This article was downloaded by: [University of Aberdeen] On: 17 June 2013, At: 00:05 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

ROAD SAFETY FROM THE PERSPECTIVE OF DRIVER GENDER AND AGE AS RELATED TO THE INJURY CRASH FREQUENCY AND ROAD SCENARIO a

b

Francesca Russo Ph.D P.Eng , Salvatore Antonio Biancardo P.Eng & Gianluca Dell'Acqua Ph.D P.Eng

c

a

Dipartimento di Ingegneria Civile, Edile e Ambientale , University of Naples “Federico II” , Via Claudio 21, I-80125 , Naples , Italy Phone: +39 0817683372 Fax: +39 0817683372 b

Dipartimento di Ingegneria Civile, Edile e Ambientale , University of Naples “Federico II” , Via Claudio 21, I-80125 , Naples , Italy Phone: +39 0817683372 c

Dipartimento di Ingegneria Civile, Edile e Ambientale , University of Naples “Federico II” , Via Claudio 21, I-80125 , Naples , Italy Phone: +39 0817683934 Accepted author version posted online: 14 Jun 2013.

To cite this article: Francesca Russo Ph.D P.Eng , Salvatore Antonio Biancardo P.Eng & Gianluca Dell'Acqua Ph.D P.Eng (2013): ROAD SAFETY FROM THE PERSPECTIVE OF DRIVER GENDER AND AGE AS RELATED TO THE INJURY CRASH FREQUENCY AND ROAD SCENARIO, Traffic Injury Prevention, DOI:10.1080/15389588.2013.794943 To link to this article: http://dx.doi.org/10.1080/15389588.2013.794943

Disclaimer: This is a version of an unedited manuscript that has been accepted for publication. As a service to authors and researchers we are providing this version of the accepted manuscript (AM). Copyediting, typesetting, and review of the resulting proof will be undertaken on this manuscript before final publication of the Version of Record (VoR). During production and pre-press, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal relate to this version also.

PLEASE SCROLL DOWN FOR ARTICLE Full terms and conditions of use: http://www.tandfonline.com/page/terms-and-conditions 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. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae, and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand, or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material.

ACCEPTED MANUSCRIPT ROAD SAFETY FROM THE PERSPECTIVE OF DRIVER GENDER AND AGE AS RELATED TO THE INJURY CRASH FREQUENCY AND ROAD SCENARIO Francesca Russo Assistant Professor, Ph.D., P.Eng. Dipartimento di Ingegneria Civile, Edile e Ambientale

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

University of Naples “Federico II” Via Claudio 21, I-80125 Naples, Italy Phone: +39 0817683372 Fax: +39 0817683946 E-mail: [email protected] Salvatore Antonio Biancardo P.Eng. Dipartimento di Ingegneria Civile, Edile e Ambientale University of Naples “Federico II” Via Claudio 21, I-80125 Naples, Italy Phone: +39 0817683372 E-mail: [email protected] Gianluca Dell’Acqua Aggregate Professor, Ph.D., P.Eng. Dipartimento di Ingegneria Civile, Edile e Ambientale University of Naples “Federico II”

1

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT Via Claudio 21, I-80125 Naples, Italy Phone: +39 0817683934 E-mail: [email protected]

ROAD SAFETY FROM THE PERSPECTIVE OF DRIVER GENDER AND AGE

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

AS RELATED TO THE INJURY CRASH FREQUENCY AND ROAD SCENARIO Objective: The objective of the research is to develop safety performance functions (SPFs) on two-lane rural roads to predict the number of injury crashes per year per 108 vehicles/km on the road segment using a study on the influence of the human factors (gender/age/number-of-drivers) and road scenario (combination of infrastructure and environmental conditions found at the site at the time of the crash) on the effects of a crash by varying the dynamic. Countermeasures are suggested to reduce the injury crash rate and they can include different awareness campaigns and structural measures on the segments of road. Methods: An 8-year period was analyzed of which 5 years of crash information were used to calibrate and specify SPFs while the remaining 3 years were used to check the reliability of the equations. Before moving to the calibration phase, a technique to filter anomalous injury crash rates was adopted by using a method widely used in geotechnical engineering that is based on estimates of ranges of values that can be considered fluctuations of the “regular” measures compared with values estimated as “abnormal” for each homogeneous scenario. Because of over dispersion of crash data, generalized estimating equations and additional log linkage equation were adopted to calibrate SPFs. Akaike information criterion and Bayesian information criterion

2

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT were used for checking the reliability of the models. Results: Six SPFs were calibrated: for head-on/side collisions, it was built one equation for circular curves and one for tangent segments; for rear-end collisions, it was built one equation for daylight and one for the hours of darkness; for single-vehicle run-off-road crashes, it was built one equation for wet road surface conditions and one for dry road surface conditions. An original

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

numerical variable “SLEH” was designed to calibrate safety models reflecting the identified road “Surface”

(dry/wet),

“Light”

conditions

(day/night),

geometric

“Element”

(tangent

segment/circular curve) and “Human” factors (gender/age/number drivers) all together when the crash happened, as provided by related police reports. The validation procedure succeeded. It emerged that males and females are involved in crashes of varying degrees of frequency, depending on the driving scenario that presents itself and the gender of the other drivers involved in the crashes. Several different dangerous scenarios were identified: female only drivers on a dry road surface in daylight on tangent segments increased the risk for head-on/side collisions; male only drivers on a wet road surface in daylight on circular curves increased the risk for single vehicle crashes; and crashes involving both female and male drivers on a dry road surface in daylight on a circular curve increased the risk for head-on/side collisions. Conclusion: According to the current study, based on the network approach for the economic resources’ allocation and road safety strategies planning, calibration of injury crash rate prediction models for specific target collision type is important because of the range of harms that are caused by different collision types.

3

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT From these studies it is apparent that age/gender drivers considered together further refines how those factors contribute to crashes. Countermeasures (structural road interventions and/or safety awareness campaigns) can be planned to reduce the highest rate of injury crash for each gender of drivers and road scenarios: the awareness campaigns cannot be generalized or vague, but must be organized by age and gender, since the study has shown crash dynamics alter as these factors

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

change, with consideration for the varying psychological traits of the driver groups. Before and after safety evaluations can be used to check the safety benefits of improvements carried out on the roadways, within budget constraints for improvement or safety compliance investments for future operation. Supplemental materials are available for this article. Go to the publisher’s online edition of Traffic Injury Prevention to view the supplemental file.

Keywords: driver behavior, head-on collision, rear impact, side impact, countermeasures. INTRODUCTION

Females at the wheel, constant danger. This is a well known Italian proverb that labels females as poorly skilled drivers. The origin of this adage can be traced to the first licensed female driver in Italy, who was originally licensed in the United States (May, 1899). A few hours into her first drive on Italian roads, she caused a crash; the Italian myth of the female as an unskilled driver was born. The reality is that the female driver as a hazard is nothing more than an urban myth; females are often rewarded with favorable automobile insurance rates because they are a more reliable and safe driver than their male counterparts.

4

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT It is accepted in the scientific community that males and females differ in terms of their psychological attributes. For example, females tend to be more ambivalent, and complex when viewing and interacting with the world. Ambivalence, as it relates to the interaction between feelings and actions is not unequivocal, especially in the context of transportation choices. The objective of the research-work presented here is to develop safety performance functions

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

(SPFs) to predict the injury crash frequency over traffic exposure (injury crash rate) using a gender/age/number-of-drivers analysis of those involved in injury crashes on two-lane rural roads using a procedure to reduce the injury crash frequency to select acceptable countermeasures for road safety targets as valuable policy-making instruments. The study aims to integrate, innovate and bring new results in addition to those already obtained in a previous experimental analysis (Russo et al. 2012, Dell’Acqua et al. 2011). It’s recalled that the Highway Safety Manual (HSM 2000) published by the American Association of State Highway and Transportation Officials (AASHTO) gives SPFs to estimate the number of crashes over a specific roadway over a specific time period, and safety countermeasures for that highway. The HSM provides predictive models for rural two-lane highways, giving estimates for total crashes. Because the SPF equations in the HSM were developed on the basis of data from a subset of states, HSM recommends that local agencies either (a) develop SPFs for their local conditions or (b) use a calibration procedure to adjust the HSM SPFs to reflect local conditions. In the paper presented here SPFs have been calibrated, specified and validated for three main crash types (head-on/side collisions, rear-end collisions, single-vehicle run-off-road crash) as identified on the investigated road network taking into account specific combinations of road

5

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT geometric-function-environmental circumstances and human factors (gender/age/number drivers) which influence the consequences of a crash type. If a critical crash type (HSM 2000) on the road network is known, the critical geometric and environmental conditions can be defined by fixing the value of the variables in SPFs, so as to predict the value of the injury crash rate. Solutions (structural road interventions and/or safety awareness campaigns) can be planned to

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

reduce the highest rate of injury crash for each gender of drivers: before and after safety evaluations can be used to check the safety benefits of improvements carried out on the roadways, within budget constraints for improvement or safety compliance investments for future operation. LITERATURE REVIEW

Legislative changes and public media campaigns to prevent impaired driving are often cited as explanations for the reduction in the rate of impaired crashes over the past 25 years in most of the industrialized world. Other factors may have contributed to these reductions, such as changes in the age and sex distribution of the driver population (Macdonald 2003). The gender analysis is considered a potential moderator of the relationships between selfreported driver aggression and demographic variables, together with general and driving-related risk factors (Wickens et al. 2012). It was observed in the scientific literature that crash rate differences between male and female drivers is connected to their dissimilar psycho-parameters: some experimental analyses have indicated that male crashes rates are significantly higher than female crash rates for the lack of attention and impatience among male drivers (Al-Balbissi 2003).

6

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT However, the research on this wide topic is fragmented and a multidimensional perspective is lacking. Some researchers are trying to investigate multiple relationships between driving styles and personality traits and how these interactions can change by age and gender (Poò and Ledesma 2013). The crash risk for different age and gender categories of road users was largely studied.

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

Disaggregated models, for example, were suggested for analyzing the differences in risk between various age-gender categories, together with their trend over time. Van den Bossche et al. (2007) studied the road risk expressed as the number of fatalities per capita, by means of state space models. It was found on the collected data that road risk changes over the age groups according to a U-shaped curve, and that men generally have higher risk than women. Further, the risk decreases over time, but not at the same rate for all age-gender groups. The highest yearly reduction in risk was found for the oldest and youngest road users. Bener et al. (2008) examined the effect of the driver gender on road traffic crashes by a questionnaire for participants including socio-demographic information, driver behavior questionnaire, driver skill inventory and seatbelt use. Analyses showed that females report a higher number of violations, and lapses, while male drivers report higher crash rates as compared to female drivers. No significant association was found between male and female drivers in terms of errors. Chen et al. (2010) performed generalized linear models to examine crash trends over time by severity of driver injury, adjusting for age, gender, rurality of residence, and socioeconomic status in New South Wales (NSW), Australia. Overall, there has been a significant decline in young driver crashes in NSW over the last decade. Regardless of injury severity, males' risk of

7

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT crash has reduced more than female young drivers, but drivers aged 17 continue to be at higher risk. The severity for crash type was also investigated in the scientific literature. Separate male and female multinomial logit models of injury severity (with possible outcomes of no injury, injury, and fatality) were, for example, estimated for young (ages 16 to 24), middle-aged (ages 25 to

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

64), and older (ages 65 and older) drivers for single-vehicle crashes involving passenger cars (Islam et al. 2006). The crash risk resulting from an overtaking maneuvers was also explored for different drivers by gender and age. Farah (2011) observed, for example, that younger male drivers apparently seem to be more involved in risky situations than the other groups due to shorter overtaking critical gaps, higher speeds, and closer following from the front vehicle. Female and older drivers were found to have larger critical gaps and lower driving speeds, which explain their longer overtaking time durations. Furthermore there is large body of research that analyzes the effect of passenger age and gender on young driver fatal crash risk (Hough et al. 2008), and also the weight of age and experience of drivers in driving mileage (McCartt et al. 2009). Nevertheless, the study on the influence of the human factors and road scenario (combination of infrastructure and environmental conditions found at the site at the time of the crash) on the effects of a crash by varying the dynamic is lacking. In this paper Safety Performance Functions are suggested for three main crash types (head-on/side collisions, rear-end collisions, singlevehicle run-off-road crash) highlighting the role played by the above factors in the prediction of the crash rate for selecting specific countermeasures.

8

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT DATA COLLECTION The crash data involved in the safety per gender investigation of two-lane rural roads in Southern Italy took into account almost 3,700 km of network; one-half were used to calibrate SPFs, attempting to predict for each homogeneous road segment the injury crash rate as the number of injury crashes per year per km per 108 vehicles, and the other half to check the effectiveness of

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

the injury crash rate prediction models in the validation procedure using Residuals analysis. The homogeneous road segment is characterized by a constant curvature change rate (CCR in gon/km) defined as the sum of the absolute values of angular changes in horizontal alignment divided by total length of road section. Crash data were made available by the Administration of the Province of Salerno. An 8-year period (2003-2010) was analyzed, of which 5 years (2003 – 2007) of crash information were used to calibrate and specify SPFs while the remaining 3 years (2008 – 2010) were used to check the reliability of the equations subdivided according to driver gender and injury crash type, as will be shown below, by comparing the predicted injury crash rates with the actual injury crash rates not included in the calibration phase of the models. A total of 303 road segments with the same curvature indicator were used, and 197 of these were used in the first phase while 106 were used in the second phase. Table A1 shows the descriptive statistics of the features observed on the analyzed roadway segments. The number of the investigated crashes on more than 3,700 km of studied road network is equal to 2,242 between 2003 and 2010, of which 1,597 were injury (2,592 injuries and 67 deaths, with mean value for the crash rate of 12.03, maximum value of 120.81, and minimum value of 0.42) while 645 caused property damage only

9

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT By analyzing the crash database (2003-2010 years) and police reports, three main crash types were identified involving at most three vehicles: 52% head-on/side collisions on 1,740 km of the total road network analyzed, 27% of single-vehicle crashes (vehicle exits the roadway and either strikes a fixed object, overturns, or collides with vehicles parked on the road side or entering the roadway) on 1,190 km, and 21% of rear-end collisions on 774 km.

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

The actual Italian population of licensed drivers on the investigated road network is 446,278 drivers; the overall distribution of background traffic is 48% males drives vs. 52% female drivers. During the eight year study period 1,945 vehicles were involved in injury crashes. The highest proportion of crashes involved cars (74.45%), of which 73% were driven by males. Two wheeled vehicles were involved in 20.62% of all accidents during the study period, of which 96% were driven by males. The remaining 4.94% of all accidents involved trucks, tractors, buses, and agricultural vehicles, of which 96% were driven by males. Table 1 shows in detail the summary of the studied crash counts from 2003 to 2010 using crash type and gender analysis. In particular, the category “crashes with male and female drivers” (Table 1) refers to three identified conditions: a) one male and one female drivers; b) two females and one male drivers; c) two men and one female drivers. After a careful analysis of crash data wide variety of factors appear to influence or be associated with the crash dynamic: gender and age of the driver, mean lane width, horizontal curvature indicator, mean speed. Table A2 incorporates gender into the analysis, while varying the crash type and road scenario for which is shown the corresponding percentage of injury crashes (n) for each crash type, and relative minimum, mean, and maximum value of injury crash rate (ICR) during the 8-year period

10

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT (2003-2010) for homogeneous segments equal to the number of injury crashes per year per 108 vehicles/km (crash frequency over traffic exposure). According to the results in Table A2, it can be observed that during the study period the mean crash rate values for head-on/side collisions are higher than the estimated crashes for the remaining crash types, and head-on/side collisions are the more frequently dangerous critical crash type for all drivers, regardless of gender.

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

It was observed that the maximum value for injury crash rate using gender analysis is the headon/side collisions crash type during the specific period both when female only drivers are involved in dry road surface + daylight + tangent scenario (max crash rate of 120.81) and when female and male drivers are involved in dry road surface + daylight + curve (max crash rate of 78.29) crashes. However, the maximum value for the injury crash rate when male only drivers are involved in crashes is reached for single-vehicle run-off-road crashes on wet road surface + daylight + curve (max crash rate of 118.05). The minimum value for injury crash rate when female only drivers are involved in crashes for the relevant period is recorded for rear-end collision type in wet road surface + daylight + tangent scenario (crash rate of 0.53), while when female and male drivers are involved, and when male only drivers are involved, the scenario is single-vehicle run-off-road crash with wet road surface + night + curve scenario in the first case (crash rate of 0.53), and dry road surface + night + curve in the second case (crash rate of 0.42). DATA ANALYSIS AND RESULTS Calibration Phase A total length of the analyzed network equal to 1,846.74 km was used in this calibration phase involving 5 years of the crash database (2003 – 2007), for a total of 829 injury crashes, which led

11

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT to 28 deaths (64% head-on/side collisions, 32% single-vehicle run-off-road crashes and 4% rearend collisions) and 1,174 injuries (61% head-on/side collisions, 13% single-vehicle run-off-road crashes and 27% rear-end collisions). In particular, it was observed of the total number of deaths recorded over the 5 years data used for calibrating the models, 57% were caused by crashes involving male only drivers, 4% female only drivers, and 39% with male and female drivers

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

involved. In addition, of the total number of injuries recorded over the 5 years, 52% were caused by crashes where male only drivers were involved, 4% with female only drivers, and 44% with male and female drivers involved. Figure 1 shows some of the plotted risk type density diagrams for the head-on/side collisions that allowed us to thoroughly study the safety gender issue and to identify significant factors in the crash phenomenon to calibrate accurate and detailed SPFs on circular curves and tangent segments. The diagrams shown in the Figures 1, A1 and A2 (Figures A1 and A2 refer to the risk type density diagrams for rear-end collisions and single-vehicle runoff-road crashes, respectively) are divided according to the number and gender of drivers involved in the injury crashes; they show by some shades’ bands different ranges of injury crash rates in connection with particular combinations of road scenario (see first column of the Table A2), mean speed and lane width, driver age. Because of overdispersion of crash data, generalized estimating equations and additional log linkage equation were adopted to calibrate six SPFs. Different SPFs varying the crash type better reflected the influence of gender and infrastructure/environmental circumstances on the crash dynamics, well functional and well-designed to maximize the reliability of the predicted injury crash rate during the validation phase, as follows:  for head-on/side collisions two subsystems (curves and tangents) were defined and it was built

12

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT one equation for circular curves and one for tangent segments;  for single-vehicle run-off-road crashes two subsystems (dry and wet road surfaces) were defined and it was built one equation for wet road surface conditions and one for dry road surface conditions;  for rear-end collisions two subsystems (daylight and hours of darkness) were defined and it

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

was built one equation for daylight and one equation for darkness. To reflect specific combinations of road geometric-function-environmental circumstances and human factors (gender/age/number drivers) which influence the consequences of a crash type, the authors designated an original numerical variable “SLEH” to denote various crash settings. This variable SLEH was used to calibrate safety models reflecting the identified road “Surface” (dry/wet), “Light” conditions (day/night), geometric “Element” (tangent segment/circular curve) and “Human” factors (gender/age/number drivers) all together when the crash happened, as provided by related police reports. The values associated with the SLEH variable, presented as one of the explanatory variables of the safety performance functions, are considered while varying the crash type, gender/age/number of drivers, scenario and infrastructure features. This process permits interested parties to examine the risk associated with how specific road/environmental/human factors can influence the consequences of an investigated crash type on the studied road segment, and assign a code to the SLEH variable. This code is obtained from a careful analysis of the risk density diagrams. Analyzing all possible maps deriving from a plot of all the possible combinations of the cited features, the value for the SLEH variable is obtained, as shown below and in Table

2. For each substrate (specific combination of

infrastructure/environmental circumstances) with a specific gender/age/number of drivers’ set,

13

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT the SLEH code is higher for the high crash frequency investigated substrates in terms of collected injury and death, and lower for those with lower injury crash frequency. The values of the “SLEH” were set for each scenario as follows:  counts of the road scenarios number (nj) for each subsystem j-th (6 subsystems were defined)  calculating for each substrate i-th of the subsystem j-th the mean injury crash rate (ICRi,j) over

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

the period of the study  identification for each subsystem of the substrate with highest mean injury crash rate (ICRmax,j)  SLEH code assignment for each i-th (CSLEH_i) scenario in the j-th subsystem with code= nj for the i-th substrate in the j-th subsystem with highest mean injury crash rate, and for all those remaining, a value CSLEH_i = Eq. (1) is as follows:

 ICRi , j CSLEH _ i  ni    ICRmax, j 

  

(1)

Before moving to the calibration phase, a technique to filter anomalous injury crash rates for each subsystem was adopted, calculated using the Vivatrat method (1979) widely used in geotechnical engineering. The method is based on estimates of ranges of values that can be considered fluctuations of the “regular” measures compared with values estimated as "abnormal" for each homogeneous substrate (in our study, 8 scenarios) within each homogeneous subsystem (in our study, 6 subsystems identified in Table 2). Akaike information criterion (AIC) and Bayesian information criterion (BIC) were used to check the reliability of the models. In addition

14

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT to these criteria, other goodness-of-fit measures for GEE models were used as shown in Table 3 where Eqs from 2 to 7 illustrate SPFs and the explanatory variables meaning for predicting Y that is the number of injury crashes per year per 108vehicles/km on the homogeneous road segment, according to the explanation above on gender analysis. AIC is a measure of the goodness of fit of a statistical model and it describes the tradeoff between the accuracy and the complexity of a

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

model. It includes a penalty that discourages over fitting; thus, the best models are those with lowest AIC value together with the smallest amount of the explanatory variables: AIC outperforms BIC when unstable conditions appear such as small/large sample with high standard deviation levels. Therefore, AIC is preferred to BIC for the type of data to be analyzed. The safety equations in Table 3 show how all injury crash rate prediction models are statistically suitable with comparable AIC and Pearson dispersion values. However, the best goodness of fit towards to the specific subsystems refer to three safety equations: a) Eq. (3) for the head-on/side collisions on the tangent segments; b) Eq. (4) for the single-vehicle run-off-road crashes on dry road surfaces; and c) Eq. (7) for the rear-end collisions during hours of darkness. Each explicative variable inserted into the models has a different effect on predicting the injury crash rate, as may be observed from the value of the coefficient and the algebraic sign. The only variable always negatively correlated to the Y dependent variable is the mean width of the travel lanes. The equations in Table 3 are to apply independently one by one on all investigated road homogeneous segments which are made up of n tangent segments and n circular curves. Subsequently, different solutions can be considered: a) to sum all results derived from the application of the performed SPFs for different crash types, geometric and environmental conditions for studying the global safety state on the whole road network; b) to analyze the

15

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT global effect for specific crash types on the whole road network; c) to study the crash injury rates for specific crash types under definite “Light” (day/night) and road “Surface” (dry/wet) conditions, gender/age/number of drivers, mean speed and roadway width, curvature indicator, by using definite equations in Table 3. Validation Phase

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

A total length of the analyzed network equal to 1,853.26 km was used in the validation procedure to test the reliability of the SPFs covering three years of crash data from 2008 to 2010 as follows:  46% are head-on/side collisions with  675 injuries of which 62% involved crashes with male only drivers, 2% with female only drivers, 36% with female+male drivers, and  22 deaths of which 60% involve crashes with male only drivers, 5% with female only drivers, 35% with female+male drivers;  37% for single-vehicle run-off-road crashes with  370 injuries of which 71% involve crashes with male only drivers, 14% female only drivers, 15% with female+male drivers, and  16 deaths of which 81% involved crashes with male only drivers; 19% involved crashes where female+male drivers are involved;  17% for rear-end collisions with  231 injuries of which 43% involved crashes with male only drivers, 2% female only drivers, 55% involved crashes where female+male drivers are involved, and  1 death for male only driver crashes.

16

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT In particular it was noted that for head-on/side collisions 45% occurred on circular curves and 55% on tangent segments, while for single-vehicle run-off-road crashes 65% occurred on dry road surface and 35% on wet road surface, and for rear-end collisions 77% took place during daylight and 23% during hours of darkness. The validation procedure consisted of the residuals analysis (residual is the difference between

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

predicted and real value of injurious crash rate) and it allows to recognize regions where the predictive models over- or underestimate the injury crash rates. Table A3 shows the values returned from the validation procedure for each of the six models, where µ is the mean value and σ is the standard deviation of the residuals, MAD is the mean absolute deviation equal to the sum of the absolute values of the residuals divided by the number of road segments, MSE is the mean squared error. The diagram of cumulative squared residuals plotted on the basis of AADT graphically show the absence of vertical jumps that would indicate real injury crash rates dissimilar to the predicted values using SPFs: in this case more investigations to decide whether to use these observations or not are necessary. In particular, it was noted for suggested SPFs all residuals fall within the range [µ - 2σ; µ + 2σ] where µ is the mean value and σ is the standard deviation of the residuals: safety prediction models are statistically significant as confirmed by the low value of mean value of the residual of the injury crash rate (min. value equal to 0.08 for Equ.2, max value equal to 2.50 for Equ.7) and MAD (min. value equal to 2.37 for Equ.7, max value equal to 7.54 for Equ.6), and the absence of jumps in the diagram of cumulated squared residuals. For the predictive model with the highest mean residual in Table A3 (rear-end collisions during hours of darkness), two little jumps in the cumulative squared

17

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT residuals’distribution of the injury crash rates were observed: a maximum jump of 0.65 for 2,500 vehicles per day and 5,500 vehicles per day (see Figure A3). DISCUSSION SECTION In this paper six SPFs are suggested for predicting the number of injury crashes per year per 10 8 vehicles/km on the road segment i by plotting several crash risk density diagrams using a driver

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

gender analysis, driver number analysis and crash type with all possible combinations of influencing factors as “Surface” conditions (dry/wet), “Light” conditions (day/night), geometric “Element” (tangent segment/circular curve), mean speed, road width and curvature indicator. A validation procedure was carried out on basis of the residuals analysis (residual is the difference between predicted and real value of injurious crash rate) that confirmed the effectiveness of the crash database’s partition in six subsystems (see Table 1) to better reflect the influence of gender and infrastructure/environmental circumstances on the crash dynamics and, consequently, on the predicted injury crash rate. The safety predictive models are an essential tool and useful aid in assessing the best solution of road maintenance work, the reliability of the hypothesized road safety targets and they may be used as valuable policy instruments. The number of possible strategies is equal to the number of potential combination of the explanatory variables in the model on which it can really work to improve road safety conditions and of which it can recognize the effect induced on safety by varying the value. Countermeasures for reducing critical injury rates for crash type can be suggested including awareness campaigns and real road structural operations. Awareness campaigns include direct actions on the population, such as training, pamphlets, and safety events tailored to the specific driver safety issues identified by age and gender. Countermeasures directed toward structural

18

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT operations act on geometric and functional features of the road and concern the permanent structure. This operation type depends only on proper maintenance work being carried out over time and all drivers will obtain long-term benefits from structural work on roads; we can have routine maintenance works such as reconstruction of the upper layer of the road surface, rearrangement of vertical and horizontal road markings, clean edges and side gutters, pruning,

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

etc and emergency maintenance works by varying geometric/infrastructural/functional features on an investigated road. The appropriate strategy could be a combination of countermeasures linked to the cost-benefit rations from economic and social perspectives. It can be analyzed, for example, the safety conditions when male only drivers are involved in a single-vehicle run-off-road crash in wet road surface+daylight+curve conditions. It applies Eq. (5) and substitutes the MW variable for the value of total real road width examined, the SLEH variable with the value shown in the first column “Code SLEH” of Table 3, which is equal in this case to 10, and finally the MS variable with the value of the mean speed on the geometric element analyzed. In this way, we obtain the mean value of the injury crash rate on the examined road element for the defined conditions and countermeasures can be developed if the value is high compared with an expected average value on the studied road network. Specific awareness campaigns for male drivers (for the specific conditions previously shown) can be implemented if these geometric, environmental, functional circumstances are reflected in a higher frequency of crashes. For a critical scenario that involves females only, then specific strategies can be defined; such as training or posters targeted at specific age groups or genders, which emphasize the factors on the road that influence crashes and encourage a safe, prudent and attentive driving style.

19

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT Disciplinary actions for unsafe driving can vary among driver groups, depending on whether a certain critical crash type with high frequency and severity is most frequently caused by male or female only drivers. Safe driving campaigns can be tailored to the specific needs of target gender and age groups depending on the features of the identified scenarios after applying Eq. (2) to Eq. (7) (if the most critical crash type is not known) to a specific homogeneous segment and

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

identifying the most associated conditions that influence crashes (road surface/light/geometric element, mean speed and road width, CI) on which the awareness campaign is to be based and the specific variables that must have a greater impact during the campaign for safer driving in order to illustrate where the greatest risk comes from. If the critical crash type on the road network is known, the critical geometric and environmental conditions can be defined by fixing the SLEH variable in accordance with the value in Table 2. Knowing the highest value of mean speed and the lowest value of road width, it can be applied the corresponding equation for these conditions from among those shown from Eq. (2) to Eq. (7). Since the predicted value of the highest injury crash rate is known, it can be now decided to propose structural intervention, such as shown below, unlike the previous case where safe driving campaigns were preferred. For example in the same previous conditions (single-vehicle run-off-road crashes on wet road surface+daylight+curve conditions) it can be lowered the predicted injury crash rate with variations which can be carried out on the roadway width, speed (Torbic et al. 2010), and in the final analysis, on the road surface type (Dell’Acqua et al.2012). By using a detailed approach to conduct assessed before and after safety evaluations, it is possible to maximize the safety benefits of improvements on an existing highway within specific budget constraints.

20

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT CONCLUSIONS AND FUTURE DEVELOPMENT The objectives of this research are to calibrate safety performance functions (SPFs) to predict the injury crash frequency over traffic exposure (injury crash rate) using a gender/age/number-ofdrivers analysis of those involved in injury crashes on two-lane rural roads in the Southern Italy and crash type analysis. Acceptable countermeasures have been suggested for road safety targets

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

as structural operations and actions on the community with targeted awareness campaigns for safer driving that highlight major risk factors by gender and age. Crash data over an 8-year period (2003-2010) were used, of which 5 years (2003 – 2007) of crash information were used to calibrate and specify the SPFs while the remaining 3 years (2008 – 2010) were used to check the reliability of the equations. In particular 3,700 km of road network were examined with 2,242 crashes from 2003 to 2010 of which 1,597 were injury (2,601 injuries and 67 deaths, with a mean injury crash rate of 12.03, a max value of 120.81, and a min value of 0.42) and 645 with only property damage of which 49% were head-on/side collisions, 30% were single-vehicle crashes, and 21% were rear-end collisions over 773.60 km. It emerges that males and females are involved in crashes of varying degrees of severity and frequency, depending on the driving scenario that presents itself; as well as the average speed on the homogeneous segment, road curvature and the gender of the other drivers involved in the crashes. All these conditions were taken into account for the specific calibration and validation of SPFs that reflect these specific conditions. A network approach was used to create six injury crash rate prediction models by implementing the GEE method with a negative binomial distribution and additional log linkage equation. Goodness-of-fit measurements were used to determine the reliability of the results: one SPF on the circular curves and another one on tangent

21

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT segments for head-on/side collisions; one SPF on a dry road surface and another one on a wet road surface for single-vehicle crashes; one SPF during daylight hours and another one for the hours of darkness for rear-end collisions. Gender It was observed that a scenario for female only drivers with the highest frequency of crashes

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

occurs on dry road surface in daylight on tangent segments in the case of head-on/side collisions. For male only drivers the risk is a wet road surface in daylight on the circular curves for single vehicle crashes. If both female and male drivers are involved, the risk is a dry road surface in daylight on a circular curve for the head-on/side collisions. Our results show, according to Tables 1 and 2, females in DDC, DNC, WDC and WNC scenarios are more cautious with single vehicle crashes circumstances than males as well as for WDT and DNT scenarios with rear-end collisions and WDC scenario with head-on/side collisions. Age How

driver

age

influences

a

crash,

is

also

analyzed

by

gender,

driver

and

environmental/infrastructure conditions. (Figures 1-A1-A2 and Table 2): a) for female only drivers involved in a crash, the age with highest risk is 30-50 years range in DDT scenario for head-on/side collisions, and DNT scenario for single-vehicle crashes and for rear-end collisions; b) for male drivers only the age with highest risk is 20 - 30 years range in WDC scenario for single-vehicle crashes, 20 - 70 years range in DDC for head-on/side collisions, and 25 – 80 years for rear-end collisions in DDC and WNC (until 50 years) scenarios; c) when both female and male drivers are involved, the risk is for 20 – 60 years range for head-on/side collisions in WDC

22

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT scenario, 18 – 40 years range for single-vehicle crashes in WDT scenario, and 20 – 45 years range for rear-end collisions in DNT scenario. Final Observations According to the current study, based on the network approach for the economic resources’ allocation and road safety strategies planning, calibration of injury crash rate prediction models

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

for specific target collision type is important because of the range of harms that are caused by different collision types. Therefore, each homogeneous road segment can be checked for incidence and frequency of each crash dynamic, as a statistically independent event, on the basis of the crash data collected for the study period. Then, it can eventually add up the results on each road segment or on the whole road network during an examined period to gain an overview of road safety condition and to plan detailed countermeasures for road scenarios with more frequently expected crash with positive consequences on the less frequent crash scenarios. The countermeasures can include different awareness campaigns promoting safer and more comfortable driver behavior, and structural measures on the segments of road (varying the width of the lanes, curvature indicators, operating speed) which will benefit both the more cautious users and make the infrastructure/environmental circumstances less dangerous. The awareness campaigns cannot be generalized or vague, but must be organized by age and gender, since our study has shown crash dynamics alter as these factors change, with consideration for the varying psychological traits of the driver groups. REFERENCES Al-Balbissi AH. Role of Gender in Road Accidents. Traffic Inj Prev. 2009; 4(1): 64-73.

23

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT Bener A, Crundall D. Role of gender and driver behaviour in road traffic crashes. International Journal of Crashworthiness. 2008; 13(3): 331-336. Chen HI, Senserrick T, Chang HY, Ivers RQ, et al. Road Crash Trends for Young Drivers in New South Wales, Australia, from 1997 to 2007. Traffic Inj Prev. 2010; 11(1): 8-15. Dell’Acqua G., De Luca M., Russo F. Procedure for Making Paving Decisions with Cluster and

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

Multicriteria Analysis. Transp Res Rec. 2012; 2282: 57-66. Dell’Acqua G, Russo F. Safety performance functions for low-volume roads. The Baltic Journal of Road and Bridge Engineering. 2011; 6(4): 225-234. Farah H. Age and Gender Differences in Overtaking Maneuvers on Two-Lane Rural Highways. Transp Res Rec. 2011; 2248: 30-36. Highway Safety Manual. American Association of State Highway and Transportation Officials (AASHTO), Washington D.C., 2010. Hough JA, Cao X, Handy SL. Exploring Travel Behavior of Elderly Females in Rural and Small Urban North Dakota An Ecological Modeling Approach. Transp Res Rec. 2008; 2082: 125131. Islam S, Mannering F. Driver aging and its effect on male and female single-vehicle accident injuries: Some additional evidence. In J Safety Res. 2006; 37: 267-276. Macdonald S. The influence of the Age and Sex Distributions of Drivers on the Reduction of Impaired Crashes: Ontario, 1974-1999. Traffic Inj Prev. 2003; 4(1): 33-37. McCartt AT, Mayhew DR, Braitman KA, Ferguson SA, Simpson HM. Effects of Age and Experience on Young Driver Crashes: Review of Recent Literature. Traffic Inj Prev. 2009; 10(3): 209-219.

24

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT Poó F. M., Ledesma R. D. A study on the relationship between personality and driving style. Traffic Inj Prev. 2013; in press. Russo F., Mauro R., Dell’Acqua G. Rural highway design consistency evaluation model. Procedia - Social and Behavioral Sciences 2012; 53: 953 – 961. Torbic DJ, Hutton JM, Bokenkroger CD, et al. Guidance on Design and Application of Rumble

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

Strips. Transp Res Rec. 2010; 2149: 59-69. Van den Bossche FAM, Wets G, Brijs T. Analysis of Road Risk by Age and Gender Category: Time Series Approach. Transp Res Rec. 2007; 2019: 7-14. Vivatrat V. Cone Penetration in clays. Ph.D. Thesis MIT Cambridge, Mass, USA; 1979. Wickens CM, Mann RE, Stoduto G, Butters JE, Ialomiteanu A, Smart RG. Does gender moderate the relationship between driver aggression and its risk factors?. Accid Anal Prev. 2012; 45: 10-18.

Table 1 - Summary of the Crash Count from 2003 to 2010 ITEM Injury crashes Number of injuries Number of deaths Number of crashes with property damage only crashes with male only drivers [%] crashes with female only drivers [%] crashes with male and female drivers [%]

ITEM

injury crashes

Total number Percentage for each crash type [%]

HEAD-ON/SIDE COLLISION 826 1,456 41

STATISTICS FOR CRASH TYPE SINGLE-VEHICLE RUN-OFF-ROAD CRASH 439 540 24

332 596 2

282

228

135

58

67

47

1

15

2

41

18

51

STATISTICS FOR GENDER, CRASH TYPE AND SUBSYSTEM HEAD-ON/SIDE SINGLE-VEHICLE RUN-OFF-ROAD COLLISION CRASH subsystems IV I SUBSYSTEM: II SUBSYSTEM: III SUBSYSTEM: SUBSYSTEM: Crashes on Crashes on Crashes on dry Crashes on wet circular curves tangent segments road surface road surface 379 447 285 154 45.88

REAR-END COLLISION

54.12

64.92

25

35.08

REAR-END COLLISION

235

VI SUBSYSTEM: Crashes during night hours 97

70.78

29.22

V SUBSYSTEM: Crashes during daylight

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT crashes with male only drivers

crashes with male and female drivers

62

68

66

65

60

49

624

424

252

120

180

107

13

12

15

4

1

1

4

3

16

12

3

11

15

12

54

18

8

11

2

-

-

-

-

-

34

29

18

23

37

40

130

360

53

43

220

70

6

8

5

-

-

1

Table 2 - SLEH Code according to Gender, Crash Type and Scenario

SUBSYSTEM II: CRASHES ON TANGENT SEGMENTS

SUBSYSTEM I: CRASHES ON CIRCULAR CURVES

Subsystem

HEAD-ON / SIDE COLLISION ONLY MALE ONLY FEMALE DRIVERS DRIVERS Code Code Age Age SLEH SLEH Range Range

Substrate (road scenario) Dry road surface + Daylight + Curve (DDC) Wet road surface + Daylight + Curve (WDC) Wet road surface + Night + Curve (WNC) Dry road surface + Night + Curve (DNC) Dry road surface + Night + Tangent segment (DNT) Wet road surface + Night + Tangent segment (WNT) Wet road surface + Daylight + Tangent segment (WDT) Dry road surface + Daylight + Tangent segment (DDT)

20-70

6.3

-

-

22-80

4.78

30-57

3.28

20-54

4.14

-

-

21-55

3.81

-

-

23-56

5.55

18-45

3.57

21-51

4.2

-

-

29-61

3.15

20-80/2746

4.01

18-65

4.39

32-43

11

MALE+FEMALE DRIVERS Age Range 19-80 man 21-53female 20-60 male 20-55female 24-54 male 19-47female 19-43 male 18-46female 18-65 male 18-70female 35-38 male 20-40female 23-52 male 24-43female 18-80 male 22-56female

Code SLEH 6.07 9 2.67 1.57 6 5.72 5.93 4.97

SUBSYSTEMI: CRASHES ON WET ROAD SURFACE

SINGLE-VEHICLE RUN-OFF-ROAD CRASH

SUBSYS TEM II: CRASHE S ON DRY ROAD SURFAC E

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

crashes with female only drivers

Percentage within each subsystem [%] Number of injuries Number of deaths Percentage within each subsystem [%] Number of injuries Number of deaths Percentage within each subsystem [%] Number of injuries Number of deaths

Wet road surface + Daylight + Curve (WDC) Wet road surface + Daylight + Tangent segment (WDT) Wet road surface + Night + Curve (WNC) Wet road surface + Night + Tangent segment (WNT) Dry road surface + Daylight + Curve (DDC) Dry road surface + Daylight + Tangent segment (DDT) Dry road surface + Night

20-30

10

23-55

3.19

39-70

5.67

-

-

19-40

7.5

21-30

2.19

26-46

3.52

38

4.86

18-70

5.09

21-80

1.99

19-51

4.73

18-75

7.63

30-55

1.46

26

23-40 male 28-39 female 18-40 male 26-40 female 20-45 male 18-31 female

3.55 7.35 3.98

-

-

-

-

32-46 male

2.56

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT + Curve (DNC) Dry road surface + Night + Tangent segment (DNT)

23-60

7.92

30-52

9

31-45 female 19-51 male 19-71 female

3.16

30-80

7

-

-

19-68

3.48

-

-

28-49

5.63

30-50

1.01

25-50

9

-

-

19-58

6.03

30-48

1.54

25-40

2.90

-

-

37-63

1.70

-

-

19-45 male 19-35 female

3.22

24-78 male 19-60 female 24-60 male 21-45 female 30-50 male 20-50 female 25-45 male 20-45 female 35-55 male 22-53 female 38-67 male 30-50 female

2.61 4.08 2.12 5.53 1.41 3.02

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

SUBSYSTEM II: NIGHT CRASHES

SUBSYSTEM I: DAYTIME CRASHES

REAR-END COLLISION Dry road surface + Daylight + Curve (DDC)/ Wet road surface + Daylight + Curve (WDC) Dry road surface + Daylight + Tangent segment (DDT) Wet road surface + Daylight + Tangent segment (WDT) Wet road surface + Night + Curve (WNC) Dry road surface + Night + Tangent segment (DNT) Dry road surface + Night + Curve (DNC) Wet road surface + Night + Tangent segment (WNT)

Table 3 - SPFs from the Driver Gender Perspective for three Injury Crash types

Crash type

Subsystem

head-on/side collisions

crashes on circular curves crashes on tangent segments

SPF

Y  e

0.096MW  0.5CI  0.18SLEH  0.02MS 

Y  e

0.077MW  0.31SLEH  0.015MS 

27

Eq. (2)

Eq. (3)

Std. Dev. of injury crash rates’ sample

AIC

BIC

Pearson dispersion

11.12

7.13

-987.98

0.78

7.81

6.71

112.79

0.09

ACCEPTED MANUSCRIPT

ACCEPTED MANUSCRIPT singlevehicle runoff-road crashes

rear-end collisions

SLEH MW MS

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

CI

crashes on dry road surfaces crashes on wet road surfaces crashes during daylight crashes during hours of darkness

Y  e Y  e

0.23MW  0.24SLEH  0.45MS 

0.07 MW  0.17 SLEH  0027 MS 

Y  e

0.15 MW  0.27 SLEH  0.04MS 

Y  e

0.32 MW  0.17 SLEH  0.05MS 

Eq. (4)

7.38

6.08

312.22

0.07

Eq. (5)

12.76

10.01

115.08

7.1∙10

Eq. (6)

8.18

6.85

221.43

0.51

Eq. (7)

4.30

5

99.20

0.24

-5

This variable assumes the value in the “SLEH Code” column of Table 2 depending on the type of crash to be studied, gender and age of drivers on a road segment to be analyzed (circular curve or tangent segment), on a dry or wet road surface, and in varying light conditions (day or night) mean width of the travel lanes plus shoulders in meters mean speed at each analyzed homogeneous roadway segment in km/h measurement of the curvature change rate for each homogeneous road segment. Value between 1 and 3 is defined in order to describe CI: 1 for low road horizontal curvature (CCR < 50gon/km), between 1 and 2 for medium curvature (50gon/km ≤ CCR ≤ 300gon/km), and between 2 and 3 for high curvature (CCR > 300gon/km)

a) 2 males drivers

b) 3 males drivers

28

ACCEPTED MANUSCRIPT

Downloaded by [University of Aberdeen] at 00:05 17 June 2013

ACCEPTED MANUSCRIPT

c) 2 females drivers

d) female + male drivers

e) female+ male drivers

f) 2 females drivers

Figure 1 - Summary of the Studied Injurious Crashes from 2003 to 2007 – Headon/Side collisions

29

ACCEPTED MANUSCRIPT