LRT Passengers’ Responses to Advanced Passenger

Draft

LRT Passengers’ Responses to Advanced Passenger

Information System (APIS) in case of Information Inconsistency and Train Crowding

Journal: Canadian Journal of Civil Engineering

Manuscript ID cjce-2017-0559.R1

Manuscript Type: Article

Date Submitted by the Author: 06-Dec-2017

Complete List of Authors: Kattan, Lina; University of Calgary Schulich School of Engineering

Bai, Yuan; University of Calgary Schulich School of Engineering

Is the invited manuscript for consideration in a Special

Issue? : N/A

Keyword: multinomial logistic model, stated preference, realtime transit information, crowded train, inconsistent information

https://mc06.manuscriptcentral.com/cjce-pubs

Canadian Journal of Civil Engineering

Draft

1

1

LRT Passengers’ Responses to Advanced Passenger Information System (APIS) in case of 2

Information Inconsistency and Train Crowding 3

4

Lina Kattan PEng, PhD 5 Professor 6 Urban Alliance Professor in Transportation Systems Optimization 7 Department of Civil Engineering, Schulich School of Engineering 8 University of Calgary 9 e-mail: [email protected] 10 11

12

Yuan Bai, MEng, 13 Department of Civil Engineering, Schulich School of Engineering 14

University of Calgary 15

email: [email protected] 16 17

Abstract: 18

This research explores and attempts to understand transit riders’ behavioural responses towards real-time 19

transit information for two specific situations: the presence of inconsistent information on transit service 20

recovery and the effects of crowded trains during rush hours. A survey was designed and conducted to 21

collect Light Rail Transit (LRT) riders’ behavioural responses in Calgary, Alberta. Multinomial logit 22

models were developed and calibrated to explore the effects of the described scenarios on riders’ 23

responses. The results led to the conclusion that socio-economic attributes, experience with APIS system, 24

familiarity with public transit in general and Calgary’s LRT system in particular, and the characteristics 25

of origin LRT stations had strong influences on travellers’ behavioural responses. It was also determined 26

that travellers’ actions vary significantly depending on the purpose of the trip, time of the trip, and 27

weather conditions. 28

29

30

Page 1 of 34



Draft

2

Introduction: 31

Over the past two decades, public transit agencies across North America and Europe have increasingly 32

implemented Advanced Passenger Information Systems (APIS). APIS offers considerable benefits in 33

terms of customer satisfaction in the short term (Gooze et al. 2013) and a possible increase in ridership in 34

the long term (Tang and Thakuriah 2012; and Body 2007). APIS improves overall satisfaction and makes 35

transit service a more attractive option for travellers. For instance, up-to-date transit information 36

decreases the uncertainty in waiting time (Rahman et al. 2013), creates a perception that the system is 37

safer and more secure (Dziekan and Kottenhoff 2007) and helps transit riders productively manage their 38

waiting time (Russell 2012). 39

40

Given the role that APIS can play on a transit system level of service and performance, it is 41

crucial to understand the attributes that influence a transit rider’s decisions in response to real-time transit 42

information. However, passengers’ responses change from situation to another. This paper explores 43

transit riders’ behavioural responses to APIS for two particular situations: crowded trains and the 44

existence of conflicting transit arrival information. There might be adjusting strategies, such as letting a 45

crowded bus go by if the display showed another arriving shortly. In particular, if disseminated 46

information on service recovery is inconsistent, it might lead to different responses from the riders. 47

Further investigations of modelling behavioural riders’ response to APIS is conducted by including the 48

effect of trip type and station characteristics. 49

The paper is organized as follows: In the next section, related studies in the APIS area are 50

reviewed. Section 3 gives a brief description of the study area selected to conduct this research. Section 4 51

discusses the questionnaire design and data collection techniques, and Section 5 summarizes the 52

descriptive statistics. Section 6 discusses the analysis of transit riders’ responses to APIS in the two 53

scenarios and the interpretation of the analysis results. Section 7 presents the concluding remarks based 54

on the analysis, proposes some policy recommendations, and outlines the scope for potential future 55

research. 56

Page 2 of 34



Draft

3

57

Literature Review 58

Zhang (2010) proposed a decision tree for a traveller’s change in trip-specific behaviour because of a 59

long expected wait time as indicated by APIS. When APIS notifies a transit rider of a long wait time, 60

he/she may forgo the trip or shift to another mode. For instance, the Stopwatch project in Southampton 61

found that 12.6% of respondents would leave the bus stop because of the delays (Warman 2003). Still, 62

passengers may change from the intended transit path for two reasons. First, long waiting times from 63

APIS might cause passengers to change stops to take a transit route with less total travel time (sum of 64

walking, waiting, and riding time). Second, when there are two transit lines at one stop, passengers may 65

choose the other route if it has less total waiting and riding time. OneBusAway conducted a stated 66

preference survey and found that 78% of respondents reported they were more likely to walk to a 67

different stop if there were delays (Rutherford et al. 2010). The survey also indicated that the most 68

popular choice was to wait at the same stop but for a different bus route, followed by walking to stops 69

further ahead on the current route. In addition, approximately 42% of respondents indicated they walked 70

to another stop for exercise, which is beneficial for health reasons (Ferris et al. 2010). Decisions that 71

transit riders made depended on trip purpose, type and duration of the disruption, trip time, and weather 72

conditions (Bai and Kattan 2014). In addition, socio-economic attributes, experience with APIS, and 73

experience with transit also influenced the passengers’ decisions. 74

While staying on the intended path, productively managing their time around stop is another 75

choice a passenger can make a choice to cope with a long wait. In the literature, 20% to 38.5% of 76

respondents said they would leave the stop and return when the bus is due (Nijkamp et al. 1996; SAIC 77

2003; Caulfield 2004). Undoubtedly, knowing the expected wait time would help riders plan their 78

activities while waiting (Russell 2012). King county metro OneBusAway users routinely comment about 79

their ability to grab a cup of coffee because they know there is a 10-minute late (Rutherford et al. 2010). 80

In general, some transit users keep busy, for example by making phones calls or reading, while waiting. 81

Bai and Kattan (2014) found that while the decision of productively managing their waiting time around 82

Page 3 of 34



Draft

4

the stop varies significantly depending on the type and duration of service interruption, it changes only 83

slightly based on trip purpose. In the case of a 10-minutes Light Rail Transit (LRT) delay, 47.57% and 84

49.54% of respondents stated they would use their time more effectively (e.g. reading a newspaper or 85

using the phone) while waiting for commute and non-commute trips, respectively. However, in the case 86

of service interruption with no expected recovery time, only 15.43% and 17.16% of riders said they 87

would wait for the LRT while actively using their time for commute and non-commute trips, 88

respectively. Consequently, psychological and behavioural responses to APIS can differ significantly 89

based on the presented situation. 90

The effect of conflicting information regarding train arrivals on a transit rider’s behavioural and 91

psychological responses is largely ignored in the literature. Mulley et al. (2017) pointed out to need of 92

examining the non-homogeneity of real time passenger transit information. Accurate, consistent 93

information disseminated from two different media sources has a positive effect on behavioural 94

reinforcement and, thus, a driver’s compliance with real-time traffic information (Kattan et al. 2012). 95

However, the effect of conflicting information from different sources is less understood. Gooze et al 96

(2013) highlighted the importance of reliable real-time transit arrival information. They found that riders 97

expected an average margin of error of 4-6 minutes for arrival predictions. However, infrequent riders, 98

compared to frequent riders, expected an even lower margin of error. They found that passengers with a 99

lower tolerance for errors might decide to take public transit less often. Rahman et al (2013) found that 100

the margin of error increases with a longer transit headway. Cat and Gkioulou (2017) pointed out that 101

while the provision of real-time passenger information has the potent to reduce travel uncertainty, its 102

impacts depends on the underlying service reliability and its perceived credibility. Thus, different 103

behavioural responses may be caused by a greater information error gap, especially when the information 104

is communicated from different sources. 105

Cats (2011) suggested the presence of interdependence between crowded trains and APIS. 106

Crowding might result in unrealistic passenger waiting time as passengers need to wait for a complete 107

headway or headways to be able to board. Various studies indicate that crowding is a concern in places 108

Page 4 of 34



Draft

5

like Los Angeles (Davidson et al. 2011), London (Wardman and Whelan, 2011), Delft (Pel et al. 2013), 109

and Dhaka (Katz and Garrow 2012). A stated-preference study conducted by Hensher et al. (2011) 110

showed the significant impact of train crowding in increasing travelers’ value of time savings. Research 111

on passengers’ responses to overcrowding often focused on the reaction to the crowded train, the 112

relationship between the level of overcrowding and stress, and incentives to reduce overcrowding. 113

Preston et al (2016) found that the willingness of passengers to change their behavior to avoid crowding. 114

Based on a simulation analysis conducted by Drabicki et al. (2017), the provision of crowding 115

information in real-time was shown to significant impact en-route path choices. These choices were also 116

found to be strongly related with network congestion level, passengers' behavior in terms of sensitivity to 117

crowding and characteristics of information provision. According to the Pel et al. (2013), MVA 118

Consultancy (2007), Maunsell (2007), and Davisdon et al. (2011), a user reaction to overcrowding can be 119

any of the following: changing departure times to avoid crowded conditions; using another departure or 120

arrival station; changing to a different route; switching to another mode; choosing less crowded 121

carriages; standing on the platform in the exact position that the train doors line up; and upgrading to 122

another class In a recent study using smart card data, Kim et al. 2015) found that travelers do take into 123

consideration crowding conditions in their route choices. In light of real time transit information, 124

passengers may choose different alternatives when faced with crowded trains, for example, by letting a 125

crowded train go by or changing routes if the display showed another one arriving shortly. Cats (2011) 126

suggested that the provision of traffic information with crowding situation could have big effects on 127

passenger route choice decisions. In a survey conducted in Korea, Kim et al. (2009) found that 128

information on crowdedness on a bus decreases the probability that a rider will choose to board on the 129

incoming bus. Based on a recent pilot study at a Stockholm metro station, Zhang (2017) found that the 130

provision of real time information on train crowding is successful in reducing the passengers boarding 131

the first, most crowded car by 4.3% points on trains that were crowded on arrival. A stated preference 132

work undertaken in the UK (Pritchard 2017) showed that the provision of improved information could in 133

some cases be effective, in helping passengers make informed decision to avoid train crowding. 134

Page 5 of 34



Draft

6

Thus, two situations that affect passengers’ responses to APIS, which are overlooked in the 135

literature and are the focus of this paper are as follows: 136

1. Inconsistent recovery information in the case of transit service interruption 137

2. Interaction between real-time information and train crowding 138

Study Area Profile 139

This research uses the Calgary’s LRT system as a case study. Calgary’s LRT system had an average of 140

285,000 riders per weekday in the fourth quarter of 2012, making it the busiest light rail system in North 141

America (American Public Transportation Association 2012). In a 2012 transit customer satisfaction 142

survey, the crowding level of the LRT system was the only attribute that did not rate well among most 143

respondents. 144

145

As of August 2014, the Calgary LRT system operates on 59.9 kilometers of track and is organized into 146

two routes: Route 201 Tuscany/Somerset-Bridlewood (red line in Figure 1) and Route 202 147

Saddletowne/69 Street W (blue line in Figure 1). The current average headway is 300 seconds during 148

peak periods and 600 seconds during off-peak periods. 149

150

Calgary and Region Travel and Activity Survey (CARTAS 2012) revealed the total number of transit 151

trips per person is statistically the same as in 2001 (City of Calgary 2012). Overall, most transit users 152

(73%) are still making two transit trips per day. According to 2015 data collected by Calgary Transit, 153

yearly transit ridership reached 110 million customers in 2016. 17% of Calgary transit users reported 154

only using the bus service, 37 % only using the LRT and 46% using both bus and LRT (Calgary Transit 155

2017). The transit mode share for home to work trips located in the city Central Business District (CBD) 156

increased from 36% to 46% from 2001 to 2011. However, the city has the following modal splits: 157

vehicle - 77%, public transit - 8.6%, walking - 12.4%, and bicycle - 2%. The overall low transit usage is 158

explained by the predominance of low-density areas underserved by public transport. 159

Page 6 of 34



Draft

7

160

On November 1, 2011, Calgary Transit introduced APIS to most LRT stations. APIS allows Calgary 161

Transit to disseminate information to riders on the estimated arrival time of the next three LRTs arriving 162

at a given stop, possible delays, and service disruptions. This information is disseminated in real time 163

through the digital signs and automated public address systems. This information was not disseminated 164

through mobile applications until April 2015. 165

166

167

Figure 1: Calgary LRT map (Source: The City of Calgary. Reprinted with permission) 168

169

Transit service interruption is generally due to unavoidable situations like medical emergencies, 170

weather conditions, traffic accidents, mechanical failures, or electrical failures. When Calgary LRT 171

service is interrupted for a prolonged period, the transit staff announces expected arrival time through an 172

audio broadcasting system. This estimated service recovery time could be inconsistent with the 173

information disseminated by at-stop APIS. In other words, an audio announcement might broadcast that 174

the LRT service will be restored in 20 minutes, while the APIS might indicate that real-time estimated 175

arrival time is 10 min. This discrepancy in the expected arrival time is due to the fact that the APIS 176

system relies on a commercially available software that estimates the LRT arrival based on the average 177

LRT speed collected at various intervals. Thus, in case of an unexpected service disruption, while the 178

Page 7 of 34



Draft

8

train driver would report this incident by calling directly the Transit Control center, the APIS would 179

continue to automatically displaying the expected arrival as the transit staff does not have the ability to 180

override the APIS. 181

182

Data Collection 183

In this study, a questionnaire-based survey was developed to analyze transit riders’ behavioural responses 184

and psychological perceptions to real-time information provided by APIS. The questionnaire took an 185

average of 10 minutes to complete and was approved by the Office of Research Ethics at the University 186

of Calgary. Two major destinations are targeted in the survey, University of Calgary and downtown 187

Calgary, as they are the two major destinations in the city that are well served by LRT/transit. The online 188

survey and face-to-face interview were conduct from February to April 2014, almost one and half year 189

after the implementation of APIS. The full questionnaire includes 29 questions and can be found in Bai 190

(2014). The collected information is divided into four broad categories: 191

− Travel characteristics, such as frequency of trips using LRT, primary mode of travel 192

− Riders’ experience with the LRT and APIS, such as the familiarity with the LRT service frequency, 193

familiarity with APIS, and perceived accuracy of APIS 194

− Demographic information, such as age, gender, income, household vehicle ownership 195

− Transit riders’ behavioural responses to different scenarios for different trip purposes (work/school 196

and other trips) and weather conditions (15℃ to 25℃, no rain expected; and -25 ℃ to -5℃, no snow 197

expected) 198

In addition, information on trip attributes, such as in-vehicle-travel time (IVT), number of transfers, and 199

station characteristics were captured indirectly from the questionnaire by indicating the trip station origin 200

and trip destination. The LRT IVT and number of bus transfers from origin and destination stations were 201

obtained from the Plan A trip tool on the Calgary Transit website for each origin-destination pair. 202

203

Page 8 of 34



Draft

9

The respondents’ origin and destination stations in the survey covered the majority of the LRT stations in 204

Calgary, except the Barlow/Max Bell and 39 Ave stations. Four groups of variables were collected to 205

capture the essential characteristics for each LRT station: LRT station bus accessibility, station types, 206

location of the station, and surrounding land use and road characteristics, as follows: 207

208

Station characteristics Parameters

LRT station bus

accessibility • Bus accessibility level index (AI)

Station types and location • In-community platforms or large enclosed steel and glass structures

• Whether station is located in the city centre, near a major activity

centre, close to a community centre, in the inner city, and in an

established community

• Whether the station is the end-of-line terminal

Surrounding land use

Road characteristics

• Whether sufficient parking is provided around the LRT stations

• Whether there are coffee shops, grocery stores, convenience stores or

restaurants located within 500 meters of the LRT stations

• Road and street network: skeletal road, arterial road, and city centre

209

Transit riders’ behavioural responses to APIS were collected in four scenarios: 210

211

1. A regular weekday with APIS showing that the next LRT was delayed, but would arrive in 10 minutes 212

2. An interruption of the LRT service due to weather or an incident, but no estimated recovery time was 213

displayed 214

Page 9 of 34



Draft

10

3. Riders’ responses to scenarios when conflicting information was presented 215

4. An oncoming crowded LRT with APIS estimated the next LRT arrival time was 5 minutes. 216

217

This paper focuses on analyzing scenarios 3 and 4; analyses of scenarios 1 and 2 were published in a 218

previous paper (Bai and Kattan 2014). Scenarios 3 and 4 were investigated separately for different trip 219

purposes, time, and weather conditions. Calgary’s climate falls under the “humid continental” 220

classification with long, cold winters and short, moderately warm summers. Such weather conditions 221

significantly affect travellers’ attitudes to waiting time. 222

The scenario related to inconsistent APIS and audio information was presented as follows in Table 1 223

below: 224

225

Table 1: Conflicting Information Scenario 226

Page 10 of 34



Draft

11

227

228

229

The scenario related to incoming crowded LRT only focuses on peak period when trains are crowded. 230

This scenario was presented as follows in Table 2: 231

Table 2: Scenario related to Crowded LRT 232

Page 11 of 34



Draft

12

233

Five hundred and five responses were collected from LRT riders. Two approaches were used to obtain 234

the responses from the riders: an online survey and face-to face interviews. The online survey link was 235

created using the Survey Monkey application, and was distributed to staff and students at the University 236

of Calgary and to staff at the City of Calgary and Calgary Transit. The face-to-face interviews were 237

conducted on LRT stations/trains and in the university area. Sixty-nine responses, 14% of the total 238

sample, were gathered from the online survey. In addition, 436, which accounted for 86% of the total 239

sample, were gathered from the face-to-face interviews. 240

241

Descriptive Statistics 242

Table 3 provides the descriptive statistics. The data for most of variables were recorded in categories; 243

thus, a set of dummy variables was created for each variable. 244

245

246

247

248

249

250 Table 3: Descriptive statistics of the variables considered in the study 251

Variables Description Percentage Variables Description Percentage

Primary mode of travel Perceived accuracy

Vehicle 32.08% Almost always no

difference 38.42%

Page 12 of 34



Draft

13

252

The majority of the respondents (51%) were under 25 years of age. This indicates that the sample over-253

represented the age group under 25-year old. CARTAS also found transit users are more likely to be 254

younger (City of Calgary 2012). In addition, 50% of the respondents were male and these results are 255

similar to the findings of the 2012 Customer Satisfaction and Non-User Survey conducted by Calgary 256

Transit. As described in the next section, a weight factor was assigned to each observation, based on the 257

estimated Calgary transit customers and the age and gender cross tabulation available from Calgary 258

Transit’s 2012 customer satisfaction survey, to obtain an unbiased estimation in the multinomial logit 259

regression. 260

261

In the questionnaire, the respondents indicated their frequency of using the LRT, their main mode of 262

travel, and their familiarity with the LRT service. Fifty-seven percent of respondents reported using the 263

Transit 64.54%

Sometime LRT arrives

earlier than displayed

time

15.64%

Other 3.36%

Sometime LRT arrives

later than displayed

time

39.61%

Missing 0.02% Do not know 5.54%

Frequency of using LRT Missing 0.79%

< 1 day/week 17.80% Age

1-4 days/week 24.80% Under 25 50.90%

≥ 5 days/week 57.50% 25 to 34 19.60%

Familiarity with LRT service 35 to 44 16.60%

Very familiar 53.70% 45 to 54 8.70%

Familiar 41.40% Over 55 4.00%

Not familiar 4.90% Missing 0.20%

Familiarity with APIS Gender

Very familiar 56.80% Male 49.70%

Familiar 39.60% Female 50.10%

Not familiar 3.60% Missing 0.20%

Driver's license Vehicle ownership continuous variable,

Having 85.15% Mean = 1.87, SD = 1.11

Not having 14.46%

Missing 0.40%

Page 13 of 34



Draft

14

LRT more than 5 days/week. The results for the main mode of travel showed that a high percentage of 264

respondents use public transit. The results also show that about 32% of respondents used cars as their 265

main mode of travel. This low percentage of vehicle users could be explained by the fact that the survey 266

targeted transit users. The survey also targeted two major destinations, downtown and the University of 267

Calgary campus, which are well served by transit. 96% of respondents indicated being familiar with the 268

real-time information display. 269

270

Responses to the examined Scenarios 271

Figures 2 a and b summarizes the responses according to scenario and trip purpose. When the LRT 272

service was interrupted with inconsistent audio and APIS, most respondents stated that they would still 273

wait for the LRT service to be restored (49.46% and 51.90 % for commute and non-commute trips, 274

respectively). Fewer respondents choose to change to a car or taxi (26.56% and 25.62% for commute and 275

non-commute trips, respectively), walk to another LRT station (4.73% and 6.30% for commute and non-276

commute trips, respectively), Take a bus from the LRT station (18.28% and 15.20% for commute and 277

non-commute trips, respectively) and other (0.97% and 0.98% for commute and non-commute trips, 278

respectively). Walking to the next station is only a realistic option for downtown stations. Respondents 279

were also asked which source of information they would trust more if they encounter conflicting 280

information. Around 43% of the respondents stated they would trust the audio announcements more and 281

roughly 20% of respondents stated they would not trust the audio announcements nor the station 282

displays. Interestingly, responses were similar even when the discrepancy between the two information 283

sources was larger. 284

285

In the case of crowded trains, the difference in behaviour between regular commutes and non-regular 286

commutes significantly varies. The majority of respondents stated that they would squeeze into the first 287

train in the case of commute trips (48.68%), while the majority opted to wait for the next train in the case 288

of non-commute trips (64.90%). 289

Page 14 of 34



Draft

15

290

Modeling Behavioural Responses using Multinomial Logit Modelling 291

292

Since the actions selected by the travellers in the hypothetical scenarios were nominal in nature, the 293

multinomial logit model was used for statistical modeling. Several multinomial logit models analyzed 294

travellers’ responses for the different trip purposes in each scenario. 295

296

For model formulation, the probability of a respondent, n, being involved in an action, i, can be written 297

as: 298

�� = �� > �� ⩝ ∈ � (1) 299

300

Where P�� is the probability of a respondent being involved in an action, ; 301

�� is a function of covariates that determines the likelihood of respondent, �, being involved in 302

action ; 303

� is the set of all possible actions that are available to the respondents. 304

The covariate function is linear: 305

306

�� = �� + �� (2) 307

308

Where β� is a vector of the estimated coefficients; 309

�� are the measurable characteristics that can determine the categories of the possible actions; 310

ε�� is an error term that is used to address the unobserved factors that influence the actions taken 311

by the individuals. 312

Therefore, the logit function for respondent, �, choosing action, , can be written as: 313

P�� =�� !

∑ �� !⩝�∈# (3) 314

Page 15 of 34



Draft

16

315

Figure 2 a: Respondents’ responses to conflicting information scenario 316

317

318

Figure 2b: Respondents’ responses when a crowded LRT arrives and the next LRT arrival time is 319

indicated as 5 minutes 320

Figure 2 a: Respondents’ responses to the examined scenarios 321

0%

5%

10%

15%

20%

25%

30%

35%

Change to a car or

a taxi

Take a bus from

the LRT station

Walk to another

LRT station to

catch the same

LRT

Do something

while waiting for

the next LRT

(reading a

newspaper, use the

phone, etc.)

Just wait for the

next LRT

Other

Commuter trips (work/ school) during peak hours

Non-commuter trips (such as shopping and recreation) during off-peak hours

0%

10%

20%

30%

40%

50%

60%

70%

Squeeze into the first

LRT

Let the crowded LRT

go by, and take the

next one

Walk or take the LRT

to the other LRT

station

Go somewhere else

and return back when

the LRT is likely to be

less crowded

Other

Commuter trips (work/ school) during peak hours

Non-commuter trips (such as shopping and recreation) during off-peak hours

Page 16 of 34



Draft

17

Each observation has an assigned weight to correct sampling bias and to obtain an unbiased estimation. 322

The following formula is implemented in logit software using a weight factor, ω, for each observation to 323

show how the log-likelihood (LL) function is weighted. 324

325

LL = ∑ ω�ln�P��∈( (4) 326

Where ω� is the weight of a respondent, which is the inverse of the probability of the observation. The 327

population used in the analysis was based on the estimated Calgary transit customers and the age and 328

gender cross tabulation available from Calgary Transit’s 2012 customer satisfaction survey. 329

330

Multinomial Logit Model Results and Discussions 331

One of the underlying assumptions in a multinomial logit model is the independence of irrelevant 332

alternatives (IIA). Alternative 4, do something while waiting for the next LRT, and alternative 5, just 333

wait for the next LRT, are interrelated. When the IIA assumption is violated, one of the simplest 334

solutions is grouping the relevant alternatives together. There were few responses for alternatives 2, 335

taking a bus from the LRT station, and 3, walking to another LRT station to catch the same LRT, and 6, 336

other. Since alternatives 2 and 3 were related to transit use, these two options were merged into “other 337

transit” options, and the “other” alternative was removed. 338

The calibrated models summarize the impacts of travel habits, travellers’ experience with the 339

Calgary LRT APIS, demographics, station characteristics, and weather conditions on the selected 340

alternatives. Multinomial logit model estimations are provided in Table 4. 341

Coefficients were estimated for a large number of variables using STATA data analysis and 342

statistical software (www.stata.com). Coefficients with p values greater than 0.1 were insignificant and 343

were not included in the final model. As suggested by Kockelman and Kweon (2002), variables with low 344

statistical significance may be retained in the model if they belong to categorical factors that had some 345

significant effect. Thus, some of the variables that were not significant were still retained in the model, as 346

long as at least one of the variables for the same factor was statistically significant. This approach may 347

Page 17 of 34



Draft

18

reduce the efficiency of the estimates, which were adjusted by using a more liberal confidence level of 348

90%, instead of the traditional 95%. In the following sections, the analysis was based on the calibrated 349

multinomial logit model. 350

351

Analysis of riders’ stated responses to inconsistent information 352

The alternatives were as follows: changing to a car or a taxi, waiting for the next LRT, and other transit. 353

Waiting for the next LRT was selected as the reference category (i.e. Alternative 1) for the conflicting 354

information scenario. 355

Keeping the explanatory variables constant at zero, the model showed that, compared to waiting 356

for the next LRT option, there was a decreased preference for changing to a car or a taxi or other transit 357

options for both trip purposes. For both types of trips, the familiarity of APIS, gender, weather 358

condition, LRT travel time, bus transfer time, availability of park and ride lots, and station were not 359

shown to be statistically significant and, thus, were dropped from the model. The impacts of other 360

explanatory variables are provided in Table 4. 361

362

Experience with LRT 363

The model concluded that, with reference to respondents whose primary mode of transport is an 364

automobile, transit users had a higher likelihood of selecting other transit options and a lower likelihood 365

of selecting the vehicle or taxi option for both examined trip purposes, which was expected. 366

367

Frequent riders (over 4 days/week) were less likely to select the vehicle/taxi or other transit options for 368

both commuter and non-commuter trips compared to infrequent LRT users (< 1 day/week). This higher 369

tolerance of frequent riders to wait supports the earlier finding by Gooze et al (2013). 370

371

Page 18 of 34



Draft

19

The model demonstrated that, relative to travellers who had an average familiarity with LRT service 372

frequency, travellers with no familiarity with LRT service frequency had a high probability of changing 373

to a vehicle or taxi during their commuter/non-commuter trips. 374

375

Experience with APIS 376

Travellers who viewed APIS as accurate had a low probability of choosing the vehicle or taxi options 377

during their commuter/non-commuter trips, although the results were statistically marginally significant. 378

Respondents who often looked at APIS were likely to choose the vehicle/taxi options during their 379

commute and non-commute trips, although the results were only marginally significant. 380

381

Demographic information 382

With reference to young transit riders (under 25 years), older riders (over 25 years) are more likely to 383

change to a vehicle or taxi during their commuter trips. This finding is mainly associated with the 384

occupation type as younger riders (under 25 years) are mainly students who might not own a car and/or 385

are less willing to pay for a more expensive transport alternative (e.g. taxi). 386

387

With reference to travellers who had two or more vehicles per household, other travellers were shown to 388

be more likely to change to the other transit options for both commuter and non-commuter trips; 389

however, this was only significant for respondents who did not own any vehicles per household. In 390

addition, the model further concluded that travellers with a driver’s license were more likely to change to 391

a vehicle or taxi compared to travellers without a driver’s license for both trip purposes. 392

393

Station characteristics 394

Relative to the stations without park and ride lots, respondent waiting at the other stations were more 395

likely to choose a vehicle or taxi option during their non-commuter trips. It is understandable that the 396

Page 19 of 34



Draft

20

access time to a vehicle for those respondents waiting at stations with park and ride lots was lower than 397

for those respondents waiting at stations without available parking. 398

399

Respondents waiting at end-of-line terminal stations were less likely to change to a vehicle or taxi option 400

compared to respondents waiting at other stations. In terms of the bus AI index, with an increasing AI 401

index, the probability of selecting other transit options also increased. This increase was expected as 402

more bus routes or high bus service frequency would attract travellers to choose the bus option when 403

LRT service was interrupted. 404

405

Relative to the respondents waiting at downtown LRT stations, respondents waiting in the LRT stations 406

in an established community were more likely to change to a vehicle or taxi and were less likely to wait 407

for LRT service recovery during their commuter trips. This finding is logical because established 408

communities have higher vehicle accessibility than do downtown communities. 409

410

Experience with conflicting information 411

The model revealed that travellers who had not experienced situations with conflicting information had a 412

high probability of choosing the vehicle or taxi option for both examined trip purposes. It is possible that 413

these respondents do not have a good knowledge of possible service recovery and thus choose to 414

abandon the transit option. This finding highlights the importance of resolving this problem of conflicting 415

information communicated from different sources. 416

417

Analysis of riders’ stated responses to an oncoming crowded LRT 418

This hypothetical scenario involved a regular weekday with an oncoming crowded LRT and APIS 419

displaying the next LRT arrival time as 5 minutes. The alternatives were squeezing in the first LRT, 420

waiting for next LRT, and changing path (i.e. walk or take the LRT to the other LRT station, go 421

somewhere else and return back when the LRT is likely to be less crowded). 422

Page 20 of 34



Draft

21

423

Squeezing in the first LRT was selected as the base category (i.e. Alternative 1) in this scenario. Keeping 424

the explanatory variables constant at zero, the model predicted that compared to squeezing in the first 425

LRT, there was a reduced preference for changing path options for both examined trip purposes. 426

427

For both trips, familiarity with the LRT service frequency and having a driver’s license were not 428

statistically significant and, thus, were dropped from the model. In the case of non-regular commutes, 429

vehicle ownership was also not statistically significant and was dropped from the model. 430

431

The impacts of other explanatory variables are provided in Table 4. They are discussed in the following 432

subsections. 433

434

Experience with the LRT 435

The multinomial logit model determined that travellers whose main travel mode was cycling or walking 436

were more likely to select waiting for the next LRT option and were less likely to select the changing 437

path option compared to car users. 438

439

Relative to non-frequent LRT riders (< 1 day/week), frequent LRT riders (1 or more days/week) 440

preferred to change paths during their commuter trips, although the finding was only significant for riders 441

who use the LRT 5 or more days per week. This finding might be explained by the fact that frequent 442

users are usually familiar with other transit routes. 443

444

Experience with APIS 445

Respondents who were familiar with APIS were more likely to change path during their commuter trips 446

when the oncoming LRT was crowded if the estimated LRT arrival time was 5 minutes. However, 447

respondents who were not familiar with APIS were less likely to change path. 448

Page 21 of 34



Draft

22

449

Relative to respondents who viewed APIS as accurate, travellers who viewed APIS as unreliable were 450

more likely to wait for the next LRT for both examined trip purposes. This finding seems 451

counterintuitive. Future research needs to examine more closely the interaction between train crowding 452

levels and APIS. 453

454

Demographic information 455

With reference to young transit riders (under 25 years), older riders (over 25 years) preferred to take the 456

next LRT during commuter trips. This result may be because younger riders place lower expectation on 457

comfort level. 458

In terms of gender, male respondents preferred to change their travel plans during their non-commuter 459

trips. 460

461

Weather conditions 462

Travellers were more likely to take the next LRT or change their plans on a typical summer day (15℃ -463

25℃ with no expected rain) than when the temperature was between -25℃ and -5℃ with no expected 464

snow for both examined trip purposes. 465

466

Trip characteristics 467

In terms of the LRT IVT, as IVT increased, the probability of waiting for the next LRT increased. 468

Therefore, travellers with a long travel time have high expectations regarding comfort level and are, thus, 469

less likely to accept crowded trains for long travel times. 470

471

Page 22 of 34



Draft

23

Station characteristics 472

The respondents waiting at stations with enclosed steel and glass structures were more likely to choose 473

the next LRT option during their non-commuter trips, compared with the respondents waiting at an in-474

community platform. 475

476

Respondents waiting at end-of-line terminals were less likely to change their path than those waiting at 477

other stations, although the finding was only significant for commute trips. This finding was expected 478

because the LRT to arrive at an end-of-line terminal was expected to be empty. 479

480

Relative to the respondents waiting at the downtown LRT stations, respondents waiting at other LRT 481

stations were less likely to change path, although the finding was only significant for inner city stations; 482

walking distance to a public place, such as a coffee shop, shopping mall and restaurants in the downtown 483

is shorter than that in other areas. 484

Discussion: 485

In this paper, four multinomial logit models were developed and calibrated to explore the factors 486

affecting trip decision-making for commuter and non-commuter trips. 487

In the scenario when inconsistent information was presented: 488

• The following factors were found to have a significant impact on riders’ behavioural choices for 489

changing to a vehicle or taxi for both commuter and non-commuter trips were identified as: 490

vehicle being the primary transportation mode, infrequent LRT riders (< 1 day/week), frequency 491

of looking at the APIS display, no previous experience in encountering conflicting information, 492

having a driver’s license, and waiting at end-of-line terminal stations. 493

• The following factors had positive influences for inducing riders to choose other transit 494

alternatives for both examined trip purposes: the primary mode of transportation being transit, 495

having a household vehicle ownership of 0 and waiting at city centre stations. 496

Page 23 of 34



Draft

24

• The following factors were found to be contributing factors in a rider’s choice for waiting for the 497

LRT for both trip purposes: frequent LRT rider (5 or more days/week), not having a driver’s 498

license and waiting at terminal station. 499

For the second scenario, which involved an oncoming crowded LRT, while APIS displayed the next LRT 500

arrival time as 5 minutes: 501

• The following factors had a significant contribution to positively influence riders’ choice for 502

squeezing in the first LRT for both commuter and non-commuter trips: the primary 503

transportation mode being a vehicle, young riders (under 25 years), cold weather, and not waiting 504

at end-of-line terminal stations. 505

• The factors that had positive influences for waiting for the next LRT for both examined trip 506

purposes were: the primary mode of transportation being other, a low perceived APIS accuracy, 507

older adults (over 25 years), warm weather, waiting at end-of-line terminal stations, and a long 508

LRT IVT. 509

• The factors that positively influenced riders’ choices for changing paths for both trip purposes 510

were: having a primary ‘other’ mode of transportation (like a bicycle), warm weather conditions, 511

and station at close proximity to the city centre. 512

Conclusion: 513

Several ATIS studies investigated the underlying factors affecting riders’ trip decision-making. However, 514

the literature has not extensively examined the contributing factors affecting the travellers’ trip-specific 515

behavioural responses to APIS in the case of overcrowded trains or conflicting information with a 516

significant gap in real-time transit arrival. This study focused on examining and understanding the factors 517

that affect transit riders’ behavioural responses to APIS in these specific situations. 518

The outcomes of this study provide useful insights to transit agencies for further development of 519

APIS. In the case of the LRT service interruption with inconsistent real-time information, vehicle 520

availability had a significant impact on travellers’ mode-switch behaviour. The provision of designated 521

Page 24 of 34



Draft

25

parking locations around LRT stations for car-sharing programs such as car2go service may further 522

increase the likelihood of using such alternative when LRT service is running late or is disrupted. In 523

addition, in the case of the presence of conflicting information, when audio announcements broadcasted 524

the expected time when LRT service was restored, more riders would choose to wait for the LRT. Thus, 525

the provision of audio information is crucial in reducing the uncertainty of waiting time and convincing 526

riders to wait for LRT. As a significant percentage of respondents (40.82 to 44.84 % for non-commute 527

and commute trips respectively) state that they would change to car or taxi when conflicting information 528

is presence, Calgary Transit should work on treating this situation and avoid that it occurs in the first 529

place. Transit staff in transit control center should find a technical solution to override the APIS software 530

and dissipate service recovery time that is consistent with the information disseminated by audio 531

announcement. 532

533

Waiting for the next LRT to arrive when the train was crowded was less likely to be chosen by 534

respondents in cold weather (-25°C to -5°C, no snow expected) than in warm weather (15°C to 25°C, no 535

rain expected). Calgary’s LRT stations range from simple platforms to large enclosed steel and glass 536

structures. Some of the platforms do not provide a heated waiting area. Calgary Transit might consider 537

having heated waiting areas on LRT platforms to make the waiting time easier for riders who prefer to 538

wait for the next train rather than squeeze into a crowded train. Making waiting time more convenient is 539

especially important for riders who are travelling for relatively long distances on the train and place 540

importance on trip comfort level. In addition, the provision of next train anticipated crowding conditions 541

with train arrival information is vital to help passengers better plan their trip and productively manage 542

their waiting time (e.g. run for some errands, make phone calls, etc.) 543

544

In addition, many respondents showed an interest in other transit alternatives. However, a lack of 545

familiarity with the transit service may deter some riders from taking this option. Thus, providing 546

information on alternative transit routes at LRT stations may help transit riders to access other transit 547

Page 25 of 34



Draft

26

choices during their trip. Information can be made more visible by including transit maps in addition to 548

the existing LRT maps at LRT stations. These references would help travellers feel more comfortable in 549

choosing alternatives transit modes. 550

551

This study can be expanded in several ways to better understand transit riders’ responses to APIS. A 552

Stated preference (SP) approach was applied to model discrete behaviour choices. One way to validate 553

the SP survey results is through a revealed preference survey. Furthermore, the travellers’ most frequent 554

origin stations were used in the model. However, the return stations were excluded from this research; 555

including that information in further study would provide more insights. In addition, future studies 556

should also examine passengers’ responses as affected by other factors, such as waiting time uncertainty, 557

acceptable prediction error, train crowding levels and the interaction effect between information 558

inconsistency and the crowding level of trains. 559

560

561

Acknowledgment: 562

This paper was supported by the Urban Alliance professorship funds from the City of Calgary. The 563

authors would like to thank Mr. Scott Hale from Calgary Transit for his help in this research. 564

565

566

567

568

569

570

571

572

573

Page 26 of 34



Draft

27

574

575

Page 27 of 34



Draft

28

Table 4. Multinomial logit model estimation results

Scenario Conflicting information Crowding train

Trip purpose Commuter trips (i.e.

Work/School)

Non-commuter trips Commuter trips (i.e.

Work/School)

Non-commuter trips

Action taken Change to

vehicle

Other transit Change to

vehicle

Other transit Waiting for

next LRT

Changing path Waiting for

next LRT

Changing path

Main mode (Reference: vehicle)

Transit -0.56** 0.72** -0.59** 0.68** 0.32 -0.2 0.12 0.25

Other -0.49 -1.62* -0.51 -1.61* 1.01* -18.86** 1.47* -18.64**

Frequency of using LRT (Reference: <1 Day/week)

1-4 Days/week -0.44 -0.51 -0.53 -0.55 -0.01 1.04

> 5 Days/week -0.62* -0.37 -0.72** -0.4 -0.35 2.22**

Familiarity of APIS (Reference: Familiar)

Very familiar -0.21 -0.22 -0.24 -0.21 0.2 1.03* 0.09 0.28

Not familiar 0.95* -0.52 1** -0.47 0.92 -18.55** -0.01 -19.65**

Perceived accuracy (Reference: Accurate)

Earlier -0.3 -0.13 -0.3 -0.11 0.49* 1.33** 0.05 0.88*

Later -0.13 0.81** -0.18 0.8** 0.47** -0.43 0.52** 0.35

Do not know -0.94** -0.14 -0.98** -0.22 -0.95* -18.98** -0.77 -0.66

Frequency of looking APIS (Reference: Sometimes (i.e. 1 or 2 times)

Often 0.8** 0 0.73** 0 -0.43 -0.41 -0.27 -0.88*

Never 1.03 -17.8** 0.86 -17.2** -0.17 -14.69** -0.72 -18.32**

Age group (Reference: Under 25)

25-34 0.44* 0.07 0.21 -0.53 -0.06 -0.63

35-44 0.42 -0.11 0.8** -0.93 0.2 -0.89

Over 45 0.4 0.23 1.01** 0.39 0.35 -1.29*

Vehicle ownership (reference: no)

1 car -0.32 0.73** -0.25 0.72** 0.4 -19.88**

over 1 car -0.1 0.19 0.03 0.19 0.36 1.23**

Driver’s license (Reference: No)

Yes 0.74* -0.04 0.84** -0.03

Gender (Reference: Female)

Male 1.07** 0.12 -0.14 0.84**

Weather condition (Reference: -25 to -5, no snow expected)

15 to 25, no rain

expected 1.3** 0.31 1.09** 0.66*

LRT IVT (in

minute)

0.03** -0.06* 0 -0.03

bus transfer -0.68** 0.92** -0.08 0.51*

Page 28 of 34



Draft

29

number

Park’n’Ride

Yes 1.07** 0.12 0.98** 1.47* 1.46** 0.03

Coffee shop (Reference: No)

0.55* 0.39

Station (Reference: In-community platform)

Station 0.08 0.8*

AI index 0 0.02* -0.01 0.02* 0 0.04 0 0.03*

End-of-line

Yes -0.5* -0.47 -0.47* -0.44 0.24 -1.89** 0.43 -0.4

Category( Reference: city centre)

Major activity

centre

0.36 0.01 -0.74 -0.11 -1.01 -0.95 -1.14 -1.47

Community activity

centre

0.83 0.34 -0.1 0.27 -0.19 -1.02 -0.58 -0.11

inner city -18.28** 1.82 -18.59** 1.68 1.13 -12.98** 0.06 -20.81**

Established 1* 0.85 -0.52 0.51 -0.38 -0.91 -0.97 -0.91

Road and street network (Reference: Skeletal road)

Arterial -0.68** 1.21* -0.82** 0.65

City centre

Frequency of receiving conflicting information (Reference: Sometimes)

Often -0.14 -0.51* -0.07 -0.48

Never 0.64** -0.17 0.54* -0.19

Constant -1.82** -2.25** -0.52 -1.82** -1.29 -6.01** 0.36 -1.87

Model Summary Population size= 155615.59

Design df= 823

F(52,772)=124.66

Prob > F =0.0000

Population size= 155615.59

Design df= 823

F(46,778)=128.81

Prob > F =0.0000


Design df= 662

F(54,609)= 76.97

Prob > F =0.0000


Design df= 664

F(50,615)=136.56

Prob > F =0.0000

Note: *, and ** denote statistically significant at α = 0.10 and 0.05.

Page 29 of 34



Draft

30

Reference:

- American Public Transportation Association, 2013. Ridership data 2012.

http://www.apta.com/resources/statistics/Pages/RidershipArchives.aspx. Accessed November 2015.

- Bai Y. and Kattan L. (2014). Modelling Riders’ Behavioral Response to Real-time information on

LRT Stations. Transportation Research Record: Journal of the Transportation Research Board.

Volume 2412 / Travel Behavior 2014, Vol. 1. pp. 82-92.

- Bai Y., 2014. Modelling Riders' Behavioural Responses to Real-Time Information at LRT Stations in

Calgary, Alberta, Master thesis, University of Calgary.

- Body, A. 2007. Regional update: signal pre-emption and improvements to real-time passenger

information are playing a key role in improving passenger transport patronage and reducing road

congestion in the Auckland region. ITS International 13 (1), 33–34.

- Caulfield, B. and M. Mahony, 2009. A Stated Preference Analysis of Real-Time Public Transit Stop

Information. Journal of Public Transportation, Volume 12, Issue Number 3.

- Caulfield, J. and O'Mahony, M. M., 2004. The provision of on street passenger information via real

time passenger. Proceedings of the IEEE Conference on Road Transport Information and Control.

London, UK.

- Cats, O. 2011. Dynamic Modelling of Transit Operations and Passenger decisions. PhD Dissertation.

Stockholm: KTH – Royal Institute of Technology.

- Calgary Transit: www.calgarytransit.com. Accessed December 2017

- Cats, O. and Gkioulou Z., 2017. European Journal of Transportation Logistics. 6: 247.

o https://doi.org/10.1007/s13676-014-0070-4.

- City of Calgary: City of Calgary Calgary and Region: Travel and Activity Survey Final Report,

November 2012.

http://www.calgary.ca/Transportation/TP/Documents/forecasting/CARTAS_FinalReport_112012.pdf

. Accessed December 2017.

Page 30 of 34



Draft

31

- Davidson, B., Vovsha, P., Abedini, M., Chu, C., and Garlan, R., 2011. Impact of capacity, crowding,

and vehicle adherence on public transport ridership: Los Angeles and Sydney experience and

forecasting approach. Proceedings of the Australasian Transport Research Forum, Adelaide,

Australia.

- Drabicki A., Kucharski R., Cats O. and Fonzone A., 2017. Proceedings of the 5th IEEE International

Conference on Models and Technologies for Intelligent Transportation Systems (MT-ITS). June

2017. Naples, Italy

- Dziekan, K., Kottenhoff, K., 2007. Dynamic at-stop real-time information displays for public

transport: effects on customers. Transportation Research Part A 41, 489–501.

- Eboli L. and Mazzulla G., 2015. Relationships between rail passengers’ satisfaction and service

quality: a framework for identifying key service factors. Journal of Public Transport, Vol.

7, Issue 2, pp 185–201

- Ferris, B., Watkins, K., and Borning, A., 2010. One bus away: results from providing real-time

arrival information for public transit. Proceedings of CHI 2010. Atlanta, G.A. USA.

- Gooze A., Watkins A. E. and Borning A., 2013. Benefits of Real-Time Transit Information and

Impacts of Data Accuracy on Rider Experience. Transportation Research Record. Vol. 2351, pp. 95–

103.

- Hensher DA, Rose JM, Collins AT, 2011. Identifying commuter preferences for existing modes and

a proposed Metro in Sydney, Australia with special reference to crowding. Public Transport. Vol.3,

pp.109–147

- Kattan L., de Barros, A. G. and Saleemi, H., 2013. Travel behavior changes and responses to

advanced traveler information in prolonged and large-scale network disruptions: a case study of west

LRT line construction in the city of Calgary. Transportation Research, Part F - Traffic Psychology

and Behaviour, Volume 21F, pp. 90 – 102.

- Kattan L., Hurul Habib K. and Islam, T., 2012. Information Provision and Driver Compliance to

ATIS Application: Case Study on the Interaction between Variable Message Sign and Other Sources

Page 31 of 34



Draft

32

of Traffic Updates in Calgary, Canada, Canadian Journal of Civil Engineering, 2011, 38(12), pp.

1335-1346.

- Katz, D. and Garrow, L. A., 2012. The impact of bus door crowding on operations and safety.

Journal of Public Transportation, 15(2), 71-93.

- Kockelman K and Kweon Y. J., 2002. Driver Injury Severity: An Application of Ordered Probit

Models. Accident Analysis and Prevention 34 (4) pp: 313-321.

- Kim J-K. Lee B. and Oh S., 2009. Passenger choice models for analysis of impacts of real-

time bus information on crowdedness. Transportation Research Record, 2112, 119-126.

- Kim KM, Hong S-P, Ko S-J, Kim D., 2015. Does crowding affect the path choice of metro

passengers? Transportation Research Part A. Vol 77, pp. 292–304

- Maunsell, F. 2007. Demand management techniques - peak spreading. London, UK: Department for

Transport, Transport for London and Network Rail.

- MVA Consultancy, 2007. Understanding the Passenger, Valuation of Overcrowding on Rail

Services, London, UK: Department of Transport.

- Mulley M., Clifton GT, Balbontin C. and Ma L., 2017. Information for travelling: Awareness and

usage of the various sources of information available to public transport users in NSW.

Transportation Research Part A: Policy and Practice. Volume 101, pp. 111-132.

- Nijkamp, P., Pepping, G., and Banister, D., 1996. Telematics and transport behaviour. Springer-

Verlag, Berlin.

- Pel, A, Bel N.H. and Pieters M (2014). Including passengers’ response to crowding in the Dutch

national train passenger assignment model. Transportation Research Part A: Policy and Practice.,

Vol. 66, pp. 111–126.

- Preston J., Pritchard J., and WatersonB., 2016. Train overcrowding: investigating the use of better

information provision to mitigate the issues Transportation Research Record, 2649, pp. 1-19.

Page 32 of 34



Draft

33

- Pritchard J., 2017. Providing Improved Crowding Information to Provide Benefits for Rail

Passengers and Operators. In: Stanton N. (eds) Advances in Human Aspects of Transportation.

AHFE 2017. Advances in Intelligent Systems and Computing, vol 597. Springer, Cham.

- Rahman, MM, Wirasinghe, SC and Kattan, L., 2013. Users' views on current and future real time

bus information systems. Journal of Advanced Transportation Apr/2013, Volume 47, Issue 3, pp. 336

- 54

- Raman, M., Schweiger, C., Shammout, K., and Williams, D., 2003. Guidance for developing and

deploying real-time traveller information systems for transit. Federal Transit Administration, U.S.

Department of Transportation, National Technical Information Service/NTIS, Springfield, Virginia,

USA.

- Russell, M. L., 2012. Travel time use on public transport: what passengers do and how it affects.

Ph.D. Dissertation, Philosophy, University of Otago, Dunedin, New Zealand.

- Rutherford, G., Wang, Y., Watkins, K., and Malinovskiy, Y., 2011. Perceived and actual wait time

measurement at transit stops using Bluetooth (No. TNW2012-09). USA: Transportation Northwest,

Washington State Department of Transportation.

- SAIC: Science Applications International Corporation, 2003. Oregon regional Intelligent

Transportation Systems (ITS) Integration program draft phase III report: transit tracker information

displays (Contract No DTFH61-96-C-00098). USA: Department of Transportation.

- Tang, L., and Thakuriah P., 2012. Ridership effects of real-time bus information system: a case study

in the City of Chicago. Transportation Research Part C, 22, 146-161.

- Wardman, M., and Whelan, G., 2011. Twenty years of rail crowding valuation studies: evidence and

lessons from British experience. Transport Reviews, 31, (3), 379-398.

- Warman, P., 2003. Measured impacts of real-time control and information systems for bus services.

London, UK: Department for Transport.

- Zhang, F., 2010. Traveller responses to real-time transit passenger information systems, Ph.D.

Dissertation, School of Architecture, University of Maryland, M.D. USA.

Page 33 of 34



Draft

34

- Zhang Y., Jenelius E. and Kottenhof K., 2017. Impact of real-time crowding information: a

Stockholm metro pilot study. Journal of Public Transport DOI 10.1007/s12469-016-0150.

Page 34 of 34

