|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 Ka
controlfield tag 001 001983336
007 cr mnu|||uuuuu
008 081230s2008 flu sbm 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0002338
Safety evaluation of freeway exit ramps
h [electronic resource] /
by Hongyun Chen.
[Tampa, Fla.] :
b University of South Florida,
ABSTRACT: The primary objective of the study is to evaluate safety performances of different exit ramps used in Florida and nationally. More specific, the research objectives include the following two parts: (1) to evaluate the impacts of different exit ramp types on safety performance for freeway diverge areas; and (2) to identify the different factors contributing to the crashes happening on the exit ramp sections. To achieve the research objectives, the research team investigated crash history at 424 sites throughout Florida. The study area includes two parts, the freeway diverge area and the exit ramp sections. For the freeway diverge areas, exit ramp types were defined based on the number of lanes used by vehicular traffic to exit freeways.^ Four exit ramp types were considered here including single-lane exit ramps (Type 1), sing-lane exit ramps without a taper (Type 2), two-lane exit ramps with an optional lane (Type 3), and two-lane exit ramps without an optional lane (Type 4). For the exit ramp sections, four ramp configurations, including diamond, out connection, free-flow loop and parclo loop, were considered. Cross-sectional comparisons were conducted to compare crash frequency, crash rate, crash severity and crash types between different exit ramp groups. Crash predictive models were also built to quantify the impacts of various contributing factors. On the freeway diverge areas, it shows that Type 1 exit ramp has the best safety performance in terms of the lowest crash frequency and crash rate.^ The crash prediction model shows that for one-lane exit ramp, replacing a Type 1 with a Type 2 will increase crash counts at freeway diverge areas by 15.57% while replacing a Type 3 with a Type 4 will increase crash counts by 10.80% for two-lane ramps. On the exit ramp sections, the out connection ramps appear to have the lowest average crash rate than the other three. The crash predictive model shows that replacing an out connection exit ramp with a diamond, free-flow, and parclo loop will increase crashes counts by 26.90%, 68.47% and 48.72% respectively. The results of this study will help transportation decision makers develop tailored technical guidelines governing the selection of the optimum design combinations on freeway diverge areas and exit ramp sections.
Thesis (M.S.C.E.)--University of South Florida, 2008.
Includes bibliographical references.
Text (Electronic thesis) in PDF format.
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
Title from PDF of title page.
Document formatted into pages; contains 112 pages.
Co-adviser: Pan Liu, Ph.D.
Co-adviser: Jian Lu., Ph.D.
Generalized regression model.
x Civil Engineering
t USF Electronic Theses and Dissertations.
Safety Evaluation of Freeway Exit Ramps by Hongyun Chen A thesis submitted in partial fulfillment of the requirements for the degree of Master of Science in Civil Engineering Department of Civil and Environmental Engineering College of Engineering University of South Florida Co-Major Professor: Pan Liu, Ph.D. Co-Major Professor: Jian Lu., Ph.D. Pei-sung Lin, Ph.D. Edward Mierzejewski, Ph.D. Date of Approval: March 5, 2008 Keywords: Lane Balance, Exit Ramp, Ramp Configuration, Hypothesis Test, Generalized Regression Model Copyright 2008, Hongyun Chen
DEDICATION This work is dedicated by my dearest parents, Zongxiang Chen and Jinfang Xie.
ACKNOWLEDGEMENTS I would like to thank Dr Pan Liu and Dr Jian Lu, the co-major profess ors who have helped me a lot in the completion of the thesis and the academic pr ogram. I also wish to thank Dr Pei-sung Lin and Dr Edward Mierzejewski, who have patiently guided me through the thesis process. This thesis is part of the research proj ect sponsored by the Florida Department of Transportation. The statistical offices o f FDOT are greatly appreciated for providing the important data of the project. I would like to thank the Graduate Research Assistants at the Department of Civil a nd Environmental Engineering of the University of South Florida as well for their assistances in field data c ollections.
i TABLE OF CONTENTS LIST OF TABLES v LIST OF FIGURES viii ABTRACT x CHAPTER ONE INTRODUCTION 1 1.1 Background 1 1.2 Research Subject 4 1.2.1 Freeway Diverge Areas 4 1.2.2 Entire Exit Ramp Sections 9 1.3 Research Objectives 12 1.4 Research Approach 12 1.5 Research Tasks 13 1.6 Thesis Outline 14
ii CHAPTER TWO LITERATURE REVIEW 16 2.1 General Freeway and Ramp Guidelines 16 2.2 Freeway Diverge Areas 17 2.3 Exit Ramp Section 23 CHAPTER THREE METHODOLOGY 28 3.1 Crash Frequency and Crash Rate 28 3.1.1 Crash Frequency 28 3.1.2 Crash Rate 30 3.2 Crash Type and Crash Severity 31 3.2.1 Crash Type 32 3.2.2 Crash Severity 33 3.3 Cross-Sectional Comparison Approach 33 3.4 The Hypothesis Test 34 3.4.1 Hypothesis Tests on the Equality of Two Means 36 3.4.2 Hypothesis Tests on Proportionality 37 3.5 Statistical Predictive Model 38 CHAPTER FOUR DATA COLLECTION 42 4.1 Site Selection Criteria 42 4.2 Segment Length Definition 44 4.2.1 Freeway Diverge Length 45 4.2.2 Exit Ramp Length 48
iii 4.3 Selected Sites Information 52 4.3.1 Freeway Diverge Areas 55 4.3.2 Exit Ramp Sections 56 4.4 Site Selection Procedures 56 4.4.1 Site Selection Procedure 1 56 4.4.2 Site Selection Procedure 2 57 4.4.3 Site Selection Procedure 3 58 4.5 Section Number, Milepost and Site Identification Number 58 4.6 Crash Database 59 4.7 Combination of Crash Data with Site Information 60 CHAPTER FIVE DATA ANALYSIS 61 5.1 Outline of Data Analysis 61 5.2 Freeway Diverge Areas 62 5.2.1 Comparison of Average Crash Frequency and Crash Rate 62 5.2.2 Comparison of Target Crash Types 68 5.2.3 Comparison of Crash Severity 73 5.2.4 Crash Predictive Model 76 5.3 Exit Ramp Sections 82 5.3.1 Comparison of Average Crash Frequency and Crash Rate 82 5.3.2 Comparison of Target Crash Types 87 5.3.3 Comparison of Crash Severity 91 5.3.4 Crash Predictive Model 93
iv CHAPTER SIX SUMMARY, CONCLUSION, AND RECOMMENDATION 100 6.1 Summary 100 6.2 Conclusions 102 6.2.1 Freeway Diverge Areas 102 6.2.2 Freeway Exit Ramp Sections 103 6.3 Applications and Recommendations 104 6.3.1 Applications 104 6.3.2 Recommendations 105 REFERENCES 107 APPENDICES 111 Appendix A: Site Pictures Examples 112
v LIST OF TABLES Table 1. FDOT Districts Distributions for Selected Sample Sites 54 Table 2. Sites Resource Distributions for Freeway Diverge Areas 5 5 Table 3. Summary of Average Crash Frequency and Crash Rate for Four Exit Ramp Types 65 Table 4. Summary Hypotheses Tests of Average Crash Frequency and Crash Rate for Four Exit Ramp Types 68 Table 5. Summary of Average Crash Numbers by Target Crash Types for Four Exit Ramp Types 69 Table 6. Summary of Average Crash Rates by Target Crash Types for Four Exit Ramp Types 70 Table 7. Z Statistics for Proportionality Tests of Three Target Crash Typ es for Four Exit Ramp Types 72 Table 8. Summary of Average Crash Numbers by Crash Severity for Four Exit Ramp Types 73 Table 9. Summary of Average Crash Rates by Crash Severity for Four Exit Ramp Types 74
vi Table 10. Z Statistics for Proportionality Tests by Crash Severity for Four Exit Ramp Types 76 Table 11. Description of Initially Considered Independent Variables on Freeway Diverge Areas 77 Table 12. Regression Results for Crash Prediction Model on Diverge Areas 79 Table 13. Summary of Average Crash Frequency and Crash Rate for Four Exit Ramp Configurations 84 Table 14. Summary Hypotheses Tests for Average Crash Frequency and Crash Rate for Four Exit Ramp Types 86 Table 15. Summary of Average Crash Numbers by Target Crash Types for Four Exit Ramp Configurations 88 Table 16. Summary of Average Crash Rates by Target Crash Types for Four Exit Ramp Configurations 88 Table 17. Z Statistics for Proportionality Tests of Target Crash Types for Four Exit Ramp Configurations 90 Table 18. Summary of Average Crash Numbers by Crash Severity for Four Exit Ramp Configurations 91 Table 19. Summary of Average Crash Rates by Crash Severity for Four Exit Ramp Configurations 92 Table 20. Z Statistics for Proportionality Tests by Crash Severity for Four Exit Ramp Configurations 93 Table 21. Description of Initially Considered Independent Variables on Exit Ramp Sections 95
vii Table 22. Regression Results for Crash Prediction Model on Exit Ramp Sections 96
viii LIST OF FIGURES Figure 1. Type 1 Exit Ramp: Parallel from a Tangent Single-lane Exit Ra mp 6 Figure 2. Type 2 Exit Ramp: Single-lane Exit Ramp without a Taper 7 Figure 3. Type 3 Exit Ramp: Two-lane Exit Ramp with an Optional Lane 7 Figure 4. Type 4 Exit Ramp: Two-lane Exit Ramp without an Optional Lane 8 Figure 5. Typical Four Exit Ramp Configurations 10 Figure 6. Type 1 Exit Ramp Section Length: Parallel from a Tangent Sing-lane Exit Ramp 46 Figure 7. Type 2 Exit Ramp Section Length: Sing-lane Exit Ramp without a Taper 46 Figure 8. Type 3 Exit Ramp Section Length: Two-lane Exit Ramp with an Optional Lane 47 Figure 9. Type 4 Exit Ramp Section Length: Two-lane Exit Ramp without an Optional Lane 47 Figure 10. Exit Ramp Segment Lengths for Four Ramp Configurations 49 Figure 11. Florida Interstate Highway System 53 Figure 12. Florida District Map 53 Figure 13. SPSS Example Format from FDOT Crash Database 60
ix Figure 14. Example of Combining Database 60 Figure 15. Comparison of Average Crash Frequency among Four Exit Ramp Types 64 Figure 16. Comparison of Average Crash Rate among Four Exit Ramp Types 64 Figure 17. Site Picture for I-95 Southbound Exit 74 66 Figure 18. Comparison of Percentages by Target Crash Types for Four Exit Ramp Types 71 Figure 19. Comparison of Percentages by Crash Severity for Four Exit Ramp Types 75
x SAFETY EVALUATION OF FREEWAY EXIT RAMPS Hongyun Chen ABSTRACT The primary objective of the study is to evaluate safety perf ormances of different exit ramps used in Florida and nationally. More specific, the rese arch objectives include the following two parts: (1) to evaluate the impacts of differ ent exit ramp types on safety performance for freeway diverge areas; and (2) to identify t he different factors contributing to the crashes happening on the exit ramp sections. To ac hieve the research objectives, the research team investigated crash history at 424 si tes throughout Florida. The study area includes two parts, the freeway diverge area a nd the exit ramp sections. For the freeway diverge areas, exit ramp types were defined based on t he number of lanes used by vehicular traffic to exit freeways. Four exit ramp t ypes were considered here including single-lane exit ramps (Type 1), sing-lane exit ramps w ithout a taper (Type 2), two-lane exit ramps with an optional lane (Type 3), and two-lane ex it ramps without an optional lane (Type 4). For the exit ramp sections, four ramp config urations, including diamond, out connection, free-flow loop and parclo loop, were considered. Cross-sectional comparisons were conducted to compare crash frequency crash rate, crash severity and crash types between different exit ra mp groups. Crash predictive models were also built to quantify the impacts of various contributing factors. On the
xi freeway diverge areas, it shows that Type 1 exit ramp has the best safety performance in terms of the lowest crash frequency and crash rate. The crash prediction model shows that for one-lane exit ramp, replacing a Type 1 with a Type 2 will increase c rash counts at freeway diverge areas by 15.57% while replacing a Type 3 with a Type 4 will increase crash counts by 10.80% for two-lane ramps. On the exit ramp sections the out connection ramps appear to have the lowest average crash rate than the other three. The crash predictive model shows that replacing an out connection exit ramp with a diamond, free-flow, and parclo loop will increase crashes counts by 26.90%, 68.47% and 48.72% respectively. The results of this study will help transportation decision makers develop tailored technical guidelines governing the selection of the optimum design combinations on freeway diverge areas and exit ramp sections.
1 CHAPTER ONE INTRODUCTION 1.1 Background Freeways play important roles in the highway system around the count ry. In the United States, the interstate highway system, which composes les s than 2% of the total urban highway mileage, carries more than 20% of the traffic b y the end of 2006. Freeways provide the specific traffic facility which allows the traffic run smoothly in the roadway network at the highest level. They are constructed accor ding to the highest highway design standards and regulated public movements by full control s of traffic elements such as capacity, post speed, geometrics fundamentals, and level of se rvice. Exit ramps are the only control accesses used for traffic ex iting freeways. They also serve as transitions from freeways to secondary crossroads which could be freeways, major or minor arterials, or local streets. The design of freew ay exit ramps could significantly impact the safety and operation performances on fre eways, exit ramps and crossroads. The AASHTO Green Book (A Policy on the Design of Ge ometric Highways and Streets) (12) mentioned that complex design components make ramps vary from simple to comprehensive layouts so that each ramp site should be st udied and planned carefully. Freeway diverge areas are the specific segments that divide the freeway traffic exiting from or continuing on the freeway mainlines. Freeways conne ct with exit ramps
2 by several different diverge types called exit ramp types in this study. These types cause different results of safety performances on the freeway diverg e areas by different ways. Exit ramp section is another important concern in this study. Exit ramps provide limitaccesses from freeways to other freeways, lower-speed arter ials or local streets. A few factors, such as geometrics, traffics, and local conditions, have diff erent relationships with crashes. These facts include more than deceleration distance s, exit ramp lengths, design speeds, operating speeds, speed differences, exit ramp confi gurations, or road conditions. Better understanding the relationships among them would help i mprove the safety, efficiency, mobility, accessibility, and accommodation aspe cts for both freeway diverge areas and exit ramp sections. Ramp Management and Control Ha ndbook(14), published by U.S. Department of Transportation (DOT) and Federal Hi ghway Administration (FHWA) in 2004, aims to manage ramp policies, strate gies and technologies as to improve safety on the exit ramp and the influent ial areas. Ramp management strategies control the flow vehicles exiting a fre eway not only on the exit ramps, but also on the freeway neighboring areas. A before and afte r evaluation of ramp crashes in Minneapolis found that the number of peak period crashes on fre eways and ramps increased 26% when there was no ramp control strategy in 2001. This case revealed the reality that resolutions to the deficiencies on the f reeway diverge areas and exit ramp sections can help to improve safety. Successful managements on the two research segments, freeway di verge areas and exit ramp sections, could obtain benefits on society, economics and cu ltures and gain satisfactions on safety improvements. However, the impacts of exi t ramp types on the safety performance of freeway diverge areas have not been we ll studied or documented
3 until recently. Few have focused on the impacts of the types of e xit ramps concerning the lane balance problems such as the number of lanes used by traffic to exit freeways. The details of the relationship between the lane balance and safety a re not well understood. Since the limit work that has been performed, a few tentative concl usions might to be drawn. It can assume that potential improvements will lead to fewe r crashes, thus enhance safety on the freeway diverge areas. On the exit ramp s ections, the various influential factors on the safety performance at entire exi t ramp sections need to be revised and re-conducted since previous studies have a few limitations. For example, some predictive crash models concerned different ramp configurations and ramp length, however the control types of ramp terminals did not contain in these m odels (3). Some models combined the off ramps and on ramps. The combination might ignore t he dissimilar operating factors between the two different kinds of ramps. Several types of exit ramps are used for traffic to exit freeways on the diverge areas. The increasing vehicular crashes in freeway diverge areas lift up the need to select the best exit ramp designs to improve safety on freeway diverge ar eas. The problem is relatively new and highly demanded in todays highway system. Fo r the exit ramp sections, little focus has been put on the safety issues in the Sta te of Florida. So this study would conduct comprehensive crash comparisons and analyses on freeway e xit ramp sections for the whole state. The results of two research par ts, freeway diverge areas and exit ramp sections in this study, will help transportation decisi on makers develop tailored technical guidelines governing the selection of the optimum exit ra mp types and combinations of related factors to be used on our freeway diverge ar eas and exit ramp sections.
4 1.2 Research Subject On the freeway diverge areas, the most commonly used freeway exi t ramps include two-lane exit ramps with an optional lane, two-lane exit ra mps without optional lane, single-lane exit ramps with widening to two lanes on the ramp beyond the exit gore, and three basic number of through lanes changed to two through lanes with one lane reduced and designated as the exit lane. Drivers exiting a fr eeway must decrease vehicle speeds and weave to the deceleration lane toward the entrance of the exit ramp. Different types of exit ramps require drivers to make distinctive decisions to complete related maneuvers both for exiting and continuing with the freeway. As a res ult, different exit ramp design may impact the safety and operational performance of f reeway diverge areas in different ways. On the exit ramp sections, different ramp confi gurations such as diamond, out connection, free flow, and parclo flow and other factors such as widening lanes, pavement paintings, and terminal controls might confuse drivers a s well. These mixed influential features on the exit ramp cause existing proble ms and situations more multifaceted. This study processes to quantitatively evaluate the safety features of two issues. 1.2.1 Freeway Diverge Areas None of the studies for the past two decades focused on the lane bal ance problems on the freeway diverge area which directly connects the mainline seg ment to exit ramps. AASHTO Green book defines the lane balance as the number of appro ach lanes on the highway after the exit should equal to the number of lanes on the high way beyond the exit, plus the number of lanes on the exit, minus one. The fundamental arr angement of a
5 freeway segment is the designation of the basic number of lanes w hich should be consistency along the freeway. The basic number of lanes might be added or deleted where the traffic volumes increase or decreased at some degr ees. On the freeway diverge area, part traffic on the freeways beyond the exits leave the freeway and so that the volumes change in this segment. The one or two outer lanes may drop to t he exit lanes so that the number of lanes on the freeway mainline sections did not bala nce ahead of or after the exits. This would not only cause confusions for the exiti ng vehicles but also for the continuing vehicles on the freeways. The lane-balanced and unbalanc ed exit ramps require drivers take different maneuvers. Even considering the lane balanced exit ramps or the unbalanced exit ramps respectively, different numbers of exi t lanes on the freeway segments have different characteristics as well. The study would focus on the lane balance issues which are innovated and original in the freeway exit ramps st udies. The exit ramp type is defined by the number of lanes used for traffic to exit freeways. They could be single-lane exit ramps or two-lane exi t ramps. After reviewing the sites in the whole Florida interstate highway systems, expr essways, turnpikes and parkways, four types are used frequently for the state. So four diff erent groups based on the types of exit ramps are characterized for the study. For c onvenience, they were set as Type 1 exit ramps (Type 1), Type 2 exit ramps (Type 2), Type 3 exit ramps (Type 3) and Type 4 exit ramps (Type 4) respectively. The definitions of ea ch type of exit ramp are described below and illustrated in Figure 1 through Figure 4 below. 1) Type 1 exit ramp Parallel from a tangent single-lane exit ramp shown in Figure 1: It is a full width parallel from tangent that leads to ei ther a tangent or flat exiting curve which includes a decelerating taper. The horizontal and vertica l alignment of
6 type 1 exit tamps were based on the selected design speed equal or less than the intersecting roadways. No direct drop lanes on the mainline section s beyond or after exits. The outer lane with a tangent would be a drop lane to the ex its and become the though lane on the exit ramp section. 2) Type 2 exit ramp Single-lane exit ramp without a taper shown in Figure 2: This type is when the outer lane becomes a drop lane at the exit gore forming a lane reduction. A paved and striped area beyond the theoretical gore were pre sent at this type of exit ramps to provide a maneuver and recovery area. No addi tional lane was added when compared with Type 1. 3) Type 3 exit ramp Two-lane exit with an optional lane shown in Figure 3: This type includes two exit lanes while a large percentage of traffic volum e on the freeway beyond the painted nose would leave at this particular exit. An auxili ary lane to develop the full capacity of two lane exit was developed for 1500 feet The entire operations in this type of exit ramps took place over a significant length of the freeway in most cases. The outer one of the two exit lanes direc tly drops to the exit ramps. But the inner lane of the two exit lanes, which is an opti onal lane, has two alternatives by continuing on the freeway or running off the freeways. 4) Type 4 exit ramp Two-lane exit without an optional lane is shown in Figure 4: It is used where one of the through lanes, the outer lane, is reduced and anoth er full width parallel from tangent lane developed with a taper is also force d to exit. It differs as from Type 3 exit ramps as Type 4 exit ramps do not enclose the optional lane.
7 From the figures, they indicate that Type 1 and Type 3 are l ane balanced ones while Type 2 and Type 4 are lane unbalanced exit ramps. In practice, there is a type 5 exit ramp which is a two-lane exit ramp without optional lane and without a taper, which is not widely used in Florida and the samples we found are too small to draw defensible conclusions. FIGURE 1. Type 1 Exit Ramp: Parallel from a Tangent Single-lane Exit Ramp
8 FIGURE 2. Type 2 Exit Ramp: Single-lane Exit Ramp without a Taper FIGURE 3. Type 3 Exit Ramp: Two-lane Exit Ramp with an Optional Lane
9 FIGURE 4. Type 4 Exit Ramp: Two-lane Exit Ramp without an Optional Lane 1.2.2 Entire Exit Ramp Sections The entire exit ramp section from the beginning of pointed nose, which diverge the freeways and ramps, to the end of ramp terminal is another re search concern. This study is to acquire an adaptable, practical, and integral transition system from the freeway to the secondary crossroad. Ramp designing contains many possible influential factors such as ramp configurations, ramp design speed, lane numbers, ramp term inal control types, ramp length, or ramp curvatures. Ramp configurations are usually considered as the ramp types in t he previous studies. Bauer and Harwoods (3) analyses show that diverse ramp conf iguration designs have significantly dissimilar impacts on the safety performanc e especially for off ramps. Typically various configurations accommodate to the ramp sites by the features of site locations. In order to clearly indicate the safety performance with related parameters, the ramp configuration was considered one of them. Four widely used configurat ions in Florida are identified in the study. They were briefly defined a s diamond exit ramps, out
10 connection exit ramps, free-flow loop exit ramps and parclo loop exit ramps. From Fi gure 5-A to Figure-D illustrate the four ramp configurations which des cribe the shape of ramps in simplified modes. Figure 5-A is a diamond exit ramp which is a one-way road with both left and right turnings at terminals. Figure 5-B is an out connection exit ramp which only supplies the single turn at the ends of exit ramps. Figure 5-C and 5-D are two classic loop ramps that make at le ast 270 degrees of turning movements to the secondary roads. Free-flow loop ramps are de signed as full cloverleaf ramps with or without collector or distributor roads on the ramp segments. The parclo loop exit ramp is a partial cloverleaf ramp which has a pr eference to provide an arrangement setting the right exiting vehicles. This configura tion could give either one or two turning ways at the exit terminals while the exit ramps location meets the requirements to provide enough design radii, space, curvatures and relat ed geometric criteria.
11 Freeway EXIT Freeway EXIT Freeway EXIT Figure 5-A. Diamond Exit Ramps Figure 5-B. Outer Connection Exit Ramps Figure 5-C Free-flow Loop Exit Ramps Figure 5-D Parclo Loop Exit Ramps FIGURE 5. Typical Four Exit Ramp Configurations Freeway EXIT
12 1.3 Research Objectives The objective of the study is to evaluate safety performances of different exit ramps used in Florida and nationals. The research objectives can divide into two parts. The first one is to evaluate how the impacts of different exit ramp types on the safety performance of freeway diverge areas. The second one focuses on ident ifying the different factors contributing to the crashes happening on the ex it ramp sections. This study developed quantitative evaluations and comparisons on the freeway di verge areas and exit ramp sections accordingly. Statistical analyses among four types of exit ramps on the freeway diverge areas, parallel from a tangentsingle-lane exit ramp, single-lane exit ramp without a taper, twolane exit ramp with an optional lane and two-lane exit ramp without an optional lane, are conducted. The four different ramp configurations and other parameter s on the entire exit ramp sections are examined as well to find their effects on the safety features for the entire exit ramps. Base on the result in this study, it would be a way to judge wha t kind of geometric, traffic, and combinations of the correlated conditions have t he best safety performance on the freeway diverge sections and entire exit r amp sections. This is also a practical step to guide the methods of safety improvements on freew ay diverge areas and exit ramp sections. The results could also be applied in design guide lines, handbooks or research projects. 1.4 Research Approach Previous studies were revised and potential safety measurements f or this study were selected. Crash histories at selected freeway segment s were investigated and crash
13 data were collected. Cross-sectional comparisons were conducted to compare the safety impacts the two segments of freeway diverge areas and exit r amp sections respectively. On the basis of the collected crash data for the diverge areas, st atistical analyses were conducted to quantitatively evaluate the impacts of different types of exit ramps on the safety performance of freeway diverge areas and different r amp configurations on exit ramp sections. In addition, crash prediction models were developed to ide ntify the factors that contribute to crashes at selected sites. The results of thi s study will help transportation decision makers develop tailored technical guidelines gove rning the selection of the optimum exit ramp to be used on our freeways and r ecommend the optimal design characteristics both on the diverge areas and the entire exit ra mps. 1.5 Research Tasks In order to achieve research purposes, following tasks were made to obtain rational conclusions. Existing methods and technologies were gathered t o reach the goals of two research subjects. Possible applications were identified t he in the research fields. After summarizing these potential measurements, useful method from previous studies were selected and detailed developments were conducted for this study These methods and developments need to be feasible to perform and practice. The ana lysis process should be correct and reasonable. The results base on this study can be applied to other exit ramp managements. In this study, four steps containing ten mai n tasks were categorized to well organize the research procedures as following: 1) Step 1: Task 1: Literature Search and Review; Task 2: Field Observation;
14 Task 3: Field Operation Plan; 2) Step 2: Task 4: Site Selection; Task 5: Field Data Collection; Task 6: Data Reduction; 3) Step 3: Task 7: Data Analysis; Task 8: Research Results; 4) Step 4: Task 9: Conclusions and discussions; Task 10: Final Report. Step 1, classifying the first three tasks, mainly focused on goin g over the past safety performance measurements and methods, discovering the possibil ity of the potential applications, viewing sites, building up study purposes and arr anging work plans. Step 2, from task 4 to task 6, gathered the site data and arran ged them to do the further analysis. This step is a very tough and tedious one since the study needs large sample sizes to get reasonable results and all the related data need to be f ound at available methods. The third step applied the main approaches to conduct safety e valuations procedures. The final step concluded the research findings and summari zed the whole research study in the final report in the thesis. These four ste ps contained all the needed tasks for this research study and have been proved successfully in past project s. 1.6 Thesis Outline This thesis contains six chapters, one reference part and one appendix s ection at all. Chapter 1 provides an overview and the research objective for the s tudy. Chapter 2 presents a brief description of previous study and related topics for the research subjects
15 in order to acquire an advanced study. Chapter 3 summarizes the te chniques applied in this project, which included a detailed description of the proposed methods and basic concepts using in data analysis procedure. Chapter 4 describes the procedures of site data collection and reduction. Chapter 5 presents the procedures of crash anal ysis, results of crash investigation and impacts of selected variables. The final chapter, Chapter 6, emphasizes the summaries, conclusions and recommendations from this s tudy to assist other agencies, public works, engineers better understanding the safet y issues of the freeway diverge areas and entire exit ramp sections. The list of references follows the final chapter and one appendix lists the sample site photos for the research subjects to illustrate different exit ramps applied in the State of Florida.
16 CHAPTER TWO LITERATURE REVIEW This chapter summarizes previous studies and findings related to th e project. Two parts, diverge areas on the freeway mainline sections and entire exit ramp sections, consist of the study subjects are discussed respectively to desc ribe the integral evaluations of the previous discoveries in the research field. 2.1 General Freeway Guidelines Freeways provide the primary transportation networks and roadway sy stems by achieving the highest functional hierarchy of highway systems by design purposes. The grand reliance on the facilities requires safer and more eff icient implements on existing freeways and their related infrastructure systems to improve the safety p erformances. The AASHTO Green Book (A Policy on the Design of Geometric Hig hways and Streets) (12) designs the key requirements on the highway faci lities such as the ramps, interchanges and frontage roads. In order to accommodate high traffi c demands of safety on freeways, exit ramps and secondary crossroads, designing proper ha ndlings of freeways and ramps are essential in the highway systems. M any factors impacts safety performances on freeways and their adjacent facilities. Also, the crash is a direct index on safety evaluations. The wide variety of site geometric conditions traffic volumes,
17 highway types, and design layouts could eliminate or increase confli ct points at some degrees while crashes related to conflict points at some levels. During the past several decades, some design regulations mentioned the importance of safety performance of freeway diverge areas and exit ramp sections. Current state and national literature reviews include freeway and ramp management handbooks, guidelines of optimal geometric designs from Highway Capacity Manual and AASHTO, reports from National Cooperative Highway Research Pro gram (NCHRP) and Different State Departments of Transportation, proceedings from Transportation Symposium, papers from transportation engineering journal, etc. Additiona lly, useful books and publications were also collected to do analysis in the project and current rules, regulations, standards, and practices in Florida were evaluated a nd summarized for the two research subjects in the sequent sections. 2.2 Freeway Diverge Areas During the past several decades, though some studies have mentioned the freeway exit ramps, none of them focused on the impacts of the number of lanes us ed by traffic to exit freeways. Closely reviewed the literature, there is l ittle direct paper or evaluation in safety performance of diverge areas which has been researche d before. In previous studies, ramp types are usually defined by ramp configurations s uch as diamond, loop, directional, outer connector, and other instead of the lane balance is sues for the diverge sections. Though many design handbooks and guidelines focused on the relations hips of geometric elements and collision causes, they did not mention the inf luence of lane balances on the freeway diverge areas.
18 In 1969, Cirillo et al. (9) did a purely innovative investigation on the tra ffic crash study on the interstate system for that period. They found that the r elationship could be established between fatality crashes and geometric elements. T he geometric factor included several types of interchanges, paved shoulders, sight dist ance, delineators, surface types, and other variables. After about thirty years, Gar ber and Fontaine (7) developed a guideline given name as Guidelines for Preliminary Se lection of the Optimum Interchange Type for a Specific Location to search the operational and safety characteristics for the optimal ramp design. The newest instr uction is the ITE Freeway and Interchange Geometric Design Handbook edited by Joel (17) in 2006. The handbook focuses on geometric and operational characteristics of freew ays and interchanges. The book recognized that geometric design procedures for freeways and interchanges may vary. It also provides the evidence that is va lued as an accompaniment of the AASHTO Greenbook (12), the Highway Capacity Manual (HCM) (13), and Traffic engineering Handbook 5 th Edition (16). In 1998, Bared et al. (1) developed a generalized regression model known as Poisson Model to estimate the crash frequency for the decelera tion lanes plus the entire ramps as a function of ramp AADT, mainline freeway AADT, dec eleration lane length and ramp configurations. The ramp configurations considered in that stud y include diamond, parclo loop, free-flow loop, and outer connecter. The model showed that t he crash frequency on freeway ramps increased with the ramp and fre eway AADT and decreased with the increase of the deceleration lane length. A 100 ft increase in deceleration lane length will result in a 4.8% reduction in cras h frequency. The coefficients of the model also indicated that off-ramps suffered f rom more crashes as
19 comparing to on-ramps. However, this study did not consider the number of l anes using for traffic leaving freeways. This problem is essential in t he driving behavior because the balanced lanes and unbalanced lanes require drivers to take different operating manne rs. Later, Bauer and Harwood (3) built up several regression models to determine the relationships between traffic accidents, highway geometric desig n elements and traffic volumes. The statistical modeling approaches used in the research i ncluded Poisson and Negative Binomial regressions. It was found that the ramp AADT e xplained most of the variability in the crash data report at selected sites. Other variables found to be significant in crash prediction models contained freeway AADT, area type (rura l, urban), ramp type (on, off), ramp configurations, and lengths of ramp and speed-change la ne (deceleration lanes, acceleration lanes). Other models have been built to find out the functions of different variables in different kind of models. The independent variable s are crash frequencies on the speed change lanes, entire ramp sections, the selected ramp sections, and speed change sections plus the entire ramp sections. The best fi t model was the one that combined crash frequency for the entire ramp, together with it s adjacent speedchange lanes. The significant influential factors included area type, ramp type, ramp configurations (diamond, loop, outer connector, others), length of speed-chang e lanes, and length of the entire ramps. Another main finding is that models for the total crashes achieved much better than those for the only fatal and injury crashes The models combined the on ramps and off ramps, and acceleration lanes and decelera tions lanes. Off ramps usually occur more crashes than on ramps as mentioned before; the requirements for the length, curve, and design guidelines of acceleration length and deceleration lanes vary; ramp configurations could not be the ramp types on the diverge are as. Without
20 judging these factors, models would decrease the accuracy of the conclusions, narrow the applications of the results and could not disclose the real situations But this study provided reasonable methods such as the regression models which have been pr oved strappingly employed in the safety studies (11, 19, 25, 26, 27, 28, and 31). One main program is called Highway Safety Improvement Program (18) that can help states decrease the number of crashes and provide optimal ways for arranging, applying, and estimating safety plans. From side to side of the intr oduction, all correlated issues to improve highway safety are recognized, measured, imple mented and evaluated highway planning, designs, constructions, maintenances, and operations. Moreover, past studies emphasized the safety evaluation based on previous mentioned me thods such as regression models or statistical tests that have been proved as useful methods in t he safety studies. Following paragraph listed the wide applications of these methods. Sarhan et al. (11) designed the approach to help achieving the optimum pre dictive models. The model related to the length of acceleration and decelera tion lanes based on expected collision frequency. Joanne and Sayed (25) undertook the study to quantify the relationship between the design consistencies on the roadway safety The generalized linear regression approach is used for model development as a quanti tative tool for evaluating the impact of design consistency on road safety. Garcia et al. (19) analyzed different deceleration lengths as functions of exit trajectory t ypes, speeds, and localization. Munoz and Daganzo (26) predicted the queued length at a wave speed about 13 mph in congested traffic by KW model. This method is widely use d in the safety evaluation of intersections as well as freeway sections. Maze et al. (27) analyzed the TWSC expressway intersection for crash rates, crash severity rates and fatal crash rates
21 by Poisson regression models. Keller et al. (28) divided crashes by different types as angle, left-turn, head-on, rear-end and pedestrian/bicycle by linea r regression models while speed limits were found to be important. Bernhard et al. (31) ra nked the locations and the estimated benefits of improvement by assigning fatal, i njury and PDO crashes. Hypothesis tests were conducted with normal distribution with high number of crashes and Poisson distribution with a low number of crashes. The statistica l tests were usually employed to find crash-prone sites in identifying some sites as hazardous at some a particular level of confidence. In fact, the level of confidence i s that 100% minus the Type I error. Type I error is the percentage that mistakes the safety sites for hazardous sites. Another Type II error is the percentage that mistakes the hazardous sites for safety sites. They concluded that the program would benefit to public traffi c to make the possible efforts in order to improve the safety studies. Other studies focus on revealing the geometric, traffic, or relat ed influential values to the mainline sections separately. Rakha and Zhang (20) model ed a total of 34 different weaving sections to estimate the traffic volume at weaving sections including merge and diverge areas at the appropriate boundaries on freeways. The paper demonstrated that the volume estimated by the model had a significa nt effect on drivers behavior in the mainline weaving sections. Abdel-Aty et al. (22) test ed various speed limits to evaluate the safety improvement on a section of Inters tate 4 in Orlando, FL. Real-time crash likelihood was calculated based on split models for predicting multivehicle crashes during high-speed and low-speed conditions. The improvement was proved in the case of rising medium-to-high-speed regimes on the fre eway. The paper recommends that the speed limit changes upstream and downstream should be large in
22 magnitude (15mph) and implemented within short distances (2miles) of the diverge locations. It makes obvious that speed limit have some specific eff ects on the collisions from the upstream to downstream of diverge areas on the freeways. C assidy et al. (24) noticed the problem that queuing from the segment's off-ramp spilling over and occupying its mandatory exit lane comes up frequently. The situa tion delayed the mainline vehicles as well and would increase weaving conflicts. J anson (8) examined the relationship of ramp designs and truck accident rates in Washingt on State plus a comparison to limited data from Colorado and California. The paper grouped freeway truck accidents by ramp type, crash type, and four conflict area s of each diverge ramp. The crash data were compared for these groups on the basis of number of truck crashes per location and per truck-mile of travel. The conclusion is slight di fferent from generally belief that a ramp with a lower accident rate per truck trip due to low truck volumes may still be a high-risk site. But these results could not represent t he real conditions if applied to all the passenger cars. The higher crashes number might stil l be constant with high volume since truck volume is really low and have the specific feats itself. One research study, concerning on the number of lanes used by traff ic exiting freeways was conducted by Batenhorst (10). The paper, Operational Analysis of Terminating Freeway Auxiliary Lanes with One-Lane and Two-L ane Exit Ramps: A Case Study, used three simulation software packages, the Highway Capacity Software (HCS), CORSIM and Simtraffic on the operational analysis of wea ving area at twenty locations by the level-of-service. The range of traffic and geometr ic conditions among the twenty sites varied. The findings of the case study suggest tha t a one-lane exit ramp may afford the best traffic operations apart from weaving length. T he experience gained from
23 the case study is to give support to traffic engineers to desig n efficient freeway facilities and to help researchers understanding the operational effects of geom etric design. Even though this study considered exit lane numbers on the freeway diverge areas, the better level-ofservice could not necessarily stand for better safety performance, and these two might have opposite results in some cases. Based on the studies mentioned before, the impacts of exit ramp ty pes on the safety performance of freeway diverge areas have not been we ll studied or documented until recently. Several previous studies have evaluated the safet y impacts of different ramp configurations such as diamond, loop, directional, outer connector, and other. However, these studies have not considered the lane balanced problems on t he diverge areas to regulate the number of lanes that shall be used for traffic to exit fr eeways. 2.3 Exit Ramp Section The entire exit ramp section is another concern in this study to provide a comprehensive evaluation of the safety performance on freeway e xits. Ramps are all oneway roads with one or more legs at terminals to connecting seconda ry crossroads. Different involvements of design speeds, configurations, speed differe nces among freeway and ramp section, ramp lengths or the direct connection feat ures determine different exit ramps which have dissimilar safety effects. S ome studies have focused on exit ramp sections and prior conclusions were described below. Lord and Bonneson (2) calibrated predictive models for different ra mp configurations at 44 selected sites. The ramp design configura tions addressed in this study included diagonal ramps, non-free-flow loop ramps, free-flow loop ramps, and
24 outer connection ramps. The non-free-flow (parclo flow loop) ramp experi enced twice as many accidents as other types of ramps Bauer and Harwood (3) as mentioned before modeled the Negative Binomial regression model on the entire ramp se ction as well and concluded that diamond ramp have slight less crash frequency comparing to other ramp types when other influential variables remain constant. At the same year, Khorashadi (4) used another method known as ANOVA test to forecast the relationship among ramp configurations, geometry parameters and crash frequencies. This s tudy found that the geometric elements had much weaker impacts than the ramp configura tions. McCartt et al. (6) examined 1,150 crashes occurring on heavily traveled urban inters tate ramps in Northern Virginia. The three major common crash types, run-off-road, rear-end, and sideswipe, accounted for 95% of total crashes. The countermeasures me ntioned in the study included increasing ramp design speed, increasing curve ra dii, installing surveillance systems such as detectors, cameras, and advanced message signs Abdel-Aty and Huang (21) explored an origin-destination survey to cust omers on the central Floridas expressway system. The distance tra veled to exit a ramp did not depend only on the spacing between ramps, but also on other factors, such as the trip purpose, vehicle occupancy, drivers income level, and E-Pass impleme ntation when the vehicle was equipped with an electronic toll collection system. A m ain finding was that the guide signs beyond the expressway exits had an important impact not only on unfamiliar travelers but also on the experienced drivers. Though i t was a little countintuitive, the result shows different design features on diverge are as would have an effect on familiar drivers as well. Hunter et al. (23) conducted field obse rvations on speed relationships between ramps and freeways by videotaping. Notable con clusions were
25 drawn that vehicle speeds on exit ramps were much higher than the post speed limit. Since the big difference between the ramp post speed limit and oper ating speed, some unfamiliar drivers might slow down the speed while some familiar drivers mi ght enter the exit ramp at a high speed relative far above the limit speed. That might be a vital reason why rear-end crashes account a large percent of crashes in the ramp secti ons. Some studies focused on the connections between different influential f actors which could be the ramp volumes, configurations, crashes, curvatures, and so on. These studies comprised Newells (29) Delays caused by a queue at a freeway exit ramp, Shaw and Mcshanes (30) Optimal Ramp Control for Incident Respons e, and Hunter et al.s (34) Summary Report of Reevaluation of Ramp Design Speed Cr iteria. Newell clarified that the graphical solution is more clearly illustrat ing practical issues. Shaw and Mcshane attended to optimize some measurements on the crashes to minimize the crash disruption. Hunter et al.s concluded that ramp design speed should larger than 50% of freeway speed. This conclusion accommodated to Hunter et al.s (23) r esult that operating speed on the exit ramp is higher than the post speed limit. It is obvious that many studies defined ramp configurations as ramp types. The conclusions included that free-flow ramps have more crashes than others increasing ramp volume might increase crashes, the post speed limit on the ra mp has some impacts on both local/familiar drivers or unfamiliar drivers and the operati ng speed is usually much higher than the post speed. Even several useful results are made on the exit ramp sections, but few consider the following two issues in the safety effects, ramp terminal treatments and ramp lane changing named widening on the exit ramp s egment. Widening in this study is defined as the number of lanes changing after t he pointed nose or in the
26 middle of the entire ramp. The definition of ramp terminal treatme nts in Ramp Management and Control Handbook is those can be implemented at ramp/art erial connections as to better manage traffic exiting the ramp facil ity. They normally solve the specific problems that occur at the ramps or arterials. Diverse terminal control strategies have the potentials to affect operations on the exit ramp and adjac ent arterials. Ramp terminal treatments implemented at exit ramps could reduce queue spillback from the secondary roads, decrease the potential for collisions on the freeway at the back of the queue, and improve traffic flow and safety on or near ramp facilitie s. Typically four strategies are broadly employed, signal timing improvements, r amp channelization, geometric improvements, and signing or pavement markings improvements. The advantages of using ramp terminal strategies are to better coordinate with ramp terminal signal timing, to offer sufficient storage spac e either for left turn or right turn vehicles and to accommodate consistently on both exit ramps and sec ondary crossroads. The method of signal timing adjustments aims to prevent que ue spillback to the freeway facility beyond exit ramps. Ramp channelization ca n increase capacity, supply enough storage space or a separate lane adjacent to the broad -spectrum lane, and delineate separate traffic movements. Geometric improvements ma nage sight distances, horizontal and vertical curves, and any other geometric deficiencies Signing and pavement marking improvements deal with guiding motorists of downward conditions and facilitating vehicle movements. Implementations of ramp termin al treatments reducing delay and queuing length, decreasing conflict points, enhancing sa fety and minimizing impact both on upstream and downstream highways and arteri als. The functions vary by implemented treatments. Alternatively, negative impacts with different
27 terminal treatments varied by the each site. Those might i ncrease trip length, cause supplementary travel time, or extend queuing and signal delay. Accor dingly, different terminal control designs or different combinations of terminal desig ns might have various powers on the safety aspects of entire ramp sections. Retting et al. (32) endeavored to reduce urban crash rate by building potential countermeasures to t he five most common crash types in fourteen cities. For the vast combinations of the cr ashes about (69%-81%) in each type via dissimilar cities, the author suggested that signal timing, sign visibility, sight distances would be the improvement measure to enhance safety in general soluti ons. This study would consider the terminal control methods to expose the i mpacts of terminal control types on safety. One study conducted by Bared et al. (5) comparing crashes between single point and tight diamond ramps related cras hes on the cross road only. Single point diamond interchange is diamond ramp free-connects to the cross roads No triangle median occurs at the terminals. Tight diamond intercha nge differs to single point diamond interchange since there is a triangle median separat ion at the termination to split different traffic movements for left turns or right t urns. Crash data were subtracted from 27 tight diamond sites and 13 single point sites in W ashington to build a Negative Binomial model of total crashes on the exit ramp and cros s-road flow. However, the safety comparison did not reveal a significant difference be tween the two types of interchanges for total crash. This study only compared one termi nal treatment as ramp channelization; however the sites number here is not sufficient enoug h to do a regression model. The lanes widening is another issue as one of the strategi es in the exit ramps. Several ramps from the field observations show that it will wide to two or more lanes after the pointed noses which separate the freeway mainline sections and ramp sec tions.
28 CHAPTER THREE METHODOLOGY This chapter describes the selected methodologies which would be appl ied in this research study. The principles for selecting the main methods c oncern on how the functions are, whether they are practical or easily applied t o the data base, and what the potential results are. The research subjects included two parts defining as freeway diverge areas and entire ramp sections separately. After reviewing pr ior studies, guidelines, handbooks and related researches, useful methodologies and important param eters are identified for the safety analysis. The main approaches used include d the cross-sectional comparison method, hypothesis tests, and generalized regression models. 3.1 Crash Frequency and Crash Rate Crash frequencies or crash rates are two indicators that ar e generally used in the safety studies to compare different treatments or groups. This r esearch project would calculate both of them for further analysis. 3.1.1Crash Frequency Crash frequency is the real number of crashes that have happened at a certain location or segment in a particular time or time interval. It i s commonly used for several
29 benefits. Firstly, the crash data are easy to get and simple t o calculate. Next, the meaning behind is straightforward so that governmental officials, engineers and public could understand it readily. The third virtue is that it could represent diverse selected places in one parameter and could change directly while the selected lengt hs or vicinity of the segments changed. The resource of the noticed crashes is only from police long form crash report which describes specific features for each crash. F lorida Traffic Crash Analysis Report (CAR System) provides detailed crashes and upda tes the database each year. The mathematics mean value of crash frequency is labeled as the average number of crashes. With different groups or managements, the average number of crashes was calculated based on the number of sample sites. In statistical a ssumption, the mean value normally is the most proficient estimator for the population groups. The following equation defines the average crash number with a specific group, C, as: N c Cn i i==1 (1) Where, C =average number of crashes for the sites with a particular group; i c = number of crashes at site i in the group; N = total number of sites within the group. For the diverge areas, four exit ramp types are classified so that four groups were chosen to compare the mean values of crash frequency. Besides, three a dditional values stand for the accuracy and variations of the mean values. The media n value is the middle rate in a series of data that have been ranked in order to sc ale and part the sites into two
30 identical fractions. The maximal and minimum values are the large st and smallest crash number in a specific group. The four additional variables imply the variation of the each sample and the mean values. If the median value is much larger or smaller than mean value, the distribution curves of crash number indicate biasness in the judgment. In order to get reasonable mean value, usually the four values, mean, median, m aximum, and minimal are calculated respectively to represent the distributions of the num ber of crashes. 3.1.2 Crash Rate In this study, crash rate is defined as crashes per milli on vehicles per vehicle miles traveled for a specific section. Crash rates are used a s a criterion for more truthful for segments under the same geometric and traffic conditions to nar row the impacts of these important factors. The crash rate, r, for a particular freeway segment can be calculated in the following formula: L V T A r = 365 000 000 ,1 (2) Where, r = crash rate at a freeway segment (crashes per million vehicles per m ile); A= number of report crashes (crashes per year), T= number of years; V= average daily traffic volume (vehicles per day); L= length of the freeway segment (miles). It is believed that the crash frequency tends to increase as the average daily traffic (ADT) goes up even through many other factors affecting the si tuation. In this study, the corresponding ADT for each site was obtained from annual Florida tr affic information
31 CDs. The time frame is determined for the database in continuous ye ars when site characters have not been changed in the period. The average crash r ates, which are the arithmetic means of crash rates, were calculated for the four groups in the freeway diverge areas. The statistical assumption is similar to the a verage number of crash as mentioned before. The average crash rate, R, is defined as: N r R n i i= = 1 (3) Where, R =average number of crashes rates with a particular group; i r = number of crashes rates at segment i in the group; N = total number of sites within the group. The median, maximal, and minimal values are measured as well to observe the distributions of crash rates. 3.2 Crash Type and Crash Severity Since the objectives are to estimate the safety impacts among 4 exit ramps on diverge area and along the entire exit ramp sections, the total number of crash, crash severity, and crash types having the highest percentages to the tota l crashes were chosen for each group. Crash severity that is widely used in the safety analysis can be classified to two categories: PDO (Property-damage-only) and injury/fatal crashes.
32 3.2.1 Crash Type In the crash database maintained by FDOT, crash type is defi ned by the first harmful event of at-fault vehicles. The comparison of crash types will help to identify driver behaviors that are related with the types of exit ramps A total number of 40 crash types are concluded in the Floridas CAR system. The most thre e highest crash types occur on diverge areas are rear-end crash, side-swipe crash and an gle crashes. Rear-end crash and side-swipe crash counted for about 60% of total crashes, 46% rear-end crashes and 16% side-swipe crashes. The target crash types on the exit r amp sections are rear-end crash, angle crash and side-swipe crash as well. Rear-end crashes which regularly take place while the first vehicle stopped or suddenly slowed down and the following vehicle had a collision with the fi rst vehicle in the rear piece of the vehicle. The severity of these crashes ca n range from minor to severe depending on the speed of the following vehicle that hits the first vehicle. Sideswipe crash is another common crash type in this study and usual ly happens when changing lanes, misdirection of exiting freeway, or vehicle weaving. The severity of this type is also ranged from minor to severe. The one vehicle crossing the passageway or changing directions i n the road might conflict with another vehicle. They are frequently set as angle crashes. Angle crashes are also commonly noticed on the misdirected vehicles. The severity of the crashes usually causes severe crashes than rear-end crashes. Comparing to other ty pes, the three types mentioned above is the most concerned types in this diverge area and exit ramp sect ions
33 3.2.2 Crash Severity Usually, crash severity level is recorded for each police reporte d crash. Three major levels of crash severity generally defined in the study c an be classified to three categories: 1) Property-damage-only (PDO) crashes; 2) Injury crashes; 3) Fatal crashes; In a property-damage-only crash, only properties are damaged but no person is hurt; in an injury crash, at least one person is lightly hurt becaus e of the crash; in a fatality crash, at least one person is dead within 90 days afte r the crash which was the most concerned problems in many other studies and this study as well. 3.3 Cross-Sectional Comparison Approach The cross-sectional comparison analysis is satisfactory to provide adequate and reasonable consequences. It is long believed that cross-sectional approa ch is a logical and efficient technique of judging the safety effects. The cross -sectional method has been proved valuable and has been performed on a number of prior studies that invol ved median alternatives, right turns followed by u-turn to direct le ft turns and truck accidents at freeway ramps. In transportation fields, traffic engineers have experimental judgments as long as the most influential factors such as section length, a verage daily traffic (ADT), speed, ramp length are well controlled. Cross-sectional analyses to evaluate different treatments are fairly reliable for the results. Briefly, reliable conclusions could be got within this measurement. In other words, this method compares the sa fety of two
34 different groups of sites with and without the treatment under investig ation. It is necessary to select similar geometric conditions in order to ge t the reliable results in comparing site histories of different types. In this study, cross-sectional comparison was conducted to measure freeway diverge areas with different types of exit ramps and exit ram p sections with four configurations. This approach involves comparing crash frequency, crash r ate, crash type, and crash severity of a group with a treatment, to that of a group of with other treated sites. As mentioned before, the selected freeway segments were divided into four groups based on the types of the exit ramps. On the basis of the collecte d crash data, statistical analysis was conducted to quantitatively evaluate the safety impac ts of different types of freeway exit ramps. The major assumption behind this comparison was that all other charac teristics in the sites remained the same during the study period. The significa nt geometric and control factors considered in this study included deceleration lengt h, ramp length, average daily traffic(ADT), posted speed limit, number of lanes i n the freeway, surface conditions, shoulder conditions and so on. By comparing crash through statist ical testing, conclusions could be reached regarding the relative safe treatment among different treatments. 3.4 The Hypotheses Test Hypothesis tests are utilized to test whether the observed diff erences of the selected variables such as mean values, variance values, or proport ion values between two or more groups have significantly variation in a statistical term. Assumptions of
35 observing the sample data were calculated in the hypothesis test ing to measure the suppositions whether they have under similar features. If the result s did not support the assumptions, then the assumed suppositions are considered doubtful. The formula of hypothesis testing includes two competing statistical hypothes es: a null hypothesis (H 0 ) and an alternative hypothesis (H a ). The null hypothesis is a postulation that one parameter of a population is true under sufficient statistical terms. The c ontrast postulation of the null hypothesis is an alternative hypothesis. It is assumed that a ll the other situations that did not covered by the situations under null hypothesis. The test result is to reject or fail to reject the null hy pothesis under the specific conditions based on the statistical distributions while they reply upon Z, t, F or 2 distribution. The decision of whether rejecting the null hypothesis is based on the statistic value range on the statistical distribution mentioned before at a s tatistical term named as the significant level Typically the level of confidence as 1is applied to determine the statistical confidence instead of The procedures of conducting a hypothesis test including four steps: 1) Step 1: Select Null HypothesisH 0 Select an Alternative Hypothesis H a ; 2) Step 2: Determine the level of confidence (1)*100%; 3) Step 3: Calculate the statistical value; 4) Step 4: Compare the statistical value to the critical value on the distribution, and decide to reject or fail to reject the null hypothesis H 0 ; The following two parts describe the detailed procedures to conduct hy pothesis tests on the equality of two means and the proportionality analysis.
36 3.4.1 Hypotheses on the Equality of Two Means Mean values of two different populations were tested to get reasonabl e conclusions whether to reject or not reject the null hypothesis. Th e average crash numbers and rash rates from one group to another group were examined i f they are significantly different. Assumed that two populations say X 1 and X 2 where X 1 has an unknown mean 1 and known variance 1 2 and X 2 has an unknown mean 2 and known variance 2 2 The purpose is to test whether the two populations have the same mean 1 and 2 The first step is to build the null hypothesis H 0 and an alternative hypothesis H a : 2 1 0: m m = H (4) 2 1: m m aH (5) The procedure is based on the fact that the difference in the sa mple mean, X 1 X 2 of two populations of interest with a sample size of n 1 and a sample size of n 2 separately, 2 1 _X X will fit the normal distribution of: 2 1 _X X -~N ( 1 2 1 2 /n 1 + 2 2 / n 2 ) (6) The second step is to choose the level of confidence. In this study 90% is used and equals 10%. The third step is to calculate the statistical value Z 0 (n 25) or 0 t (n<25): 2 2 2 1 2 1 2 1 0 n n X X Zs s+ = (7) ) 1 1 ( 2 1 2 2 1 0 n n s X X t p + = (8)
37 The final step is to compare the calculated value with the crit ical value Z /2 or 2/ t The null hypothesis could be rejected if: Z 0 > Z /2 or Z 0 < Z /2 (9) 0 t > 2/ t or 0 t < 2/ t (10) If the variance 2 is unknown, it can be replaced by the square of the standard devia tion of the sample size n which is S 2 as following: n X X S i n ) ( 1 2 @ (11) If the sample sizes is less or equal to 25, the populations are app roximately t distribution with a pooled variance, 2 p s based on sample variance 2 1 s and 2 2 s The formula is given by: 2 )1 ( )1 ( 2 1 2 2 2 2 1 1 2 + + n n s n s n S p (12) 3.4.2 Hypotheses Tests on the Proportionality Analysis On the basis of the collected crash data, statistical analysi s was conducted to quantitatively evaluate the crash type and crash severity on the s afety effects. The proportionality hypothesis test was utilized in this study to compar ing target crash types and crash severity between different freeways diverge sections. Proportionality test is often used to test the significance of the percentages between two populations or samples. Let p 1 and p 2 be the proportions of a particular type of crashes in two different groups. Assuming that the total crash counts in these two groups are m and n respectively, for testing the null hypothesis:
38 H 0 : p 1 = p 2 (13) Versus H 1 : p 1 p 2, (14) H 0 can be rejected if: 2/ 1 1 2 2 1 2 ) 1( ) 1(aZ n p p m p p p p Z + = (15) 3.5 Statistical Predictive Model Crash prediction models were developed for this study at selecte d freeway segments and entire ramp sections respectively. The purpose to use r egression predictive models is to identify the factors that contribute to the crashes and quantify the effects on crashes at selected sites. This research project would draw on t he generalized linear regression models to mold crash number. Generalized linear models have been widely used for modeling crashe s at safety studies (1, 3, 11, 19, 25, 26, 27, 28, and 31) at intersections, roadways or freeways Generalized linear models are the expansion forms of the classic al linear regression models. The classical linear regression model assumes that the dependent variable is continuous and normally distributed with a constant variance. The assumpti on is not appropriate for crash data which are approximately Poisson distribute d and are generally non-negative, random and discrete in nature. Numerous previous studies have s uggested the use of Poisson models or Negative-Binomial (NB) Models for modeling crash da ta (1, 3). The Poisson model assumes that the dependent variable is Poisson dis tributed. Using
39 a Poisson model, the probability that a particular freeway segme nt i or an exit ramp section experiences y i crashes during a fixed time period is given by: ) ( ) ( i y i i i i y e y p y Y pimm= = = i =1, 2, 3, , n (16) Where, i = the expected number of crashes for segment i; y i = the probability that a particular segment i. A logarithm link function connects to a linear predictor The link function and the linear predictor determine the functional forms of the crash pre diction model. If the linear predictor is a linear function of the explanatory variables, the fitted c rash prediction model takes the functional form as below: ) ... exp( 2 2 1 1 0 ik k i i i x x xb b b b m+ + + + = (17) Where, 0 1 , k = coefficients of explanatory variables; x i1 x i2 x ik = explanatory variables. If the linear predictor is a linear function of the logarithm of the explanatory variables, the functional form is given below: kik i i i x x xb b bb m...2 12 1 0 = (18) The Poisson model assumes that the mean of the crash counts equals the variance. The assumption is usually too stringent considering the fact that t he variance is often greater than the mean. In this condition, overdispersion will be observe d and the estimated coefficients of the Poisson model are biased. An altern ative to deal with the over dispersed data is to use the negative binomial model. The negative binomial model
40 assumes that the crash counts are Poisson-gamma distributed. The pro bability density function of Poisson-gamma structure is given by: 1 1 1 1 ) ( ) ( ) ( 1 1 + + G + G = = a i y i i i i i a a a a y a y y Y p im m m, i =1, 2, 3 n (19) Where y i = the crash count at segment i, i = the expected number of crashes for segment i, = the dispersion parameter. The dispersion parameter determines the variance of the Poisson-gam ma distribution. Usually can be estimated either by the Moment Method or by the Maximum Likelihood Method. Two parameters are often used for evaluating the goodness-of-fi t of a generalized linear model. These two parameters are the scaled deviance (SD ) and the Pearsons 2 statistic. For an adequate model, the two statistics should be chisquare distributed with ( N p ) degrees of freedom, where N is the number of observations and p is the number of parameters in the model. The scaled deviance equals twice the di fference between the log-likelihood under the maximum model and the log-likelihood under the reduced model. The scaled deviance can be calculated as: )) log( ) (log( 2 s L L SD =b (20) Where L s = the likelihood under the maximum model; L = the likelihood under the reduced model. The Pearsons 2 statistic can be calculated as:
41 2 1 2' = =n i i i iy s Pearsons m c (21) where y i = the crash count at segment i, i = the expected number of crashes for segment i; i = the estimation error for segment i. It is usually assumed that the crash data are approximately nor mally distributed. Thus, the scaled deviance SD and Pearsons 2 statistic for an adequate model should be approximately chi-square distributed with ( N p ) degrees of freedom, where N is the number of observations and p is the number of parameters in the model.
42 CHPATER FOUR DATA COLLECTION This chapter focuses on illustrating the data collection procedures that include the selected sites and relative sites information. Both freewa y diverge areas and entire exit ramp sections are reviewed and the criteria for classifying the site segments and segment lengths are explained. Detailed methods of identifying road secti ons in FDOTs system, subtracting specific site database, and tackling with the cras h data for each site were depicted in this chapter as well. 4.1 Site Selection Criteria The study focuses on the safety effects of the freeway diverge areas and entire exit ramp sections. In order to obtain reasonable results, criteri a to identify the site segments are really important in order to narrow the unstable and unr elated factors. The criteria were listed below for both freeway diverge areas and freeway exit ramp sections: 1) All the objects are on the freeway diverge areas or exit ramps ; 2) Freeways defined here are the highway segments with the highe st level of service and full control of accesses; 3) Only right exit ramps are considered in the sites which means a ll exits should be at the right hand of the directions on freeways;
43 4) The impacts of left exit ramps are not incorporated in this s tudy as they have significant different features to right exits; 5) A sufficient and significant curb, bar, or other facilities in the median separates two directions; 6) The right-shoulder of freeways and exit ramps should be clear, no si ght obstruction, and no dangerous facilities; 7) The grade variations are smallest so that no grade varieties a re considered in both sections; 8) The freeway segments should be homogeneous segments without large horizontal or vertical curves distinctions since this research would narrow the other parameters that not compared; 9) All sites are in Florida States from District one to Distr ict seven plus an additional Florida Turnpikes generally named as District eight. Two dissimilar sections are selected so that they both have spec ial requirements for the segments. The following items list the special site r equirements at the freeway diverge areas: 10) The minimal posted speed limit on the freeway mainline section shoul d be larger than 50 mph; 11) The upstream and downstream distances from the deceleration lanes are long enough so that influential factors up or down from the deceleration lanes are minimal; 12) Deceleration lanes are calculated from the beginning of the tape r or widening points to the painted nose;
44 13) Four different ramp types on the diverge areas have different number of lanes at freeways, but the research segments remain same. The exit ramp sections that connect the diverge areas and continue unt il the beginning of secondary roads should meet subsequent extra criteria: 14) The exit amp lengths begin from the painted nose and end at the last part of terminals; 15) All exit ramp suggested or post speed limits is larger is 25 tha n mph no matter the ramp configurations or ramp length. Following these criteria ensures that the candidate list of fie ld study sites could be obtained without low speed limits in the freeway diverge areas a nd large difference of speed limits on entire ramp sections. This would make the same cha racters except the concentration variables to do the statistical analysis. The lane width is an interesting parameter in this study so that the lane widths are not necessar ily synchronized in the sites selection procedures. From the field studies, all the pref erred segments would go for the interstate highway systems, expressways, turnpikes, and parkways in Flori da. 4.2 Segment Length Definition Two research sections are defined in this section, the freeway diverge areas and the entire exit ramp sections. The segment length of diverge are as include the deceleration areas and the adjacent vicinities that have relate d effects for traffic exiting or continuing on freeways. The decision is based on both previous studies and sit e observation experiences. The exit ramp length includes the entire ra mp sections no matter the ramp configurations, ramp terminal control types or other fact ors. No more regions are taken into concerns as the ramp sections are continuous to the diverge areas.
45 4.2.1 Freeway Diverge Area Length The freeway diverge segment in this study is a section of freew ay which contains a deceleration lane and its adjacent section. The segment leng th for the freeway diverge area consists of two continuous sections, including (1) a 1500 ft section l ocated in the upstream of the painted nose and (2) a 1000 ft section located in the downstream of the pained nose. Thus, the length of the freeway diverge segment in this study equals 2500 ft for each site. The definition of the freeway diverge segment for each type of exit ramp is also given in Figure 6 through Figure 9. They illustrate the w hole study section that combines the declaration areas and their surrounding areas. Using different influential distances in the upstream of painted nos e could result in different safety analysis results. If the selected distance is too l ong, crashes reported for selected freeway segments may include some mainline crashes which are not directly related to exit ramps. If the selected distance is too short, how ever, the selected freeway segment is not long enough to cover the entire influential area of ex it ramps. In previous studies, the selected influential distance located upstream of the painted nose ranged from 1000 ft to 2000 ft (1, 11, 12). The HCM (13) suggests 1500 ft beyond the painted nose in the simulation software including Corsim and Highway Capacity Soft ware (HCS). In addition, the length of deceleration lane at selected diverge sites varies from 26 ft to 918 ft. Our field observations show that, when the distance to painted nose is greater than 1500 ft, the exit ramp type does not impact behaviors of mainline drivers in an obvious way. Due to these reasons, a 1500 ft section was selected as the influential area located upstream of pained nose and 1000 ft downstream the painted nose on the freew ay mainline sections.
46 FIGURE 6. Type 1 Exit Ramp Length: Parallel from a Tangent Single-lane Exit Ramp FIGURE 7. Type 2 Exit Ramp Length: Single-lane Exit Ramp without a Taper
47 FIGURE 8. Type 3 Exit Ramp: Two-lane Exit Ramp with an Optional Lane FIGURE 9. Type 4 Exit Ramp: Two-lane Exit Ramp without an Optional Lane
48 4.2.2 Exit Ramp Section Length The crash frequency is related to the segment length since diffe rent distances might have different effects on the number of crashes when other si tuations are equal. Usually, longer distances might have more crash potentials than s horter distances. Resende and Benekohal (35) did a comprehensive study on the influence of segment lengths and the geometric variables on crash rates. The paper proved th e essences of different segment lengths. The entire ramp section is the length of the exit ramp itself. The definition means that the painted nose is the beginning of exit ramp and the end of te rminals is the closing stages for the exit ramp. It varies slightly from past st udies conducted by Lord and Bonneson (2), Bauer and Harwood (3), Khorashadi (4), McCart et al. (6), and Janson et al. (8). Some studies excluded the terminal sections from the entire exit ramps. However, different termination styles would influence the beyond sections as well as the adjacent sections. Some adjusted the exit ramp sections plus the upstream de celeration lanes. This study would separate these two continuous sections because the diverge areas and ramp sections have dissimilar crash features and prominent influential f actors. The mixed of these two might get incorrect results. Even Bauer and Harwood (3) did conside r the entire ramp sections, they ruled out the all the rear-end crashes for the ramps. It might misrepresent the crash distribution and lead to misunderstand of the othe r factors to the rear-end crashes which are generally highly occurred in the exi t ramps. As a result, the clarity of ramp length here uses the definition described befor e. The following Figure 10 from A to D present the ramp segment lengths for four ramp config urations as mentioned above.
49 Figure 10-A. Diamond Exit Ramp Segment Length
50 Figure 10-B. Out Connection Exit Ramp Length
51 Figure 10-C. Free-flow Loop Exit Ramp Segment Length Figure 10-D. Parclo Loop Exit Ramp Segment Length FIGURE 10. Exit Ramp Segment Lengths for Four Ramp Configurations
52 From the four figures, four bold lines added to each one illustrate t he study field for exit ramp sections. Even they have special design patterns as they appear, the principles are unique. This is intended to obtain useful results and rai se the accuracy of the analysis. 4.3 Selected Sites Information In this study, crash data were collected at research segments in the State of Florida. After checking the available sites, the site resources are limited. In this reason, all the freeways are examined in order to get reasonable sampl e sites. Following the sites criteria before, a total of 12 Interstate Highways, 10 expresswa ys, 1 turnpike and 1 parkway are overviewed and sites are collected on these freeway s. These freeways provide high service level with high design standards. Figure 11 be low lists the most important four interstate highways. Interstate Highway 75 (I-75) and Interstate Highway 95 (I-95) are both north-south directions while Interstate 4 (I-4) a nd Interstate Highway 10 (I-10) are east-west directions. Other highways connect intraregion or inter-regions as to provide better traffic operations at limited accesses. Florida divided eight districts for the whole state, from Distric t One to District Eight. District One through District Seven have their local off ices to manage each district respectively. District eight is the toll roads that are built, m anaged and maintained by all Florida areas. FIGURE 12, the District Map, gives an idea about t he seven districts allocation in the Florida. The figure is original from FDOT Comm unity Traffic Safety Teams (CTST). These selected freeways are dispensed in all the eight districts and Table 1 lists the detailed information of each district.
53 FIGURE 11. Florida Interstate Highway System FIGURE 12. Florida District Map
54 Table 1. FDOT Districts Distributions for Selected Sample Sites District Number Freeways One I-75, I-4; Two I-295, I-10, I-75, I-95; Three I-10, I-110; Four I-595, I-75, I-95; Five I-4, I-75, I-95, Bee Line Exp, East-West Expressway, Central Florida Greenway Expressway; Six I-395, I-75, I-95, I-195, Dolphin Expressway, 826 State Highway, Palmetto Expressway, Florida Turnpike, Don Shula Expressway; Seven I-375, I-75, I-275, I-175,I-4, Veterans Expressway, S Crosstown Expressway, N Memorial Expressway; Eight Florida Turnpike, Polk Parkway;
55 4.3.1 Freeway Diverge Areas The task of site collection is the most time-consuming and tedious work in this study. Hundreds of sites are available and each site needs to c heck patiently and review carefully to make sure all the related data are correct. A rea photos for each site were pulled together. However, some sites are under reconstructions or ha ve been closed for some time during the study period. Some sites did not have detail ed site information such as AADT, especially at some expressways. Since some sites did not have full information, they did not meet the sites requirements as mentioned before. These sites might be large curvatures, low post speed limit as 45 mph, grade variation much higher than the expect one and so on. After reviewing the area photos for freeway diverge a reas in the State of Florida. 424 sites were selected for the freeway diverge segments. Among these sites, 220 sites are Type 1 exit ramps-parallel from a tangent single -lane exit; 96 sites are Type 2 exit ramps-single lane exit ramp without a taper; 77 sites ar e Type 3 exit ramps-two lane exit ramp with an optional lane; and 31 sites are Type 4 exit r amps-two lane exit ramp without an optional lane. Table 2 lists the site resources for each type. Table 2. Sites Resource Distributions for Freeway Diverge Areas Resource Exit Ramp Type Total Size Interstate Highways Expressways Turnpikes Parkways 1 220 220 0 0 0 2 96 96 0 0 0 3 77 59 16 2 0 4 31 17 11 2 1
56 4.3.2 Exit Ramp Segments The work of sites gathering on the ramp sections is labor intensive as well. Since the exit ramp sections are sequential to the diverge areas, the sample size basically equals to freeway diverge sites with available data. However several sites did not have ramp ADT because there are no detectors there. These sites are exc luded from the exit ramp sites. So a total of 389 sites are determined as the sample siz e for the entire exit ramp segments. 4.4 Site Selection Procedures The processes of site selection can be explained in three steps, field study, site information collection, and site review. Field study is the first step to collect raw data as geometric data, site notification data and other related factors Based on these data, the sites ID could be obtained from Florida road identification systems : Straight-Line Diagram (SLD) and Florida Traffic Information CDs. Finally, al l the selected sites are checked again to acquire available sites. 4.4.1 Site Selection Procedure 1 Step 1 Field Study: Field study collects site location and ge ometric conditions which match the requirements and criteria. The photograph maps were obt ained from each district traffic information CD. For each site, simple sketches with geometric information were checked to find the following information: 1) Major freeway names; 2) Freeway directions;
57 3) Ramp types; 4) Deceleration lane lengths; 5) Number of lanes in freeways; 6) Post Speed Limits on freeways; 7) Upstream 1500 ft distances measurements from the painted nose; 8) Downstream 1000 ft distances measurements from the painted nose; 9) Exit ramp directions; 10) Ramp lengths; 11) Number of lanes in the ramp; 12) Ramp suggested or post speed limit; 13) Number of lanes changing on the ramp sections; 14) Ramp terminal control types; 15) Secondary road name; 16) Distances from the first upstream intersection on the secondary road; 17) Distances from the first downstream intersection on the secondary road; 18) Number of lanes on the secondary roads. 4.4.2 Site Selection Procedure 2 Step 2 Extracting Road ID: SLD and Florida Traffic Informati on (FTI) annual CDs were obtained from corresponding FDOT district offices. The road mileposts and road identification numbers for each site were gathered from SLD a nd ADT each year were subtracted from traffic information CDs. These kinds of inform ation are listed below:
58 19) Section and subsection number of the freeways; 20) Section and subsection number of exit ramp sections; 21) Milepost on the beginning and end of the segment length for diverge areas; 22) Milepost on the beginning and end of the segment length for exit ramps; 23) Site number for freeways; 24) Site number for exit ramps. 4.4.3 Site Selection Procedure 3 Step 3 Site Review: Each site and the related information wer e checked again to prove that all the data are correct and confirm that no significa nt reconstruction had taken place at the selected study sites during the study period. 4.5 Section Number, Milepost and Site Identification Number The section number and milepost for each selected freeway segment was obtained from the SLD provided by the Florida Department of Transportation. T he purpose of using section numbers and mileposts is to consist with FDOT crash database. Each section number contains eight digital codes which were used to identify one specif ic road. The first two digital codes are the county number for each distr ict. The subsequent three digital numbers are section numbers and the last three digits are the subsection numbers. While looking for a location in a site, section number is not enough. The milepost was additional information to recognize the position on the roadway segment. Mileposts are made from the beginning of a road way from south to north or from we st to east. For example, I-75 in Hillsborough County (section number 10 075 000) begins at the
59 Manatee/Hillsborough county line as milepost 0.000 and ends as milepost 36.25 at Pasco/Hillsborough County. Site ID is another index in the annual FTI CDs which contained s everal essential parameters including AADT, peak hour factor, and other volume related da ta. Six numbers are combined. The first two are the county number and the rest four digits are the sites recognized ID. The site ID for I-75 at Bruce B. Dow ns exits is 10 0153. The AADT for this section could be obtained from AADT annual report through site ID. 4.6 Crash Database Based on the range in mileposts of each segment, crash data reported was obtained from the crash database maintained by the State of Florida In 2003, the FDOT renamed all the freeways exit ramps for the whole state. Acc ordingly, the crash database updated the exit ramp numbers for the entire database. Due to this reason, crash data for freeway exit ramps before 2004 include a lot of missing informati on and, as a result, cannot be used in this study. A three-year time frame, from 2004 t hrough 2006, was selected to obtain crash data. Eighty-six variables are enclos ed in the FDOT crash database including site identification, time of crashes, traffi c conditions, geometric conditions, crash detailed information as location, direction, crash t ype, severity and so on. The software SPSS would be used to examine the crash data. Figur e 13 shows the SPSS format from FDOT crash database for one site.
60 Figure 13. SPSS Example format from FDOT crash database 4.7 Combination of Crash Data with Site Information Each site has a specific database consisted of geometric var iables, traffic data and relative crash information. The Excel file will be used to arr ange the format of each location for useful variables. The following Figure 14 shows part da ta from the combining database for some sites. FIGURE 14. Example of Combining Database
61 CHAPTER FIVE DATA ANALYSIS Detailed procedures and results of crash data analyses were per formed in this chapter. As mentioned before, freeway diverge areas and entire e xit ramp sections are two separate research subjects in the study. Quantitative invest igations were conducted to find out crash characteristics and the contributing factors in order to evaluate safety performances both on the freeway diverge areas and exit ramp sections. 5.1 Outline of Data Analysis Crash data for freeway diverge areas and exit ramps are anal yzed independently as to evaluate the safety performances on the two research sec tions in this study. As mentioned previously, the cross-sectional comparisons were conducted to c ompare the effects of the four exit ramp types on the safety performance of freeway diverge areas and effects of ramp configurations on the safety performance of t he exit ramp sections respectively. On the freeway diverge areas, a total of 424 sampl e sites were collected. The sample size was divided into four groups according to the four diffe rent exit ramp types as mentioned before. Group 1 has 220 sites for Type 1 exit ramp s, Group 2 has 96 sites for Type 2 exit ramps, Group 3 has 77 sites for Type 3 exit ramps and Group 4 has 31 sites for Type 4 exit ramps. On the exit ramp sections, a tot al of 389 sites with 247
62 sites for diamond ramps, 93 sites for out collection ramps, 26 sites f or free-flow loop ramps, and 23 sites for parco loop ramps were categorized. Two cra sh predictive models were developed for the two research subjects to find the contributing factors to the crashes occurring at diverge areas and exit ramp sections. First, average crash frequency and crash rate for each group on the two research subjects were calculated. Statistical tests were conducted t o compare each section at a 90% confidence level one by one. Second, each group had the sample sites cla ssified by target crash types that have three most crash frequencies a mong all the crash types. Then the average crash number and crash rates by target crash types were calculated by using crash data from 2004 to 2006 and the corresponding statistical test s were performed. Third, crash severity categories such as PDO (property-damageonly), injury and fatality were compared with corresponding average crash number and crash rate by each ramp configuration. The comparisons were followed by statistical sig nificance tests at 90% confidence level which is believable and commonly used in crash analys is. Finally, two predict models were built to find the predictive crash number under som e definite conditions according to the independent variables. 5.2 Freeway Diverge Areas 5.2.1 Comparison of Average Crash Frequency and Crash Rate A total of 13968 crashes were reported at selected freeway diver ge segments for three years from 2004 to 2006. The crash frequency at selected sites varies from 0 to 60 with a mean of 11.01 crashes per year. Summary statistical anal yses of crash frequency and crash rate for four exit ramp groups were illustrated in Tabl e 3. The average crash
63 frequency and crash rate for different exit ramp groups were compa red in Figure 15 and 16. Average crash frequency is the mean value of all the crashes in one group each year. In this study, crash rate is defined in the methodology chapter, se t as crashes per million vehicles per mile. The average daily traffic for each site was collected and the segment length was identified equally for each site. For example, if s ite I has 10 crashes for the three years from 2004 to 2006, segment length is 0.47 miles (2500 ft), and the ADT is 10,000 vehicles per day, the crash rate for this site I could be calculated as fol lowing: Crash Rate for the Site I = 94.1 47.0 000 10 3 365 10 000 000 ,1 = miles vpd years days crashes The average crash rate for a particular group is calculated by the mean value of crash rates for all sites. As shown in Figure 15 and 16, the type 1 exit ramp group has the best safety performance in terms of the lowest average crash fr equency and crash rate comparing to other exit ramp types. The figures also show that the type 2 exit ramp group has the highest average crash frequency and crash rate. The tr ends of average crash frequency and crash rate among the four types showing in the figure s are sequent. Type 1 and Type 2 have the lowest and highest average crash frequency and crash rate among the 4 groups while the average crash frequency and crash rate f or Type 3 and Type 4 is a little higher than Type 1 and a little lower than Type 2. Table 3 listed the detail ed analysis such as mean, median, max and min values for each group. On average, the sites in type 2 exit ramps group report the most average crash frequency as 15.4 crashes per year in freeway diverge segments. As compared those in Type 1 exit ramp group, sites in Type 2 exit ramp group reports 75% more crashes per year for one lane e xit ramp. The average crash rate at sites with Type 2 is also 35.6% higher when compa ring those with Type 1
64 0 2 4 6 8 10 12 14 16 Type 1Type 2Type 3Type 4 Exit Ramp TypeAverage Crash Frequency (crashes/year) per year. For two lane exits, Type 3 appears 20% and 14% less ave rage crash frequency and crash rate than Type 4. Figure 15. Comparison of Average Crash Frequency among Four Exit Ramp Types Figure 16. Comparison of average Crash Rate among Four Exit Ramp Types 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 Type 1 Type 2 Type 3 Type 4 Exit Ramp TypeCrash Rate (crashes/mvm)
65 Table 3. Summary of Average Crash Frequency and Crash Rate for Four Exit Ramp Types Crash Frequency (No. of crashes per year) Crash Rate (No. of crashes per million vehicles per mile) Type 1 2 3 4 1 2 3 4 No. of Sites 220 96 77 31 220 96 77 31 Total No. of Crashes per year 1934 1481 824 417 1934 1481 824 417 Average No. of Crashes 8.8 15.4 10.7 13.45 0.45 0.61 0.48 0.61 St. Deviation 6.23 13.8 8.14 11.3 0.36 0.45 0.32 0.37 Median 4.7 13.2 8.67 12.3 0.36 0.48 0.42 0.55 Max 54 30 31 60 1.36 1.98 1.18 1.24 Min 0 0 1 0 0 0 0.061 0 The site with the highest crash frequency is located on Intersta te Highway 95 (I-95) in District 4 along the southbound. Figure 17 below showed the site picture. During the three-year time period, 179 crashes were reported at selected f reeway segments. 101 are injury plus fatal crashes and the others are PDO crashes. Fiel d observation was made to the particular site to identify the undesirable driving behaviors contr ibuting to the high crash frequency. The segment is located on a five-lane freeway with a posted speed limit of 55 mph. The exit ramp is found to be a type 4 exit ramp which is a two-lane exit ramp without an optional lane. The annual daily traffic volume (ADT) on the f reeway is 224,000 vehicles per day. The reasons that had most crashes might be the t raffic volume
66 was higher than usual, and the exit ramp type in the site caused more weaving maneuvers in diverge areas. Drivers who mistakenly entered the exit lane need to merge back into through lanes to continue on the freeway; while vehicles exiting fre eways may need to change up to four lanes to weave to the outer exit lane. Some severe weaving conflicts have been observed at the site that indicates a high potential crash prone area. FIGURE 17. Site Picture for I-95 Southbound Exit 74 In order to compare whether the average crash frequencies and cras h rates for the four exit ramp types have significant different from each other, hypothesis tests were applied to evaluate the samples. For example, the statistical Z test to compare the average crash frequency for Type 1 exit ramp and Type 2 exit ramp was performed as foll owing: 1) The mean values for two populations Type 1 exit ramp and Type 2 ex it ramp are 1 and 2 ; 2) Mean value and standard deviation of the two samples for Type 1 exit ramp are 8.8 and 6.73 respectively, while those for Type 2 exit ramp are 15.4 and 13.8;
67 3) The sample numbers for Type 1 exit ramp and Type 2 exit ramp a re 220 and 96 accordingly; 4) The null hypothesis is H 0 : 1 = 2 the alternative hypothesis is H a : 1 2 ; 5) Assuming the difference in the sample means of the two population f it the normal distribution and 90% confidence level was chosen for this study; 6) 06.3 96 8. 13 220 23.6 4. 15 8.8 2 2 0 = + = Z ; 7) The critical value for 2/aZ is 1.645 which is smaller than 3.06 so that the null hypothesis is rejected; 8) The conclusion could be get as the average crash number for Type 1 a nd Type 2 exit ramp is significant different at a 90% confidence level. The average crash frequency and crash rate for each population we re tested at a 90% confidence level. Table 4 listed all the results for the hypoth esis tests. The comparison of the average number of crashes for Type 1 and Type 2 exit ramp showing 1:2 is significantly different at a 90% confidence level meani ng YES in the table. For average crash frequency, Type 1 shows significant different fr om the other three types while Type 2 has significantly different average crash frequency with Ty pe 3 but not with Type 4 exit ramps. The results were consistent for average cr ash frequency and crash rate except comparing Type 1 and Type 3 exit ramps. This might be the c ause that crash rate has limited the traffic volume impacts. For one lane exit ramp, Type 1 exit ramp is much safer than Type 2 exit ramp. For two-lane exit ramp, Type 3 group did appear significant difference with Type 4 exit ramp on average crash rate.
68 Table 4. Summary Hypotheses Tests of Average Crash Frequency and Crash Rate for Four Exit Ramp Types Statistics Results for Two Mean Tests: 90% Crash 1:2 1:3 1:4 2:3 2:4 3:4 Frequency YES YES YES YES NO NO Rate YES NO YES YES NO YES 5.2.2 Comparison of Target Crash Type Three target crash types as mentioned before, rear-end crashes, angle crashes and sideswipe crashes, were compared for each exit ramp type to find the cras h characteristics among the four ramp types. Table 5 lists the total numbers of cra shes, percentages of total crashes, average crash numbers, standard deviations and median val ues for the four ramp types by three target crash types. The average crash numbe rs for rear-end crashes and sideswipe crashes among the four types have larger differenc es among each other while those for angle crashes have minor distinctions among the four ramp types. In Table 6, the average crash rate for Type 1 and Type 3 equal of 0.21 crashes per million vehicles per mile per year for rear-end crashes. But Type 2 and Type 4 have 30% a nd 34.4% more crashes than these two types. Figure 18 illustrates that the percentage of rear-end crashes for 4 type s are 45.97%, 48.41%, 41.26%, and 44.60%. Type 3 group counts less percentage than the other three groups. It is reasonable that two-lane exit ramp with an operationa l lane will provide more spaces for vehicles acceding or decreasing speed in the di verge area than single-
69 lane exit ramp. With the optional lane, some unfamiliar drivers or t hese drivers on the wrong lanes would have an opportunity to either continue or leave the free way mainline segments. The sideswipe crashes is the crash type that have the second largest crash number. Table 5 shows the percentage of each group for sideswipe crash es is 15.82%, 15.67%, 15.05% and 16.31%. That might be a result of the additional weaving maneuvers for Type 4 comparing to Type 3. As Type 4 exit ramp group, some driv ers are willing to continue on the freeways when they may misunderstand the inner lane o f two exits as a through lane. When they found it was an exit lane, they might take some dangerous maneuvers such as quickly reducing speed, immediately changing la nes, or even completely stopping which often cause more sideswipe crashes happeni ng to continue driving on freeways. Type 3 appears less rear-end and sideswipe c rashes than other three exit ramp types. Table 5. Summary of Average Crashes Numbers by Target Crash Types for Four Exit Ramp Types Target Crash Types Statistics Type 1 Type 2 Type 3 Type 4 No. of Crashes per year (% of Total) 899 (45.97%) 717 (48.41%) 340 (41.26%) 186 (44.60%) Average No. of Crashes 4.09 8.06 4.42 6.00 Standard Deviation 7.50 8.75 4.40 7.05 Rear-end Crashes Median 2 6 3 6
70 Table 5 (continued) No. of Crashes per year (% of Total) 152 (7.88%) 121 (8.19%) 76 (9.22%) 27 (6.47%) Average No. of Crashes 0.69 1.26 0.99 0.87 Standard Deviation 0.91 1.16 0.79 0.89 Angle Crashes Median 0.67 1.33 1 1 No. of Crashes per year (% of Total) 306 (15.82%) 232 (15.67%) 124 (15.05%) 68 (16.31%) Average No. of Crashes 1.39 2.42 1.61 2.19 Standard Deviation 3.52 2.10 1.43 1.97 Sideswipe Crashes Median 1 2.33 1.33 2.67 Table 6. Summary of Average Crashes Rates by Target Crash Types for Four Exit Ramp Types Target Crash Type Statistics Type 1 Type 2 Type 3 Type 4 Average No. of Crashes 0.210 0.300 0.210 0.320 Standard Deviation 0.250 0.291 0.225 0. 350 Rear-end Crashes Median 0.120 0.170 0.260 0.130 Average No. of Crashes 0.054 0.054 0.050 0.053 Standard Deviation 0.100 0.028 0.029 0.032 Angle Crashes Median 0.040 0.050 0.055 0.050 Average No. of Crashes 0.091 0.115 0.118 0.098 Standard Deviation 0.118 0.111 0.067 0.054 Sideswipe Crashes Median 0.060 0.090 0.120 0.060
71 Figure 18. Comparison of Percentages by Target Crash Types for Four Exit Ramp Types Proportionality tests were then conducted to compare the percentage s of difference among the four groups. The procedures of proportionality test are similar to Z tests mentioned before. For example, the portions of rear-ends crashe s to total crashes for Type 1 exit ramps and Type 2 exit ramps were tested as following: 1) The two populations, Type 1 exit ramp and Type 2 exit ramp, have the pe rcentages of rear-end crashes to the total crashes as 1 p and 2 p ; 2) The percentages of rear-end crashes to the total crashes for the two samples of Type 1 exit ramp and Type 2 exit ramp are 1 p 45.97%, and 2 p 48.41%; 3) Type 1 exit ramp has 220 sites and Type 2 exit ramp has 96 sites;
72 4) The null hypothesis is H 0 : 1p -2p =0, the alternative hypothesis is H a :1p 2p 0 ; 5) Assuming the difference of proportions for rear-end crashes in the t wo samples fits the normal distribution and 90% confidence level was chosen for this study; 6) 40.0 96 ) 41. 48 100 ( 41. 48 220 ) 97. 45 100 ( 97. 45 41. 48 97. 45 = + = Z ; 7) The critical value for 2/aZ is 1.645 which is much larger than Z* so that the n ull hypothesis can not be rejected; 8) The conclusion is that the proportions of rear-end crashes for Type 1 and Type 2 exit ramp is not significantly different at 90% confiden ce level; All the results are given in Table 7. The results o f the proportionality tests show that the percentages of both rear-end and angle/rig ht-turn crashes among the four exit ramp groups on the freeway diverge areas did not ha ve statistically significant differences with 90% level of confidence. This conclusion indic ated that the three crash types having the highest crashes did not differ a lot for the fo ur types. Table 7. Z Statistics for Proportionality Tests by Target Crash Types for Four Exit Ramp Types Proportionality Tests:90% Crash Type 1:2 1:3 1:4 2:3 2:4 3:4 Rear-end NO NO NO NO NO NO Angle NO NO NO NO NO NO Sideswipe NO NO NO NO NO NO
73 5.2.3 Comparison of Crash Severity Among the total crashes reported for selected freew ay segments, 7518 property damage only (PDO) crashes, 6333 injury crashes and 117 fatal crashes were included. In this study, crash severity was compared among diffe rent exit ramp groups by comparing percentages of PDO crashes and injury plus fatal cr ashes. Summary statistics for crash severity for different exit ramp groups are given i n Table 8 and 9 and compared in Figure 19. For one lane exit ramp, Type 1 exit ramp has le ss average crash frequency and crash rate for both PDO crashes and injury plus fatality crashes than the type 2 exit ramp group. Also, Type 3 exit ramp appears less average crash f requency and average crash rate for the two crash severity categories for two-lane exit ramps. As compared in Figure 19, the percentage of injury plus fatality crashes does not significantly differ from each other among different exit ramp groups. Type 2 exit ramp has slightly higher percentage of injury plus fatality crashes comparing to Type 1 ex it ramp for one lane exit ramp and Type 4 exit ramp is a bit higher than Type 3 exit r amp for that as well. Table 8. Summary of Average Crash Number by Crash Severity for Four Exit Ramp Types Crash Severity Statistics Type 1 Type 2 Type 3 Type 4 No. of Crashes (% of Total) 1072 (55.43%) 771 (52.06%) 444 (53.88%) 219 (52.52%) Average No. Of Crashes 4.87 8.03 5.77 5.23 Standard Deviation 6.92 7.64 4.82 6.57 PDO Median 3.67 13.80 4.67 9.00
74 Table 8. (Continued) No. of Crashes (% of Total) 862 (44.57%) 710 (47.94%) 380 (46.12%) 198 (47.48%) Average No. Of Crashes 3.92 7.40 4.94 6.39 Standard Deviation 5.38 7.18 4.16 5.18 Injury/ Fatality Crashes Median 2.33 6 3.33 4.67 Table 9. Summary of Average Crash Rates by Crash Severity for Four Exit Ramp Types Crash Severity Statistics Type 1 Type 2 Type 3 Type 4 Average No. of Crashes 0.325 0.342 0.276 0.356 Standard Deviation 0.292 0.314 0.204 0.231 PDO Median 0.205 0.245 0.24 0.32 Average No. of Crashes 0.204 0.287 0.238 0.278 Standard Deviation 0.155 0.2 0.167 0.174 Injury/ Fatality Crashes Median 0.17 0.23 0.18 0.28
75 0% 10% 20% 30% 40% 50% 60% 70% Type 1Type 2Type 3Type 4 Ramp Type Severity Crashes/Total Crashes PDO Injury/Fatal FIGURE 18. Comparison of Percentages by Crash Severity for Four Exit Ramp Types Proportionality tests were also conducted for testi ng the differences in crash severity among four exit ramp groups. The crash dat abase includes 6420 injury plus fatality crashes for three years time frame. The nu ll hypothesis of the proportionality test is that the percentages of injury plus fatal crashe s in different exit ramp groups are equal. The conclusions of Z statistics for the proportiona lity tests are listed in Table 10. The calculating procedures are same as target crash typ es mentioned above. Based on the Z statistic tests, there is no evidence to reject the null hypothesis with 90% level of confidence. The results suggest that, even the exit ramp types significantly impacts the average crash frequency and average crash rate, the differences of their impacts on crash severity are not statistically significant.
76 Table 10. Z Statistics of Proportionality Tests by Crash Severity for Four Exit Ramp Types Z Statistics for Proportionality Tests Crash Severity 1:2 1:3 1:4 2:3 2:4 3:4 PDO NO NO NO NO NO NO Injury/Fatal NO NO NO NO NO NO 5.2.4 Crash Predictive Model In this study, a crash prediction model was develop ed to identify the factors that contribute to the crashes reported at selected free way segments and to quantify the safety impacts of different types of freeway exit ramps. C onsidering the available data source, a total of 404 observation sites were used in the mod el. Since some sites did not have ramp ADT and ramp design speeds. The variables were beli eved significantly important to have potential crashes. The dependent variable of t he model is the average crash frequency per year reported at selected freeway div erge areas. Seventeen independent variables were initially considered when building t he crash prediction model. The four exit ramp types were defined as three indicator var iables. The initially selected independent variables are described in Table 11. Th e value of each variable are also listed in the table.
77 Table 11. Description of Initially Considered Independent Variables on Freeway Diverge Areas Independent Variable Value Frequency Type 2 exit ramp 1 Type 2 exit ramp 0 Otherwise 92 Type 3 exit ramp 1 Type 3 exit ramp 0 Otherwise 75 Type 4 exit ramp 1 Type 4 exit ramp 0 Otherwise 22 Number of lanes on mainline 1 One lane on mainline 2 Two lanes on mainline 3 Three lanes on mainline . n N lanes on mainline 404 Number of lanes on exit ramps 1 One lane on mainline 2 Two lanes on mainline 3 Three lanes on mainline . n N lanes on mainline 404 Length of deceleration lanes Distance of the decele ration lanes (mi) 404 Length of entire exit ramps Distance for the entire ramp from the painted nose to the end of ramp (mi) 404 ADT per year in thousand on freeway sections Average ADT in thousands for three years 2004~2006 404 ADT per year in thousand on exit ramp sections Average ADT in thousands for three years 2004~2006 404 Speed difference between mainline and exit ramp Maximal speed limit difference (mi/h) 404 Road surface condition 0 Dry 1 Wet 404 Land type 0 Primarily business 1 Primarily residential 404
78 Table 11 (continued) Road surface type 0 Blacktop 1 Concrete 404 Right shoulder type 0 Paved 1 Unpaved 404 Right shoulder width Width for the right shoulder ( ft) 404 Post speed on mainline Maximal speed limit (mi/h) 4 04 Post or suggested speed on ramp Maximal speed limit (mi/h) 404 The crash modeling starts from a Poisson model. For an adequate model, the scaled deviance and Pearsons 2 divided by the degrees of freedom shall be close t o one. These two values are used to detect overdispersion or underdispersion in the Poisson regression model. Values greater than 1 indicate ov erdispersion, while values smaller than 1 indicate underdispersion. In this study, the Pearsons 2 divided by the degrees of freedom was found to be 10.50, indicating the fact that the crash data are overdispersed and NB models shall be used. Stepwise regression me thod was used to select independent variables in the model. Seven variables were not fo und to be statistically significant. As a result, these variables were not included into the model. The best model contains ten independent variables. The regression results of th e best model are given in Table 12. As shown in the table 12, the scaled deviance and Pear sons 2 divided by the degrees of freedom are 1.12 and 1.27 which are reasonably clos e to one, indicating the fact that the model is adequately fitted. The final equation of t he model is given as follows:
79 (22) Where, Y = expected average crash frequency in a fr eeway diverge area (crashes/year), X 1 = 1 if the site has a Type 2 exit ramp, 0 others; X 2 = 1 if the site has a Type 3 exit ramp, 0 others; X 3 = 1 if the site has Type 4 exit ramp, 0 others; X 4 = Number of lanes on the mainline sections; X 5 = Length of the deceleration lanes (mile); X 6 = Length of the entire exit ramp (mile); X 7 = ADT per year in thousands on mainline sections; X 8 = ADT per year in thousands on exit ramp sections; X 9 = Speed difference between the post speed limit on mainline and exit ramp sections (mph); X 10 = Post speed limit on mainline sections (mph); Table 12. Regression Results for Crash Prediction Model for Diverge Areas Criteria for Goodness of Fit Criteria DF Value Value/DF Deviance 393 441.5189 1.12 Scaled Deviance 393 441.5189 1.12 Pearson Chi-Square 393 501.1979 1.27 Scaled Pearson 393 501.1979 1.27 Log Likelihood 38746.0924 ) 0301 .0 0.0614 0223 .0 0679 .0 9385 .0 3470 .1 1302 .0 2244 .0 1354 .0 1416 .0 1523 .3 exp(10 9 8 7 6 5 4 3 2 1X X X X X X X X X X Y + + + + + + + + =
80 Table 12 (continued) Analysis of Parameter Parameter Coefficient Standard error 2 Pr > 2 Intercept 3.1523 0.4205 132.12 <0.0001 Type 2 exit ramp 0.1416 0.1066 0.19 0.0610 Type 3 exit ramp 0.1345 0.1239 0.38 0.0536 Type 4 exit ramp 0.2240 0.1033 0.80 0.0543 Number of lanes on mainline 0.1302 0.0512 4.41 0.10 02 Length of deceleration lanes 1.3470 1.2667 1.02 <0. 0001 Length of entire ramp -0.9385 0.1616 35.46 <0.0001 ADT in thousands on mainline 0.0679 0.0079 73.66 <0 .0001 ADT in thousands on ramp 0.0223 0.0049 21.00 <0.000 1 Speed difference 0.0614 0.0023 69.68 <0.0001 Post speed limit on mainline -0.0301 0.0188 12.56 0 .0129 Dispersion 0.4365 0.0339 All selected independent variables were statistical ly significant with 90% confidence level. The coefficients of the model sho w that the crash counts at freeway diverge areas increase with the mainline lane numbe r, the deceleration lane length, mainline ADT, ramp ADT and post speed limit differe nce between mainline sections and ramp sections, however decrease with the entire ram p length, and post speed limit on mainline. With the more numbers of lanes on the fre eway segments, the potential conflict
81 points will increase so that the chances occurring crashes increase. ADT both on freeway mainline areas and exit ramp sections would increas e the opportunities occurring crashes. It is consistent with previous studies (1, 3). Anot her two positive variables are the deceleration lengths on diverge areas and the post speed limit differences. It was long believed that crash number would decrease if longer deceleration lengths were applied. However, recently a study presented in last Interna tional Symposium on Highway Geometric Design indicated the hypothesis is not co rrect. The study also proved that longer deceleration length might increase the numbe r of weaving maneuvers and cause more potential crashes than short distances. Speed differences between mainline sections and exit ramp sections have positive influences on the crashes as well. It is intuitive as the larger variations on posted speed, more difficu lt for vehicles to control operating speeds. Some vehicles might lose controls as hard d riving maneuvers. From the model, it points out fewer crashes with lo nger exit ramp length. It make sense that longer ramp length would diminish the im pacts of exit ramps on the freeway diverge areas. The coefficient for the posted speed limit is negative, implying that crash counts increase with the increase of the posted spe ed limit of the freeway. This result is a little bit counter-intuitive. A possible explanatio n is that the variable posted speed limit is correlated with other variables which were not incl uded into the model. For example, it is very possible that a freeway with higher posted spe ed limit is also designed according to higher standards. Thus, higher posted speeds may al so imply wider lane width, better lighting conditions, better signing or pavement mar king; and these missing variables could reduce crash freeway at freeway diverge areas
82 The coefficients for the three indicator variables are all positive, indicating the fact that the site with the type 1 exit ramp has th e least numbers of crashes. This conclusion is consistent with the result of our cro ss-sectional comparison. The coefficients of the model can be used to quantify t he safety impacts of different types of freeway exit ramps. Based on the model, replacing a type 1 exit ramp with a type 2 exit ramp will increase crash counts at freeway diverge areas by exp (0.1416-0)-1=15.57%. Replacing a type 3 ramp with a type 4 ramp will inc rease crash counts at freeway diverge areas by exp (0.2244-0.1354)-1=10.80%. 5.3 Exit Ramp Section 5.3.1 Crash Characteristics Four different exit ramp configurations were groupe d for each category to evaluate the impacts on the safety performance. A t otal of 2520 crashes were stated for the entire segments for three years from 2004 to 20 06. The sites were grouped for four configurations simply named as D (Diamond), O (Outconnector), F (Free-flow Loop) and P (Parclo Loop). The group D has 247 sites, the group O has 93 sites, the group F has 26 sites and the group P has 23 sites. The average crash frequencies for the four groups are 2.20, 2.32, 2.21 and 1.00 crashes per site per year. Summary statistics for average crash frequency and average crash rate by four exit ramp configuration groups were given in Table 13. Average crash frequency is the mean value of all th e crash frequencies in one group for each year. Crash rate is defined in the m ethodology chapter as crashes per million vehicles per mile. The volume for each site was collected and segment length was
83 set as the whole ramp length for the site. The proc edures of calculating each exit ramp site were similar to the diverge areas. For example if site II has 5 crashes for the three years from 2004 to 2006, the entire ramp length is 0.25 miles (1320 ft), and the ADT is 5,000, the crash rate for this site II could be cal culated as following: Crash Rate for the Site II = 65.3 25.0 000 ,5 3 365 5 000 000 ,1 = miles vpd years days crashes The average crash rate for a ramp configuration gro up is calculated by the mean value of crash rate for all sites. In Table 13, the average crash frequencies indicate the parclo loop group has the less average crash frequency, however the average crash rates point out that the out connection group has the best safety perfor mance while considering the ramp volume and ramp length. The average crash rate is m ore reliable as it eliminates the impacts of different ramp volumes and ramp distance s. The free-flow loop group has more potential crashes in terms of the maximum aver age crash rate comparing to the other three exit ramp types. The average crash rate for the free-flow loop group is almost 162%, and 69% more than the out connection group an d the diamond group. This result shows different ramp configurations might influence the exit ramps in different ways and the free-flow ramp would have more chances to occur crashes. The conclusion is consistent with previous studies (1, 3, and 5). In the past researches (1, 3), diamond ramps had the best safety performances comparing to other ramp configurations. But the out connection ramps have less average crash rate t han the diamond ramps. This might be the reason that the out connection ramps in Florida are widely used as the freeway interchanges that have high design standards than n ormal exits. These improved standards might be better sign locations before and after the entrances of exit ramps,
84 better road conditions, or less variations along th e exit ramps. Table 13 also listed the detailed statistical analysis results such as the t otal crashes per year, mean value, median value, and max and min values for each group in the exit ramp sections. For the loop exits, parclo loop ramps reported 16.7% less averag e crash rate than free-flow loop exit ramps. Table 13. Summary of Average Crash Frequency and Crash Rate for Four Exit Ramp Configurations Crash Frequency (No. of crashes per year) Crash Rate (No. of crashes per million vehicles per mile) Type D O F P D O F P No. of Sites 247 93 26 23 247 93 26 23 Total No. of Crashes 544 216 57 23 544 216 57 23 Average No. of Crashes 2.20 2.32 2.21 1.00 3.47 2.24 5.86 4.88 Standard Deviation 2.46 3.44 2.20 1.09 6.35 3.89 8.33 8.9 Median 1.33 1.33 2 0.67 1.86 0.85 2.16 2.20 Max 11 22 8 4 77.11 22.25 37.28 41.51 Min 0 0 0 0 0 0 0 0 In order to compare whether the average crash frequ encies and crash rates for the four exit ramp configurations have significant diff erences from each one, hypothesis tests were used to evaluate two populations. For example, the statistical Z or t test of average
85 crash rates for the diamond ramp group and the out connection ramp group was performed as following: 1) The mean values for two populations the diamond exi t ramp and the out connection exit ramp are 1 and 2 ; 2) Mean value and standard deviation for the diamond e xit ramp configurations are 3.47 and6.35, while those for the out-connector exit ram p are 2.24 and 3.89; 3) 247 sites are diamond exit ramps and 93 sites are o ut connection sites; 4) The null hypothesis is H 0 : 1 = 2 the alternative hypothesis is H a : 1 2 ; 5) Assuming the difference in the sample means of the two population fit the normal distribution and 90% confidence level was chosen fo r this study; 6) 97.4 93 89.3 247 35.6 24.2 47.32 2 0= + = Z ; 7) The critical value for 2/aZ is 1.645 which is smaller than 4.97 so that the nul l hypothesis is rejected. 8) The conclusion could be got as the average crash ra te for the diamond exit ramps and the out-connector exit ramps is significant differe nt at 90% confidence level. The average crash frequency and crash rate for each population were tested at a 90% confidence level. Considering the sample size for parclo loop group i s less than 25, t tests were chosen to use for this particular group as mentioned in the methodology parts. The basic procedures are same instead of the functi onal form which has been described in the methodology part. Table 14 listed all the resul ts for the hypothesis tests The comparison of the average number of crashes for the diamond exit ramps and the out
86 connection exit ramps showing D:O is significantl y different at 90% confidence level meaning YES in the table. For average crash rate, the out connection exit ramps have significant difference to the other three configura tions. The out connection ramps have the least average crash rate so that it has the bes t safety performance among the four exit ramp configurations at 90% confidence level. The fr ee-flow ramps have the highest average crash rate and the hypothesis tests documen ted this ramp configuration appears more dangerous than the diamond ramps and out conne ction ramps. However, the difference between the free-flow ramps and parclo r amps is not significant at 90% confidence level. Table 14. Statistical Hypotheses Tests of Average Crash Frequency and Crash Rate for Four Exit Ramp Configurations Statistics for Two Mean Tests: 90% Crash Type D:O D:F D:P O:F O:P F:P Frequency NO NO YES NO YES YES Rate YES YES NO YES YES NO
87 5.3.2 Target Crash Types Three target crash types that have the three highes t crash numbers, rear-end crashes, angle crashes and sideswipe crashes, were compared for each ramp configuration among the four exit ramp configurations types. Tab le 15 lists the total numbers of target crashes, percentages of target crashes to total cra shes, average crash numbers, standard deviations and median values for the four configura tions by three target crash types. The average crash numbers for rear-end crashes and angle crashes among the four configurations have larger differences between each other while the sideswipe crashes have minor distinction among the four configuration s. In Table 16, the average crash rates for diamond ramps have highest per million ve hicles per mile per year for rear-end crashes. Free-flow ramps have a little higher avera ge crash rate than the other three configurations for angle crashes and sideswipe cras hes. This is because diamond interchanges did not include large curves and most of crashes happened by the operating speed differences between vehicles. But the loop ra mps such as free-flow loops have a 360 degree changing on the ramp sections alliance. Usually post or suggested speed limits on these ramps are smaller than diamond ramp s, the causation of crashes are more related to the large variations of the alignments o n the ramp itself. This geometric design feature lead to more angle and sideswipe crashes on the free-flow ramps.
88 Table 15. Summary of Average Crashes Numbers by Target Crash Types for Four Exit Ramp Configurations Crash Severity Statistics D O F P No. of Crashes (% of Total) 274 (50.37%) 80 (37.04%) 14 (24.56%) 8 (34.78%) Average No. of Crashes 1.11 0.96 0.54 0.35 Standard Deviation 1.71 1.78 2.48 1.58 Rear-end Crashes Median 0.4 0.33 0 0 No. of Crashes (% of Total) 44 (8.81%) 19 (8.80%) 13 (22.81%) 1 (4.35%) Average No. of Crashes 0.18 0.20 0.50 0.04 Standard Deviation 0.24 0.17 0.36 0.11 Angle Crashes Median 0.18 0 0 0 No. of Crashes (% of Total) 30 (5.50%) 10 (4.63%) 11 (19.30%) 2 (8.70%) Average No. of Crashes 0.15 0.11 0.42 0.09 Standard Deviation 0.32 0.3 0.22 0.16 Sideswipe Crashes Median 0 0 0 0 Table 16. Summary of Average Crash Rates by Target Crash Types for Four Exit Ramp Configurations Crash Severity Statistics D O F P Average No. of Crashes 1.52 0.61 0.59 0.67 Standard Deviation 2.78 1.31 1.23 1.01 Rear-end Crashes Median 0.43 0 0 0
89 Table 16 (Continued) Average No. of Crashes 0.29 0.19 0.90 0.06 Standard Deviation 0.26 0.66 0.76 0.21 Angle Crashes Median 0 0 0 0 Average No. Of Crashes 0.28 0.05 0.76 0.11 Standard Deviation 0.45 0.16 0.98 0.32 Sideswipe Crashes Median 0 0 0 0 Proportionality tests were then conducted to compar e the percentages of difference among the ramp configuration groups. The procedures of proportionality tests are mentioned before in the diverge areas. For exam ple, the portions in rear-ends crashes for the diamond exit ramps and the out connection e xit ramps were tested as follows: 1) The two populations of the diamond exit ramps and t he out-connector exit ramps have the percentages of rear-end crashes to the tot al crashes 1p and 2p ; 2) The percentages of rear-end crashes to the total cr ashes for the two samples of Type 1 exit ramp and Type 2 exit ramp are 1 p and 2 p ; 3) 247 sites are diamond exit ramps and 93 sites are o ut connection sites; 4) The null hypothesis is H 0 : 1p -2p =0, the alternative hypothesis is H a :1p 2p 0 ; 5) Assuming the difference of proportions for rear-end crashes in the sample fit the normal distribution and 90% confidence level was ch osen for this study; 6) 25.2 93 ) 04. 37 100 ( 04. 37 247 ) 37. 50 100 ( 37. 50 04. 37 37. 50 = + = Z ;
90 7) The critical value for 2/aZ is 1.645 which is much larger than Z* so that the n ull hypothesis can be rejected; 8) The conclusion is the proportions of rear-end crash es for the diamond exit ramps and the out-connector exit ramps is significantly diffe rent at a 90% confidence level. Table 17 exhibited all the statistical tests result s for target crash types of exit ramp configurations. The diamond exit ramps have significant higher aver age rear-end crash rate than the other three types at 90% confidence l evel; while free-flow loop exit ramps have higher the average crash rates for angle and s ideswipe crashes than the diamond exit ramps and out connection exit ramps. But the free-f low loop exit ramps and parclo loop exit ramps did not have significant difference on a verage sideswipe crash rate. This conclusion is consistent with the reason mentioned above as loop exit ramps have more opportunities occurring sideswipe crashes due to th e continuous changeable on the ramp. Table 17. Z Statistics for Proportionality Tests by Target Crash Types for Four Exit Ramp Configurations Z Statistics for Proportionality Tests Crash Type D: O D:F D:P O:F O:P F:P Rear-end YES YES YES YES NO NO Angle NO YES NO YES NO YES Sideswipe NO YES NO YES NO NO
91 5.3.3 Crash Severity Summary statistics for crash severity for different exit ramp configuration groups are given in Table 18 and 19. Even free-flow loop a nd parclo loop exit ramps have less average crash frequency for crash severity than the other two configurations. They both have higher average crash rates on crash severity a nd percentages in injury/fatality crashes to total number of crashes. Table 18. Summary of Average Crash Numbers by Crash Severity for Four Exit Ramp Configurations Crash Severity Statistics D O F P No. of Crashes (% of Total) 305 (56.07%) 119 (55.09%) 20 (35.09%) 8 (34.78%) Average No. of Crashes 1.23 1.28 0.77 0.35 Standard Deviation 1.44 1.61 1.12 1.69 PDO Median 0.7 0.67 0.24 0.60 No. of Crashes (% of Total) 239 (43.93%) 97 (44.91%) 37 (64.63%) 15 (65.22%) Average No. of Crashes 0.97 1.04 1.42 0.65 Standard Deviation 1.21 1.15 1.30 0.69 Injury/ Fatality Crashes Median 0.30 0.67 1 0.40
92 Table 19. Summary of Average Crash Rates by Crash Severity for Four Exit Ramp Configurations Crash Severity Statistics D O F P Average No. of Crashes 1.91 1.12 3.16 2.06 Standard Deviation 3.96 2.17 4.39 5.14 PDO Median 0.93 0.30 1.65 0 Average No. of Crashes 1.56 0.99 2.70 2.82 Standard Deviation 2.69 2.04 4.27 4.79 Injury/ Fatality Crashes Median 0.74 0.32 0.79 0.94 Proportionality tests were also conducted to test t he differences in crash severity among different configuration groups. The null hypo thesis of the proportionality test is that the percentages of PDO or injury plus fatality crashes in different groups are equal. The results of Z statistics for the proportionality tests are listed in Table 20. The calculating procedures are as same as target crash type mentioned above. Based on the Z statistic tests, there is no evidence to reject the null hypothesis with 90% level of confidence. The results suggest that the impacts of different exit ramp configurations on crash severity are statistically significant especi ally for those loop exit ramps and nonloop exit ramps. Free-flow loop exit ramps and parc lo loop exit ramps have higher percentage of injury plus fatality crashes but less percentage of PDO crashes comparing to diamond exit ramps and out connection exit ramps at 90% confidence level. Loop exit ramps seem to have more chances occurring high seve rity crashes. This is reasonable as angle and sideswipe crashes usually cause higher cr ash severity than rear-end crashes.
93 Table 20. Z Statistics for Proportionality Tests by Crash Severity for Four Exit Ramp Configurations Z Statistics for Proportionality Tests Crash Type D:O D:F D:P O:F O:P F:P PDO NO YES YES YES YES NO Injury/fatal NO YES YES YES YES NO 5.3.4 Crash Predictive Models Another crash prediction model was developed to ide ntify the factors that contribute to the crashes reported at selected exit ramp segments. Considering the available data source, a total of 388 observation s ites were included in the model. One site did not have ramp design speeds which were bel ieved significantly important to crashes. The dependent variable of the model is the average crash frequency per year reported at selected exit ramp sections. Nineteen i ndependent variables were initially considered when building the crash prediction model The initially selected independent variables are described in Table 21. The value of e ach variable are also listed in the table. The four exit ramp configurations were defined as t hree indicator variables. The crash modeling starts from a Poisson model. For an adequate model, the scaled deviance and Pearsons 2 divided by the degrees of freedom shall be close t o one. These two values are used to detect overdispersion or underdispersion in the Poisson regression model. Values greater than 1 indicate ov erdispersion, while values smaller than 1 indicate underdispersion. In this study, the Pearsons 2 divided by the degrees of freedom was found to be 5.84, indicating the fact t hat the crash data are overdispersed
94 ) 0580 .0 0129 .0 0978 .0 2470 .0 3679 .0 6861 .0 0062 .0 2608 .0 2973 .0 4392 .0 2253 .0721 0. -1 exp( 11 10 9 8 7 6 5 4 3 2 1 X X X X X X X X X X X Y + + + + + + + = and NB models shall be used. Stepwise regression me thod was used to select independent variables in the model. Eight variables were not found to be statistically significant. As a result, these variables were not included into the model. The best model contains eleven independent variables. The regression results of th e best model are given in Table 22. As shown in the table 22, the scaled deviance and Pear sons 2 divided by the degrees of freedom are 1.18 and 1.06 which are reasonably clos e to one, indicating the fact that the model is adequately fitted. The final equation of t he model is given as follows: (23) Where, Y = expected average crash frequency in an e xit ramp section (crashes/year), X 1 = 1 if the site has an out connection exit ramp, 0 others; X 2 = 1 if the site has a free-flow loop exit ramp, 0 others; X 3 = 1 if the site has parclo loop exit ramp, 0 other s; X 4 = Length of the entire exit ramp (mile); X 5 = Number of lanes on the ramp sections; X 6 = 1 if the number of lanes widening after the entr ance of exit ramps, 0 no; X 7 =Upstream distances between exit ramp terminal and first intersection (mile); X 8 = ADT per year in thousands on exit ramp sections; X 9 = Ramp shoulder width (mile) ; X 10 =Post speed limit on mainline (mph); X 11 = Post or suggested speed limit on exit ramp sectio ns (mph);
95 Table 21. Description of Initially Considered Independent Variables on Exit Ramp Sections Independent Variable Value Frequency Out-connector exit ramp 1 out-connector exit ramp 0 Otherwise 93 Free-flow loop exit ramp 1 free-flow loop exit ramp 0 Otherwise 26 Parclo loop exit ramp 1 parclo loop exit ramp 0 Otherwise 23 Number of lanes on mainline 1 One lane on mainline 2 Two lanes on mainline 3 Three lanes on mainline . n N lanes on mainline 388 Length of entire ramp Distance for the entire ramp from the painted nose to the end of ramp (mi) 388 Number of lanes on exit ramps 1 One lane on mainline 2 Two lanes on mainline 3 Three lanes on mainline . n N lanes on mainline 388 Widening 0 No widening on the ramp 1 Exit ramp widening on the exit ramp Section 388 Signal 0 No signal control 1 Signal control Ramp terminal 388 Channalization 0 No channalization 1 Ramp terminal is channalization 388 Secondary upstream intersection Distance between ramp terminal and the first upstream intersection 388 Secondary downstream intersection Distance between ramp terminal and the first downstream intersection 388 ADT per year in thousand on exit ramp sections Average ADT in thousands for three years 2004~2006 388
96 Table 21 (continued) Road surface condition 0 Dry 1 Wet 388 Land type 0 Primarily business 1 Primarily residential 388 Road surface type 0 Blacktop 1 Concrete 388 Right shoulder type 0 Paved 1 Unpaved 388 Right shoulder width Width for the right shoulder ( ft) 388 Post speed on mainline Maximal speed limit (mi/h) 3 88 Post or suggested speed on ramp Maximal speed limit (mi/h) 388 Table 22. Regression Results for Crash Prediction Model for Exit Ramp Secti ons Criteria for Goodness of Fit Criteria DF Value Value/DF Deviance 375 441.8539 1.1783 Scaled Deviance 375 441.8359 1.1783 Pearson Chi-Square 375 397.9857 1.0613 Scaled Pearson 375 397.9857 1.0613 Log Likelihood 3221.6867
97 Table 22 (continued) Analysis of Parameter Parameter Coefficient Standard error 2 Pr > 2 Intercept -1.0721 0.8577 0.6089 0.1113 Out-connect exit ramp -0.2253 0.1577 0.0837 0.0530 Free-flow loop exit ramp 0.4392 0.2428 0.9150 0.07 04 Parclo loop exit ramp 0.2973 0.2897 0.2704 0.0946 Length of entire ramp -0.2608 0.3117 0.3502 0.0428 Number of lanes on exit ramp -0.0062 0.1477 0.2833 0.0335 Widening 0.6861 0.1466 0.9732 <0.0001 Secondary Upstream 0.3679 0.1689 0.6990 0.0294 ADT in thousands on ramp 0.2470 0.0860 0.4155 0.004 1 Should width -0.0978 0.0775 0.0540 0.0266 Post speed limit on mainline 0.0129 0.0093 0.0311 < 0.0001 Post or suggested speed limit on the ramp section 0.0580 0.0133 0.840 <0.0001 Dispersion 1.1143 0.0993 All selected independent variables were statistical ly significant with a 90% confidence level. The coefficients of the model sho w that the crash counts at exit ramp sections increase with the mainline lane number r amp ADT, post speed limit both on mainline sections and ramp sections, distances from ramp terminals to the first upstream intersection, and widening, but decrease with the r amp length, the exit ramp lane number,
98 and ramp shoulder type. With the increase of number of lanes on the exit ramp sections, the situation is different from diverge areas. Sinc e more number of lanes on the ramp sections might diminish vehicle distributions on th e ramp sections which are particular transition from freeway sections to the secondary r oads. The desperation of vehicles would diminish conflict points on the ramp section. With long ramp length, the impacts of freeway diverge areas and secondary cross roads would be minimal, so fewer crashes would occur comparing these short distance ramps th at both freeways and cross roads have influences on the ramp itself. With larger sho uld width, drivers have more flexible spaces while dangerous situations happened especial ly for loop exit ramps that need more space to avoid angle and sideswipe crashes. ADT exit ramp sections would increase the opportuni ties occurring crashes. It is consistent with previous studies. Post speed limits both on mainline and ramp sections have positive influences on the crashes. Since ramp speed is much lower than freeway segments, such as 25-40 mph, drivers would continua lly maintain high speed on the ramp section while the post speed limit is high; however usually ramp sections did not have a high design standard comparing to freeways. This wo uld mistake drivers so that chances of having potential crashes would rise. Another two positive variables are the widening conditions and distance from ramp terminals to firs t upstream intersection. It is institutive that widening would cause more merging or diverging maneuvers which were generally the main reasons of happening crashes. The coeffici ent of distance from ramp terminals to first upstream intersection is 0.3679 which has a significant increase in crash frequency while the increasing the distances. It means if the intersection is far away the ramp terminals, it would raise the chances of happening crashes. If the intersection is nearby
99 the ramp terminals, more attentions would paid at t hose intersection areas as most drivers are more sensitive to intersections than the normal driveways or roadways. The coefficients for the three indicator variables have different signs, indicating the fact that the site with the out connection exit ramp has the least numbers of crashes. This conclusion is consistent with the result of ou r cross-sectional comparison. The coefficients of the model can be used to quantify t he safety impacts of different exit ramp configurations. Based on the model, the sign of out connection exit ramp is negative. It can concluded that replacing a diamond exit ramp wi th an out connection exit ramp, will reduce crashes in the sections by exp (0.2253)-1=26 .90%. However, replacing a diamond exit ramp with a free-flow loop ramp and a parclo l oop ramp will increase crash counts at exit ramp by exp (0.4392)-1=56.86%, and exp (0.2973 )-1= 35.62%. Thus, we can calculate the increasing percentages for replacing an out connection exit ramp with 68.47% and 48.72%. While only concerning on the loo p exit ramp, replacing a parclo loop exit ramp with a free-flow loop exit ramp woul d increase crash counts by exp (0.4392-0.2973)-1=15.66%.
100 CHAPTER SIX SUMMARY, CONCLUSIONS, AND RECOMMENDATIONS 6.1 Summary The objective of this study is to evaluate the impa cts of different exit ramp types on the safety performance. Two research subjects, f reeway diverge areas and exit ramp sections, were selected. Impacts of different exit ramp types on the diverge areas and different ramp configurations on the exit ramps wer e analyzed respectively. This study developed quantitative evaluations and comparisons on the freeway diverge areas and exit ramp sections correspondingly. The results of this study will help transportation decision makers develop tailored technical guidelin es governing the selection of the optimum exit ramp types to be used on our freeways and exit ramps. For the freeway diverge areas, in order to find the impacts of exit ramp types on the safety performance of freeway diverge areas, la ne balance issues were considered to determine the exit ramp types on the freeway diver areas. The exit ramp types were defined by the number of lanes used by traffic to e xit freeways. Four different types of exit ramps were considered in this study. For conve nience, they are defined as Type 1, Type 2, Type 3, and Type 4 exit ramps. Among these exit ramp types, Type 1 and Type 2 are one-lane exit ramps, while Type 3 and Type 4 ar e two-lane exit ramps. Type 1 is a parallel from a tangent single-lane exit ramp. Type 2 is a single-lane exit ramp without a
101 tangent. Type 3 is a two-lane exit with an optional lane and Type 4 is a two-lane exit without an optional lane. A total of 424 freeway se gments were collected in the State of Florida, 220 sites for Type 1 exit ramps, 96 sites for Type 2 exit ramps, 77 sites for Type 3 exit ramps and 31 sites for Type 4 exit ramps. Th e selected sites were divided into four groups based on the types of exit ramps. Crash data were selected for three years, from 2004 to 2006 for each site. Cross-sectional compari son was conducted for comparing the crash frequency, crash rate and crash severity betw een different exit ramp groups. Three target crash types that have the three most crashes were chosen from all the crash types. They are rear-end crashes, sideswipe crashes and an gle crashes. The average crash number and crash rate was calculated by each exit r amp type on each freeway diverge site. The hypothesis tests were conducted for four exit ramp types to compare whether significant differences for average crash frequency and crash rate are present between the four exit ramp types at 90% confidence level. Crash severity was grouped by two categories, property-damage-only crashes and injury /fatality crashes for four exit ramp types. The average crash frequency and crash rate f or each target crash type and crash severity were calculated by four exit ramp types on the freeway diverge areas as well. Proportionality tests were performed for the target crash types and two crash severity categories by four exit ramp types. A crash predict ion model containing 404 sites was developed to identify the factors that contribute t o the crashes reported at selected freeway segments and to quantify the safety impacts of different freeway exit ramps. On the exit ramp sections, the exit ramp configurat ions were grouped by four regular categories, which are diamond exit ramps, o ut connection exit ramps, free-flow loop exit ramps and parclo loop exit ramps. A total of 389 exit ramp sites were collected
102 in the State of Florida, 247 sites for the diamond exit ramps, 93 sites for the out connection exit ramps, 26 sites for the free-flow l oop exit ramps and 23 sites for the parclo loop exit ramps. Crash data were selected for the same years in the diverge areas, from 2004 to 2006 for each site. Cross-sectional co mparison was also conducted for comparing crash frequency, crash rate and crash sev erity between different exit ramp configuration groups. Rear-end crashes, sideswipe c rashes and angle crashes are the target crash types that have the three most crashes among all the crash types. Crash severity was grouped by two categories, property-da mage-only crashes and injury/fatality crashes. The hypothesis tests were completed respectively at 90% confidence level. A negative binomial crash prediction model including 388 sites was developed to identify the factors that contribute to the crashes reported at selected exit ramp segments. 6.2 Conclusions In this thesis, two research parts, freeway diverge areas and exit ramp sections are analyzed separately. The conclusions would describe separately for the two parts. 6.2.1 Freeway Diverge Areas Based on the research analysis, the conclusions on freeway diverge areas can be obtained as following: 1) Type 1 exit ramp has the best safety performance in terms of the lowest crash frequency and crash rate on freeway diverge areas. However, statistical tests show that crash severity and crash types did not have si gnificant differences among the four exit ramp types on the freeway diverge areas at 90% confidence level.
103 2) The predictive model was built. The coefficients of the model show that the crash counts at freeway diverge areas increase with the m ainline lane number, the deceleration lane length, mainline ADT, ramp ADT an d post speed limit difference between mainline sections and ramp sections, howeve r decrease with the entire ramp length, post speed limit on mainline sections and s urface type. 3) The model also quantifies the impacts of different exit ramp types. For one-lane freeway exit ramp, replacing a type 1 exit ramp wit h a type 2 exit ramp will increase crash counts at freeway diverge area by 15.57%. For two-lane exit ramps, replacing a type 3 ramp with a type 4 ramp will increase crash counts at freeway areas by 10.80%. 6.2.2 Freeway Exit Ramp Sections Summary of safety evaluation on exit ramp sections were given in following conclusions: 1) The results of average crash rates on four ramp con figurations show that the out connection group has the best safety performance. T he free-flow loop group has more dangerous in terms of the greatest average crash ra te comparing to the other three exit ramp types. 2) Statistical tests suggest that the loop exit ramps have significant higher crash severity level than non-loop exit ramps at 90% confidence le vel. Three target crash types, which have the three highest crash numbers, are rea r-end crash, angle crash and sideswipe crash. Diamond exit ramps have significan t higher average rear-end crash than the other three types; while free-flow loop ex it ramps have higher average crash rates for angle and sideswipe crashes than the non loop exit ramps.
104 3) The coefficients of the model show that the crash c ounts at exit ramp sections increase with the mainline lane number, ramp ADT, p ost speed limit both on mainline sections and ramp sections, distances from ramp ter minals to the first upstream intersection, and widening, but decrease with the r amp length, the exit ramp lane number and ramp shoulder type. 4) The coefficients for ramp configurations indicate t he fact that the site with the out connection exit ramp has the least numbers of crash es. Based on the model, replacing an out connection exit ramp with a diamond exit ram p, a free-flow loop ramp and a parclo loop ramp will increase crash counts at exit ramp sections by 26.90%, 68.47%, and 48.72%. For the loop exit ramp, replacing a par clo loop exit ramp with a freeflow loop exit ramp would increase crash counts by 15.6%. 6.3 Applications and Recommendations 6.3.1 Applications This study conducted statistical methods and tests to evaluate safety performances of freeway exit ramps on two parts, freeway diverge areas and exit ramp sections. On the freeway diverge areas, four typical exit ramp types used in Florida were compared and it was found that a parallel from a tangent single-lan e exit ramp has the best safety performances among the four exit ramp types. On the exit ramp sections, four widely used exit ramp configurations were selected and com pared in the State of Florida. The study provided technical specifications for transpo rtation agencies to develop tailored guidelines or practical design instructions. Transp ortation engineers, researchers and investigators would benefit from the study as well. The contributing factors to crashes
105 and their impacts were identified and concluded. Th e results of this study would help transportation decision makers select the optimal e xit ramp types and design combinations in our freeway mainline segments under different site situations. 6.3.2 Recommendation Four types of freeway exit ramps were considered on the freeway diverge areas, the crash data analysis results between one lane ex it ramps (Type 1 and Type 2 exit ramps) and two-lane exit ramps (Type 3 and Type 4 e xit ramps) confirm the general assumption that lane balanced exit ramps would be s afer than those not lane balanced exit ramps on the freeway diverge areas (12). In practic e, however, there is also a type 5 exit ramp which is a two-lane exit ramp without optional lane and without a taper. This exit ramp is not widely used in Florida and the samples we found are too small to draw defensible conclusions. To select the optimal exit ramp type, the safety pe rformance of freeway ramp section, more study need to focus on ramp terminal design and control and the diverge deflection angle. These two variables are very impo rtant factors which need to be considered more specific. The authors recommend tha t future studies could be made on these issues. Another important consideration is the conflict stu dies on these sites to further refine the methodology. In addition, operational an alysis and simulation analysis need to be applied. Operational impact and safety impacts s hould look closely to determine the practical design for both freeway diverge areas and exit ramp sections.
106 REFERENCES  Joe Bared, Greg L. Giering and Davey L. Warren, Safety of Evaluation of Acceleration and Deceleration Lane Lengths ITE Journal, May 1999, p 50-54  Dominique Lord and James A. Bonneson, Calibration of Predictive Models for Estimating Safety of Ramp Design Configurations Journal of Transportation Research board, No. 1908, Transportation Research Board of t he National Academies, Washington, D.C., 2005, p 88-95  K.M Bauer, and D.W. Harwood, Statistical Models of Accidents on Interchange Ramps and Speed-Change Lanes FHWA-RD-97-106, FHWA, U.S. Department of Transportation, 1998.  Ahmed Khorashadi, Effect of Ramp Type and Geometry on Accidents FHWA/CA/TE-98/13, California Department of Transpor tation, 1998  Joe Bared, Alvin Powell, Evangelos Kaisar and R amanujan Jagannathan, Crash Comparison of Single Point and Tight Diamond Interc hanges Sources Journal of transportation Engineering, American Society of Civ il Engineers, May 2005, p 379-381  Anne T. McCartt, Veronika Shabanova Northrup, R ichard A. Retting, Types and Characteristics of Ramp-Related Motor Vehicle Crash es On Urban Interstate Roadways In Northern Virginia Journal of Safety Research, 35, 2004, p 107114  N. Garber and M. Fontaine, Guidelines for Preliminary Selection of the Optimum Interchange Type for a Specific Location. VTRC-99-r15  Bruce N.Janson, Wael Awad and Juan Robles, Truck Accidents at Freeway Ramps: Data Analysis and High-Risk Site Identification, Journal of Transportation and Statistics, January 1998, p 75-92.
107  Cirillo, J.A., SK. Dietz, and R.L. Beatty, Analysis and Modeling of Relationships between Accidents and the Geometric and Traffic Cha racteristics of the Interstate System, Federal Highway Administration, August 1969.  Ralph A. Batenhorst and Jef G. Gerken, Operational Analysis of Terminating Freeway Auxiliary Lanes with One-Lane and Two-Lane Exit Ramps : A Case Study, MidContinent Transportation Symposium 2000 Proceedings  Mohamed Sarhan, Yasser Hassan, and Abd El Hali m, Design of Freeway Speed Change Lanes: Safety-Explicit Approach, Transportation Research Board, Washington, D.C., 2006  A Policy on Geometric Design of Highways and S treets, American Association of State Highway and Transportation Officials, Washing ton, DC, 2004  Highway Capacity Manual, Transportation Resear ch Board, National Research Council, Washington, D.C., 2000  Ramp Management and Control Handbook, U.S Depa rtment of Transportation, Federal Highway Administration, 2006  Freeway Management and Operations Handbook, U. S Department of Transportation, Federal Highway Administration, 2004  Pline, James Traffic Engineering Handbook 5 th Edition, Institute of Transportation Engineers. 01-Oct-1999 / 704 pages, ISBN: 093540332 9.  Leisch, Joel P, Freeway and Interchange Geometric Design Handbook, Institute of Transportation Engineering, 2006  John D. Coursey, Highway Safety Improvement Program US DOT & FHWA-TS81-218  Alfredo Garcia, and Mario A. Romero, Experimental Observation of Vehicle Evolution on a Deceleration Lane with Different Len gths, Transportation Research Board, Washington D.C., 2006
108  Hesham Rakha and Yihua Zhang, (2006), Analytical Procedures for Estimating Capacity of Freeway Weaving, Merge, and Diverge Sec tion, Journal of Transportation Engineering ASCE  Mohamed Abdel-Aty and Yile Huang, (2004), Exploratory Spatial Analysis of Expressway Ramps and its Effect on Route Choice, Journal of Transportation Engineering ASCE  Mohamed Abdel-Aty, Jeremy Dilmore, and Albinde r Dhindsa, Evaluation of variable speed limits for real-time freeway safety improvement, (2006), Accident Analysis and Prevention 38, 335-345.  Michael Hunter, Randy Machemehl, and Alexei Ts yganov, (2001), An Operational Evaluation of Freeway Ramp Design, Transportation Research Board of the National Academies, Volume 1751 / 2001, Page 90-100.  Michael J. Cassidy, Shadi B. Anani, and John M Haigwood, (2000), Study of Freeway Traffic Near an Off-Ramp, California PATH Working Paper, UCB-ITS-PWP2000-10.  Joanne C.W. Ng and Tarek Sayed, (2004), Effect of geometric design consistency on road safety, Canadian Journal of Civil Engineering 31 no2 218-27  Juan Carlos Munoz, and Carlos Daganzo, (2000), Experimental Characterization of Multi-lane Freeway Traffic Upstream of an Off-ramp Bottleneck, California PATH Working Paper, UCB-ITS-PWP-2000-13.  Tom H. Maze, Garrett Burchett, Thomas M. Welch and Neal R. Hawkins, (2006), Safety Performance of Two-Way Stop-Controlled Expre ssway Intersection, Institute of Transportation Journal.  Joanne Keller, Mohamed Abdel-Aty, and Patrick A. Brady, (2006), Type of Collision and Crash Data Evaluation at Signalized I ntersection Institute of Transportation Journal.
109  G.F. Newell, (1998), Delays caused by a queue at a freeway exit ramp, Transportation Research Part B 33 (1999) 337-350.  Leonard Shaw and William R. Mcshane, (1972), Optimal Ramp Control for Incident Response, Research and Training in Urban Transportation, Urba n Mass Transportation Administration of the Department of Transportation, Britain.  Kristen L. Sanford Bernhardt, and Mark R. Virk ler, (2002), Improving the Identification, Analysis and Correction of High-Cra sh Locations, ITE Journal.  Richard A. Retting, Allan F. Williams, David F Preusser, and Helen B. Weinstein, (1995), Classifying Urban Crashes for Countermeasure Develo pment, Accident Analysis and Prevention, Vol. 27, No.3, pp.283-294.  John Neter, Michael H. Kutner, Christopher J. Nachtsheim, and William Wasserman, (2002), Applied Linear regression Models, the third edition  Michael Hunter, Randy Machemehl and Alexei Tsy ganov (2000), Reevaluation of Ramp Design Speed Criteria: Summary Report, Center for Transportation Research, The University of Texas at Austin, Texas Department of Transportation Research and Technology Transfer Section/Construction Division, No. 1732-S.  Resende, P. and Benekohal, R.F. (1997), Effects of Roadway Section Length on Accident Modeling Challenges, Innovations and Opportunities: Procee dings on Traffic Congestion and Traffic Safety in the 21 st Century, New York, pp. 403-409
111 Appendix A: Site Picture Examples Type 2 Exit Ramp with Parcolo Configuration Type 3 Exit Ramp with Diamond Configurations
112 Type 1 Exit Ramp with out connection Configuration Type 2 Exit Ramp with Parclo Loop Configuration