|USFDC Home | USF Electronic Theses and Dissertations||| RSS|
This item is only available as the following downloads:
xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam 2200385Ka 4500
controlfield tag 001 002029508
007 cr mnu|||uuuuu
008 090917s2009 flu s 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0002924
Study on crash characteristics and injury severity at roadway work zones
h [electronic resource] /
by Qing Wang.
[Tampa, Fla] :
b University of South Florida,
Title from PDF of title page.
Document formatted into pages; contains 76 pages.
Thesis (M.S.C.E.)--University of South Florida, 2009.
Includes bibliographical references.
Text (Electronic thesis) in PDF format.
ABSTRACT: In USA, despite recent efforts to improve work zone safety, the number of crashes and fatalities at work zones has increased continuously over several past years. For addressing the existing safety problems, a clear understanding of the characteristics of work zone crashes is necessary. This thesis summarized a research study focusing on work zone traffic crash analysis to investigate the characteristics of work zone crashes and to identify the factors contributing to injury severity at work zones. These factors included roadway design, environmental conditions, traffic conditions and vehicle/driver features. Especially, special population groups, which divided into older, middle Age, and young, were inspected. This study was based on history crash data from the Florida State, which were extracted from the Florida CAR (Crash Analysis Reporting) system. Descriptive statistics method was used to find the characteristics of crashes at work zones. After then, an injury severity predict model, using the ordered probit regression technology, was developed to investigate the impacts of various factors on different the injury severity at work zones. From the model, it can be concluded that some factors, including the road section with curve, alcohol/drugs involved, a high speed, angle crash and too young or old drivers are more likely to increase the probability of angle crashes. Based on the magnitudes of the variable coefficients, the factor of maximum posted speed have a great impact to injury severity, which shows restriction to driving speed is principle countermeasure for improving work zone safety.
Mode of access: World Wide Web.
System requirements: World Wide Web browser and PDF reader.
Advisor: Jian Lu, Ph.D.
Ordered probit regression
x Civil & Environmental Engineering
t USF Electronic Theses and Dissertations.
Study On Crash Characteristics And Injury Severity At Roadway Work Zones by Qing Wang A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Civil Engineering Department of Civil & Environmental Engi neering College of Engineering University of South Florida Major Professor: Jian Lu, Ph.D. Abdul Pinjari, Ph.D. Yu Zhang, Ph.D. Zhenyu Wang, Ph.D. Date of Approval: March 26, 2009 Keywords: ordered probit regression, descriptive statistics, age groups, traffic safety, marginal effects Copyright 2009, Qing Wang
ACKNOWLEDGMENTS It is with great pride that I thank the brilliant minds affiliated with the Department of Civil and Environment Engineering at the University of South Florida. I would like to give special thanks to my major professor, Dr. Jian John Lu, for the guidance he has provided. In addition, I would like to thank committee members Dr. Abdul Pinjari, Dr. Yu Zhang, and Dr. Zhenyu Wang. This thesis would not have been possible without your contributions.
i TABLE OF CONTENTS LIST OF TABLES iii LIST OF FIGURES iv ABSTRACT vi INTRODUCTION 1 1.1 Background 1 1.2 Research Objectives and Approaches 6 1.3 Organization 6 LITERATURE REVIEW 8 2.1 Previous Studies on Work Zone Crashes 8 2.2 Previous Studies on Crash Severity Model 11 2.2.1 Log linear Model 12 2.2.2 Logit Model 12 2.2.3 Probit Model 16 DESCRIPTIVE STATISTICS ANALYSIS 21 3.1 The Trend of Crashes 21 3.2 Distribution of Crashes by Drivers Age 22 3.3 Distribution of Crashes by Crash Severity 23 3.4 Distribution of Crashes by Climatic Environmental Conditions 24 3.5 Distribution of Crashes by Crash Types 26 3.6 Distribution of Crashes by Contributing Factors 27 3.7 Predominant Factors for Other Variables 31 CRASH SEVERITY MODEL 33 4.1 Methodology 33 4.1.1 Crash Severity Models 33 4.1.2 Ordered Probit Regression 34 4.1.3 Criteria for Ordered Probit Models 39 22.214.171.124. z Test 39 126.96.36.199. Pseudo 2 R 40 188.8.131.52. Likelihood Ration (LR) Test 40
ii 4.1.4 Interpretation of Model Coefficients 41 184.108.40.206. The Partial C hange in y 41 220.127.116.11. Partial Change in Predicted Probabilities 42 4.2 Data Collection 43 4.2.1 Data Base 43 4.2.2 Data Description 45 4.3 Work Zone Crash Injury Severity Model 49 4.3.1 Estimation Procedure 49 4.3.2 Cross Tabulations between Explanatory Variables and Crash Severity 50 4.3.3 Estimation Results 51 4.3.4 Interpretation 53 18.104.22.168. Signs 55 22.214.171.124. Magnitude of Coefficients 55 126.96.36.199. Detailed Interpretations 56 4.3.5 Possible Countermeasures to Improve Work Zone Safety 58 SUMMARY 60 5.1 Summary 60 REFERENCES 64 BIBLIOGRAPHY 67 APPENDICES 68 Appendi x A: Variables and Codes of Work Zone Crash 69
iii LIST OF TABLES Table 4.1 Definition and Description of Crash Severity Level 34 Table 4.2 Tables from Florida Traffic Crash Records Database 44 Table 4.3 Description of Selected Variables for Model Development 45 Table 4.4 Description Statistic of Continuous Variables 47 Table 4.5 Frequencies of Discrete Variables 48 Table 4.6 Cross Tabulation between explanatory Variables and Crash Severity 50 Table 4.7 Estimation of Ordered Probit Regression for Work Zone Crash Severity M odel 52 Table 4.8 Partial Change in y* 53 Table 4.9 Partial Change in Predicted Probabilities 54 Table 4.10 Analysis of the Coefficient Signs 55 Table 4.11 Ranked Magnitudes of Coefficients 56
iv LIST OF FIGURES Figure 1.1 Component Parts of a Work Zone 2 Figure 3.1 Work Zone Crashes and Work Zone Fatal Crashes in Florida 21 Figure 3.2 Distribution of Work Zone and Nonwork Zone Crashes by Age Group 22 Figure 3.3 Distribution of Work Zone Crashes by Crash Severity 23 Figure 3.4 Distribution of Work Zone Crashes by Lighting Conditions 24 Figure 3.5 Distribution of Work Zone Cra shes by Weather Conditions 25 Figure 3.6 Distribution of Work Zone Crashes by Road Surface Conditions 26 Figure 3.7 Distribu tion of Work Zone Crashes by Crash Types 27 Figure 3.8 Distribution of Work Zone and Nonwork Zone Crashes by Crash Types 27 Figure 3.9 Distribution of Work Zone Crashes by Contributing Factors 28 Figure 3.10 Distribution of Work Zone Rear end Crashes by Contributing Factors 29 Figure 3.11 Distribution of Work Zone Angle Crashes by Contributing Factors 30 Figure 3.12 Distribution of Work Zone Sideswipe Crashes by Contributing Facto rs 30 Figure 3.13 Distribution of Work Zone Crashes by Alcohol/Drug Involved 31 Figure 3.14 Distribution of Work Zone Crashes by Heavy Vehicle Involved 32
v Figure 3.15 Distribution of Work Zone and Nonwork Zone Crashes by Heavy Vehicle Involved 32 Figure 4.1 Latent Variables to the Observed Categories 35 Figure 4.2 Distribution of y Given x for the Ordered Regression Model 35
vi Study on Crash Characteristics and Injury Severity at Roadway Work Zones Qing Wang ABSTRACT In USA, despite recent efforts to improve work zone safety, the number of crashes and fatalities at work zones has increased co ntinuously over several past years. For addressing the existing safety problems, a clear understanding of the characteristics of work zone crashes is necessary This thesis summarized a research study focusing on work zone traffic crash analysis to investigate the characteristics of work zone crashes and to identify the factors contributing to injury severity at work zones These factors included roadway design, environmental conditions, traffic conditions and vehicle /driver features. Especially, special po pulat ion groups, which divided into older, middle Age, and young, were inspected This study was based on history crash data from the Florida State which were extracted from the Florida CAR (Crash Analysis Reporting) system D escriptive statistics method was used to find the characteristics of crashes at work zones. After then, an injury severity predict model, using the ordered probit r egression technology, was developed to investigate the impacts of various factors on different the injury severity at wor k zones From the model, it can be concluded that some factors,
vii including the road section with curve, alcohol/drugs involved, a high speed, angle crash and too young or old drivers are more likely to increase the pro bability of angle crashes. Based on the magnitudes of the variable coefficients, the factor of maximum posted speed have a great impact to injury severity, which shows restriction to driving speed is principle countermeasure for improving work zone safety.
1 CHAPTER ONE INTRODUCTION 1.1 Background In Highway Capacity Manual 2002, the definition of work zone is a segment of highway in which maintenance and construction operations impinge on the number of lanes available to traffic or affect the operational characteristics of traffic flowi ng through the segment. It should be typically marked by signs, channelizing devices, barriers, pavement marking, and/or work vehicles. It extends from the first warming sign or highintensity rotating, flashing, oscillating, or strobe lights on a vehicle to the End Road Work sigh or the last temporary traffic control device. The Manual on Uniform Traffic Control Devices lists five distinct areas within a work zone. Each of these has a specific purpose and may vary in size and location depending on the specifics of each work zone. The five areas are: advance warning area, transition area, activity area, buf fer space, and termination area ( Figure 1.1). The advance warning area is the section of highway where road users are informed about the upcoming work z one or incident area. The transition area is that section of highway where road users are redirected out of their normal path. Transition areas usually involve strategic use of tapers, which because of their importance are discussed
2 Figure 1.1 Component Parts of a Work Zone
3 separately in detail. The activity area is the section of the highway where the work activity takes place. It is comprised of the work space, the traffic space, and the buffer space. The work space is that portion of the highway close d to road users and set aside for workers, equipment, and material, and a shadow vehicle if one is used upstream. Work spaces are usually delineated for road users by channelizing devices or, to exclude vehicles and pedestrians, by temporary barriers. Typi cally, the buffer space is formed as a traffic island and defined by channelizing devices. When a shadow vehicle, arrow panel, or changeable message sign is placed in a closed lane in advance of a work space, only the area upstream of the vehicle, arrow pa nel, or changeable message sign constitutes the buffer space. The termination is the end area of work zone. Work zone safety has always been a high priority issue in highway systems but remains unsatisfactory in USA. Based on the statistics from FHWA ( Fed eral Highway Administration ), in 2007, there were 835 work zone fatalities, which represent 2.0% of all roadway fatalities for the year. Over four out of every five work zone fatalities were motorists. In addition, there are over 40,000 injuries at work zones. The total cost of highway work zone injuries calculates to $9.25 billion per year. The highway work zone fatalities per billion dollars spent, are at list 4 times more than in total constructio n ( Maze et al., 2000 ). Estimating between 1995 and 1997, t he direct costs of highway construction zone accidents were as high as $6.2 billion per year, and the average cost is $3687 per accident ( Mohan and Gautam, 2002)
4 To improve work zone safety, four fields need to be approached contemporaneously: engineering, education, enforcement, and coordination with public agencies. Engineering: This focuses on standardization and evaluation. The standardization part is for traffic control and safety devices in work zone areas. The MUTCD ( Manual on Uniform Traffic Contro l Devices) is the national safety standards to control traffic through work zones, and the NCHRP350 ( National Cooperative Highway Research Program Report 350 Recommended Procedures for the Safety Performance Evaluation of Highway Features ) contains the f ederal standards and guidelines for all work zone safety devices. The national guidelines regarding planning and implementing work zones is keeping update to address the changing times of more traffic more congestion, greater safety issues, and more work z ones. Education: Public awareness is improved through a variety of activities like clearinghouse website ( www.workzonesafety.org ) ; training courses for federal, state, l ocal and tribal highway engineers; conferences, CDs; guidebooks; brochures (for the general public and highway practitioners); bilingual safety public outreach materials; and press events such as National Work Zone Awareness Week. Enforcement: Engineers in federal highway work closely with state highway to identify appropriate engineering safety countermeasures for highrisk locations new roads. They also work with the enforcement community such as the IACP ( International Association of Chiefs of Police ). S peed enforcement is a top safety concern in work
5 zones since it has critical relationship with crash severity. In Maryland, Michigan and Virginia, VSL ( Variable speed Limits) demonstration projects which determine appropriate speeds for work zones and change them when conditions change were to analyze variations on speed and accompany driver behavior. Association: Working with emergency medical services, police and fire organizations can ensure that public safety is maintained at high levels and access for emergency vehicles is possible during work zone operations. AASHTO ( American Association of State Highway and Transportation Officials ), ATSSA ( American Traffic Safety Services Association ) and FHWA found the National Work Zone Awareness Week in April eve ry year to bring national attention to motorist and worker safety and mobility issues in work zones. Beside this, lots of other publications like Basic Traffic Control for Utility Operations manual and Strategic Highway Safety Plan are the productions by m ore than one partner or sponsor. Researching the characteristics of crashes is the very first step of learning the deficiencies of work zone safety and countermeasures. In addition, studying the characteristic differences between each crash injury severit y level may cause the discovery of factors influencing injury severity change, which could benefit the development of traffic controls for reducing the proportion of highseverity crashes in total crashes.
6 1.2 Research Objectives and Approaches The main objectives of this study are to investigate the characteristics of accidents in work zones to identify the factors contributing to injury severity levels and to study how the se factor s influence injury levels. For more specifically, this study follows t hese steps: (1) Review the previous researches in the field of work zone crash characteristics and injury severity models. (2) Determine the most promising model for model development part by comparing various models in literature review part. (3) Investigate the differences of characteristics such as crash severity, e nvironmental conditions, crash types and contributing factors among three driver age groups. (4) D evelop a crash severity model for the identificati on of the most significant factors contribu ting to the injury severity levels. 1.3 Organization This thesis consists of five chapters. Chapter 1 provides a brief introduction of the research, including the background of the research, research objective and approaches. Chapter 2 discusses the past studies in both work zone crash characteristics and crash injury severity models, and chooses the most appropriate model to develop the work zone injury severity model for this study. Chapter 3 compares the descriptive characteristics of
7 work zone crashes in three age groups, including the crash severity, environmental conditions and some other contributing factors. A crash injury severity model is produced and interpreted; the factors that influence crash severity levels are found are given in chapter 4. F inally, chapter 6 provides a summary and the conclusion of this research.
8 CHAPTER TWO LITERATURE REVIEW 2.1 Previous Studies on Work Zone Crashes Many studies have been performed on accident experience within work area in the United States. Most of them focus on the crash characteristic in diverse work zone types, crash severity levels, and different locations within work zone. Ullman et al. (2005) presented an analysis of the safety effects of night work activity upon crashes at two types of constr uction projects in Texas. The first project type involved both day and night work, whereas the other project type involved pavement resurfacing activities performed only at night. They found that crashes increased more significantly during periods of work activity than during periods when the work zone was inactive. Overall, the increase during work activity was somewhat higher at night than during the day. Researchers also found that crashes increased more at night than during the day at the hybrid project s even when the work zone was inactive, presumably reflecting a disproportionate influence of the temporary geometrics and traffic control upon nighttime travel at these sites. 77 fatal work zone crash sites throughout Texas from Feb. 2003 to Apr. 2004 we re analyzed by Schrick (2004). Based on these investigations, researchers concluded that
9 only 8 percent of the investigated crashes had a direct influence from the work zone, whereas 39 percent of the investigated crashes had an indirect influence from the work zone. Researchers also concluded that 45 percent of the investigated crashes appeared to have no influence from the work zone (included in this subset are the 16 percent of the investigated crashes which occurred in work zones that were work zones in name only, such as work zones that consisted only of project limit signing). The characteristics of highway work zone collisions and their detailed locations within work zones were studied by Garber and Zhao (2002) to enhance the selection of effective countermeasures. The objective was to determine the distribution and characteristics of crashes in specific areas within a work zone and to compare selected characteristics of work zone crashes with those of non work zone crashes. In their study, the differ ent locations in the work zone were referred to as the advance warning area, transition area (taper), longitudinal buffer area, activity area, and termination area. Based on the crash percentages regarding location, severity, and collision type, the resear chers concluded several major findings. First, the activity area had the highest number of crashes and the highest number of fatal crashes while the termination area was the safest area in terms of numbers of crashes. Second, property damageonly (PDO) cra shes were the predominant severity type, followed by the injury crashes. Third, rear end crashes were predominant for all areas and all road types except for the termination area, where all crashes were angle crashes. Fourth, as traffic moved from the tran sition area to the
10 work area, the proportions of rear end and samedirection sideswipe crashes decreased and the proportions of fixedobject, off road, and angle crashes increased, although rear end crashes were still predominant. Last, most nighttime work zone crashes were in the activity area and the severities of nighttime and daytime work zone crashes were not significantly different. In 2000, Daniel et al. performed a study which was expanded further to examine the difference between fatal crash activ ity within work zones compared with fatal crashes in nonwork zone locations. Using data from three work zone locations in Georgia, fatal crash activity within work zones also was compared with nonfatal crashes within work zones. Finally, fatal crash activ ity was examined to determine the influence of the work zone activity on the frequency of fatal crashes. The overall findings of the study indicate that the work zone influences the manner of collision, light conditions, truck involvement, and roadway func tional classification under which fatal crashes occur. The study also indicates that fatal crashes in work zones are more likely to involve another vehicle than nonwork zone fatal crashes, and fatal crashes in work zones are less influenced by horizontal and vertical alignment than are nonwork zone crashes. Khattak et al. (2002) created a unique dataset of California freeway work zones that included crash data (crash frequency and injury severity), road inventory data (average daily traffic and urban/rura l character), and work zone related data (duration, length, and location). Crash rates and crash frequencies were investigated in the prework zone and
11 duringwork zone periods. For the freeway work zones investigated in this study, the total crash rate in the duringwork zone period was 21.5% higher (0.79 crashes per million vehicle km) than the pre work zone period (0.65 crashes per million vehicle km). Compared to the pre work zone period, the increase in noninjury and injury crash rates in the during w ork zone period was 23.8% and 17.3%, respectively. Next, crash frequencies were investigated using negative binomial models, which showed that frequencies increased with increasing work zone duration, length, and average daily traffic. Wang et al. (1996) discussed the primary questions that safety researches are attempting to answer. The results were presented of an investigation to (a) determined what is known about the magnitude of highway work zone crashes, (b) examined characteristics of highway work z one crashes using the Highway Safety Information System, (c) investigated how work zone accidents are reported on police accident report forms and within state accident report systems, (d) identified critical voids in the knowledge of the relative safety of work zones, and (e) examined possible ways to address unfulfilled information needs related to work zone safety. 2.2 Previous Studies on Crash Severity Model Researchers have employed many statistical techniques to analyze crash severity level. Among t hese techniques were log linear logit, and probit models.
12 2.2.1 Loglinear Model Using 1994 and 1995 crash data from Florida, Abdel A ty et al. (1998) used loglinear technique to examine relationships between driver age and crash characteristics. The thre e injury severities in their study were no injury, injury and fatality, and their results suggest that injury severity is positively associated with age; they also concluded that middle age drivers are more likely to be involved in some crashes, but older drivers are more likely to be involved in fatal crashes. Kim et al. (1995) used log linear models to predict automobile crash and injury severity. The results suggested that alcohol or drug use and lack of seat belt use increase the odds of more severe crashes and injuries. 2.2.2 Logit Model Logistic regression models were developed by Donnell and Mason (2004) using both an ordinal and a nominal response. The results indicate ed that modeling crash severity as an ordinal response provided appropriate resul ts for cross median crashes, whereas a nominal response was more appropriate for median barrier crashes. Explanatory variables such as pavement surface conditions, use of drugs or alcohol, presence of an interchange entrance ramp, horizontal alignment, cra sh type, and average daily traffic volumes affect crash severity. The analysis results m ight be used by practitioners to understand the trade off between geometric design decisions and median related crash severity. Approximately 0.7% median barrier crashe s on the
13 Interstate system resulted in a fatality, whereas 43% were property damageonly crashes and about 56% were injury crashes. More than 17% of crossmedian collisions were fatal, and 67% involved injury. Modeling severity as a discrete outcome invol ves estimating the probability that a vehicular crash has a certain severity by determining the likelihood of outcomes given that a crash has occurred. Lee and Chang (2002) estimated the severity of run offroad crashes in the state of Washington, again by using the nested logit model. Temporal, environmental, driver, roadway, and roadside characteristics were used to estimate property damage and possible injury probabilities for rural run offroad crashes conditioned on no evident injury. The findings indi cated that wet pavement surfaces resulted in possible injury, drivers younger than 25 were more likely to be involved in injury crashes, alcohol impaired drivers were more likely to be involved in injury crashes, and crashes in the presence of a horizontal curve were more likely to involve an injury. Dissanayake and Lu (2002) used binary logistic regression model takes the following form Factors that prove most influential in predicting severity in young driver crashes included influence of alcohol or drugs, ejection in the crash, point of impact, crash location, existence of horizontal curve or vertical grades at the crash site, speed of the vehicle, and restraint device usage. Krull, Khattak, and Council (2000) used logit models to analyze driver injury severity involved in a single vehicle crash. Threeyear crash data from Michigan and
14 Illinois were analyzed to explore the effect of rollover, while controlling for roadway, vehicle, and driver factors. Results showed that driver injury severity increases with: (a) failure to use a seatbelt, (b) passenger cars as opposed to pick up trucks, (c) alcohol use, (d) daylight, (e) rural roads as opposed to urban, (f) posted speed limit, and (g) dry pavement as opposed to slippery pavement. Chang and Mannering (1999) estimated a nested logit model to study the occupancy crash injury severity relationship. Crash data of principle arterials, state highways, and interstates in Seattle, Washington, during 1994 were used in the analysis. The dependent variable was the crash severity, which represents the most severe level of injury sustained by any vehicle occupant involved in the crash. The occupancy can be significant because vehicles with large occupancies have an increased likelihood of having someone seriously inju red. Separate models were estimated for non truck involved crashes and for nontruck involved crashes. Results showed that increased severity was more likely for truck involved crashes, high speed limits, crashes occurring when a vehicle is making a right or left turn, and rear end types of collisions. Shankar, Mannering, and Barfield (1996) estimated a nested logit model to analyze crash severity of single vehicle crashes on rural freeways. All possible nesting structures (which examine possible correlati on among the choices) were considered and statistically tested by the likelihood ratio test. The authors found that a nestedlogit model, which
15 treated property damage only (no injury) and possible shared characteristics of injury crashes, fits the data best. Shankar and Mannering (1996) used a multinomial logit specification for estimating motorcycle rider crash severity likelihood conditioned on the occurrence of a crash. Five levels of severity are considered: property damage only, possible injury, evident injury, severe injury, and fatality. Crash data were 5 year statewide data on single vehicle motorcycle crash from the state of Washington. Results showed that the multinomial logit formulation is a promising approach to evaluate the determinants of motorcycle crash severity. Nassar, Saccomanno, and Shortreed (1994) estimated a nested logit model to predict crash severity. Three separate models were calibrated for three crash situations: single vehicle, twovehicle, and multivehicle crashes. Factors that affect the level of damage experienced by individuals involved in traffic crashes include a crash dynamic term, seating position, seat belt use, vehicle condition, vehicle mass, driver condition, and driver action. Road surface condition was insignif icant in the models. Bad weather conditions may prompt drivers to slow down and keep a safe distance from other vehicles.
16 2.2.3 Probit Model Abdel Aty and Keller (2005) produced ordered probit models for crash severity level and used the tree based regr ession to explore the factors which affect injury level. T he results of this research show ed that when attempting to forecast the number of expected crashes of different severity levels, it is imperative that models are developed for each level of collisio n instead of aggregating crash types to predict the overall severity level. While the ordered probit model approach had been adopted, as did many previous researchers, using the treebased regression for each severity level improved our understanding of the specific factors and their importance for each severity level. Furthermore, the results showed that crashes reported on short forms are important and should therefore be retained and included in crash databases. Ignoring this data could lead to biasing t he results by under reporting crashes of certain severity or type that could be related to specific explanatory factors. Other crash types or severities might appear to have higher percentages, and therefore, their effect could be artificially exaggerated. Khattak and Targa (2004), Khattak et al. (2002, 2003) used ordered probit models to predict the injury level for crashes occurring at construction zones and involving trucks, to predict injury severity for single vehicle truck rollovers, and to determine vehicle, roadway, driver, crash, and environmental characteristics that influence the severity level of older drivers involved in crashes, respectively.
17 Abdel Aty (2003) applied the ordered probit models to predict crash injury severity on roadway secti ons, signalized intersections and toll plazas. Models explained a drivers violation was significant in the case of signalized intersections. Alcohol, lighting conditions, and the existence of a horizontal curve affected the likelihood of injuries in the r oadway sections model. A variable specific to toll plazas, vehicles equipped with Electronic Toll Collection (ETC), had a positive effect on the probability of higher injury severity at toll plazas. Other variables that entered into some of the models wer e weather condition, area type, and some interaction factors. This study illustrates the similarities and the differences in the factors that affect injury severity between different locations. Kockelman and Kweon (2002) described the use of ordered probi t models to examine the risk of different injury levels sustained under all crash types, twovehicle crashes, and single vehicle crashes. The results suggested that pickups and sport utility vehicles are less safe than passenger cars under singlevehicle crash conditions. In two vehicle crashes, however, these vehicle types were associated with less severe injuries for their drivers and more severe injuries for occupants of their collision partners. Other conclusions also were presented; for example, the results indicated that males and younger drivers in newer vehicles at lower speeds sustain less severe injuries. Toshiyuki and Shankar (2002) used a bivariate ordered response probit model to study driver and most severely injured passenger severity in collision with fixed objects in Washington State. Results showed that icy roadway surface and rain decrease the
18 probability of more severe driver injury. The type of fixed objects significantly affects drivers injury severity. Guardrails have different effects on drivers injury whether the collisions are with its face or with its leading end. Proper use of a restraint system significantly decreases the probability of more severe driver injury. Male and younger drivers have a lower probability of more severe injury, probably because of their physical strength. Also, drivers unconsciousness causes more severe driver injury. Duncan, Khattak, and Council (1999) used ordered probit modeling to examine the occupant characteristics and roadway and environmental conditions that influence injury severity in rear end crashes involving truckpassenger car collisions. Two models were developed, one with the basic variables and the other including interactions among the independent variables. Results revealed that an in creased severity risk exists for higher speed crashes, those occurring at night, for women, when alcohol is involved, and for crashes when a passenger car rear ends a truck at a large differential speed between the two vehicles. Khattak (1999) applied the ordered probit model to examine the effect of information (accuracy of information conveyed by brake and turning lights) and other factors on rear end crash propagation and the propensity of driver injury in such crashes. Results on injury severity showed that in a two vehicle crash, the leading driver is more likely to be injured, whereas, in a three vehicle crash, the driver in the middle is likely to
19 be more severely injured. Furthermore, as rear end crashes propagate from two vehicles to three vehicl es the last driver is relatively less severely injured. Klop (1998) examined the impacts of physical and environmental factors on the severity of injury to bicyclists in North Carolina. Using the ordered probit model, the effect of a set of roadway, envir onmental, and crash variables on injury severity was explored. Separate models were estimated for rural and urban locations. Results indicated that straight grades, curved grades, darkness, and fog significantly increase injury severity. Renski, Khattak, and Council (1998) estimated ordered probit models to explore the effects of policy variables on injury severity. Results showed that highway segments where speed limits were raised by 10 mph resulted in a higher probability of increased severity than thos e raised by only 5 mph. No significant changes in injury severity were found for the highway segments where speed limits were raised from 65 to 70 mph. In assessing the probabilities of four levels of injury severity as a function of driver attributes, O Donnell and Connor (1996) compared ordered logit and ordered probit specifications. Their results suggest that injury severity rises with speed, vehicle age, occupant age (squared), female gender, blood alcohol levels over 0.08 percent, nonuse of a seatbelt, manner of collision (e.g., headon crashes), and travel in a light duty truck. And, according to their comparison of effects, seating position of crash victims was most relevant (e.g., the left rear seat of the vehicle was found to be most dangerous) a nd
20 gender least relevant. Many of their results are echoed in the models presented here; the key distinction is that here collision partners and crashtype are examined and emphasized. Hutchinson (1986) developed an ordered probit model to study occupants injury severity when involved in traffic crashes. British crash data for 19621972 had been processed to give a cross tabulation of the severity of injury to the driver and to the front seat passenger in four types of single vehicle crashes (overturning and nonoverturning, each in rural and urban areas). Results showed that passengers tend to be more seriously injured than drivers in nonoverturning, but that there is no difference in overturning crashes.
21 CHAPTER THREE DESCRIPTIVE STATISTICS ANALYSIS 3.1 The Trend of Crashes The trend of work zone crashes and fatal crashes are ascending continuously from 2002 to 2006 in Florida (see Figure 3.1). The average annual increase rate of work zone crashes is 18.8%, and the fatal crashes in 2006 are 64.4% more than one in 2002. This trend indicates that the work zone safety in Florida remained a serious concern. Figure 3.1 Work Zone Crashes and Work Zone Fatal Crashes in Florida
22 3.2 Distribution of Crashes by Drivers Age Figure 3.2 shows the age dis tribution of the atfault drivers for work zone and nonwork zone crashes. The drivers are divided into three age groups: Young Age (less than 25), Middle Age (25 64) and Elderly Age (greater than 65). In work zone area, the middle age drivers cause the highest proportion (67%) of crashes, while the elderly drivers are only responsible for 9% of the crashes. The driver group having the second highest crash rate (24%) is the young age drivers. Compared to work zone crashes, middle age drivers in nonwork z one area have a lower possibility of occurring crashes (63%). Figure 3.2 Distribution of Work Zone and Nonwork Zone Crashes by Age Group
23 3.3 Distribution of Crashes by Crash Severity The distribution of work zone crashes by crash severity is shown i n Figure 3.3, which indicates that the middle age drivers involved the highest percentage in the no injury crashes which is 49%, and always has the lowest percentage in other severity levels. While in the more severe level crashes, elderly drivers contribu te more than the other two age groups (Incapacitating Injury: Old Drivers 9% and Fatal Injury: Old Drivers 2%). Figure 3.3 Distribution of Work Zone Crashes by Crash Severity
24 3.4 Distribution of Crashes by Climatic Environmental Conditions Climatic Environmental conditions include lighting conditions, weather conditions, and road surface conditions. Figure 3.4 summarizes the distribution of crashes by lighting conditions. Most crashes occur when lighting condition is good. Elderly drivers is most likely to having crashes under good lighting condition (daylight), and only has 18% crashes under non daylight condition including dawn, dusk and dark conditions. In contrast, the difference of crash rate between these two lighting conditions in young driver s is not remarkable. Figure 3.4 Distribution of Work Zone Crashes by Lighting Conditions
25 The results of analysis of the distribution of work zone crashes by weather and road surface conditions are shown in Figures 3.5 and 3. 6 respectively. The re sults indicate that in all three age groups only a small proportion of work zone crashes occur in bad weather or bad road surface conditions. In contrast to the common sense, the adverse weather and road conditions do not have significant influence on the work zone fatal crashes. Figure 3.5 Distribution of Work Zone Crashes by Weather Conditions
26 Figure 3.6 Distribution of Work Zone Crashes by Road Surface Conditions 3.5 Distribution of Crashes by Crash Types As illustrated in Figure 3. 7, the top three work zone crash types in all age groups are the same. There are rear end, angle and sideswipe which are defined as the principle crash types in this study. In young and middle age groups, the percentage of rear end crashes is obviously higher than angl e and sideswipe crashes. Elderly age group shows higher rate in angle crashes than others. Compared work zone and nonwork zone crash type s in Figure 3.8, read end and sideswipe crashes are more likely to be occurred in work zone area.
27 Figure 3.7 Distri bution of Work Zone Crashes by Crash Types Figure 3.8 Distribution of Work Zone and Nonwork Zone Crashes by Crash Types 3.6 Distribution of Crashes by Contributing Factors Figure 3.9 represents the distribution of contributing factors by all drivers a nd each age group. Among all drivers, careless driving, the most predominant contributing factor, is responsible for 43% of total crashes. Another predominant factor is failed to yield right of way (11%) followed by no improper driving action (10%) and improper lane change (7%) respectively. In young and middle age group, the distributions are basically same as
28 which of all drivers, except that young drivers show slightly higher rate in careless driving (48%), and the second and third factors which are not variant too much in rate. But in elderly age group, the rate of first factor is just 34% and second one is more than 10% higher than other two age groups. Figure 3.9 Distribution of Work Zone Crashes by Contributing Factors Figure 3.10 to 3.12 expre ss the distribution of predominant contributing factors over the principal crash types. The most predominant contributing factor for rear end crash es is careless driving (average 74% in all three age groups). A difference between elderly age group and the other two age groups i s that improper lane change is not a predominant
29 contributing factor for older age drivers but it is for young age drivers and middle age drivers Figure 3.10 Distribution of Work Zone Rear end Crashes by Contributing Factors Fa iled to yield right of way is the most predominant contributing factors for angle crashes. In elderly age group, the rate of this crash type is significantly higher than young and middle age groups; otherwise the rate of careless driving is less than other s. For sideswipe crashes, the improper lane change is the most frequent contributing factor in middle (36%) and elderly (40%) age group, and second most one is careless driving (19% for both groups). However, for young drivers, the top two factors have no much difference (27% for improper lane change and 30% for careless driving).
30 Figure 3.11 Distribution of Work Zone Angle Crashes by Contributing Factors Figure 3.12 Distribution of Work Zone Sideswipe Crashes by Contributing Factors
31 3.7 Predominant Factors for Other Variables The distributions of alcohol/drug involved and heavy vehicle (heavy truck and truck tractor) involved are given in Figure 3.13, 3.14 and 3.15. Old drivers are seldom influenced by alcohol/drug (only 1% involved), and most work zone crashes for young age group is not included by heavy vehicle. But heavy vehicle is more easily related to work zone crashes (14%) than nonwork zone crashes (7%). Figure 3.13 Distribution of Work Zone Crashes by Alcohol/Drug Involved
32 Figure 3.14 Distribution of Work Zone Crashes by Heavy Vehicle Involved Figure 3.15 Distribution of Work Zone and Nonwork Zone Crashes by Heavy Vehicle Involved
33 CHAPTER FOUR CRASH SEVERITY MODEL 4.1 Methodology As stated in previous papers In contrast to the multinomial models which neglect the datas ordinarily and require more parameters estimated and nested logit models that produce better results but have complexity in identifying the nesting structure, the ordered probit models with a relatively simple approach recognize the indexed nature of various response variables. They are recommended to analyze the crash severity levels. 4.1.1 Crash Severity Models The crash severity model in this study w as developed to investigate the factors that affect crash severity in work zone area. The dependent variable in the model is injury severity level, and the independent variables are the factors which have significant influence on the crash severity. The crash injury severity is a typical ordinal variable which could be categorized at five levels from the least severe level to the most severe level (shown in Table 4.1)
34 Table 4.1 Definition and Description of Crash Severity Level Level Definition Description 1 No Injury there is no reason to believe any person received bodily harm from the crash 2 Possible Injury No visible signs of injury but complaint of pain or momentary unconsciousness 3 Non incapacitating I njury Visible injuries from the such as bruises, abrasions, limping, etc. 4 Inca pacitating Injury Any visible signs of injury from the crash and person(s) had to be carried from the scene. 5 Fatal Injury an injury sustained in a motor vehicle crash that results in death within 90 days 4.1.2 Ordered Probit Regression The ordered probit model is as followed: i i ix y (4.1) w here iy is the latent and continuous m easure of crash injury severity; i is the number of crash es faced by this severity level; ix is a vector of parameters to be estimated ; i is a random error term which assumed to follow a normal distribution with mean 0 and variance 1. The pdf (Probability Density Function) is 2 exp 2 12 (4.2) and the cdf (Cumulative distribution Function) is dt 2 exp 2 12 (4.3) The observed and coded discrete crash injury severity variable y is determine d from the model as follows:
35 If If 1 If 3 If 2 If 11 1 2 3 2 2 1 1 i n n i n i i i iy n y n y y y y (4.4) This mapping from the latent variable to the observed crash injury severity class is illustrated in Figure 4.1. Figure 4.1 Latent Variables to the Observed Categories Figure 4.2 Distribution of y Given x for the Ordered Regression Model
36 Consider Figure 4.2 which shows the distribution of y for four values of x The errors are distributed normally a round the regression line x x y E The Probability of outcome m corresponds to the area of the error distribution between the cutpoints 1 m and m This area is computed as follows. First consider th e formula for the probability that y = 1. We observe y = 1 when y falls between 0 and 1 This implies that i i i ix y x y1 0Pr 1 Pr (4.5) Substituting i i ix y i i i ix x x y1 0Pr 1 Pr (4.6) Then, subtracting x within the inequality, i i i i ix x x x y 1 0Pr 1 Pr (4.7) The probability that a random variable is between two values is the difference between the cdf evaluated at these values. Therefore, i i i i i ix x x x x y 0 1Pr Pr 1 Pr i ix x 0 1F F (4.8) These steps can be generalized to compute the probability of any observed outcome y = m given x : i m i m i ix x x m y 1F F Pr (4.9) When computing x y 1 Pr the second term on the right hand side drops out since 0 F F0 x x ; when computing x J y Pr the first term equal 1 since
37 1 F F x xJ Thus, for a model with four observed outcomes, such as shown in Figure 4.2, the formula s for the ordered probit model are i i ix x y 11 Pr i i i ix x x y 1 22Pr i i i ix x x y 2 33Pr (4.10) i n i n i ix x x n y 11 Pr i n i ix x n y 11 Pr where i is an individual; 1, 2, 3n1, n are response alternatives; is the standard normal cumulative distribution function. Since y is latent, its mean and variance cannot be estimated. The variance is ident ified by using that 1 Var x While these assumptions identify the variance, the mean of y is still unidentified. The consequences of this can be seen by considering the model x y with cutpoints m. Think of and the s as the true parameters in the sense that they were used to generate the observed data. Define an alternative set of parameters: and m m (4.11) where is an arbitrary constant. The probability that m y is identical, whether the true or alternative parameters are used:
38 x x x m ym m 1F F Pr x xm m 1F F i m i mx x 1 1F F (4.12) Since both sets of parameters generate the same value for the probability of an observed outcome, there is no way to choose between the two sets of parameters using the observed data: a chan ge in the intercept in the structural model can always be compensated for by a corresponding change in the thresholds. That is to say, the model is unidentified. While there are an infinite number of assumptions that could be made to identify the model, only two are commonly used: ( 1) Assume that 01 This involves setting 1 in Equation 4.11. ( 2) Assume that 0 This involves setting in Equation 4.11. Both assumptions identify the model by imposing a constraint on one of the parameters. The different identifying assumptions lead to what are known as different parameterizations of the model. The choice of which parameterization to use is arbitrary and does not affect the s (except for 0) or associated significance tests. Further, as known by Equation 4.12, the probabilities are not affected by the identifying assumption. However, understanding the different parameterizations i s important since different software uses different parameterizations. Programs such as LIMDEP uses the first assumption, while programs such as Markov, SASs LOGISTIC, and Stata use the second
39 one. The choice of parameterization does not affect estimates of the slopes, but does affect the estimates of 0 and the s. 4.1.3 Criteria for Ordered Probit Models 188.8.131.52. z Test z Test is used to test the statistical significance of individual estimated coefficient in ordered porbit models. Maximum likelihood estimators possess a number of desirable properties when certain general conditions apply. Independent and identically distributed observations, and independence of the i x and the model err ors (the i) are all that is required. With these conditions satisfied, the maximum likelihood estimator is asymptotically unbiased (consistent), is normally distributed, and has the smallest variance among all consistent and asymptoti cally normal estimators. The t ratios for the null hypothesis 0 H that 0 i and the test statistic is k zi i (4.11) where i i s the estimator of i; and i is the i th coefficient of the model; i is the estimator of standard deviation of the coefficient i; i is number of observations. If 0 H is true, the coefficient i of the model is not statistically significant. If 0H is rejected at a confidence level (usually is 0.05), the coefficient i is significant to the re sponse.
40 184.108.40.206. Pseudo 2 R A Pseudo 2 R is often used as a goodness of fit measure in non linear models. They look like 2R in the sense that they are on a similar scale, ranging from 0 to 1, but they cannot be interpreted as one would interpret an ordinary least squares ( OLS ) 2 R and different Pseudo 2 R can arrive at very different values. Here, the Pseudo 2R is provided as ercept fullM L M L Rint 2 ln ln 1 (4.12) where full M is the model with predictors; ercept M int is the model without predictors; L is the estimated likelihood. A likelihood fall s between 0 and 1, so the log of a likelihood is less than or equal to zero. If a model has a very low likelihood, then the log of the likelihood will have a larger magnitude than the log of a more likely model. Thus, a small ratio of log likelihoods indic ates that the full model is a far better fit than the intercept model. 220.127.116.11. Likelihood Ration (LR) Test The likelihood ratio test is a statistical test of the goodness of fit between two models. It relies on a test statistic computed by taking the rat io of the maximum value of the likelihood function under the constraint of the null hypothesis to the maximum with that constraint relaxed. The null hypothesis is 0 :0H, where is the intercept. This statistic is gi ven as
41 ned unconstrai d constraineM L M L G ln ln 22 (4.13) where d constraineM L is the likelihood of the constrained model; ned unconstraiM L is the likelihood of the unconstrained model. This LRT statistic approximately follows a chi squa re distribution The degree of freedom is equal to the number of additional parameters in the unconstrained model. If the null hypothesis is rejected (the confidence level is usually 0.05), it can be concluded that at least one independent variable has sig nificant influence for the dependent variable. 4.1.4 Interpretation of Model Coefficients 18.104.22.168. The Partial Change in y In the ordered regression model, x y (4.16) and the partial change in y with respect to kx is k kx y (4.17) Since the model is linear in y, the partial change can be interpreted as: for a unit increase in kx, y is expected to change by k units, holding all other variables constant. Because the variance of y cannot be estimate from the observed data, the meaning of a change of k units in y is unclear. Interpretations should be based on ystandardized coefficients.
42 If y is the unconditional standa rd deviation of the latent y, then the y standardized coefficient for kx is y k Sy k (4.18) which can be interpreted as: for a unit increase in kx, y in expected to increase by Sy k standard deviations, holding all other variables constant. ystandardized coefficients indicate the effect of an in dependent variable in its original unit of measurement. This is sometimes preferable for substantive reasons and is necessary for binary independent variables. The variance of y can be estimated by the quadratic form: Var ar V 2 xy (4.19) where x ar V is the covariance matrix for the x s computed from the observed data; contains Maximum Likelihood (ML) estimates; and 1 Var in the ordered probit model. 22.214.171.124. Partial Change in Predicted Probabilities The predicted probability that x m y given is x x x m ym m F F Pr (4.20) Taking the partial derivative with respect to kx,
43 k m k m kx x x x x x m y 1F F Pr x f x fm k m k 1 x f x fm m k 1 (4.21) The partial change or marginal effect is the slope of the curve relating kx to x m y Pr holding all other variables constant. The sign of the marginal effect is not ne cessarily the same as the sign of since x f x f m m 1 can be negative. Indeed, it is possible for the marginal effect of kx to change signs as kx changes. In general, the marginal effect does not indicate the change in the probability that would be observed for a unit changes in kx However, if an independent variable varies over a re gion of the probability curve that is nearly linear, the marginal effect can be used to summarize the effect of a unit change in the variable on the probability of an outcome. 4.2 Data Collection 4.2.1 Data Base The data set used to fit the ordered probit model was extracted from the Florida Crash Analysis Reporting (CAR) system. CAR system is a relational database for State System crashes consisting of nine tables which contain different data relevant to a certain facet of a traffic crash ( Table 4.2). It maintains electronic crash records based on crashes reported on the longform crash report. That t he variable FIRST ROAD CONDITION
44 CRASH COD is equal to 04 (road under repair/construction) is used as the indicator of work zone crashes. In this study, the work zone crash data set contain ed all the work zone crashes from 2002 to 2006. Some variables in the database were selected for modeling. They may include ordinal variables, nominal variables, or continuous variables. In order to get better result perform ance all categorical variables should be purposely converted to binary ones (dummy variable) The continuous variables need to be normalized (by dividing by each maximum value) to have values which lie between 0 and 1. The reason for this is that the dummy variables have means between 0 and 1, and ordered multiple choice models are almost never estimable if the variables are of very different magnitudes (Greene 1993). All the missing values are deleted from database. Appendix A lists the description of ever y original variable in this work zone crashes database. Table 4.2 Tables from Florida Traffic Crash Records Database File Name Description Events Contains information about the crash event (i.e. date, time, harmful events, etc.). This is the "parent file of the database Drivers Contains information about each driver involved in the crash demographic and causal). Passengers Contains information about each passenger involved in the crash (demographic and causal) Pedestrians Contains information about each pedestrian involved in the crash (demographic and causal). Property Contains information about property (other than vehicles) damaged in the crash Vehicles Contains information about each vehicle involved in the crash.
45 Table 4.2 (Continued) File Name Description Violations Contains information about citations issued to drivers or pedestrians involved in crashes (limited to the first eight citations issues per party). ComVeh The newest table, contains information about commercial vehicles and car riers involved in crashes. DOT Contains Department of Transportation location and road data. 4.2.2 Data Description For developing the work zone crash injury severity model, 10 variables ( Table 4.3) are selected. The dependent variable is the crash inju ry severity which has 5 levels from no injury to fatal injury at an ascending order The other independent variables can be categorized as 4 classes: environmental condition, roadway condition, drivers condition, and c rash related information Table 4.3 Description of Selected Variables for Model Development Variable Description Type Value Definition ACCISEV Crash Severity Level Ordinal 1 No Injury 2 Possible Injury 3 Non incapacitating injury 4 Incapacitating Injury 5 Fatal Injury Envi ronmental Condition LGHTCOND If the crash occurred un der the good lighting condition (daylight condition) Binary 0 No 1 Yes Roadway Condition CURVE If there is a curve at the crash location Binary 0 No 1 Yes
46 Table 4.3 (Continued) URBAN If the crash occurred in a urban area Binary 0 No 1 Yes MAXSPEED Maximum Posted Speed Limit Continuous SECTADT Section average annual daily traffic Continuous Driver's Condition AGE_AT_FA ULT At fault driver's age Categorical 1 Young (1524) 2 Middle (25 64) 3 Old ( 65) ALDGUSE_ AT_FAULT If at fault driver was under influence of alcohol or drugs Binary 0 No 1 Yes Crash Related VEHTYPE If heavy vehicle ( heavy truck and truck tractor ) was involved Binary 0 No 1 Yes HARMEVN Crash Type Categorical 1 Rear end 2 Angle 3 Sideswipe 4 Other Types Table 4.4 describes the minimum value, maximum value, range, mean, and standard deviation of the two continuous variables. The minimums, maximums, ranges means and standard deviations of the original unnormalized variables can be obtained easily by multiplying the values in Table 4.4 by the appropriate scaling factors (the original maximum values in each variable) The range of AADT in work zone area is very large f rom 300 1 000 289 0045 0 vehicles per day to 000 289 000 289 1 vehicles per day. The minimum speed limit is 15 70 2143 0 miles per hour, the maximum one is 70 70 1 miles per hour, and mean value is 52 70 7455 0 miles per hour.
47 Table 4.4 Description Statistic of Continuous Variables Varibale N Minimum Maximum Range Mean Std. Deviation Scaling Factors SECADT 14217 0.0045 1 0.9955 0.2205 0.1774 289000 MAXSPEED 14217 0.2143 1 0.7857 0.7455 0.15984 70 Table 4.5 illustrates the discrete variables frequency statistic. When the crash injury severity increases, the frequency of crashes decreases. The total percentage of slight injury crashes (ACCISEV = 1, 2, and 3) in work zone area is 90.41%. Incapacitating injur y crash only holds 7.94%, and the fatal crash has the least proportion which is 1.66%. More than one third of work zone crashes (34.17%) occur under the not good lighting condition (nondaylight), and 85.84% of them in the urban area. Only 8.10% of locations where work zone crash happen has curve, 14.62% work zone crashes occur with heavy vehicle involvement, and 5.15% drivers are influenced by drugs or alcohol. The top three crash types here are rear end (37.15%), angle (12.04%), and sideswipe (11.26%). The distribution of at fault drivers age group is 23.61% young age drivers, 66.81% middle age drivers, and 9.59% old age drivers.
48 Table 4.5 Frequencies of Discrete Variables Variable Value Frequency Percent Sample Size 14217 ACCISEV 1 2 3 4 5 6477 3555 2820 1129 236 45.56 25.01 19.83 7.94 1.66 LGHTCOND 0 1 4858 9359 34.17 65.83 CURVE 0 1 13065 1152 91.90 8.10 URBAN 0 1 2013 12204 14.16 85.84 AGE_AT_FAULT 1 2 3 3356 9498 1363 23.60 66.81 9.59 ALDGUSE_AT_FAULT 0 1 13485 732 94.85 5.15 VEHTYPE 0 1 12139 2078 85.38 14.62 HARMEVN 1 2 3 4 5282 1712 1601 5622 37.15 12.04 11.26 39.55
49 4.3 Work Zone Crash Injury Severity Model 4.3.1 Estimation Procedure This section presents the estimation results of the work zone crash severity model for all work z one crashes. At first, cross tabulation analysis i s performed to check the distribution of explanatory variables across injury severity levels and ensure enough observations in each cell. And AGE_AT_FAULT variable was transformed to three dummy variables: YOUNG_AGE ( AGE_AT_FAULT = 1 ), MIDDLE_AGE ( AGE_AT_FAULT = 2 ), and OLD_AGE ( AGE_AT_FAULT = 3 ). Be similar another categorical variable HARMEVN was converted to four dummy variables: REAR END ( HARMEVN =1), ANGLE ( HARMEVN = 2), SIDESWIPE ( HARMEVN = 3), and OT HERS ( HARMEVN = 4). After then the ordinal probit regression model was developed using the O PROBIT procedure in the STATA software package. In the procedure, the s tepwise option was added for selecting independent variables for which the significant level i s greater than 95% The theory of variable selection is: a t first, there was no variable in this ordered probit model, then the variables whose pvalue is less or equal to 0.05 were added into the model one by one.
50 4.3.2 Cross Tabulations between Ex planatory Variables and Crash Severity In order to obtain a better understanding about the selected explanatory variables, cross tabulations of binary or categorical variables with crash severity were developed and given in Tables 4.6. Table 4.6 Cross Tabulation between explanatory Vari ab les and Crash Severity Frequency Row % Value Crash severity Total 1 2 3 4 5 LGHTCOND 0 2131 1118 1013 452 144 4858 43.9% 23.0% 20.9% 9.3% 3.0% 100% 1 4346 2437 1807 677 92 9359 46.4% 26.0% 19.3% 7.2% 1.0% 100% Total 6477 3555 2820 1129 236 14217 45.6% 25.0% 19.8% 7.9% 1.7% 100% CURVE 0 5950 3316 2574 1016 209 13065 45.5% 25.4% 19.7% 7.8% 1.6% 100% 1 527 239 246 113 27 1152 45.7% 20.7% 21.4% 9.8% 2.3% 100% Total 6477 3555 2820 1129 236 14217 45.6% 25.0% 19.8% 7.9% 1.7% 100% URBAN 0 787 408 491 252 75 2013 39.1% 20.3% 24.4% 12.5% 3.7% 100% 1 5690 3147 2329 877 161 12204 46.6% 25.8% 19.1% 7.2% 1.3% 100% Total 6477 3555 2820 1129 236 14217 45.6% 25.0% 19.8% 7.9% 1.7% 100% AGE_AT_ FAULT 1 1391 891 748 267 59 3356 41.4% 26.5% 22.3% 8.0% 1.8% 100% 2 4496 2330 1795 730 147 9498 47.3% 24.5% 18.9% 7.7% 1.5% 100% 3 590 334 277 132 30 1363 43.3% 24.5% 20.3% 9.7% 2.2% 100% Total 6477 3555 2820 1129 236 14217 45.6% 25.0% 19 .8% 7.9% 1.7% 100%
51 Table 4.6 (Continued) Frequency Row % Value Crash severity Total 1 2 3 4 5 HARMEVN 1 2110 1782 1043 314 33 5282 39.9% 33.7% 19.7% 5.9% 0.6% 100% 2 708 410 388 172 34 1712 41.4% 23.9% 22.7% 10.0% 2.0% 100% 3 1183 226 144 43 5 1601 73.9% 14.1% 9.0% 2.7% 0.3% 100% 4 2476 1137 1245 600 164 5622 44.0% 20.2% 22.1% 10.7% 2.9% 100% Total 6477 3555 2820 1129 236 14217 45.6% 25.0% 19.8% 7.9% 1.7% 100% ALDGUSE_AT_FAULT 0 6140 3445 2696 1050 154 13485 45.5% 25.5% 20.0 % 7.8% 1.1% 100% 1 337 110 124 79 82 732 46.0% 15.0% 16.9% 10.8% 11.2% 100% Total 6477 3555 2820 1129 270 14251 45.4% 24.9% 19.8% 7.9% 1.9% 100% VEHTYPE 0 5141 3223 2553 1030 192 12139 42.4% 26.6% 21.0% 8.5% 1.6% 100% 1 1336 332 267 99 44 2078 64.3% 16.0% 12.8% 4.8% 2.1% 100% Total 6477 3555 2820 1129 236 14217 45.6% 25.0% 19.8% 7.9% 1.7% 100% 4.3.3 Estimation Results The estimation of results of the ordinal probit regression is given in Table 4.7. The sample size is 14,217 observ ations, and the Likelihood Ratio (LR) test statistic falls into the rejection area (p value = 0 < 0.05). That means the overall explanatory variables of the model have significant influence on the responses (crash severity levels) at a statistical signif icance level 95% Except for ANGLE all slope coefficients are significant at a confidence level 0.05. Although the p value of ANGLE is little greater
52 than 0.05, the variable was still included in the model since angle crash was an important crash type a nd more variables increase the explanation ability of the model. Table 4.7 Estimation of Ordered Probit Regression for Work Zone Crash Severity Model Ordered probit regression Log likelihood = 17861.331 Number of obs ervation = 14217 LR chi2(12) = 1094.6 Prob > chi2 = 0.000 Pseudo R2 = 0.0297 ACCISEV Coef. Std. Err. z P>z [95% Conf.Interval] LGHTCOND 0.0981 0.0206 4.77 0.000 0.1384 0.0578 CURVE 0.0818 0.0344 2.38 0.018 0.01432 0.149 4 URBAN 0.1768 0.0308 5.74 0.000 0.2372 0.1164 VEHTYPE 0.3846 0.0295 13.02 0.000 0.4425 0.3267 ALDGUSE_AT_FAULT 0.2096 0.0430 4.87 0.000 0.1252 0.2939 YOUNG_AGE 0.0506 0.0224 2.26 0.024 0.0067 0.0945 OLD_AGE 0.1229 0.0326 3.77 0.000 0.0590 0.1867 REAREND 0.0752 0.0217 3.47 0.001 0.1178 0.0327 ANGLE 0.0569 0.030 5 1.87 0.062 0 .002 8 0.1166 SIDESWIPE 0.7253 0.0363 19.98 0.000 0.7964 0.6541 SECADT 0.3851 0.0656 5.87 0.000 0.5136 0.2565 MAXSPEED 0.770 2 0.074 2 10.38 0.000 0.6248 0.9156 /cut point1 /cut point 2 /cut point 3 /c ut point 4 0.0434 0.7261 1.5236 2.3867 0.0677 0.0679 0.0686 0.0722 0.0892 0.1761 0.5931 0.8591 1.3892 1.6579 2.2452 2.5281 Based on the estimated results in Table 4.7, the probability models for five crash injury severity level s are given as:
53 i i ix x y 11 Pr i i i ix x x y 1 22 Pr i i i ix x x y 2 33 Pr (4.22) i i i ix x x y 3 44 Pr i i ix x y 41 5 Pr where is the cutpoint, and is the coefficient of the corresponding variable. 4.3.4 Interpretation The crash severity model estimated by the ordinal probit regression has the same slope coefficients across all severity levels. For example, the coefficient for LGHTCOND is 0.0981 and the standardized coeffici ent for it is 0.0931, which means that the presence of day light ( LGHTCOND = 1) tends to reduce the injury severity of work zone crashes, and when driving in daylight condition, the probability of having a higher injury severity crash is 0.0931 standard d eviations lower than in non daylight condition, holding all other variables constant Table 4.8 and 4.9 shows the estimated results of the partial changes in y and in predicted probabilities for this ordered model respectively. Table 4.8 Partial Change in y* ACCISEV Coef. z P>z y standardized coef. LGHTCOND 0.0981 4.773 0.000 0.0931 CURVE 0.0818 2.38 0.018 0.0777 URBAN 0.1768 5.74 0 .000 0.1678 VEHTYPE 0.3846 13.02 0.000 0.3650 ALDGUSE_AT_FAULT 0.2096 4.87 0.000 0.1989 YOUNG_AGE 0.0506 2.26 0.024 0.0480
54 Table 4.8 (Continued) OLD_AGE 0.1229 3.77 0.000 0.1166 REAREND 0.0752 3.47 0.001 0.0714 ANGLE 0.0569 1.87 0.062 0.0541 SIDESWIPE 0.7253 19.98 0 .000 0.6882 SECADT 0.3851 5.87 0.000 0.3654 MAXSPEED 0.7702 10.38 0.000 0.7308 Table 4.9 Partial Change in Predicted Probabilities No Injury Possible Injury Non incapacitating I njury Incapacitating Injury Fatal Injury LGHTCOND P>z 0.0388 0.000 0.0053 0.000 0.0179 0.000 0.0123 0.000 0.0034 0.000 CURVE P>z 0.0323 0.017 0.0040 0.005 0.0149 0.017 0.0105 0.022 0.0029 0.029 URBAN P>z 0.0694 0.000 0.0074 0.000 0.0318 0.000 0.0234 0.000 0.0068 0.000 VEHTYPE P>z 0.1524 0.000 0.0333 0.000 0.0693 0.000 0.0403 0.000 0.0096 0.000 ALDGUSE_AT_ FAULT 0.0817 0.0072 0.0374 0.0286 0.0086 P>z 0.000 0.000 0.000 0.000 0.000 YOUNG_AGE 0.0200 0.0027 0.0092 0.0063 0.0017 P>z 0.024 0.017 0.024 0.026 0.030 OLD_AGE 0.0483 0.0056 0.0222 0.0160 0.0045 P>z 0.000 0.000 0.000 0.000 0.001 REAREND 0.0299 0.0045 0.0137 0.0092 0.0024 P>z 0.001 0.001 0.001 0.000 0.000 ANGLE 0.0225 0.0030 0.0104 0.0072 0.0020 P>z 0.061 0.039 0.061 0.068 0.076 SIDESWIPE 0.2793 0.0792 0.1231 0.0632 0.0138 P>z 0.000 0.000 0.000 0.000 0.000 SECADT 0.1527 0.0222 0.0703 0.0475 0.0127 P>z 0.000 0.000 0.000 0.000 0.000 MAXSPEED 0.3054 0.0443 0.1406 0.0950 0.0255 P>z 0.000 0.000 0.000 0.000 0.000
55 126.96.36.199. Signs In the T able s 4.8, the variables recording daylight condition, urban area, heavy vehicle involved, rear end crash type, si deswipe crash type and average annual daily traffic ha ve negative coefficients, that means when the value of these variables increase the crash injury severity is more likely to be slight. In contrast, the increase of other variables with positive coeffi cients tends to make a higher probability of more severe injury crashe s The summary is in Table 4.10 Table 4.10 Analysis of the Coefficient Signs Independent Variable Sign Influence for Crash Severity Level LGHTCOND Decrease CURVE + Increase U RBAN Decrease VEHTYPE Decrease ALDGUSE_AT_FAULT + Increase YOUNG_AGE + Increase OLD_AGE + Increase REAREND Decrease ANGLE + Increase SIDESWIPE Decrease SECADT Decrease MAXSPEED + Increase 188.8.131.52. Magnitude of Coefficients The injury severity level y is specified as a linear function of the independent variables, the relative magnitudes of estimated variable coefficients are, in most cases, a measure of the relative impacts of these v ariables on the average severity level of injury
56 severity ( ODonnell and Connor, 1996) For example, the increase in injury severity of an old driver is about 2.43 times higher than the increase in injury severity of a young driver, all other things being equal, because the estimated coefficient of the variable OLD_AGE ( 1229 0 ) is about 2.43 larger than the estimate of the coefficient of the variable YOUNG_AGE ( 0506 0 ). Then, the estimated variable coefficients can be compa red in this way and the influences of different variables on average injury severity level can be ranked (see Table 4.11). Table 4.11 Ranked Magnitude s of Coefficients Rank Independent Variable Coefficient (Pos itive) Independent Variable Coefficient (Negat ive) 1 MAXSPEED 0.7702 SIDESWIPE 0.7253 2 ALDGUSE_AT_ FAULT 0.2096 SECADT 0.3851 3 OLD_AGE 0.1229 VEHTYPE 0.3846 4 CURVE 0.0818 URBAN 0.1768 5 ANGLE 0.0569 LGHTCOND 0.0981 6 YOUNG_AGE 0.0506 REAR END 0.0752 184.108.40.206. D etail ed I nterpretations ( 1) Under good lighting conditions (such as daylight ) the work zone crash severity is more likely to decrease. (2) A curved design at the work zone section s which means the driving condition turns to be difficult, is easily to result in a severe sever cr ash. (3 ) In urban work zone area, the level of crash injury tends to decrease. It may because of the lower driving speed.
57 (4) Heavy vehicle involved can induce to less sever crash es. This is not the same as we think usually. The reason might be that the most people drive carefully when there is a truck around them. (5) Alcohol and drugs tend to increase the crash injury severity level. (6) In two special age groups, young age drivers who are more aggressive and have less experience and old age drivers whose physical, visual, and cognitive abilities may deteriorate are easily involved into severe crashes. But the influence of the old age is more than which of the young age. (7) Two major crash types in work zone area, readend and sideswipe may not contribute directly hurt to drivers, so if these two types of crashes happen, the probability of having injury would decrease. The condition of angle crash type occurring is totally contrary. The impact of the sideswipe crashes is much more than the impact of the rear end crashes (0.7253 / 0.0752 = 9.64). (8) The increase of maximum speed limit tends to increase the crash severity level and the condition is totally contrary to the variable AADT. (9) According to the different magnitudes of estimated variable coeff icients, the increase of maximum posted speed ( 7702 0 ) has the highest impact to increase the crash severity level, which is the 3.67 times higher than the second ranked variable ALDGUSE_AT_FAULT ( 2096 0 ). In contrast, the sideswipe crash type ( 7253 0 ) has the highest impact to reduce the crash severity level, which is the 1.88
58 times higher than the second ranked variables SECADT ( 3851 0 ), and VEHTYPE ( 3846 0 ). 4.3.5 Possi ble Countermeasures to Improve Work Zone Safety Since the explanatory variables are the factors which have significant influence on the crash severity, the countermeasures can be suggested based on the variables in the models. (1 ) Driving in daylight can r educe crash severity level, so a good lighting condition is important for work zone safety, especially during the nighttime periods. When nighttime work is being performed, floodlights should be used to illuminate in work zones, but the disabling glare condition for approaching road users which might be produced should be noticed. (2) Be careful the work zone transition beginning in existing horizontal curve. We can keep continuous curve radii on work zone transitions which can help drivers from overestimating the appropriate speed, resulting in fewer runoff -the road crashes, or move transition upstream so that it does not start in an existing horizontal curve instead. (3) Speed limit is to keep drivers at a constant safe speed in work zones. Several other signs besides regular speed limit sign such as speed feedback signs and changeable message signs with radar (CMR) can be used
59 Speed feedback signs usually measure using radar and display an individual vehicles speed. These signs can only display speed but several have the capability of displaying other text, such as Slow Down. CMR displays warning messages when a vehicle is traveling at an unsafe speed. The standard message on the CMS unit changes when a vehicle is traveling faster than the program med speed, typically 3 mph above the speed limit. The messages used might included: YOU ARE SPEEDING, SLOW DOWN, HIGH SPEED, SLOW DOWN, REDUCE SPEED IN WORK ZONE, and EXCESSIVE SPEED, SLOW DOWN.
60 CHAPTER FIVE SUMMARY 5.1 Summary The main objec tives of this study are to investigate the characteristics of accidents in work zones to identify the factors contributing to injury severity levels and to study how the factor influence injury levels. To achieve this purpose, two different statistics a re processed. One is descriptive statistics and the other ordered regression modeling. Descriptive statistic analysis was used to get the distribution of work zone crashes over three age groups for various factors which were paid attentions by researchers. In this part, crash severity level, environmental conditions, crash types, contributing factors, heavy vehicle involvement, and alcohol/drugs involvement were discussed over age groups, in some characteristics even the distribution between work zone and nonwork zone were compared. The main results are: (1 ) In work zone area, the middle age drivers cause the highest proportion (67%) of crashes, while in non work zone area they have a lower possibility of occurring crashes (63%). (2) Middle age drivers inv olved the highest percentage in the no injury crashes which is 49%, and always has the lowest one in other crashes. While in the more severe level crashes, elderly drivers contribute more than the other two age groups
61 ( 3) Rear end, angle and sideswipe are the principle crash types in all three age groups. In young and middle age groups, the percentage of rear end crashes is obviously higher than angle and sideswipe crashes, and elderly age group shows higher rate in angle crashes than others. Read end and sideswipe crashes are more likely to be occurred in work zone area. ( 4) The most predominant factor for work zone crashes is careless driving, and others are failed to yield right of way, no improper driving action and improper lane change in all age groups. But in elderly age group, the distribution (proportion and rank) has slight difference. In the distribution of predominant contributing factors over the principal crash types, careless driving, failed to yield right of way, and improper lane change are three most predominant contributing factor for rear end, angle, and sideswipe crashes respectively. ( 5) Heavy vehicle is more easily related to work zone crashes (14%) than nonwork zone crashes (7%). Most driver especially old driver is not influenced by alcohol/drugs. Crash severity is an important criterion reflecting the cost of work zone crashes in social and economy, and affected by various factors including drivers characteristics, vehicle characteristics, environmental factors, and roadway features. A full understanding of the impacts of the factors on the crash severity is beneficial to select proper countermeasure for reduc ing the crash severity at work zones and decrease the loss of construction/maintenance on roadway. A probit regression for or dinal output was used to
62 estimate the crash severity models for overall work zone crashes. Based on the results of crash severity modeling and analysis, some conclusions can be obtained: (1 ) According to the ordered probit model for work zone crash severit y, lighting condition, road section with curves, urban or rural area, heavy vehicle involved, alcohol/drug involvement, young and old age group, three predominant crash types, AADT and maximum posted speed have the main influence to work zone crash severit y. (2) The factors of daylight condition, urban, rear end crash type, sideswipe crash type and high average annual daily traffic are more likely to reduce th e severity of work zone crashes. (3) In contrast to the common sense, heavy vehicle involved could induce work zone crash severity. Thats maybe because of driving carefully when there is a truck or tractor around. (4) Based on the magnitudes of the variable coefficients, the variables of maximum posted speed and the sideswipe crash have the major impa ct to crash severity level. That shows restriction to driving speed is principle factor for work zone safety. Based on these statistical analyses for work zone crashes, several countermeasures can be given: (1 ) Floodlights needs to be used to illuminate in work zones in the nighttime in order to build a good lighting condition. (2) Discourage traffic control plan designs that include transition areas for the work
63 zone on an existing horizontal curve, and encourage that the transition be accomplished on a tangent section instead. (3) Speed limit signs are very important for work zone safety. Some dynamic signs like changeable message signs with radar and speed feedback signs have better effectiveness to reduce driver speed.
64 REFERENCES Abdel Aty, M Analysis of driver injury severity levels at multiple locations using ordered probit models. 34 (5), 597 603., 2003. Abdel aty, M., Chen, C., and Schott, J. R. An assessment of the effect of driver age on traffic accident involvement using log linear mo dels. Accident Analysis and Prevention. 30(6) 851 861., 1998. Abdel Aty, M., Keller, J. Exploring the overall and specific crash severity levels at signalized intersections. Accident; analysis and prevention. 37(3), 41725., 2005. Chang, L., & Mannering, F. Analysis of Injury Severity and Vehicle Occupancy in Truck and Nontruck Involved Accidents. A ccident Analysis and Prevention. 31, 579592., 1999. Daniel, J., Dixon, J., and Jared, D. Analysis of fatal crashes in Georgia work zones. Transportation Re search Record 1715. Transportation Research Board, Washington, D .C ., pp. 1823., 2000. Dissanayake, S., and Lu, J., Analysis of Severity of Young Driver Crashes: Sequential Binary Logistic Regression Modeling. In Transportation Research Record: Journal of the Transportation Research Board, No. 1784, TRB, National Research Council, Washington, D.C., pp. 108114., 2002. Donnell, E., and Mason, J. Predicting the Severity of MedianRelated Crashes in Pennsylvania by Using Logistic Regression. Transportation R esearch Record, No. 1897, TRB, National Research Council, Washington, D.C., pp. 5563., 2004. Duncan, C., Khattak, A., & Council, F. Applying the Ordered Probit Model to Injury Severity in TruckPassenger Car Rear End Collisions. Transportation Research R ecord, vol. 1635, pp. 63 71., 1999. Garber, N. J. and Zhao, M. Distribution and Characteristics of Crashes at Different Locations within Work Zones in Virginia. Research Report VTRC 02R12, Virginia Transportation Research Cou ncil, Charlottesville, Virgi nia, 2002.
65 Hutchinson, T. Statistical Modeling of Injury Severity,With Special reference to Driver and Front Seat Passenger in Single Vehicle Crashes. Accident Analysis and Prevention, 18(2), 157 167., 1986. Khattak, A. Effect of Information and Other Factors on Multivehicle Rear end Crashes: Crash Propagation and Injury Severity. Presented at the 78th Annual Meeting of the Transportation Research Board, Washington, D.C., 1999. Khattak, A.J., Khattak, A.J., and Council, F.M. Effects of work zone presence on injury and noninjury crashes, Accident Analysis and Prevention Vol. 34(1), pp. 1929., 2002. Khattak, A., Pawlovich, M., Souleyrette, R., Hallmark, S. Factors related to more severe older driver traffic crash injuries. J. Transportation Eng. 128 ( 3), 243249., 2002. Khattak, A., Schneider, R., Targa, F., Risk factors in large truck rollovers and injury severity: analysis of singlevehicle collisions. In: Presented at the 82nd Annual Meeting of the Transportation Research Board, Washington, DC., 20 03. Khattak, A., Targa, F., Injury severity and total harm in truck involved work zone crashes. In: Presented at the 83rd Annual Meeting of the Transportation Research Board, Washington, DC., 2004. Kim, K., L. Nitz, J. Richardson, and L. Li. Personal and Behavioral Predictors of Automobile Crash and Injury Severity. Crash Analysis and Prevention. Vol. 27, No. 4, pp. 469481., 1995. Klop, J. Factors Influencing Bicycle Crash Severity on TwoLane Undivided Roadways in North Carolina. Presented at the 78th Annual Meeting of the Transportation Research Board, Washington, D.C., 1998. Kockelman, K., Kweon, Y. Driver injury severity: an application of ordered probit models. Accident Analysis and Prevention. 34 (3), 313321(9) ., 2002. Krull, K., Khattak, A., & Council, F. Injury Effects of Rollovers and Events Sequence in Single Vehicle Crashes. Presented at the 80th Annual Meeting of the Transportation Research Board, Washington, D.C., 2000. Maze, T. Kamyab, A. and Schrock, S. Evaluation of Work Zone Speed Re duction Measures. Final Contract Report. Iowa Department of Transportation, IA, 2000.
66 McFadden, J., Yang, W. T., and Durrans, S. R. Application of Artificial Neural Networks to Predict Speeds on TwoLane Rural Highways. Transportation Research Record. No. 1751, Transportation Research Board, National Research Council, Washington D.C., 2005. Mohan, S. B. and Gautam, P. Cost of Highway Work Zone Injuries. Practical Periodical on Structural Design and Construction. Vol. 7, No.2, 2002, pp68 73., 2002. Nas sar, S., Saccomanno, F., and Shortreed, J Road Accident Severity Analysis: A Micro Level Approach. Canadian Journal of Civil Engineering, 21, 847 855., 1994. ODonnell, C., and Connor, D. Predicting the Severity of Motor Vehicle Accident Injuries Using Models of Ordered Multiple Choices. Accident Analysis and Prevention, 28(6), 739 753., 1996. Renski, H., Khattak, A., and Council, F Impact of Speed Limit Increases on Crash Injury Severity: Analysis of Single Vehicle Crashes on North Carolina Interstat e Highways. Presented at the 78th Annual Meeting of the Transportation Research Board, Washington, D.C. 1998. Schrick, S., et al. An analysis of Fatal Work Zone Crashes in Texas. Texas Transportation Institute. No. 040281., 2004. Shankar, V., and Mann ering, F. An Exploratory Multinomial Logit Analysis of Single Vehicle Motorcycle Accident Severity. Journal of Safety Research, 27(3), 183 194., 1996. Shankar, V., Mannering, F., and Barfield, W. Statistical Analysis of Accident Severity on Rural Freeway s. A ccident Analysis and Prevention. 28(3), 391401., 1996. Toshiyuki, Y., and Shankar, V. Bivariate OrderedResponse Probit Model of Drivers and Passengers Injury Severities in Collision with Fixed Objects. Presented at the 81th Annual Meeting of the T ransportation Research Board, Washington, D.C., 2002. Ullman, G., et al. Analysis of Crashes at Active Night Work Zones in Texas Presented at the Annual Meeting of the Transportation Research Board, Washington, D.C., 2005. Wang, J., Hughes, W.E., Council F.M., and Paniati, J.F. Investigation of highway work zone crashes: what we know and what we dont know, Transportation Research Record 1529, pp. 5462., 1996.
67 BIBLIOGRAPHY FHWA. Manual on Uniform Traffic Control Devices for Streets and Highways. 2003 Edition, Federal Highway Administration. U.S. Department of Transportation, 2003. Greene H. Econometric Analysis. Second Edition, 1993. Long. J. Regression Models for Categorical and Limited Dependent Dependent Variables, 1997. Long, J. and Freese J. Regression Models for Categorical Dependent Variables Using Stata. Second Edition, 2006. Powers, D. and Xie, Y. Statistical Methods for Categorical Data Analysis, 2000 Transportation Research Board, Highway Capacity Manual, 4th edition, National Res earch Council, Washington, D.C., 2000. Washington, S., Karlaftis, M., and Mannering, F. Statistical and Econometric Method for Transportation Data Analysis, 2003
69 Appe ndix A: Variables and Codes of Work Zone Crash Tab le A 1 Variable of Work Zone Crashes Variable Description Type YEAR The year of work zone fatal crash Nominal TIME The time of work zone fatal crash Nominal AGE The age of driver at fault Ordinal VEHMOVEMENT The movement of vehicle at fault before acci dent Nominal CRASHTYPE The type of crash Nominal VEHICLETYPE Heavy vehicle involved? Nominal FUNCLASS The function of roads Nominal TRWAYCHR Road Characteristics (level / curve?) Nominal MAXSPEED The speed limit Continue SECTADT The AADT of the secti on of work zones Continue TYPESUR The type of road surface Nominal SITELOCA Site Location Nominal LIGHTCONDITION Light condition Nominal WEATHERCONDITION Weather condition Nominal ROADSURFACE Road surface condition Nominal VISION Vision Obstructed No minal RDACCESS Access control type Nominal SURWIDTH The width of roads Continue CONTRIBUTINGFACTORS The contributing factors Nominal TRAFCONT Traffic Control Nominal Table A 2 Codes for TIME Codes Description 1 6:0010:00 2 10:0016:00 3 16:0020:00 4 20:006:00
70 Appe ndix A (Continued) Table A 3 Codes for AGE Codes Description 1 <19 2 20 24 3 2534 4 3544 5 4554 6 55 64 7 >65 Table A 4 Codes for VEHMOVEMENT Codes Description 01 STRAIGHT AHEAD 02 SLOWING/STOPPED/STALLED 03 MA KING LEFT TURN 04 BACKING 05 MAKING RIGHT TURN 06 CHANGING LANES 07 ENTERING/LEAVING PARKING SPACE 08 PROPERLY PARKED 09 IMPROPERLY PARKED 10 MAKING U TURN 11 PASSING 12 DRIVERLESS OR RUNAWAY VEH. 77 ALL OTHERS 88 UNKNOWN
71 Appe ndix A (Continued) Table A 5 Codes for CRASHTYPE Codes Description 01 COLL. W/MV IN TRANS. REAR END 02 COLL. W/MV IN TRANS. HEAD ON 03 COLL. W/MV IN TRANS. ANGLE 04 COLL. W/MV IN TRANS. LFT TURN 05 COLL. W/MV IN TRANS. RGT TURN 06 COLL. W/MV IN TRANS. SIDESWIP 07 COLL. W/MV IN TRANS. BAKD INTO 08 COLL. W/PARKED CAR 09 COLLISION WITH MV ON ROADWAY 10 COLL. W/ PEDESTRIAN 11 COLL. W/ BICYCLE 12 COLL. W/ BICYCLE (BIKE LANE) 13 COLL. W/ MOPED 14 COLL. W/ TRAIN 15 COLL. W/ ANIMAL 16 MV HIT SIGN/SIGN POST 17 MV HIT UTILITY POLE/LIGHT POLE 18 MV HIT GUARDRAIL 19 MV HIT FENCE 20 MV HIT CONCRETE BARRIER WALL 21 MV HIT BRDGE/PIER/ABUTMNT/RAIL 22 MV HIT TREE/SHRUBBERY 23 COLL. W/CONSTRCTN BARRICDE/SGN 24 COLL. W/TRAFFIC GATE 25 COLL. W/CRASH ATTENUATORS 26 COLL. W/FIXED OBJCT ABOVE ROAD 27 MV HIT OTHER FIXED OBJECT 28 COLL. W/MOVEABLE OBJCT ON ROAD 29 MV RAN INTO DITCH/CULVERT 30 RAN OFF ROAD INTO WATER 31 OVERTURNED 32 OCCUPANT FELL FROM VEHICLE 33 TRACTOR/TRAILER JACKNIFED 34 FIRE 35 EXPLOSION 36 DOWNHILL RUNAWAY 37 CARGO LOSS OR SHIFT
72 Appe ndix A (Continued) Table A 5 (Continued) 38 SEPARATION OF UNITS 39 MEDIAN CROSSOVER 77 ALL OTHER (EXPLAIN) Table A 6 Codes for VEHICLETYPE Codes Description 00 UNKNOWN/NOT CODED 01 AUTOMOBILE 02 PAS SENGER VAN 03 PICKUP/LIGHT TRUCK (2 REAR TIR) 04 MEDIUM TRUCK (4 REAR TIRES) 05 HEAVY TRUCK (2 OR MORE REAR AX) 06 TRUCK TRACTOR (CAB) 07 MOTOR HOME (RV) 08 BUS (DRIVER + 9 15 PASS) 09 BUS (DRIVER + > 15 PASS) 10 BICYCLE 11 MOTORCYCLE 12 MOPED 13 ALL TERRAIN VEHICLE 14 TRAIN 15 LOW SPEED VEHICLE 77 OTHER 88 PEDESTRIAN NO VEHICLE Table A 7 Codes for TRWAYCHR Codes Description 1 STRAIGHT LEVEL 2 STRAIGHT UPGRADE/DOWNGRADE 3 CURVE LEVEL 4 CURVE UPGRADE/DOWNGRADE
73 Appe ndix A (Continued) Table A 8 Codes for TYPESUR Codes Description 01 SLAG/GRAVEL/STONE 02 BLACKTOP 03 BRICK/BLOCK 04 CONCRETE 05 DIRT 77 ALL OTHER Table A 9 Codes for SITELOCA Codes Description 01 NOT AT INTERSECTION/RRX/BRIDGE 02 AT INTERSECTION 03 INFLUENCED BY INTERSECTION 04 DRIVEWAY ACCESS 05 RAILROAD CROSSING 06 BRIDGE 07 ENTRANCE RAMP 08 EXIT RAMP 09 PARKING LOT/TRAFFIC WAY 10 PARKING LOT AISLE OR STALL 11 PRIVATE PROPERTY 12 TOLL BOOTH 13 PUBLIC BUS STOP ZONE 77 ALL OTHER Table A 10 Codes for LIGHTCONDITION Codes Description 01 DAYLIGHT 02 DUSK 03 DAWN 04 DARK (STREET LIGHT) 05 DARK (NO STREET LIGHT) 88 UNKNOWN
74 Appe ndix A (Continued) Table A 11 Codes for WEATHERCONDITION Codes Description 01 CLEAR 02 CLOUDY 03 RAIN 04 FOG 77 ALL OTHER 88 UNKNOWN Table A 12 Codes for ROADSURFACE Codes Description 01 DRY 02 WET 03 SLIPPERY 04 ICY 77 ALL OTHER 88 UNKNOWN Table A 13 Codes for VISION Codes Description 01 VISION NOT OBSCURED 02 INCLEMENT WEATHER 03 PARKED/STOPPED VEH ICLE 04 TREES/CROPS/BUSHES 05 LOAD ON VEHICLE 06 BUILDING/FIXED OBJECT 07 SIGNS/BILLBOARDS 08 FOG 09 SMOKE 10 GLARE 77 ALL OTHER (EXPLAIN)
75 Appe ndix A (Continued) Table A 14 Codes for RDACCESS Codes Description 1 FULL 2 PARTIAL 3 NONE Tab le A 15 Codes for CONTRIBUTINGFACTORS Codes Description 01 NO IMPROPER DRIVING/ACTION 02 CARELESS DRIVING 03 FAILED TO YEILD RIGHT OF WAY 04 IMPROPER BACKING 05 IMPROPER LANE CHANGE 06 IMPROPER TURN 07 ALCOHOL UNDER INFLUENCE 08 DRUGS UNDER INFLUE NCE 09 ALCOHOL DRUGS UNDER INFLUENCE 10 FOLLOWED TOO CLOSELY 11 DISREGARDED TRAFFIC SIGNAL 12 EXCEEDED SAFE SPEED LIMIT 13 DISREGARDED STOP SIGN 14 FAILED TO MAINTAIN EQUIP/VEHIC 15 IMPROPER PASSING 16 DROVE LEFT OF CENTER 17 EXCEEDED STATED SPEED LIMIT 18 OBSTRUCTING TRAFFIC 19 IMPROPER LOAD 20 DISREGARDED OTHER TRAFFIC CONT 21 DRIVING WRONG SIDE/WAY 22 FLEEING POLICE 23 VEHICLE MODIFIED 24 DRIVER DISTRACTION 77 ALL OTHER (EXPLAIN)
76 Appe ndix A (Continued) Table A 16 Codes for TRAFCON T Codes Description 01 NO CONTROL 02 SPECIAL SPEED ZONE 03 SPEED CONTROL SIGN 04 SCHOOL ZONE 05 TRAFFIC SIGNAL 06 STOP SIGN 07 YIELD SIGN 08 FLASHING LIGHT 09 RAILROAD SIGNAL 10 OFFICER/GUARD/FLAGMAN 11 POSTED NO UTURN 12 NO PASSING ZONE 77 A LL OTHER