1

Multi-objective Evaluation in Countermeasure Selection at Two-Way Stop

Controlled Intersections Considering Traffic Operation, Safety and Environment

By

Zhao Yang, Ph.D., Assistant Professor

National Key Laboratory of Air Traffic Flow Management

College of Civil Aviation, Nanjing University of Aeronautics and Astronautics

Jiangjun Road No. 29, Nanjing 211106, China

Tel: +86-13605152958

Email: yangzhao@nuaa.edu.cn

Yuanyuan Zhang, Ph.D. (corresponding author)

Safe Transportation Research & Education Center

Institute of Transportation Studies, UC Berkeley

2614 Dwight Way, Mail Code #7374

Berkeley, CA 94720-7374

Tel: 315 706 6231

Email: yyz.ucb@gmail.com

Renwei Zhu, Graduate Research Assistant

China Academy of Urban Planning & Design, Shanghai 200335, China

Tel: +86-13915998295

Email: zrenwei_work@163.com

and

Yin Zhang, Ph.D., Assistant Professor

College of Civil Aviation, Nanjing University of Aeronautics and Astronautics

Jiangjun Road No. 29, Nanjing 211106, China

Tel: +86-13584092095

Email: yoyozhying@163.com

Total number of words = 5244+ 250*9=7494

January 8-12, 2017

Paper submitted to the 96th Annual Meeting of the Transportation Research Board

2

Multi-objective Evaluation in Countermeasure Selection at Two-Way Stop Controlled 1

Intersections Considering Traffic Operation, Safety and Environment 2 3

by Zhao Yang, Yuanyuan Zhang, Renwei Zhu and Yin Zhang 4

5

ABSTRACT: 6

This study aims to develop a procedure to conduct multi-objective evaluation in traffic countermeasure 7

(CM) selection process at two-way stop-controlled (TWSC) intersections. To illustrate the procedure, 8

the economic benefits of three vehicle safety related CMs were calculated considering not only the 9

safety impacts but also the operational and environmental impacts. First, for each countermeasure, 10

VISSIM simulation models were developed to obtain the average delay, vehicle emission and fuel 11

consumption for the intersection both before and after the treatment. The traffic operational impacts 12

were calculated as the change in delay costs. The environmental impacts were calculated as the change 13

in vehicle emission and fuel consumption costs. Next, the safety impacts were calculated as the crash 14

reduction benefits for different CMs using safety performance functions (SPFs) and crash modification 15

factors (CMFs). Finally, the life cycle cost (LCC) method was used to combine the different 16

components in the total benefit. The Monte Carlo (MC) simulation method was used to conduct 17

uncertainty analysis by using random sampling from probability descriptions of uncertain input 18

variables to generate a probabilistic description of results. The findings showed first, that the 19

operational and environmental impacts accounted for a large proportion of the total impacts, which can 20

significantly affect the selection of CMs. Second, the rankings of the CMs differ depending on whether 21

the safety impacts alone are considered, or whether the safety, operational and environmental impacts 22

are considered together. The study illustrates the detailed process of evaluating projects considering 23

multiple objectives. This process offers policy and decision makers a solid and practical reference of 24

how to conduct multi-objective evaluation. The findings also explain how different objectives can 25

countervail with each other in improving motorist safety at TWSC intersections. 26

27

KEYWORDS: Traffic operation; safety; environment; countermeasure; multi-objective 28 29

30

3

INTRODUCTION 31 32

In transportation project decision making process, decision makers need to select the most competing 33

project from a set of alternatives. To realize a fully integrated and sustainable transportation system, 34

decision makers need information about the impacts of alternatives on multiple aspects of the 35

transportation system, the environment, and society. The “Transportation Strategic Plan 2012-2016” set 36

up the goals to provide safe, efficient, convenient, and sustainable transportation choices (1). This 37

requires project analysts to consider the multi-objective implications in designing and providing 38

transportation choices and services (2-4), particularly for those projects having countervailing effects 39

on different objectives (i.e., an operational enhancement that speeds up traffic flow may have negative 40

effects on traffic safety (5)). 41

42

Mobility, safety and environment are three important components in multi-objective evaluation for 43

both state and regional plans. Accommodating one objective without negatively affecting other 44

objectives is important when selecting traffic improvements. Therefore, to achieve the multi-objective 45

planning goal, it is necessary to examine the trade-offs among different objectives in order to make the 46

most informed choices among potential CMs. 47

48

In recent years, a few studies have investigated the trade-off impacts among different objectives for a 49

specific project. For example, Sharma et al. examined the impact of signal countdown timer on safety 50

and efficiency of signalized intersections by comparing the queue-discharge characteristics and 51

red-light violations in the presence and absence of timers. Results showed that the information 52

provided at the start of green (end of red) enhances efficiency while increasing incidence of red-light 53

violations (6). Stevanovic et al. examined the trade-off between efficiency and safety in the 54

development of signal timing plans to improve traffic safety. The results indicated that with the 55

increase in cycle length, the total number of conflicts was reduced while the delay was increased (7). 56

Mendonca evaluated the trade-offs between noise pollution and traffic safety of traffic-noise abatement 57

approaches. Results revealed that low noise pavements combined with all-electric and hybrid vehicles 58

might pose a severe threat to the safety of vulnerable road users (8). 59

60

Although some research has been conducted to examine the trade-off impacts among different 61

objectives for a specific project, most of them only provided the separate impact of the project for each 62

objective, without combining them together. But to prioritize the potential treatments, impacts on 63

different objectives need to be combined into one uniformed value to evaluate the multi-objective 64

effects of specific projects (2-4, 9) and to compare potential options. The “User and Non-User Benefit 65

Analysis for Highways” manual (5) presents the concept of combining multiple objectives by 66

converting them into monetary values. However, to date, there are few applications to streamline the 67

entire process of quantifying multiple objectives of a project, and combining them together to evaluate 68

the composite impacts for a specific project. 69

70

To this end, this study presents an analytic approach to select best option of traffic improvements by 71

taking into account multiple impacts of operation, safety and environment on motorists. To achieve this 72

goal, the paper illustrated the procedure through the case of incorporating different objectives into the 73

process of evaluating CMs at TWSC intersections. First, the impacts of different CMs on the traffic 74

operation, traffic safety and environment were quantified. Then, the combined impact for each CM 75

incorporating different objectives was calculated. Finally, the selected CMs were ranked considering 76

4

different objectives under various traffic scenarios. The framework implemented in the study illustrates 77

the solid process of prioritizing the alternative projects considering multi-objective impacts. The 78

framework can also be expanded to integrate additional objectives. 79

80

SAFETY-RELATED COUNTERMEASURES AT TWSC INTERSECTIONS 81 82

Two-way stop-controlled intersection is a common type of intersections in the United States. In a 83

TWSC intersection, vehicles from the two minor approaches to the intersection are required to stop and 84

all the other approaches are uncontrolled. Stopped vehicles on the minor roads are required to wait until 85

there is a sufficient gap in traffic before proceeding. Forcing vehicle to have a full stop is beneficial for 86

safety, but it causes the delay and emission raise on minor roads. In this way, operation, environment, 87

and safety could countervail with each other. A few traffic treatments are proposed to reduce stops and 88

improve traffic operation at TWSC intersections, i.e. conversion from a TWSC intersection into a 89

yield-controlled (YC) intersection. On the other hand, as part of an effort to reduce crashes at 90

intersections, some safety treatments are commonly implemented at TWSC intersections, such as 91

conversion from a TWSC intersection into an all-way stop controlled (AWSC) intersection or a 92

roundabout. These treatments are illustrated as follows: 93

94

CM 1: conversion from a TWSC intersection into a yield-controlled intersection (TWSC-YC). 95

Yield control is an intermediate form of control between normal right-of-way under no sign control and 96

stop sign control. Vehicles arriving at the minor approaches of the intersection do not have to stop 97

before proceeding if there is no traffic on major roads. Prior research has indicated that there could be 98

large savings in fuel consumption, vehicle operating costs, motorists delay, and vehicle emissions if 99

yield control were substituted for stop control at appropriate locations (10). 100

101

CM 2: conversion from a TWSC intersection into an all-way stop-controlled intersection 102

(TWSC-AWSC). An all-way stop-controlled intersection, also referred to as a four-way stop-controlled 103

intersection, is an intersection at which all vehicles are required to stop before proceeding through the 104

intersection. Vehicles generally have the right-of-way to proceed through the intersection in the order 105

that they arrived. This CM is implemented due to the perceived safety benefits, relative low cost, and 106

ease of implementation at the original TWSC intersections with safety problems (11). 107

108

CM 3: conversion from a TWSC intersection into a modern roundabout. A roundabout is a type of 109

circular intersection or junction in which road traffic flows almost continuously in one direction around a 110

central island. A great benefit of roundabouts is that they eliminate perpendicular/T-bone crashes (12). 111

112

DATA COLLECTION 113 114

To calibrate and validate the VISSIM simulation models, data were collected at six TWSC intersections 115

in the state of California. The following criteria were applied in the site selection process: 116

The major street had bidirectional six lanes with a non-traversable median. The minor street had 117

bidirectional two lanes with undivided medians; 118

There were few pedestrians crossing streets; 119

There were limited numbers of bicyclists; 120

Lane widths should be at least 9 feet; 121

The approach grade was level; and 122

5

The selected sites should be 250 feet away from upstream traffic signals. 123

124

The selected sites are given in Table 1. A video camera was set up in the field for recording traffic data. 125

The video cameras were inconspicuously mounted to avoid altering motorist behaviors. Field data 126

collection was conducted during weekday peak periods under fine weather conditions. In total, the 127

research team recorded 24 hours of traffic data in the field. 128

129

The recorded video tapes were later reviewed in the laboratory for obtaining traffic data. For each 130

movement, two reference lines were marked in the video as the location where the motorists generally 131

started to decelerate and the location where the motorists accomplished the acceleration to normal 132

speed (Figure 1). The following pieces of information were collected: (1) the vehicle volume and traffic 133

composition for each movement; (2) the exact time at which the rear wheel of a vehicle crossed 134

reference line A and B. With the recoded data, the travel time for each vehicle from A to B could be 135

obtained. 136

137

METHODOLOGY 138 139

To estimate the impacts of different CMs, the operational, safety and environmental components of the 140

total effects that arise from each CM were calculated. The original TWSC intersection is treated as the 141

base condition. VISSIM simulation models were used to evaluate the operational and environmental 142

impacts with different treatments. The operational impact is calculated as the change in total delay 143

costs for all the vehicles at the treated intersection observed in the calculation time compared with 144

those for the base condition. Similarly, the environmental impact with each treatment is calculated as 145

the change in vehicle emission costs and fuel consumption costs at the intersection compared with 146

those for the base condition. For safety performance, the Highway safety manual (HSM) provides 147

analytical tools and techniques to predict the average crash frequency at TWSC intersections (13). 148

Crash modification factors can be used to estimate cash reduction due to each treatment. 149

150

As the safety, operational and environmental elements of project impacts are quantified in different 151

units of measurements (i.e., delay is measured in seconds, traffic safety is measured as the number and 152

severity of crashes, vehicle emission is measured in grams per unit of time, and fuel consumption is 153

measured in gallon per unit of time), they are converted to a common unit of measurement so that they 154

can be compared with each other. The impacts are spread over future years and the net impacts in each 155

year are discounted to a present value. The User and Non-User Benefit Analysis for Highways (Red 156

Book) presents a method for converting benefit components with various units into monetary values 157

and aggregating the annual benefits/costs across years (5). 158

159

Because the operational impacts and environmental impacts are obtained from the simulation tool by 160

hour, the estimated costs are extrapolated to daily then to annual data. To account for the traffic volume 161

fluctuation during a day, the default hourly traffic distribution provided by FHWA is used in this study, 162

as shown in Table 2 (14). It is assumed that the distribution is consistent through the whole week. The 163

estimated hourly delay savings were aggregated through a day, and then through a year. 164

165

Operational Impacts 166 To obtain the impacts on delay and vehicle emission with the installation of different treatments, 167

VISSIM simulation model was developed for the selected TWSC intersections. The VISSIM 168

6

simulation models were calibrated using the traffic and geometric data measured at the selected site. 169

Travel time for different movements was used to validate the model. To take into account the stochastic 170

nature of simulation results, the VISSIM simulation models were run for multiple times with different 171

random number seeds and each run lasted for one hour period. The mean absolute percent error (MAPE) 172

was used to measure the differences between the field measured and the simulated travel time. The 173

MAPE value can be estimated as: 174

1

1i i

f s

i

f

l

i

t tMAPE

l t

(1) 175

where 176

MAPE = mean absolute percent error between the field measured and the simulated travel time; 177

l= number of movements; 178

tif = the field measured average travel time for movement i which is estimated by using the delay model 179

(sec); 180

tis = the simulated travel time for movement i (sec). 181

182

The calibrated VISSIM simulation model yielded a MAPE value of 8.69% for the average travel time 183

at the selected intersection, indicating that the calibrated simulation model provides reasonable 184

estimates. With the calibrated VISSIM simulation models for TWSC intersections, some similar 185

simulation models were established by changing the traffic control methods for the base condition, 186

including YC intersections, AWSC intersections and roundabouts, as shown in Figure 2. 187

188

The simulation model for each treated intersection was established based on the VISSIM simulation 189

model for the base condition. A node was created in VISSIM so that data could be collected in this area. 190

The outputs of the VISSIM simulation model include average delay (sec), emissions CO (g), emissions 191

NOx (g), emissions VOC (g), and fuel consumption (gal). 192

193

The delay reduction benefit associated with each CM is estimated as the additional value of total delay 194

for the treated intersection over the total delay for the base condition, which is shown as follows: 195

, , ,Ti h B h Ci hB VOT T T (2) 196

24

, ,

1

365Ti yearly Ti h

h

B B

(3) 197

where 198

BTi, h = delay reduction benefits during hour h with CM i (i=1, 2, 3) ($); 199

VOT = the value of time for motorists ($/person hours); 200

TCi,h = total delay during hour h with CM i (h); 201

TB,h = total vehicle delay during hour h for the base condition (h); 202

BTi, yearly = the annual delay reduction benefits for CM i compared with that of the base condition ($). 203

204 It should be noted that in the simulation models, the selected intersection has moderate traffic volume 205

(around 400 vehicles per hour per lane during peak hours), so that we assume that the traffic shift from 206

the intersection to other paths is minor. 207

208

Environmental Impacts 209 Similarly, the environmental impact was calculated as the change in vehicle emission and fuel 210

consumption costs, which are shown as follows: 211

7

, , , , ,+Ei h j Bj h Cij h F B h Ci h

j

B C E E C F F (4) 212

24

, ,

1

365Ei yearly Ei h

h

B B

(5) 213

where 214

BEi, h = the environmental impact during hour h for CM i ($); 215

BEi, yearly = the annual environmental impact for CM i ($); 216

Cj = cost for emission j ($/U.S. ton); 217

ECij,h = total amount of emission j during hour h for CM i (U.S. ton); 218

EBj,h = total amount of emission j during hour h for the base condition (U.S. ton); 219

CF= fuel price ($/gallon); 220

FCi,h = total fuel consumption during hour h for CM i (gallon); 221

FB,h = total fuel consumption during hour h for the base condition (gallon). 222

223

Safety Impacts 224 To estimate the safety impact of each treatment, the change in number of crashes due to the CM was 225

calculated. Safety performance function (SPF) was used to estimate the predicted average crash 226

frequency for a specific site type using a regression model developed from data for a number of similar 227

sites. HSM provides a series of SPFs to estimate the average crash frequency of a TWSC intersection 228

with specified base conditions based on average daily traffic volumes of major and minor roads (5). 229

The predicted average crash frequency includes multiple-vehicle crashes and single-vehicle crashes. 230

Crash modification factors (CMF) are then used to adjust the SPF estimates of predicted average crash 231

frequencies to determine the effects of individual geometric design and traffic control treatments. The 232

equations are shown as follows (5): 233

exp( ln( ) ln( ))k k minbkmv kmajN AADT AADa b c T (6) 234

exp( ' ' ln( ) ' ln( ))minbksv majk k kN AADT AADa b c T (7) 235

SPFk bkmv bksvN N N (8) 236

ikbik SPFkN N CMF (9) 237

, ( )k SPFk bikSi yearly

k

C N NB (10) 238

where 239

k = crash severity (T= all, F = fatal/injury, O= property damage only); 240

Nbksv = predicted average number of single-vehicle collisions for base conditions; 241

Nbkmv = predicted average number of multiple-vehicle collisions for base conditions; 242

AADTmaj = average daily traffic volume (veh/day) for major road (both directions of travel combined); 243

AADTmin = average daily traffic volume (veh/day) for minor road (both directions of travel combined); 244

Nspfk = predicted total average crash frequency for crash type k for base conditions (excluding 245

vehicle-pedestrian and vehicle-bicycle collisions); 246

Nbik = predicted crash frequency for CM i for crash type k (excluding vehicle-pedestrian and 247

vehicle-bicycle collisions); 248

CMFik = crash modification factors for CM i for crash type k, as shown in Table 3 (15); 249

ak, bk, ck, a’k, b’k, c’k = regression coefficients for crash type k, as shown in Table 4 (13); 250

BSi, yearly = the annual safety impact for CM i ($); 251

Ck = crash cost estimates for crash type k, as shown in Table 4 (5). 252

8

253

Life-Cycle Cost Analysis 254 To compare the overall performance of different treatments, the operational, safety and environmental 255

impacts are quantified using monetized expressions in order to establish a common unit. The life-cycle 256

cost analysis (LCC) is used to evaluate the overall long term economic efficiency for investment 257

alternatives (5, 16). The net present value (NPV) for each CM is calculated by combing all costs and 258

benefits or returns associated with a transportation project into a single present value over the life cycle. 259

A positive NPV indicates that the project is economically efficient. The procedure to estimate the NPV 260

of each treatment is illustrated as follows: 261

262

Step 1: Identify the evaluating objectives and performance measurements. In our case, the objectives 263

include traffic operation, traffic safety and environment impacts. 264

265

Step 2: Estimate the change in annual crash costs, delay costs, vehicle emission and fuel consumption 266

costs using equations (2) to (10). 267

268

Step 3: Convert the annual crash costs, delay costs, vehicle emission and fuel consumption costs into a 269

present value. The factor to convert a series of uniform future values to a single present value (P/A, m, 270

n) is calculated as follows: 271

(1 ) 1

( / , , )(1 )

n

n

mP A m n

m m

(11) 272

, , ( / , , )NPV Ti Ti yearlyB B P A m n (12) 273

, , ( / , , )NPV Si Si yearlyB B P A m n (13) 274

, , ( / , , )NPV Ei Ei yearlyB B P A m n (14) 275

where 276

m= discount rate; 277

n= year in service life of the countermeasure; 278

BNPV,Ti = the NPV of the operational impact for CM i ($); 279

BNPV,Si = the NPV of the safety impact for CM i ($); 280

BNPV, Ei= the NPV of the environmental impact for CM i ($); 281

282

Step 4: Calculate the NPV of each CM, which is the sum of the present value of the operational, safety 283

and environmental impacts over the implementation costs. 284

, , , ,+ +NPVi NPV Ti NPV Si NPV Ei NPV IiB B B B B (15) 285

where 286

BNPVi = the NPV of overall impact for CM i ($); 287

BNPV,Ii = the NPV of implementation costs for CM i ($). 288

289

The procedure to conduct the multi-objective evaluation is summarized in Figure 3. 290

291

To account for the variability of the input data in the LCC analysis, the MC method was used to 292

conduct uncertainty analysis by using random sampling from probability descriptions of uncertain input 293

variables to generate a probabilistic description of results. With the uncertainty analysis results, the 294

decision maker knows not only the full range of possible values, but also the relative probability of any 295

9

particular outcome actually occurring. An example of the values or distributions of input variables is 296

shown in Table 5. 297

298

RESULTS 299

300

Sensitivity Analysis 301 Traffic volume is one of the crucial parameters that may affect the impact of different objectives. The 302

tradeoff between those objectives depends on traffic volume from both major and minor roads. So this 303

study evaluates the combined impacts of each treatment under different traffic volume scenarios. With 304

the established VISSIM simulation models, sensitivity analysis can be conducted to identify the effects 305

of traffic volume on travel time, vehicle emission and fuel consumption with each treatment. 306

307

As mentioned in HSM, the SPFs are applicable to the intersection with AADT of the major road 308

ranging from 0 to 46,800 veh/d and AADT of the minor road ranging from 0 to 5,900 veh/d. 309

Meanwhile, as shown in Table 1, the critical volume for selected CMFs ranges from 680 veh/d to 310

15,400 veh/d for both the major and minor roads. Therefore, in the sensitivity analysis, the AADT of 311

the major road ranges from 680 veh/d to 15,400 veh/d, and the AADT of the minor road ranges from 312

680 veh/d to 5,900 veh/d. Field observation showed that the percentage of left-turn and right-turn 313

traffic from the major road and minor road of the selected site are around 20%. The NPV for each CM 314

was calculated under different traffic volume conditions with the same traffic composition. 315

316

Figure 4 illustrates the comparison of annual delay cost, emission cost and fuel consumption cost 317

before and after the treatment. As shown in Figure 4, the annual delay cost can be up to 1 million 318

dollars per year under the selected traffic volume condition. The annual emission cost can reach 60,000 319

dollars per year. And for annual fuel consumption cost the number can increase to 800,000 dollars per 320

year. So for each countermeasure, the annual delay cost always accounts for the largest proportion 321

(over 60%) of the total annual cost, followed by the annual fuel consumption cost (over 30%) and 322

annual emission cost (less than 10%). 323

324

Generally, the annual delay cost, emission cost and fuel consumption cost increase with the raise in 325

AADT on both the major roads and the minor roads, and the increase becomes more sensitive as traffic 326

volumes get higher. As shown in Figure 4, the slopes are gentle when the volumes are low, and become 327

steeper as traffic volumes increase. The slope in Figure 4 (a) turns to be the steepest, indicating that the 328

annual delay cost is the most sensitive to the volume changes. 329

330

Along with the conversion from the base condition to different CMs, the cost changes in different ways. 331

Using CM2 (convert TWSC to AWSC) always costs more for each objective at each volume condition. 332

As compared with the base condition (TWSC), all the three types of costs become higher with the 333

implementation of CM 2 for most traffic volume scenarios. In Figure 4, as the AADT from the major 334

street or minor street changes, the grey layer of CM2 are all above the black layer of base condition. 335

The greatest increase of annual cost by conversion from base condition to CM2 increase occurs with 336

the largest AADT from the major road and the lowest AADT from the minor road. This is reasonable 337

because when major road traffic volume is far more than the minor road, stopping vehicles on major 338

roads results in much more costs. On the contrary, changing from the base condition to CM1 and CM3 339

reduces the costs for every objective. The greatest cost reduction occurs with the greatest AADT from 340

the major road and the minor road. As shown in Figure 4, the red layer of CM1 and blue layer of CM3 341

10

are all below the black layer of base condition. This is reasonable because yielding and roundabout do 342

not need to fully stop vehicles so that the delay and costs from breaking are reduced. In addition, the 343

red layer and blue layer are very close to each other with changing the vehicle volume on major and 344

minor roads. The only difference is when the volumes become very high. It means the costs between 345

CM1 and CM3 are similar until the major and minor roads become very busy. In this scenario, CM1 346

costs more than CM3. 347

348

Ranking of the Selected CMs 349 To combine the different components to obtain the total benefits of each countermeasure, MC method 350

was used to estimate the NPVs before and after the treatment. The range and distribution of NPV can 351

be estimated for each traffic volume condition. Figure 5 illustrates the distribution of NPVs when 352

AADT of major road is 7,100 veh/d and AADT of the minor road is 5,650 veh/d as an example. For 353

each CM, the distribution of NPV incorporating the operational, safety and environmental impacts is 354

compared with the NPV incorporating the safety impact only. As shown in Figure 5, the mean NPV is 355

the highest for CM 2 (TWSC-AWSC), followed by CM 3 (TWSC-Roundabout) and CM 1 (TWSC-YC) 356

when considering the safety impact only. However, when traffic efficiency and environmental impacts 357

are incorporated, the mean NPV for CM 3 (TWSC-Roundabout) is the highest, followed by CM1 358

(TWSC-YC) and CM 2 (TWSC-AWSC). The results indicate that as a countermeasure to improve 359

motorist safety at a TWSC intersection, CM 2 (TWSC-AWSC) tends to be the most effective. However, 360

when considering the external impacts, this countermeasure greatly increases the total delay and stops 361

for motorists, and thus causes negative impacts to traffic flow and the environment. 362

363

CONCLUSION 364 365

This study presents a procedure for evaluating traffic treatments considering multiple objectives. The 366

treatments are safety related countermeasures at TWSC intersections. The multiple objectives 367

considered are traffic efficiency, traffic safety and environmental impacts. The change in total delay, 368

vehicle emission and fuel consumption for each CM is obtained using VISSIM simulation tools. The 369

safety benefits are quantified using a method from HSM based on SPF and CRF. To combine the 370

different objectives, the NPV of each selected CM is calculated, including the change in delay costs, 371

crash costs, vehicle emission costs and fuel consumption costs. On the basis of the estimated results 372

and analysis, the following conclusions can be made: 373

374

AADT of the major and minor roads may influence the external impacts of different treatments. 375

For each CM, the delay costs, vehicle emission costs and fuel consumption costs increase with the 376

increase in major and minor road AADT. The external costs are increased for CM2 (TWSC-AWSC) 377

while decreased for CM1 (TWSC-YC) and CM3 (TWSC-Roundabout). 378

379

For each CM, trade-offs may exist among different objectives. CM1 improves traffic operation and 380

reduces vehicle emissions but may result in some safety problems. CM2 improves traffic safety while 381

increasing vehicle delay and emissions. CM3 improves traffic operation, traffic safety and reduces 382

vehicle emissions under most traffic volume scenarios. When combining all the impacts together, the 383

mean NPV is positive for CM3, negative for CM2, and around zero for CM1 when AADT of the major 384

street equals 7,100 veh/d and AADT of the minor street equals 5,650 veh/d. 385

386

11

Combining multiple objectives in the evaluation could result in reprioritization of traffic 387

improvement projects. This study illustrates the common evaluation for treatments by ranking potential 388

CMs based only on the consideration of the safety impacts. However, by incorporating the external 389

impacts, the prioritization of the selected CMs may change. This is because the total delay, vehicle 390

emission and fuel consumption savings benefits comprise a large proportion of the estimated NPVs, 391

which when added into the total benefits can change the dominant trending of the benefits patterns. 392

393

DISCUSSION 394 395

Although there are specific resources to provide quantification of the impacts of each objective, 396

combining them into the practice of evaluation results in greater challenges than expected. This study 397

summarizes these challenges for other researchers: 398

399

The present documentation provides a series of equations and default parameter values that can be 400

directly applied in conducting traffic efficiency and safety analyses. However, when applying these to 401

other research, data from local studies should be used to calibrate the parameters of the equations. 402

When selecting data from different studies, the best value of each parameter should be chosen. If 403

research to evaluate the operational or safety impacts is unavailable, it is necessary to establish 404

estimation models or simulation techniques. 405

406

The results of this study rely on the values obtained from existing studies, including CMF, VOT, 407

crash cost and emission cost estimates. These estimates may vary over time. For example, the emission 408

cost may change given the penetration of vehicles equipped with new technology. Thus, the values 409

should be updated as new research results become available. 410

411

One of the limitations of this study is that the effects of a treatment are likely to have an impact on 412

traffic flow. The traffic volume and traffic conditions may change after the treatment implemented. In 413

this paper, the selected intersection has moderate traffic volume in simulation models, so that this 414

impact was ignored in the evaluation. However, if the intersection is a key point in the roadway 415

network, the change in traffic flow should be considered. Not only the single location but also the 416

larger area network should be evaluated. The authors recommend that the future research can be 417

expanded by considering this issue. 418

419

Another limitation of this study is that only the impacts of different CMs for motorists are 420

considered is this study. The impacts for other roadway users are not incorporated. When considering 421

multiple objectives, it is necessary to simultaneously consider multiple road users. Among the 422

objectives, those that might have countervailing impacts under different traffic conditions or among 423

different roadway users should be investigated more thoroughly. 424

425

ACKNOWLEDGMENTS 426 427

This research was sponsored by the National Natural Science Foundation of China (Grant No. 428

51608268), the Natural Science Foundation of Jiangsu Province (BK20150747) and the Fundamental 429

Research Funds for the Central Universities (NJ20160016). 430

431

12

REFERENCES 432

433 [1]. U.S. Department of Transportation (2014). Transportation for a new generation, strategic plan for 434

fiscal years 2012-2016. Washington, D.C. [Accessed April 18, 2015]. 435

[2]. Hickman, R., Saxena, S., Banister, D., Ashiru, O. (2012) Examining transport futures with 436

scenario analysis and MCA. Transport Res A-Pol, 46, 560-575. 437

[3]. Sælensminde, K. (2004) Cost-benefit analyses of walking and cycling track networks taking into 438

account insecurity, health effects and external costs of motorized traffic. Transport Res A-Pol, 38, 439

593-606. 440

[4]. Sohn, K. (2011) Multi-objective optimization of a road diet network design. Transport Res A-Pol, 441

45, 499-511. 442

[5]. American Association of State Highway Transportation Officials (AASHTO). (2010a) User and 443

Non-User Benefit Analysis for Highways. Washington, D.C. 444

[6]. Sharma，A., Vanajakshi, L., Girish, V., Harshitha, M. (2012) Impact of Signal Timing 445

Information on Safety and Efficiency of Signalized Intersections. Journal of Transportation 446

Engineering, 138(4), pp.467-478. 447

[7]. Stevanovic, A., Stevanovic, J., Kergaye, C. (2013) Optimization of traffic signal timings based on 448

surrogate measures of safety. Transportation Research Part C, pp. 159-178. 449

[8]. Mendonc C., Freitas, E., Ferreirac, J., Raimundoc, I., Santosa, J. (2013) Noise abatement and 450

traffic safety: The trade-off of quieter engines and pavements on vehicle detection. Accident 451

Analysis and Prevention, 11-17. 452

[9]. Metropolitan Transportation Commission (MTC). Transportation 2035 Plan for the San 453

Francisco Bay Area, 2009. http://www.mtc.ca.gov/planning/2035_plan/. [Accessed October 30, 454

2015]. 455

[10]. Transportation Research Board (TRB). (1989) NCHRP 320: Guidelines for Converting Stop to 456

Yield Control at Intersections. National Research Council, Washington, D.C. 457

[11]. Carrie L. Simpson & Joseph E. Hummer (2010) Evaluation of the Conversion from Two-Way 458

Stop Sign Control to All-Way Stop Sign Control at 53 Locations in North Carolina, Journal of 459

Transportation Safety & Security, 2:3, 239-260. 460

[12]. Federal Highway Administration (FHWA). Roundabouts: An Information Guide, 461

FHWA-RD-00-67, Exhibit 5.2, pp 106, Washington, DC, June 2000. 462

(www.tfhrc.gov/safety/00068.htm). 463

[13]. American Association of State Highway Transportation Officials (AASHTO). (2010b) Highway 464

Safety Manual. Washington, D.C. 465

[14]. Walls III, J. and Smith, M. R. (1998) Life-Cycle Cost Analysis in Pavement Design. Report No. 466

FHWA-SA-98-079. 467

[15]. Crash Modification Factors Clearing House. http://www.cmfclearinghouse.org. [Accessed 468

October 30, 2015]. 469

[16]. Caltrans. Life-Cycle Benefit-Cost Analysis Economic Parameters 2012. 470

http://www.dot.ca.gov/hq/tpp/offices/eab/benefit_cost/LCBCA-economic_parameters.html. 471

[Accessed June 30, 2016]. 472

473

13

Nomenclature 474

475 List of symbols and abbreviations 476

477

CM countermeasure 478

TWSC two-way stop-controlled 479

YC yield-controlled 480

AWSC all-way stop controlled 481

SPF safety performance function 482

CMF crash modification factor 483

LCC life cycle cost 484

MC Monte Carlo 485

MAPE mean absolute percent error 486

NPV net present value 487

HSM Highway safety manual 488

AADT average annual daily traffic 489

L number of movements 490

tif the field measured average travel time for movement i which is estimated by using the 491

delay model (sec) 492

tis the simulated travel time for movement i (sec) 493

BTi, h delay reduction benefits during hour h with CM i (i=1, 2, 3) ($) 494

VOT the value of time for motorists ($/person hours) 495

TCi,h total delay during hour h with CM i (h) 496

TB,h total vehicle delay during hour h for the base condition (h) 497

BTi, yearly the annual delay reduction benefits for CM i compared with that of the base condition 498

($) 499

BEi, h the environmental impact during hour h for CM i ($) 500

BEi, yearly the annual environmental impact for CM i ($) 501

Cj cost for emission j ($/U.S. ton) 502

ECij,h total amount of emission j during hour h for CM i (U.S. ton) 503

EBj,h total amount of emission j during hour h for the base condition (U.S. ton) 504

CF fuel price ($/gallon) 505

FCi,h total fuel consumption during hour h for CM i (gallon) 506

FB,h total fuel consumption during hour h for the base condition (gallon) 507

k crash severity (T= all, F = fatal/injury, O= property damage only) 508

Nbksv predicted average number of single-vehicle collisions for base conditions 509

Nbkmv predicted average number of multiple-vehicle collisions for base conditions 510

AADTmaj average daily traffic volume (veh/day) for major road (both directions of travel 511

combined) 512

AADTmin average daily traffic volume (veh/day) for minor road (both directions of travel 513

combined) 514

ak, bk, ck regression coefficients for multiple-vehicle collisions 515

a’k, b’k, c’k regression coefficients for single-vehicle collisions 516

Nspfk predicted total average crash frequency for crash type k for base conditions (excluding 517

vehicle-pedestrian and vehicle-bicycle collisions) 518

14

Nbik predicted crash frequency for CM i for crash type k (excluding vehicle-pedestrian and 519

vehicle-bicycle collisions) 520

CMFik crash modification factors for CM i for crash type k 521

BSi, yearly the annual safety impact for CM i ($) 522

Ck crash cost estimates for crash type k 523

m discount rate 524

n year in service life of the countermeasure 525

BNPV,Ti the NPV of the operational impact for CM i ($) 526

BNPV,Si the NPV of the safety impact for CM i ($) 527

BNPV, Ei the NPV of the environmental impact for CM i ($) 528

BNPVi the NPV of overall impact for CM i ($) 529

BNPV,Ii the NPV of implementation costs for CM i ($) 530

531

532

15

533

List of Figures 534 535

Figure 1 An example of the selected TWSC intersection 536

537

Figure 2 VISSIM simulation models 538

539

Figure 3 The procedure to conduct multi-objective evaluation 540

541

Figure 4 Cost comparison before and after the treatment 542

543

Figure 5 Distribution of NPVs (AADTmaj=7,100veh/d, AADTmin=5,650veh/d): (a) considering 544

multiple objectives and (b) considering the safety impacts only 545

546

547

List of Tables 548 549

Table 1 Selected sites for field data collection 550

551

Table 2 Default hourly traffic distribution 552

553

Table 3 Crash modification factors for selected CMs 554

555

Table 4 Regression coefficients 556

557

Table 5 Probability distribution of input values 558

559

16

560 Figure 1. An example of the selected TWSC intersection 561

562

17

563 Figure 2. VISSIM simulation models 564

565

18

Identify evaluating objectives

- Traffic Operation

- Traffic safety

- Environment

Identify evaluating objectives

- Traffic Operation

- Traffic safety

- Environment

Safety impactSafety impact

Prioritize the alternativesPrioritize the alternatives

Environmental impactEnvironmental impact

Calculate the NPV of each CMCalculate the NPV of each CM

, ( )k SPFk bikSi yearly

k

C N NB

24

, ,

1

365Ei yearly Ei h

h

B B

SPFk bkmv bksvN N N

ikbik SPFkN N CMF

exp( ln( ) ln( ))k k minbkmv kmajN AADT AADa b c T

exp( ' ' ln( ) ' ln( ))minbksv majk k kN AADT AADa b c T

Operational impactOperational impact Safety impactSafety impact Environmental impactEnvironmental impact

, , ( / , , )NPV Ti Ti yearlyB B P A m n , , ( / , , )NPV Si Si yearlyB B P A m n

, , ( / , , )NPV Ei Ei yearlyB B P A m n

Convert to a present valueConvert to a present value

(1 ) 1( / , , )

(1 )

n

n

mP A m n

m m

, , , ,+ +NPVi NPV Ti NPV Si NPV Ei NPV IiB B B B B

Operational impactOperational impact

24

, ,

1

365Ti yearly Ti h

h

B B

, , ,Ti h B h Ci hB VOT T T

, , ,

, ,+

Ei h j Bj h Cij h

j

F B h Ci h

B C E E

C F F

566 Figure 3. The procedure to conduct multi-objective evaluation 567

568

19

569 Figure 4. Cost comparison before and after the treatment 570

571

20

572 Figure 5. Distribution of NPVs (AADTmaj=7,100veh/d, AADTmin=5,650veh/d): (a) considering 573

multiple objectives and (b) considering the safety impacts only 574 575

576

21

Table 1. Selected sites for field data collection 577

Sites City Type Wa (ft)

1 San Pablo Ave & Harrison St Berkeley, California TWSC 12

2 San Pablo Ave & Hearst Ave Berkeley, California TWSC 12

3 San Pablo Ave & Jones St Berkeley, California TWSC 12

4 San Pablo Ave & Camelia St Berkeley, California TWSC 12

5 Shattuck Ave & Virginia St Berkeley, California TWSC 10

6 Shattuck Ave & Berkeley Way Berkeley, California TWSC 10 aWidth of median nose. 578

579

22

580

TABLE 2. Default hourly traffic distribution 581

Hour % ADTa Hour % ADT Hour % ADT Hour % ADT

0-1 1.2 6-7 5.1 12-13 5.6 18-19 5.9

1-2 0.8 7-8 7.8 13-14 5.7 19-20 3.9

2-3 0.7 8-9 6.3 14-15 5.9 20-21 3.3

3-4 0.5 9-10 5.2 15-16 6.5 21-22 2.8

4-5 0.7 10-11 4.7 16-17 7.9 22-23 2.3

5-6 1.7 11-12 5.3 17-18 8.5 23-24 1.7 a Average daily traffic 582

583

584

23

Table 3. Crash modification factors for selected CMs 585

Countermeasures

CMF

Crash

Type

Crash

Severity

Major Road

Traffic

Volume

(veh/d)

Minor Road

Traffic

Volume

(veh/d)

Mean Std.

Error

CM 1 TWSC-YC 2.27 1.26 All All Not Specified Not Specified

CM 2 TWSC-AWSC 0.319 0.022 All All Not Specified Not Specified

CM 3 TWSC-

Roundabout

0.71 0.11 All All

680-15,400 680-15,400 0.19 0.09 All

Serious

Injury,

Minor

Injury

586

587

24

Table 4. Regression coefficients 588

Crash Type Crash Severity Intercept AADTmaj AADTmin

Multiple-Vehicle Crashes

ak bk ck

T -8.90 0.82 0.25

F -11.13 0.93 0.28

O -8.74 0.77 0.23

Single-Vehicle Crashes

a'k b'k c'k

T -5.33 0.33 0.12

F NbFsv= 0.28×NbTsv

O -7.04 0.36 0.25

589

590

25

Table 5 . Probability distribution of input values 591

Parameters

Types of

Probability

Distribution

Values

Average fuel price ($/Gallon) (16) Constant 3.714

VOT ($/person hours)( 16) Constant 12.50

Crash cost estimates

($)

(13)

Fatality (K) Constant 4,008,900

Disabling injury (A) Constant 216,000

Evident injury (B) Constant 79,000

Fatal/Injury (K/A/B) Constant 158,200

Possible injury (C) Constant 44,900

PDO (O) Constant 7,400

Emission cost

estimates ($)

(16)

Carbon Monoxide

(CO) Constant 75

Nitrogen Oxide

(NOx) Constant 17,300

Volatile Organic

Compounds (VOC) Constant 1,210

Implementation cost ($)

(11, 12)

Uniform

(min, max)

CM 1 Uniform (4,430,

5,000)

CM 2 Uniform (4,430,

5,000)

CM 3 Uniform(194,00

0, 500,000)

Annual mantaintance fees ($) Constant

CM 1 200

CM 2 200

CM 3 1,000

Interest rate

Triang (min,

most likely,

max)

Triang (0.03,0.04,0.05)

CMF Normal (avg,

std) See Table 3

592