EXPLORING FACTORS CONTRIBUTING TO CRASH SEVERITY OF MOTORCYCLES AT SUBURBAN ROADS

Transcription

1 EXPLORING FACTORS CONTRIBUTING TO CRASH SEVERITY OF MOTORCYCLES AT SUBURBAN ROADS Amin Ariannezhad Graduate Student of Transportation Engineering and Planning Department of Civil Engineering Sharif University of Technology Address: Azadi Avenue, P.O. Box -, Tehran, Iran Phone: +- Hesamoddin Razi-Ardakani (corresponding author) PhD Candidate of Transportation Engineering and Planning Department of Civil Engineering Sharif University of Technology Address: Azadi Avenue, P.O. Box -, Tehran, Iran Phone: +- Mohammad Kermanshah Professor of Transportation Engineering and Planning Department of Civil Engineering Sharif University of Technology Address: Azadi Avenue, P.O. Box -, Tehran, Iran Phone: +-() Word Count: (Words), 0 (Tables) = Resubmitted on November st, 0 Submitted for Presentation at the rd Annual Meeting of the Transportation Research Board

2 Ariannezhad, Razi-Ardakani and Kermanshah 0 0 ABSTRACT Recently, severity of motorcycle crashes has been considered by different researchers. One of the main reasons of such considerations is greater vulnerability of these users as compared to other vehicle drivers. In recent years, the number of fatalities caused by motorcycle crashes, particularly in suburban roads, has become a concerning issue in Iran since motorcyclists accounted for percent of all traffic-related deaths in Iran in 0. This study is the first research in Iran in which the factors associated with crash severity are identified. In this paper the crash data from police-reported motorcycle crashes in suburban roads of Iran during 00 and 00 is used to estimate ordered logit model to identify the factors affecting severity of suburban motorcycle crashes. In order to better understand the effect of variables on crash severity, the value of pseudo-elasticity has also been calculated for all the variables in the model. Results mainly show that factors such as occurring the crash at weekends, during winter and fall, during the dawn, in foggy and clear weather, in non-administrative areas, rider age above 0, rider without proper license, lack of helmet, motorcycle at-fault, speeding and overtaking as well as collision with bus, heavy vehicle, pedestrian and single vehicle crashes increase the crash severity of motorcyclist. Besides, head on crashes, fatigue and sleepiness, inexcusable haste, violating the rules, road imperfection and curved level roads cause increased in motorcycles crash severity. Finally, as a result of this research, several policy recommendations are presented for improving motorcycles safety at suburban roads. Key words: Motorcycle Crash Severity, Ordered Logit Model, Pseudo-Elasticity

3 Ariannezhad, Razi-Ardakani and Kermanshah INTRODUCTION Investigating crash severity and casualties has become a universal problem for both developed and developing countries. According to the World Health Organization report,. million people are killed in driving accidents every year and approximately 0 to 0 million people are also injured (). It is important to point out that % of the mentioned fatalities take place in countries with low income while they possess only % of the total vehicles around the world (). Among developing countries, Iran as a low income country shows an extremely high crash fatality rate. On average, 000 Iranian citizens die every year due to driving accidents, for which the overall direct and indirect damages account for % of the total Gross Domestic Products of the country, while on average this number is about to % in developing countries (). The average crash fatality rate in 00 is. persons per 00,000 populations which is increased up to. person in Iran, rating this country as the top in fatality rate out of countries. United States, Japan, Switzerland and China have a fatality rate of.,.,. and 0., respectively (). According to the Iranian Legal Medicine Organization, people passed away in motorcycle accidents for the year of 0, which accounts for % of the total crash fatalities in that year. Out of the mentioned number, people were male and the remaining people were female. The age group of to reserved the highest rank in motorcycle fatality rate by.% while people older than 0 or between 0- years old with % rate were allocated in the subsequent ranks. Moreover,.0%,.% and 0.% of motorcycle fatal crashes happened in suburban, urban and other areas, respectively, while the highest number of accidents took place in cities compared with all other roads. The above statistics show that motorcycle accidents which happen at suburban areas are more dangerous, and result in higher fatalities as compared to other areas crashes. Thus, given data are divided into urban and suburban categories and due to the greater importance of the former category, the modeling of crash severity is done based on suburban accidents. Due to the considerable number of motorcycle accidents in Iran specifically in suburban roads areas as compared to other countries, the necessity for safety studies and improving the current situation in this developing country is imperative. While there have been several studies regarding accidents and their respective causes, no credible investigation has been previously done about modeling motorcycle crash severity in Iran. LITERATURE REVIEW Since the number of motorcycle accidents and fatalities has increased in the recent years, many researchers have focused on reducing the number and severity of such accidents. To this end, many studies have been performed to investigate the factors that impact the motorcyclists injuries in motorcycle accidents. Quddus et al. () have identified the effective factors responsible for the intensity of injuries to motorcycles and damages to vehicles in motorcycle accidents and categorized them in and levels for injuries and damages, respectively. Based on their results, the following factors increase crash severity: motorcyclists not possessing Singaporean citizenship, higher engine capacity, collision with pedestrians and fixed vehicles, driving in early hours of the day, having pillion passenger and

4 Ariannezhad, Razi-Ardakani and Kermanshah rider at-fault. Yannis et al. () have also investigated the combined effects of age and engine capacity on crash severity based on the database gathered in Greece in 00. Shankar and Mannering () studied a specific aspect of accidents by investigating crash severity in single-vehicle accidents and categorizing them into levels. For this purpose, they used the Washington State crash database over a period of years and by utilizing descriptive variables such as environmental specifications, road conditions, vehicle and driver characteristics identified the factors that affect crash severity. Additionally, in a similar study Savolainen and Mannering () determined the factors associated with crash severity in single and two vehicle motorcycle crashes to be the aging of motorcyclists, alcohol consumption, not wearing safety helmet, over speeding and darkness for both cases. On the other hand, some researchers have studies accidents in particular places. For instance, Pai and Saleh (0) studied motorcycle crash severity at T-junctions in three different ordered probit models. Moreover, Pai () showed in 00 that not giving the right of way to motorcycles by other vehicles at T-junctions led to higher angle crash severity at these intersections. Haque and Chin () studied the effects of road conditions, environmental specifications, traffic conditions, type of maneuver, human factors and driver characteristics on the vulnerability of motorcyclists at angle collisions at intersections. Also Schneider and Savolainen () developed multinomial logit models for modeling single and multi-vehicle crash severity for accidents at both intersections and non-intersections. Their results demonstrate that high severity crashes are followed by high speeds and alcohol consumption, as well as collisions with fixed object, angle and head-on collisions. Use of safety helmet significantly reduces accident severity in all cases. In some studies, researchers figured out the reasons behind fatal accidents by categorizing the crash severity levels into fatal and non-fatal motorcycle crashes. Chang and Yeh () and Shaheed and Dissanayake () used this type of data for modeling motorcycles crash severity. Furthermore, Jackson and Mello () studied the difference in injury patters among various age groups for motorcycle accidents and came to the conclusion that older motorcyclists are more vulnerable to injuries than younger ones. In other studies such as Chimba et al. () and Rifaat et al. (), researchers compared the strength of different models as well as predicting factors that affect crash severity. Rifaat et al. developed three severity models in order to determine the parameters that increase motorcycle crash severity which are logit model, Heterogeneous choice model and partially constrained generalized ordered logit model. Based on their obtained results, the outcome of the three models were similar to each other, proving that they all benefit from high prediction strength. The following factors increased motorcycle crash severity in all three models: over speeding, collision with truck, angle collisions, alcohol consumption, peak hours and darkness. Chimba et al. developed multinomial logit and probit models to predict the severity of motorcycle accidents, both of which had similar experimental coefficients in terms of value and sign. However, they assessed the multinomial probit model as the more efficient one. According to their research, factors such as increased number of lanes, alcohol and drug consumption, high posted speed limit, curved and grade locations, driving with insufficient daylight, collision with truck and buses and multi-vehicle accidents increased crash severity.

5 Ariannezhad, Razi-Ardakani and Kermanshah Finally, some researchers investigated crash severity in urban (), suburban (0) and urban and suburban () locations. Lapparent () studied the severity of motorcycle accidents in large and dense French urban areas during the year of 00 using Bayesian method based on the Multinomial-Dirichlet model. On the other hand, Geedipally et al. () utilized two distinct models to contrast crash severity in urban and rural areas of Texas, US. They concluded that factors such as alcohol usage, horizontal and vertical curves increase crash severity in both urban and rural models, while use of safety helmets and appropriate lighting conditions decrease crash severity. Furthermore, motorcyclists aged above increase and single-vehicle collisions, angle collisions, intersections and highways decrease crash severity in rural areas. In their findings about fatality risks in motorcycle collisions with trees and fixed objects in rural places, Bambach and Grzebieta (0) figured out that over speeding, motorcyclists above years old, darkness and curved roads lead to fatal crashes in motorcycle collisions with trees. Safety helmets also significantly decline fatal accidents. In all of the above studies, discrete choice models such as multinomial logit model (,,,, ), ordered probit (, 0, ), nested logit (), binary logit (,,, 0) and ordered logit () have been utilized. In this study, ordered logit model is used to modeling motorcycle crash severity in suburban areas. Unlike other countries, no significant effort has been taken to study the factors associated with motorcycle crash severity in Iran. Consequently, the results of this research will have profound effects on determining factors leading to increased crash severity in suburban areas of Iran and could result in policy makings which improve safety of motorcyclists in these areas. METHODOLOGY In case that the dependent variable is ordinal, using ordered response models are more common in comparison to multinomial ones, due to the sequential nature of alternatives (choices). Estimating fewer coefficients and accordingly more clear (easier) interpretation of results is the advantage of these models. Because of the mentioned nature of injury levels, ordered models are reported as most prevailing methods for analyzing the accidents severity (). Assumption of these models is that a continuous unobserved (invisible) variable named latent variable is related to a discrete observable one (). Main equation of ordered models has the following general form: Y x * n n n Where * Y n is the unobserved latent variable measuring the injury severity of nth collision, x n is explanatory variable (driver and crash characteristics), is the Vector of coefficients associated with explanatory variables and n is random error term. The relationship between Yn j if and only if In which j * Y n and Y n (observed severity category) can be expressed as follows: * j Y n and j are thresholds for latent variable that define the range of accident severity and j is crash severity categories. Based on severity levels defined for this study and given the equation () and (), Y n Can be calculated as: () ()

6 Ariannezhad, Razi-Ardakani and Kermanshah () Where it is assumed that 0 and levels, j, for a given accident is calculated as: Pr( Y j X ) F( X ) F( X ) 0 n n j n j n. Thus the probability of each severity Where F represents the cumulative distribution of term. If F has logistic distribution with zero mean and variance of (), the ordered Logit model is achieved (). Ordered logit model is developed through maximum likelihood method which can be formulated according to the equation (): N J log L m log[ F( X ) F( X )] n j 0 nj j n j n Where mnj if Yn j, otherwise. Maximization is done with respect to the constraints 0 and J. An important assumption associated with the ordered Logit model is parallel regression assumption that is also called proportional odds or parallel lines. Based on this assumption each explanatory variable has the same impact (influence) on different severity category. Goodness of fit statistic,, is a common measure of overall model fit. is defined for the models that are developed through maximum likelihood method (e.g. Logit model), as follows: LL( ) () LL(0) Where LL(β) is the log likelihood at the point of convergence with estimated coefficients β and LL(0) is the initial log likelihood with all coefficients equal to zero. The statistic varies between zero and one, where values close to one represent better prediction of model. Since the values of variable coefficients do not fully explain the effect of variables on the probability of observing each alternative, an elasticity parameter is usually utilized to explain the results. For binary variables, pseudo-elasticity parameter is calculated (). Since the above variables can only take zero and one, a % change in their values is not defined and as a result, the change in the probability of observance of each alternative is determined based on the change in the value of each variable from zero to one. The pseudo-elasticity parameter is calculated as follows: P P [ ] [ 0] j j xkj Pj xkj Ex () kj P [ x 0] kj kj To further improve the accuracy of this parameter, an average value of the pseudoelasticity parameter over a range of observations should be calculated (). ()

7 Ariannezhad, Razi-Ardakani and Kermanshah DATA The suburban accidents which occurred in 00 and 00 in Iran and involved at least one motorcycle have been investigated in this research. This information is obtained through traffic police database which has been prepared through collecting completed forms of K concerning registered accidents. K forms are completed at the scene of the accident by traffic experts who are responsible for investigating accidents and hence, record all information regarding all occurred accidents in Iran in addition to their causes. These data are constituted of main parts consisting properties of the vehicles involved in accident, drivers characteristics, road characteristics, information regarding the circumstances in which the accident took place and quality of driving and instruction rules observance. Information prepared through police databases was obtained in two separate databases in Microsoft access software, one database including information regarding the accident and the other containing drivers characteristics involved in the accident. In order to prepare the required data, the two mentioned databases were combined in MATLAB software based on serial numbers and series as the indicator of each observation. Obtained data containing each accident characteristics accompanied by driver or drivers characteristics involved are considered as final prepared data. It should be noted that data concerning traffic conditions such as traffic volume and speed were not available and hence, their relevant variables are not included in the model. Among accidents which were available during 00 and 00, those in which at least one motorcycle is involved were chosen for further researches. It is noteworthy that motorcycle crashes in which more than two vehicles were involved have been omitted from the accident database. Then, according to the research s goal, urban accidents were omitted from the database and thus, accidents which occurred outside cities were remained. In present study, urban crash was defined as the collisions occurring in any area within the boundaries of a city. Crashes occurring outside of cities were classified as suburban crashes Incomplete observations and cases, in which part of the required information was missing, were omitted and thus crashes were ultimately examined. Table presents descriptive statistics for the final data of suburban motorcycle crashes. The model dependent variable which is accidents severity is defined in three levels of damage only, injury and fatality. Table indicates that damage only accidents, injury and fatality constitute.%,.% and.% of all existing data, respectively. It should be mentioned that severity of each accident is defined as severity of injury to the most injured person in the accident. In this regard, fatal accident is defined as the one in which at least one person loses his/her life. Injury accident is defined as the one in which at least one person is injured (with no fatality) and damage accident is defined as the one in which nobody is injured and is only accompanied by property damage. The factors under study embody a wide range of variables such as: Conditionalenvironmental characteristics (crash time, crash season, light condition, weather condition), motorcycle specifications (age, type of license, education, helmet use, fault status, rider s fault), crash characteristics (other vehicle type involved in crash, crash type, collision type, crash position, human factor) and road conditions (pavement condition, area usage, road deficiency, obstacle view, road alignment).

8 Ariannezhad, Razi-Ardakani and Kermanshah TABLE Sample Characteristics Variable Percentage Dependent variable Damage only.% Injury.% Fatal.% Conditional and environmental characteristics Accident day of week Weekend.% Weekday 0.% Seasonal condition Spring 0.% Summer.% Fall.0% Winter.% Light condition Daytime.% Nighttime.% Dawn 0.% Dusk.% Weather condition Clear.% Rainy.% Foggy 0.% Snowy.0% Others 0.% Motorcyclist characteristics rider s gender Male.% Female 0.% Rider s age Up to 0.% -.% 0 above.% License type Motorcycle driving license.% Other driving licenses.% No driving license.% Driver s education Illiterate.% Below high school diploma.0% High school diploma.% University degree.% Other degrees.% Helmet use Wearing.% Not-wearing.%

9 Ariannezhad, Razi-Ardakani and Kermanshah Variable Percentage Fault status Rider is at-fault.% Rider is not at-fault.% Rider s offense Disregarding the right of way.% Deviation to right and left.% Going through no entry routes.% Unsafe longitudinal distance.0% Unsafe lateral distance.% Inattention to front.% Insufficient driving skill 0.% Loss of control.0% Speeding.% Overtaking and lane deviation.% driving in the opposite direction.% Sudden redirection.% Crash characteristics Collision partner Passenger car.% pickup.% Minibus or van.% Bus 0.% Heavy vehicle.% Motorcycle.% bicycle.% Motorcycle alone.% Crash Type Hit parked vehicle.% Hit fixed object.% Single vehicle.% With animal 0.% With pedestrian.% With other vehicle.% Collision Type Sideswipe.% Angle.% Head on.% Rear end.% Other types.% Collision location In road.% Other locations.% Human factor Violating the rules.% Fatigue and sleepiness.% Inexcusable haste.% Unfamiliar with roads.%

10 Ariannezhad, Razi-Ardakani and Kermanshah Variable Percentage Alcohol or drug consumption 0.% Other human factors.% No human reason for crash.% Roadway characteristics Surface condition Dry.% Not dry.% Area usage Industrial.% Administrative 0.% Non-administrative.0% Other areas.% Road imperfection With imperfection 0.% Without imperfection.% Vision obstruction Rider/driver has vision obstruction.% Without vision obstruction.% Road alignment Gradient and curved.% Level and curved.0% Gradient and straight.% Level and straight.% License variable in this research includes motorcyclists with motor license, other licenses and no license at all. As shown in Table,.% of motorcyclists did not have any license which includes a significantly large number of motorcyclists. Unfortunately, riding motorcycles without permit is a problem which endangers the safety of Iranian motorcyclists significantly. Rider offence variable contains misconducts committed by the motorcyclists before the accident and are reported by police in accident forms. This variable is part of the motorcyclist characteristic variable and includes elements. Human factor variable consists of human factors that led to motorcycle accidents as specified by police, and could come from either parties involved in the accident. Such factors include violating the driving laws, fatigue and sleepiness, unexcused haste, unfamiliarity with the road, disregarding the right of way and alcohol/drug consumption. Area usage variable includes industrial, commercial-administrative, nonadministrative and other zones. It should be pointed out that non-administrative zones consist of residential, agricultural and recreational zones in suburban areas. These zones have similar coefficients in the final model and thus, have been categorized under the non-administrative zone. The following road imperfections which have been identified by police have been categorized and utilized in the model as road imperfection variable: sign incompleteness, non-standard longitudinal and lateral slope, non-standard curve, lack of shoulder and side parking, non-even asphalt, non-standard or lack of road barrier, speed bump and limited lateral size of road.

11 Ariannezhad, Razi-Ardakani and Kermanshah 0 RESULTS AND DISCUSSIONS Considering the importance of the existing variables drawn from the pertinent researches, proper variables were introduced to the model, and their coefficients were estimated. It is noteworthy that variables were retained if they were significant at least at the 0. level according to the t statistics of their estimated coefficients. The final model was selected subsequent to considering the different combinations of the variables. The Stata () statistical software was employed in this research. Table shows motorcycles Crash Severity Estimation Results. The factors affecting the severity of motorcycle accidents in non-urban roads are comprehendible through interpretation of coefficients in this model. The positive coefficients represent the increased possibility of severe crashes. The goodness of fit coefficient was 0. for the model developed for motorcycles crash severity. TABLE Suburban Motorcycles Crash Severity Estimation Results Variable Coefficient t Conditional and environmental characteristics Accident day of week Weekend 0.***. Seasonal condition Fall 0.***. Winter 0.**.0 Light condition Nighttime 0.**. Dawn.0***. Weather condition Clear 0.**. Foggy.***.0 Motorcyclist characteristics Rider s age Up to 0.**. 0 above 0.***. License type Motorcycle driving license -0.** -. No driving license 0.0***. Driver s education High school diploma -0.0** -. University degree -.** -. Helmet use Not-wearing.***. Fault status Rider is at-fault.***. Rider s offense Deviation to right and left 0.**. Inattention to front 0.0***. Loss of control 0.***. Speeding.***. Overtaking and lane deviation 0.***.

12 Ariannezhad, Razi-Ardakani and Kermanshah 0 Variable Coefficient t Crash characteristics Collision partner Passenger car -0.* -. Bus.**. Heavy vehicle 0.0***. Crash Type Hit fixed object 0.***. Single vehicle.*. With pedestrian 0.***. Collision Type Angle 0.**. Head on 0.0***.0 Collision location In road -0.** -. Human factor Violating the rules 0.***. Fatigue and sleepiness.***. Inexcusable haste 0.***. Roadway characteristics Area usage Non-administrative.***. Road imperfection With imperfection 0.0*. Road alignment Level and curved 0.*. Model Observations Loglikelihood at start -. Loglikelihood at convergence *** p<0.0, ** p<0.0, * p<0. Conditional and Environmental Properties To begin with, the conditional and environmental properties and their effects on the severity of accidents are investigated based on the modeling results. As stated previously, date, season, lighting and weather conditions are the factors responsible for conditional and environmental properties. The modeling results show that the severity of accidents is higher on weekends. Similar results have been reported by Quddus et al. () and also by Pai and Saleh (0) who investigated the severity of accidents at T-like intersections. During fall and winter, the severity of motorcycle crashes increases in suburban areas. This result is in contrary with a research done by Rifaat () on motorcycle accidents in Calgary, Canada, where he concluded that crash severity declines during the winter. In general, traffic flow speed is higher in suburban areas than other roads and thus, drivers drive at higher speeds. However, they may not tend to take enough precaution during the abnormal

13 Ariannezhad, Razi-Ardakani and Kermanshah weather condition of fall and winter seasons as compared to the other seasons. Therefore, the results can be linked to harder cruising control of vehicles during the adverse weather conditions of such seasons in contrast with spring and summer. The results of modeling show that the severity of accidents which happen during night and dawn is higher than those of daytimes due to the better visibility and line of sight of drivers in daytime. Moreover, it seems that locating and observing motorcycles by other drivers in suburban areas are a more difficult task in dark because of smaller size and lack of sufficient lighting of motorcycles, leading to higher injuries of motorcycle accidents at nights. These results also conform to the results of the researches done by Savolainen and Mannering (), Chimba et al. (), Rifaat (), Bambach and Grzebieta (0), Pai and Saleh (0) and Shaheed and Dissanayake (). According to the modeling results, accidents happening during dawn affect the severity of accidents. It could be appears that fatigue and sleepiness of motorcyclists or other drivers during such times of the day contribute to the higher severity of crashes. Also, the delay in adapting to the change in lighting conditions at dawn may causes drivers to lose their full ability in detecting and reacting to road dangers, leading to more severe accidents. Among all variables relating to the weather conditions, the worst weather condition contributing to accidents is foggy weather. With no doubt, the visibility of drivers declines in foggy weather, making it harder to see other vehicles which could lead to more severe accidents. Moreover, the coefficient responsible for clear weather condition is significant in the final model which causes higher motorcycle crash severity. It should be noted that Pai and Saleh (0) and Lapparent () researches show similar results for clear weather conditions. Motorcyclist Characteristics According to the table of results, the gender of motorcyclists had no significant effect on the severity of motorcycle crashes in suburban areas, while in some papers men drivers led to higher severity (0, ) and in others led to lower severity (,, ) of accidents. Motorcyclists aged below and above 0 increases the severity of accidents as compared to middle-age drivers ( to 0 years old). Results showing more severe accidents by older drivers have also been obtained in other researches (, 0,, -). Since divers of age 0 and above have lower concentration and maneuverability and also higher vulnerability due to their age, justifies more severe accidents of this age group. Young motorcyclists are expected to show dangerous behaviors. It seems that occurrence of more severe crashes among this age group is resulted from this risky behaviors. The model shows that while motorcyclists who had proper motorcycle driving license caused lower crash severity as compared to those with other driving licenses, motorcyclists with no driving license experienced higher injuries in accidents. The results show the profound influence of riding after obtaining driving license specific to motorcycles compared with riders with miscellaneous driving licenses or no license at all. Consequently, it can be concluded that more strict authority reinforcement regarding driving licenses of motorcyclists should be in place. Also, disallowing the use of motorcycles without proper permit and imposing strict fines for offenders could act as intensives for such drivers to obtain motorcycle driving license and results in safety of motorcyclists in suburban roads.

14 Ariannezhad, Razi-Ardakani and Kermanshah Another factor which was studied in the modeling of motorcycle accidents is the education level of motorcyclists. Based on the obtained results, motorcyclists with diploma or university degree are involved in accidents with lower severity compared to the ones with no education. This could be traced to the fact that higher literacy is followed by greater knowledge about the outcome and possible casualties of careless and dangerous driving habits in such a way that the severity of accidents is lower among this group of drivers. The use or non-use of safety helmet is another variable that has been investigated in many studies regarding motorcycle accidents. In this study, lack of safety helmets has dramatically increased crash severity. Head injuries and the lack of helmet which would decrease the impact to the head is the most important reason for the obtained result, which have also been confirmed in other studies (,,,, 0, ). The positive coefficient of rider at-fault variable shows that accidents in which the motorcyclist was proved to be at-fault had higher severity as compared to other crashes. In their studies, Savolainen and Mannering () also found at-fault motorcyclist as a factor which increases crash severity in multi-vehicle crashes involving motorcycles. Quddus et al. () concluded similar results. Another variable related to motorcycle accidents is motorcyclist s offence before the accident which has been introduced in the modeling. According to the modeling results, the following factors had the most effect on increasing the crash severity: speeding, overtaking and lane deviation, loss of motorcycle control, inattention to front and deviation to right and left. In several studies, it has been found that over speeding is the factor leading to higher accident severity (,,, ). Crash Characteristics With regards to the other vehicle involved in accidents, collisions with bus and other large vehicles resulted in more severe accidents based on the modeling results. Similar results have been reported by other researchers (, 0, ). Consequently, accidents involving motorcycles colliding with passenger cars led to low crash severity. The difference between the volume and weight of buses and other large vehicles compared with motorcycles, together with higher vulnerability of motorcyclists due to lack of proper guarding leads to higher casualties of motorcycle drivers when colliding with such vehicles in suburban areas. Moreover, bus drivers in general suffer from lower line of sight (as compared to passenger cars) due to the larger size of their vehicles. This factor is intensified when considering motorcycles and thus, contributes to more severe motorcycle accidents. The results of the model show that among different types of motorcycle crashes, single vehicle crashes, collision with pedestrians and collision with fixed objects have contributed to higher casualties, of which collision with pedestrians had the most adverse effect. This result was also obtained in the studies performed by Quddus et al. (). Expectedly, pedestrian collisions in residential areas and rural areas in the vicinity of roads showed more severe results due to inattention of the drivers to the speed limits imposed in such areas. On the other hand, in his study about the severity of motorcycle accidents in large and dense French urban areas, Lapparent () concluded that collision with pedestrians

15 Ariannezhad, Razi-Ardakani and Kermanshah results in lower crash severity, while an increase in crash severity in collisions with fixed objects has also been reported by other researchers (,,, ). Another important aspect when studying accidents is the type of collision of motorcycles with other vehicles. Compared with other types, head-on and angle collisions increase crash severity the most. The same results for head on collisions were obtained in (, 0, ) and for angle collisions in Rifaat () study. Accidents which take place in positions except in road center have positive coefficient which proves that such accidents encompass higher crash severity compared to those that take place in road center. Among human factors, disobeying driving rules, fatigue and sleepiness and inexcusable haste (for either motorcyclists or drivers of other vehicles) have led to an increase in crash severity. Road Characteristics As shown by table, pavement condition has not altered crash severity significantly in suburban roads. Accidents which happened at non-administrative areas (such as residential, agricultural and recreational locations) had higher severity compared with crashes taken place at areas with other usages. There are many villages and residential, recreational and agricultural areas located in the suburban areas of the country. Local transit and transportation (pedestrians, bikers, etc) in these areas along with recklessness of the residents and inattention of motorcyclists to speed limits are the reasons why crash severity is higher in such areas. The existence of road imperfection in suburban areas leads to more severe accidents compared to non-defective roads. This result reveals that road defectiveness is of great importance in suburban areas and safe driving in these locations requires safe and proper road infrastructure due to the high speed of vehicles. In addition, vision obstruction has no effect on crash severity. Roads with geometry characteristics of curved and level cause higher crash severity. High crash severity in curved roads has been reported in other researches (,,,, 0). Higher speeds of motorcyclists in level roads and their inability to control the motorcycle in road turns could be the reasons of high crash severity in these segments of roads. Moreover, motorcycles suffer from weak handling when turning due to their single axis and thus, curves can act as a factor of motorcycle imbalance at higher speeds and cause more severe crashes. Elasticity Analysis To better understand the effect of each of the variables on crash injury severity, elasticity values are defined for different levels of severities. Since all variables are dummy, pseudoelasticity is calculated based on the relation () and the mean of elasticity values has been used in all observations. These values show the percentage change of the probability of occurrence of each alternative when converting the value of each dummy variable from zero to one. The obtained elasticity values from ordered logit model are shown in Table.

16 Ariannezhad, Razi-Ardakani and Kermanshah TABLE Average Pseudo-Elasticities of the Variables Affecting Motorcycles Crash Severity Variable Elasticity (%) Damage only Injury Fatality Conditional and environmental characteristics Accident day of week Weekend -.%.%.% Seasonal condition Fall -.%.% 0.% Winter -0.0%.%.% Light condition Nighttime -.%.%.% Dawn -.%.%.% Weather condition Clear -.%.%.% Foggy -0.%.%.% Motorcyclist characteristics Rider s age Up to -.%.%.% 0 above -.%.%.% License type Motorcycle driving license 0.% -.% -.0% No driving license -.%.%.% Driver s education High school diploma.% -.% -.% University degree.% -.% -.% Helmet use Not-wearing -.%.%.% Fault status Rider is at-fault -.%.0%.% Rider s offense Deviation to right and left -.%.%.% Inattention to front -.%.0%.% Loss of control -.%.%.% Speeding -.%.%.% Overtaking and lane deviation -.%.%.% Crash characteristics Collision partner Passenger car.% -.% -.% Bus -.%.0% 0.% Heavy vehicle -.%.% 0.% Crash Type Hit fixed object -.%.%.% Single vehicle -.%.%.% With pedestrian -.0%.%.% Collision Type Angle -.%.%.% Head on -.%.%.% Collision location In road.% -.% -.% Human factor Violating the rules -0.%.%.0% Fatigue and sleepiness -.%.% 0.0%

17 Ariannezhad, Razi-Ardakani and Kermanshah Variable Elasticity (%) Damage only Injury Fatality Inexcusable haste -.%.%.% Roadway characteristics Area usage Non-administrative -.%.0% 0.% Road imperfection With imperfection -.%.%.% Road alignment Level and curved -.%.%.% Among various lighting conditions, dawn time had the most profound effect on the severity of motorcycle accidents in suburban areas. Accidents happening during dawn increased fatality by.%. More attention by motorcyclists and other drivers to driving at these hours of the day can have positive effects on increasing traffic safety. Next in the ranking is driving at night which increases crash severity. Among different weather conditions, foggy weather was the most dangerous condition in such a way that fatality and injury probability of these accidents increased by.% and.%, respectively. As mentioned previously, low visibility and difficulty in observing motorcyclists by other drivers could be the reasons behind these results. Cautious driving and extra attention paid by motorcyclists and other drivers in foggy weather is required in order to decrease crash severity in suburban areas. Clear weather is the second worst weather condition which has led to higher motorcycle crash severity. By studying the elasticity values about motorcyclists characteristics represent the higher vulnerability of motorcyclists aged above 0 as compared to other age groups. The tabulated results show that the mentioned age group increases crash fatality by.% when compared with middle-aged motorcyclists ( to 0 years old). According to the results of the investigation of the type of motorcyclists riding license, it was observed that having proper motorcycle license has decreased motorcycle accident severity. The tabulated elasticity results show that motorcyclists without riding license, when compared to motorcyclists who have license other than proper motorcycle license, are expected to experience more fatality by.%. Another interesting result obtained by studying the elasticity values of motorcycle license is the fact that motorcyclists who had motorcycle license increased the possibility of damage-only accidents by 0.%, and decrease accident fatality by.0%. Motorcyclists with university degree had the most profound effect on crash severity in such a way that such motorcyclists decreased crash fatality and injury by.% and.% respectively, as compared to illiterate motorcyclists. Other significant factors in increasing motorcycle crash fatality are safety helmets and the role of driver in accidents. The table of elasticity results shows that by not using safety helmet, motorcyclists increase crash fatality and injury by.% and.%, respectively. Also at-fault motorcyclists boost crash fatality and injury by.% and.0%, respectively, when compared with non at-fault motorcyclists. Among offences committed by motorcyclists, over speeding increased crash severity the most. According to the results, driving above speed limit in suburban areas increased fatality and injury by.% and.%, respectively, as compared to driving within permitted speed limit. Moreover, overtaking and deviation, loss of the vehicle control,

18 Ariannezhad, Razi-Ardakani and Kermanshah inattention to the front and deviation to left or right allocated the next positions in increasing crash fatality after over speeding by.%,.%,.% and.%, respectively. Considering motorcycles collisions partner, collisions with bus had most crash severity and increased fatality and injury by 0.% and % as compared to collision with other vehicles. Next in line is collision with other heavy vehicles which increased fatality and injury by 0.% and.%, respectively. Moreover, collision with passenger cars decreased crash fatality by.% and increased damage-only accidents by.%. The worst types of accidents include single vehicle crashes, collision with pedestrians and fixed object which increased fatality by.%,.% and.% respectively. Also head-on and angle collisions increased fatality by.% and.%. Fatigue and sleepiness resulted in worst crash severity among human-related factors by increasing crash fatality by 0%. Thus, it can be concluded that drivers are required to pay attention to their sleeping hours and healthcare while driving in suburban areas. In addition, Inexcusable haste and violating traffic rules increased fatality by.% and %, respectively. Among various area usages, non-administrative areas (recreational, agricultural and residential) increased crash fatality by 0.%. Consequently, cautious driving in these areas is suggested. CONCULUSION This study is aimed at analyzing the severity of crashes involving at least one motorcycle. For this purpose, ordered logit model for the analysis of factors affecting crash severity has been used. By analyzing motorcycle crash databases, it was found that suburban accidents were relatively more dangerous than those of other areas and thus, the 00 and 00 Iranian suburban crash information was used. Finally, the model was determined based on the significance of estimated coefficients, followed by the analysis of factors affecting crash severity. Also, to determine the value of each factor, elasticity values of coefficients were calculated and explained. Modeling results indicate that accidents taking place during weekends, fall and winter seasons, nights and dawn times, clear and foggy weather associated with higher severity, while based on elasticity values foggy weather and dawn had the most effect on increased in crash fatality. In few studies, dawn and dusk variables were distinguished from day and nighttime variables in the model. Based on the obtained results, additional attention should be directed towards the effects of the mentioned times of day on motorcycle crash severity and further investigation regarding this subject in future studies is recommended. Motorcyclists below the age of and above the age of 0 increase crash severity. Elasticity analysis shows that older motorcyclists have a higher probability of experiencing fatal crashes than younger ones. Therefore, a separate study about crash severity of elderlies is recommended. It was also found that motorcyclists without license experiences severe accidents, while those with motorcycle license and diploma or university degree experience crashes with low severity. In this study, disuse of safety helmet as well as those accidents in which motorcyclists were known to be at-fault increase crash severity significantly. Among motorcyclists offences, over speeding, overtaking and deviation, loss of vehicle control, inattention to the front and deviation to right and left increased crash severity the most,

19 Ariannezhad, Razi-Ardakani and Kermanshah respectively. Motorcycle collisions with bus and other heavy vehicles increased and collisions with passenger cars decreased crash severity, respectively. Also single vehicle crashes, collision with pedestrians and fixed objects, head-on and angle collisions led to higher crash severity, with pedestrian collisions having the highest effect. Among human related factors, violating the traffic rules, fatigue and sleepiness and Inexcusable haste committed by either motorcyclists or other drivers caused higher crash severity. Accidents which happened at non-administrative areas (recreational, agricultural and residential), roads with imperfections and curved level roads had higher crash severity. Policy Recommendations According to the modeling results, the following policy modifications are recommended in order to decrease motorcycle crash severity in suburban areas. - Upgrading lighting systems of motorcycles for a better functionality in fog which results in better detection of motorcyclists by other drivers. - When considering safety improvement policies, elderlies should take priority. - Reinforcement authorities should put into higher effort to control license of motorcyclists. Also, prohibiting the use of motorcycles without proper license and more strict fines for convicts could act as incentives for motorcycles to obtain motorcycle license which increases safety of motorcyclists in suburban roads. - More efficient planning to increase motorcyclists with safety helmet and also executing more strict laws against guilty are required. - Speed detection and control of motorcyclists in suburban residential areas and informing local residents about cautious commuting could results in the safety of local residents. - Road safety in Iran needs to be improved by investing in road infrastructure, inspecting safety of roads and constructing roads without defection. In addition, speed limits in curved segments of roads should be applied. REFERENCES. WHO. A Decade of Action for Road Safety: A Brief Planning Document. World Health Organization, 0. WHO. Road Traffic Injuries. Fact sheet N, World Health Organization, 0. Hejazi, R., M. N. Shamsudin, A. Radam, K. A. Rahim, Z. Z. Ibrahim and S. Yazdani. Estimation of Traffic Accident Costs: A Prompted Model. International Journal of Injury Control and Safety Promotion, Vol. 0, No., 0, pp. -.. WHO. Mortality: Road Traffic Deaths by Country. Fact sheet N, World Health Organization, 0. Iranian Legal Medicine Organization. Accessed March, 0.. Quddus, M. A., R. B. Noland and H. C. Chin. An Analysis of Motorcycle Injury and Vehicle Damage Severity Using Ordered Probit Models. Journal of Safety Research, Vol., No., 00, pp. -.

20 Ariannezhad, Razi-Ardakani and Kermanshah Yannis, G., J. Golias and E. Papadimitriou. Driver Age and Vehicle Engine Size Effects on Fault and Severity in Young Motorcyclists Accidents. Accident Analysis & Prevention, Vol., No., 00, pp. -.. Shankar, V. and F. Mannering. An Exploratory Multinomial Logit Analysis of Single-Vehicle Motorcycle Accident Severity. Journal of Safety Research, Vol., No.,, pp. -.. Savolainen, P. and F. Mannering. Probabilistic Models of Motorcyclists' Injury Severities in Single- and Multi-Vehicle Crashes. Accident Analysis and Prevention, Vol., No., 00, pp Pai, C. W. and W. Saleh. Exploring Motorcyclist Injury Severity in Approach-Turn Collisions at T-Junctions: Focusing on the Effects of Driver's Failure to Yield and Junction Control Measures. Accident Analysis and Prevention, Vol. 0, No., 00, pp. -.. Pai, C. W. Motorcyclist Injury Severity in Angle Crashes at T-Junctions: Identifying Significant Factors and Analysing What Made Motorists Fail to Yield to Motorcycles. Safety Science, Vol., No., 00, pp Haque, M. M. and H. C. Chin. A Mixed Logit Analysis on the Right-Angle Crash Vulnerability of Motorcycles at Signalized Intersections. Presented at rd Annual Meeting of the Transportation Research Board 00.. Schneider, W. H. and P. T. Savolainen. Comparison of Severity of Motorcyclist Injury by Crash Types. In Transportation Research Record: Journal of the Transportation Research Board, No., 0, pp Chang, H. L. and T. Yeh. Risk Factors to Driver Fatalities in Single-Vehicle Crashes: Comparisons between Non-Motorcycle Drivers and Motorcyclists. Journal of Transportation Engineering, Vol., No., 00, pp. -.. Shaheed, M. and S. Dissanayake, in: Risk Factors Associated with Motorcycle Crash Severity in Kansas, Presented at 0th Annual Meeting of Transportation Research Board, 0.. Jackson, T. L. and M. J. Mello. Injury Patterns and Severity among Motorcyclists Treated in Us Emergency Departments, 00 00: A Comparison of Younger and Older Riders. Injury Prevention, 0.. Chimba, D. and T. Sando. Multinomial Probability Assessment of Motorcycle Injury Severities. Advances in Transportation Studies, Vol., 00.. Rifaat, S. M., R. Tay and A. de Barros. Severity of Motorcycle Crashes in Calgary. Accident Analysis & Prevention, Vol., No. 0, 0, pp. -.. De Lapparent, M. Empirical Bayesian Analysis of Accident Severity for Motorcyclists in Large French Urban Areas. Accident Analysis & Prevention, Vol., No., 00, pp Bambach, M. and R. Grzebieta, in: Fatality Risk Mitigation for Rural Motorcycle Collisions with Trees and Utility Poles, 0.. Geedipally, S. R., P. A. Turner and S. Patil. Analysis of Motorcycle Crashes in Texas with Multinomial Logit Model. In Transportation Research Record: Journal of the Transportation Research Board, No., 0, pp. -.. Savolainen, P. T., F. L. Mannering, D. Lord and M. A. Quddus. The Statistical Analysis of Highway Crash-Injury Severities: A Review and Assessment of Methodological Alternatives. Accident Analysis and Prevention, Vol., No., 0, pp. -.. Greene, W. H. and D. A. Hensher Modeling Ordered Choices: A Primer and Recent Developments. Cambridge: Cambridge University Press, 00.. Washington, S. P., M. G. Karlaftis and F. L. Mannering Statistical and Econometric Methods for Transportation Data Analysis. Chapman and Hall/CRC, 00.