Identifying the Determinants of Light Rail Mode Choice for Medium- and Long-Distance Trips

Size: px
Start display at page:

Download "Identifying the Determinants of Light Rail Mode Choice for Medium- and Long-Distance Trips"

Transcription

1 Identfyng the Determnants of Lght Ral Mode Choce for Medum- and Long-Dstance Trps Results from a Stated Preference Study Leve Creemers, Maro Cools, Hans Tormans, Peter-Jan Lateur, Davy Janssens, and Geert Wets The ntroducton of new publc transport systems can nfluence socety n a multtude of ways rangng from modal choces and the envronment to economc growth. Ths paper examnes the determnants of lght ral mode choce for medum- and long-dstance trps (10 to 40 km) for a new lght ral system n Flanders, Belgum. To nvestgate these choces, the effects of varous transport system specfc factors (.e., travel cost, n-vehcle travel tme, transt punctualty, watng tme, access and egress tme, transfers, and avalablty of seats) as well as the travelers personal trats were analyzed by usng an alternatng logstc regresson model, whch explctly takes nto account the correlated responses for bnary data. The data used for the analyss stem from a stated preference survey conducted n Flanders. The modelng results are n lne wth lterature: most transport system specfc factors as well as socoeconomc varables, atttudnal factors, perceptons, and the frequency of usng publc transport contrbute sgnfcantly to the preference for lght ral transt. In partcular, the results ndcate that the use of lght ral s strongly nfluenced by travel cost and n-vehcle travel tme and to a lesser extent by watng and access egress tme. Seat avalablty appeared to play a more mportant role than dd transfers n decdng to choose lght ral transt. The fndngs of ths paper can be used by polcy makers as a frame of reference to make lght ral transt more successful. The mportance of transport as one of the key prerequstes of any modern socety cannot be downplayed. Transport enables people to reach servces and to mantan contacts and socal nteractons. Unfortunately, transport also has many negatve mpacts such as safety problems (e.g., traffc casualtes), envronmental pressure (e.g., greenhouse gas emssons), and economc losses (e.g., tme lost due to congeston) (1). To make transport more sustanable, transport Transportaton Research Insttute, Hasselt Unversty, Wetenschapspark 5, Bus 6, BE-3590 Depenbeek, Belgum. Addtonal afflatons for M. Cools: Centre for Informatcs, Modelng, and Smulaton, Hogeschool-Unverstet Brussel, Warmoesberg 26, BE-1000 Brussels, Belgum; and Research Foundaton Flanders, Egmontstraat 5, BE-1000 Brussels, Belgum. Correspondng author: M. Cools, maro.cools@hubrussel.be. Transportaton Research Record: Journal of the Transportaton Research Board, No. 2275, Transportaton Research Board of the Natonal Academes, Washngton, D.C., 2012, pp DOI: / systems that provde good alternatves to car use must be developed. In an attempt to allevate the negatve effects of car use and to acheve travel behavor that s more sustanable, the Flemsh publc transport company De Ljn s preparng an nvestment program to ntroduce a regonal lght ral network to provde adequate publc transport at medum dstances (10 to 40 km) (2, 3). Nevertheless, at ths moment, the concept of lght ral s stll relatvely unknown n Flanders (the Dutch-speakng northern regon of Belgum), snce ths mode of transportaton has not been mplemented yet. Consequently, there s a clear need to assess the mpact of a lght ral system n Flanders. The development of lght ral systems may have many effects, as ndcated n the lterature. They may range from shfts n modal splt and mproved accessblty to urban development and economc growth. Bllngs (4) nvestgated the mpact of a new lght ral system on property values n Charlotte, North Carolna, and demonstrated an ncrease n real estate prces wthn 1.6 km of the statons. Lght ral nvestments can therefore serve as an economc development tool. Senor (5) nvestgated the mpacts of a lght ral project n the Greater Manchester regon (Unted Kngdom) on travel behavor and concluded that the lght ral project contrbuted sgnfcantly to the declnng share of bus trps and work trps by car. Mackett and Edwards (6) analyzed the mpacts of 46 urban publc transport systems around the world, ncludng a seres of lght ral systems. In most nstances, they found a reducton n car use after mplementaton of the transt systems. In addton to the mpacts on modal splt, they reported mportant envronmental and accessblty effects. In general, these mpacts were postve (e.g., a reducton n ar polluton and mproved access to the cty center). However, n a few cases, negatve effects occurred (e.g., an ncrease n nose polluton). Furthermore, they consdered mpacts on urban development and land use. Transt nfrastructure stmulated ndustry and urban development around statons. Ths paper examnes the determnants of lght ral mode choce for medum- and long-dstance trps (10 to 40 km) for a new lght ral system n Flanders. To nvestgate these choces, the effects of a multtude of transport system specfc factors, such as travel cost, n-vehcle travel tme, transfers, avalablty of seats, access and egress tme, watng tme, and transt punctualty, as well as the travelers personal trats, are analyzed wth an alternatng regresson model (7, 8). 30

2 Creemers, Cools, Tormans, Lateur, Janssens, and Wets 31 Lterature Revew Key Determnants of Mode Choce n General In ths secton, an overvew of the factors that nfluence mode choce s gven. The overvew focuses on two categores: specfc factors of publc transport systems and personal trats and atttudes of the travelers. Wth regard to the frst category, t was found that relable travel tmes contrbute sgnfcantly to publc transport choce. Schramm et al. (9) and Van Loon et al. (10) ndcated that more relable travel tmes lead to an ncrease n transt rdershp. In addton, Zhang et al. (11) and Outwater et al. (12) found that punctualty of transt systems adds sgnfcantly to the mode choce decson process. In contrast, only one study could be found n whch the effect of relable travel tmes on mode choce turned out to be not sgnfcant (13). In addton to relable travel tmes, other transport system specfc factors, such as travel cost, n-vehcle travel tme, watng tme, access and egress tme, transfers, and avalablty of seats, affect mode choce. Mattson et al. (14), Outwater et al. (12), and Hensher and Rose (15) emphaszed the mportance of travel costs (.e., transt fares). Ben-Akva and Morkawa (16) llustrated that the varous travel tme components (.e., access and egress tme, n-vehcle tme, and watng tme) all contrbute sgnfcantly n explanng mode choce. Furthermore, they ndcated that travelers prefer modes that offer suffcent comfort (heatng, ar-condtonng, suffcent legroom, etc.). Outwater et al. (12) emphaszed the decreasng effect of transfers on transt choce. Fnally, the mportance of seat avalablty was stressed by Berlare et al. (17), who ndcated that suffcent free seats ncrease transt use. In addton to factors specfc to the transport system, varous personal trats and atttudes of travelers sgnfcantly nfluence mode choce. Age, gender, car ownershp, and ncome are often reported as factors nfluencng transt rdershp (e.g., 18 20). Personal atttudes have been cted as nfluencng the mode choce decson process (e.g., 21, 22). These studes show that people wth negatve atttudes toward publc transport and postve atttudes toward car use are less nclned to use publc transt. Fnally, Mattson et al. (14) found that ndvduals wth at least some transt experence are more lkely to choose publc transt and other alternatve modes. Key Determnants of Lght Ral Rdershp The drvng characterstcs for the specfc case of lght ral rdershp are pnponted n ths secton. They can be classfed nto four categores: system-specfc, socoeconomc, polcy-related, and regonal characterstcs. Ths paper focuses on the frst and second categores. However, for successful mplementaton of the lght ral system, the other factors must be taken nto account as well. Wth respect to the system-specfc attrbutes, one of the most mportant factors s the servce level, measured as the frequency or the tme span covered. In general, the hgher the level of servce, the hgher the lght ral rdershp (19, 23 25). Furthermore, travel costs have been cted as one of the key drvers: Kan and Lu (25) reported that tcket costs are negatvely related to lght ral rdershp. Next, speed contrbutes to rdershp: lower speed s related to hgher rdershp (23, 26). Ths negatve relaton appears to be llogcal, but t can be explaned by the fact that dwell tmes ncrease as loadngs rse. Accordngly, routes wth hgh rdershp tend to be slower. In addton, Kuby et al. (24) found that hgh-qualty connectons (short walkng dstances between modes combned wth coordnated and closely scheduled arrval and departure tmes) to other forms of publc transport contrbute to the success of the system. Tcket ntegraton (a sngle tcket for varous transport modes) between publc transport modes s also cted as a success factor (19, 23, 26). Wth regard to socoeconomc characterstcs, Mackett and Babalk-Sutclffe (19) found that hgh car ownershp and hgh ncomes reduce lght ral rdershp. Wth regard to polcy-related attrbutes, Mackett and Babalk-Sutclffe (19) showed that offerng (temporally) free travel for target groups ncreases rdershp. Furthermore, they ndcated that marketng campagns enlarge the travelers knowledge of the lght ral system, whch n turn augments rdershp levels. Fnally, wth regard to regonal characterstcs, a number of studes emphasze the mportance of land use features (e.g., 19, 23 25). Areas wth hgh employment and retal and resdental denstes generate more trps than do regons wth low denstes. Moreover, lght ral systems are more lkely to be successful when they serve areas wth economc growth. In ts turn, the development of a lght ral system contrbutes to the economc and urban development of the regon as t generates attractve locatons for retal settlements (19). Data A stated preference (SP) survey was conducted to dentfy the determnants of lght ral mode choce for medum- and long-dstance trps for a new lght ral system n Flanders (the northern part of Belgum). In 2010, the regon had about 6.2 mllon nhabtants. Flemsh resdents make 2.9 trps per day on average; the majorty of the trps (66.8%) are carred out by car. Slow modes account for 26.4% of the trps, whle publc transport has a share of 5.3% (27). In SP approaches, respondents ndcate ther preferences among a set of alternatves for varous hypothetcal stuatons (28 30). SP surveys are common n travel behavor research and have been extensvely appled to the analyss of modal choces (28). SP approaches allow researchers to dentfy behavoral responses to new transport optons and travel condtons that have not yet been experenced. Such responses are not (yet) revealed on the market (30). However, SP data have one major drawback. They descrbe only what an ndvdual clams he or she would do n a gven scenaro, whch does not always correspond to the actual or revealed behavor (28, 30). One reason for ths msmatch s that respondents mght gve socally desrable answers. Despte ths dsadvantage, SP approaches have already proved successful n capturng ndvdual preferences under new choce stuatons. Louvere et al. (31) showed that stated behavor s a good approxmaton of actual (revealed) behavor when socally desrable answers are controlled for. In the current research, socally desrable answers are mtgated by takng the frequency of publc transt use and the atttudes toward the varous transport modes as controls for nherent preferences. The SP survey was conducted on a person-based level from early December 2010 to late January 2011 and was completed by random ndvduals over 18 years of age. The majorty of the questonnares were dstrbuted over the Internet. However, smlar tradtonal paperand-pencl questonnares were handed out to counteract the sample bas arsng when only web-based data are collected (32, 33). In total, the survey collected valuable nformaton from 492 respondents. The SP questonnare conssted of three parts n whch the respondents had to ndcate ther preference for, respectvely, (a) car use versus lght ral transt, (b) bus transt versus lght ral transt, and (c) tran transt versus lght ral transt. Each part contaned eght

3 32 Transportaton Research Record 2275 You lve n the center of a small cty called A. You want to perform a work actvty. For ths, you need to travel to the center of a larger cty called B. The towns are 30 km apart. Whch alternatve do you prefer, gven followng trp characterstcs? Total Travel Tme Egress Tme Watng Tme Cost Transfers Seats Alternatve 1 Bus 43 mn 59 mn 5 mn 6 mn 2 No Free seats Alternatve 2 Lght ral 25 mn 35 mn 5 mn 5 mn 5 Yes No free seats Bus Lght ral FIGURE 1 Example of alternatves n a hypothetcal stuaton. hypothetcal stuatons wth varyng trp characterstcs. The trp characterstcs ncluded total travel tme, access and egress tme, watng tme, travel cost, transfers, and avalablty of seats. Access and egress tmes were defned as the necessary tmes for travelng to and from the staton, respectvely. In contrast to trp characterstcs, trp dstance and trp motve remaned constant across the hypothetcal stuatons. Trp dstance was fxed at 30 km, snce the goal of lght ral transt s to provde transport at the regonal level. Trp purpose was set as the most frequently performed purpose ndcated by the respondent. Ths could be a work trp, a shoppng trp, or a lesure trp. In total, each respondent was confronted wth 24 (= 3 8) stuatons. Fgure 1 shows an example of such a hypothetcal stuaton n the survey. For each hypothetcal stuaton, the respondents had to choose between exactly two alternatves. Ths was a conscous choce, snce research had shown that augmentng the number of alternatves n the experment would enlarge the cogntve burden of the survey, and the respondents would gnore some of the nformaton (34). In addton to the SP questons, the survey quered some socoeconomc varables n a personal questonnare (e.g., age, gender, ncome, household sze, number of chldren, owned vehcles). Informaton about the frequency of usng dfferent transport modes was also obtaned. Atttudes toward varous transport modes were surveyed, as well as the mportance that respondents attached to, respectvely, a fast, a convenent, an nexpensve, an envronmentally frendly, and a safe trp. Respondents perceptons of the varous modes wth regard to comfort, envronment, safety, and speed were quered. The respondents expected values of travel tme, watng tme, access and egress tme, cost, and number of transfers n a trp of 30 km were surveyed and used as a bass for comparson of the values offered n the hypothetcal stuatons. Fnally, nformaton was gathered about the mportance that respondents attach to specfc features of the staton or stop locatons, such as lghtng, guarded bcycle parks, and dynamc nformaton. Table 1 gves an overvew of the defntons and the correspondng measurement unts of the varables that were collected n the survey. Because of the large number of varables, only the varables that are ncluded n the fnal models (Tables 2 and 3 of the results secton) are presented here. To attan an optmal correspondence between the survey sample composton and the Flemsh populaton, the observatons n the sample were weghted. The weghts were calculated by matchng the margnal dstrbutons of the sample wth the margnal dstrbutons of the populaton on the bass of the key person-level attrbutes age, sex, and martal status. Methodology As stated earler, the man research objectve of ths paper s the assessment of the mpact of varous transport system specfc factors (such as travel cost, n-vehcle travel tme, transfers, avalablty of seats, access and egress tme, watng tme, and transt punctualty) and the travelers personal trats on modal choce n the presence of lght ral transt. The prevous secton ndcated that each respondent had to gve the preferred mode (a bnary choce) for a number of hypothetcal stuatons. Ths mples that multple answers for a sngle respondent were recorded and that correlaton among these repeated observatons cannot be dsregarded. Therefore, a modelng approach that takes nto account correlated responses for bnary data s needed. The model adopted to fulfll ths requrement s a generalzed estmatng equatons (GEE) model for bnary data wth the logt lnk functon. The mean response s modeled as a logstc regresson model, whch s defned as follows (7): π* log θ* β* () 1 1 π* = + X where π* 1 π* = odds, θ* = ntercept, β* = vector of model parameters to be estmated, and X = vector of explanatory varables. The above equaton can be rewrtten as the well-known lkelhood functon of a bnary logt model: ( ) ( ) exp θ* + X β* π* = 1+ exp θ* + X β* ( 2)

4 Creemers, Cools, Tormans, Lateur, Janssens, and Wets 33 TABLE 1 Overvew of Varables Collected n the Survey Varable Defnton Measurement Unt Transport System Specfc Varables Cost Total cost for the traveler when usng LRT (nclude access and egress mode costs) Access egress tme Necessary tme to travel to and from LRT staton Mnutes Seat avalablty Avalablty of suffcent free seats on LRT vehcle Yes/no Transfers Need to make transfers durng the trp Yes/no In-vehcle travel tme Total travel tme on LRT vehcle Mnutes Transt punctualty Varaton n travel tmes (e.g., due to delays) Mnutes Watng tme The total tme spent watng at boardng staton Mnutes Socoeconomc Varables Age Years passed snce brth Years Sex Gender Man or woman Number of cars The number of cars n the household Absolute values Frequency of usng publc transport Regularty of publc transport use Daly, weekly, monthly, several tmes a year, never Atttudnal Varables Atttude toward car Feelngs or mnd-set toward the car 7-pont Lkert scale (1 = very postve,..., 7 = very negatve) Atttude toward tran Feelngs or mnd-set toward the tran 7-pont Lkert scale (1 = very postve,..., 7 = very negatve) Atttude toward tram Feelngs or mnd-set toward the tram 7-pont Lkert scale (1 = very postve,..., 7 = very negatve) Percepton of bus, tram, or metro wth regard to comfort To what extent do people fnd a bus, tram, or metro trp comfortable? 8 7-pont Lkert scale (1 = very comfortable,..., 7 = not comfortable at all) Percepton of tran wth regard to cost To what extent do people fnd a tran trp nexpensve? 7-pont Lkert scale (1 = very cheap,..., 7 = not cheap at all) Importance of fast travelng How mportant s fast travelng to the traveler? 7-pont Lkert scale (1 = very mportant,..., 7 = not mportant at all) Importance of nexpensve travelng How mportant s nexpensve travelng to the traveler? 7-pont Lkert scale (1 = very mportant,..., 7 = not mportant at all) Expected watng tme of a 30-km trp (relatve) Note: LRT = lght ral transt. Expected watng tme of an magnary 30-km trp Relatve values (dfference n expected watng tmes between bus and LRT) Equaton 1 shows that the estmated parameters must be nterpreted as the change n the predcted logged odds for a one-unt change n the correspondng explanatory varable. The odds can be defned as the probablty of an event dvded by the probablty of no event. In ths paper, the probablty of an event equals the lkelhood of choosng lght ral transt. The most common way to nterpret the parameter estmates s accordng to the odds ratos (ORs). An OR can be obtaned by takng the exponent of the parameter estmate (e β ). If the OR s smaller (greater) than 1, t represents a decrease (ncrease) n the odds of an event (.e., choosng lght ral). Ths mples that the probablty decreases (ncreases) sgnfcantly for every unt ncrease n the correspondng explanatory varable. Parameter estmates can also be construed by examnng ther sgns. A postve (negatve) sgn mples an ncrease (decrease) n the lkelhood of an event for every ncrease n the correspondng explanatory varable. GEE models take nto account the correlaton between dfferent observatons of the same subject (.e., repeated answers by the same respondent) by explctly modelng the correlaton structure of the repeated observatons. Correlaton structures specfy how observatons wthn a subject or cluster are correlated wth each other. For bnary data, the correlaton between the jth and the kth response s by defnton the followng (18): ( ) = Corr Y, Y j k ( j k ) j k ( ) Pr Y = 1, Y = 1 µµ µ ( 1 µ j ) µ k 1 µ k j () 3 However, the above formula has one mportant dsadvantage. The correlaton s constraned to be wthn lmts that depend n a complcated way on the means of the data. In contrast, the OR s not constraned by the means and s therefore preferred. The OR s defned as follows (8): ( ) = OR Y, Y j k ( ) ( = = ) ( j k ) ( j = k = ) Pr Yj = 1, Yk = 1 Pr Yj 0, Yk 0 Pr Y = 1, Y = 0 Pr Y 0, Y 1 ( 4) The latter mplementaton of GEE s called alternatng logstc regresson (ALR). In general, ALR models the assocaton between responses wth log ORs nstead of wth correlatons, as do ordnary GEE (8).

5 34 Transportaton Research Record 2275 TABLE 2 Results of Overall Sgnfcance Type III Test of Travel Mode Choce Model Car Versus Lght Ral Bus Versus Lght Ral Tran Versus Lght Ral Parameter DF χ 2 P-Value Sgn. χ 2 P-Value Sgn. χ 2 P-Value Sgn. Transport System Specfc Varables Cost <.0001 *** <.0001 *** <.0001 *** Access egress tme <.0001 *** <.0001 *** * Seat avalablty <.0001 *** <.0001 *** <.0001 *** Transfers <.0001 *** NS <.0001 *** In-vehcle travel tme <.0001 *** <.0001 *** <.0001 *** Transt punctualty NS NS NS Watng tme <.0001 *** <.0001 *** *** Socoeconomc Varables Age NS NS NS Sex NS ** * Frequency of usng publc transport ** / / / / / / Number of cars * / / / / / / Atttudnal Varables Atttude toward car <.0001 *** / / / / / / Atttude toward tram ** * / / / Atttude toward tran 1 / / / / / / * Importance of nexpensve travelng 1 / / / *** / / / Importance of fast travelng 1 / / / * / / / Percepton of tran wth regard to cost 1 / / / / / / ** Percepton of bus, tram, or metro wth ** / / / * regard to comfort Expected watng tme of a 30-km trp (relatve) 1 / / / * / / / Note: DF = degrees of freedom; sgn. = sgnfcance; / = not applcable. * P-value <.05, ** P-value <.01, *** P-value < 0.001, NS = not sgnfcant. Three models were estmated to assess the mpact of varous transport system specfc factors: the bnary choce between car use and lght ral transt (Model 1), between bus transt and lght ral transt (Model 2), and between tran transt and lght ral transt (Model 3). Snce transport system specfc attrbutes may not explan mode choce fully, other varables (such as personal trats and atttudes) that may have an nfluence were added as control varables. When the models were bult, forward selecton was used to fnd the most relevant varables n the model. Forward selecton adds varables to the model one at a tme. At each step, each varable that was not already n the model s tested for ncluson. The most sgnfcant varable s then added to the model as long as ts P-value remans below the sgnfcance level of.05. The fnal models were assessed for multcollnearty by usng tolerance and varance nflaton factor values, but no problems occurred. The results of the model estmatons are presented below. Results Overall Results The overall sgnfcance tests for the fnal models are gven n Table 2. The table ndcates that, n all three models, almost all transport system specfc factors sgnfcantly affect the choce of lght ral transt (P-values are below.05). An excepton s the punctualty of lght ral transt, whch appears not to be sgnfcant n any of the three models. In addton, the varable transfers shows no sgnfcant effect when the choce between bus transt and lght ral transt s modeled. Other factors that nfluence mode choce were taken nto account n the models as well. Table 2 ndcates that varous socoeconomc varables, atttudnal factors, and perceptons, as well as the frequency of usng publc transport (only Model 1), sgnfcantly nfluence the preference for lght ral transt. Sex and age are not always sgnfcant but were kept n the fnal models to control for Type I errors (also known as false postves) (35). It was also found that the expected watng tme for lght ral transt for a 30-km trp was sgnfcant when the choce between bus transt and lght ral transt was modeled. The expected watng tme s relatve: t s the dfference between the expected watng tme for bus and the expected watng tme for lght ral. If the value s greater than zero, the watng tme for bus transt s larger than the watng tme for lght ral transt for the same trp and vce versa. Parameter Estmates The parameter estmates for the bnary mode choce models are shown n Table 3. As stated earler, the most common way to nterpret these

6 Creemers, Cools, Tormans, Lateur, Janssens, and Wets 35 TABLE 3 Parameter Estmates for Bnary Travel Mode Choce Model Model 1. Car (0) Versus Lght Ral (1) Model 2. Bus (0) Versus Lght Ral (1) Model 3. Tran (0) Versus Lght Ral (1) Parameter Estmate SE OR Estmate SE OR Estmate SE OR Intercept / / / Transport System Specfc Varables Cost Access egress tme Free seats Yes No Ref. Ref. Ref. Ref. Ref. Ref. Ref. Ref. Ref. Transfers Yes a No Ref. Ref. Ref. Ref. Ref. Ref. Ref. Ref. Ref. In-vehcle travel tme Transt punctualty a a a Watng tme Socoeconomc Varables Age a a a Sex Man a Woman Ref. Ref. Ref. Ref. Ref. Ref. Ref. Ref. Ref. Frequency of usng publc Transport Daly / / / / / / Weekly a / / / / / / Monthly a / / / / / / Several tmes a year a / / / / / / Never Ref. Ref. Ref. / / / / / / Number of cars / / / / / / Atttudnal Varables Atttude toward car / / / / / / Atttude toward tram / / / Atttude toward tran / / / / / / Percepton of bus, tram, or metro / / / wth regard to comfort Percepton of tran wth regard / / / / / / to cost Importance of nexpensve travelng / / / / / / Importance of fast travelng / / / / / / Expected watng tme of a 30-km trp (relatve) / / / / / / Note: / = not applcable; ref. = reference category, no parameter estmated. a Not sgnfcant at the.05 level. parameter estmates s accordng to the OR. The OR of travel cost n the car versus lght ral model s Ths represents a decrease n the odds for lght ral use of 36% for every 1 ncrease n tcket prce and mples that the probablty of choosng the lght ral opton decreases sgnfcantly for every ncrease n tcket prce and that people wll be more lkely to take the car. Smlar conclusons can be drawn for the remanng two models. When tcket prces rse, people are less lkely to choose the lght ral opton and are more lkely to choose the bus or tran alternatve. Next, the OR for the varable access egress tme shows that a 1-mn ncrease n access egress tme wll decrease the odds for lght ral by 7.5%, 16%, and 6% for the car, bus, and tran models, respectvely. Thus, every ncrease n lght ral s access egress tme sgnfcantly lowers the lkelhood of usng lght ral. Smlar conclusons can be drawn for lght ral s n-vehcle travel tme and watng tme. The OR mples that an ncrease n these tme components leads to a sgnfcantly lower probablty of lght ral use. Furthermore, when lght ral vehcles have suffcent seats avalable, people s probablty of optng for lght ral ncreases n all three models. The odds of usng lght ral when seats are avalable are 1.7, 2.4, and 5.2 tmes the odds when no seats are avalable for the car, bus, and tran models, respectvely. The opposte holds for the varable transfers. An nterpretaton of the OR ndcates that the lkelhood of usng lght ral decreases sgnfcantly when one has to make transfers. Ths s not the case n the bus versus lght ral model, where the reverse s true. However, as mentoned above, ths effect s not sgnfcant.

7 36 Transportaton Research Record 2275 A number of atttudnal factors and perceptons contrbuted sgnfcantly to the choce of lght ral transt. The OR ndcates that a postve atttude toward the car (Model 1) wll decrease the lkelhood of lght ral use and wll ncrease the probablty of car use. Smlar conclusons can be drawn for the atttude toward tran (Model 3), whereas the opposte s true for the atttude toward tram (Model 1 and Model 2). A postve atttude toward tram wll enhance the lkelhood of usng lght ral. Moreover, people who beleve that travelng by bus, tram, or metro s not comfortable (Model 1 and Model 3) are less lkely to travel by lght ral. People who beleve that tran s expensve (Model 3) are more lkely to use lght ral, whle people who attach great mportance to nexpensve travelng (Model 2) are more lkely to take the bus and are less nclned to use lght ral. People who attach great mportance to fast travel (Model 2) have hgher probabltes of usng lght ral. Wth regard to socoeconomc factors, t appears that men are more nclned to use lght ral than women (Models 2 and 3) and that a hgh number of cars n the household (Model 1) wll lower the probablty of usng lght ral. Fnally, t appears that current frequent publc transport users (Model 1) are more nclned to choose lght ral and have a lower probablty of choosng the car. Dscusson of Results In the precedng secton, the relatonshp between transt punctualty and mode choce was shown not to be sgnfcant at the.05 level. Ths was surprsng snce the majorty of studes n the lterature ndcated the opposte. The nsgnfcance of transt punctualty n the current study can be accounted for by the fact that the devatons of the travel tmes n the survey were small (3 to 5 mn) n comparson wth the overall travel tme of the 30-km trps. The effect of larger devatons on lght ral mode choce s not explored n ths paper. Thus, the concluson s confned to the fact that small devatons n travel tmes have no sgnfcant nfluence on lght ral mode choce for medum- and long-dstance trps. The results of the other transport system specfc factors (.e., travel cost, n-vehcle travel tme, watng tme, access and egress tme, transfers, and avalablty of seats) are n lne wth the lterature: these factors all affect mode choce sgnfcantly n the way one would expect. Only the varable transfers shows no sgnfcant effect when the choce between bus transt and lght ral transt s modeled. A possble reason s that people may mplctly assume that f a transfer for lght ral s requred, one would also be requred for usng the bus, whch would negate the overall effect of transfers. Moreover, Table 2 ndcates that travelers are strongly nfluenced by the cost of lght ral (large χ 2 -values, same number of degrees of freedom). Travel cost s the most mportant factor when the choces between car use and lght ral transt and between tran transt and lght ral transt are modeled. From the χ 2 -values of the tme components, t can be nferred that people are most nfluenced by n-vehcle travel tme and are nfluenced to a lesser extent by watng and access and egress tmes (although the nfluence of those tmes s stll hghly sgnfcant). Furthermore, t appears that travelers pay more attenton to the avalablty of seats than to transfers (larger χ 2 -values, same number of degrees of freedom). The latter fndngs can be explaned by the fact that the correspondng n-vehcle travel tme s large compared wth the total travel tme and by the fact that for medum- and long-dstance trps a lack of empty seats s perceved as unfavorable. Varous personal trats of the travelers contrbuted sgnfcantly to the choce for lght ral transt. Ths agan s n accordance wth the general lterature concernng transt mode choce. The lterature has demonstrated that, n addton to age, gender, and car ownershp, ncome s a man determnant of publc transt mode choce. Nevertheless, ncome was not ncluded n the fnal models presented n ths paper. Income and number of cars are closely correlated, whch mples that hgher ncomes make ownng a car more feasble. As a result, the ncome effect s ndrectly ncluded n the models by means of the number of cars varable. The fndngs wth regard to atttudes are also confrmed by the lterature. Table 3 ndcates that a postve atttude toward the car (Model 1) wll decrease the lkelhood of usng lght ral, whereas a postve atttude toward the tram (Models 1 and 2) wll enhance the lkelhood of usng lght ral. Ths can be explaned by the fact that a tram s also a publc transport mode that mght be vewed as a good approxmaton to lght ral. The results ndcated that persons who attach great mportance to fast travelng are more nclned to use lght ral transt than to use bus servce. Ths s confrmed by Scherer (36), who found that travelers are more attracted to lght ral transt than to bus transt, even f both servces offer the same level of servce. Scherer explans the dfference n rdershp by suggestng that lght ral transt s consdered to be faster than bus servce because t has ts own rght-of-way. In addton, the results ndcated that travelers who regard trans as expensve (Model 3) are more lkely to use lght ral, whle travelers who attach great mportance to nexpensve travelng (Model 2) are more lkely to take the bus and are less nclned to use lght ral. These results may ndcate that people see lght ral as an expensve but fast publc transportaton mode. Polcy Recommendatons The fndngs n ths paper provde nsght nto the success factors of a (new) lght ral system for medum- and long-dstance trps and can be helpful to polcy makers n makng lght ral more successful. Travel cost and n-vehcle travel tme were dentfed as the most mportant factors for travelers n choosng to use lght ral. On the bass of the assumpton that polcy makers wll prmarly am at shftng car users to lght ral transt for trps of moderate length (10 to 40 km) wth urban or suburban destnatons, the fndngs of ths research suggest that ther flankng measures durng ntroducton of the lght ral network should be orented toward the cost-effectveness and mmunty to congeston of ths travel mode. Travelers can be convnced to exchange ther cars for a lght ral tran by drawng ther attenton to the low travel cost per klometer compared wth the real cost of drvng (ncludng fxed costs such as nsurance and deprecaton). To ths end, subsdzng (perhaps temporarly) lght ral trps for partcular target groups (e.g., commuters, large famles, persons wth low ncomes) can help ncrease the chances of successful ntroducton of a lght ral system (19). Regonal authortes can play a major role n ths respect through ntensve promotonal campagns and by encouragng destnaton ctes to partcpate n a system of thrd-party payers (37). In addton to the polcy measures above, accompanyng the ntroducton of a lght ral network wth a car restrant polcy

8 Creemers, Cools, Tormans, Lateur, Janssens, and Wets 37 may help ncrease the success of the ntroducton (19). Road prcng and hgher road taxes may be part of such a polcy. The bundlng of road prcng wth mproved publc transportaton alter natves ncreases the acceptablty and consequently the effectveness of road prcng (38). Constranng n-vehcle travel tme s mportant to polcy makers n makng lght ral transt more successful. Lmtng the number of stops to the absolute mnmum and careful consderaton of stop locatons can make mportant contrbutons. In addton, reducng dwell tme at stops by elmnatng onboard tcket sales by the drver can sgnfcantly lessen total run tme (39). Onboard tcket vendng machnes, vendng machnes at statons, and tcket sales by new technologes such as short message servce, rado frequency dentfcaton, and electronc cash systems can be good alternatves. Conclusons and Further Research The mpact of varous transport system related factors as well as soco demographc varables on the use of lght ral transt for medum- and long-dstance trps n Flanders, Belgum, was nvestgated. Results from an ALR model confrm that most of these factors sgnfcantly nfluence the use of lght ral. The results are n lne wth nternatonal lterature. Hence, the key varables for lght ral mode choce appear to be stable across dfferent contexts, whch mples that best-practce examples mght be applcable across dfferent geographcal contexts. The research fndngs can be used by polcy makers n makng the mplementaton of lght ral transt more successful. The results of the paper ndcate that there would be a shft toward lght ral, but whether the shft can be characterzed as major s uncertan unless the addtonal polces that were dscussed n the polcy recommendatons secton are also mplemented. The effects of these measures are not analyzed n ths paper and are an area for further research. In addton, mportant changes n land use and urban development around the statons can be expected. Hence, constructon of a model ntegratng travel mpacts wth these land use and urban developments could be ntrgung. References 1. Steg, L. Can Publc Transport Compete wth the Prvate Car? Internatonal Assocaton of Traffc and Safety Scences Research, Vol. 27, 2003, pp Regonet Lmburg. Regonaal Openbaar Vervoer n de Provnce Lmburg. Conceptstreefbeeld. Vervoersmaatschappj De Ljn, Mechelen, Belgum, Varna. Ontwerpstreefbeeld Spartacus Sneltramljn 1. Vervoersmaatschappj De Ljn, Hasselt, Belgum, Bllngs, S. B. Estmatng the Value of a New Transt Opton. Regonal Scence and Urban Economcs, Vol. 41, No. 6, 2011, pp DOI: /j.regscurbeco Senor, M. L. Impacts on Travel Behavour of Greater Manchester s Lght Ral Investment (Metrolnk Phase 1): Evdence from Household Surveys and Census Data. Journal of Transport Geography, Vol. 17, 2009, pp Mackett, R. L., and M. Edwards. The Impact of New Urban Publc Transport Systems: Wll the Expectatons Be Met? Transportaton Research Part A, Vol. 32, No. 4, 1998, pp Allson, P. D. Logstc Regresson Usng SAS: Theory and Applcaton. SAS Insttute, Inc., Cary, N.C., SAS/STAT User s Gude. SAS Insttute, Inc., Cary, N.C., Schramm, L., K. Watkns, and S. Rutherford. Features That Affect Varablty of Travel Tme on Bus Rapd Transt Systems. In Transportaton Research Record: Journal of the Transportaton Research Board, No. 2143, Transportaton Research Board of the Natonal Academes, Washngton, D.C., 2010, pp Van Loon, R., P. Retveld, and M. Brons. Travel-Tme Relablty Impacts on Ralway Passenger Demand: A Revealed Preference Analyss. Journal of Transport Geography, Vol. 19, 2011, pp Zhang, J., A. Fujwara, and S. Then. Capturng Travelers Stated Mode Choce Preferences Under Influence of Income n Yangon Cty, Myanmar. Journal of Transportaton Systems Engneerng and Informaton Technology, Vol. 8, No. 4, 2008, pp Outwater, M. L., G. Sptz, J. Lobb, M. Campbell, B. Sana, R. Pendyala, and W. Woodford. Characterstcs of Premum Transt Servces That Affect Mode Choce. Transportaton, Vol. 38, No. 4, 2011, pp Ahern, A. A., and N. Tapley. The Use of Stated Preference Technques to Model Modal Choces on Interurban Trps n Ireland. Transportaton Research Part A, Vol. 42, 2008, pp Mattson, J., D. Peterson, D. Rpplnger, W. Thoms, and J. Hough. An Assessment of Demand for Rural Intercty Transportaton Servces n a Changng Envronment. In Transportaton Research Record: Journal of the Transportaton Research Board, No. 2145, Transportaton Research Board of the Natonal Academes, Washngton, D.C., 2010, pp Hensher, D. A., and J. M. Rose. Development of Commuter and Non- Commuter Mode Choce Models for the Assessment of New Publc Transport Infrastructure Projects: A Case Study. Transportaton Research Part A, Vol. 41, No. 5, 2007, pp Ben-Akva, M., and T. Morkawa. Comparng Rdershp Attracton of Ral and Bus. Transport Polcy, Vol. 2, 2002, pp Berlare, M., K. W. Axhausen, and G. Abay. The Acceptance of Modal Innovaton: The Case of Swssmetro. Presented at 1st Swss Transport Research Conference, Monte Vertà, Ascona, Swtzerland, Chatterjee, K. Modellng the Dynamcs of Bus Use n a Changng Travel Envronment Usng Panel Data. Transportaton, Vol. 38, No. 3, 2011, pp Mackett, R., and E. Babalk-Sutclffe. New Urban Ral Systems: A Polcy Based Technque to Make Them More Successful. Journal of Transport Geography, Vol. 11, No. 2, 2003, pp Popur, U., K. Proussaloglou, C. Ayvalk, F. Koppelman, and A. Lee. Importance of Traveler Atttudes n the Choce of Publc Transportaton to Work: Fndngs from the Regonal Transportaton Authorty Atttudnal Survey. Transportaton, Vol. 38, No. 4, 2011, pp Murray, S. J., D. Walton, and J. A. Thomas. Atttudes Towards Publc Transport n New Zealand. Transportaton, Vol. 37, No. 6, 2010, pp Domarch, C., A. Tudela, and A. González. Effect of Atttudes, Habt and Affectve Apprasal on Mode Choce: An Applcaton to Unversty Workers. Transportaton, Vol. 35, No. 5, 2008, pp Curre, G., A. Ahern, and A. Delbosc. Explorng the Drvers of Lght Ral Rdershp: An Emprcal Route Level Analyss of Selected Australan, North Amercan and European Systems. Transportaton, Vol. 38, No. 3, 2011, pp Kuby, M., A. Barranda, and C. Upchurch. Factors Influencng Lght-Ral Staton Boardngs n the Unted States. Transportaton Research Part A, Vol. 38, 2004, pp Kan, J., and Z. Lu. Secrets of Success: The Large Increases n Transt Rdershp Acheved by Houston and San Dego Transt Provders. Transportaton Research Part A, Vol. 33, 1999, pp Hass-Klau, C., and G. Crampton. Future of Urban Transport: Learnng from Success and Weakness Lght Ral. Envronmental and Transport Plannng, Brghton, Unted Kngdom, Cools, M., K. Declercq, D. Janssens, and G. Wets. Onderzoek Verplaatsngsgedrag Vlaanderen 4.2 ( ): Tabellenrapport [Travel Behavor Research Flanders 4.2 ( ): Table Report]. Hasselt Unversty, Depenbeek, Belgum, Hensher, D. A. Stated Preference Analyss of Travel Choces: The State of Practce. Transportaton, Vol. 21, No. 2, 1994, pp

9 38 Transportaton Research Record Favre D Arcer, B., O. Andan, and C. Raux. Stated Adaptaton Surveys and Choce Process: Some Methodologcal Issues. Transportaton, Vol. 25, No. 2, 1998, pp Wardman, M. A Comparson of Revealed Preference and Stated Preference Models of Travel Behavour. Journal of Transport Economcs and Polcy, Vol. 22, No. 1, 1998, pp Louvere, J. J., D. A. Hensher, and J. D. Swat. Stated Choce Methods: Analyss and Applcaton. Cambrdge Unversty Press, New York, Couper, M. P., A. Kapteyn, M. Schonlau, and J. Wnter. Noncoverage and Nonresponse n an Internet Survey. Socal Scence Research, Vol. 36, No. 1, 2007, pp Hayslett, M. M., and B. M. Wldemuth. Pxels or Pencls? The Relatve Effectveness of Web-Based Versus Paper Surveys. Lbrary and Informaton Scence Research, Vol. 26, No. 1, 2004, pp Hensher, D. A. How Do Respondents Process Stated Choce Experments? Attrbute Consderaton Under Varyng Informaton Load. Journal of Appled Econometrcs, Vol. 21, 2006, pp Frank, K. A. Impact of a Confoundng Varable on a Regresson Coeffcent. Socologcal Methods Research, Vol. 29, No. 2, 2000, pp Scherer, M. Is Lght Ral More Attractve to Users Than Bus Transt? Arguments Based on Cognton and Ratonal Choce. In Transportaton Research Record: Journal of the Transportaton Research Board, No. 2144, Transportaton Research Board of the Natonal Academes, Washngton, D.C., 2010, pp De Wtte, A., C. Machars, P. Lannoy, C. Polan, T. Steenberghen, and S. Van de Walle. The Impact of Free Publc Transport: The Case of Brussels. Transportaton Research Part A, Vol. 40, No. 8, 2006, pp Cools, M., K. Brjs, H. Tormans, E. Moons, D. Janssens, and G. Wets. The Soco-Cogntve Lnks Between Road Prcng Acceptablty and Changes n Travel-Behavor. Transportaton Research Part A, Vol. 45, No. 8, 2011, pp Dorbrtz, R., M. Lüth, U. A. Wedmann, and A. Nash. Effects of Onboard Tcket Sales on Publc Transport Relablty. In Transportaton Research Record: Journal of the Transportaton Research Board, No. 2110, Transportaton Research Board of the Natonal Academes, Washngton, D.C., 2009, pp The Lght Ral Transt Commttee peer-revewed ths paper.