This article was downloaded by: [FU Berlin] On: 19 November 2014, At: 06:35 Publisher: Taylor & Francis Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Traffic Injury Prevention Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/gcpi20
Hospitalizations and Fatalities in Crashes With Light Trucks a
Pinar Karaca-Mandic & Jinhyung Lee
b
a
University of Minnesota, School of Public Health, Division of Health Policy and Management , Minneapolis , Minnesota b
Jiann Ping-Hsu College of Public Health , Georgia Southern University , Statesboro , Georgia Accepted author version posted online: 14 Jun 2013.Published online: 17 Dec 2013.
Click for updates To cite this article: Pinar Karaca-Mandic & Jinhyung Lee (2014) Hospitalizations and Fatalities in Crashes With Light Trucks, Traffic Injury Prevention, 15:2, 165-171, DOI: 10.1080/15389588.2013.803279 To link to this article: http://dx.doi.org/10.1080/15389588.2013.803279
PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions
Traffic Injury Prevention (2014) 15, 165–171 C Taylor & Francis Group, LLC Copyright ISSN: 1538-9588 print / 1538-957X online DOI: 10.1080/15389588.2013.803279
Hospitalizations and Fatalities in Crashes With Light Trucks PINAR KARACA-MANDIC1 and JINHYUNG LEE2 1 2
University of Minnesota, School of Public Health, Division of Health Policy and Management, Minneapolis, Minnesota Jiann Ping-Hsu College of Public Health, Georgia Southern University, Statesboro, Georgia
Downloaded by [FU Berlin] at 06:35 19 November 2014
Received 21 February 2013, Accepted 5 May 2013
Objective: This study examined 2-car crashes including one passenger car and one light truck (van, minivan, pickup truck, or sport utility vehicle) and investigated the likelihood of hospitalization, hospitalization charges, and the likelihood of fatality of an occupant by vehicle type differentiating between passengers and drivers. Methods: We used unique data from Minnesota’s Crash Outcome Data Evaluation System (CODES) linked with hospital discharge data from 2004 to 2005. We focused on police-reported crashes that involved 2 vehicles, one car and one light truck. First, we estimated models to predict the likelihood of hospitalization. Next, we estimated models to predict hospitalization charges among the hospitalized. Finally, we modeled the likelihood of fatality. In all models, we distinguished between the 2 vehicle types and controlled for a broad range of occupant, crash, and vehicle characteristics. We estimated separate models for passengers and drivers. Results: We found that in a crash between a car and a light truck, drivers of light trucks were less likely to be hospitalized (odds ratio [OR] = 64%; 95% confidence interval [CI], 59–70%) and killed (OR = 35%; 95% CI, 18–68%) relative to the drivers of cars. Similarly, passengers of light trucks had a lower likelihood of hospitalization (OR = 66%; 95% CI, 57–77%) and fatality (OR = 14%; 95% CI, 3–54%) relative to the passengers of cars. Among hospitalized occupants, we did not find statistically significant differences in hospital charges between light truck drivers and car drivers, but hospital charges for hospitalized light truck passengers were 59% (95% CI, 40–87%) of the hospital charges of hospitalized car passengers. Conclusions: Though previous studies have shown high fatality costs associated with light trucks, this study is the first to explore the hospitalization costs associated with these vehicles. The existing traffic liability systems (tort or no-fault systems) likely fail to fully make light trucks accountable for costs they impose on other cars, pedestrians, and other road occupants. Our findings suggest the importance of a close examination of a broad range of cost implications even beyond hospitalization and fatality costs to evaluate the optimal amount of corrective taxes or other corrective policies in future research. Supplemental materials are available for this article. Go to the publisher’s online edition of Traffic Injury Prevention to view the supplemental file. Keywords: light trucks, hospitalization, fatality, traffic safety
Introduction Over the last 2 decades, there has been a significant increase in the number of light trucks, including vans, minivans, pickup trucks, and sport utility vehicles (SUVs), on the roads. For example, based on the Census Bureau’s Vehicle Inventory and Use Surveys (U.S. Census Bureau 1999a; U.S. Census Bureau 1999b; U.S. Census Bureau 2002), there was an increase of 60.5% in the number of registered minivans from 1992 to 1997 and an increase of 24.1% from 1997 to 2002. The percentage increases in the number of registered SUVs were even larger: 92.7% from 1992 to 1997 and 55.8% from 1997 to 2002. An important contributor to the increased popularity of light trucks is the fact that they provide improved protection
Address correspondence to Pinar Karaca-Mandic, PhD, University of Minnesota School of Public Health, Division of Health Policy and Management, 420 Delaware St. SE, MMC 729, Minneapolis, MN 55455. E-mail: [email protected]
to their own passengers (Coate and VanderHoff 2001). However, these vehicles likely impose significant damage on the passengers of the other cars they crash into, as well as to pedestrians, bicyclists, and motorcyclists. For example, Meyer and Gomez-Ibanez (1981) found that conditional on a crash between a passenger car and a light truck, occupants of large vehicles were 29 percent less likely to be seriously injured. Similarly, Wenzel and Ross (2005) showed that SUVs and pickup trucks imposed greater driver fatalities on passenger cars and that the fatality risks increase with the size of the larger vehicle. A large body of literature on crashworthiness shows that in a crash between a car and a light truck, the taller body, higher mass, and stiffer front structure of the light trucks relative to the cars result in vehicle incompatibility, especially in frontto-front and front-to-side crashes, disadvantaging the cars. Though vehicle incompatibility is associated with a higher probability of fatality in occupants of the cars relative to light trucks, advances in vehicle designs over time have resulted in the improved safety of cars involved in crashes with light trucks (Braitman et al. 2007; Teoh and Nolan 2012).
Downloaded by [FU Berlin] at 06:35 19 November 2014
166 Several studies have quantified the fatality costs associated with light trucks. White (2004) estimated that for every additional one million drivers of SUVs there would be between 34 and 93 additional fatalities involving passenger car occupants, pedestrians, bicyclists, or motorcyclists per year. The value of these lost lives would range from $242 to $652 million per year. Similarly, Anderson (2008) found that a one percentage point increase in the share of light trucks (which he defined as SUVs and pickup trucks) on the roads was associated with a 0.34% increase in annual traffic fatalities, or 143 additional fatalities per year. Further, he found that whereas 20% of the increase in traffic fatalities accrued to the occupants of light trucks, 80% accrued to the occupants of other vehicles and pedestrians (external costs). Recently, Anderson and Auffhammer (2011) found that the baseline fatality probability increased 47% by being hit by a 1000-pound heavier vehicle and estimated the total external costs from fatalities to be $93 billion per year. Similarly, Gayer (2004), Martinelli and DiasdadoDe-La-Pena (2008), and Li (2012) confirmed the general finding that light trucks cause more damage to the occupants of other vehicles but less damage to their own occupants. Li (2012) estimated that the external costs associated with a light truck were around $2444 during the vehicle’s lifetime. In this article, we estimate the magnitude of hospitalization costs as well as the fatality costs that light trucks cause to occupants of cars and to their own occupants. From a policy perspective, to design market-based incentives or regulations to make light trucks accountable for the additional cost they impose on other cars, it is important to understand the broad range of costs associated with these vehicles. Though previous studies have quantified the costs these larger vehicles impose on other occupants and their own occupants in terms of fatalities, the cost implications in terms of hospitalizations are unknown. Naturally, the cost of a fatality is considerably higher than the cost of a hospitalization, given that the value of a statistical life, as estimated by the U.S. Department of Transportation, is estimated to be around $5.8 million (Anderson and Auffhammer 2013). However, the rate of injuries and hospitalizations after a crash is substantially higher than the fatality rate. In 2007, of the 6 million police-reported crashes in the United States, 30% involved an injury, and less than 1% involved a fatality (NHTSA 2007).
Material and Methods Data We used a unique database, the Crash Outcome Data Evaluation System (CODES), 2004 and 2005, which contains linked data from the Minnesota Hospital Association, the Emergency Medical Services (EMS) Regulatory Board, and the Minnesota departments of Health, Public Safety, and Transportation. The crash data within CODES come from the Department of Public Safety and provide information on all police-reported crashes and on all persons involved in the crash (age, gender, drinking status, driver/pedestrian status, seat belt use, etc.) as well as detailed information on the crash
Karaca-Mandic and Lee (location, time, road type, road conditions, number of vehicles involved, etc.). These police-reported crash data were linked with hospital data, which included information on hospital charges for the hospitalized crash victims. Ideally, each crash associated with a hospitalization would have a relevant medical record that can be linked, and the absence of a linkage should indicate the absence of a hospitalization event. However, in reality, the crash records may not be linked with a medical record despite a hospital event, or a matched hospital record may not indicate a crash. As a result, the CODES project uses a probabilistic record linkage method to account for missing unique identifiers for injured persons and medical events. The detailed algorithm for the linkage is described in the U.S. Department of Transportation, NHTSA (1996) technical report (Johnson and Walker 1996). Instead of a single set of linked records, Minnesota CODES data are provided as multiple (5) sets of imputed, linked records using a full Bayesian model as described in McGlincy (2004). We focused on police-reported crashes that involved 2 vehicles, one car and one light truck. Car was defined as a passenger car, and the definition of light trucks in the data included pickup trucks, vans, minivans, and SUVs. Outcome Variables We considered 3 outcome variables: hospitalization of vehicle occupants, hospital charges for hospitalized individuals, and fatality outcome of vehicle occupants. Hospitalization was characterized into a binary indicator based on whether or not the individual was taken to the hospital. Hospital charges represented an aggregate of charges for all hospital treatments, from acute care to discharge. An acute care period lasts until the patient is discharged to home, transitional care, or rehabilitation. If the patient was discharged but readmitted within 2 days, we considered charges associated with the readmission as well. Finally, the fatality outcome was a binary indicator extracted from a variable that characterized injury severity (no apparent injury, possible injury, moderate injury, severe injury, and killed) for each vehicle occupant in the crash. Explanatory Variables We identified each type of vehicle (car or a light truck) involved in the crash. We also extracted information on the characteristics of the vehicle occupants involved in the crash (age and gender, seat belt use, physical condition such as normal, driving under the influence of alcohol or drugs, fatigued, other physical disabilities, etc.), as well as information on the characteristics of the vehicle (make/model, model year, number of passengers in the vehicle, and auto insurance status of the vehicle) and characteristics of the crash (weather conditions such as clear, snow, rain, etc.; road conditions such as dry, wet, ice, etc.; the time of crash such as daylight, sunrise, sunset, etc.). All of our analyses controlled for the weight of the occupant’s own vehicle and the weight of the other vehicle involved in the crash. Data on vehicle weight were available for 62% of the vehicles. In 34% of the crashes, both the weight of the occupant’s own vehicle and the weight of the other vehicle
Crashes With Light Trucks
167
Table 1. Models of hospitalization, hospital charges, and driver fatalitiesa
Downloaded by [FU Berlin] at 06:35 19 November 2014
Explanatory Variables Light truck Gender of occupant (male = 1, female = 0) Age 0–18 19–45 46–64 65 and over Urban Time of accident Light Sunrise Sunset Dark Road conditions during the accident Dry Wet Snow Ice Physical condition of occupant Normal Alcohol or drug Occupant belt use No belt Proper seat belt use Other/improper seat belt use Number of occupants in the vehicle Weight of own vehicle (in 1000 lb) Weight of other vehicle (in 1000 lb) Own vehicle model fixed effects Own vehicle year of make fixed effects
Specification 1
Specification 2
Specification 3
Hospitalization (1/0)
Total hospital charges conditional on hospital admission ($)
Fatality (1/0)
Logistic
Generalized linear model (log link, gamma family)
Logistic
95% Confidence interval
Adjusted odds ratio
95% Confidence interval
−0.14 0.00
(−0.37, 0.09) (−0.18, 0.18)
0.35∗∗∗ 1.00
(0.18, 0.68) (0.64, 1.57)
(1.02,1.24) (1.23, 1.54) (1.46, 1.91) (0.56, 0.64)
Ref 0.15 0.67∗∗∗ 0.90∗∗∗ −0.04
(−0.19, 0.48) (0.27, 1.07) (0.36, 1.44) (−0.22, 0.14)
Ref 0.91 1.55 3.62∗∗∗ 0.16∗∗∗
(0.47, 1.77) (0.75, 3.21) (1.70, 7.72) (0.09, 0.31)
Ref 1.26∗ 1.24∗∗ 1.23∗∗∗
(0.97, 1.63) (1.02, 151) (1.14, 1.35)
Ref 0.72∗∗ −0.08 −0.08
(0.00, 1.43) (−0.52, 0.37) (−0.30, 0.14)
Ref 2.01 1.22 1.94∗∗
(0.45, 9.05) (0.42, 3.53) (1.13, 3.33)
Ref 0.90∗ 0.60∗∗∗ 0.64∗∗∗
(0.80,1.01) (0.50, 0.72) (0.55, 0.74)
Ref 0.20 0.27 0.22
(−0.12, 0.52) (−0.16, 0.69) (−0.19, 0.63)
Ref 0.53 0.59 0.70
(0.19, 1.45) (0.19, 1.82) (0.31, 1.57)
Ref 2.75∗∗∗
(2.32, 3.25)
Ref 0.95∗∗
(0.12, 1.76)
Ref 1.82
(0.59, 5.62)
Ref 0.20∗∗∗ 0.28∗∗∗ 1.12∗∗∗
(0.18, 0.23) (0.12, 0.62) (1.08, 1.16)
Ref −0.61∗∗∗ −1.26∗∗∗ 0.01
(−0.86, −0.36) (−1.61, −0.91) (−0.09, 0.12)
Adjusted odds ratio
95% Confidence interval
0.64∗∗∗ 0.59∗∗∗
(0.59, 0.70) (0.56, 0.63)
Ref 1.13∗∗ 1.38∗∗∗ 1.67∗∗∗ 0.60∗∗∗
0.97 1.01 Included Included
(0.92, 1.01) (0.96, 1.05)
Coefficient estimate
−0.08∗∗ 0.13∗∗ Included Included
(−0.13, −0.02) (0.03, 0.23)
Ref 0.05∗∗∗ Omitted 0.95 1.22∗ 1.02 Included Included
(0.03, 0.08) (0.72, 1.26) (0.99, 1.50) (0.75, 1.38)
Standard errors are clustered at the crash level. All other explanatory variables listed in the text and in Table A1 are included in the models but not reported in this table. They are available from the authors upon request. ∗ P < .10. ∗∗ P < .05. ∗∗∗ P < .01.
were available. We used “unknown” indicators for when we did not know the weight of the occupant’s own vehicle or the weight of the other vehicle. Statistical Methods We estimated a model of hospitalization at the occupant-level separately for drivers and passengers. We used logistic regression to relate the binary variable characterizing hospitalization and whether the vehicle was a car or a light truck, occupant characteristics, vehicle characteristics, and crash characteristics listed above. For unknown values of any given characteristic, we included an unknown category and included the unknown indicators in the statistical models. Next, we estimated the hospital charges at the occupant level for occupants who were hospitalized. Mod-
els were estimated separately for drivers and passengers. We used a generalized linear model with log link and gamma family because hospital charges are typically not normally distributed (Duan et al. 1983; Manning and Mullahy 2001). Because the hospital discharge data are linked to crash data via probabilistic linkage, which provided 5 possible matches for each hospitalization, we used appropriate multiple imputation regression models (“mim”) in Stata software, version 12 (StataCorp 2009). The “mim” command allows for essentially repeating the hospitalization analyses 5 times and combining the parameter estimates across all analyses (Carlin et al. 2008). We used the same explanatory variables to characterize occupant, vehicle, and crash characteristics. An additional characteristic we considered was the payer source for the hospitalization (private insurance; government insurance, including
168
Karaca-Mandic and Lee
Table 2. Models of hospitalization, hospital charges, and passenger fatalitiesa
Downloaded by [FU Berlin] at 06:35 19 November 2014
Explanatory Variables Light truck Gender of occupant (male = 1, female = 0) Age 0–18 19–45 46–64 65 and over Urban Time of accident Light Sunrise Sunset Dark Road conditions during the accident Dry Wet Snow Ice Physical condition of occupant Normal Alcohol or drug Occupant belt use No belt Proper seat belt use Other/improper seat belt use Number of occupants in the vehicle Weight of own vehicle (in 1000 lb) Weight of other vehicle (in 1000 lb) Own vehicle model fixed effects Own vehicle year of make fixed effects
Specification 1
Specification 2
Specification 3
Hospitalization (1/0)
Total hospital charges conditional on hospital admission ($)
Fatality (1/0)
Logistic
Generalized linear model (log link, gamma family)
Logistic
Adjusted odds ratio
95% Confidence interval
Coefficient estimate
95% Confidence interval
Adjusted odds ratio
95% Confidence interval
0.66∗∗∗ 0.64∗∗∗
(0.57, 0.77) (0.58, 0.70)
−0.53∗∗∗ −0.14
(−0.92, −0.14) (−0.44, 0.17)
0.14∗∗∗ 0.37∗∗∗
(0.03, 0.54) (0.17, 0.78)
Ref 1.19∗∗∗ 1.61∗∗∗ 2.14∗∗∗ 0.75∗∗∗
(1.07,1.34) (1.37, 1.88) (1.78, 2.59) (0.67, 0.85)
Ref −0.02 0.38∗ 0.77∗∗ −0.04
(−0.32, 0.27) (−0.04, 0.79) (0.37, 1.17) (−0.28, 0.21)
Ref 0.73 1.22 7.04∗∗∗ 0.34∗∗
(0.28, 1.88) (0.39, 3.83) (2.67, 18.59) (0.13, 0.86)
Ref 1.21 0.98 1.24∗∗∗
(0.68, 2.15) (0.69, 1.37) (1.08, 1.43)
Ref 1.61∗∗∗ −0.02 0.25∗
(0.39, 2.82) (−0.67, 0.63) (−0.03, 0.53)
Ref Omitted 1.86 1.39
(0.30, 11.44) (0.63, 3.02)
Ref 0.88 0.58∗∗∗ 0.64∗∗∗
(0.72, 1.07) (0.42, 0.81) (0.50, 0.82)
Ref −0.10 −0.24 −0.22
(−0.54, 0.35) (−0.92, 0.45) (−0.74, 0.29)
Ref Omitted Ref 0.24∗∗∗ 0.47∗∗∗ 1.03 0.99 1.08∗∗ Included Included
Ref Omitted
(0.20, 0.28) (0.30, 0.76) (0.98, 1.10) (0.92, 1.06) (1.01, 1.17)
Ref −0.83∗∗∗ −1.06∗∗ −0.08 −0.009 0.032 Included Included
Ref 0.88 1.02 0.86
(0.30, 2.55) (0.34, 3.10) (0.24, 3.16)
Ref Omitted
(−1.36, −0.29) (−2.04, −0.08) (−0.27, 0.08) (−0.096, 0.079) (−0.097, 0.160)
Ref 0.06∗∗∗ 0.32 1.26 1.15 0.38∗∗ Included Included
(0.03, 0.12) (0.04, 2.90) (0.89, 1.77) (0.84, 1.58) (0.18, 0.80)
aStandard
errors are clustered at the crash level. All other explanatory variables listed in the text and in Table A1 are included in the models but not reported in this table. They are available from the authors upon request. ∗ P < .10. ∗∗ P < .05. ∗∗∗ P < .01.
Medicare/Medicaid/MN Care; a commercial insurance; and self-payment). Finally, we estimated a model of fatality at the occupant level. Using a logistic regression, we related the binary fatality variable to the same explanatory variables used in the hospitalization model. In all of the models, we clustered standard errors at the crash level to account for factors unobserved by us but that may be correlated for vehicles and occupants in a given crash.
Results There were 184,040 motor vehicle crashes in Minnesota in 2004 and 2005. Of these crashes, 32% involved only one vehicle, 61% involved 2 vehicles, and the rest involved more
than 2 vehicles. Of the one-vehicle crashes, 52% involved a car and 45% involved a light truck. Of the 112,624 two-vehicle crashes, 85,679 involved at least one car. Among those with at least one car, 44,394 also involved a light truck. These crashes involved 126,737 occupants (88,422 drivers and 38,315 passengers). Seven percent of drivers and 16% of passengers were hospitalized. About 0.10% of the drivers and passengers were killed. Average hospital charges for hospitalized occupants were $6045 for drivers and $4818 for passengers. Fifty-seven percent of drivers and 45% of passengers were male. Nearly 11% of drivers and 25% of passengers were under 18 years old; 8% of drivers and 6% passengers were over age 65 (Table A1; see online supplement). Table 1 presents model estimates for drivers corresponding to the 3 outcomes: hospitalization (specification 1), hospital charges for those hospitalized (specification 2), and
Downloaded by [FU Berlin] at 06:35 19 November 2014
Crashes With Light Trucks
169
fatality (specification 3). Based on the first specification, the odds of hospitalization for light truck drivers were 64% (95% confidence interval CI, 59–70%) of the odds of hospitalization for car drivers. In the second specification, the coefficient estimate represents the expected change in log-transformed hospital charges from a unit change in the independent variable. The reported estimate on the light truck indicator was not statistically significant, which suggests that there was no statistically significant difference between hospital charges for hospitalized light truck drivers and car drivers. Based on the third specification, the odds of fatality for light truck drivers were 35% (95% CI, 18–68%) of the odds of fatality for car drivers. Other factors associated with a higher odds of hospitalization included older age, sunset or darkness (relative to daytime), alcohol or drug impairment, and number of occupants in the vehicle. Wet, snowy, or icy road conditions (relative to dry road conditions) and seat belt use were associated with a lower odds of hospitalization. Older age and darkness were associated with higher odds of fatality, and proper seat belt use was associated with lower odds of fatality. Hospital charges among those hospitalized decreased with the use of seat belts and weight of occupant’s own vehicle and increased with the weight of the other vehicle. Table 2 presents similar model estimates for passengers. We found that the odds of hospitalization for light truck passengers were 66% (95% CI, 57–77%) of the odds of hospitalization for car passengers. Hospital charges for hospitalized light truck passengers were 59% (95% CI, 40–87%) of the hospital charges of hospitalized car passengers (e−0.53 . = 0.59). The odds of fatality for light truck passengers were 14% (95% CI, 3–54%) of the odds of fatality for car passengers. In Figure 1 we present the adjusted rate of hospitalizations by vehicle type (light truck vs. car) and occupant type (driver vs. passenger) using estimates from the first specification. These rates were obtained by computing the probability Odds for each vehicle and occupant of hospitalization as 1+Odds configuration and averaging across all observations. On average, the adjusted rate of hospitalization was 13.8% for light truck passengers, 19.1% for car passengers, 6.3% for light truck drivers, and 10.5% for car drivers. The higher probability of hospitalization among passengers relative to drivers is an interesting finding and could be because passengers are less likely to be properly restrained relative to drivers and unrestrained
Light Truck Passenger
occupants are more likely to be ejected from the vehicle and hospitalized (Singleton and Qin 2003). Although our models control for seat belt use, a substantial portion of the occupants are coded as unknown (24% of drivers, 45% of passengers, Table A1). Using model estimates from specification 2, we predicted the adjusted hospital charges for hospitalized occupants. Given that the dependent variable was log-transformed, we predicted the level of hospital charges for different vehicle types and then averaged them across all hospitalization observations. Adjusted hospital charges, on average, were $2855 for a hospitalized light truck passenger, $4874 for a hospitalized car passenger, $10,778 for a hospitalized light truck driver, and $12,362 for a hospitalized car driver (results not shown in the figure). Rear seat seating, in general, is associated with lower likelihood of severe injury and fatality relative to front seat position (Smith and Cummings 2004), which may explain higher hospital charges for hospitalized drivers relative to hospitalized passengers conditional on vehicle type. To quantify the average expected hospital charges for an average driver or a passenger by vehicle type, we multiplied the adjusted probability of hospitalization presented in Figure 1 with the adjusted hospital charges for those hospitalized. As presented in Figure 2, adjusted hospital charges averaged across all of the occupant observations were $394 for a light truck passenger (0.138 × 2855), $931 for a car passenger (0.191 × 4874), $679 for a light truck driver (0.063 × 10,778), and $1298 for a car driver (0.105 × 12,362). Figure 3 presents adjusted fatality rates by vehicle and occupant type. Relying on model estimates from specification 3 in Tables 1 and 2 and, using the same approach as for hospitalization rates, the average adjusted rate of fatality was 0.09% for light truck passengers, 0.64% for car passengers, 0.10% for light truck drivers, and 0.32% for car drivers (Figure 3). Having estimated the probability of fatality by each vehicle–occupant type configuration, we quantified the fatality costs (value of lives lost) by multiplying these probabilities with $5.8 million, the value of a statistical life used by the U.S. Department of Transportation (Anderson 2008; Anderson and Auffhammer 2011; Li 2012; White 2004). The expected fatality cost was estimated to be $5220 for light truck passengers, $37,120 for car passengers, $5800 for light truck drivers, and $18,560 for car drivers (Figure 4).
13.8% Light Truck Passenger
Car Passenger
394
19.1% Car Passenger
Light Truck Driver
Car Driver
6.3%
Light Truck Driver
10.5%
Fig. 1. Adjusted rate of hospitalizations (%) (color figure available online).
Car Driver
931
679
1,298
Fig. 2. Hospital charges averaged across occupants ($) (color figure available online).
170
Karaca-Mandic and Lee Light Truck Passenger
0.09%
Car Passenger
Light Truck Driver
0.64%
0.10%
Car Driver
0.32%
Fig. 3. Adjusted rate of fatalities (color figure available online).
Downloaded by [FU Berlin] at 06:35 19 November 2014
Discussion and Conclusion Our findings suggested that drivers of light trucks had lower adjusted rates of hospitalization relative to drivers of cars. Similarly, passengers of light trucks had lower adjusted rates of hospitalization relative to passengers of cars. Hospital charges for hospitalized light truck passengers were lower than the hospital charges for hospitalized car passengers. There was no statistically significant difference between the hospital charges for hospitalized light truck drivers and car drivers. Drivers of light trucks had lower fatality rates relative to the drivers of cars. Similarly, fatality rates for light truck passengers were lower than the fatality rates for car passengers. Our estimates of the fatalities are comparable to previous studies. For example, Anderson (2008) reported that 80% of the increase in traffic fatalities due to the increased share of light trucks on the roads accrued to the occupants of vehicles other than light trucks and by pedestrians. Crashworthiness literature estimated that occupants in cars were 6 times more likely to die in a crash with a light truck than in a crash with another car because of vehicle incompatibility issues (Teoh and Nolan 2012). Our fatality estimates suggest that the fatality rate was 3 to 6 times higher for the car occupants relative to light truck occupants (drivers: 0.32% versus 0.10%; passengers: 0.64% versus 0.09%). Unlike previous literature (Anderson 2008; Anderson and Auffhammer 2011; Gayer 2004; Li 2012; Martinelli and Diasdado-De-La-Pena 2008; Wenzel and Ross 2005; White 2004), we considered injury costs in terms of hospital charges as well as fatality costs. We found that average hospitalization costs were nonnegligible because they represented 2.5–11.7% of the average fatality costs (7.5% for light truck passengers; 2.5% for car passengers; 11.7% for light truck drivers; 7.0% for car drivers; cf. Figures 2 and 4). It is important to note that the hospital charges we estimate represent only a por-
Light Truck Passenger
5,220
Car Passenger
Light Truck Driver
Car Driver
37,120
5,800
18,560
Fig. 4. Average fatality costs ($) (color figure available online).
tion of the total costs related to injury. For example, other monetary and nonmonetary costs of injury, such as pain and suffering, disability, productivity losses, and wage and income losses, are not accounted for in our estimates. For 2009, the National Safety Council estimated the average comprehensive costs per incapacitating injury, accounting for such factors, was $216,800, and the corresponding estimate per nonincapacitating injury was $55,300, and per possible injury it was $26,300. Naturally, the comprehensive cost of fatality was the highest at $4,300,000 (NSC 2011). However, the likelihood of injury (in our case measured in terms of hospitalization) was substantially higher than the likelihood of fatality as reported in Figures 1 and 3. The existing traffic liability systems (tort or no-fault systems) likely fail to fully make light trucks accountable for costs they impose on other cars, pedestrians, and other road occupants. This is especially the case in a no-fault state that limits the liability of drivers for damage or injury to others. The presence of such costs that are not accounted for by markets or regulations could lead to an inefficiently larger share of light truck on the roads. Evaluating the optimal amount of corrective taxes or other corrective policies is beyond the scope of this article. One of the limitations of our study is that we have not considered the cost implications of light trucks on pedestrians, bicyclists, and motorcyclists, which previous literature found to be substantial. Calculations of optimal taxes would require quantifying the cost implications of light trucks not only on passenger car occupants but also on other types of vehicle occupants, pedestrians, bicyclists, and motorcyclists. Moreover, one needs to consider the previously listed broad range of cost implications beyond hospitalization and fatality costs. We focused on 2-vehicle accidents that involved one car and one light truck and compared the levels of injury to occupants of cars and light trucks while stratifying our analyses by drivers and passengers. An alternative approach could, for example, also include 2-vehicle crashes that involved 2 cars and compare injury to the occupants of cars that crashed with another car versus with a light truck. Similarly, one could include 2-vehicle crashes of 2 light trucks. Though such approaches would improve the generalizability of our results, we are most interested in understanding what happens to the occupants of cars versus light trucks when they crash. Moreover, crashes between a car and light truck were very common in the data. Of the 112,624 two-vehicle crashes, 85,679 involved at least one car. Among those with at least one car, 44,394 also involved a light truck. Crashes that involved 2 light trucks comprised 13% of the 2-vehicle crashes. Another limitation of our study is that data are specific to the State of Minnesota and may not be generalizable to other states. From a legislative perspective, Minnesota has a no-fault auto liability system, as opposed to the more common tort system, which limits the driver’s liability from their actions and may reduce incentives to drive safely. The literature provides mixed evidence on whether driving behavior is different in no-fault liability systems. Though several studies found little or no evidence that crash rates are different in no-fault states (Derrig et al. 2002; Heaton and Helland 2010; Kochanowski and Young 1985; Loughran 2001), others point to empirical evidence that no-fault laws led to significant increases in
Downloaded by [FU Berlin] at 06:35 19 November 2014
Crashes With Light Trucks traffic fatalities (Cohen and Dehejia 2004; Cummins et al. 2001; Landes 1982). The concern regarding whether no-fault liability systems lead to reduced incentives to drive carefully is especially pronounced for drivers who drive heavier and larger vehicles, which are shown to cause more significant damage to other motor vehicles when they are involved in a crash. The no-fault system does not penalize the drivers of such vehicles. In fact, on the contrary, drivers of smaller vehicles suffer more damage and yet have to bear most of these costs themselves. If this is the case, our estimates of the injury and fatality costs on car drivers and passengers in Minnesota may be higher than those observed in states with tort liability systems. Interestingly, there have been substantial improvements in vehicle design aimed at improving vehicle compatibility. Recent data show that voluntary design standards as a result of a 2003 agreement by the major auto manufacturers have largely been successful. A comparison of 2-vehicle crash data for 2001–2004 to 2008–2009 showed that car crash partner fatality rate (fatality rate in the other vehicle in crash) declined broadly for all vehicle types in the latter period, but the reduction was particularly more pronounced for SUVs and pickup trucks colliding with passenger cars (Teoh and Nolan 2012). In our data, the average vehicle model year for cars and light trucks were 1996.6 (SD = 4.9) and 1997.9 (SD = 4.8), respectively. The evidence based on these most recent data suggests that our findings may not generalize to newer light truck designs. As more recent data become available, it also becomes important to monitor changes in improved vehicle designs as a result of greater adherence to voluntary agreements among vehicle manufacturers and their implication for the costs of light trucks.
Acknowledgments This research was supported by the Center for Urban and Regional Affairs of the University of Minnesota through a Faculty Interactive Research Program. Funding was used for research assistant support. The funding organization played no role in the conduct of this study. We greatly acknowledge the Minnesota Department of Health Injury & Violence Prevention Unit for facilitating access to data used in this study.
References Anderson M. Safety for whom? The effects of light trucks on traffic fatalities. J Health Econ. 2008;27:973–989. Anderson M, Auffhammer M. Pounds that kill: The external costs of vehicle weight. 2013. Rev Econ Stud. (in press). Braitman KA, Ferguson SA, Elharam K. Changes in driver fatality rates and vehicle incompatibility concurrent with changes in the passenger vehicle fleet. Public Health Rep. 2007;122:319–327. Carlin JB, Galati JC, Royston P. A new framework for managing and analyzing multiply imputed data in Stata. Stata J. 2008;8:49–67. Coate D, VanderHoff J. The truth about light trucks. Regulation. 2001;24:22–27. Cohen A, Dehejia R. The effect of automobile insurance and accident liability laws on traffic fatalities. J Law Econ. 2004;47:357–359. Cummins JD, Weiss MA, Phillips RD. The incentive effects of no-fault automobile insurance. J Law Econ. 2001;44:427–464. Derrig RA, Segui-Gomez M, Abtahi A, Liu LL. The effect of population safety belt usage rates on motor vehicle-related fatalities. Accid Anal Prev. 2002;34:101–110.
171 Duan N, Manning WM, Morris C, Newhouse JP. A comparison of alternative models for the demand for medical care. J Bus Econ Stat. 1983;1:115–126. Gayer T. The fatality risks of sport-utility vehicles, vans, and pickups relative to cars. J Risk Uncertain. 2004;28:103–133. Heaton P, Helland E. No-fault insurance and automobile accidents. The selected works of Paul Heaton. 2010. Available at: http:// www.rand.org/content/dam/rand/pubs/monographs/2010/RAND MG860.pdf. Accessed May 6, 2013. Johnson SW, Walker J. The Crash Outcome Data Evaluation System (CODES). Washington, DC: US Department of Transportation, National Highway Traffic Safety Administration; 1996. DOT HS 808 338. Kochanowski PS, Young MV. Deterrent aspects of no-fault automobile insurance: some empirical findings. J Risk Insur. 1985;52:269– 288. Landes EM. Insurance, liability, and accidents: a theoretical and empirical investigation of the effect of no-fault accidents. J Law Econ. 1982;25:49–65. Li S. Traffic safety and vehicle choice: quantifying the effects of the “arms race” on American roads. J Appl Econ. 2012;27:34–62. Loughran DS. The Effect of No-Fault Automobile Insurance on Driver Behavior and Automobile Accidents in the United States. Santa Monica, CA: RAND Corporation; 2001. Manning WG, Mullahy J. Estimating log models: to transform or not to transform? J Health Econ. 2001;20:461–494. Martinelli DR, Diasdado-De-La-Pena M. Safety Externalities of SUVs on Passenger Cars: An Analysis of the Peltzman Effect Using FARS Data. Morgantown, WV: Department of Civil and Environmental Engineering; 2008. McGlincy M. A Bayesian record linkage methodology for multiple imputation of missing links. 2004. Available at: http://www.amstat. org/sections/srms/proceedings/y2004/files/Jsm2004-000683.pdf. Accessed May 2012. Meyer JR, Gomez-Ibanez JA. Autos, Transit and Cities. Cambridge, MA: Harvard University Press; 1981. National Highway Traffic Safety Administration (NHTSA). Fatality Analysis Reporting System (FARS) Encyclopedia: Crashes. 2007. Available at: http://www-fars.nhtsa.dot.gov/Main/DidYouKnow. aspx. Accessed August 2011. National Safety Council (NSC). Estimating the Cost of Unintentional Injuries. 2011. Available at: http://www.nsc.org/news resources/ injury and death statistics/Pages/EstimatingtheCostsofUnintention alInjuries.aspx. Accessed August 2011. Singleton M, Qin H. Risk Factors for Death or Hospitalization Among Occupants of Passenger Motor Vehicles That Were Severely Damaged in Crashes in Kentucky, 2000–2001. Lexington, KY: Kentucky Injury Prevention & Research Center; 2003. Smith KM, Cummings P. Passenger seating position and the risk of passenger death or injury in traffic crashes. Accid Anal Prev. 2004;36:257–260. StataCorp. Stata Statistical Software: Release 11. College Station, TX: StataCorp LP; 2009. Teoh ER, Nolan JM. Is passenger vehicle incompatibility still a problem? Traffic Inj Prev. 2012;13:585–591. U.S. Census Bureau. 1997 Economic Census: Vehicle Inventory and Use Survey. 1999a. Available at: http://www.census.gov/prod/ec97/ 97tv-mn.pdf. Accessed August 2011. U.S. Census Bureau. 1997 Economic Census: Vehicle Inventory and Use Survey, Geographic Area Series. 1999b. Available at: http://www. census.gov/prod/ec97/97tv-us.pdf. Accessed August 2011. U.S. Census Bureau. Vehicle Inventory and Use: Products. 2002. Available at: http://www.census.gov/svsd/www/vius/products.html. Accessed August 2011. Wenzel TP, Ross M. The effects of vehicle model and driver behavior on risk. Accid Anal Prev. 2005;37:479–494. White M. The arms race on American roads: the effect of sport utility vehicles and pick-up trucks on traffic safety. J Law Econ. 2004;47:333–356.
Traffic Injury Prevention Publication details, including instructions for authors and subscription information: http://www.tandfonline.com/loi/gcpi20
Hospitalizations and Fatalities in Crashes With Light Trucks a
Pinar Karaca-Mandic & Jinhyung Lee
b
a
University of Minnesota, School of Public Health, Division of Health Policy and Management , Minneapolis , Minnesota b
Jiann Ping-Hsu College of Public Health , Georgia Southern University , Statesboro , Georgia Accepted author version posted online: 14 Jun 2013.Published online: 17 Dec 2013.
Click for updates To cite this article: Pinar Karaca-Mandic & Jinhyung Lee (2014) Hospitalizations and Fatalities in Crashes With Light Trucks, Traffic Injury Prevention, 15:2, 165-171, DOI: 10.1080/15389588.2013.803279 To link to this article: http://dx.doi.org/10.1080/15389588.2013.803279
PLEASE SCROLL DOWN FOR ARTICLE Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”) contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensors make no representations or warranties whatsoever as to the accuracy, completeness, or suitability for any purpose of the Content. Any opinions and views expressed in this publication are the opinions and views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content should not be relied upon and should be independently verified with primary sources of information. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands, costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly or indirectly in connection with, in relation to or arising out of the use of the Content. This article may be used for research, teaching, and private study purposes. Any substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distribution in any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found at http:// www.tandfonline.com/page/terms-and-conditions
Traffic Injury Prevention (2014) 15, 165–171 C Taylor & Francis Group, LLC Copyright ISSN: 1538-9588 print / 1538-957X online DOI: 10.1080/15389588.2013.803279
Hospitalizations and Fatalities in Crashes With Light Trucks PINAR KARACA-MANDIC1 and JINHYUNG LEE2 1 2
University of Minnesota, School of Public Health, Division of Health Policy and Management, Minneapolis, Minnesota Jiann Ping-Hsu College of Public Health, Georgia Southern University, Statesboro, Georgia
Downloaded by [FU Berlin] at 06:35 19 November 2014
Received 21 February 2013, Accepted 5 May 2013
Objective: This study examined 2-car crashes including one passenger car and one light truck (van, minivan, pickup truck, or sport utility vehicle) and investigated the likelihood of hospitalization, hospitalization charges, and the likelihood of fatality of an occupant by vehicle type differentiating between passengers and drivers. Methods: We used unique data from Minnesota’s Crash Outcome Data Evaluation System (CODES) linked with hospital discharge data from 2004 to 2005. We focused on police-reported crashes that involved 2 vehicles, one car and one light truck. First, we estimated models to predict the likelihood of hospitalization. Next, we estimated models to predict hospitalization charges among the hospitalized. Finally, we modeled the likelihood of fatality. In all models, we distinguished between the 2 vehicle types and controlled for a broad range of occupant, crash, and vehicle characteristics. We estimated separate models for passengers and drivers. Results: We found that in a crash between a car and a light truck, drivers of light trucks were less likely to be hospitalized (odds ratio [OR] = 64%; 95% confidence interval [CI], 59–70%) and killed (OR = 35%; 95% CI, 18–68%) relative to the drivers of cars. Similarly, passengers of light trucks had a lower likelihood of hospitalization (OR = 66%; 95% CI, 57–77%) and fatality (OR = 14%; 95% CI, 3–54%) relative to the passengers of cars. Among hospitalized occupants, we did not find statistically significant differences in hospital charges between light truck drivers and car drivers, but hospital charges for hospitalized light truck passengers were 59% (95% CI, 40–87%) of the hospital charges of hospitalized car passengers. Conclusions: Though previous studies have shown high fatality costs associated with light trucks, this study is the first to explore the hospitalization costs associated with these vehicles. The existing traffic liability systems (tort or no-fault systems) likely fail to fully make light trucks accountable for costs they impose on other cars, pedestrians, and other road occupants. Our findings suggest the importance of a close examination of a broad range of cost implications even beyond hospitalization and fatality costs to evaluate the optimal amount of corrective taxes or other corrective policies in future research. Supplemental materials are available for this article. Go to the publisher’s online edition of Traffic Injury Prevention to view the supplemental file. Keywords: light trucks, hospitalization, fatality, traffic safety
Introduction Over the last 2 decades, there has been a significant increase in the number of light trucks, including vans, minivans, pickup trucks, and sport utility vehicles (SUVs), on the roads. For example, based on the Census Bureau’s Vehicle Inventory and Use Surveys (U.S. Census Bureau 1999a; U.S. Census Bureau 1999b; U.S. Census Bureau 2002), there was an increase of 60.5% in the number of registered minivans from 1992 to 1997 and an increase of 24.1% from 1997 to 2002. The percentage increases in the number of registered SUVs were even larger: 92.7% from 1992 to 1997 and 55.8% from 1997 to 2002. An important contributor to the increased popularity of light trucks is the fact that they provide improved protection
Address correspondence to Pinar Karaca-Mandic, PhD, University of Minnesota School of Public Health, Division of Health Policy and Management, 420 Delaware St. SE, MMC 729, Minneapolis, MN 55455. E-mail: [email protected]
to their own passengers (Coate and VanderHoff 2001). However, these vehicles likely impose significant damage on the passengers of the other cars they crash into, as well as to pedestrians, bicyclists, and motorcyclists. For example, Meyer and Gomez-Ibanez (1981) found that conditional on a crash between a passenger car and a light truck, occupants of large vehicles were 29 percent less likely to be seriously injured. Similarly, Wenzel and Ross (2005) showed that SUVs and pickup trucks imposed greater driver fatalities on passenger cars and that the fatality risks increase with the size of the larger vehicle. A large body of literature on crashworthiness shows that in a crash between a car and a light truck, the taller body, higher mass, and stiffer front structure of the light trucks relative to the cars result in vehicle incompatibility, especially in frontto-front and front-to-side crashes, disadvantaging the cars. Though vehicle incompatibility is associated with a higher probability of fatality in occupants of the cars relative to light trucks, advances in vehicle designs over time have resulted in the improved safety of cars involved in crashes with light trucks (Braitman et al. 2007; Teoh and Nolan 2012).
Downloaded by [FU Berlin] at 06:35 19 November 2014
166 Several studies have quantified the fatality costs associated with light trucks. White (2004) estimated that for every additional one million drivers of SUVs there would be between 34 and 93 additional fatalities involving passenger car occupants, pedestrians, bicyclists, or motorcyclists per year. The value of these lost lives would range from $242 to $652 million per year. Similarly, Anderson (2008) found that a one percentage point increase in the share of light trucks (which he defined as SUVs and pickup trucks) on the roads was associated with a 0.34% increase in annual traffic fatalities, or 143 additional fatalities per year. Further, he found that whereas 20% of the increase in traffic fatalities accrued to the occupants of light trucks, 80% accrued to the occupants of other vehicles and pedestrians (external costs). Recently, Anderson and Auffhammer (2011) found that the baseline fatality probability increased 47% by being hit by a 1000-pound heavier vehicle and estimated the total external costs from fatalities to be $93 billion per year. Similarly, Gayer (2004), Martinelli and DiasdadoDe-La-Pena (2008), and Li (2012) confirmed the general finding that light trucks cause more damage to the occupants of other vehicles but less damage to their own occupants. Li (2012) estimated that the external costs associated with a light truck were around $2444 during the vehicle’s lifetime. In this article, we estimate the magnitude of hospitalization costs as well as the fatality costs that light trucks cause to occupants of cars and to their own occupants. From a policy perspective, to design market-based incentives or regulations to make light trucks accountable for the additional cost they impose on other cars, it is important to understand the broad range of costs associated with these vehicles. Though previous studies have quantified the costs these larger vehicles impose on other occupants and their own occupants in terms of fatalities, the cost implications in terms of hospitalizations are unknown. Naturally, the cost of a fatality is considerably higher than the cost of a hospitalization, given that the value of a statistical life, as estimated by the U.S. Department of Transportation, is estimated to be around $5.8 million (Anderson and Auffhammer 2013). However, the rate of injuries and hospitalizations after a crash is substantially higher than the fatality rate. In 2007, of the 6 million police-reported crashes in the United States, 30% involved an injury, and less than 1% involved a fatality (NHTSA 2007).
Material and Methods Data We used a unique database, the Crash Outcome Data Evaluation System (CODES), 2004 and 2005, which contains linked data from the Minnesota Hospital Association, the Emergency Medical Services (EMS) Regulatory Board, and the Minnesota departments of Health, Public Safety, and Transportation. The crash data within CODES come from the Department of Public Safety and provide information on all police-reported crashes and on all persons involved in the crash (age, gender, drinking status, driver/pedestrian status, seat belt use, etc.) as well as detailed information on the crash
Karaca-Mandic and Lee (location, time, road type, road conditions, number of vehicles involved, etc.). These police-reported crash data were linked with hospital data, which included information on hospital charges for the hospitalized crash victims. Ideally, each crash associated with a hospitalization would have a relevant medical record that can be linked, and the absence of a linkage should indicate the absence of a hospitalization event. However, in reality, the crash records may not be linked with a medical record despite a hospital event, or a matched hospital record may not indicate a crash. As a result, the CODES project uses a probabilistic record linkage method to account for missing unique identifiers for injured persons and medical events. The detailed algorithm for the linkage is described in the U.S. Department of Transportation, NHTSA (1996) technical report (Johnson and Walker 1996). Instead of a single set of linked records, Minnesota CODES data are provided as multiple (5) sets of imputed, linked records using a full Bayesian model as described in McGlincy (2004). We focused on police-reported crashes that involved 2 vehicles, one car and one light truck. Car was defined as a passenger car, and the definition of light trucks in the data included pickup trucks, vans, minivans, and SUVs. Outcome Variables We considered 3 outcome variables: hospitalization of vehicle occupants, hospital charges for hospitalized individuals, and fatality outcome of vehicle occupants. Hospitalization was characterized into a binary indicator based on whether or not the individual was taken to the hospital. Hospital charges represented an aggregate of charges for all hospital treatments, from acute care to discharge. An acute care period lasts until the patient is discharged to home, transitional care, or rehabilitation. If the patient was discharged but readmitted within 2 days, we considered charges associated with the readmission as well. Finally, the fatality outcome was a binary indicator extracted from a variable that characterized injury severity (no apparent injury, possible injury, moderate injury, severe injury, and killed) for each vehicle occupant in the crash. Explanatory Variables We identified each type of vehicle (car or a light truck) involved in the crash. We also extracted information on the characteristics of the vehicle occupants involved in the crash (age and gender, seat belt use, physical condition such as normal, driving under the influence of alcohol or drugs, fatigued, other physical disabilities, etc.), as well as information on the characteristics of the vehicle (make/model, model year, number of passengers in the vehicle, and auto insurance status of the vehicle) and characteristics of the crash (weather conditions such as clear, snow, rain, etc.; road conditions such as dry, wet, ice, etc.; the time of crash such as daylight, sunrise, sunset, etc.). All of our analyses controlled for the weight of the occupant’s own vehicle and the weight of the other vehicle involved in the crash. Data on vehicle weight were available for 62% of the vehicles. In 34% of the crashes, both the weight of the occupant’s own vehicle and the weight of the other vehicle
Crashes With Light Trucks
167
Table 1. Models of hospitalization, hospital charges, and driver fatalitiesa
Downloaded by [FU Berlin] at 06:35 19 November 2014
Explanatory Variables Light truck Gender of occupant (male = 1, female = 0) Age 0–18 19–45 46–64 65 and over Urban Time of accident Light Sunrise Sunset Dark Road conditions during the accident Dry Wet Snow Ice Physical condition of occupant Normal Alcohol or drug Occupant belt use No belt Proper seat belt use Other/improper seat belt use Number of occupants in the vehicle Weight of own vehicle (in 1000 lb) Weight of other vehicle (in 1000 lb) Own vehicle model fixed effects Own vehicle year of make fixed effects
Specification 1
Specification 2
Specification 3
Hospitalization (1/0)
Total hospital charges conditional on hospital admission ($)
Fatality (1/0)
Logistic
Generalized linear model (log link, gamma family)
Logistic
95% Confidence interval
Adjusted odds ratio
95% Confidence interval
−0.14 0.00
(−0.37, 0.09) (−0.18, 0.18)
0.35∗∗∗ 1.00
(0.18, 0.68) (0.64, 1.57)
(1.02,1.24) (1.23, 1.54) (1.46, 1.91) (0.56, 0.64)
Ref 0.15 0.67∗∗∗ 0.90∗∗∗ −0.04
(−0.19, 0.48) (0.27, 1.07) (0.36, 1.44) (−0.22, 0.14)
Ref 0.91 1.55 3.62∗∗∗ 0.16∗∗∗
(0.47, 1.77) (0.75, 3.21) (1.70, 7.72) (0.09, 0.31)
Ref 1.26∗ 1.24∗∗ 1.23∗∗∗
(0.97, 1.63) (1.02, 151) (1.14, 1.35)
Ref 0.72∗∗ −0.08 −0.08
(0.00, 1.43) (−0.52, 0.37) (−0.30, 0.14)
Ref 2.01 1.22 1.94∗∗
(0.45, 9.05) (0.42, 3.53) (1.13, 3.33)
Ref 0.90∗ 0.60∗∗∗ 0.64∗∗∗
(0.80,1.01) (0.50, 0.72) (0.55, 0.74)
Ref 0.20 0.27 0.22
(−0.12, 0.52) (−0.16, 0.69) (−0.19, 0.63)
Ref 0.53 0.59 0.70
(0.19, 1.45) (0.19, 1.82) (0.31, 1.57)
Ref 2.75∗∗∗
(2.32, 3.25)
Ref 0.95∗∗
(0.12, 1.76)
Ref 1.82
(0.59, 5.62)
Ref 0.20∗∗∗ 0.28∗∗∗ 1.12∗∗∗
(0.18, 0.23) (0.12, 0.62) (1.08, 1.16)
Ref −0.61∗∗∗ −1.26∗∗∗ 0.01
(−0.86, −0.36) (−1.61, −0.91) (−0.09, 0.12)
Adjusted odds ratio
95% Confidence interval
0.64∗∗∗ 0.59∗∗∗
(0.59, 0.70) (0.56, 0.63)
Ref 1.13∗∗ 1.38∗∗∗ 1.67∗∗∗ 0.60∗∗∗
0.97 1.01 Included Included
(0.92, 1.01) (0.96, 1.05)
Coefficient estimate
−0.08∗∗ 0.13∗∗ Included Included
(−0.13, −0.02) (0.03, 0.23)
Ref 0.05∗∗∗ Omitted 0.95 1.22∗ 1.02 Included Included
(0.03, 0.08) (0.72, 1.26) (0.99, 1.50) (0.75, 1.38)
Standard errors are clustered at the crash level. All other explanatory variables listed in the text and in Table A1 are included in the models but not reported in this table. They are available from the authors upon request. ∗ P < .10. ∗∗ P < .05. ∗∗∗ P < .01.
were available. We used “unknown” indicators for when we did not know the weight of the occupant’s own vehicle or the weight of the other vehicle. Statistical Methods We estimated a model of hospitalization at the occupant-level separately for drivers and passengers. We used logistic regression to relate the binary variable characterizing hospitalization and whether the vehicle was a car or a light truck, occupant characteristics, vehicle characteristics, and crash characteristics listed above. For unknown values of any given characteristic, we included an unknown category and included the unknown indicators in the statistical models. Next, we estimated the hospital charges at the occupant level for occupants who were hospitalized. Mod-
els were estimated separately for drivers and passengers. We used a generalized linear model with log link and gamma family because hospital charges are typically not normally distributed (Duan et al. 1983; Manning and Mullahy 2001). Because the hospital discharge data are linked to crash data via probabilistic linkage, which provided 5 possible matches for each hospitalization, we used appropriate multiple imputation regression models (“mim”) in Stata software, version 12 (StataCorp 2009). The “mim” command allows for essentially repeating the hospitalization analyses 5 times and combining the parameter estimates across all analyses (Carlin et al. 2008). We used the same explanatory variables to characterize occupant, vehicle, and crash characteristics. An additional characteristic we considered was the payer source for the hospitalization (private insurance; government insurance, including
168
Karaca-Mandic and Lee
Table 2. Models of hospitalization, hospital charges, and passenger fatalitiesa
Downloaded by [FU Berlin] at 06:35 19 November 2014
Explanatory Variables Light truck Gender of occupant (male = 1, female = 0) Age 0–18 19–45 46–64 65 and over Urban Time of accident Light Sunrise Sunset Dark Road conditions during the accident Dry Wet Snow Ice Physical condition of occupant Normal Alcohol or drug Occupant belt use No belt Proper seat belt use Other/improper seat belt use Number of occupants in the vehicle Weight of own vehicle (in 1000 lb) Weight of other vehicle (in 1000 lb) Own vehicle model fixed effects Own vehicle year of make fixed effects
Specification 1
Specification 2
Specification 3
Hospitalization (1/0)
Total hospital charges conditional on hospital admission ($)
Fatality (1/0)
Logistic
Generalized linear model (log link, gamma family)
Logistic
Adjusted odds ratio
95% Confidence interval
Coefficient estimate
95% Confidence interval
Adjusted odds ratio
95% Confidence interval
0.66∗∗∗ 0.64∗∗∗
(0.57, 0.77) (0.58, 0.70)
−0.53∗∗∗ −0.14
(−0.92, −0.14) (−0.44, 0.17)
0.14∗∗∗ 0.37∗∗∗
(0.03, 0.54) (0.17, 0.78)
Ref 1.19∗∗∗ 1.61∗∗∗ 2.14∗∗∗ 0.75∗∗∗
(1.07,1.34) (1.37, 1.88) (1.78, 2.59) (0.67, 0.85)
Ref −0.02 0.38∗ 0.77∗∗ −0.04
(−0.32, 0.27) (−0.04, 0.79) (0.37, 1.17) (−0.28, 0.21)
Ref 0.73 1.22 7.04∗∗∗ 0.34∗∗
(0.28, 1.88) (0.39, 3.83) (2.67, 18.59) (0.13, 0.86)
Ref 1.21 0.98 1.24∗∗∗
(0.68, 2.15) (0.69, 1.37) (1.08, 1.43)
Ref 1.61∗∗∗ −0.02 0.25∗
(0.39, 2.82) (−0.67, 0.63) (−0.03, 0.53)
Ref Omitted 1.86 1.39
(0.30, 11.44) (0.63, 3.02)
Ref 0.88 0.58∗∗∗ 0.64∗∗∗
(0.72, 1.07) (0.42, 0.81) (0.50, 0.82)
Ref −0.10 −0.24 −0.22
(−0.54, 0.35) (−0.92, 0.45) (−0.74, 0.29)
Ref Omitted Ref 0.24∗∗∗ 0.47∗∗∗ 1.03 0.99 1.08∗∗ Included Included
Ref Omitted
(0.20, 0.28) (0.30, 0.76) (0.98, 1.10) (0.92, 1.06) (1.01, 1.17)
Ref −0.83∗∗∗ −1.06∗∗ −0.08 −0.009 0.032 Included Included
Ref 0.88 1.02 0.86
(0.30, 2.55) (0.34, 3.10) (0.24, 3.16)
Ref Omitted
(−1.36, −0.29) (−2.04, −0.08) (−0.27, 0.08) (−0.096, 0.079) (−0.097, 0.160)
Ref 0.06∗∗∗ 0.32 1.26 1.15 0.38∗∗ Included Included
(0.03, 0.12) (0.04, 2.90) (0.89, 1.77) (0.84, 1.58) (0.18, 0.80)
aStandard
errors are clustered at the crash level. All other explanatory variables listed in the text and in Table A1 are included in the models but not reported in this table. They are available from the authors upon request. ∗ P < .10. ∗∗ P < .05. ∗∗∗ P < .01.
Medicare/Medicaid/MN Care; a commercial insurance; and self-payment). Finally, we estimated a model of fatality at the occupant level. Using a logistic regression, we related the binary fatality variable to the same explanatory variables used in the hospitalization model. In all of the models, we clustered standard errors at the crash level to account for factors unobserved by us but that may be correlated for vehicles and occupants in a given crash.
Results There were 184,040 motor vehicle crashes in Minnesota in 2004 and 2005. Of these crashes, 32% involved only one vehicle, 61% involved 2 vehicles, and the rest involved more
than 2 vehicles. Of the one-vehicle crashes, 52% involved a car and 45% involved a light truck. Of the 112,624 two-vehicle crashes, 85,679 involved at least one car. Among those with at least one car, 44,394 also involved a light truck. These crashes involved 126,737 occupants (88,422 drivers and 38,315 passengers). Seven percent of drivers and 16% of passengers were hospitalized. About 0.10% of the drivers and passengers were killed. Average hospital charges for hospitalized occupants were $6045 for drivers and $4818 for passengers. Fifty-seven percent of drivers and 45% of passengers were male. Nearly 11% of drivers and 25% of passengers were under 18 years old; 8% of drivers and 6% passengers were over age 65 (Table A1; see online supplement). Table 1 presents model estimates for drivers corresponding to the 3 outcomes: hospitalization (specification 1), hospital charges for those hospitalized (specification 2), and
Downloaded by [FU Berlin] at 06:35 19 November 2014
Crashes With Light Trucks
169
fatality (specification 3). Based on the first specification, the odds of hospitalization for light truck drivers were 64% (95% confidence interval CI, 59–70%) of the odds of hospitalization for car drivers. In the second specification, the coefficient estimate represents the expected change in log-transformed hospital charges from a unit change in the independent variable. The reported estimate on the light truck indicator was not statistically significant, which suggests that there was no statistically significant difference between hospital charges for hospitalized light truck drivers and car drivers. Based on the third specification, the odds of fatality for light truck drivers were 35% (95% CI, 18–68%) of the odds of fatality for car drivers. Other factors associated with a higher odds of hospitalization included older age, sunset or darkness (relative to daytime), alcohol or drug impairment, and number of occupants in the vehicle. Wet, snowy, or icy road conditions (relative to dry road conditions) and seat belt use were associated with a lower odds of hospitalization. Older age and darkness were associated with higher odds of fatality, and proper seat belt use was associated with lower odds of fatality. Hospital charges among those hospitalized decreased with the use of seat belts and weight of occupant’s own vehicle and increased with the weight of the other vehicle. Table 2 presents similar model estimates for passengers. We found that the odds of hospitalization for light truck passengers were 66% (95% CI, 57–77%) of the odds of hospitalization for car passengers. Hospital charges for hospitalized light truck passengers were 59% (95% CI, 40–87%) of the hospital charges of hospitalized car passengers (e−0.53 . = 0.59). The odds of fatality for light truck passengers were 14% (95% CI, 3–54%) of the odds of fatality for car passengers. In Figure 1 we present the adjusted rate of hospitalizations by vehicle type (light truck vs. car) and occupant type (driver vs. passenger) using estimates from the first specification. These rates were obtained by computing the probability Odds for each vehicle and occupant of hospitalization as 1+Odds configuration and averaging across all observations. On average, the adjusted rate of hospitalization was 13.8% for light truck passengers, 19.1% for car passengers, 6.3% for light truck drivers, and 10.5% for car drivers. The higher probability of hospitalization among passengers relative to drivers is an interesting finding and could be because passengers are less likely to be properly restrained relative to drivers and unrestrained
Light Truck Passenger
occupants are more likely to be ejected from the vehicle and hospitalized (Singleton and Qin 2003). Although our models control for seat belt use, a substantial portion of the occupants are coded as unknown (24% of drivers, 45% of passengers, Table A1). Using model estimates from specification 2, we predicted the adjusted hospital charges for hospitalized occupants. Given that the dependent variable was log-transformed, we predicted the level of hospital charges for different vehicle types and then averaged them across all hospitalization observations. Adjusted hospital charges, on average, were $2855 for a hospitalized light truck passenger, $4874 for a hospitalized car passenger, $10,778 for a hospitalized light truck driver, and $12,362 for a hospitalized car driver (results not shown in the figure). Rear seat seating, in general, is associated with lower likelihood of severe injury and fatality relative to front seat position (Smith and Cummings 2004), which may explain higher hospital charges for hospitalized drivers relative to hospitalized passengers conditional on vehicle type. To quantify the average expected hospital charges for an average driver or a passenger by vehicle type, we multiplied the adjusted probability of hospitalization presented in Figure 1 with the adjusted hospital charges for those hospitalized. As presented in Figure 2, adjusted hospital charges averaged across all of the occupant observations were $394 for a light truck passenger (0.138 × 2855), $931 for a car passenger (0.191 × 4874), $679 for a light truck driver (0.063 × 10,778), and $1298 for a car driver (0.105 × 12,362). Figure 3 presents adjusted fatality rates by vehicle and occupant type. Relying on model estimates from specification 3 in Tables 1 and 2 and, using the same approach as for hospitalization rates, the average adjusted rate of fatality was 0.09% for light truck passengers, 0.64% for car passengers, 0.10% for light truck drivers, and 0.32% for car drivers (Figure 3). Having estimated the probability of fatality by each vehicle–occupant type configuration, we quantified the fatality costs (value of lives lost) by multiplying these probabilities with $5.8 million, the value of a statistical life used by the U.S. Department of Transportation (Anderson 2008; Anderson and Auffhammer 2011; Li 2012; White 2004). The expected fatality cost was estimated to be $5220 for light truck passengers, $37,120 for car passengers, $5800 for light truck drivers, and $18,560 for car drivers (Figure 4).
13.8% Light Truck Passenger
Car Passenger
394
19.1% Car Passenger
Light Truck Driver
Car Driver
6.3%
Light Truck Driver
10.5%
Fig. 1. Adjusted rate of hospitalizations (%) (color figure available online).
Car Driver
931
679
1,298
Fig. 2. Hospital charges averaged across occupants ($) (color figure available online).
170
Karaca-Mandic and Lee Light Truck Passenger
0.09%
Car Passenger
Light Truck Driver
0.64%
0.10%
Car Driver
0.32%
Fig. 3. Adjusted rate of fatalities (color figure available online).
Downloaded by [FU Berlin] at 06:35 19 November 2014
Discussion and Conclusion Our findings suggested that drivers of light trucks had lower adjusted rates of hospitalization relative to drivers of cars. Similarly, passengers of light trucks had lower adjusted rates of hospitalization relative to passengers of cars. Hospital charges for hospitalized light truck passengers were lower than the hospital charges for hospitalized car passengers. There was no statistically significant difference between the hospital charges for hospitalized light truck drivers and car drivers. Drivers of light trucks had lower fatality rates relative to the drivers of cars. Similarly, fatality rates for light truck passengers were lower than the fatality rates for car passengers. Our estimates of the fatalities are comparable to previous studies. For example, Anderson (2008) reported that 80% of the increase in traffic fatalities due to the increased share of light trucks on the roads accrued to the occupants of vehicles other than light trucks and by pedestrians. Crashworthiness literature estimated that occupants in cars were 6 times more likely to die in a crash with a light truck than in a crash with another car because of vehicle incompatibility issues (Teoh and Nolan 2012). Our fatality estimates suggest that the fatality rate was 3 to 6 times higher for the car occupants relative to light truck occupants (drivers: 0.32% versus 0.10%; passengers: 0.64% versus 0.09%). Unlike previous literature (Anderson 2008; Anderson and Auffhammer 2011; Gayer 2004; Li 2012; Martinelli and Diasdado-De-La-Pena 2008; Wenzel and Ross 2005; White 2004), we considered injury costs in terms of hospital charges as well as fatality costs. We found that average hospitalization costs were nonnegligible because they represented 2.5–11.7% of the average fatality costs (7.5% for light truck passengers; 2.5% for car passengers; 11.7% for light truck drivers; 7.0% for car drivers; cf. Figures 2 and 4). It is important to note that the hospital charges we estimate represent only a por-
Light Truck Passenger
5,220
Car Passenger
Light Truck Driver
Car Driver
37,120
5,800
18,560
Fig. 4. Average fatality costs ($) (color figure available online).
tion of the total costs related to injury. For example, other monetary and nonmonetary costs of injury, such as pain and suffering, disability, productivity losses, and wage and income losses, are not accounted for in our estimates. For 2009, the National Safety Council estimated the average comprehensive costs per incapacitating injury, accounting for such factors, was $216,800, and the corresponding estimate per nonincapacitating injury was $55,300, and per possible injury it was $26,300. Naturally, the comprehensive cost of fatality was the highest at $4,300,000 (NSC 2011). However, the likelihood of injury (in our case measured in terms of hospitalization) was substantially higher than the likelihood of fatality as reported in Figures 1 and 3. The existing traffic liability systems (tort or no-fault systems) likely fail to fully make light trucks accountable for costs they impose on other cars, pedestrians, and other road occupants. This is especially the case in a no-fault state that limits the liability of drivers for damage or injury to others. The presence of such costs that are not accounted for by markets or regulations could lead to an inefficiently larger share of light truck on the roads. Evaluating the optimal amount of corrective taxes or other corrective policies is beyond the scope of this article. One of the limitations of our study is that we have not considered the cost implications of light trucks on pedestrians, bicyclists, and motorcyclists, which previous literature found to be substantial. Calculations of optimal taxes would require quantifying the cost implications of light trucks not only on passenger car occupants but also on other types of vehicle occupants, pedestrians, bicyclists, and motorcyclists. Moreover, one needs to consider the previously listed broad range of cost implications beyond hospitalization and fatality costs. We focused on 2-vehicle accidents that involved one car and one light truck and compared the levels of injury to occupants of cars and light trucks while stratifying our analyses by drivers and passengers. An alternative approach could, for example, also include 2-vehicle crashes that involved 2 cars and compare injury to the occupants of cars that crashed with another car versus with a light truck. Similarly, one could include 2-vehicle crashes of 2 light trucks. Though such approaches would improve the generalizability of our results, we are most interested in understanding what happens to the occupants of cars versus light trucks when they crash. Moreover, crashes between a car and light truck were very common in the data. Of the 112,624 two-vehicle crashes, 85,679 involved at least one car. Among those with at least one car, 44,394 also involved a light truck. Crashes that involved 2 light trucks comprised 13% of the 2-vehicle crashes. Another limitation of our study is that data are specific to the State of Minnesota and may not be generalizable to other states. From a legislative perspective, Minnesota has a no-fault auto liability system, as opposed to the more common tort system, which limits the driver’s liability from their actions and may reduce incentives to drive safely. The literature provides mixed evidence on whether driving behavior is different in no-fault liability systems. Though several studies found little or no evidence that crash rates are different in no-fault states (Derrig et al. 2002; Heaton and Helland 2010; Kochanowski and Young 1985; Loughran 2001), others point to empirical evidence that no-fault laws led to significant increases in
Downloaded by [FU Berlin] at 06:35 19 November 2014
Crashes With Light Trucks traffic fatalities (Cohen and Dehejia 2004; Cummins et al. 2001; Landes 1982). The concern regarding whether no-fault liability systems lead to reduced incentives to drive carefully is especially pronounced for drivers who drive heavier and larger vehicles, which are shown to cause more significant damage to other motor vehicles when they are involved in a crash. The no-fault system does not penalize the drivers of such vehicles. In fact, on the contrary, drivers of smaller vehicles suffer more damage and yet have to bear most of these costs themselves. If this is the case, our estimates of the injury and fatality costs on car drivers and passengers in Minnesota may be higher than those observed in states with tort liability systems. Interestingly, there have been substantial improvements in vehicle design aimed at improving vehicle compatibility. Recent data show that voluntary design standards as a result of a 2003 agreement by the major auto manufacturers have largely been successful. A comparison of 2-vehicle crash data for 2001–2004 to 2008–2009 showed that car crash partner fatality rate (fatality rate in the other vehicle in crash) declined broadly for all vehicle types in the latter period, but the reduction was particularly more pronounced for SUVs and pickup trucks colliding with passenger cars (Teoh and Nolan 2012). In our data, the average vehicle model year for cars and light trucks were 1996.6 (SD = 4.9) and 1997.9 (SD = 4.8), respectively. The evidence based on these most recent data suggests that our findings may not generalize to newer light truck designs. As more recent data become available, it also becomes important to monitor changes in improved vehicle designs as a result of greater adherence to voluntary agreements among vehicle manufacturers and their implication for the costs of light trucks.
Acknowledgments This research was supported by the Center for Urban and Regional Affairs of the University of Minnesota through a Faculty Interactive Research Program. Funding was used for research assistant support. The funding organization played no role in the conduct of this study. We greatly acknowledge the Minnesota Department of Health Injury & Violence Prevention Unit for facilitating access to data used in this study.
References Anderson M. Safety for whom? The effects of light trucks on traffic fatalities. J Health Econ. 2008;27:973–989. Anderson M, Auffhammer M. Pounds that kill: The external costs of vehicle weight. 2013. Rev Econ Stud. (in press). Braitman KA, Ferguson SA, Elharam K. Changes in driver fatality rates and vehicle incompatibility concurrent with changes in the passenger vehicle fleet. Public Health Rep. 2007;122:319–327. Carlin JB, Galati JC, Royston P. A new framework for managing and analyzing multiply imputed data in Stata. Stata J. 2008;8:49–67. Coate D, VanderHoff J. The truth about light trucks. Regulation. 2001;24:22–27. Cohen A, Dehejia R. The effect of automobile insurance and accident liability laws on traffic fatalities. J Law Econ. 2004;47:357–359. Cummins JD, Weiss MA, Phillips RD. The incentive effects of no-fault automobile insurance. J Law Econ. 2001;44:427–464. Derrig RA, Segui-Gomez M, Abtahi A, Liu LL. The effect of population safety belt usage rates on motor vehicle-related fatalities. Accid Anal Prev. 2002;34:101–110.
171 Duan N, Manning WM, Morris C, Newhouse JP. A comparison of alternative models for the demand for medical care. J Bus Econ Stat. 1983;1:115–126. Gayer T. The fatality risks of sport-utility vehicles, vans, and pickups relative to cars. J Risk Uncertain. 2004;28:103–133. Heaton P, Helland E. No-fault insurance and automobile accidents. The selected works of Paul Heaton. 2010. Available at: http:// www.rand.org/content/dam/rand/pubs/monographs/2010/RAND MG860.pdf. Accessed May 6, 2013. Johnson SW, Walker J. The Crash Outcome Data Evaluation System (CODES). Washington, DC: US Department of Transportation, National Highway Traffic Safety Administration; 1996. DOT HS 808 338. Kochanowski PS, Young MV. Deterrent aspects of no-fault automobile insurance: some empirical findings. J Risk Insur. 1985;52:269– 288. Landes EM. Insurance, liability, and accidents: a theoretical and empirical investigation of the effect of no-fault accidents. J Law Econ. 1982;25:49–65. Li S. Traffic safety and vehicle choice: quantifying the effects of the “arms race” on American roads. J Appl Econ. 2012;27:34–62. Loughran DS. The Effect of No-Fault Automobile Insurance on Driver Behavior and Automobile Accidents in the United States. Santa Monica, CA: RAND Corporation; 2001. Manning WG, Mullahy J. Estimating log models: to transform or not to transform? J Health Econ. 2001;20:461–494. Martinelli DR, Diasdado-De-La-Pena M. Safety Externalities of SUVs on Passenger Cars: An Analysis of the Peltzman Effect Using FARS Data. Morgantown, WV: Department of Civil and Environmental Engineering; 2008. McGlincy M. A Bayesian record linkage methodology for multiple imputation of missing links. 2004. Available at: http://www.amstat. org/sections/srms/proceedings/y2004/files/Jsm2004-000683.pdf. Accessed May 2012. Meyer JR, Gomez-Ibanez JA. Autos, Transit and Cities. Cambridge, MA: Harvard University Press; 1981. National Highway Traffic Safety Administration (NHTSA). Fatality Analysis Reporting System (FARS) Encyclopedia: Crashes. 2007. Available at: http://www-fars.nhtsa.dot.gov/Main/DidYouKnow. aspx. Accessed August 2011. National Safety Council (NSC). Estimating the Cost of Unintentional Injuries. 2011. Available at: http://www.nsc.org/news resources/ injury and death statistics/Pages/EstimatingtheCostsofUnintention alInjuries.aspx. Accessed August 2011. Singleton M, Qin H. Risk Factors for Death or Hospitalization Among Occupants of Passenger Motor Vehicles That Were Severely Damaged in Crashes in Kentucky, 2000–2001. Lexington, KY: Kentucky Injury Prevention & Research Center; 2003. Smith KM, Cummings P. Passenger seating position and the risk of passenger death or injury in traffic crashes. Accid Anal Prev. 2004;36:257–260. StataCorp. Stata Statistical Software: Release 11. College Station, TX: StataCorp LP; 2009. Teoh ER, Nolan JM. Is passenger vehicle incompatibility still a problem? Traffic Inj Prev. 2012;13:585–591. U.S. Census Bureau. 1997 Economic Census: Vehicle Inventory and Use Survey. 1999a. Available at: http://www.census.gov/prod/ec97/ 97tv-mn.pdf. Accessed August 2011. U.S. Census Bureau. 1997 Economic Census: Vehicle Inventory and Use Survey, Geographic Area Series. 1999b. Available at: http://www. census.gov/prod/ec97/97tv-us.pdf. Accessed August 2011. U.S. Census Bureau. Vehicle Inventory and Use: Products. 2002. Available at: http://www.census.gov/svsd/www/vius/products.html. Accessed August 2011. Wenzel TP, Ross M. The effects of vehicle model and driver behavior on risk. Accid Anal Prev. 2005;37:479–494. White M. The arms race on American roads: the effect of sport utility vehicles and pick-up trucks on traffic safety. J Law Econ. 2004;47:333–356.
