Effects of lane departure warning on police-reported crash rates. August Jessica B. Cicchino Insurance Institute for Highway Safety

Transcription

1 Effects of lane departure warning on police-reported crash rates August 2017 Jessica B. Cicchino Insurance Institute for Highway Safety

2 Abstract Objective: To evaluate the effects of lane departure warning (LDW) on single-vehicle, sideswipe, and head-on crashes. Method: Police-reported data for the relevant crash types were obtained from 25 U.S. states for the years Observed counts of crashes with fatalities, injuries, and of all severities for vehicles with LDW were compared with expected counts based on crash involvement rates for the same passenger vehicles without LDW, with exposure by vehicle series, model year, and lighting system standardized between groups. For relevant crashes of all severities and those with injuries, Poisson regression was used to estimate the benefits of LDW while also controlling for demographic variables; fatal crashes were too infrequent to be modeled. Results: Without accounting for driver demographics, vehicles with LDW had significantly lower involvement rates in crashes of all severities (18%), in those with injuries (24%), and in those with fatalities (86%). Adding controls for driver demographics in the Poisson regression reduced the estimated benefit of LDW only modestly in crashes of all severities (11%, p<0.05) and in crashes with injuries (21%, p<0.06). Conclusions: Lane departure warning is preventing the crash types it is designed to address, even after controlling for driver demographics. Results suggest that thousands of lives each year could be saved if every passenger vehicle in the United States were equipped with a lane departure warning system that performed like the study systems. Practical applications: Purchase of LDW should be encouraged, and, because drivers do not always keep the systems turned on, future efforts should focus on designing systems to encourage greater use and educating consumers about the benefits of using the systems. Keywords: Crash avoidance technology; collision warning; driver assistance; single-vehicle run-off-road; sideswipe; head-on collision 1

3 1. Introduction Crashes resulting from lane departures can be among the deadliest collisions. In 2015, nearly 13,000 people died in single-vehicle run-off-road, head-on, and sideswipe crashes where a passenger vehicle left the lane unintentionally (Insurance Institute for Highway Safety, 2017). Technology designed to help drivers avoid unintentional lane departures can prevent these crashes. Electronic stability control, which can prevent lane departures due to loss of control, has been shown to reduce fatal single-passengervehicle crash risk by 49 percent (Farmer, 2010). Other technologies aim to keep drivers from drifting out of lanes, either by providing warnings or steering corrections when they cross a lane line without signaling or by actively centering them within their lanes. Lane departure warning first became available in the United States on the Infiniti FX35 in model year 2005 and is becoming increasingly available on new passenger vehicles. In model year 2017, lane departure warning was available on 63 percent of new U.S. passenger vehicle series as standard (6%) or optional (57%) equipment (Highway Loss Data Institute [HLDI], 2016d). Jermakian (2011) estimated if every passenger vehicle in the United States were equipped with lane-keeping technology that prevented all relevant crashes, up to 3 percent of crashes of all severities, 5 percent of crashes with serious or moderate injuries, and 23 percent of fatal crashes involving passenger vehicles could be prevented. If all large trucks were equipped with these systems, they could potentially prevent 3 percent each of crashes of all severities and with serious or moderate injuries, and 6 percent of fatal crashes involving these vehicles (Jermakian, 2012). However, real-world evidence on the effectiveness of lane departure warning systems has been mixed. Hickman et al. (2015) studied the effect of lane departure warning on large trucks using carriercollected data from 14 U.S. fleets. Trucks equipped with lane departure warning had crash rates per mile traveled in single-vehicle run-off-road, head-on, and sideswipe crashes deemed relevant to the technology that were 48 percent lower than the rate for trucks without lane departure warning. Sternlund, Strandroth, Rizzi, Lie, and Tingvall (2017) used induced exposure to investigate the effectiveness of lane departure warning and prevention (i.e., systems that warn or provide steering 2

4 correction) on Volvo passenger cars in Sweden. The ratio of single-vehicle and head-on crash involvements with injuries, which were deemed relevant to lane departure warning and prevention, to rear-struck crash involvements with injuries, which were assumed to be unaffected by the technologies, was compared between Volvo vehicles with lane departure warning or prevention and the same vehicle models without the optional systems. Crashes relevant to lane departure warning and prevention were further limited to those occurring on roads with speed limits of kph and that were not covered with ice or snow to align with system limitations. Volvo s lane departure warning and prevention systems reduced relevant crashes by 53 percent. Positive benefits for lane departure warning have not yet been seen among passenger vehicles in the United States. HLDI (2012, 2016a, 2016b, 2016c, 2017a, 2017b) compared insurance claim rates for passenger vehicles insured in the United States with lane departure warning from a number of manufacturers to the same make, series, and model year vehicles without it, controlling for other collision avoidance systems on the vehicles and characteristics of the rated driver on the insurance policy. Lane departure warning always came bundled with a front crash prevention system (forward collision warning with or without autonomous emergency braking) on the vehicles from most manufacturers that HLDI examined, which made it difficult to tease out system-specific effects. On Mercedes-Benz and Mazda vehicles, where lane departure warning did not always come bundled with another system and effects could be isolated, no benefits for the system were found. It is possible that the crash types that can potentially be prevented by lane departure warning, which make up a relatively small proportion of the total crash population (Jermakian, 2011), appear too infrequently in HLDI s database to be detected by an overall claim rate analysis. Effects of the system on crash types affected by the system could not be examined in HLDI s work because their data do not contain detailed information on crash circumstances. The goal of the current study was to examine the effect of lane departure warning on single-vehicle, head-on, and sideswipe crashes relevant to the system in the United States using police-reported crash data, where crash types could be identified. Crash involvement rates per insured vehicle year in crash types relevant to lane departure warning of all 3

5 severities, with injuries, and with fatalities were compared between vehicles with lane departure warning and the same make, series, and model year vehicles without the optional system. 2. Method 2.1 Vehicles Vehicle series and model years included in the analyses are listed in Table 1. Study vehicles were limited to those where lane departure warning was offered as an optional feature and the presence or absence of the system on individual vehicles at the VIN (vehicle identification number) level was known. The study focused on vehicles with optional lane departure warning systems because these systems were rarely offered as standard equipment at the time of the study. VINs of General Motors (Buick, Cadillac, Chevrolet, and GMC), Mazda, Mercedes-Benz, and Volvo vehicles equipped with various collision avoidance technologies, including lane departure warning, were obtained from manufacturers. Collision avoidance systems on Honda and Subaru vehicles that were examined in this study were decodable from the VIN. Lane departure warning systems from Honda, Subaru, and Volvo warned drivers with audible beeps, while Mercedes-Benz vehicles warned with steering wheel vibrations. Mazda offered a choice of beeps or a sound mimicking driving over a rumble strip. Some General Motors vehicles offered only a beeping warning and others offered a choice of a beeps or directional seat vibrations. Systems were operational beginning at speeds ranging from mph. Collision avoidance systems other than lane departure warning were offered on many study vehicles. The presence of headlight and other nighttime visibility systems that could potentially impact lane departure crashes that occur in the dark were controlled for in analyses. These systems were: General Motors: high-intensity discharge headlights, curve-adaptive high-intensity discharge headlights, high-beam assist, cornering lamps Honda: LED headlights Mazda: curve-adaptive high-intensity discharge headlights, high-beam assist (packaged with forward collision warning) 4

6 Mercedes-Benz: high-intensity discharge headlights, curve-adaptive high-intensity discharge headlights, high-beam assist, cornering lamps, night vision system Volvo: curve-adaptive high-intensity discharge headlights When an advanced headlight type was optional, the base headlights were halogen for all manufacturers. Subaru offered high-intensity discharge headlights as an option, but their presence or absence on individual vehicles was unknown; thus, headlights were not accounted for in analyses involving Subaru vehicles. LED headlights were standard on some General Motors series and could not be controlled for separately because they were never optional equipment. Other optional collision avoidance systems were not expected to affect crashes relevant to lane departure warning and were not accounted for in analyses. These features were sometimes, but not always, packaged with lane departure warning, and included side-view assist, rear cross-traffic alert, rearview cameras, front and rear parking sensors, rear automatic emergency braking, and active parallel parking assistance on General Motors vehicles; passenger-side blind spot cameras and rear parking sensors on Honda vehicles; low-speed autonomous emergency braking, side-view assist, rear cross-traffic alert, and rearview cameras on Mazda vehicles; precharged brakes, side-view assist, front and rear parking sensors, rearview cameras, and active parallel parking assistance on Mercedes-Benz vehicles; side-view assist, rear cross-traffic alert, and rearview cameras on Subaru vehicles; and forward collision warning, side-view assist, rearview cameras, and rear parking sensors on Volvo vehicles. All study vehicles had electronic stability control as a standard feature. Lane departure warning always came packaged together with a forward collision warning system with or without autonomous emergency braking on study vehicles from all manufacturers but Mazda. On General Motors vehicles, it always came paired with either forward collision warning alone or with autonomous emergency braking; on Honda vehicles with forward collision warning alone; and on Mercedes-Benz, Subaru, and Volvo vehicles with forward collision warning with autonomous emergency braking. Although forward collision warning and autonomous emergency braking systems are effective in 5

7 Table 1. Study vehicle series and model years Make Series Model years Buick LaCrosse 4D 2WD Buick LaCrosse 4D 4WD Buick Regal 4D 2WD Buick Regal 4D 4WD Cadillac ATS 4D 2WD Cadillac ATS 4D 4WD Cadillac CTS 4D 2WD 2014 Cadillac CTS 4D 4WD 2014 Cadillac Escalade 4D 2WD 2015 Cadillac Escalade 4D 4WD 2015 Cadillac Escalade ESV 4D 2WD 2015 Cadillac Escalade ESV 4D 4WD 2015 Cadillac SRX 4D 2WD Cadillac SRX 4D 4WD Cadillac XTS 4D 2WD Cadillac XTS 4D 4WD Chevrolet Impala 4D Chevrolet Suburban 4D 2WD 2015 Chevrolet Suburban 4D 4WD 2015 Chevrolet Tahoe 4D 2WD 2015 Chevrolet Tahoe 4D 4WD 2015 GMC Yukon 4D 2WD 2015 GMC Yukon 4D 4WD 2015 GMC Yukon XL 4D 2WD 2015 GMC Yukon XL 4D 4WD 2015 Honda Accord 2D Honda Accord 4D Honda Accord Crosstour 4D 2WD Mazda 3 4D Mazda 3 5D Mazda 6 4D 2WD Mercedes-Benz E Class 4D 2WD 2010 Mercedes-Benz E Class 4D 4WD 2010 Mercedes-Benz S Class Hybrid 4D 2WD 2010 Mercedes-Benz S Class LWB 4D 2WD 2010 Mercedes-Benz S Class LWB 4D 4WD 2010 Subaru Forester 4D 4WD Subaru Impreza 4D 4WD Subaru Legacy 4D 4WD Subaru Outback SW 4WD Volvo S80 4D 2WD Volvo S80 4D 4WD Volvo V70 SW 2WD Volvo XC60 4D 2WD 2010 Volvo XC60 4D 4WD 2010 Volvo XC70 SW 4WD D=two-door, 4D=four-door, 5D=five door, 2WD=two-wheel drive, 4WD=four-wheel drive, SW=station wagon, LWB=long wheelbase 6

8 preventing front-to-rear striking crashes (Cicchino, 2017), they are not expected to affect the crash types relevant to lane departure warning that were examined in the current study. 2.2 Crash data Police-reported data for crashes involving study vehicles were extracted from 25 states that provided VINs of involved vehicles with their crash data so that study vehicles could be identified. The years during which crash data with VINs could be obtained varied by state. Data were available during from Nevada and Rhode Island; during from Delaware, Idaho, Louisiana, Missouri, South Dakota, and Tennessee; during from Indiana; during from Mississippi; during from Florida, Georgia, Iowa, Kansas, Michigan, Minnesota, New Jersey, Oklahoma, Pennsylvania, Texas, Utah, Washington, Wisconsin, and Wyoming; and during from Maryland. These 25 states made up half the population and 55 percent of crash deaths in the United States during VINs were missing for 7 percent of vehicles involved in crashes in these states during these years. Crashes considered relevant to lane departure warning included single-vehicle, head-on, and sideswipe crashes where no crash-involved vehicle was changing lanes, merging, passing, turning, or backing prior to the crash, and which occurred on roads with speed limits of 40 mph or greater that were not covered with snow or ice. Similar to Sternlund et al. (2017), crashes on roadways with lower speed limits were excluded in the current study because travel speeds would be more likely to fall outside of the systems minimum speed thresholds, and crashes on snowy and icy roads were excluded because systems will not provide reliable warnings if lane markings are obscured by snow or ice. Head-on and sideswipe crashes were defined primarily by the crash type coded by police and the number of vehicles involved in the crash. Sideswipe crashes included multivehicle crashes categorized as sideswipes by the police, and head-on collisions were limited to two-vehicle crashes coded as head-on collisions. To remove crashes that police may have erroneously coded as head-on or sideswipe, direction of travel prior to the crash was also considered when it was available. Head-on crashes excluded crashes 7

9 where vehicles were traveling in the same or perpendicular directions prior to the crash and sideswipes excluded two-vehicle crashes where vehicles were initially traveling on perpendicular paths. When direction of travel was available, it did not lead to the exclusion of many crashes (4% of potential head-on and sideswipe collisions where direction was known). Direction of travel was unavailable in four study states (Louisiana, Mississippi, Nevada, and Oklahoma). Parked vehicles were not included in vehicle counts. Single-vehicle crashes excluded crashes between a single vehicle and pedestrians or bicyclists. In the 21 states where direction of travel was available, it was missing in 6% of crash involvements of the study vehicles. Crash type was missing in 5% of study vehicle crash involvements in all 25 study states, surface condition (i.e., how snowy and icy roads were identified) was missing in 3%, and speed limit was missing in 13%. Speed limits were considered invalid if they were less than 5 mph or higher than the state s maximum speed limit in that year. Vehicles with missing values on crash type or speed limit variables were treated as if they were not involved in a lane-departure-warning-relevant crash. Missing values were equally distributed between vehicles with and without lane departure warning. 2.3 Exposure data Data on vehicle exposure and characteristics of the vehicle s garaging location (density of registered vehicles in the ZIP code where vehicle is garaged) and rated driver (age, gender, marital status, and insurance risk level) were obtained from HLDI. The HLDI database includes approximately 85 percent of insured U.S. passenger vehicles. Vehicle exposure was expressed in insured vehicle days, so that a vehicle insured for 6 months would have 183 days of exposure. For simplicity, exposure is presented in tables as insured vehicle years. Vehicle feature data, crash data, and insurance exposure data were merged by matching VINs within states; because VINs were matched within states, crashes that occurred in a different state than where a vehicle was insured were not captured. Vehicles that did not appear in the HLDI database or that were insured in a different state than where they crashed were excluded from both the numerator and denominator of crash rates. 8

10 2.4 Regression models Poisson regression was used to model crash involvement rates per insured vehicle year for vehicles with lane departure warning compared with vehicles without the system, controlling for a number of factors that affect crash risk. Two separate regression models were constructed for each manufacturer: one that modeled effects on lane-departure-warning-relevant crashes of all severities and one that examined effects of the system on relevant crashes with injuries. Six manufacturers were included, resulting in 12 separate models. All models controlled for rated driver age (15-34, 35-54, 55-69, 70+, unknown), gender, marital status, and insurance risk level (standard risk, nonstandard risk, unknown); state; calendar year; and registered vehicle density per square mile (0-499, 500+) in the ZIP code where the vehicle is garaged. These covariates were chosen for consistency with previous HLDI (2012, 2016a, 2016b, 2016c, 2017a, 2017b) analyses examining the effects of these same systems on insurance claim rates. A single variable capturing the vehicle series and model year was included to restrict estimates of effects within series and model year, preventing confounding of collision avoidance feature effects with other vehicle design changes that may occur between model years. Finally, binary variables indicating the presence or absence of collision avoidance features that could potentially impact lane-departure-warning-relevant crash types were additionally included. These included lane departure warning on all models, and on the Honda, General Motors, Mazda, Mercedes-Benz, and Volvo models, the headlight and other nighttime visibility systems named earlier. Models used a logarithmic link function. Suppose C i represents the number of crash involvements, E i represents exposure (i.e., insured vehicle days), and L i represents the presence or absence of lane departure warning for vehicle i. Assuming C i is a Poisson random variable with mean E i λ i, the statistical models were specified as log λ i = β 0 + β 1 (L i ) + β 2 (covariates). Exp(β 1 ) represented a rate ratio comparing crash involvement rates for vehicles with lane departure warning to vehicles without. Over-dispersion in the Poisson models was controlled for by estimating a scale parameter in SAS (i.e., PSCALE) and adjusting statistics accordingly. Pearson scale parameters ranged from 1.2 to 3.6. Negative 9

11 binomial models were considered, but were ultimately not used because they did not converge in every case. 2.5 Pooled estimates Effects from regression models were pooled across manufacturers using meta-analysis methods outlined below, which are similar to those used in evaluations of the effectiveness of other collision avoidance technologies (Cicchino, 2017, in press; Fildes et al., 2015). Random effects models were used because the implementation of lane departure warning varied across manufacturers. Rate ratios were log-transformed. A weight was assigned to each estimate as follows: 1! " = ) % " + ' ( where % " represents the estimate s variance and ' ) ( is a function of the Q statistic that represents the systematic variation among the estimated effects. The pooled effect for lane departure warning was calculated as follows: * = +,-. "/0. "/0! "! " * " where +,- is the exponential function, * " is the logarithm of each effect estimate,! " is each estimate s weight, and 1 is the total number of estimates. Thus, because weights were primarily based on the inverse variance of the respective estimates, estimates with high variance (those with large confidence intervals, typically because of a small number of crashes or little exposure) contributed less to the pooled effect than estimates with low variance (those with small confidence intervals). Ninety-five percent confidence intervals were computed using the following equation: 95% 67 = +,-. "/0. "/0! "! " * " ± "/0! " Effect estimates indicated that vehicles with lane departure warning significantly lower crash involvement rates than comparison vehicles when estimates and their 95% confidence intervals were less than 1. Percentage reductions were expressed as the rate ratio minus 1, multiplied by

12 2.6 Observed and expected crash counts Fatal crash involvements relevant to lane departure warning could not be modeled because of low crash counts and thus a second method was used to analyze the effects of lane departure warning on fatal crash involvements. Fatal crash involvement counts in crashes relevant to lane departure warning were compared between vehicles with lane departure warning and those without while standardizing the exposure distributions of vehicles in the two groups by series, model year, and type of headlight/nighttime visibility system. This method, similar to calculating a standardized mortality ratio, limited possible confounding between the groups due to some vehicle series being more likely to have lane departure warning than others, or due to advanced headlighting or nighttime visibility systems being fit more often on vehicles with lane departure warning than without. Unlike the Poisson regression models, these analyses did not account for driver demographic characteristics or insurance risk. Although the primary purpose of this analysis was to examine fatal crash involvements, identical analyses were also performed for relevant crash involvements of all severities and with injuries so that the similarity between the primary Poisson regression analyses and this secondary method could be assessed. Crash counts that would be expected for lane departure warning-equipped vehicles if they did not have the system were computed as the product of the crash involvement rate for the non-equipped version and the insured vehicle year count for the equipped version for each combination of vehicle series, model year, and type of headlight or nighttime visibility system. Vehicle series/model year combinations with lighting systems that were not offered on both lane-departure-warning-equipped and unequipped versions were dropped from the analyses. Because the presence or absence of headlight types were not known on Subaru vehicles, headlights were not considered when analyzing data from Subaru. For example, model year 2013 Cadillac XTS 2WD vehicles with high intensity discharge headlights and without lane departure warning were involved in 27 relevant crashes of all severities and were insured for 9,577 vehicle years, resulting in a crash involvement rate of 28.2 crashes per 10,000 insured vehicle years. Model year 2013 Cadillac XTS 2WD vehicles with high intensity discharge headlights and lane departure warning were insured for 11,804 vehicle years, and thus would be expected 11

13 to be involved in 33 relevant crashes if their crash involvement rate was identical to XTS 2WD models without lane departure warning ( *11,804=33). Curve-adaptive headlights on the model year 2013 Cadillac XTS 2WD were only offered with lane departure warning, and thus vehicles with curve-adaptive headlights were not included in the analysis. If multiple lighting combinations were offered on both vehicles with and without lane departure warning of a series/model year combination, then vehicles without lane departure warning were only compared to vehicles with lane departure warning that had the equivalent headlight type. For each manufacturer and for all study vehicles combined, rate ratios were computed by dividing the sum of the observed crash counts for vehicles with lane departure warning by the sum of expected counts. A rate ratio significantly less than 1 was taken as evidence that vehicles with lane departure warning have lower involvement rates in relevant crashes than the same vehicles without the system. Ninety-five percent confidence intervals on the rate ratios were computed as follows (Silcocks, 1994): lower limit = β (O, E + 1)/[1- β (O, E + 1)] and upper limit = β (O + 1, E)/[1- β (O + 1, E)], where O is the sum of observed crashes, E is the sum of expected crashes, and β p (x, y) is the p th percentile from the beta distribution with parameters x and y. 3. Results Study vehicles were involved in a total of 94,892 crashes of all severities, 22,384 crashes involving an injury to anyone in the crash, and 197 fatal crashes. Of these, 5,433 crashes of all severities (6% of total), 986 injury crashes (4%), and 39 fatal crashes (20%) were considered relevant to lane departure warning. Most relevant injury crashes resulted in moderate or minor injuries (87% had a maximum severity of B or C on the KABCO scale). Involvement rates in relevant crashes of all severities, with injuries, and with fatalities were lower among vehicles with lane departure warning than among vehicles without for each manufacturer (Table 2). For the crash involvements relevant to lane departure warning, Table 2 presents descriptive measures on raw crash count and involvement rates that do not control for various exposure factors (described below) that may impact crash risk. It should be kept in 12

14 Table 2. Crash involvement rates in crashes relevant to lane departure warning (LDW) of all severities, with injuries, and with fatalities among vehicles with and without the system (not controlling for factors that can affect crash risk) Manufacturer System Insured vehicle years All severities Injury crashes Fatal Rate Crashes (x100,000) Rate Crashes (x100,000) Rate Crashes (x100,000) General Motors LDW 307, No LDW 187, Honda LDW 286, No LDW 389, Mazda LDW 8, No LDW 147, Mercedes-Benz LDW 13, No LDW 110, Subaru LDW 77, No LDW 447, Volvo LDW 7, No LDW 138, All manufacturers LDW 701, No LDW 1,420, Total 2,121,487 5,

15 mind that raw crash counts associated with fewer insured vehicle years yield less reliable estimates of crash rates. Table 3 summarizes results of Poisson regression models examining the effects of lane departure warning controlling for state, calendar year, registered vehicle density of the vehicle garaging location, vehicle series/model year combination, vehicle headlight/visibility systems, and the age, gender, marital status, and insurance risk of the rated driver. On average, lane departure warning reduced crash involvement rates in relevant crashes of all severities significantly by 11 percent, and crash involvement rates in relevant injury crashes by 21 percent. The effect for injury crashes approached significance (p<0.06). Table 3. Adjusted rate ratios from Poisson regression models examining the effects of lane departure warning on crash involvement rates in crashes relevant to the system Rate ratio (95% confidence interval) Manufacturer All severities Injury crashes General Motors 0.87 (0.69, 1.08) 0.79 (0.50, 1.26) Honda 0.88 (0.76, 1.02) 0.81 (0.58, 1.13) Mazda 0.94 ( ) * Mercedes-Benz 0.65 (0.24, 1.78) 0.39 (0.07, 2.08) Subaru 0.98 (0.75, 1.30) 0.77 (0.36, 1.64) Volvo 0.90 (0.35, 2.33) 0.79 (0.04, 17.71) Combined effect 0.89 (0.80, 0.99) 0.79 (0.61, 1.01) * not enough crashes to model; Mazda is not included in combined effect To allow for examination of effects of lane departure warning on fatalities, which were too few to analyze in a regression, expected and observed crash counts in relevant crashes of all severities, with injuries, and with fatalities among vehicle series with and without lane departure warning were compared, with exposure by vehicle series, model year, and type of headlight or nighttime visibility system standardized between the groups (Table 4). Vehicles with lane departure warning were involved in 18 percent fewer relevant crashes of all severities, 24 percent fewer lane-departure-relevant crashes with injuries, and 86 percent fewer relevant fatal crashes than would be expected based on the crash involvement rates of the same series and model year vehicles without lane departure warning and with equivalent lighting systems. Reductions were significant at all severity levels. This pattern of results held 14

16 for lane departure warning-relevant crashes of all severities among each manufacturer, and for those with injuries or fatalities among 5 of the 6 manufacturers. 15

17 Table 4. Observed and expected crash involvements in crashes of all severities, with injuries, and with fatalities relevant to lane departure warning, with exposure standardized by vehicle series, model year, and headlight/nighttime visibility systems Manufacturer Crash severity Without lane departure warning With lane departure warning RR (95% confidence interval) Crashes Observed crashes Expected crashes General Motors All severities (0.74, 0.99) Injury (0.57, 1.20) Fatal Honda All severities 1, (0.71, 0.87) Injury (0.59, 0.95) Fatal (0.00, 0.63) Mazda All severities (0.46, 1.98) Injury Fatal Mercedes-Benz All severities (0.32, 1.17) Injury (0.02, 0.59) Fatal Subaru All severities (0.73, 1.15) Injury (0.41, 1.32) Fatal (0.01, 134.1) Volvo All severities (0.34, 2.31) Injury (0.01, 116.9) Fatal All manufacturers All severities 3,030 1,189 1, (0.76, 0.89) Injury (0.63, 0.91) Fatal (0.02, 0.63) 16

18 4. Discussion Results of these analyses provide evidence that lane departure warning is effective in reducing the single-vehicle, head-on, and sideswipe crashes it is designed to address among passenger vehicles in the United States. This is similar to the experiences of truck fleets in the United States and Volvo passenger cars in Sweden, although with smaller estimates of effect size on crashes of all severities and with injuries than seen in prior studies (Hickman et al., 2015; Sternlund et al., 2017). It was encouraging that a large effect for lane departure warning was seen for fatal crashes relevant to the system, although the effect was based on a small number of crashes, mainly from a single vehicle model (Honda Accord). Lane departure warning has been estimated to have the potential to prevent nearly 1 in 4 fatal crashes involving passenger vehicles (Jermakian, 2011). If the systems on other vehicles are as effective in preventing fatalities as they were among study vehicles, a substantial number of lives could be saved. However, it is somewhat perplexing that lane departure warning is proving effective given evidence that drivers do not always keep the system turned on. Reagan, Cicchino, Kerfoot, and Weast (2017) observed the activation status of lane departure warning systems that retained their prior setting at ignition on vehicles from eight manufacturers as they were serviced at 14 dealership service centers in the Washington, DC, metro area. Only 45 percent of the more than 500 vehicles observed had their lane departure warning systems turned on. Vehicles were observed from some of the manufacturers examined in the current study, including General Motors (51% of vehicles observed had lane departure warning turned on), Honda (33% turned on), Mazda (75% turned on), and Volvo (50% turned on); Volvo vehicles observed were mostly newer model years than in the current study, while General Motors, Honda, and Mazda vehicles observed included many of the models and model years analyzed here. Other studies have observed similarly low use rates of lane departure warning. Flannagan et al. (2016) tracked lane departure warning use across every ignition cycle over the course of a year among nearly 2,000 owners of General Motors vehicles nationwide through the vehicles OnStar systems, and found that the systems were turned on for about 50 percent of driving time. In an earlier pilot study using 17

19 the same methodology as Reagan et al. (2017) but limited to observations of Honda vehicles, Reagan and McCartt (2016) reported that lane departure warning was activated on 40 percent of Honda Accord models observed, which was the only Honda model evaluated in the current study. The activation status of lane departure warning in study vehicles at the time of the crash was unknown, and it is possible that use of lane departure warning was higher among drivers at greater risk of being involved in a lane departure crash and in situations where such crashes were most likely to occur. For example, multiple studies have found that observed and self-reported use of lane departure warning and prevention is significantly higher among men than women (Eichelberger & McCartt, 2016; Flannagan et al., 2016; Reagan et al., 2017), and men in the current study were overrepresented in crashes relevant to lane departure warning relative to their involvement in all police-reported crashes (41% of drivers in all police-reported crashes vs. 51% of drivers in lane-departure-warning-relevant crashes of all severities, 50% in relevant crashes with injuries, and 71% in relevant crashes with fatalities were male). Flannagan et al. (2016) reported that lane departure warning use increased when drivers were traveling at speeds greater than 55 mph compared with speeds of mph, which are speeds at which serious crashes relevant to lane departure warning are most probable. Drivers disuse of lane departure warning may explain why the system s effects on relevant crashes of all severities and with injuries were modest. If half of these lane departure warning systems were turned off, then the reduction in crash involvements was only half of what it could have been. In other words, if passenger vehicle drivers in the United States left the lane departure systems on all the time, then the crash reductions could be similar to the percent reported in prior studies (Hickman et al., 2015; Sternlund et al., 2017). One way to increase use of lane departure warning is by using vibrating rather than auditory warnings. Flannagan et al. (2016) and Reagan et al. (2017) found that lane departure warning use was markedly higher for systems with vibrating warnings compared with auditory warnings. For example, when the seat vibration setting was selected by the driver instead of beeps in Flannagan et al. s (2016) study, system usage doubled (from 32% to 64%) relative to drivers that only had the option of beeps. 18

20 Because drivers must respond appropriately to warnings for them be effective, another factor that may have limited the size of the system s benefit is the physical state of the driver. Cicchino and Zuby (2017) examined the physical state of 631 drivers involved in lane drift crashes in the National Motor Vehicle Crash Causation Survey (NMVCCS), in which in-depth crash investigations were performed on a national sample of vehicles involved in crashes where emergency medical services were called and a passenger vehicle was towed that occurred between 6 a.m. and midnight. Thirty-four percent of the drivers were physically incapacitated in some way, with 17 percent medically incapacitated and 17 percent asleep, and an additional 14 percent had a blood alcohol concentration of 0.08 percent or above, non-incapacitating medical issue, or another physical factor that led to the crash. These proportions were higher when considering only crashes involving fatal or serious injuries. Wiacek, Fikenscher, Forkenbrock, Mynatt, and Smith (2017) examined fatal lane-drift crashes in NMVCCS and found that fewer drivers were sleeping or otherwise incapacitated, although the physical states of many were unknown. Lane departure warnings cannot help medically incapacitated drivers, but they can possibly prevent sleeping drivers from lane drifts. Warnings have been shown to improve lane positioning and reduce lane departures of fatigued drivers in simulators (De Rosario, Soler, Tenas, Dominguis, & Echeverria, 2010; Gaspar et al., 2017; Kozak et al., 2006; Rimini-Doering, Altmueller, Ladstaetter, & Rossmeier, 2005). However, their ability to wake real-world sleeping drivers in enough time to prevent a crash is unknown. Lane departure warning was optional equipment on study vehicles. Vehicles with the systems could be substantially more expensive than vehicles without. Vehicles with lane departure warning also more often were equipped with other kinds of collision avoidance technologies than vehicles without. Regression analyses controlled for differences in lighting systems when known and for driver factors that may affect risk of crashes relevant to lane departure warning, but, nevertheless, drivers who choose to purchase these vehicles may differ even after controlling for these factors. Backing assistance systems, side-view assist systems, and forward collision warning/autonomous emergency braking systems, which 19

21 were not accounted for in regression analyses, are not designed to affect the crash types examined in this study. Smaller effect sizes on crashes of all severities and injury crashes were found in regression models controlling for driver factors than in the analyses of expected and observed crash rates, which indicates that driver factors did account for some of the differences between vehicles with and without lane departure warning. This suggests that the estimate for the effect on fatal crashes, which was not modeled, would also decrease if these factors were controlled. Although relevant crash types excluded crashes where drivers purposefully left their lanes prior to the crash by turning, changing lanes, merging, or passing, lane-drift crashes and crashes that occurred off of the roadway could not be isolated with the information provided in the crash data. Thus, crashes considered relevant likely included some that would not be affected by lane departure warning, including crashes involving loss of control, single-vehicle crashes where the vehicle did not depart the roadway, and head-on and sideswipe crashes where the crash partner and not the subject vehicle departed the lane. Since lane-drift crashes could not be isolated from these other crash types, it is likely that the effects estimated in this study are underestimates of the true effects of the system on relevant crashes. Crash counts were too small to compare lane departure warning systems with different characteristics, and systems can differ considerably by manufacturer in the warning modality, default state, and sensitivity settings (see Reagan et al., 2017 for examples). This variability could have contributed to the differing pattern of effect sizes by manufacturer seen in this study. Lane departure crashes are among the most deadly, and it has been estimated that lane departure warning and prevention systems have the potential to prevent up to 23 percent of all fatal passenger vehicle crashes reported to the police (Jermakian, 2011). The current study provides the first evidence in the United States that lane departure warning is preventing crashes among passenger vehicles. If every passenger vehicle in the United States were equipped with a lane departure warning system that performed like the study systems, thousands of lives each year could be saved. Further efforts should be made to design lane departure warning features in a manner that increases system usage and to make consumers aware of the safety benefits of the system to enhance effectiveness. 20

22 ACKNOWLEDGEMENTS The author would like to thank Adrian Lund, Chuck Farmer, and David Zuby of the Insurance Institute for Highway Safety, and Matt Moore of the Highway Loss Data Institute, for their input that improved the design and statistical methods used in this study. She is also grateful to Lisa Henke of the Highway Loss Data Institute, who provided exposure and vehicle feature data, and Jason Rubinoff and JoAnn Wells of the Insurance Institute for Highway Safety for their assistance in obtaining and formatting state crash data. Pennsylvania data used herein was supplied by the Pennsylvania Department of Transportation. The Pennsylvania Department of Transportation specifically disclaims responsibility for any analyses, interpretations, or conclusions drawn in this publication. This work was supported by the Insurance Institute for Highway Safety. REFERENCES Cicchino, J.B. (2017). Effectiveness of forward collision warning and autonomous emergency braking systems in reducing front-to-rear crash rates. Accident Analysis & Prevention, 99(Pt A), doi: /j.aap Cicchino, J.B. (in press). Effects of rearview cameras and rear parking sensors on police-reported backing crashes. Traffic Injury Prevention. doi: / Cicchino, J.B., & Zuby, D.S. (2017). Prevalence of driver physical factors leading to unintentional lane departure crashes. Traffic Injury Prevention, 18(5), doi: / De Rosario, H., Soler, A., Tenas, J., Dominguis, M., & Echeverria, M. (2010). Effective patterns of sleep warnings based on the physiologic and behavioural reaction of users. In European Conference on Human Centred Design for Intelligent Transport Systems (49-58). Berlin, Germany. Eichelberger, A.H., & McCartt, A.T. (2016). Toyota drivers' experiences with dynamic radar cruise control, pre-collision system, and lane-keeping assist. Journal of Safety Research, 56, doi: /j.jsr Farmer, C.M. (2010). Effects of electronic stability control on fatal crash risk. Arlington, VA: Insurance Institute for Highway Safety. Fildes, B., Keall, M., Bos, N., Lie, A., Page, Y., Pastor, C., Pennisi, L., Rizzi, M., Thomas, P., & Tingvall, C. (2015). Effectiveness of low speed autonomous emergency braking in real-world rear-end crashes. Accident Analysis & Prevention, 81, doi: /j.aap Flannagan, C., LeBlanc, D., Bogard, S., Kazutoshi, N., Narayanaswamy, P., Leslie, A., Kiefer, R., Marchione, M., Beck, C., & Lobes, K. (2016). Large-scale field test of forward collision alert and lane departure warning systems (Publication No. DOT HS ). Washington, DC: National Highway Traffic Safety Administration. Gaspar, J.G., Brown, T.L., Schwarz, C.W., Lee, J.D., Kang, J., & Higgins, J.S. (2017). Evaluating driver drowsiness countermeasures. Traffic Injury Prevention, 18(S1), S58-S63. doi: / Hickman, J.S., Guo, F., Camden, M.C., Hanowski, R.J., Medina, A., & Mabry, J.E. (2015). Efficacy of roll stability control and lane departure warning systems using carrier-collected data. Journal of Safety Research, 52, doi: /j.jsr Highway Loss Data Institute. (2012). Volvo collision avoidance feaures: initial results. HLDI Bulletin, 29(5). 21

23 Highway Loss Data Institute. (2016a) Honda Accord collision avoidance features: a 2016 update. HLDI Bulletin, 33(32). Highway Loss Data Institute. (2016b). Buick collision avoidance features: an update. HLDI Bulletin, 33(18). Highway Loss Data Institute. (2016c). Mercedes-Benz collision avoidance features: a 2016 update. HLDI Bulletin, 33(23). Highway Loss Data Institute. (2016d). Vehicle information report, HLDI facts and figures: vehicle fleet. Arlington, VA. Highway Loss Data Institute. (2017a) Subaru collision avoidance features. HLDI Bulletin, 34(10). Highway Loss Data Institute. (2017b). Mazda collision avoidance features: HLDI Bulletin, 34(8). Insurance Institute for Highway Safety. (2017). [Unpublished analysis of Fatality Analysis Reporting System]. Jermakian, J.S. (2011). Crash avoidance potential of four passenger vehicle technologies. Accident Analysis & Prevention, 43(3), doi: /j.aap Jermakian, J.S. (2012). Crash avoidance potential of four large truck technologies. Accident Analysis & Prevention, 49, doi: /j.aap Kozak, K., Pohl, J., Birk, W., Greenberg, J., Artz, B., Blommer, M., Cathey, L., & Curry, R. (2006). Evaluation of lane departure warnings for drowsy drivers. In Proceedings of the Human Factors and Ergonomics Society Annual Meeting ( ). Los Angeles, CA: Sage Publications. Reagan, I.J., Cicchino, J.B., Kerfoot, L.B., & Weast, R.A. (2017). Crash avoidance and driver assistance technologies are they used? Arlington, VA: Insurance Institute for Highway Safety. Reagan, I.J., & McCartt, A.T. (2016). Observed activation status of lane departure warning and forward collision warning of Honda vehicles at dealership service centers. Traffic Injury Prevention, 17(8), doi: / Rimini-Doering, M., Altmueller, T., Ladstaetter, U., & Rossmeier, M. (2005). Effects of lane departure warning on drowsy drivers performance and state in a simulator. In Proceedings of the Third International Driving Symposium on Human Factors in Driver Assessment, Training, and Vehicle Design (88-95). Silcocks, P. (1994). Estimating confidence limits on a standardised mortality ratio when the expected number is not error free. Journal of Epidemiology and Community Health, 48(3), doi: /jech Sternlund, S., Strandroth, J., Rizzi, M., Lie, A., & Tingvall, C. (2017). The effectiveness of lane departure warning systems-a reduction in real-world passenger car injury crashes. Traffic Injury Prevention, 18(2), doi: / Wiacek, C., Fikenscher, J., Forkenbrock, G., Mynatt, M., & Smith, P. (2017). Real-world analysis of fatal run-out-of-lane crashes using the National Motor Vehicle Crash Causation Survey to assess lane keeping technologies. In Proceedings of the 25th Enhanced Safety of Vehicles International Conference. Washington, DC: National Highway Traffic Safety Administration. 22