The Effectiveness of Drinking and Driving Policies for Different Alcohol-Related Fatalities: A Quantile Regression Analysis

To understand the impact of drinking and driving laws on drinking and driving fatality rates, this study explored the different effects these laws have on areas with varying severity rates for drinking and driving. Unlike previous studies, this study employed quantile regression analysis. Empirical results showed that policies based on local conditions must be used to effectively reduce drinking and driving fatality rates; that is, different measures should be adopted to target the specific conditions in various regions. For areas with low fatality rates (low quantiles), people’s habits and attitudes toward alcohol should be emphasized instead of transportation safety laws because “preemptive regulations” are more effective. For areas with high fatality rates (or high quantiles), “ex-post regulations” are more effective, and impact these areas approximately 0.01% to 0.05% more than they do areas with low fatality rates.
Keywords: quantile regression analysis, alcohol-related traffic fatalities, policies
Go to:

Introduction

Driving under the influence of alcohol has long been a severe social problem in the United States. In 2009, a study by the National Highway Traffic Safety Administration (NHTSA) indicated that approximately 30 people died in alcohol-related collisions per day (approximately 11,000 deaths per year); that is, one person dies in an alcohol-related collision every 48 min. Additionally, this horrifying figure was the result of already improved traffic safety conditions (the data provided by the NHTSA showed that in approximately 1982, nearly 30,000 people died in alcohol-related collisions in the U.S. per year, which accounted for 60% of the overall traffic crashes. Today that percentage has dropped to 38%). In 1980, Mothers Against Drunk Driving (MADD) was founded in the U.S., dedicating itself to urging state and federal governments to enact a series of drinking and driving policies that significantly reduced alcohol-related fatalities in the U.S. Since then, government officials and scholars have conducted numerous investigations and studies on the effectiveness of drinking and driving policies in reducing alcohol-related fatalities.

The data used in the studies on drunk driving consist of three categories: Cross-sectional data (e.g., Beck et al., [1]; Paschall, [2]; Phelps, [3]), time-series data (e.g., Whetten-Goldstein et al., [4]; Villaveces et al., [5]), and panel data (e.g., Chang et al., [6]; Lovenheim and Slemrod, [7]; Hingson et al., [8]; Ruhm, [9]; Males, [10]; Cook & Tauchen, [11]; Saffer and Grossman, [12]). Two estimation methods were used in these traditional econometric studies: (1) the ordinary least square (OLS) method that estimates the conditional mean function of dependent variables; and (2) the least absolute deviation (LAD) method that estimates the conditional median function of dependent variables. These two estimation methods emphasize the central tendency distribution of dependent variables and they both address the data at a macro or comprehensive level instead of examining individual quantiles. However, an observation of the alcohol-related fatality data show that we must study the development tendency of the alcohol-related fatalities of individual quantiles in addition to the central tendency development of alcohol-related fatalities. The reasons are as follows:
1.1. High Consistency of the U.S. Alcohol-Related Fatalities

Figure 1 shows that although U.S. alcohol-related fatalities have declined significantly, the states with high rates of alcohol-related fatalities in 1982 had maintained comparatively high levels in 2009 (e.g., CA, TX, and FL); the opposite situation was also true (e.g., in UT, VT, and RI). Based on this phenomenon, we suspect that drinking and driving policies that showed mean effectiveness had different effects for varying quantiles or alcohol-related fatality rates, preventing the values for states in Quadrant 3 (i.e., the states that maintained high rates of alcohol-related fatalities) from moving toward Quadrant 1 (i.e., the states that had shown high alcohol-related fatalities transformed into states with low alcohol-related fatalities).
Figure 1
Figure 1
Alcohol-related fatalities in 2009 (horizontal axis) and 1982 (vertical axis).
1.2. Regional Difference in United States Alcohol-Related Fatalities

Figure 2 shows that the states with relatively high alcohol-related fatalities are situated in the west and the south, whereas the states with relatively low alcohol-related fatalities are situated in the northeast, indicating that U.S. alcohol-related fatality are regional. Chang et al. [6] indicated that the drinking and driving policies in different regions had varying effects (In Chang et al. [6], the U.S. was divided into Far West, Great Lakes, Mid East, New England, Plains, Rocky Mts., Southeast, and Southwest). We concluded that different drinking and driving policies had different effects depending on the level of alcohol-related fatalities.
Figure 2
Figure 2
Fatalities as a percentage of total fatalities in crashes involving at least one driver with a BAC=0.08+, 2006 (Source: NHTSA).

This finding provided strong motivation to examine the effectiveness of drinking and driving policies under different alcohol-related fatality rates. To effectively discuss the effects of various drinking and driving policies on alcohol-related fatalities in different quantiles, we used the quantile regression method proposed by Koenker and Bassett [13] for estimation. The simple concept of the advantage of quantile regression, relative to the ordinary least squares regression, is that the quantile regression estimates are more robust against outliers in the response measurements [14]. Other advantages of the quantile regression method include that it makes no distribution assumptions on the population; it supplements the insufficiency of the traditional regression methods, which focuses only on the mean value of alcohol-related fatalities to estimate and interpret drinking and driving policy parameters; and finally, in our study, it specifically identifies the differing effect levels of drinking and driving policies on alcohol-related fatalities in different quantiles. Thus, we believe that applying QR model has advantages over the traditional method.

The structure of this study is as follows: research motivation and objectives are introduced in Section 1. In Section 2, we provide background information for the alcohol control- and road safety-related policies and review empirical studies on drunk driving. In Section 3, we explain the methodology and the quantile regression model. In Section 4, we provide the results of the empirical study; we follow the steps provided in the research methodology to implement empirical research and introduce, interpret, and analyze the results from the conducted empirical study. Finally, in Section 5, we present a conclusion in which the empirical results are integrated and conclusions and suggestions for future studies are provided.