÂ VariableÂ GroupÂ ReductionÂ inÂ PredictedÂ FatalitiesÂ AccountedÂ byÂ theÂ VariableÂ GroupÂ ReductionÂ inÂ ObservedÂ FatalitiesÂ AccountedÂ byÂ theÂ VariableÂ GroupÂ AllÂ VariablesÂ 100%Â â97%Â 100%Â â83%Â EconomicÂ â82%Â â80%Â â88%Â â73%Â SafetyÂ ExpendituresÂ 0%Â 0%Â â2%Â â2%Â RoadwayÂ CapitalÂ ExpendituresÂ â1%Â â1%Â +4%Â +3%Â RegulatoryÂ â3%Â â3%Â â2%Â â1%Â VehicleÂ SafetyÂ â13%Â â13%Â â12%Â â10%Â Table 8-3 provides a comparison the average percent of reduction in predicted and observed fatalities accounted for by each grouping of count model variables. The reduction of fatalities was based on the difference between the number of fatalities in 2007 and the number of actual observed fatalities or number predicted by each model for the years 2008 through 2012. The results indicate that economic factors had by far the largest influence on the predicted reduction in fatalities, and also accounted for a 73% to 80% of the observed reduction in fatalities. The next most significant contributor was improvements to vehicle safety. Regulatory changes had a much smaller impact. Safety spending had almost no influence in the MCS model and a small but not-statistically-significant effect in the MNCS model. Roadway Capital Expenditures had a mixed effect. The MCS model suggests a decrease of 1% due to capital expenditures but the MNCS model suggests that spending was associated with an increase but neither of the influences was statistically significant. Figures 8-2 through 8-6 provide a visual representation of the proportion of predicted and observed fatality reductions accounted for by each of the groupings of variables (economic, safety expenditures, capital expenditures, regulatory, and vehicle effects) for each of the models. The top of each figure shows the predicted results from the MNCS model and the bottom from the MCS model. For example, in Figure 8-2, the line segment A depicts the reduction in traffic fatalities predicted by the economic variables, while the line segment B depicts the predicted reduction from all the variables in the model. Line segment
Page 64 C represents the observed drop in fatalities from 2007. The ratio of A to B expressed as a percentage gives the proportion of the total predicted fatality reduction accounted for by economic factors. The ratio of A to C, expressed as a percentage gives the proportion of the observed drop in fatalities accounted for by economic factors. Figures 8-3 through 8-6 are interpreted similarly, however with a minor difference, for each of the respective model parameter groups. More specifically, in figures 8-3 through 8-6, the ratio calculated as (A A /B represents the proportion of the total predicted fatality reduction . a) MNCS model b) MCS model Figure 8-2 Quantification of Economic Effects on Fatalities
Page 65 a) MNCS model b) MCS model Figure 8-3 Quantification of Effects of Safety Expenditure on Fatalities
Page 66 a) MNCS model b) MCS model Figure 8-4 Quantification of Effects of Capital Expenditure on Fatalities
Page 67 a) MNCS model b) MCS model Figure 8-5 Quantification of Effects of Regulatory Effects on Fatalities
Page 68 a) MNCS model b) MCS model Figure 8-6 Quantification of Effects of Vehicle Improvements on Fatalities Effect of Economic Variables An examination of Figure 8-2 through 8-6 reveals that both models indicated that a significant portion of the drop in traffic fatalities over the period was accounted for by economic factors. According to both models, the set of economic factors accounted for 82% to 88% of the difference between the level of fatalities in 2007 and the level of fatalities predicted by the model over the period 2008 to 2012 and 73% to 80% of the decline in observed fatalities when compared to fatalities in 2007.
Page 69 Effect of Safety Expenditures State safety expenditures were compiled for each state. The expenditures were normalized by miles of highway in order to control for differences in the sizes of states. Safety expenditures included law enforcement, education, safety-related capital investments, and HSIP obligations and were entered into the model the year after (lagged) the expenditure was made (e.g., because the effects are not immediate, the expenditures in 2007 were evaluated in relation to fatalities in 2008). These expenditures were estimated to have made a minimal contribution to the dramatic decline in traffic fatalities over the period. According to MNCS and MCS models, on average over the period 2008 to 2012, state safety expenditures accounted for 0% to 2% of fatalities predicted by the model. However, the effect of the lagged safety spending was not statistically significant in these data. Thus, one cannot state with confidence that there was an effect from it. There are several possible reasons for this result: 1) the effect of highway programs may tend to be more long-term and cumulative, 2) short-term effects are rarely visible, 3) the effects were overwhelmed by the much larger impact of the economic recession, and 4) the effect of safety programs do not vary linearly with the amount of expenditures. Effect of Roadway Capital Expenditures Roadway capital expenditures were compiled for each state and normalized by miles of highway in order to control for differences in the sizes of states. Capital expenditures included non-safety-related capital investments and were entered into the model the year after (lagged) the expenditure was made (e.g., because the effects are not immediate, the expenditures in 2007 were evaluated in relation to fatalities in 2008).. The MNCS model predicted that these expenditures were associated with an increase in fatalities. This is a surprising and somewhat counterintuitive finding, which could be attributed to the huge amount of variance in the data. In contrast, the MCS model, which fixed state effects, predicted that these expenditures were associated with a decrease in fatalities, albeit small and non-significant. In fact, the variable was not statistically significant in either model and it cannot be said with confidence that the effect was either positive or negative. It is possible that this result was an artifact of the modeling process and the coefficients were compensating for the effects of other related factors. It is also possible that an increase in speeds associated with improved roadways or increased travel in rural areas could possibly contribute to an increase in fatalities. It is not possible to sort out with the current data why an increase was predicted by the model. The effect was reversed in the MCS model, which may indicate a correlation between roadway spending and state characteristics. It is possible, for instance, that capital spending was related to economic conditions within a state. Much state DOT funding comes from fuel sales. Increased fuel sales would generate more funding for capital projects but also be associated with high VMT and thus exposure to fatal crashes. Thus, the models may not be able to clearly predict the influence of roadway capital funding.
Page 70 According to the MNCS and MCS models, on average over the period 2008 to 2012, the effect of roadway capital investments was not large, ranging from an increase of 4% to a decline of 1% in the predicted level of traffic fatalities. Effect of Regulatory Policies In the model, the effect of DUI, Safety Belt and Motorcycle Helmet laws were taken into consideration using rating indexes developed for each state. The rating indexes increased as the laws become stricter. Each categoryâDUI, Safety Belt and Motorcycle Helmet Lawsâwere considered separately in the model and a combined effect was evaluated to assess the effectiveness of regulatory policies.. Both models show a marginal change in the prediction through the inclusion of these ratings. According to both models, on an average over the period 2008 to 2012, the influence was not large, and ranged from 2% to 3% decline in fatalities. None of the effects of the individual parameters were statistically significant in the MNCS model. However, DUI and motorcycle helmet law ratings were statistically significant in the MCS model. Effect of Changes in Vehicle Fleet Safety The effect of the influence of improved safety (more crashworthy designs, improved occupant protection, stability control, etc.) in vehicles was incorporated into the model as the proportion of the U.S. vehicle fleet that was model year 1991 or newer. Because data on fleet penetration could not be obtained for each state, this value was assumed to be the same across all the states. The proportion of the predicted reduction in fatalities that can be attributed to vehicle fleet safety, on average over the period 2008 to 2012, was 13% for the MCS model and 12% for the MNCS model. Effects of Individual Factors within the Economic Grouping Table 8-4 lists the economic factors, the change in each parameter between 2007 and 2011, and the associated fatality effect in the MNCS and MCS models. The effects of individual factors were determined by calculating the average value of the variable in the period from 2008 through 2012 and using the difference between that average and the value in 2007 to determine the resulting impact of that change on fatalities.
Page 71 Table 8-4 Effects of Individual Economic Factors VariableÂ ChangeÂ inÂ ParameterÂ ValueÂ fromÂ 2007Â toÂ 2011Â MNCSÂ modelÂ MCSÂ modelÂ PredictedÂ ChangeÂ inÂ FatalitiesÂ StatisticallyÂ SignificantÂ atÂ 5%Â level?Â PredictedÂ ChangeÂ inÂ FatalitiesÂ StatisticallyÂ SignificantÂ atÂ 5%Â level?Â RuralÂ VMTÂ ProportionÂ asÂ percentÂ ofÂ totalÂ â0.8Â %Â â103Â YesÂ 95Â YesÂ StateÂ GrossÂ DomesticÂ ProductÂ perÂ CapitaÂ â$6301Â perÂ personÂ â617Â YesÂ â1236Â YesÂ UnemploymentÂ RateÂ forÂ 16Â toÂ 24Â yearÂ oldsÂ +6.39%Â â3305Â YesÂ â3125Â YesÂ PumpÂ priceÂ +$0.55/gallonÂ â877Â NoÂ 127Â NoÂ PerÂ CapitaÂ BeerÂ ConsumptionÂ â0.08Â Gal./personÂ â835Â YesÂ â1312Â YesÂ MedianÂ incomeÂ â$3760Â 2677Â YesÂ â466Â YesÂ Â Unemployment Rate for 16 to 24. The unemployment rate for 16 to 24 year olds had the biggest effect on the decline in traffic fatalities from 2007 through 2012. In both models, the overall magnitude of the effect was similar, and both were statistically significant. In both models, the effect of the increase in teen and young adult unemployment accounted for between 50% and 52.% of the total reduction predicted by the set of economic factors. Because unemployment itself is not a traffic risk in and of itself, there has to be a mechanism that acts upon risk. It is suggested that the increase in teen and young adult unemployment led to reduced driving by this group, and because this group tends to have higher crash rates (risk) (see, e.g., (Massie, Campbell et al. 1995), reduced the risk per unit of overall VMT. Both MNCS and MCS models indicated that a 5% increase in the unemployment rate for 16 to 24 year olds resulted in a 5.7% to 6.4% decrease in fatalities. Pump Price. Increases in the pump price of gasoline were associated with reductions in fatalities in the model without state fixed effects (MNCS). However, when state fixed-effects were considered, then the influence of pump price was reversed and reductions were associated with an increase in fatalities. This may be because the state-fixed effect was correlated with pump price and in some way tended to overestimate without an adjustment. The effect of pump price was not significant in the MCS model and marginally significant in the MNCS model. Accordingly, it cannot be said with certainty that the effect was not zero. The manner in which pump price may lower risk is that higher prices may discourage driving amongst riskier populations, which could be young people without jobs, or older drivers with limited resources. It may also tend to discourage riskier rural driving because of the longer distances travelled and increased fuel consumption on those trips. In addition, it may reduce discretionary, leisure travel, particularly among teens and young adults, and those in the lower income quintiles.
Page 72 The average national pump price over the period from 2007 to 2012 is shown in Figure 8-7. Note that the pump price fluctuated in a manner not entirely consistent with the trend in fatalities over that time. This may contribute to the mixed results obtained in the two different modeling scenarios and the lack of statistical significance of the effect of pump price. Other studies, as noted in Section 6.7, have found a relationship between fuel prices and traffic fatalities (Grabowski and Morrisey 2004; Grabowski and Morrisey 2006; Morrisey and Grabowski 2011). Figure 8-7 Fuel prices, constant 2013 dollars, 2007-2012 Beer Consumption. The models indicated a consistent relationship between a reduction in per capita beer consumption and a reduction in traffic fatalities, both of which were statistically significant. In the model without state fixed effects (MNCS), the reduction in beer consumption predicted 10% of the total reduction in fatalities due to economic factors, and about 13% in the state fixed effects model. The potential mechanism for this parameter to affect risk is fairly direct, as the likelihood of drunk driving was likely to diminish as consumption of alcoholic beverages declined. The modeling indicated that a tenth of a gallon increase in per capita beer consumption would be associated with a 2.8% increase in fatalities in the model without state fixed-effects and a 4.1% increase in the model that includes state fixed-effects. The national trend in beer consumption from 2007 to 2012 was shown in Figure 6-18 above. Over the period, beer was the primary alcoholic beverage consumed and, in contrast to wine and spirits, consumption declined by 2.5% in 2009 and 2.7% in 2010. Gross Domestic Product per Capita. In both models, a decrease in per capita gross domestic product was associated with a decrease in traffic fatalities. The effect was statistically significant in both models. In the MNCS model this variable accounted for about 8% of the decline in fatalities attributed to economic factors and 15% of the decline from economic variables in the MCS model.
Page 73 Per capita GDP was likely connected to traffic fatalities by influencing the types and amounts of travel (VMT). GDP per capita is a measure of economic activity, much of which is realized through the transport of goods and services. Higher rates of economic activity may also be associated with greater leisure and discretionary driving, which may be riskier, while also allowing more travel by riskier groups (such as younger novice drivers). Similarly, declines in GDP or reduced growth could limit or reduce riskier travel. Rural VMT. The association of the proportion of rural travel of total VMT with traffic fatalities was not consistent across the models. Although the MNCS model suggested that a decline in rural VMT proportion was associated with a reduction in traffic fatalities, the MCS effect model suggested the reverse. The effects were statistically significant in both models. The percent of rural VMT in a state might be a surrogate for the degree of âruralnessâ of that state, and therefore the parameter may be correlated with the fixed effect. In that case, the proportion of rural VMT might be âadjustingâ the fixed effect and act in a different manner than when considered without those state effects. Rural travel is riskier than urban travel, so it would be expected that a decrease in the rural proportion of travel would be associated with a decrease in fatalities. The MNCS model estimated that a 10% increase in rural VMT proportion would predict a small (2.25%) increase in fatalities. The percent of rural travel was fairly stable during the period from 2007 to 2012, declining only slightly, which may also help to explain the mixed results from the models. Median Income. The association of median income with traffic fatalities also differed between the MCS and MNCS models. The MNCS model indicated that lower median income was associated with a greater number of traffic fatalities. On the other hand, the MCS model suggested the opposite. Both were statistically significant. This result, in combination with the result for rural VMT proportion, illuminates important differences between the two models, as well as the mechanisms that connected the factors in the models with safety. The MNCS (model not controlling for state) attempted to account for variance from two primary sources: the 50 states and the 12 years of the target period. States with lower median income tended to have a higher proportion of rural VMT, tended to spend less per highway mile on safety, and had higher fatal crash rates. Thus, lower median income was associated with higher risk and thus higher fatalities. The model controlling for state (MCS) focused on changes over time (the 12 year period of the study) within a state. In this model, decreases in median income likely reduced the travel of a relatively riskier population and thus reduced overall risk and therefore traffic fatalities. This may be considered an example, in the regression context, of Simpsonâs Paradox, in which a trend appears in disaggregate groups of data, but disappears or is reversed when the data are aggregated. In this case, the MNCS model may capture the long-term effect of low median income and correlated factors across states, while the MCS model may capture the short-term effect of changes in median income on who drives and how much they drive.
Page 74 Figure 8-8 depicts the national trend in median income from 2007 to 2012. Median income declined over the period and then grew in 2012. This trend was very similar to the national trend in traffic fatalities over the same period. Figure 8-8 Trends in median household income, 2007-2012 Effects of Individual Factors within the Regulatory Factors Table 8-5 lists the parameters that were grouped into the regulatory factors grouping, the change in each parameter between 2007 and 2011, and the associated effects on traffic fatalities in the MNCS and MCS models. Table 8-5 Effects of Individual Regulatory Factors Â ChangeÂ inÂ ParameterÂ ValueÂ MNCSÂ modelÂ MCSÂ modelÂ PredictedÂ ChangeÂ inÂ FatalitiesÂ StatisticallyÂ Significant?Â FatalÂ changeÂ StatisticallyÂ Significant?Â DUIÂ lawsÂ ratingÂ â1.05Â â120Â NoÂ â261Â YesÂ SafetyÂ BeltÂ lawsÂ ratingÂ â0.16Â 5Â NoÂ â28Â NoÂ MotorcycleÂ helmetÂ lawÂ ratingÂ 0Â 0Â YesÂ 0Â NoÂ Â DUI Rating. Both models indicated that an increase in DUI ratings was associated with a decline in fatalities. Although DUI rating accounted for almost all of the effect of the regulatory grouping, it only predicted some 2% of the reductions of all the parameters associated with the predicted decline in fatalities in the MNCS model and about 3% in the MCS model. The parameter was statistically significant in the MCS model but not significant in the MNCS model. A one point increase in the rating (i.e., increased stringency of the laws related to drunk driving) was associated with a decrease of 0.3% in the
Page 75 MNCS model and 0.7% in the MCS model. The mechanism relating DUI laws to risk was to deter impaired driving through increasing the severity of consequences for driving while impaired. The rating steadily increased over the period, reflecting the adoption of stricter DUI laws. Safety Belt Rating. The models suggested a very small and inconsistent effect of changes in safety belt ratings. A small increase in fatalities was associated with an increase in the Safety Belt Rating in the MNCS model and small decrease in the MCS model. The parameter was not statistically significant in either model. There was only a slight increase in the rating during the period, which may help to explain why the effect was not significant and inconsistent between the models. Motorcycle Helmet Rating. During the period from 2007 to 2012, no overall change in the motorcycle helmet rating occurred and therefore it is not associated with the decline in fatalities (see Section 6.6). 8.2 Variable influence estimated from the change model Figure 8-9 illustrates the magnitude of the effect of changes in the different parameters in the change model on the number of traffic fatalities, 2007-2011. In 2007, there were 41,259 traffic fatalities. In 2011, there were 32,479 traffic fatalities, which was down -21.3% from 2007. The factors in the change model accounted for about 12.4 percentage points of the decline, leaving 8.9 percentage points of the decline not accounted for.
Page 76 Figure 8-9 Change model prediction of variable effect on reduction in traffic fatalities, 2007-2011 The decline in total VMT accounted for only a modest reduction in traffic fatalities over the period. The model predicted that the 2.3% reduction in total VMT was associated with a 1.2% reduction in total fatalities. This finding aligns well with the overall expectation from Figure 4-2, which suggested that the substantial drop in fatalities could not be explained by the modest drop in total VMT. Much of the reduction in VMT occurred in rural areas, while urban roads, which tended to have lower fatal crash rates, either plateaued or increased slightly (Figure 6-8), so in addition to the effect of the change in overall VMT, there was a slight shift in the distribution of travel from rural roads to urban roads. The percentage of travel on rural roads declined by 1.6%, resulting in a 0.1% decline in traffic fatalities, predicted by the change model. Overall, change in VMT was not a primary contributor to the decline in traffic fatalities during the period. Economic factors Economic factors explained most of the decline in traffic fatalities over the period. Economic effects were captured mainly by four measures: unemployment for teens and young adults (ages 16 to 24), GDP per capita, median household income, and the pump price of auto fuel. The primary factor in the change model was teen and young adult unemployment. There was a 55.7% increase in the teen and young adult
Page 77 unemployment rate, 2007 to 2011, which the change model predicted produced a 6.1% decline in the number of traffic fatalities. Declines in GDP per capita and median income also contributed significantly. The 7.5% reduction in GDP per capita was predicted to reduce fatalities by 1.2%; the 4.3% reduction in median income reduced fatalities by 2.2%. The slight increase in the pump price of gasoline (up 2.6%) was predicted to result in a decrease in traffic fatalities by only 0.1%. In total, economic factors accounted for 9.5 percentage points of the total of 12.4 predicted by the model. This result from the change model aligns well with the two count models discussed above. In those models, economic factors accounted for 80-90% of the total effect predicted by the models. In the change model, the set of economic factors accounted for 76.6% of the total effect predicted by the model. Note that reductions in median income produced greater reductions in traffic fatalities than the larger changes in GDP per capita. GDP per capita is interpreted as characterizing the overall growth (or contraction) in the economy, normalized by population to account for population differences between the states. Median household income is interpreted as capturing the median economic status of households, and by inference the lower quintiles of the population. If median income declined, it is also likely that the household income of lower quintiles also declined. Even though the decline in median income was proportionally less than GDP per capita, the change model predicted a larger effect on traffic fatalities. This suggests that lower income households disproportionately contributed to the reduction in traffic fatalities, somewhat in the same way that the increase in unemployment among teens and young adults may have contributed to the fact that teens and young adults accounted for a disproportionate share of the reduction in traffic fatalities documented in section 6.1, especially Figure 6-3. Some studies have suggested that lower socioeconomic groups have significantly higher mortality rates in traffic crashes and may have benefited from the recession through reduced travel and reduced risky travel (Cotti and Tefft 2011; Harper, Charters et al. 2015). It should be noted that the finding here is of an association between declines in median income and traffic fatalities. Further research is needed to explore the mechanism through which they are connected, e.g., through reduced travel, reduced risky travel, shifts in the mode of travel, or some other means. Driver and vehicle factors One mechanism through which the economic factors discussed above affected the number of traffic fatalities was through changing driver behavior, either by reducing the travel of riskier driver populations or inducing them to drive more safely, both for economic reasons. The change model also included three other factors that may be regarded as reflecting driver behavior: safety belt usage rates, beer consumption per capita, and wine consumption per capita. The parameters for each were not statistically significant, and the effects were small. The safety belt use rate increased by 2.4% over the period, resulting in a predicted 0.1% decline in traffic fatalities. Many studies have shown the relationship between safety belt use and occupant protection. The fact that the coefficient was small and non-significant in the present study does not mean that there was no contribution from increasing belt use; instead, it points up the focus and context of the study. Safety
Page 78 belt use, from observational studies, showed a slow and steady increase over the entire period. Increased belt use was among the long-term factors that operated to set the baseline level of safety over the period. The study, however, was focused on the substantial drop in traffic fatalities after 2007. There was no significant change in the rate of increase of safety belt usage, just the continuation of the prior trend of increasing belt usage. The penetration of more recent vehicle models into the fleet showed a similar pattern and effect. The percentage of post-1991 model year vehicles in the passenger vehicle fleet was used as a surrogate for the long-term spread of occupant protection and collision-mitigation features, as documented by Kahane (2015). These features are part of the long-term base of safety that cumulatively reduced crash risk, but because the fleet turns over slowly, the effects were incremental and long-term. In these data, the effect was positive, but small and not statistically significant. Beer consumption per capita declined over the period by 3.5% and the change model associated that decline with an 0.7% decline in traffic fatalities. The reduction in beer consumption may have been related to the recession, as consumers had less income to spend on beer, or it may have been related to other factors (changing consumer tastes, for example). The predicted result of the decline in beer consumption per capita was in the expected direction, although the effect was small, reducing traffic fatalities by 0.7%. The effect of wine consumption in the model, however, was somewhat counterintuitive. The change model parameter coefficient for wine consumption was small and negative, indicating that increased wine consumption was associated with a small (0.1%) reduction in traffic fatalities. As indicated, the parameter was non-significant, meaning there was no statistical support that the coefficient differed from zero. Regulatory factors As in the count models, the parameter for DUI laws was statistically significant and associated with a decrease in traffic fatalities. There was a 4.0% increase in the DUI rating index over the period, and the change model predicted a small, 0.7%, decline in the number of traffic fatalities associated with this change. This result aligns with the expected mechanism of DUI laws, i.e., increased stringency of laws was intended to discourage driving under the influence of alcohol and thus reduce crashes and fatalities. The result also was of similar magnitude to the effect in the count models above, though it was statistically significant in only one (MCS) of the count models. In any case, while these laws helped set the long-term level of safety, as indicated by the modeling result here, their contribution to the substantial drop in traffic fatalities after 2007 was small and incremental. In contrast, there was effectively no change in the index of motorcycle helmet laws, therefore there could be no effect of a change in the index.
Page 79 Highway expenditures The results of the change model for state highway expenditures on capital improvements and gross safety spending were also mixed, and paralleled the results from the two count models. The effects in the change model were small, not statistically significant, and mixed. The change in capital spending on state highways in the change model was slightly negative, but not statistically significant. The effect of gross safety spending (including capital spending related to safety, law enforcement, and safety education programs) was actually positive, meaning associated with an increase in fatalities, but also not statistically significant. It is likely that the mixed result was a statistical artifact and not meaningful, even directionally. The two variables were strongly correlated (0.78), so they overlap in explaining the variance. In any case, the effect of spending on roadways, with the exception of law enforcement, would be expected to be lagged and cumulative, not expressed over a short time period. Accordingly, it would not be expected that incremental changes in highway spending would have an immediate and dramatic effect on traffic fatalities. Improvements in infrastructure roll out incrementally and over a long period of time. The change measures were lagged by one year in the modeling, but still were both small in terms of predicted effect and statistically insignificant. Factors related to drivers can change substantially over a short period of time, as was the case for the great increase in the unemployment rate of teens and young adults. But it is very difficult to change infrastructure significantly in a short period. 8.3 Implications for Reducing Fatalities and Crash Risk This study found that the changes in the economy were the primary factors in the substantial drop in traffic fatalities from 2008 through 2011. Details of the results suggest possible areas for interventions that may be pursued productively to reduce traffic fatalities. ï· Teens and young adults accounted for almost 48% of the reduction in traffic fatalities, 2008 through 2012. It was suggested here that the mechanism was economic constraints reduced total travel and risky (discretionary and leisure) travel. The increase in teen and young adult unemployment rates was the primary economic factor in the statistical models of the decline in traffic fatalities. It has long been known that teens and young adults have disproportionately high crash risk, but the results from this study suggest that their behavior can be significantly modified over the short run, substantially reducing fatalities. ï· The findings related to median household income raise the possibility of an income effect. This finding warrants further investigation to determine if lower income groups disproportionately reduced their driving, engaged is less risky discretionary driving, or some combination of the two. There is some recent evidence that lower socioeconomic groups, as measured by educational attainment, tend to have higher traffic mortality rates (Harper, Charters et al. 2015), We believe further research is needed to understand the influence of changes in household income and the effect on the amount and types of travel. One goal would be to determine if safety interventions aimed at lower income groups may have a disproportionately positive effect, similar to reducing crash risk among teens and young adults.
Page 80 ï· DUI laws showed a significant positive effect in reducing traffic fatalities, even over the short term of this study and even within the substantial impact of the economic contraction. Reduced beer consumption similarly showed a significant, positive effect. It is clear that continuing to focus on reducing drunk driving can have a disproportionate effect on reducing traffic fatalities. ï· Rural VMT bears a higher risk of fatal crashes across all road types; reduction in the proportion of rural VMT was significant in the model that controlled for state effects. Programs aimed at reducing the risk of rural travel can substantially reduce traffic fatalities. ï· It may be difficult to discern in any given year the effects of safety countermeasures, due to the significant influence of other factors on traffic fatalities. There is a need to more fully assess and document safety advances from countermeasures because these other factors may obscure them. ï· The results here clearly illustrate that factors outside the authority of safety professionals can have highly significant impacts on the level of highway safety. In the short term, shocks in the economy can swamp the effect of safety interventions. It is clear that exogenous factors such as economic trends should be accounted for in setting realistic goals and evaluating traffic safety programs. Most of the factors incorporated into the models have been shown in other contexts to be effective in increasing traffic safety and reducing crashes and fatalities. That some were not statistically significant in the analysis here does not show that they are not related to safety. The specific problem at hand was explaining the sharp decline in traffic fatalities after 2007. The mixed results for some of the factors in the model likely was that their contribution was small relative to the other factors in the model, and their effect on traffic safety was more stable and long-term.