Distracted Driving: Can Higher Cigarette Prices Reduce Non-Drinking Driving Fatalities Among Teens? Vinish Shrestha Emory University Department of Economics October, 2014 1
Abstract Motor vehicle accidents impose an enormous economic cost to a society. Deaths from motor vehicle crashes are a leading cause of unintentional deaths in the U.S. In this study, I evaluate the effect of higher cigarette prices on non-alcohol-related fatal accidents among16-to 20-year old drivers. Using state level panel data from the Fatality Analysis Reporting Systems (FARS), I find that a dollar increase in a cigarette pack is associated with a reduction in driving accidents by 5 to 7 percent among 16-to 20-year old drivers. This translates to approximately 5 less fatal accidents per year involving 16-to 20-year old drivers. I conduct several robustness checks and falsification exercises to test the validity of the findings. The logical explanation of the results is that higher cigarette prices reduce smoking rates among youths, decrease the incidence of smoking while operating a vehicle and thus, reduce instances of distracted driving and the risk of motor vehicle accidents. 2
Introduction Deaths from motor vehicle accidents are the leading cause of deaths due to unintentional injuries (33,687 deaths in 2010; CDC, WISQARS). The U.S. Department of Transportation s National Highway Traffic Safety Administration (NHTSA) estimates economic costs of motor vehicle crashes of $900 for each person living in the U.S. in 2010. 1 Nine percent of the total fatal crashes in 2010 are estimated to be due to distraction (NHTSA, 2010). Interruptions while driving can be caused by several factors such as cell phone usage, moving object in a vehicle, other occupants, eating or drinking, and smoking. While estimating the number of distracted fatal crashes, NHSTA attributes smoking as one of the determinants. Among all ages, 16-to 20-year olds are mostly vulnerable to traffic fatalities and distracted driving (Traffic Safety Facts, 2010). 2 Perhaps, due to the lack of experience in performing dual activities smoking and driving younger adults may be more prone to smoking related driving accidents. Little is known about whether smoking can lead to motor vehicle accidents. As shown in Figure 1, an upswing in cigarette prices following the Master Settlement Agreement (MSA) coincides with a reduction in number of non-alcohol fatal driving accidents among 16-to 20-year olds. Over the past two decades, many jurisdictions have banned smoking in bars, restaurants, and workplaces vehicles maybe one of the few places left, besides at home. In fact, this can increase one s propensity to smoke while driving. In an attempt to reduce distracted driving, since 2007, states have passed specific laws prohibiting text messages while operating a moving vehicle and are actively enforcing distracted driving laws. 3 Although, several countries have enacted smoking bans in private vehicles, these laws are implemented mainly to protect children from second hand smoke. 4 However, the smoking risk associated with motor vehicle accidents while operating a vehicle has not received serious attention. Several 1 The calculation of economic costs excludes costs associated with loss of life, pain, decreased quality of life and productivity due to injuries. 2 Eleven percent of the all drivers involved in fatal crashes were reported as distracted at the time of the crash (NHTSA, 2010). 3 Source: http://www.huffingtonpost.com/kendell-poole/states-take-action-to-red_b_4221018.html 4 Australia, Bahrain, Canada, Cyprus, Mauritus, South Africa, United Arab Emirates, and several states of United States have enacted smoking bans in private vehicles. The majority of these bans are imposed with a motive of preventing children from second hand smoking. 3
moments in a process of smoking can result in an increased risk of accident: 1) Lighting the Cigarette, 2) Reaching or looking for a cigarette, 3) Cigarette blowing back into the vehicle; and 4) Dropping the cigarette (Stutts, Reinfurt, & Rodgman, 2001). In the realm of policy sector, understanding the effects of smoking on driving accidents can provide further guidance in possible ways to reduce driving crashes and negative externalities associated with accidents. In this study, I evaluate the effects of smoking on non-alcohol related fatal driving accidents among 16-to 20-year olds by using the data from the Fatality Analysis Reporting Systems (FARS) of National Highway Traffic Safety Administration (NHTSA). 5 Smokers or people at risk of smoking might in general have specific personality traits which can increase the likelihood of getting in accidents. 6 To reduce such bias, I use cigarette prices as a plausibly exogenous factor and explore the effects of higher cigarette prices on non-alcohol related fatal driving accidents. In addition, evaluating whether increases in cigarette prices can reduce motor vehicle crashes is relevant for policy purposes. The reasonable pathway is higher cigarette prices reduces smoking rate, which decreases the incidence of smoking while operating a vehicle. If cigarette smoking is a factor that contributes to motor vehicle accidents, substantial increases in cigarette prices should reduce non-alcohol related accidents; hence, leading to fewer fatalities. I particularly focus on individuals of ages 16 to 20 due to two reasons: 1) Prior literature has shown that among all ages, people of this age group are most responsive to increases in cigarette prices (Sloan and Trogdon, 2004); and 2) Individuals of this age group are likely to be novice drivers; thus, acquiring less experience in performing dual activities-smoking and driving. Following the intuition that higher cigarette prices should not affect driving behavior of older individuals, I conduct a falsification test for individuals of 30 years old and over. 5 The study concentrates on non-alcohol related fatalities due to avoid the strong association between smoking and drinking. For example, if individuals substitute cigarettes for alcohol due to higher cigarette prices, this will raise the risk of driving fatalities. However, an increase in risk is not because of smoking behind the wheels but is drinking related. Hence, higher cigarette prices can influence drunk-driving fatalities by increasing the risk of drunk driving accidents. 6 Sloan et al. (2004) show that non-smokers are risk averse compared to smokers. 4
Economists favor higher cigarette prices as means to reduce cigarette smoking. A strand of literature focuses on effects of smoking reduction in public health outcomes such as increased life expectancy due to cutbacks in smoking related diseases, and improved birth outcomes (Sloan et al., 2004; Markowitz, 2007). This study has two main contributions. First, this study aims at determining whether smoking is responsible for driving accidents. As a second contribution, the study examines whether a reduction in smoking due to higher cigarette prices can curtail driving accidents. Such knowledge is important for public policy purposes if smoking is a determinant of diving fatalities, laws prohibiting cigarette smoking while driving can reduce motor vehicle crashes; hence, preventing deaths from unintentional injuries as well as curbing the economic cost associated with driving accidents. The findings of this paper indicate that higher cigarette prices are associated with a reduction in non-alcohol related fatal accidents among 16-to 20-year olds. The estimates are not negligible a dollar increase in price per pack of cigarette is associated with a reduction in fatal accidents among 16-to 20- year olds by 5 to 7 percent. Performing calculation at the mean suggests approximately 5 less fatal accidents among this age group. I conduct several tests to check the validity of the findings. Results from first falsification exercise indicate that higher cigarette prices have no effect on fatal crashes associated with 30 year olds and over. In a second falsification exercise, I isolate the fatal cases to those that are most likely to result due to distracted driving single-vehicle, sole occupant. The evidence indicates that increases in cigarette prices reduce single-vehicle, single-occupant crashes but has no effect on multiplevehicles, multiple-occupants crashes. Results from falsification exercises and several robustness checks provide suggestive evidence that the main findings of this paper are not spurious. I. Background Driver s inattention is one of the major determinants of traffic accidents. About 20 to 50 percent of crashes are related to some form of inattention (Goodman, Bents, Tijerina, Wirwille, Lerner, & Benel, 1997; Ranney, Garrott, & Goodman, 2001). Performing secondary task can increase risk of driving crashes as it is cognitively demanding to be involved in dual activities. Participating in secondary activity 5
can distract a driver, reduce attention level, and lessen reaction time required to respond to unexpected road hazards (Morton, Ouimet, Zhang; 2011). Texting is considered a major determinant of distracted driving crashes. The majority of focus in the realm of distracted driving has been provided to curtail texting or cell phone usage while operating a vehicle. Since 2007, several states have passed specific laws prohibiting text messages while driving. 7 Abouk and Adams (2013) find that texting enforced as primary offense can reduce single vehicle, singleoccupant accidents; whereas, the secondary offense laws have no effect on driving crashes. The authors find that the effect of laws prohibiting texting is short lived. Abouk and Adams conclude that their results are mainly being driven by announcement effect rather than the laws itself. Perhaps, not as serious as texting, it is reasonable that smoking while driving can increase risk of motor vehicle accidents. Unfortunately, the existing literature in driving fatalities and accidents provide scant knowledge regarding this matter smoking is only speculated to increase risk of driving accidents; however, the empirical results are non-existing. Over the past decades, several states have enacted smoke free air laws prohibiting smoking in bars, restaurants, and workplaces. Such location bans in smoking can in return increase the propensity of a smoker to smoke while operating a vehicle. Smoking creates a number of distractions including visual, cognitive, and manual interruptions. Some common scenarios of smoking-related distractions could be: lighting the cigarette, when the driver might have to let both hands off the steering wheel; reaching for a cigarette or cigarette lighter while driving; and expelling ash either into the ashtray or out of the window during the process of smoking. A study conducted by National Institutes of Health reports that cigarette smokers averaged 12.0 seconds of distraction; whereas, cell phone users averaged 10.6 seconds of distraction. Specifically, young adults who are novice drivers can be vulnerable to distracted driving crashes due to smoking. A study conducted by Klauer et al. (2014) finds that adolescents using cell phones were more likely than adult experienced drivers using cell phones to drive through a red or yellow light. 7 These laws include cell phone usage while operating a motor vehicle being considered as primary or secondary offenses. Washington was the first state to start in January 2008, 32 other states enacted laws to prohibit cell phones while driving by 2012. 6
Situation can potentially be similar in the context of smoking. Younger individuals tend to be more risky than the older ones; hence, they are more likely to take risks while driving. Cigarette smoking may further exacerbate the scenario. In addition, younger adults are more responsive to higher cigarette prices compared to older adults. Sloan and Trogdon (2004) demonstrate that price elasticity of cigarette for 18-to 20-year olds is -0.27, and decreases in magnitude with age. Past studies indicate that increases in cigarette prices reduce both smoking prevalence and conditional quantity demanded of cigarettes among youth. The majority of studies focusing on smoking habits among youths conclude that youths are up to three times as responsive to prices as are adults (Tauras et al., 2001). Older individuals find it hard to quit smoking due to already established pattern of habit; hence, are less responsive to higher cigarette prices. Considering the vulnerability of novice drivers and responsiveness of younger adults to increases in cigarette prices, this paper focuses on 16-to 20-year olds. If smoking increases the likelihood of road accidents, individuals smoking while driving can be underestimating risk of his/her action. In such a case, the smoking driver fails to internalize costs associated with his/her conduct. For example, during the process of smoking while driving, if a driver crashes onto an innocent bystander, the driver involved in smoking fails to internalize the cost of hitting the bystander. Smoking related driving crashes increases social costs of smoking due to externality imposed on the bystander and his/her family. Similarly, externality can arise in the forms of property damage, road congestion, and medical expenditures. Failure to account for such externalities leads to excess quantity demanded of cigarettes than optimal quantity demanded, after accounting for the external costs. This results to a market failure which may validate government intervention to make smokers internalize the cost of smoking while driving. Thus, understanding whether smoking increases the risk of driving accidents can aid policy making process. II. Data and Empirical Methods A. Data 7
The primary data used in this study comes from the Fatality Analysis Reporting Systems (FARS) of NHSTA (1998 to 2006). 8 The reported data in FARS is a nationwide census of fatal motor vehicle crashes. The main variable of interest is the number of non-alcohol related fatal accidents in a state at a given year which involved divers of age group 16 to 20. As a falsification exercise, I estimate the effect of cigarette price increases on non-alcohol related fatal accidents among individuals 30 years and over. This practice allows comparison across age groups to provide evidence on whether some other underlying trends are affecting the results pertaining to 16-to 20- year olds. I follow NHSTA procedure which is used to generate their official statistics, to calculate the number of non-alcohol driving crashes for 16-to 20-year olds for each state and a given year (DOT HS 809 403, NHTSA; 2002). The main variable of interest is the number of fatal accidents which involves at least one driver of age 16 to 20 given that the blood alcohol level (BAC) of that driver is zero. 9 I use three measures of non-alcohol related driving fatalities: 1) Log rate of driving fatalities, 2) Log of driving fatalities; and 3) Count of driving fatalities. To calculate the rate, similar to the one used in Adams et al. (2011), I compute the percentage of individuals at risk by dividing the respective year and state specific crash counts by age-specific population data. The age specific population data is extracted from the U.S. Census Bureau. Although Federal mandates require measure of BAC level in cases of fatal accidents, for roughly half the accidents BAC levels are unreported. The missing BAC level in the FARS dataset is imputed by using a general location model which models the probability of having a positive BAC level (See Subramanian & Utter, 1998 for more detail). The imputed value of BAC level depends on factors such as age, gender, safety belt or helmet use, license expiration, prior traffic convictions, day of the week, time of the day, the role of the vehicle in the accident, whether the car remains on the road, they type of vehicle driven, and whether police at the accident believed drinking was involved (Adams et al., 2011). 8 I limit my analysis until 2006 to isolate the effects of laws regarding cell phone usage while operating a vehicle, which were enacted starting from 2007. 9 This gives a measure of total number of accidents involving drivers of age 16 to 20 who were not driving under the influence. 8
The imputation of BAC level can lead to measurement error issue if higher cigarette prices are correlated to actual drinking-related driving accidents. Two possible scenarios can arise due to the strong association between cigarettes and alcohol: 1) Higher cigarette prices increasing drunk-driving fatalities; and 2) Increases in cigarette prices decreasing drunk-driving accidents. The former is a result of individuals substituting alcohol for cigarettes, and latter is a case when cigarettes and alcohol are complements. If higher cigarette prices increase actual incidences of drinking and driving, it raises the probability of alcohol-related driving fatalities to be considered as non-drinking fatal accidents. This in fact will underestimate the effect of increases in cigarette prices on non-alcohol driving fatal accidents. On the other hand, if higher cigarette prices decreases drunk-driving (if alcohol and cigarettes are complements), a reduction in drunk-driving fatal accidents can falsely be acknowledged as a reduction in non-drinking fatal accidents. Such measurement error can overestimate the effect of higher cigarette prices on non-drinking fatal accidents. However, using similar timeframe as of this study, Shrestha (2014) finds that increases in cigarette prices following the MSA led young adults to substitute cigarettes for alcohol. Furthermore, the effect of substitution is concentrated among heavy drinkers. He concludes that if anything, increases in cigarette prices are positively associated with drunk driving fatal accidents. Data for cigarette prices comes from the Tax Burden on Tobacco (Orzechowski and Walker, 2011). The prices used are the weighted averages (by the market share) for a pack of 20 cigarettes and are inclusive of taxes. Cigarette prices are then converted to 2006 dollars by using the Consumer Price Index (CPI). The price per pack of cigarettes is considered than tax on that pack. Price captures the exogenous variations generating from differences in transportation costs, retailing costs; and Herfindahl index among states (see Chou et al., 2006 for details). As a form of smoke free air laws (SFA), smoking bans in bars are considered in the main model; whereas smoking bans in private and workplaces are considered in alternative specifications. The data for smoking ban is extracted from project ImpacTeen (http://www.impacteen.org/tobaccodata.htm). Other variables considered in this study are: 1) Beer taxes; 2) Gas Prices; 3) Minimum Wage; 4) Percentage of 16-to 20-year olds wearing seatbelt; 5) Per-capita 9
income; 6) State unemployment rate; and 7) States population. The data sources and discussion regarding these variables are included in the Appendix section. Figure 1 shows the trend in non-alcohol related driving fatalities along with cigarette prices. The graph demonstrates that non-alcohol driving fatalities for 16-to 20-year olds start declining after increases in cigarette prices. Following the Master Settlement Agreement, cigarette prices increased overtime across all states. Table 1 provides the summary statistics of the variables used in this study. The determinants of non-alcohol related driving fatalities can be given by: D ss = f(c ss, X ss, S, Y) (1) where, D ss represents non-alcohol related fatal accidents in state s at time t, C ss is per capita cigarette sales, X represent other time varying characteristics of the states which are likely to determine motor vehicle accidents, and S aaa Y represent state and year fixed effects, respectively. Relying upon the estimates of equation (1) can be problematic due to the potential endogeneity of cigarette consumption. 10 Substituting C in equation (1) by plausibly exogenous factor of smoking, cigarette prices, the study concentrates on the reduced form estimation, which is given by: D ss = f(p ss, X ss, S, Y) (2) where, all the variables are similar to those from equation (1) with an exception of P ss representing price per cigarette pack. Besides facilitating a relatively more exogenous factor of smoking (cigarette prices), findings from the reduced form model portrayed by equation (2) has direct policy relevance. B. Empirical Methods The effect of increases in cigarette prices on non-alcohol related driving fatalities is identified by using within state variation in cigarette prices over time. Specifically, the estimation strategy is based on following specification: D ss = δ 1 + δ 2 Pc ss + δ 3 X ss + λ s + η t + e ss (3) 10 Sloan et al. (2004) concludes that smokers are usually less risk averse compared to non-smokers. Hence, smokers might be willing to take risk while driving, regardless of if he/she smokes while diving. In such a case, it is difficult to identify whether it is the risk taking personality or smoking while operating a vehicle that increases risk of accidents. 10
where, D ss is the measure of non-drinking fatal motor vehicle accidents, Pc ss is the real cigarette prices in state s at time t, X represents state-level time varying characteristics which might influence fatal accidents, λ s is state fixed effects, η t is year fixed effects, and e ss is the error term. Three main measures of non-drinking fatal motor vehicle accidents are used: 1) Log rate of motor vehicle accidents, 2) Log of count of fatal accidents, and 3) Count of motor vehicle accidents. The first two measures are estimated by using OLS with fixed effect; whereas, the third measure uses Poisson count model with fixed effects. For all three cases, robust standard errors to functional form misspecification are estimated. To control for the within cluster correlation, standard errors are clustered at the state level. Including state and year fixed effects in the model can be viewed as an extension to the differences-in-differences frameworks that allows for multiple groups and multiple time periods rather than analyzing a typical two groups (control and treatment) and two period (pre-and-post) setting (Wooldrige, 2001). For an unbiased estimate of increases in cigarette prices, δ 2 in equation (3); there should be no contemporaneous state-level trends that are correlated with increases in cigarette prices and non-alcohol related driving fatalities. State level time varying variables included in X of equation (3) are expected to control for other factors which could affect driving accidents. The identification strategy used in this study depends on an assumption that after the inclusion of fixed effects and other time varying variables, the states experiencing higher magnitude increases in cigarette prices are comparable to those states whose cigarette prices increased by relatively smaller amounts. Even after controlling for fixed effects and other time varying variables, the concern that specification might be failing to account for some unobserved trend in non-alcohol related fatalities correlated with cigarette prices still remains. Though it is unlikely that cigarette prices are endogenous in this case, I test the presence of policy exogeneity by using the lead of cigarette prices. If cigarette prices are endogenous, or in other words, if increases in cigarette prices are large in magnitude among the states already experiencing reduction in driving fatalities, then estimates on lead of cigarette prices should resemble the estimates from equation (3). Thus, the lead effect will provide evidence on whether states 11
experiencing higher cigarette prices had prior differential trends in driving fatalities compared to states with lower increases in cigarette prices. An additional concern could be some other factors, such as, severe weather condition, road construction, and intensity of pedestrians and bicyclists on the road, influencing accident rates. 11 There is little reason to believe that changes in frequency of bicyclists and pedestrians will be systematically correlated to increases in cigarette prices of states. However, states may contribute a certain percent of their tax increases in cigarettes to improve road conditions. It is reasonable that any improvements in road quality or road construction should affect risk of driving accidents among older individuals as well. Nevertheless, it can be argued that due to years of experience in driving, older individuals may be more equip with detrimental road situations; hence, improvement in road conditions may not affect their risk of accidents with a same intensity compared to accidents related to youths. To capture such ongoing changes in a state that may be difficult to observe, I test alternative set of specifications which additionally includes state-specific linear and cubic time trends respectively in equation (3). It has to be noted that inclusion of state-specific time trends limit identification information; however, such models are robust. 12 III. Results A. Basic Results Table 2, Panel A, shows the results after estimating equation (3) where the dependent variable is a measure of non-alcohol related driving fatalities. Model (1) includes log rate of accidents as the dependent variable, model (2) uses log of count of accidents, and model (3) pertains to counts of accidents. The first two models are estimated by OLS with fixed effects; whereas, model (3) relates to Poisson fixed effect estimates. The coefficient on all three models can be interpreted as semi-elasticities. The coefficient on cigarette prices across all three models in Panel A, Table 2, indicates that higher cigarette prices are associated with a reduction in non-alcohol related driving fatalities. The 11 NHTSA estimates that 7 percent of total economic costs related to motor vehicle accidents can be attributed to accidents involving pedestrians and bicyclists. 12 The state-specific trends controls for time varying state specific unobserved changes, which potentially can be correlated to trend in motor vehicle crashes and cigarette prices. 12
coefficient on cigarette prices of model (1) suggests that a dollar increase in cigarette prices is associated with a reduction in non-alcohol related fatal accident rate by 7 percent. Similarly, the effects of cigarette prices from model (2) and model (3) suggest that a dollar increase in a cigarette pack is associated with reduction in non-driving fatal accidents by 7.19 and 5.83 percent respectively. The coefficients on cigarette prices are significant at a 5 percent level across all three models in Panel (A), Table 2. Panel B in Table 2 presents results from a falsification exercise pertaining to non-alcohol related driving fatalities for individuals of 30 years and older. The intuition of this falsification test is that increases in cigarette prices should not affect older individuals due to the less responsive nature to higher cigarette prices and established experience in driving. Any results suggesting a negative effect of increases in cigarette prices on non-alcohol related driving fatalities for this age group would be suggestive of unobserved underlying trends driving the results in Panel A, Table 2. However, the coefficients on cigarette prices reported in Panel B, Table 2, are close to zero and statistically insignificant at the conventional levels. The results in Panel B provide suggestive evidence that the results in Panel A, Table 2, are not spuriously driven. The regression results of Table 2 can be complemented by Figure 2 and Figure 3 which demonstrates trends in non-alcohol related driving fatalities for 16-to 20-year olds and 30 years and over, respectively. The figures are divided according to the size of increases in cigarette prices. Graph representing low states includes those states whose magnitude of increases in cigarette prices fall below 50 th percentile (unconditional), and high states pertains to the states whose magnitude of increases in cigarette prices fall above 50 th percentile (including the 50 th percentile). The graph for low states in Figure 2 shows a noisy pattern of non-alcohol related driving fatalities. In contrast, graph for high states show a clear pattern of a reduction in non-alcohol related driving fatalities among 16-to 20-year olds following increases in cigarette prices. In contrast, both the graphs for ages 30 years and older ( low states and high states ) in Figure 3 show no evident pattern of reduction in non-alcohol related driving fatalities following the rise in cigarette prices. In fact, the graph for low states shows that non-driving fatal accidents mirror an increasing trend in cigarette prices. 13
Table 3 provides additional results after including smoking bans in private workplaces and restaurants separately in the model where log rate of accidents is used as a dependent variable. The SFA variables are included separately due to the potential of multicollieniearity, as states tend to pass SFA laws pertaining to private workplaces, restaurants, and bars at the same time. The coefficients on cigarette prices are similar to that of Table 2. The coefficient on smoking bans in restaurants is positive; however, statistically insignificant at any conventional levels. The coefficient on smoking ban in bars indicates that bar bans are associated with an increase in fatal non-alcohol related accidents. Although, statistically insignificant, magnitude of the coefficient on smoking bans in bars is not ignorable and needs discussion. Results from Table 3, column (3), suggests that smoking bans in bars are associated with 7.79 percent increase in rate of fatal accidents. It is not surprising to notice a negative coefficient on percentage of 16-to 20-year olds wearing seatbelt across all three models in Panel A, Table 2; and in Table 3. Though large in magnitude, the effects are statistically insignificant at the conventional levels which could possibly be driven by lack of within state variation across states. B. Additional Falsification Tests and Robustness Checks The results so far indicate that increases in cigarette prices are associated with a reduction in nonalcohol related motor vehicle fatalities. However, the concern still exist that unobserved trends in driving fatalities correlated with cigarette prices might be driving the results. I conduct an additional falsification exercise and several robustness checks to test the validity of findings reported in Table 2, Panel A. Smoking in an operating vehicle is likely to increase single-vehicle, single-occupant accidents, compared to vehicle consisting with multiple passengers. First, a smoker may not decide to smoke while operating a vehicle due to concerns of secondhand smoke, given that an occupant does not smoke. Second, even if he/she does smoke, an occupant can be handy in the process of smoking (for instance, finding the lighter and lighting up the cigarette). As an additional falsification exercise, I conduct analysis after dividing the crashes between single vehicle, single occupant and those involving multiple vehicles and occupants. The results are presented in Table 4. Panel A pertains to single occupant, single vehicle 14
crashes; whereas, Panel B show results for multiple vehicle or multiple occupants. The findings from Panel A indicates that increases in cigarette prices are associated with a reduction in single vehicle, single occupant driving crashes. In contrast, as expected, the coefficients in Panel B are small and close to zero. Table 4 provides evidence that the findings in Table 2, Panel A, are driven by reduction in single-vehicle, single-occupant accidents ones that are more likely to be smoking related. Table 5 shows the results for 16-to 20-year olds once lead of real cigarette prices is used. The coefficients on lead of cigarette prices should not have any effect on non-alcohol related driving fatalities, as intuitively today s prices cannot affect yesterday s outcome. The coefficients on lead cigarette prices in Table 5 are close to zero and statistically insignificant at any conventional levels. This provides strong suggestive evidence that states experiencing higher magnitude increases in cigarette prices following the MSA are comparable to the states having relatively lower increases in cigarette prices. Additionally, the results from Table 5 provide suggestive evidence regarding the exogeneity of cigarette prices. Table 6 includes state-specific linear time trend as an additional control. The inclusion of statespecific linear time trend accounts for unobserved time varying factors across state which may be influencing both trends in driving accidents and cigarette prices. The coefficients on cigarette prices in Panel A, Table 5, show that inclusion of state-specific linear time trend leaves the results unchanged. However, the magnitude of coefficients on cigarette prices across all models is doubled compared to those in Table 2. Though specification used to estimate results of Table 6 is robust, it comes with a cost of limiting variation in a sample when evaluating the effect of increases in cigarette prices on non-driving fatal accidents. This might explain why the coefficients on cigarette prices are enlarged compared to Table 2, Panel A. Next, I limit the sample to the states enacting text messaging ban and enforcing primary offense between 2007 and 2010. 13 This demonstrates states attitude or urgency in reducing distracted driving fatalities. Table 6 shows results from sub-sample restricted to the states enforcing primary offense for text 13 Primary offense means that the law enforcement officials can stop someone who is suspected of texting while operating a vehicle even though no other offense is committed. 15
messaging between 2007 and 2010. The coefficients on cigarette prices are negative and similar in magnitude to those in Table 2, Panel A. DeCicca et al. (2006) demonstrate that smoking sentiments of the states increased over the 1990s. One possible scenario is that anti-smoking sentiments may decrease smoking rate of 16-to 20-year olds; hence, decrease smoking incidence in motor vehicles. This could itself lead to a reduction in driving fatalities. Failure to account for a measure of anti-smoking sentiments can confound the estimates of cigarette prices. To account for smoking sentiments, I include a measure of anti-smoking sentiments in an additional specification. 14 The results (not shown but available upon request) after an inclusion of antismoking sentiments are virtually unchanged. IV. Conclusion This is the first paper to show that increases in cigarette prices are negatively associated with nonalcohol related driving fatalities among 16-to 20-year olds. In contrast, I find no statistically significant evidence of increases in cigarette prices reducing non-alcohol related driving fatalities for older population. The driving accidents associated with older population are not expected to be affected by higher cigarette price due to their established pattern of smoking habit and inbuilt experience in smoking while driving. The second falsification exercise conducted reveals that higher cigarette prices affect single-vehicle, single-occupant accidents; ones which are likely to be due to distracted driving. Incorporating results from falsification tests and several robustness exercises, the findings of this paper indicates a real effect of cigarette prices increases on non-alcohol related driving fatalities among youths. The estimated effects are not ignorable the findings suggest that a dollar increase in a pack of cigarette is associated with a reduction of non-drinking related fatal accidents by 5.8 percent to 7.8 14 Adapting DeCicca et al. s (2008) strategy, I use attitudes regarding smoking in various places to measure a state s anti-smoking sentiments. The data is obtained from 1995-1996, 1998-1999, 2000-2001, 2002-2003, and 2006-2007 waves of the Current Population Survey Tobacco Use Supplement (TUS-CPS). For bars, restaurants, workplaces, and sporting events, individuals are asked if smoking should be allowed, allowed in some areas, or not allowed. Also, respondents are asked to report their smoking environment at home. I use a principal factor analysis to obtain one latent variable, which represents the anti-smoking sentiments. For each wave, the estimated factor is normalized to have a mean of zero. Estimated factors are then averaged by state and year and I linearly interpolate and extrapolate the estimated factors for the missing years as required. 16
percent for 16-to 20-year olds. The logical explanation is higher cigarette prices reduces smoking rate among youths, which reduces propensity of smoking while operating a moving vehicle. This leads to an increased attention level while driving, which reduces the risk of driving accidents. The results from this paper indicate that youths smoking while operating a vehicle can be underestimating the risk of smoking while driving. The evidence suggests that discouraging smoking while operating a vehicle through the means of increases in cigarette taxes or by prohibiting smoking in motor vehicles may reduce non-drinking related accidents among 16-to 20-year olds. 17
Refrences Adams, Scott, and Chad Cotti. "Drunk driving after the passage of smoking bans in bars." Journal of Public Economics 92.5 (2008): 1288-1305. Adams, Scott, McKinley L. Blackburn, and Chad D. Cotti. "Minimum wages and alcohol-related traffic fatalities among teens." Review of Economics and Statistics 94.3 (2012): 828-840. Abouk, Rahi, and Scott Adams. "Texting bans and fatal accidents on roadways: Do they work? Or do drivers just react to announcements of bans?." American Economic Journal: Applied Economics 5.2 (2013): 179-199. Eby, David W., and Lidia P. Kostyniuk. Driver distraction and crashes: An assessment of crash databases and review of the literature. No. HS-043 601, UMTRI-2003-12. 2003. Goodman, M.J., Bents, F.D., Tijerena, L., Wierwille, W., Lerner, N., & Benel, D. (1997). An investigation of the safety implications of wireless communications in vehicles(dot HS 806-635). Washington, D.C.: National Highway Transportation Safety Administration.http://www.nhtsa.dot.gov/people/injury/research/wireless/ Klauer, Sheila G., et al. "Distracted driving and risk of road crashes among novice and experienced drivers." New England journal of medicine 370.1 (2014): 54-59. Markowitz, Sara. "Where there's smoking, there's fire: The effects of smoking policies on the incidence of fires in the USA." Health economics (2013). National Highway Traffic Safety Administration. "Traffic Safety Facts: 2011 Data: Alcohol-impaired driving." DOT HS 811 (2012): 700. National Highway Traffic Safety Administration. "Traffic Safety Facts: 2010 Data: Alcohol-impaired driving." DOT HS 811 (2012): 700. National Highway Traffic Safety Administration. "Traffic Safety Facts: 2003 Data:Children." DOT HS 809 National Highway Traffic Safety Administration. "Traffic Safety Facts: 2010 Data: Distracted Driving." DOT HS 811 (2012): 650. National Highway Traffic Safety Administration. New NHTSA Study Shows Motor Vehicle Crashes Have $871 Billion Economic and Societal Impact on U.S. Citizens. http://www.nhtsa.gov/about+nhtsa/press+releases/2014/nhtsa-study-shows-vehiclecrashes-have-$871-billion-impact-on-u.s.-economy,-society (accessed October 2, 2014) Ranney, T.A., Garrot, R., and Goodman, M. J. (2001). NHTSA Driver Distraction Research: Past, Present, and Future. 17th ESV Conference, Washington DC, USA. Simons-Morton, Bruce G., et al. "Crash and risky driving involvement among novice adolescent drivers and their parents." Journal Information 101.12 (2011). 18
Sloan, Frank A., and Justin G. Trogdon. "The impact of the master settlement agreement on cigarette consumption." Journal of Policy Analysis and Management 23.4 (2004): 843-855. Shrestha, Vinish. Do Young Adults Substitute Alcohol for Cigarettes Learning from the Master Settlement Agreement. Working Paper (2014) Stutts, J.C., Reinfurt, D.W., Staplin, L., Rodgman, E.A., 2001. The role driver distraction in traffic crashes. Report Prepared for AAA Foundation for Traffic Safety. Retrieved June 10, 2003 from http://www.aaafoundation.org/pdf/distraction.pdf. Subramanian, Rajesh. Transitioning to multiple imputation-a new method to estimate missing blood alcohol concentration (BAC) values in FARS. No. HS-809 403,. 2002. Tauras, John A., Patrick M. O'Malley, and Lloyd D. Johnston. Effects of price and access laws on teenage smoking initiation: a national longitudinal analysis. No. w8331. National Bureau of Economic Research, 2001. Wooldridge, Jeffrey M. Econometric Analysis of Cross Section and Panel Data. MIT press, 2010. 19
Appendix Data Sources A. Beer Taxes: Data for beer taxes comes from the Alcohol Policy Information Systems (APIS). Using CPI, beer taxes are converted to 2006 dollars. B. Gas Prices: Gas price data is extracted from the Office of Highway Policy Information, and gas prices are converted to 2006 dollars. C. Minimum Wage: Minimum wage data is obtained from the January edition of Monthly Labor Review. If changes in minimum wage occurred in middle of the year, weighted average of new and old minimum wages is used, based on the number of months in a year the new minimum wage was in effect. Using CPI, minimum wage is converted to 2006 dollars. D. Percentage of 16-to 20-year olds wearing seatbelts: I construct a state-specific measure of percentage of youths wearing seatbelt while in a car by referring to data from the Youth Risk Behavior Surveillance System. The question asked is, How often do you wear a seatbelt while driving a car? If the person answered most of the time or always, I consider them as individuals inclined to put on a seatbelt while driving. YRBSS conducts survey every other year. The missing values are replaced by state-specific means. E. Per-capita income: State-level per capita income is obtained from the Bureau of Economic Analysis. F. State unemployment rate: Data for state-level unemployment is extracted from the Bureau of Labor Statistics. G. State population: State-level population data is obtained from the U.S. Census Bureau. I use the log of state-level population as a control variable. 20
Figure 1 Non-Alcohol Related Driving Fatalities 105 110 115 120 125 Trend in Cigarette Prices and Driving Fatalities (16-to 20-year olds) 1998 2000 2002 2004 2006 year 2.5 3 3.5 4 Cig Price in 2006 dollars non-alcohol driving fatalities real cigarette price 21
Figure 2 Trend in Cigarette Prices and Driving Fatalities (Low States,16-to 20-year olds) Trend in Cigarette Prices and Driving Fatalities (High States, 16-to 20-year olds) Non-Alcohol Related Driving Fatalities 134 136 138 140 142 2.5 3 3.5 4 Cig Price in 2006 dollars Non-Alcohol Related Driving Fatalities 80 85 90 95 100 2.5 3 3.5 4 4.5 Cig Price in 2006 dollars 1998 2000 2002 2004 2006 year 1998 2000 2002 2004 2006 year non-alcohol driving fatalities real cigarette price non-alcohol driving fatalities real cigarette price 22
Figure 3 Trend in Cigarette Prices and Driving Fatalities (Low States,30 years and up) Trend in Cigarette Prices and Driving Fatalities (High States, 30 year and up) Non-Alcohol Related Driving Fatalities 640 660 680 700 720 2.5 3 3.5 4 Cig Price in 2006 dollars Non-Alcohol Related Driving Fatalities 430 440 450 460 470 2.5 3 3.5 4 4.5 Cig Price in 2006 dollars 1998 2000 2002 2004 2006 year 1998 2000 2002 2004 2006 year non-alcohol driving fatalities real cigarette price non-alcohol driving fatalities real cigarette price 23
Table 1. Summary Statistics Variable Mean Std. Dev. non-alcohol related accidents (16-to 20-years) 117.54 116.53 percentage at risk 0.00 0.00 log of non-alcohol related accidents 4.27 1.10 real cigarette prices (in 2006 dollars) 3.58 0.79 real beer taxes (in 2006 dollars) 27.40 21.48 drvlic21 0.45 0.50 smoking ban in bars 0.06 0.24 log(per capita income) 10.32 0.18 unemployment rate 4.70 1.16 minimum wage (in 2006 dollars) 6.07 0.64 bac 8 0.62 0.49 bac10 0.38 0.49 state population 18900000 23700000 log(state population) 16.11 1.20 proportion wearing seatbelt 0.66 0.13 population of 18 to 24 year olds 550195.20 619113.80 log(population of 18 to 24 year olds) 12.73 1.01 number of observations = 459 24
Table 2. Determinants of Non-Alcohol Related Driving Fatalities Panel A (16-to 20-year) model (1) model (2) model (3) real cigarette price (in 2006 dollars) -0.0778** -0.0719** -0.0583** (0.0360) (0.0330) (0.0290) real beer taxes (in 2006 dollars) 0.0005 0.0005-0.0004 (0.0007) (0.0007) (0.0017) smoking ban in bars 0.0763 0.0879-0.0252 (0.0757) (0.0767) (0.0762) minimum wage (in 2006 dollars) -0.0072 0.0119 0.0664 (0.0361) (0.0384) (0.0446) blood alcohol concentration (0.08) -0.0776-0.0026-0.0047 (0.0625) (0.0271) (0.0258) seatbelt -0.1110-0.1026-0.1796 (0.1414) (0.1395) (0.1477) N 459 459 459 Panel B (30 years and up) model (1) model (2) model (3) real cigarette price (in 2006 dollars) -0.0326-0.0134 0.0090 (0.0238) (0.0237) (0.0142) real beer taxes (in 2006 dollars) 0.0038*** 0.0037*** 0.0032** (0.0008) (0.0006) (0.0013) smoking ban in bars 0.0074 0.0264-0.0599 (0.0553) (0.0562) (0.0414) minimum wage (in 2006 dollars) -0.0652*** -0.0275 0.0037 (0.0221) (0.0221) (0.0226) blood alcohol concentration (0.08) 0.0383 0.0236 0.0198 (0.0404) (0.0193) (0.0140) seatbelt -0.0297-0.0183-0.1641* (0.1180) (0.1076) (0.0857) N 459 459 459 Note: All regression includes both state and fixed effects. The dependent variable for model (1), model (2), and model (3) are log of accident rate (count of accident/age-specific population), log of count of accidents, and count of accidents. Model (1) and model (2) are estimated by OLS with fixed effects. Model (3) is estimated by Poisson fixed effect model. Additionally, models control for log of per capita income, state unemployment rate, log of state population, population of 16-to 20-year olds (not in model 1), and percentage of 16-to 20-year olds wearing seatbelt. Standard errors reported in parenthesis are clustered at the state level to allow for heteroskedasticity and correlation within each cluster. * indicates p<0.10, ** indicates p<0.05, and *** represents p<0.01. 25
Table 3. Including Venue Specific SFA laws 16-to 20-year olds model (1) model (2) model (3) real cigarette price (in 2006 dollars) -0.0663** -0.0730** -0.0808** private workplace smoking bans -0.0147 (0.0328) (0.0347) (0.0341) (0.0803) smoking bans in restaurants 0.0297 (0.0661) smoking ban in bars 0.0779 (0.0754) real beer taxes 0.0002 0.0003 0.0005 (0.0006) (0.0007) (0.0007) minimum wage (in 2006 dollars) 0.0091 0.0026-0.0071 (0.0298) (0.0350) (0.0361) blood alcohol concentration (0.08) -0.0110-0.0106-0.0113 (0.0257) (0.0262) (0.0268) seatbelt -0.1199-0.1177-0.1089 (0.1403) (0.1398) (0.1405) N 459 459 459 r2 0.8689 0.8690 0.8696 Note: All regression includes both state and fixed effects. The dependent variable for model (1), model (2), and model (3) are log of accident rate (count of accident/age-specific population), log of count of accidents, and count of accidents. Model (1) and model (2) are estimated by OLS with fixed effects. Model (3) is estimated by Poisson fixed effect model. Additionally, models control for log of per capita income, state unemployment rate, log of state population, population of 16-to 20-year olds (not in model 1), and percentage of 16-to 20-year olds wearing seatbelt. Standard errors reported in parenthesis are clustered at the state level to allow for heteroskedasticity and correlation within each cluster. * indicates p<0.10, ** indicates p<0.05, and *** represents p<0.01. 26
Table 4. Falsification Exercise 2 (Single Vehicle, Single Occupant Versus Multiple Vehicle, Multiple Occupants) Panel A (single occupant, single vehicle) model (1) model (2) model (3) real cigarette price (in 2006 dollars) -0.0827** -0.0761** -0.0597** 27 (0.0350) (0.0323) (0.0274) real beer taxes (in 2006 dollars) 0.0001 0.0001-0.0008 (0.0007) (0.0008) (0.0018) smoking ban in bars 0.0600 0.0714-0.0319 (0.0751) (0.0759) (0.0701) minimum wage (in 2006 dollars) -0.0093 0.0100 0.0652 (0.0353) (0.0372) (0.0419) blood alcohol concentration (0.08) -0.0671-0.0055-0.0078 (0.0618) (0.0262) (0.0253) seatbelt -0.1500-0.1420-0.2282 (0.1317) (0.1300) (0.1418) N 459 459 459 Panel B (multiple vehicles or multiple occupants) model (1) model (2) model (3) real cigarette price (in 2006 dollars) -0.0015-0.0020-0.0007* (0.0024) (0.0024) (0.0004) real beer taxes (in 2006 dollars) -0.0170*** -0.0168*** -0.0023 (0.0031) (0.0033) (0.0022) smoking ban in bars -0.1407-0.2016-0.1000 (0.3066) (0.2979) (0.1137) minimum wage (in 2006 dollars) -0.0292-0.1428 0.0788 (0.1189) (0.1100) (0.0638) blood alcohol concentration (0.08) 0.0131-0.0867 0.0236 (0.1749) (0.1139) (0.0433) seatbelt -0.1736-0.2126-0.2165 (0.3049) (0.3257) (0.1899) N 459 459 459 Note: All regression includes both state and fixed effects. The dependent variable for model (1), model (2), and model (3) are log of accident rate (count of accident/age-specific population), log of count of accidents, and count of accidents. Model (1) and model (2) are estimated by OLS with fixed effects. Model (3) is estimated by Poisson fixed effect model. Standard errors reported in parenthesis are clustered at the state level to allow for heteroskedasticity and correlation within each cluster. * indicates p<0.10, ** indicates p<0.05, and *** represents p<0.01.
Table 5. Using the Lead of Cigarette Prices 16-to 20-year olds model (1) model (2) model (3) real cigarette price (in 2006 dollars) -0.0081-0.0006-0.0249 (0.0460) (0.0434) (0.0254) real beer taxes (in 2006 dollars) 0.0002 0.0001-0.0010 (0.0009) (0.0009) (0.0018) smoking ban in bars 0.0478 0.0537-0.0857 (0.1299) (0.1307) (0.0929) minimum wage (in 2006 dollars) 0.0081 0.0216 0.0884* (0.0400) (0.0442) (0.0532) blood alcohol concentration (0.08) -0.0957-0.0255-0.0014 (0.0755) (0.0360) (0.0268) seatbelt -0.1033-0.0976-0.1894 (0.1553) (0.1541) (0.1473) N 408 408 408 Note: All regression includes both state and fixed effects. The dependent variable for model (1), model (2), and model (3) are log of accident rate (count of accident/age-specific population), log of count of accidents, and count of accidents. Model (1) and model (2) are estimated by OLS with fixed effects. Model (3) is estimated by Poisson fixed effect model. Additionally, models control for log of per capita income, state unemployment rate, log of state population, population of 16-to 20-year olds (not in model 1), and percentage of 16-to 20-year olds wearing seatbelt. Standard errors reported in parenthesis are clustered at the state level to allow for heteroskedasticity and correlation within each cluster. * indicates p<0.10, ** indicates p<0.05, and *** represents p<0.01. 28
Table 6. Determinants of Non-Alcohol Related Driving Fatalities with State Linear Time Trends Panel A (16-to 20-year) model (1) model (2) model (3) real cigarette price (in 2006 dollars) -0.1622*** -0.1651*** -0.1248*** (0.0486) (0.0494) (0.0362) real beer taxes (in 2006 dollars) 0.0009 0.0022* 0.0027 (0.0011) (0.0013) (0.0017) smoking ban in bars 0.0151 0.0119-0.0034 (0.0934) (0.0986) (0.0672) minimum wage (in 2006 dollars) -0.0531-0.0403 0.0358 (0.0622) (0.0715) (0.0330) blood alcohol concentration (0.08) -0.0904-0.0174-0.0168 (0.0908) (0.0352) (0.0309) seatbelt -0.1164-0.1184-0.0844 (0.1492) (0.1480) (0.1203) N 459 459 459 Panel B(30 years and up) model (1) model (2) model (3) real cigarette price (in 2006 dollars) -0.0204-0.0202 0.0010 (0.0291) (0.0295) (0.0168) real beer taxes (in 2006 dollars) 0.0003 0.0019** 0.0017 (0.0008) (0.0008) (0.0013) smoking ban in bars -0.0011-0.0094-0.0312 (0.0651) (0.0636) (0.0364) minimum wage (in 2006 dollars) -0.0691* -0.0550* -0.0111 (0.0371) (0.0320) (0.0175) blood alcohol concentration (0.08) 0.0184 0.0149 0.0090 (0.0402) (0.0210) (0.0156) seatbelt 0.0421 0.0373-0.0805 (0.1238) (0.1145) (0.0716) N 459 459 459 Note: All regression includes both state and fixed effects. The dependent variable for model (1), model (2), and model (3) are log of accident rate (count of accident/age-specific population), log of count of accidents, and count of accidents. Model (1) and model (2) are estimated by OLS with fixed effects. Model (3) is estimated by Poisson fixed effect model. Additionally, models control for log of per capita income, state unemployment rate, log of state population, population of 16-to 20-year olds (not in model 1), and percentage of 16-to 20-year olds wearing seatbelt. Standard errors reported in parenthesis are clustered at the state level to allow for heteroskedasticity and correlation within each cluster. * indicates p<0.10, ** indicates p<0.05, and *** represents p<0.01. 29
Table 7. Limiting the Sample to States enacting text message bans (primary offense) between 2007 and 2010 16-to 20-year olds model (1) model (2) model (3) real cigarette prices -0.0798-0.0945** -0.0999** (0.0478) (0.0436) (0.0470) real beer taxes -0.0078-0.0079-0.0080 (0.0067) (0.0072) (0.0074) smoking ban in bars 0.0742 0.0919 0.1453* (0.0563) (0.0732) (0.0794) real minimum wage 0.0303 0.0537 0.0888* (0.0395) (0.0330) (0.0465) blood alcohol concentration (0.08) -0.1945*** -0.0281-0.0443 (0.0430) (0.0393) (0.0332) seatbelt -0.0786-0.0769-0.2174 (0.2208) (0.2130) (0.1817) N 216 216 216 Note: All regression includes both state and fixed effects. The dependent variable for model (1), model (2), and model (3) are log of accident rate (count of accident/age-specific population), log of count of accidents, and count of accidents. Model (1) and model (2) are estimated by OLS with fixed effects. Model (3) is estimated by Poisson fixed effect model. Additionally, models control for log of per capita income, state unemployment rate, log of state population, population of 16-to 20-year olds (not in model 1), and percentage of 16-to 20-year olds wearing seatbelt. Standard errors reported in parenthesis are clustered at the state level to allow for heteroskedasticity and correlation within each cluster. * indicates p<0.10, ** indicates p<0.05, and *** represents p<0.01. 30