Improvig Bus Service Reliability: The Sigapore Experiece Waiya Leog a*,kare Goh b, Stephae Hess c, Paul Murphy d a Lad Trasport Authority, No. 1 Hampshire Road Sigapore 219428. b Miistry of Commuicatios ad Iformatio, 140 Hill Street, Sigapore 179369. c Istitute of Trasport Studies & Choice Modellig Cetre, Uiversity of Leeds, 34-40 Uiversity Road, Leeds LS2 9JT, UK. d AECOM, AECOM House, 179 Moss Lae, Altricham, WA15 8FH, UK. * Correspodig author: Lad Trasport Authority, No. 1 Hampshire Road Sigapore 219428 Phoe: +65 6295 7726. Email: leog_wai_ya@lta.gov.sg Keywords: Bus Service Reliability, Sigapore, Excess Wait Time, Lad Trasport Authority, Bus Cotractig Model. Classificatio codes: R480, L920 ABSTRACT I February 2014, Sigapore embarked o a 2-year trial of a Bus Service Reliability Framework (BSRF) to improve e-route bus regularity ad reduce istaces of bus buchig ad prologed waitig times. Based o Lodo s Quality Icetive Cotract, the Sigapore model also imposes pealties or provides icetives to operators for icreases/reductios of Excess Wait Time (EWT) beyod a certai route-specific baselie. Drawig o isights derived from research o performace-based cotracts, this paper describes some key cosideratios surroudig this particular iovatio i Sigapore s overall bus regulatory framework. We also discuss a importat advacemet i our uderstadig of how bus users value reliability improvemets through estimates obtaied from stated preferece data. At the same time, early idicatios from the trial have bee ecouragig. 1. Itroductio For may public trasport users, service reliability is a key attribute of the travel experiece. The importace of reliability is amply demostrated by the multitude of papers cocludig that variability i travel time impacts well-beig more egatively tha the actual jourey time itself (see review by Carrio & Leviso, 2012). With regards to the bus i particular, a ureliable bus service, characterised by uequal headways or bus buchig for high frequecy services, ca lead to loger waitig time ad travel time for bus passegers. Moreover, i cases where passeger loadig o a particular bus route is already high, uequal arrival times ca mea severe crowdig o the first bus that arrives after a log headway. The geerally upleasat i-vehicle experiece adds aother layer of frustratio to passegers who have already edured a loger tha expected wait, if they are ot deied boardig i the first place. Ureliability begets further ureliability as dwell times icrease at bus stops to cater for the higher passeger movemets o ad off the bus. This paper describes Sigapore s experiece with improvig bus service reliability. Sectio 2 provides a review of reliability measures ad how reliability is achieved i various jurisdictios worldwide. Sectio 3 briefly describes Sigapore s bus idustry before discussig Sigapore s trial of its Bus Service Reliability Framework (BSRF). Sectio 4 is largely a empirical sectio which discusses commuter awareess of the 1
BSRF, stated preferece strategies to measure improvemet i reliability ad crowdig, ad outcomes to date. Sectio 5 cocludes. 2. Literature Review 2.1 Overview of Reliability Measures Used Worldwide Recogisig the cetrality of this aspect of service quality i passeger experiece, idustry regulators aroud the world have itroduced various service reliability frameworks i their performace moitorig regimes. TriMet i Portlad, Orego, uses the Bus Dispatchig System (BDS) to moitor public trasport reliability (Feg & Figliozzi, 2012). The BDS combies Automatic Vehicle Locatio ad Automatic Passeger Couters data to provide detailed iformatio o bus service performace. The two performace measures are headway deviatio ad actual headway spatial distributio. Headway deviatio looks at the differece betwee actual headway ad scheduled headway. Actual headway spatial distributio depicts the proportio of actual headways deviatig from scheduled headway agaist differet stops alog the route. Spikes ad dips o the distributio would suggest cogestio or chokepoits that require improvemets. Strathma et al. (2000) report that the BDS has decreased bus service ruig time by 3 percet after implemetatio. I Shaghai ad Jiagyi City, a ormalised average headway idex is used to determie the actual headway deviatio from the scheduled headway (Guo, Luo, Li & Feg, 2011). A idex below 100% idicates that the bus is earlier tha scheduled, while a idex above 100% idicates that the bus is behid schedule (Eq. (1)). H δi Hδ 2 H i = 100% 1 (1) where: H i H δ H δi = average headway of ormalised i-th bus stop; = the departure iterval at departure statio of two specific adjacet bus deploymets; = headway of the two specific adjacet bus deploymets at i-th bus stop; = umber of bus deploymets. I Chagzhou s Bus Rapid Trasit, four measures are used as idicators of reliability (Huo, Zhao, Li, & Hu, 2014). Oe statistic used is the coefficiet of variatio of headway, which is the stadard deviatio of headway divided by its mea. It idicates service reliability from the operator s perspective. O the other had, potetial waitig time, equivalet waitig time ad reliability buffer time idicate service reliability from the users perspective. Potetial waitig time refers to the differece betwee 95 th percetile waitig time ad mea waitig time. Equivalet waitig time is the weighted sum of mea ad potetial waitig time. Reliability buffer time is the extra time that commuters eed to provisio beyod typical jourey time to esure o time arrival at destiatio with 95% probability. 2
I Chicago, the Automatic Vehicle Locatio data is used to determie ruig time adherece ad headway regularity (Li, Wag & Barum, 2008). Ruig time adherece measures the average differece betwee actual ad scheduled ru times, while headway regularity measures the average differece betwee actual ad scheduled headways. A high metric value for these two idicators will idicate irregular bus services ad poor reliability (Eq. (2)). Ruig Time Adherece = m Actual Ru Time Scheduled Ru Time Scheduled Ru Time 100% m Actual Headway Scheduled Headway Scheduled Headway Headway Regularity = 100% (2) I Sydey, Trasport for New South Wales uses Key Performace Idicators to moitor bus services performace. It measures puctuality of buses at the commecemet of trip, mid-poit of trip ad at the last trasit stop, requirig at least 95% of the trips to be betwee 2 mi early ad 6 mi late. Trasport for Lodo (TfL) characterises Lodo bus services depedig o whether they are high or low frequecy. The reliability of high frequecy services, defied as those with headways of less tha 15 miutes, is assessed based o average excess wait time (EWT) experieced by commuters. Ureliable bus services, as evideced by irregular spacig of buses, will result i high EWT. O the other had, low frequecy services are assessed o percetage of buses departig o time accordig to bus schedules. 2.2 Achievig Reliability through Performace Based Cotracts How to meaigfully measure bus service reliability is oe, but certaily ot the oly cosideratio that regulators eed to address. Aother importat questio ivolves mechaism desig how oe might achieve eve better bus service reliability performace. Fortuately, o this latter issue, the existig literature offers substatial guidace, particularly through the use of performace-based cotracts (PBCs). Based o the extesive research, PBCs are ow used across maufacturig ad service idustries, i public ad private domais (Selviaridis ad Wystra, 2014). I the realm of bus service provisio i particular, Hesher ad Houghto (2002) proposed a system that takes ito accout various exteral costs such as costs of cogestio, amog may others, with social surplus maximisatio as the uderlyig motivatio i order to esure that bus operators deliver the optimal service level that is cosistet with the eeds of stakeholders, especially the govermet. Workig alog similar lies, Hesher ad Staley (2003) highlighted the importace of PBCs as a crucial factor that aligs commercial objectives with social objectives by rewardig operators for achievig a miimum level of service (MSL) ad for a icrease i ridership. Selviaridis ad Wystra (2014) also highlighted a form of PBCs where egative or positive icetives are give although there ca also be dead zoes for acceptable performace levels for which there is either pealty or extra rewards. A otable example of a implemetatio of PBCs i the provisio of public bus service is the Hordalad framework (Larse, 2001). I the Hordalad framework, 3
social beefits such as reduced waitig time, reduced umber of trasfers ad trasfers of riders from car to public trasport is iteralised ito the operators remueratio cotracts based o reveue kilometres ad passegers The framework attempts to iduce operators to deliver the socially optimal level of services through performace-based subsidies as part of the total paymet per passeger received by operators (Hesher ad Staley, 2003). Numerous studies have also show that reliability remais a crucial compoet of bus service quality. dell Olio, Ibeas ad Cecıń (2010) foud that bus reliability is oe of the most importat attribute alogside waitig time. They highlighted the eed for trasport compaies to place greater emphasis o puctuality ad headways betwee buses. Not oly is bus reliability a importat factor i determiig people s behaviour choices (Disey, 1999; Köig & Axhause, 2002), it is also foud to be a mai reaso behid users dissatisfactio towards public bus services (Edvardsso, 1998). I additio, bus reliability may be of a decisive factor i determiig people s tolerace level i other areas such as comfort ad vetilatio (Disey, 1999). Similarly, a key fidig ucovered by Bates, Polak, Joes, ad Cook (2001) shows that as travellers, users highly value puctuality. I order to improve service regularity, Cats (2014) highlighted the eed to icorporate reliability measures such as regularity idicators ito a icetive scheme. The icorporatio of bus service reliability ito a PBC regime for operators has bee widely implemeted aroud the world, such as i New Zealad ad i Lodo (Trasport for Lodo, 2015; Ia Wallis Associates Ltd & The TAS Partership, 2013; Vicet, 2008). I Lodo s case, a icetive mechaism embedded i its Quality Icetive Cotract is used to ecourage bus operators to provide reliable service, with a bous of up to 15 percet or deductio of up to 10 percet of the cotract price relative to the required stadards. These stadards are kow as the Miimum Performace Stadards (MPS) ad are a crucial part of Lodo s scheme as they act as a reliability bechmark for operators. Operators are ot paid for ay mileage ot operated for reasos withi the operator s cotrol, such as staff shortages or mechaical issues. Settig MPS properly is a very importat part of the cotractig process, as failure to could so could mea that bus operators would pass the risk of poor performace to the public authority by puttig the aticipated pealty i the cotract bid. Sice the implemetatio of the Quality Icetive Cotracts i 2001, EWT of high frequecy services has falle from 2.2 miutes to 1.0 miutes ad the percetage of o-time low frequecy services has rise from 68 percet to 83 percet i 2014 (Trasport for Lodo, 2014). With greater service reliability, passegers are also able to time their arrival time to coicide with bus arrival time to reduce waitig time. TfL reports that service improvemets are estimated to have accouted for some 30 percet of the growth i demad for bus services from 1997 to 2012 (Trasport for Lodo, 2014). 3. The Bus Service Reliability Framework (BSRF) i Sigapore 3.1 Overview of Sigapore s Bus Idustry Bus services i Sigapore are primarily provided by two major bus operators, SBS Trasit (SBST) ad SMRT Corporatio (SMRT). As of Jue 2015, both compaies operated over 350 bus services, with a average total daily vehicle-kilometres travelled of 856,600 bus-km. Each bus operator has bee assiged separate areas 4
of resposibility that correspod to satellite residetial tows developed by govermet plaig agecies. They operate a mixture of short itra-tow routes (called feeder services ) ad loger routes (called truk routes ). Havig bee awarded the first bus package of public bus services uder the ew Bus Cotractig Model (see Goh, Swee, & Low, 2015), Tower Trasit became Sigapore s third major bus operator as of May 2016. SBS Trasit operates 247 bus routes, with a total route etwork of 5054 km, while SMRT Corporatio operates 111 bus routes with a total route etwork of 2326 km. Before it was required to udertake a ew advisory role o public trasport matters, the Public Trasport Coucil (PTC) oversaw regulatio o, amog other thigs, bus services ad bus service operators. As part of its regulatory framework, the PTC established Quality of Service (QoS) stadards, i particular Operatig Performace Stadards which measure miimum daily or mothly operatioal deliverables, coverig aspects of bus reliability, loadig ad safety. O bus reliability stadards, the PTC required operators to esure that, o a daily basis, each service had at least 85% of its trips depart the bus iterchages ad termials ot more tha 5 mi from its scheduled headway. 3.2 Overview of the Bus Service Reliability Framework (BSRF) It has ofte bee poited out that the PTC s regulatory stace failed to sufficietly address commuters cocers about reliability as the oly key performace idicator measures reliability at origi ad ot e-route regularity. I 2014, the Lad Trasport Authority (LTA) decided to make e-route reliability a explicit service requiremet by itroducig the Bus Service Reliability Framework (BSRF) iitially as a two-year trial for the two major bus operators (Lad Trasport Authority, 2014b, 2014a). I the first phase of the trial described i this paper, 22 bus services are covered, comprisig a mixture of truk ad feeder services. The itetio of the trial is to improve e-route bus regularity, reduce istaces of bus buchig ad reduce prologed waitig times for bus users. Although the BSRF trial oly covers a modest umber of bus services, evertheless, the LTA hopes that the trial will provide a better uderstadig of bus reliability improvemets that could be made o differet types of routes, as well as the BSRF s effectiveess to get bus operators to improve service reliability. A successful roll-out of the BSRF would complemet other regulatory measures such as more strictly eforcig bus laes ad ehacig bus priority schemes ad measures. With the Bus Cotractig Model that is beig itroduced, the lessos leart from the trial will also be valuable to LTA as it is ow resposible for settig regulatory stadards for the quality of bus services i the bus packages that are tedered out. Like Lodo, Sigapore also categorises bus services by high ad low frequecy. Low frequecy services are those with the majority of headways beig more tha 15 miutes where puctuality of bus arrivals is more importat. High frequecy services are buses that arrive at frequecies of 15 miutes or less where commuters ca tur up ad go ad geerally do ot refer to the timetable. The vast majority of basic bus services i Sigapore are high frequecy services. Followig the TfL model, the BSRF assesses the regularity of a bus service usig the cocept of Excess Wait Time (EWT). EWT is the average additioal wait time actually experieced by commuters at bus stops, compared to the expected wait time if the buses arrived at regular itervals. It is defied as the differece betwee the Actual Wait Time (AWT) ad the Scheduled Wait Time (SWT), that is, EWT = AWT SWT, with AWT ad SWT defied i Eq. (3). 5
AWT 2 actual actual headway 2 headway ; SWT 2 scheduled scheduled headway 2 headway (3) The EWT methodology assumes uiform arrival rate of passegers ad icreases if there is bus buchig which results i prologed waits for the subsequet bus. Coversely, it goes dow if bus arrivals at each bus stop become more regular. As a result, commuters would experiece greater ease i boardig as the passeger load is spread more evely across the various bus trips. If a bus service arrives perfectly regularly, the EWT will be 0 mi. The EWT explicitly quatifies the commutig experiece usig the excess time experieced by commuters due to variability of the bus services. This allows useful ad direct comparisos across differet bus routes ad bus stops alog the lie. The EWT is a idicator of reliability from the commuter s perspective, rather tha the supply side perspective such as the operator s ability to dispatch buses o time (Oort, 2014). Moreover, the EWT ca be used a cosistet idicator across differet coutries eve if there are differet defiitios ad rage of stadards for puctuality. I Sigapore, for the trial, the EWT is measured durig peak ad off-peak hours from Modays to Fridays, excludig Public Holidays. The AM peak is defied as actual arrivals betwee 6.30am ad 8.29am; AM off-peak as betwee 8.30am ad 4.59pm; PM peak as betwee 5.00pm ad 6.59pm ad PM off peak as betwee 7.00pm ad 10.59pm. EWT is also measured across all the trips for a sigle bus service, ad at several critical bus stops or itermediate timig poits (ITPs). The umber of ITPs varies by route legth, with more ITPs for loger routes. EWT are also weighted heavier for the peak periods compared to the off-peak. Fially, EWT are averaged across the caledar moth by directio before summatio across both directios to obtai the overall EWT for the service for the moth. 22 bus services were chose for the BSRF trial because of public feedback of poor reliability. These iclude a mix of log ad short truk services as well as feeder services 1. Each bus service will have its ow existig baselie EWT depedig o the curret performace ad the characteristics of the route. Typically, a log truk route will have a higher existig baselie EWT. The operators performace is bechmarked to historical performace i order to esure the reasoableess of the stadards that operators eed to comply with. A holistic measuremet that assesses idividual bus services as a whole over the etire moth, ad ot idividual bus trips at specific times of the day, is used i order to esure that improvemets i reliability are i fact sustaied over the loger term. Table 1 lists the baselie EWT scores for the bus services covered i the iitial rollout of the BSRF. Although the baselie EWT bechmarks are coceptually similar to the Lodo MPS, ulike Lodo, operators i Sigapore are ot allowed to curtail service, such as 1 Seve SMRT bus services - 176, 184, 188, 302, 858, 901 ad 911 - were placed o the BSRF from February 2014. SBS Trasit Services 17, 52, 228 ad 242 were placed o the BSRF from 28 February 2014 ad Services 3, 39, 241 ad 325 from 24 March 2014. From 23 Jue 2014, aother seve bus services - SBS Trasit Services 51, 154, 292, ad 354, as well as SMRT Services 189, 853 ad 962 - were implemeted uder BSRF. 6
short turig without pealty whe there is traffic. As the operatig eviromet is differet i Sigapore, such curtailmets are ot expected to be ecessary ad operators are aware of this requiremet. Uder the Bus Cotractig Model, the govermet specifies frequecy requiremets for the route whe callig for a teder. Bidders are to submit their optimised required umber of vehicles for operatio as part of their teder proposal. Table 1: Baselie EWT scores Fleet size category Category 3 ( 20 buses) Category 2 (10 to <20 buses) Category 1 (< 10 buses) Operator SBST SMRT SBST SMRT SBST Bus Service Truk/ Feeder Legth of Directio 1 (km) Legth of Directio 2 (km) EWT Baselie (mi) 3 Truk 19.5 20.3 1.5 39 Truk 26.2 26.4 1.2 51 Truk 37.7 36.3 2.2 154 Truk 32.5 34.4 1.9 176 Truk 23.5 24.4 1.6 188 Truk 21.6 22.1 1.4 858 a Truk 73.4 0 2.1 17 a Truk 25 0 1.8 52 Truk 25.9 25.2 2.0 184 a Truk 22.4 0 1.3 189 a Truk 19.8 0 1.4 302 Feeder 7.5 0 1.0 853 Truk 31.1 32.3 1.5 911 Feeder 11.8 0 1.3 962 a Truk 16.9 0 1.4 228 Feeder 10.7 0 1.1 241 Feeder 6.2 0 1.2 242 Feeder 5.7 0 0.8 292 Feeder 5.9 0 0.9 325 Feeder 12.3 0 0.8 354 Feeder 4.5 0 1.1 SMRT 901 Feeder 11.8 0 0.9 a These truk services are loop services ad therefore have o Directio 2. 3.3 Trial of BSRF Icetive-Pealty Framework It is widely ackowledged that improvig reliability or reducig EWT is operatioally challegig. Bus operators will have to put i additioal resources, such as hirig more service cotrollers to maage bus services ad havig stadby buses to iject mid-route if there are delays to buses which are already e-route. The operators may also choose to deploy aother bus if the desigated bus that is scheduled to turaroud has bee delayed o the precedig trip due to cogestio. By ruig 7
such outside-schedule trips to maitai headway, the iitiative take by the operators is rewarded with a better EWT score. Hece, should sigificat improvemets i EWT be made, icetives are provided to allow operators to recoup costs. These icetives are calibrated i accordace with the efforts ad operatioal costs ivolved i improvig the reliability of the services. The icetive-pealty framework of the BSRF is modelled after Lodo s experiece. Operators are rewarded oly whe they achieve improvemets i bus service regularity, ad are pealised if the service is ot so. Icetives ad pealties are determied based o 6-moth average performace of each service. No icetives or pealties apply i a eutral zoe of ± 0.1 miutes o either side of the route-specific baselie EWT. Outside this eutral zoe, the mothly icetive ad pealty apply for every 0.1 miute improvemet or deterioratio i EWT score whe compared to the baselie (Table 2). The ratio of icetives to pealties is approximately i the order of 3:2 (S$1 = US$0.75). Table 2: BSRF icetive ad pealty amouts Fleet Size Category a 1 (< 10 buses) 2 (10 to <20 buses) 3 ( 20 buses) Icetive Amout per 0.1 miute improvemet per moth S$ 2000 (US$1500) S$ 4000 (US$3000) S$ 6000 (US$4500) Pealty Amout per 0.1 miute deterioratio per moth S$ 1300 (US$975) S$ 2600 (US$1950) S$ 4000 (US$3000) a This is the scheduled fleet for each bus service To allow bus operators to adjust to the ew framework, for the trial, LTA grated both operators a trasitio period from 3 February util 31 May 2014 whe EWTs will be moitored but o icetives or pealties will be applied. The trasitio period was desiged to give the bus operators more time to trai their bus drivers ad service cotrollers ad to fie tue their operatioal procedures to regulate bus arrival times. 4. Evidece ad Outcomes 4.1 Stated Preferece Surveys To provide some evidece o how bus users i Sigapore value differet aspects of the bus service they use, a Stated Preferece (SP) compoet focussig specifically o bus waitig times was icluded i the wider SP study coducted i Sigapore i 2015 to elicit a umber of ecoomic measures relatig to travel. The bus waitig time choice scearios explicitly look at how bus users value patters of waitig time at the bus stop, ad hece EWT improvemets ad ehacemets to bus service reliability. As part of the SP study, a represetative cross sectio of the Sigapore populatio were asked to record the trips they had made i the past 2 days ad were the asked about oe of these joureys i detail. They were asked to record the times ad costs ivolved i makig their jourey ad for motorised modes they were asked to provide similar iformatio about their alterative mode assumig they could ot make the jourey by their curret mode. This iformatio was the used as the basis of a umber of differet sets of SP scearios, or games. Most of the SP games were 8
withi mode so respodets were preseted with two optios relatig to their chose mode. However some respodets were preseted with a mode choice game where they chose betwee the differet travel modes, such as car, bus, MRT, ad taxi. A fial sample of 791 bus users received the Bus Waitig Time game. Below we provide the cotext behid this game, a example of how the optios were preseted plus a descriptio of the SP desig issues relatig to each desig. These are D- efficiet desigs ad have bee developed usig Ngee. The SP preseted each respodet with seve separate ulabelled biary choice tasks. Respodets were asked which of the two hypothetical optios they would choose o the basis of the iformatio preseted. I the Bus Waitig Time SP, respodets were asked to thik about the jourey they had provided iformatio about but imagiig a situatio where they had the choice betwee two future hypothetical bus services. The scheduled arrival patter of the bus was show as a referece i all games (always arrivig every 10 mi, depictig the sceario of perfect reliability) ad i each game, two sets of experimetally geerated itervals betwee buses were show as hypothetical alteratives A ad B. The bus fare they would have to pay was also show. It was assumed that buses arrived frequetly eough that they forget the timetable ; i other words, respodets were told to assume that their arrival time at the bus stop was completely arbitrary. Everythig else about their jourey remaied uchaged. The bus fare was pivoted aroud the curret sigle-trip fare reported by the respodet, with pivot levels of 30%, 15%, 0%, +15%, +30%. Several SP studies (see Li, Hesher, & Rose, 2010 for a review) have attempted to estimate the value of service reliability usig presetatios that rely o probabilities of early or late arrivals of a bus (relative to a schedule) as key attributes of the alteratives. I this work however, because most bus services i Sigapore ru at headways of less tha 15 mi ad therefore do ot have a posted schedule at the bus stop, it was ot very iformative to describe a bus service as experiecig schedule delays or of beig early, o-time or late. Istead, we decided to opt for a pictorial represetatio of reliability based o arrival itervals, such as i Fig. 1, which is also cosistet with how LTA has bee commuicatig the BSRF i public (LTA, 2014a). Moreover, the desig of the Bus Waitig Time game also allowed for the uderlyig EWT associated with each of the optios to be easily calculated ad hece, the estimatio of WTP measures for EWT improvemets, which was a key objective of the study. For the service reliability attribute, respodets were asked to cosider bus arrival patters over a time iterval of 60 mi. The scheduled arrival times, which depict the perfect sceario, do ot vary across choice tasks ad were based o a scheduled headway of 10 mi, for a frequecy of six buses per hour correspodig to six time itervals betwee arrivals. A scheduled headway of 10 mi was deemed a reasoable assumptio for the choice experimet as uder the PTC QoS stadards effective August 2009, operators are required to provide at least 80% of bus services at frequecies of ot more tha 10 miutes durig weekday peak periods. For the hypothetical scearios A ad B, the desig cotais six iterval attributes, each betwee 4 ad 16 mi ad with a coditio that the total headways sum to 60 mi i both optios. The EWT i the two hypothetical alteratives rages from 0.05 mi to 1.35 mi, with a average at 0.66 mi. 9
Figure 1: A example screeshot of a EWT choice experimet The data collected from this SP survey was ext aalysed usig advaced discrete choice models. We specified the models i valuatio space, such that the utility for alterative i would be give by V i = β fare (fare i + VTTEWT EWT i ) + ε i where ε i is a type I extreme value error term, fare i ad EWT i measure the fare ad EWT of alterative i, β fare is a estimated fare sesitivity ad VTTEWT is the moetary valuatio of a reductio i EWT. We worked with Mixed Logit models, i which we allow for radom heterogeeity across idividual decisio makers. We similarly estimated the correlatio betwee willigess to pay measures ad cost compoets. For distributioal assumptios, we largely relied o logormal distributios, with egative logormal distributios for cost compoets, ad positive logormal distributios for willigess-to-pay measures. With the log tail of the logormal distributio, a very small umber of outlyig values ca lead to extreme valuatios, ad with this i mid, attempts at very mior cesorig of the distributio were made, as discussed below. The resultig model structure is highly advaced ad flexible, with ofte large umbers of radomly distributed coefficiets, with correlatio betwee them. The degree of flexibility goes beyod what has bee used i may previous studies, where especially the estimatio of the full covariace matrix leads to substatial gais i model flexibility. These advatages were cofirmed i our empirical work, with large further gais i model fit ad more importatly also a greater ability to distiguish valuatios across idividual jourey compoets. 10
However, this icrease i flexibility also comes at the cost of icreased model complexity. Classical estimatio techiques for Mixed Logit were foud to be icapable of estimatig these models i a stable ad timely maer ad we istead tured to Bayesia estimatio. The uderlyig model structure is still MMNL, it is simply the process used to obtai estimates that chages. For a detailed discussio of Bayesia techiques for Mixed Logit, see Trai (2009, chapter 12). I our Bayesia estimatio, we made use of 1,000,000 bur-i iteratios for each model, with results the obtaied by averagig over 50,000 post-bur-i iteratios. After model estimatio, we also produced posterior estimates, thus geeratig for each idividual the most likely value for the various willigess-to-pay measures as a fuctio of the choices they were observed to have made (see Trai, 2009, chapter 11). These idividual-specific values are the used i posterior segmetatio work to attempt to ucover further determiistic heterogeeity. The model cotais oe cost compoet (fare) ad oe time compoet (EWT). We made use of a egative logormal distributio for fare ad specified the models i WTP space relative to fare, usig a positive logormal distributio for the valuatio of EWT. This leads to two radomly distributed coefficiets, where we estimate a full covariace matrix, meaig two diagoal terms (the variaces of the idividual utility compoets) ad oe off-diagoal term (the covariace betwee the two idividual utility compoets). The estimated parameters relate to the uderlyig Normal distributio, i.e. the log of the absolute values of the coefficiets (Table 3). Table 3: Estimatio results for the EWT choice experimet Parameters Estimate Respodets 791 Observatios 5,537 Estimated parameters 5 Log-likelihood -2,863.00 adj. ρ 2 0.25 Fare (uderlyig Normal mea for log of egative of coeff) Est. t-ratio 0.99 7.85 VTT EWT (uderlyig Normal mea for log of coeff) -1.89-7.13 cov(1,1) 5.52 7.18 cov(1,2) -8.83-6.99 cov(2,2) 14.24 5.98 Before studyig the results o the valuatio of EWT i detail, two core poits eed to be made. Firstly, the rages of EWT preseted i the experimet were by defiitio very arrow, give the use of a scheduled headway of 10 mi, ad a maximum ad miimum time i betwee bus arrival times of 4 ad 16 mi, respectively. This leads to a maximum EWT of just 1.35 mi, with a average of 0.67 mi. The resultig boudary EWT values, i.e. the trade-offs that respodets were faced with, raged from 7.69c/mi to 4800c/mi (i.e. SGD48). This would be the valuatio of EWT a respodet would eed to have to choose the more expesive optio i a give choice task (ad hece the oe with the lower EWT). The media accepted boudary (i.e. from choices where respodets accepted to pay more for lower EWT) was 75c/mi, while the media rejected boudary (i.e. from choices where respodets 11
refused to pay more for lower EWT) was 160c/mi, with respective meas of 114.87c/mi ad 374.05c/mi for accepted ad rejected boudaries. This thus directly explais the high estimated value of EWT we will ow tur to, which also eeds to be put ito cotext by otig that i practice, a oe miute chage i EWT is a major differece. For the valuatio of EWT, we cesored the logormal distributio at the highest accepted boudary value, which was 800c/mi, i.e. (SGD8). This is possible i this game where the choice i each sceario ca be quatified by a sigle boudary value. This cesorig led to a drop i log-likelihood to -2,921.65 uits, i.e. a drop by 58.65 uits. This is a o-trivial drop i fit, but this approach to cesorig was eeded i this game i order to obtai reasoable results. The result average valuatio of EWT is 71.74c/mi (Table 4). This is much higher tha the valuatios of i vehicle time from the same study, ad exceeds the average wage rate by a factor of more tha two. However, the value is a realistic estimate of the real valuatio, beig very close to the media accepted boudary value, ad it agai eeds to be bore i mid that achievig a miute reductio i EWT is a far bigger step tha a miute reductio i travel time. Table 4: Implied EWT valuatios Mea Std. Dev. Mea as % of wage rate Value of EWT (c/mi) 71.74 144.75 238 Wage rate from sample (SGD/hr) 18.11 I our aalysis of the posterior estimates, i.e. the most likely value for each compoet for each idividual, we ca make the followig observatios, where the small sizes for some of the subgroups eed to be take ito cosideratio (Table 5): - There is a strog idicatio of higher valuatio of EWT reductio o school trips, potetially as a result of schedulig costraits. - The EWT games show the strogest evidece of a meaigful icome effect, with higher valuatios for higher icome respodets. - Surprisigly, valuatios are higher for studets tha for other respodets, a fidig that is difficult to explai except o the basis of a higher household icome for their families ad a perceptio that it is ot them who pay for bus fares. - The valuatio is higher for those travellig i a group. Take together, the results will help LTA focus its bus reliability improvemets spatially ad temporally across sectios of the etwork where such improvemets are most likely to be valued accordig to the profile of bus users. 12
Table 5: Posterior aalysis of EWT valuatios Sample size Trip purpose Value of EWT (c/mi) No purpose 3 59.10 Home-based Other 229 60.44 Home-based School 101 90.07 Home-based Work 297 70.69 Not-home based 161 78.52 Icome levels No icome 312 69.98 Icome < 2000 168 70.02 Icome 2000-4000 142 78.74 Icome 4000-7000 62 83.00 Icome > 7000 23 90.04 Icome missig or refused Work full-time, part-time or self-employed Employmet status 84 56.57 469 72.31 housewife 104 55.94 studet 128 92.46 retired 66 48.49 uemployed or work NA 24 82.64 Travellig party size Travelled aloe 646 70.45 Travelled with others 145 77.51 4.2 Commuter awareess of the BSRF While it may be argued that bus users eed ot be burdeed with the techical details of the BSRF as log as they perceive a improvemet i bus service reliability, it is importat from the Govermet public relatios perspective that the public attributes the sigificat resources expeded o the scheme to the efforts of the Authority/regulator. I this regard, it is thought that the most saliet feature of the BSRF to the public would be the fiacial icetives/pealties for operators who exceed/fail to meet EWT baselie measures. However, a iteral Govermet survey coducted i late 2014 foud that the BSRF received just low to moderate awareess amog bus users (Figure 2).Whe asked how aware they were of the BSRF o a scale of 1 5 (1 Not at all aware, 2 Slightly aware, 3 Somewhat aware, 4 Moderately aware, 5 Extremely aware), just slightly over half of the bus users (56 percet) were at least somewhat aware of the framework, while close to a third were ot at all aware (32 percet). It is possible that the survey results are simply a reflectio of the limited scale of the trial at that time ad that a similar survey to be coducted i a year s time may yield differet results as the BSRF is ramped up across the islad. 13
Bus Service Reliability Framework Icetives ad pealties for operators to improve o bus service reliability 32 12 24 25 7 Not aware at all Slightly aware Somewhat aware Moderately aware Extremely aware Figure 2: Proportio of respodets aware of the Bus Service Reliability Framework o varyig degrees of awareess 4.3 Resultat Reliability Improvemets For the first assessmet period of Jue to November 2014, the reliability of 20 out of the 22 bus services uder trial has improved, as idicated i Table 6 (Lad Trasport Authority, 2015a). Of the bus services which improved, 18 eared icetives. The remaiig four services had performaces i the eutral zoe ad either eared ay icetive or icurred ay pealty. All truk bus services saw improvemets i EWT. Table 6: Summary of EWT improvemets i first year of BSRF trial Fleet size Bus EWT Baselie EWT for Ju to Nov Operator category Service (mi) 2014 (mi) Category 3 ( 20 buses) Category 2 (10 to <20 buses) Category 1 (< 10 buses) SBST SMRT SBST SMRT SBST SMRT EWT (mi) 3 1.5 1.1 0.4 39 a 1.2 1.1 0.1 51 2.2 1.6 0.6 154 1.9 1.4 0.5 176 1.6 1.4 0.2 188 1.4 1.1 0.3 858 2.1 1.8 0.3 17 1.8 1.4 0.4 52 2.0 1.4 0.6 184 1.3 1.0 0.3 189 1.4 1.2 0.2 853 1.5 1.0 0.5 911 a 1.3 1.4-0.1 962 a 1.4 1.3 0.1 228 1.1 0.7 0.4 241 1.2 1.0 0.2 242 0.8 0.6 0.2 292 0.9 0.7 0.2 325 0.8 0.6 0.2 354 1.1 0.6 0.5 302 a 1.0 1.0 0.0 901 0.9 0.7 0.2 a These services remaied i the eutral zoe ad did ot qualify for ay icetive or pealty. 14
Based o its performace over 6 moths, SBST eared about S$700,000 (US$525,000) for reliability improvemets to 11 services ad SMRT eared S$345,000 (US$259,000) for improvemets to seve services. I compariso, total bus fare reveue by SBST ad SMRT for the whole of 2014 was S$754 millio (US$566 millio) ad S$218 millio (US$163 millio) respectively, which works out to be a average of S$2.7 millio (US$2 millio) per route. There were o pealties deducted from both operators for this assessmet period as oe of their services had deteriorated more tha 0.1 mi to fall ito the pealty zoe. The icetives will help to offset the costs icurred by the PTOs to hire the additioal service cotrollers to support the BSRF. To further ivestigate the impact of the BSRF, we compared the daily ITP-specific EWT performace of a BSRF truk route (SMRT 853) agaist the daily ITP-specific EWT of SMRT 852, a o-bsrf truk route. SMRT 852 was chose o the basis of a sigificat route overlap with SMRT 853 from a bus iterchage i the orther part of the islad, so that factors exteral to the operator, such as road, traffic ad weather coditios, ca be largely cotrolled for whe comparig EWTs at the commo ITP. Route diagrams of SMRT 853 ad SMRT 852 with the approximate locatio of the ITP that the EWTs are based o are show i Appedix 1. A before ad after compariso relative to the start of the icetive-pealty regime i Jue 2014 shows that SMRT 853 s EWT has decreased by about 0.1 miutes, while SMRT 852 s EWT has icreased by the same magitude (Table 7). Table 7 also reports a simple differece-i-differece aalysis of the EWT of SMRT 852 relative to the EWT of SMRT 853 o a mothly basis from Jauary 2014, suggestig that BSRF has helped to improve the reliability of SMRT 853, after cotrollig for chages i operatig coditios that are proxied by the EWT scores of SMRT 852. Improvemets were detectable from April 2014 owards, such that by November of the same year, the EWT differece for SMRT 852 ad SMRT 853 has wideed by 0.7 miutes from the Jauary baselie. These are ecouragig idicatios for the BSRF trial i geeral. The success of the BSRF trial i its first year may be attributed to operators ijectig additioal resources required to moitor ad improve e-route reliability, such as havig more stadby buses ad drivers, as well as employig more service cotrollers to commuicate with drivers ad maage bus movemet. The surpassig of stadards i most of the assessed bus services does ot idicate low baselie stadards but rather the cocerted efforts of operators to improve reliability of services uder BSRF. Likewise, there were services that did ot improve despite operators puttig i their best efforts to maage these services. With these ad subsequet fidigs from the trial, LTA will cotiue to fie-tue the framework over time. 15
Table 7: Before-ad-after EWT comparisos of a selected pair of BSRF/o-BSRF bus routes SMRT 853 SMRT 852 Depedet variable: (EWT 852 EWT 853) t Costat 0.25*** (0.094) Average EWT from Ja to May 2014 before Icetive/Pealty Start Date Average EWT from Jul to Nov 2014 after icetive/pealty Start Date Differece i EWT (After Before) 0.81 0.89 0.71 1.01 0.10** (0.045) 0.12* (0.063) (0,1) variable for Feb (0,1) variable for Mar (0,1) variable for Apr (0,1) variable for May (0,1) variable for Jul (0,1) variable for Aug (0,1) variable for Sep (0,1) variable for Oct (0,1) variable for Nov 0.18 (0.14) 0.32 (0.17) 0.56*** (0.12) 0.52*** (0.12) 0.57*** (0.14) 0.74*** (0.14) 0.42*** (0.14) 0.35** (0.14) 0.70*** (0.149) Adj R 2 0.15 No. of observatios 202 Stadard errors are reported i paretheses. *, **, *** idicates sigificace at the 90%, 95% ad 99% level, respectively. 5. Coclusios This paper provides a overview of various mechaisms used by regulators to measure ad improve bus service reliability stadards, with a particular emphasis o the approach that Sigapore has take. From a empirical ad methodological perspective, aother sigificat cotributio of this paper is the estimatio of willigess-to-pay measures for EWT improvemets, the results of which will be useful from a policy perspective i helpig to determie a commesurate level of effort ad resources to be expeded o improvig bus reliability. Overall, the BSRF trial i Sigapore has yielded promisig results to date with a majority of bus services uder the trial otchig improvemets i their EWT scores. I Jue 2015, LTA aouced that the umber of bus services o the BSRF trial would be doubled to 45, ad that the trial would be exteded from February to August 2016 (Lad Trasport Authority, 2015b). LTA also aouced that a ew idicator, o-time adherece 2, which is used to measure the puctuality of bus services, will be trialled for two low frequecy services. Evetually, LTA iteds to brig all bus services uder the BSRF through the Bus Cotractig Model. Durig the tederig process, EWT baselie stadards for each bus service are clearly 2 O-time adherece (OTA) is the percetage of arrivals at selected bus stops that are withi a -2 mi/+5 mi rage of the scheduled arrival time. The OTA target is curretly set at 85%. 16
stipulated i cotractual documets, which tederers have to cosider whe plaig their timetables. With the start of operatios by Tower Trasit, Sigapore s third bus operator, o 29 May 2016, Sigapore s ew Bus Cotractig Model provides aveues for further research ito how cotract specificatios ad the ijectio of greater competitio advace the improvemet ad implemetatio of reliability stadards across the bus etwork. It would be iterestig for example to study if there are varyig degrees of success amog the various operators (icumbets ad ew etrats) i terms of ruig more reliable bus services, ad the factors that distiguish operators from each other. Amid risig commuter expectatios ad with a major push by the Govermet to provide better service ad make public trasport a choice mode, the emphasis o improvig reliability through the itroductio of ew measures ad regulatory frameworks could ot come at a more opportue time. Ackowledgemets The authors wish to ackowledge the ivaluable research assistace redered by Teck-Sig Chg, Zhi-Rog Chg, Jeffrey Je ad Zhi-Zha Lo i the preparatio of this paper. Commets from a aoymous reviewer are much appreciated. The views expressed i this paper are those of the authors ad do ot ecessarily reflect the official positio of the Lad Trasport Authority or of the Govermet of Sigapore. Ay remaiig errors are ours aloe. 17
Refereces Bates, J., Polak, J., Joes, P., Cook, A., 2001. The valuatio of reliability for persoal travel. Trasportatio Research Part E: Logistics ad Trasportatio Review, 37(2), 191-229. Carrio, C., ad Leviso, D., 2012. Value of travel time reliability: a review of curret evidece. Trasportatio Research Part A: Policy ad Practice, 46(4), 720-741. Cats, O., 2014. Regularity-drive bus operatio: Priciples, implemetatio ad busiess models. Trasport Policy, 36, 223 230. dell Olio, L., Ibeas, A., ad Cecıń, P., 2010. Modellig user perceptio of bus trasit quality. Trasport Policy, 17(6), 388 397. Disey, J., 1999. Customer satisfactio ad loyalty: The critical elemets of service quality. Total Quality Maagemet, 10(4-5), 491-497. Edvardsso, B., 1998. Causes of customer dissatisfactio studies of public trasport by the critical icidet method, Maagig Service Quality: A Iteratioal Joural, 8(3), 189 197. Feg, W., ad Figliozzi, M., 2012. Developig a bus service reliability evaluatio ad visualizatio framework usig archived AVL/APC data. Paper preseted at the Wester ITE Coferece, Sata Barbara, Califoria, USA. Goh, P., Swee, A., ad Low, J., 2015. Trasitio of Sigapore s public trasport idustry. Paper preseted at the Iteratioal Coferece Series o Competitio ad Owership i Lad Passeger Trasport, Satiago, Chile, 30 August 3 September 2015. Guo, G., Luo, H., Li, X., ad Feg, C., 2011. Headway-based evaluatio of bus service reliability. Paper preseted at the 14 th Iteratioal IEEE Coferece o Itelliget Trasportatio Systems, Washigto, DC, USA. Hesher, D.A., ad Houghto, E., 2002. Performace-Based Quality Cotracts for the Bus Sector: Deliverig Social ad Commercial Value for Moey. Trasportatio Research Part B: Methodological, 38(2), 123 146. Hesher, D.A., ad Staley, J., 2003. Performace-based quality cotracts i bus service provisio. Trasportatio Research Part A: Policy ad Practice, 37(6), 519 538. Huo, Y., Zhao, J., Li, W., ad Hu, X., 2014. Measurig bus service reliability: A example of bus rapid trasit i Chagzhou. Joural of Public Trasportatio, 17(2), 113-133. Ia Wallis Associates Ltd, ad The TAS Partership, 2013. Improvig bus service reliability. NZ Trasport Agecy research report, 527. [olie] Available from https://www.zta.govt.z/assets/resources/research/reports/527/docs/527.pdf [Accessed: 15 May, 2016] Köig, A., ad Axhause, K.W., 2002. The reliability of the trasportatio system ad its ifluece o the choice behaviour. Paper preseted at the Swiss Trasport Research Coferece (STRC), Mote Verità/Ascoa, March 20 22. Lad Trasport Authority, 2014a. To improve bus regularity, reduce bus buchig, 22 bus services uder BSRF startig from February 2014. [olie] Available from: 18
http://www.lta.gov.sg/apps/ews/page.aspx?c=2&id=5e6512c1-d964-4229-ba34-92c539a7c6b8 [Accessed: 29 Jue 2015] Lad Trasport Authority, 2014b. Early sigs that BSRF is reducig istaces of bus buchig ad prologed waitig times. [olie] Available from: http://www.lta.gov.sg/apps/ews/page.aspx?c=2&id=d95b67ae-7659-4ff0-8ec2-00957f2becc1 [Accessed: 29 Jue 2015] Lad Trasport Authority, 2015a. Bus service reliability framework (update for first assessmet period. [olie] Available from http://www.lta.gov.sg/apps/ews/page.aspx?c=2&id=edab460e-1915-43e2-a668-2e2d33792663 [Accessed: 29 Jue 2015] Lad Trasport Authority, 2015b. Expadig the bus service reliability trial. [olie] Available from http://www.lta.gov.sg/apps/ews/page.aspx?c=2&id=76c8984b-93ee-41a2-a06d- 3623c094450d [Accessed: 29 Jue 2015] Larse, O. I., 2001. Desigig icetive schemes for public trasport i Hordalad couty, Norway. Paper preseted at the 7 th Iteratioal Coferece o Competitio ad Owership of Lad Passeger Trasport, Molde, Norway. Li, Z., Hesher, D.A., ad Rose, J.M., 2010. Willigess to pay for travel time reliability i passeger trasport: a review ad some ew empirical evidece. Trasportatio Research Part E: Logistics ad Trasportatio Review, 46(3), 384-403. Li, J., Wag, M. L., ad Barum, D. T., 2008. A quality cotrol framework for bus schedule reliability. Trasportatio Research Part E: Logistics ad Trasportatio Review, 44(6), 1086-1098. Oort, N., 2014. Icorporatig service reliability i public trasport desig ad performace requiremets: iteratioal survey results ad recommedatios. Research i Trasportatio Ecoomics, 48, 92-100. Selviaridis, K., ad Wystra, F., 2014. Performace-based cotractig: a literature review ad future research directios. Iteratioal Joural of Productio Research, 53(12), 3505-3540. Strathma, J., Kimpel, T., Gerhart, R., Turer, K., Taylor, P., Callas, S., ad Griffi, D., 2000. Service reliability impacts of computer-aided dispatchig ad automatic vehicle locatio techology: A Tri-Met case study. Trasportatio Quarterly, 54(3), 85-102. Trai, K., 2009. Discrete choice methods with simulatio. Cambridge UK: Cambridge Uiversity Press. Trasport for Lodo, 2014. Travel i Lodo: Report 7. [olie] Available from https://www.tfl.gov.uk/cd/static/cms/documets/travel-i-lodo-report-7.pdf [Accessed: 11 Jue 2016] Trasport for Lodo, 2015. Lodo s Bus Cotractig ad Tederig Process. [olie] Available from http://cotet.tfl.gov.uk/uploads/forms/lbsl-tederig-ad-cotractig.pdf [Accessed: 15 May 2016] Vicet, M., 2008. Measuremet valuatio of public trasport reliability. Lad Trasport New Zealad Research Report, 339. [olie] Available from 19
https://www.zta.govt.z/assets/resources/research/reports/339/docs/339.pdf [Accessed 15 May 2016] 20
Table A.2 Route diagrams of SMRT 853 ad 852 APPENDIX 1 21