Journal of Health Economics
A R R A A
J I I
K M A F I
l M i A f I c b ( i b
Journal of Health Economics 28 (2009) 831–838
Contents lists available at ScienceDirect
Journal of Health Economics
journa l homepage: www.e lsev ier .com/ locate /econbase
ublic policies and motorcycle safety
ichael T. French a,∗, Gulcin Gumus b,c, Jenny F. Homer d
Health Economics Research Group, Department of Sociology, Department of Epidemiology and Public Health, and Department of Economics, University of Miami, 202 University Drive, Merrick Building, Room 121F, P.O. Box 248162, Coral Gables, FL 33124-2030, USA Department of Health Policy and Management, Department of Economics, Florida International University, 11200 S.W. 8th Street, HLSII 554A, Miami, FL 33199, USA IZA, Bonn, Germany Health Economics Research Group, Department of Sociology, University of Miami, Sociology Research Center, 5665 Ponce de Leon Blvd., Flipse Building, Room 104, .O. Box 248251, Coral Gables, FL 33124-0719, USA
r t i c l e i n f o
rticle history: eceived 22 October 2008 eceived in revised form 7 May 2009 ccepted 18 May 2009 vailable online 27 May 2009
a b s t r a c t
Numerous studies have examined the effectiveness of alcohol and traffic policies in reducing automobile crashes and fatalities, but only a few have analyzed the impact of state-specific policies on motorcycle safety. Given the growing popularity and inherent safety risks of motorcycle riding, this study provides a comprehensive investigation of both fatal and non-fatal injuries. State-level longitudinal data from 1990 to 2005 are analyzed to determine how various alcohol and traffic policies impact motorcycle safety and whether there are differential effects by type of injury. The results consistently show that universal helmet
eywords: otorcycle safety
lcohol and traffic policies
laws have the most significant effect on both non-fatal and fatal injuries. Mandatory rider education programs and speed limits on rural interstates significantly impact non-fatal injuries.
© 2009 Published by Elsevier B.V.
1 c 2
r a I d e i 2 t 3
“There are two types of motorcyclists: those who have fallen and those who will.” Motorcycle Safety Foundation (MSF) Instructor
Although motorcycle riding has become increasingly popu- ar in recent years, it remains a risky form of transportation.
otorcycle registrations in the U.S. increased from 4.26 million n 1990 to 6.69 million in 2006 (National Highway Traffic Safety dministration [NHTSA], 2007), while motorcycle sales increased
rom 278,000 units in 1992 to 1.1 million units in 2007 (Motorcycle ndustry Council, 2006; Welsh, 2008). The number of motorcy- le rider fatalities declined throughout the 1980s and early 1990s
ut began increasing in the late 1990s. According to the NHTSA 2008), 4810 motorcycle riders were killed and 88,000 were injured n the U.S. in 2006.1 During this same time period, the num- er of registered passenger cars increased from 123 million in
∗ Corresponding author. Tel.: +1 305 284 6039; fax: +1 305 284 5310. E-mail addresses: email@example.com (M.T. French), firstname.lastname@example.org
G. Gumus), email@example.com (J.F. Homer). 1 Figures displaying these trends can be found in French et al. (2008).
h r a s
i c o a s
167-6296/$ – see front matter © 2009 Published by Elsevier B.V. oi:10.1016/j.jhealeco.2009.05.002
990 to 136 million in 2006 while the number of passenger ar occupants killed decreased from 24,092 to 17,800 (NHTSA, 007).
A large proportion of motorcycle crashes and fatalities involve iders who lack a proper license or training, are speeding, and/or re not wearing a safety helmet (Hurt et al., 1981; NHTSA, 2008). ncreases in motorcyclist fatalities may also be related to the ecisions of several states to rescind helmet laws after Congress liminated sanctions against states without universal helmet laws n 1995 (Sass and Zimmerman, 2000; Houston and Richardson, 008). An obvious risk factor for motorcyclists that has received lit- le attention in the literature is alcohol consumption. An estimated 4 percent of all motorcyclists who were fatally injured in 2006 ad BAC levels above 0.01 g/dL (NHTSA, 2008). Riding a motorcycle equires more strength, coordination, and attention than driving an utomobile, all of which can be severely impaired after consuming everal alcoholic drinks.
In light of the increases in fatal and non-fatal motorcycle rider
njuries and the public health burden associated with motorcycle rashes, a Department of Transportation Report (U.S. Department f Transportation, 2007) recently referred to motorcycle fatalities s “our Nation’s greatest highway safety challenge.” The present tudy contributes important new information in this area by
8 ealth E
f m m m f g g d
c m o a m 1 2 S e ( l c e F b e t e K
t t m i ( ( a t A g r
s F t f a u i i t
t s i b t s a
m r c s e m M e 1 a a c v (
a ( R w a f b A c w E o a p i c a s p o 1 t g
t c T d h f e
32 M.T. French et al. / Journal of H
ocusing on three alcohol policies and three traffic policies to deter- ine whether policy interventions can be effective in improving otorcycle safety. Unlike most existing studies of automobile and otorcycle safety, we examine predictors of non-fatal as well as
atal motorcycle injuries using an extensive set of state-specific lon- itudinal data from 1990 to 2005. The study findings provide initial uidance for the formulation of policy and rider safety recommen- ations and are used to highlight areas for future research.
The role of public policies in reducing fatalities among passenger ar occupants has been studied extensively. Rigorous econometric ethods have been applied to more accurately assess the impact
f these policies by taking into account differences across states nd time, simultaneous changes in other policies, and environ- ental conditions that could influence drinking behavior (Ruhm,
996; Mast et al., 1999; Eisenberg, 2003; Morrisey and Grabowski, 005). Research indicates that more stringent BAC laws (Dee, 2001; hults et al., 2001; Eisenberg, 2003), zero tolerance laws (Shults t al., 2001; Carpenter, 2004), administrative license revocation ALR) (Grabowski and Morrisey, 2001; Freeman, 2007), and speed imits (Grabowski and Morrisey, 2007) can all reduce motor vehi- le fatalities. Two recent studies reported small or non-significant ffects of BAC laws on motor vehicle fatalities (Eisenberg, 2003; reeman, 2007). Several studies have found higher beer taxes to e associated with fewer motor vehicle fatalities (e.g., Chaloupka t al., 1993; Ruhm, 1996), but more recent research has ques- ioned the magnitude of these estimates (e.g., Dee, 1999; Mast t al., 1999; Grabowski and Morrisey, 2001; Young and Bielinska- wapisz, 2006).
This literature suffers from two major limitations. First, rela- ively few studies have examined the effects of alcohol and other raffic policies on traffic safety for specific types of vehicles such as
otorcycles. Motorcycles account for a greater proportion of fatal- ties (11 percent in 2006) than their share of registered vehicles 3 percent), indicating that this is an important area to research NHTSA, 2008). Second, the vast majority of motor vehicle studies nalyze fatality data from the Fatality Analysis Reporting Sys- em (FARS), a surveillance system administered by the NHTSA.2
lthough non-fatal injuries far outnumber fatalities, the federal overnment has not assembled a comparable and publicly available eporting system for non-fatal injuries in all 50 states.3
Carpenter and Stehr (2008) used the FARS to evaluate whether eatbelt policies reduced serious non-fatal injuries. Because the ARS collects data on crashes where at least one fatality occurred, he analysis could only assess whether seatbelt policies affect non- atal injuries that occur in crashes with at least one fatality. In n example from the motorcycle literature, Coben et al. (2007)
sed data from the Healthcare Cost and Utilization Project, which
ncluded cross-sectional hospital discharge records from 33 states n 2001. The authors showed that motorcycle-related hospitaliza- ions in states without universal helmet laws were more likely to
2 FARS contains detailed data from law enforcement reports about motor vehicle rashes that occurred on public roads in the United States and resulted in a fatality p to 30 days after the crash. 3 Since 1988, the National Automotive Sampling System General Estimates Sys-
em has collected data on motor vehicle crashes (from a nationally representative ample of police reports) that lead to a fatality, injury (possible, non-incapacitating, ncapacitating), or major property damage, but state identifiers are currently not eing made available. The National Electronic Injury Surveillance System, a program o monitor injuries related to consumer products from a nationally representative ample of 99 hospitals in the United States, only began including information on car nd motorcycle-related injuries in 2000 (Christoffel and Gallagher, 2006).
c o b c t h ( s o
b s l c s u
conomics 28 (2009) 831–838
nvolve a death during hospitalization or a more serious injury than ospitalizations in states with universal helmet laws.
As a result of data limitations, much of the existing research on otorcycle safety is either descriptive, published as government
eports, or primarily based on narrow samples from hospital dis- harge data, traffic crash records, or police records from a particular tate or a small number of states over a short period of time. Nev- rtheless, these studies have provided valuable information about otorcyclists and their patterns of risky behaviors (Hurt et al., 1981; ax et al., 1998). Research shows that motorcyclists under the influ-
nce of alcohol are less likely to use helmets (Peek-Asa and Kraus, 996; Bledsoe and Li, 2005) and more likely to speed, drive without valid license, and be involved in single-vehicle crashes (Peek-Asa nd Kraus, 1996; Shankar, 2003). Two studies suggest that motor- ycle operators become impaired (i.e., unable to safely drive their ehicles) at lower BAC levels than other motor vehicle operators Colburn et al., 1993; Sun et al., 1998).
With the exception of universal helmet laws, which are strongly ssociated with lower motorcycle fatality rates in numerous studies Sass and Zimmerman, 2000; Bledsoe and Li, 2005; Houston and ichardson, 2008; Dee, 2009), only a few studies have examined hether other state policies can be used to reduce risky behaviors
mong motorcyclists. Villaveces et al. (2003) compared motorcycle atality rates when certain alcohol-related policies were in effect etween 1980 and 1997 to rates when these policies did not exist. LR laws were associated with reductions in all types of motor- ycle fatalities while stricter BAC laws were strongly associated ith lower motorcycle fatality rates for crashes involving alcohol.
ach policy was considered separately without taking into account ther policies or factors that might affect fatality rates. Houston nd Richardson (2008) evaluated the effects of motorcycle helmet olicies on fatalities while controlling for the minimum legal drink-
ng age, 0.08 BAC per se limit, and speed limit. Of these three policy ontrols, only minimum legal drinking age was significantly associ- ted with lower fatality rates, and only in certain models. Although ome research has supported the effectiveness of rider education rograms in reducing motorcycle crashes and fatalities, estimates f the effect of mandatory programs are not available (Billheimer, 998; McGwin et al., 2004). Rider education programs are impor- ant components of motorcycle safety initiatives supported by rider roups as well as the NHTSA (NHTSA, 2008).
Sass and Zimmerman (2000) conducted one of the few studies hat used a methodology similar to ours to investigate the asso- iation between universal helmet laws and motorcycle fatalities. hey analyzed panel data from 1976 to 1997 and controlled for emographic variables, seat belt policies, speed limits, and alco- ol consumption. Accounting for state and year fixed-effects, they
ound that helmet laws, alcohol consumption, and per capita police mployment (as a measure of enforcement) were significantly asso- iated with annual adjusted motorcycle fatalities. Yet this analysis nly included data up to 1997, the year when motorcycle fatalities egan increasing again in the United States. Using more recent data ould reveal additional factors that have contributed to the upward rend in fatalities. In addition, instead of evaluating specific alco- ol policies such as BAC limits and DUI laws, Sass and Zimmerman 2000) used alcohol consumption per capita as a composite mea- ure of the impact of these alcohol policies. Consequently, the role f specific alcohol policies could not be determined.
Based on this comprehensive review of existing studies, we elieve that the current analysis contributes to the motorcycle
afety literature in several important ways. First, it features a unique ongitudinal dataset on both fatal and non-fatal motorcycle injuries ompiled from numerous government reports and personal corre- pondence with representatives from many state agencies. Second, nlike most of the motorcycle studies noted above, it evaluates
M.T. French et al. / Journal of Health Economics 28 (2009) 831–838 833
Table 1 Variable definitions and summary statistics (N = 768 unless indicated otherwise).
Variable Definition Mean St.Dev. Min Max
Total non-fatal injuries (N = 574)a Total non-fatal motorcycle rider injuries
1472 1590 69 11,043
Non-fatal injuries per 100,000 people (N = 574)a Non-fatal motorcycle rider injuries per 100,000 population of age 15 and above
37.392 15.594 8.266 138.197
Total fatal injuries Total fatal motorcycle rider injuries 56.997 65.603 1 563 Fatal injuries per 100,000 people Fatal motorcycle rider injuries per
100,000 population of age 15 and above 1.401 0.639 0.195 4.798
Motorcycle registrations per 100,000 peopleb Number of two-wheeled and three-wheeled motorcycles per 100,000 population of age 15 and above
2475 1347 614 8,850
Traffic policies Universal helmet law Mandatory helmet requirement for all
riders 0.469 0.499 0 1
Mandatory rider education program State legislated or sponsored rider education program that is mandatory for all or some riders
0.419 0.494 0 1
Speed limit on rural interstatesc Maximum legal speed limit on rural interstates (mph)
66.914 4.809 55 75
Alcohol policies BAC limit ≤ 0.08d Maximum allowable blood alcohol
concentration (BAC) of driver ≤0.08 g/dL
0.374 0.484 0 1
Zero tolerance laws Zero tolerance law with the BAC limit = 0.00 g/dL for individuals under age 21
0.177 0.382 0 1
Administrative license revocation Law enforcement can suspend or revoke a license of someone who fails/refuses to take an alcohol test after a traffic stop or vehicle crash
0.738 0.440 0 1
a Data on motorcycle non-fatal injuries for the states of New Jersey, Vermont, and Washington were not available for any year of our analysis period. In addition, state- and y
igned the h
m o k m i c a r b m
n a i a p p C
a a n t o r o o
s c t s s w b f F a M m t r
ear-specific non-fatal injury data were missing for an additional 146 observations. b Includes mopeds and scooters in states that require them to be registered. c Three state/year observations did not have an explicit speed limit; and were ass d Six state/year observations did not have an explicit BAC limit and were assigned
ultiple public policies and employs statistical methods capable f accounting for many of the relevant factors and policies. To our nowledge, the present study is the first to apply rigorous econo- etric methods to a large dataset on fatal and non-fatal motorcycle
njuries, alcohol and traffic policies, and many other state-specific ontrol variables. These estimation techniques have been applied to utomobile-specific data but have not been extended to motorcycle iders. Thus, the results provide new insight into the relationships etween alcohol policies, traffic policies, and fatal and non-fatal otorcycle injuries.
This study uses state-specific longitudinal data for the conti- ental U.S. from 1990 to 2005 to evaluate the effects of alcohol nd traffic safety policies on motorcycle rider fatal and non-fatal njuries.4 The list of variable definitions for the injury measures nd policy variables can be found in Table 1. French et al. (2008) resent the full list of sources for all variables. Consistent with the revious literature, we exclude Alaska, Hawaii, and the District of olumbia.
.1. Outcome measures
Previous studies of traffic fatalities have estimated the effect of olicies on the fatality rate (i.e., number of fatalities per capita or er vehicle mile traveled) (e.g. Ruhm, 1996; Dee, 1999, 2001; Sass
4 “Motorcyclist” in this paper is a term that refers both to motorcycle drivers and o passengers.
the highest observed speed limit of 75 mph. ighest observed BAC limit of 0.12.
nd Zimmerman, 2000; Eisenberg, 2003; Freeman, 2007; Houston nd Richardson, 2008). Since state- and year-specific data on the umber of licensed motorcycle riders and motorcycle vehicle miles raveled are not available, we evaluated the effects of public policies n three main fatal and non-fatal injury measures: total motorcycle ider fatality count, fatalities per 100,000 people aged 15 years and lder, and non-fatal injuries per 100,000 people aged 15 years and lder.
Fatality figures were requested from the FARS, the surveillance ystem administered by the NHTSA.5 As part of our robustness hecks, we used the extensive crash characteristics available in FARS o investigate whether public policies have differential impacts on ix additional outcomes (weekend, weekday, nighttime, daytime, ingle-vehicle, and multi-vehicle fatalities). Weekend fatalities ere defined as motorcycle riders killed in traffic crashes occurring
etween 6:00 p.m. on Friday and 6:00 a.m. on Monday. Weekday atalities occur between 6:00 a.m. on Monday and 6:00 p.m. on riday. Daytime fatalities occur between 6:00 a.m. and 6:00 p.m., nd nighttime fatalities occur between 6:00 p.m. and 6:00 a.m. otorcyclist fatalities that occurred in crashes involving only otorcycles are referred to as “single-vehicle fatalities” while those
hat occurred in crashes involving other types of vehicles are eferred to as “multi-vehicle fatalities.” All fatality data used in his study were based only on motorcycles and exclude scooters,
opeds, and off-road vehicles. Given the lack of a national registry or other database compa-
able to FARS for non-fatal injuries, we contacted individual state gencies to request total annual counts of non-fatal motorcycle
5 Data requests can be made through the FARS website (www.fars.nhtsa.dog.gov).
8 ealth E
i a t t l y t i a
a a l l i a i Y i o
w m i r o c m t
t ( r d
i 2 fi A m u a b m
m a 1
d q r p a t p A s i
( R i s a c t r m
a d r m w p p r a e r p y d s
s e a n fi 2
34 M.T. French et al. / Journal of H
njuries beginning in 1990. Although some states included mopeds nd scooters in their injury counts and were not able to separate hem out, these vehicles make up a very small proportion of all wo-wheeled vehicles in any state. Since a few states did not col- ect any injury data and not all states had complete data for every ear, thus the panel used in this analysis is unbalanced.6 Despite he different reporting systems for fatal and non-fatal motorcycle njuries, investigative analyses confirm that the within-state trends re similar (French et al., 2008).7
.2. Policy variables
.2.1. Alcohol policies Three binary indicators were constructed to identify whether
state had an ALR policy, a zero tolerance law (a law mandating BAC limit of 0.00 for drivers under 21 years of age), and a BAC
imit of less than or equal to 0.08 g/dL. In states with an ALR policy, icensing authorities or law enforcement can suspend or revoke an ndividual’s license if a driver fails or refuses to take an alcohol test fter a traffic stop or vehicle crash. Given concerns about the min- mal within-state variations over time in alcohol taxes (Dee, 1999; oung and Bielinska-Kwapisz, 2006), we did not include this policy n our core specifications and instead used it to test the sensitivity f our estimates.
We expect that the presence of more stringent alcohol policies ill reduce motorcycle fatalities in several ways. First, motorcyclists ay abstain from drinking before riding or may ride more carefully
f they have been drinking. Second, they might actually change their iding patterns by riding less frequently or using a different means f transportation when they plan on drinking. Finally, these policies ould influence the drinking and driving behavior of other drivers, aking the roads safer for motorcyclists and possibly decreasing
he risk of a collision with another motor vehicle.
.2.2. Traffic policies The maximum speed limit in each state was entered as a con-
inuous variable, while the presence of a universal helmet law requiring riders of all ages to use a helmet) and a mandatory ider education program (for all or some riders) were included as ichotomous measures.
Studies have reported that motor vehicle fatality rates increased n states that raised their speed limits (Grabowski and Morrisey, 007). Traveling at higher speeds makes avoiding a crash more dif- cult and, if a crash occurs, may lead to more severe consequences. lthough the alcohol policies and maximum speed limit apply to otorcycle riders as well as drivers of other types of motor vehicles,
niversal helmet policies and mandatory rider education programs re intended to affect motorcycle safety by directly impacting the ehavior of motorcycle operators. Helmet use and universal hel- et laws have consistently been associated with lower fatalities
6 Data on non-fatal motorcycle-related injuries for the states of New Jersey, Ver- ont, and Washington were not available for any year of our analysis period. In
ddition, state- and year-specific non-fatal injury data were missing for an additional 46 observations (mostly for earlier years).
7 To the extent possible, we further examined the reliability of the non-fatal injury ata. At the national level, the trends in non-fatal and fatal injury measures are uite consistent (French et al., 2008). For the entire sample, there is a strong cor- elation (0.673, N = 574, p < 0.001) between fatalities per 10,000 people and injuries er 10,000 people. A close examination of the variation in each of these measures s compared to the averages, as well as the comparison of the within-state variation o the overall variation, reveals that fatal and non-fatal injury counts display similar atterns. They also display similar trends within each state (French et al., 2008). lthough we are unable to confirm the reliability of the reporting system in each tate, we are reasonably confident that the states collected and reported non-fatal njury data consistently vis-à-vis the national fatality data.
a o t a o s
e m m
m s m
conomics 28 (2009) 831–838
Sass and Zimmerman, 2000; Bledsoe and Li, 2005; Houston and ichardson, 2008), injury severity (Rowland et al., 1996), and med-
cal costs (Max et al., 1998; Bledsoe et al., 2002). As of 2006, 47 tates had legislated motorcycle rider education programs, which re intended to prevent or reduce the likelihood of crashes. These ourses are required for certain riders (e.g., young riders) prior o licensing in some states. Universal helmet laws and mandatory ider education programs are expected to be associated with fewer
otorcycle fatalities and injuries.
.3. Control variables
A number of control variables are included in the analysis to ccount for demographic, economic, geographic, and traffic con- itions as well as motorcycle usage. The number of motorcycle egistrations is included as an exposure variable in all models since
otorcycle fatalities and injuries occur more frequently in states ith more motorcycles.8 Other control variables include the unem-
loyment rate, income per capita, average annual temperature and recipitation, gasoline prices, lane miles per mile of total public oads, highway maintenance funds per mile of total public roads, nd motor vehicle fatalities per 10,000 registered vehicles. We gen- rated two traffic density variables, one for urban and another for ural areas, by dividing the annual millions of vehicle miles traveled er 1000 residents. Demographic controls included percentage of oung drivers, percentage of white residents, percentage of resi- ents with a bachelor’s or higher degree, and average household ize.
Fatal and non-fatal motorcycle injuries exhibit both between- tate and within-state variation over time. Several previous studies xamining how public policies affect motor vehicle fatalities have ddressed unobserved heterogeneity by using panel data tech- iques and modeling these state-specific factors as time-invariant xed-effects (Ruhm, 1996; Dee, 1999; Morrisey and Grabowski, 005; Freeman, 2007).
Using an approach similar to the earlier literature on motor vehi- le fatalities, we define motorcycle injuries by state and year as a unction of the following form:
st = f (Ast, Mst, Cst) (1)
here yst indicates either fatal or non-fatal injuries for state s in ear t, Ast is a vector of alcohol policy measures, Mst is a vector of utomobile and motorcycle traffic safety policies, and Cst is a vector f other controls such as economic, demographic, and environmen- al factors. Time period t refers to calendar years from 1990 to 2005, nd state s refers to each state. Fatal and non-fatal injuries depend n the observable factors listed above as well as on unobserved tate-specific fixed-effects.
The injury rates depend on the intensity of motorcycle use in
ach state and year, for which we proxy by using the number of otorcycle registrations per 100,000 people.9 Hence, we first esti- ate the following fixed-effects linear regression:
st = �s + ıt + Astˇ1 + Mstˇ2 + Cstˇ3 + εst (2)
8 Mopeds and scooters are included in registration data in states that require these ehicles to be registered. 9 Another option we considered for the exposure variable was the number of new otorcycle units sold each year in each state. Since current sales represent only a
mall portion of the total motorcycles in use in a particular year, we decided to use otorcycle registrations instead.
M.T. French et al. / Journal of Health Economics 28 (2009) 831–838 835
Table 2 Estimation results for non-fatal and fatal motorcycle injuries.
Non-fatal injuries per 100,000 people (fixed-effects OLS)
Fatal injuries per 100,000 people (fixed-effects OLS)
Fatal injury count (conditional fixed-effects negative binomial)
(1) (2) (3) (4) (5) (6)
Universal helmet law −6.605* −7.386* −0.444* −0.415* −0.254* −0.240* (1.452) (1.472) (0.076) (0.077) (0.028) (0.031)
[0.776] [0.786] Mandatory rider education −3.333** −3.806** −0.059 −0.105 0.030 0.010
(1.471) (1.488) (0.087) (0.089) (0.044) (0.046) [1.030] [1.010]
Speed limit on rural interstates/10 −5.034* −4.137* 0.015 −0.068 0.102*** 0.034 (1.142) (1.186) (0.056) (0.063) (0.029) (0.033)
[1.108] [1.035] BAC limit ≤ 0.08 −1.031 0.117 0.049 0.010 0.007 0.005
(0.811) (0.817) (0.046) (0.048) (0.021) (0.022) [1.007] [1.005]
Zero tolerance law −0.159 0.672 0.052 0.006 0.032 0.011 (1.120) (1.119) (0.062) (0.063) (0.028) (0.028)
[1.032] [1.011] Administrative license revocation 3.427** 3.689*** 0.062 0.070 0.007 0.009
(1.407) (1.388) (0.067) (0.067) (0.033) (0.032) [1.007] [1.009]
State and year fixed-effects Yes Yes Yes Yes Yes Yes State-specific controls No Yes No Yes No Yes
Number of observations 574 574 768 768 768 768 Log-likelihood −1744.52 −1713.19 −261.52 −240.56 −2385.47 −2355.07 Notes: state-specific controls include the unemployment rate, income per capita, average annual temperature, average annual precipitation, gasoline prices, annual urban and rural millions of vehicle miles traveled per 1000 residents, lane miles per mile of total public roads, highway maintenance funds per mile of total public roads, motor vehicle fatalities per 10,000 registered vehicles, percentage of young drivers, percentage of white residents, residents with bachelor’s or higher degree, and the average household size. Each specification also includes motorcycle registrations as described in the text. For each explanatory variable in columns 1–4, we report the estimated coefficient and t e incid t
w a u c a
i a 2 u u b f d i P f fi
v b r b s b t l i
i m h m
t a t m v t v i
v p 7 t of 0.00 make up slightly less than 18 percent of all observations. Approximately 74 percent of state/year observations had an ALR policy in place while approximately a third of all observations had a BAC limit of 0.08 or less during the period of analysis. During the late
he estimated standard errors in parentheses. In columns 5 and 6, we also report th he null hypothesis that IRR = 1. ** Significance at the 5% level.
*** Significance at the 1% level.
here the fatal or non-fatal injury rates are regressed directly on lcohol and traffic policies and a set of controls, �s and ıt denote the nobserved state-specific and year-specific determinants of motor- ycle injuries, and εst is the error term, which is assumed to follow normal distribution.
Since our data reveal a small number of motorcycle fatalities n many states and years, employing count models may be more ppropriate than using fatality rates in this case (Grant and Rutner, 004; Morrisey and Grabowski, 2005). Given the nature of the nderlying data, we also estimate a model for fatal injury counts sing a conditional fixed-effects count data technique proposed y Hausman et al. (1984). In a conditional fixed-effects Poisson ramework, the count of fatalities (yst) is assumed to have a Poisson istribution with parameter �st, and the unobserved heterogene-
ty is modeled as state-specific fixed-effects denoted by �s. The oisson parameter � is a deterministic function of the observed actors listed above as well as the state-specific and year-specific xed-effects according to the following expression:
st = exp(�s + ıt + Astˇ1 + Mstˇ2 + Cstˇ3) (3)
Because the fatality counts across states exhibit considerable ariation leading to a high degree of overdispersion, the negative inomial technique was chosen for the core analysis, but Poisson egressions were included in the sensitivity analyses. The negative inomial technique is a more flexible alternative to Poisson regres-
ion in the presence of overdispersion. Both models are estimated y maximum likelihood and the estimation is conditional on the otal count of fatalities in each state. In our count data models, the ogarithm of motorcycle registrations was used as a proxy for the ntensity of motorcycle use in each state and year (i.e., exposure).
ence rate ratios [IRR] in brackets and statistical significance is based on the test of
In both linear regression and count models, the coefficients of nterest are contained in the vectors ˇ1 and ˇ2.10 The direction,
agnitude, and significance of the coefficients attached to the alco- ol and traffic policies indicate whether these policy tools have a eaningful effect on motorcycle safety.
Table 1 contains descriptive statistics (mean, standard devia- ion, and range) for the outcome and policy variables used in the nalysis. The average count of non-fatal injuries is 1472 whereas he average count of fatalities across all states and years is approxi-
ately 57. As indicated by the range and standard deviations, wide ariation exists in both of these measures, even when adjusted for he size of the population. Non-fatal injuries per 100,000 people aries between 8.3 and 138.2 across all years and states, and fatal njuries per 100,000 people ranges between 0.2 and 4.8.
In terms of traffic policies, less than half of the state/year obser- ations had a universal helmet law or mandatory rider education rogram. The speed limit on rural interstates ranged from 55 mph to 5 mph during the analysis period. Although many states had zero olerance laws during this period, those with a strict youth BAC limit
10 To conserve space, we do not report the estimated coefficients for the control ariables nor the state and year fixed effects. These results can be obtained upon equest and can be found in French et al. (2008).
8 ealth E
1 b B h
W e a r T c v a s t a e p e 2 l ( i l c a l f o
r v t a c b l m p b n i a a c t I i a
m c e d S t
o u ( u t t
e m s i s
r r s n c h c u p e l t a
l a e l e ( w a m y o o
s r c m w t a o p w c t i s s c l t not reveal any evidence of differential policy effects.
36 M.T. French et al. / Journal of H
990s, several states adopted more stringent traffic safety policies y implementing mandatory rider education programs and stricter AC limits while other states repealed their mandatory universal elmet policy and raised their maximum speed limit.
The estimation results for core models are presented in Table 2. e report the estimated coefficients together with their standard
rrors in parentheses. All specifications in this table include state nd year fixed-effects. In the first two columns, we present the esults of the fixed-effects linear model for non-fatal injury rates. he first column includes alcohol and traffic policies and motor- ycle registrations per 100,000 people, without any other control ariables. The second column presents the estimation results when rich set of state-specific control variables is added to the analy-
is. Note that the inclusion of state-specific controls does not alter he main results in this analysis. All three traffic safety policies re significantly related to non-fatal injury rates. Mandatory rider ducation programs reduce non-fatal injuries by approximately 10 ercent (p < 0.01). The estimated effect of universal helmet laws is ven larger, decreasing the non-fatal injury rate by approximately 0 percent (p < 0.01). Paradoxically, a 10 mph reduction in the speed
imit would increase the non-fatal injury rate by about 11 percent p < 0.01). It is possible that traveling at higher speeds makes avoid- ng a motorcycle crash more difficult and, if a crash occurs, may ead to a fatal rather than a non-fatal injury. It could also be the ase that more rural states, with less vehicular traffic and associ- ted hazards, are more likely to raise speed limits. Zero tolerance aws and a .08 BAC limit are not significantly associated with non- atal injuries whereas ALR laws work in the opposite direction from ur hypothesis.
Columns 3 and 4 in Table 2 present the linear fixed-effects esults with the fatality rate per 100,000 people as the dependent ariable. As discussed above, modeling fatalities as a count rather han a rate may be more appropriate, so we refrain from drawing ny conclusions in terms of the quantitative results. This specifi- ation, however, allows us to make direct qualitative comparisons etween fatal and non-fatal injury estimates. A universal helmet
aw is the only public policy that significantly influences the rate of otorcycle fatalities. The estimated coefficient on the ALR policy is
ositive (and statistically significant) in the non-fatal injury models, ut essentially zero in the fatal injury models. One possible expla- ation for the differential effect of ALR in the non-fatal and fatal
njury models could be that the severity, reporting, and other char- cteristics of non-fatal crashes are important omitted variables. If dopting policies such as the ALR reduces the overall severity of rashes, but not the frequency, then it could be that relatively more raffic crashes will lead to non-fatal injuries rather than fatal ones. n fact, both passenger car fatalities and overall motor vehicle fatal- ties are on average higher for state and year observations without n ALR policy in place.
The results of the conditional fixed-effects negative binomial odels for the count of fatal motorcycle injuries are presented in
olumns 5 and 6 of Table 2. We report the estimated coefficient,
stimated standard error (in parentheses), and the associated inci- ence rate ratios (IRR [in brackets]) for each explanatory variable.11
tatistical significance is based on a test of the null hypothesis hat there is no relationship between motorcycle fatalities and the
11 IRRs are the exponentiated coefficients and represent the difference in the rate f fatalities predicted by the model when the variable of interest is increased by one nit above its mean value while all other variables are kept constant at their means see Table 1 of French et al. (2008) for the means and units of measure for all variables sed in the analysis). A value greater than 1 indicates a positive relationship between he rate of fatalities and the particular regressor, and a value less than 1 indicates he opposite.
c m i n u r
conomics 28 (2009) 831–838
xplanatory variable (i.e., IRR is equal to 1). The results from this odel are consistent with those in columns 3 and 4, which show a
trong negative effect of universal helmet laws on motorcycle fatal- ties. None of the other alcohol or traffic policies are statistically ignificant in columns 5 and 6.12
As seen in all specifications, a universal helmet law significantly educes both fatal and non-fatal injuries (p < 0.01). Although our esults for non-fatal injuries are unique, these estimates are con- istent with the findings of previous studies, showing a significant egative relationship between universal helmet laws and motorcy- le fatalities. Sass and Zimmerman (2000) estimated that universal elmet laws lower per capita motorcyclist fatalities by about 24 per- ent. Houston and Richardson (2008) concluded that states with niversal helmet laws had rider fatality rates that were about 29 ercent lower than states without universal policies. More recent stimates by Dee (2009) reveal similar effects of universal helmet aws on motorcyclist fatalities (27 percent). Our estimates indicate hat over the period from 1990 to 2005, universal helmet laws led to 24 (20) percent reduction in fatal (non-fatal) motorcycle injuries.
In 2005, 20 of the 48 states in our sample had universal helmet aws. Total rider fatalities were 1894 for universal helmet law states nd 2472 for states without a universal helmet law. Based on the stimates from our models and additional calculations, about 489 ives could have been saved if universal helmet laws had been in ffect in all 48 states. Using $5 million as the value of a statistical life Viscusi and Aldy, 2003), the estimated mortality cost associated ith the absence of universal helmet policies in 2005 alone was
lmost $2.5 billion. It would be interesting to determine whether otorcyclists would be willing to pay an “endorsement fee” each
ear for the right to ride without a helmet, which could offset some f these costs, but such a cost–benefit analysis is beyond the scope f the present paper.
To further examine the sensitivity of the results to model pecification, we conducted several robustness checks.13 First, we e-estimated the specifications in columns 5 and 6 of Table 2 using a onditional fixed-effects Poisson model instead of a negative bino- ial model. In each case, the results were virtually identical. Next, e disaggregated the total fatality counts according to the day or
he time of the crash. One might expect the alcohol policies to have relatively larger effect on nighttime and weekend fatalities than n daytime and weekday fatalities. The rationale here is that such olicies would influence drinking behaviors more at night and on eekends when alcohol consumption is more common. On the
ontrary, the results are similar (both qualitatively and quantita- ively) for all specifications, regardless of the time or day. Finally, n an effort to identify whether the alcohol and traffic policies have imilar effects on drivers other than motorcyclists, we estimated eparate regressions for motorcycle rider fatalities in single-vehicle rashes and rider fatalities in multi-vehicle crashes involving at east one motorcycle and one other type of vehicle. Once again, he stratified results are consistent with our core models and do
Given the limited within-state variation in alcohol taxes over ime for most states, the beer tax was not included in our core spec- fications. As an additional robustness check, we re-estimated the
12 Despite the fact that we control for the number of motorcycle registrations in the onditional fixed-effects negative binomial models, some of the policies we consider ight indirectly reduce motorcycle-related fatalities by discouraging motorcycling
n general. A closer examination of motorcycle registrations per capita indicates a egative relationship with universal helmet laws. This suggests that states that adopt niversal helmet laws might inadvertently reduce motorcycle-related fatalities by educing motorcycle usage. 13 The full results of the sensitivity analyses are available upon request from the orresponding author.
c a m t c o a a f o b f c t t e c t t
a m t i r l s m
o s c m a m r c a n t a b p f c i
l f a b t (
b o E i a t
t t f S p D (
i r m m m p n a u t i b fi m d
I A H s t G M C S a s t o i h
M.T. French et al. / Journal of H
ore models with the beer tax. Coefficient estimates on all other lcohol and traffic policies are virtually unchanged in terms of sign, agnitude, and significance. While the results consistently indicate
hat the beer tax has a negative and significant impact on motorcy- le fatalities, we are not confident in the large estimated magnitude f this effect.14 The beer tax coefficient is not significant in our fully ugmented non-fatal injury specification, and we are not aware of ny studies that have estimated the effect of the beer tax on non- atal automobile injuries. In light of concerns about the magnitude f the beer tax estimates reported in other studies and the possi- ility that beer taxes are correlated with important unobservable
actors, we decided to exclude this measure from our core specifi- ations in Table 2. Finally, we added per capita beer consumption to he models to examine whether controlling for state-specific pat- erns in alcohol consumption might alter the main findings. As xpected, per capita beer consumption is positively and signifi- antly related to both fatal and non-fatal injuries, but inclusion of his variable does not meaningfully change the estimated effects of he alcohol and traffic policies.
To our knowledge, this study is the first rigorous longitudinal nalysis of the effects of public policies on both fatal and non-fatal otorcycle crashes in the U.S. Using state-specific data from 1990
o 2005, our findings suggest that several public policies can signif- cantly reduce non-fatal motorcycle injuries, including mandatory ider education programs, universal helmet laws, and lower speed imits on rural interstates. On the other hand, universal helmet laws eem to be the most reliable and effective policy tool to reduce fatal otorcycle injuries.
The primary objective of this study was to determine the effects f alcohol and traffic policies on motorcycle safety, but we also con- idered a large set of demographic, economic, and environmental ontrols, as these important state characteristics could influence otorcycle crashes as well. An extensive data collection effort from variety of sources was required to compile state-specific infor- ation on non-fatal injuries, alcohol and traffic policies, and the
ich set of controls. Data on non-fatal injuries among motorcy- lists were collected from unpublished state-specific documents, rchived data files, and personal correspondence. Information on on-fatal injuries was not available for all years and states. In addi- ion, data collection resources and procedures might differ slightly cross states. Any potential measurement error, if present, would ias the results to the extent it is systematically correlated with the olicy changes over time. A standardized source of data on non-
atal injuries for all states and years (similar to FARS) would have onsiderably reduced data collection costs and research time and mproved overall reliability of the estimates.
As in most studies of motor vehicle safety, there are additional imitations to our empirical analysis. First, data were unavailable or some potentially important predictors in our models, such as
nnual motorcycle miles traveled. Furthermore, the estimates could e biased due to endogenous policy adoption. We believe, however, hat our estimates for policies targeting all motor vehicle drivers e.g., ALR) are less likely to be endogenous than those specifically
14 A few studies have offered explanations for why an increase in the beer tax can e quite effective in reducing automobile fatalities even though these taxes display nly small within-state variations over time (Dee, 1999; Mast et al., 1999; Dee and vans, 2001; Young and Bielinska-Kwapisz, 2006). The most plausible explanation
s that beer taxes are correlated with other important and omitted state-level char- cteristics such as law enforcement, health policies, or social and political attitudes owards alcohol.
conomics 28 (2009) 831–838 837
argeting motorcycle riders (e.g., universal helmet laws). Finally, he inclusion of state and time fixed-effects cannot compensate or important omitted variables that vary within states over time. ome potentially important time-varying omitted variables include olicy enforcement and grass-roots activities by Mothers Against runk Driving (MADD), American Bikers Aiming Toward Education
ABATE), or other advocacy groups (Eisenberg, 2003). Despite these limitations, this study is original, timely, and pol-
cy relevant given the dramatic rise in the popularity of motorcycle iding and the recent volatility of gasoline prices that is encouraging
otorists to switch to fuel-efficient vehicles. Studies investigating otor vehicle safety and public policy have largely focused on auto- obiles and trucks and almost exclusively on fatal injuries. Public
olicy in this area should also be evaluated in terms of its effective- ess in reducing non-fatal injuries, which occur far more frequently nd generate high social costs. Given that many motorcyclists mis- nderstand or simply disregard the increased safety risks relative o operating an automobile (Bellaby and Lawrenson, 2001), these ndividuals may be reluctant to abandon their dangerous riding ehaviors and may underestimate the value of safety programs. Our ndings suggest that certain public policies can significantly impact otorcycle safety, and, with the exception universal helmet laws,
ifferential effects are present for fatal and non-fatal injuries.
Financial assistance for this study was provided by the National nstitute on Alcohol Abuse and Alcoholism (grant numbers R01 A13167 and R01 AA015695). We thank Ana Balsa, David Bradford, ai Fang, Alan Mathios, Oscar Mitnik, Michael Morrisey, Todd Olm-
tead, Bisakha Sen, Jody Sindelar, and two anonymous referees for heir comments and suggestions. We gratefully acknowledge Ana uzman, Max Johansen, Rosemary Kenney, Shay Klevay, Adrienne ilner, Robin Prize, Kristen Smith, Alex Strassman, Lauren Tapsell,
olleen Trifilo, Pamela Valbuena, Jamila Wade, Venessa Wilson, pencer Winkle, and state rider education coordinators for data nd research assistance; and Carmen Martinez and William Rus- ell for editorial assistance. The authors are entirely responsible for he research and results reported in this paper, and their position r opinions do not necessarily represent those of their respective
nstitutions or the National Institute on Alcohol Abuse and Alco- olism.
ppendix A. Supplementary data
Supplementary data associated with this article can be found, n the online version, at doi:10.1016/j.jhealeco.2009.05.002.
ellaby, P., Lawrenson, D., 2001. Approaches to the risk of riding motorcycles: reflec- tions on the problem of reconciling statistical risk assessment and motorcyclists’ own reasons for riding. The Sociological Review 49 (3), 368–388.
illheimer, J.W., 1998. Evaluation of the California motorcyclist safety program. Transportation Research Record No. 1640, 100–109.
ledsoe, G.H., Li, G., 2005. Trends in Arkansas motorcycle trauma after helmet law repeal. The Southern Medical Journal 98 (4), 436–440.
ledsoe, G.H., Schexnayder, S.M., Carey, M.J., Dobbins, W.N., Gibson, W.D., Hindman, J.W., 2002. The negative impact of the repeal of the Arkansas motorcy- cle helmet law. Journal of Trauma-Injury Infection & Critical Care 53 (6), 1078–1087.
arpenter, C.S., 2004. How do zero tolerance drunk driving laws work? Journal of
Health Economics 23 (1), 61–83.
arpenter, C.S., Stehr, M., 2008. The effects of mandatory seatbelt laws on seatbelt use, motor vehicle fatalities, and crash-related injuries among youths. Journal of Health Economics 27 (3), 642–662.
haloupka, F.J., Saffer, H., Grossman, M., 1993. Alcohol-control policies and motor- vehicle fatalities. Journal of Legal Studies 22 (1), 161–186.
8 ealth E
38 M.T. French et al. / Journal of H
oben, J.H., Steiner, C.A., Miller, T.R., 2007. Characteristics of motorcycle-related hos- pitalizations: comparing states with different helmet laws. Accident Analysis & Prevention 39 (1), 190–196.
hristoffel, T., Gallagher, S.S., 2006. Injury Prevention and Public Health. Jones & Bartlett Publishers, Inc., Sudbury, MA.
olburn, N., Meyer, R.D., Wrigley, M., Bradley, E.L., 1993. Should motorcycles be oper- ated within the legal alcohol limits for automobiles? The Journal of Trauma 35 (2), 183–186.
ee, T.S., 1999. State alcohol policies, teen drinking and traffic fatalities. Journal of Public Economics 72 (2), 289–315.
ee, T.S., 2001. Does setting limits save lives? The case of 0.08 BAC laws. Journal of Policy Analysis and Management 20 (1), 111–128.
ee, T.S., 2009. Motorcycle helmets and traffic safety. Journal of Health Economics 28 (2), 398–412.
ee, T., Evans, W.N., 2001. Teens and traffic safety. In: Gruber, J. (Ed.), Risky Behavior among Youths: An Economic Analysis. University of Chicago Press, Chicago.
isenberg, D., 2003. Evaluating the effectiveness of policies related to drunk driving. Journal of Policy Analysis and Management 22 (2), 249–274.
reeman, D.G., 2007. Drunk driving legislation and traffic fatalities: new evidence on BAC 08 laws. Contemporary Economic Policy 25 (3), 293–308.
rench, M.T., Gumus, G., Homer, J., 2008. Fatal and non-fatal motorcycle injuries: can public policies influence rider safety and reduce crashes? Working paper.
rabowski, D.C., Morrisey, M.A., 2001. The effect of state regulations on motor vehi- cle fatalities for younger and older drivers: a review and analysis. The Milbank Quarterly 79 (4), 517–545.
rabowski, D.C., Morrisey, M.A., 2007. System-wide implications of the repeal of the National Maximum Speed Limit. Accident Analysis and Prevention 39 (1), 180–189.
rant, D., Rutner, S.M., 2004. The effect of bicycle helmet legislation on bicycling fatalities. Journal of Policy Analysis and Management 23 (3), 595–611.
ausman, J., Hall, B., Griliches, Z., 1984. Econometric models for count data with an application to the patents-R&D relationship. Econometrica 52 (4), 909–938.
ouston, D.J., Richardson, L.E., 2008. Motorcyclist fatality rates and mandatory helmet-use laws. Accident Analysis and Prevention 40 (1), 200–208.
urt, H.H., Ouellet, J.V., Thom, D.R., 1981. Motorcycle Accident Cause Factors and Identification of Countermeasures, vol. 1: Technical Report No. DOT HS 805 862. National Highway Traffic Safety Administration, Washington, DC.
ast, B.D., Benson, B.L., Rasmussen, D.W., 1999. Beer taxation and alcohol-related
fatalities. Southern Economic Journal 66 (2), 214–249.
ax, W., Stark, B., Root, S., 1998. Putting a lid on injury costs: the economic impact of the California motorcycle helmet law. Journal of Trauma 45 (3), 550–556.
cGwin, G., Whatley, J., Metzger, J., Valent, F., Barbone, F., Rue, L.W., 2004. The effect of state motorcycle licensing laws on motorcycle driver mortality rates. Journal of Trauma 56 (2), 415–419.
conomics 28 (2009) 831–838
orrisey, M.A., Grabowski, D.C., 2005. State motor vehicle laws and older drivers. Health Economics 14 (4), 407–419.
otorcycle Industry Council, 2006. Motorcycle and scooter sales get pumped up in first half of 2006. Press release from July 20, 2006. Available online at http://www.mic.org/news072006.cfm.
ational Highway Traffic Safety Administration (NHTSA), 2007. Traffic Safety Facts—2006. A Compilation of Motor Vehicle Crash Data from the Fatality Anal- ysis Reporting System and the General Estimates System. DOT HS 810 818. U.S. Department of Transportation, Washington, DC.
ational Highway Traffic Safety Administration (NHTSA), 2008. Traffic Safety Facts—Motorcycles 2006 Data. DOT HS 810 806. U.S. Department of Transporta- tion, Washington, DC.
eek-Asa, C., Kraus, J.F., 1996. Alcohol use, driver, and crash characteristics among injured motorcycle drivers. The Journal of Trauma Injury, Infection, and Critical Care 41 (6), 989–993.
owland, J., Rivara, F., Salzberg, P., Soderberg, R., Maier, R., Koepsell, T., 1996. Motor- cycle helmet use and injury outcome and hospitalization costs from crashes in Washington state. American Journal of Public Health 86 (1), 41–45.
uhm, C.J., 1996. Alcohol policies and highway vehicle fatalities. Journal of Health Economics 15 (4), 435–454.
ass, T.R., Zimmerman, P.R., 2000. Motorcycle helmet laws and motorcyclist fatalities. Journal of Regulatory Economics 18 (3), 195–215.
hankar, U.G., 2003. Alcohol Involvement in Fatal Motorcycle Crashes. DOT HS 809 576. NHTSA Research Note. U.S. Department of Transportation, Washington, DC.
hults, R.A., Elder, R.W., Sleet, D.A., Nichols, J.L., Alao, M.O., Carande-Kulis, V.G., Zaza, S., Sosin, D.M., Thompson, R.S., Task Force on Community Preven- tive Services, 2001. Reviews of evidence regarding interventions to reduce alcohol-impaired driving. American Journal of Preventive Medicine 21 (4S), 66–88.
un, S.W., Kahn, D.M., Swan, S.K., 1998. Lowering the legal blood alcohol level for motorcyclists. Accident Analysis and Prevention 30 (1), 133–136.
.S. Department of Transportation, 2007. U.S. Department of Transportation Action Plan to Reduce Motorcycle Fatalities. DOT HS 810 855. Washington, DC.
illaveces, A., Cummings, P., Koepsell, T.D., Rivara, F.P., Lumley, T., Moffat, J., 2003. Association of alcohol-related laws with deaths due to motor vehicle and motor- cycle crashes in the United States, 1980–1997. American Journal of Epidemiology 157 (2), 131–140.
iscusi, W.K., Aldy, J.E., 2003. The value of a statistical life: a critical review of market estimates throughout the world. Journal of Risk and Uncertainty 27 (1), 5–76.
elsh, J., 2008. Posh Bikes Rev up Amid Slowdown, The Wall Street Journal, New York, NY, Mar 26. Available online at http://online.wsj.com/public/ article/SB120648834678863907-NgfXiIAjdt9k68gbHnq6v7MedbM 20080424. html?mod=tff main tff top.
oung, D.J., Bielinska-Kwapisz, A., 2006. Alcohol prices, consumption, and traffic fatalities. Southern Economic Journal 72 (3), 690–703.
Public policies and motorcycle safety
The post Journal of Health Economics appeared first on superioressaypapers.]]>