The Effectiveness of Drinking-and-Driving Policies in the American States: A Cross-Sectional Time Series Analysis for 1984-2000 LE Richardson DJ Houston 105 Middlebush Hall, University of Missouri, Columbia, MO 65211-6100, USA Background In the United States, alcohol is frequently recognized as the single most significant contributing factor in motor vehicle fatalities. The responsibility for mitigating the harmful effects of alcohol-related crashes rests primarily with the 50 states (and the District of Columbia). The linchpin in this battle against drunk-driving has been a reliance on a deterrence-based strategy. States attempt to discourage impaired driving by imposing sanctions that increase the certainty, severity, and celerity of punishment. Policies that address the certainty of punishment increase the likelihood of detecting impaired driving or the likelihood that punishment will be imposed. Among these types of tools are preliminary breath tests (PBT), implied consent laws, and illegal per se laws. A PBT grants law enforcement officers the authority to administer a breath test to a suspected drunk driver prior to arrest. The results of this test are then used by the officer to determine if there is probable cause for arrest. Implied consent laws stipulate that by virtue of applying for a driver s license an individual has consented to tests for intoxication if arrested for a driving-while-intoxicated (DWI) offense. Illegal per se laws define as a criminal offense the operation of a motor vehicle at and above a specified level of blood alcohol concentration (BAC). The lower the BAC level used to define DWI, the greater is the certainty of detection and punishment for impaired driving. States increase the severity of punishment by imposing minimum mandatory sanctions on those convicted for a DWI offense. Typically these sanctions are in the form of a monetary fine and/or a jail sentence. Additionally, revoking or suspending the license of a convicted driver adds to the costs one must pay for violating state law. In some states these minimum mandatory sanctions increase with each successive DWI conviction. Lastly, the celerity (or swiftness) with which punishment is meted out is increased by administrative license revocation (ALR or administrative per se laws), whereby the agency responsible for issuing driver licenses is authorized to revoke or suspend a license based on a specified BAC level or other criteria relating to a DWI violation. In addition to deterrence-based policies, states rely on alcohol control regulations to reduce impaired driving. Among these policies that restrict access to alcohol are the minimum legal drinking age (MLDA), open container laws, and dram shop laws. Since 1988 all 50 U.S. states have required an individual to be at least 21 years of age to purchase and consume alcohol. Prior to this time the legal age varied among the states between 18 and 21 years of age. An MLDA potentially reduces the dangerous interactive effect of being an inexperienced driver, an inexperienced drinker, and a risk taking youth. More generally, drivers of motor vehicles are restricted in their access to alcohol through open container laws that make it illegal to possess an open container of alcohol. Stricter state laws prohibit possession of an open container by any motor vehicle occupant. Another policy that limits alcohol access is a dram shop law that establishes civil liability for commercial servers of alcohol. In the event that a commercial establishment serves alcohol to a patron who is underage or clearly intoxicated, its owners and employees may
be held liable in the event the patron injures someone else in a traffic crash. Whether based in statutory or common law, dram shop laws create a responsibility to reduce the incidence of severe alcohol impairment that is borne by commercial servers of alcohol. Objectives Despite frequent adoption by states, research findings on the impact of drinking-anddriving laws are inconclusive. Consistent with the behavioral assumptions of deterrentbased policy, some research has reported enhanced traffic safety associated with policies that increase the certainty, severity, and celerity of punishment. Yet other analyses indicate that these tools have not reduced alcohol-impaired driving or have done so only temporarily following the implementation of new policy. In sum, the results of aggregate research examining the efficacy of criminal laws designed to reduce drinking-and-driving in the American states have been mixed. This present study provides a more comprehensive analysis of U.S. state drinking-anddriving policies than is typically performed. In particular, both deterrent-based and alcohol control policies will be examined, along with controls for other policies that have implications for traffic safety and state characteristics associated with the frequency of drinking-and-driving. Several measures of the affect of impaired driving will be employed to avoid the limitation inherent in any particular indicator. Furthermore, the 16-year time period studied here is longer than that examined in most studies. Three questions will be addressed. First, are drinking-and-driving policies correlated with lower motor vehicle fatality rates? Second, what types of policies appear to be the most effective? Third, how robust are the apparent effects of these policies in the presence of other traffic safety policies and state characteristics? Methodology While the goal of these state policies is to reduce impaired driving, direct measures of this behavior do not exist because most trips made by an impaired driver go undetected. Instead, most evaluations have studied the correlation between drinking-and-driving laws and traffic fatalities, the ultimate societal consequence of drunk driving. Using data from the Fatal Accident Reporting Systems, a census of fatal motor vehicle crashes in the nation, annual fatality rates are calculated separately for the 50 states and the District of Columbia over the period 1984 to 2000. Three dependent variables are used in the models below. First, the overall fatality rate is the total number of traffic fatalities per 10 billion vehicle miles traveled (VMT). Second, the single vehicle nighttime (SVN) fatality rate is the number of fatalities per 10 billion VMT that occur in crashes involving a single motor vehicle between the hours of 6 p.m. and 6 a.m. It is generally regarded that alcohol is over-represented as a causal factor in SVN crashes. Thus, the impact of state drinkingand-driving policies should be apparent in changes to the SVN fatality rate. Third, the alcohol-related fatality rate is the number of alcohol-related fatalities per 10 billion VMT. NHTSA defines an alcohol-related fatality as one that involves a driver with a known or imputed BAC level of 0.01 or higher. Because the BAC level of over half of all drivers in fatal crashes is unknown, multiple imputation methodology is employed by NHTSA to obtain estimates for these missing values. The alcohol-related fatality rate may offer a more accurate estimate of the impact impaired driving has on safety. In general, these three fatality rates adjust for the exposure to risk that varies according to state population and travel patterns, and are more normally distributed than simple fatality counts (a desirable property for regression analysis).
Annual editions of the Digest of State Alcohol-Highway Safety Related Legislation (compiled by NHTSA) serve as the source for coding state policy. Deterrent-based policies included in the analysis are grouped into three categories. First, binary variables are used to represent the following policies that increase the certainty of punishment: preliminary breath tests, implied consent laws, illegal per se BAC 0.08 (and below) laws, and illegal per se BAC 0.10 (and above) laws. Second, mandatory minimum sanctions for a first DWI conviction are severity policies. These sanctions are measured as: fines (in dollars), jail sentences (in days), and license revocation or suspension (a binary variable). Third, the celerity of punishment is represented by a binary variable for the presence of administrative license revocation (ALR) in a state. Alcohol control policies are operationalized as binary variables for open container and dram shop laws, in addition to the minimum legal drinking age (in years). Additional policies that affect traffic safety are state laws that mandate the use of seat belts and maximum speed limits. Seat belt laws provide for primary enforcement (a motorist may be ticketed merely for not wearing a seat belt) or secondary enforcement (a motorist must be stopped for another traffic offense to receive a seat belt citation). Two binary variables are included to represent primary enforcement seat belt laws and secondary enforcement seat belt laws. State maximum speed limits are measured as the highest posted speed limit on any state road. A last set of control variables represent state characteristics associated with drinking-anddriving behavior. These are: percent of 15-29 year olds in the population of 15 years and older, per capita alcohol consumption (in ethanol gallons), percent of the population 25 years and older with at least 16 years of education, unemployment rate (in percent), and personal income per capita (in 1982-84 constant dollars). In the time-series cross-sectional analysis presented below, fixed effects models utilizing the least-squares dummy variable method are estimated for each of the three fatality rates. F-tests indicate the need to include state dummy variables (for all but one state) in each of these models. Furthermore, modified Wald χ 2 statistics for panel heteroskedasticity and Wooldridge F-tests for panel AR(1) serial correlation indicate the presence of nonspherical residuals in all models. Thus, feasible generalized least squares (GLS) was used to estimate the state fixed effects models to correct for groupwise heteroskedasticity and AR(1) serial correlation with a common autocorrelation parameter for all panels. Results and Analysis Two sets of models are estimated for this analysis. Table 1 reports the results of the timeseries cross-sectional analysis for a reduced set of models that restrict explanation of fatality rates to state policies directed at impaired driving. In general, it is apparent in this reduced form that state policies are correlated with a clear reduction in state traffic fatality rates. In all three models, the estimated coefficients have negative signs indicating that the presence of these policies is related to lower fatality rates. The only exceptions to this pattern are the coefficients for implied consent laws, however, these variables do not approach any acceptable level of statistical significance. Taken together, policies that increase the certainty of detection or of punishment are significant factors in explaining reductions in traffic fatality rates. While the performance of policies related to the severity of punishment is not as strong. While all severity tools have the expected negative signs, only minimum mandatory fines are significantly related to the dependent variables. In contrast, the lone variable representing celerity of punishment (ALR) is significantly related to fatality rates. Similarly, the alcohol control policies all are
correlated with lower fatality rates, and appear to be important factors in reducing impaired driving. Based on these three models it appears that state drinking-and-driving policies have the desired effect of enhancing traffic safety, especially in the form of certainty and celerity deterrent-based tools, and alcohol control policies. When other state policies and characteristics are introduced into the estimated models the effectiveness of state drinking-and-driving policies is lessened. (See Table 2.) In contrast to the reduced models, the certainty policy tools do not emerge as significant determinants of fatality rates in the full (i.e., fully specified) models. While the illegal per se variables are negatively associated with fatality rates, only the laws that stipulate a BAC of 0.08 are statistically significant. The PBT variables are not only statistically insignificant, the sign of these coefficients flip. In contrast, as a group, the severity tools perform comparatively better in these full models. While mandatory minimum fines are statistically significant correlates of reductions in traffic fatality rates in all estimated models (reduced and full), mandatory license revocation or suspension emerges as another important determinant. Finally, ALR remains a significant correlate of fatality rate reductions. The evidence for alcohol control policies in the full models is weaker than that presented by the reduced model specification. Still, open container laws are significantly related to lower fatality rates in two of the models reported in Table 2. Similarly, dram shop laws continue to be correlated with significant reductions in all three fatality rates. However, the MLDA variable no longer is statistically significant in any of the three full models. The control variables for other state policies and characteristics generally are significantly correlated with all three traffic fatality rates, and in the directions suggested by the literature on drinking-and-driving. Discussion and Conclusion Do state drinking-and-driving policies reduce the occurrence of impaired driving? The general conclusion appears to be, yes, they do. While different policies emerged as important correlates of fatality rates depending on the model specification employed, in both sets of models the parameter estimates for the deterrent-based and alcohol control policies typically are in the expected direction and several are statistically significant in all models. Furthermore, it appears that the gains in safety attributed to drinking-and-driving policies are substantial enough to be evident in total fatality rates and not just those rates that represent fatalities in crashes where alcohol is a likely contributing factor (i.e., SNV and alcohol-related fatality rates). This finding is significant given the number of control variables entered in the fully specified models. Which state drinking-and-driving policies appear to be the most effective? Conventional wisdom about deterrent-based policies suggests that the most important component for deterring undesired behavior is the certainty of detection and punishment. Severity is generally regarded as less important and celerity is given little attention. The pattern evident in the reduced models of traffic fatality rates lends credence to this conventional wisdom. However, policies that pertain to the severity of punishment emerge as more prominent, and only illegal per se laws with a 0.08 BAC level remain significant, in the presence of other control variables. Thus, the results of the full models suggest that policies pertaining to the severity of punishment may be as important as certainty tools, if not more so, in deterring drunk driving. More important may be the finding that the celerity tool emerges as an important component of a deterrence-based strategy. Yet, this type of deterrent tool often is overlooked in evaluations and policy debates.
How robust are the statistical results pertaining to these policies? Findings related to several laws (i.e., 0.08 BAC, minimum fines, ALR, and dram shop) are fairly robust regardless of which proxy is used to measure the prevalence of impaired driving, and which control variables are included. However, other policies found significant in the reduced models became insignificant in the fully specified models, and vice versa. These results suggest that model specification must be carefully approached when evaluating the effectiveness of drinking-and-driving policies. Table 1 Time-Series Cross-Sectional Feasible GLS Estimates of Reduced Models Alcoholrelated Total fatality rate SVN fatality rate fatality rate Preliminary breath test -1.4494*** (0.5235) -0.6985*** (0.2346) -0.8651** (0.3647) Implied consent law 0.0473 (0.2720) 0.0123 (0.1304) 0.0331 (0.2159) Illegal per se BAC 0.08-3.9434*** (0.7552) -1.8899*** (0.3649) -3.2331*** (0.5769) Illegal per se BAC 0.10-2.1351*** (0.6377) -1.0257*** (0.3113) -1.4540*** (0.4968) Mandatory minimum fine for -0.0084*** (0.0011) -0.0041*** (0.0005) -0.0058*** (0.0008) Mandatory minimum jail term for -0.0580 (0.0864) -0.0442 (0.0431) -0.0150 (0.0646) License revoked or suspended for -0.4055 (0.4377) -0.0506 (0.2089) -0.2184 (0.3294) Administrative license revocation -3.0778*** (0.3146) -1.5061*** (0.1506) -2.4230*** (0.2341) Open container law -1.5429*** (0.3236) -0.5874*** (0.1592) -0.9997*** (0.2437) Dram shop law -1.6836*** (0.4536) -0.8120*** (0.2321) -1.6053*** (0.3521) Minimum legal drinking age -0.8307*** (0.1383) -0.4725*** (0.0703) -0.7258*** (0.1064) N 867 867 867 Log likelihood -1804.979-1222.882-1560.337 Wald χ 2 1456.63*** 1159.92*** 1053.72*** Adjusted R 2 0.6404 0.5666 0.5174 Note: Cell entries are unstandardized coefficients. (Numbers in parentheses are standard errors, corrected for groupwise heteroskedasticity and AR(1) serial correlation.) Intercepts for each state are estimated but are not reported here. * probability 0.10; ** probability 0.05; *** probability 0.01
Table2 Time-Series Cross-Sectional Feasible GLS Estimates of Full Models Alcoholrelated Total fatality rate SVN fatality rate fatality rate Preliminary breath test 0.1654 (0.3733) 0.1469 (0.1638) 0.3478 (0.2613) Implied consent law 0.1076 (0.2313) 0.1076 (0.1057) 0.1232 (0.1760) Illegal per se BAC 0.08-1.0880** (0.5283) -0.4516* (0.2490) -0.8783** (0.4011) Illegal per se BAC 0.10-0.6444 (0.4356) -0.2123 (0.2078) -0.2867 (0.3393) Mandatory minimum fine for -0.0029*** (0.0008) -0.0015*** (0.0004) -0.0019** (0.0006) Mandatory minimum jail term for -0.0339 (0.0527) -0.0412** (0.0196) -0.0221 (0.0429) License revoked or suspended for -0.6627** (0.3345) -0.3182** (0.1560) -0.5084** (0.2407) Administrative license revocation -0.7968*** (0.2338) -0.3750*** (0.1064) -0.5785*** (0.1763) Open container law -0.6708*** (0.2388) -0.1597 (0.1064) -0.3374* (0.1805) Dram shop law -1.0505*** (0.4027) -0.4738*** (0.1822) -1.0506*** (0.2934) Minimum legal drinking age 0.1696 (0.1173) -0.0386 (0.0556) -0.0133 (0.0860) Seat belt law: primary enforcement -1.0632*** (0.3453) -0.3859** (0.1540) -0.6153** (0.2458) Seat belt law: secondary enforcement -0.5832*** (0.2146) -0.1723* (0.1012) -0.4208*** (0.1581) Highest maximum speed limit -0.0658*** (0.0191) -0.0406*** (0.0088) -0.0659*** (0.0139) Percent 15-29 years old of population 15 years and above 0.6266*** (0.0682) 0.2370*** (0.0305) 0.4266*** (0.0489) Per capita alcohol consumption 2.2307*** (0.6380) 1.3391*** (0.2996) 2.4862*** (0.4750) Percent of population with 16+ years of education -0.1383*** (0.0335) -0.0709*** (0.0160) -0.0916*** (0.0249) Unemployment rate -0.1322** (0.0618) -0.0021 (0.0273) -0.0122 (0.0464) Personal income per capita -0.0001 (0.0001) -0.0000 (0.0001) -0.0000 (0.0001) N 867 867 867 Log likelihood -1591.73-1007.751-1313.227 Wald χ 2 4841.79*** 4515.61*** 3577.53*** Adjusted R 2 0.8343 0.7813 0.7869 Note: See Note for Table 1.