A Demand Analysis of an Ecological Public Transport Line in a Suburban Area of a Medium-Sized

A demand analysis of an ecological public transport line in a suburban area of a medium-sized town

M. Catalani 1, F. Russo **

aTerritorial Systems Department, Parthenope University of Naples, Via Medina 40, 80122 Naples, Italy bTransport Engineering Department, Federico II University of Naples, Via Claudio 21, 80122 Naples, Italy

______Abstract The objective of this paper is to give a method of analysis the effect of transport network improvement in small and medium towns .The problem can be difficult for Italy because many urban contest have not interconnected LPT or efficient transport planning system. Which approaches for modelling situations where public transport alternative is absent or irrelevant? The proposed methodology based on different logit models with a latent variable (city attitude) will simulate the residents behaviour on a sub urban corridor . By a specific SP survey on a user sample it will be calibrated a modal choice simulation of an ecological transport line.

Keywords: Modelling; Demand; Corridor ; Town; IT

1. Introduction 1992, Certu 1998). The resources allocated to funding public transport in The European urban context lends Italy, handled by the local administrative itself to various interpretations as far as regions (regioni) using national funds, its internal structure is concerned. The have reduced in real terms over recent European city rotates around a central years. The whole transport demand for area where the public buildings, the period from 2000 to 2003 in the churches, monuments, and areas for medium-sized towns increased from commerce and trade are concentrated. 26.9% to 29.1% in the big city segment Not being fixed to a centre, the (up to 250,000 inhabitants). Overall, a American city’s development produces modest willingness to make greater use low density expansion favoured by the of public transport has emerged. common use of the private car as the Analysing the causes influencing the main means of transport. From the point choice of the private car over public view of mobility, development on the means, it is clear that the main problems two continents can be said to be similar. of public transport is not so much, or not Italy, like other European cities, has a only, a matter of qualitative aspects high motorisation rate that also exists (Hillen B. and Pharoah T. 2002). They even in the big cities, where the can be the quality of service itself as the concentration and range of public speed and journey times, stop transport services is greater (Camagni R. accessibility, distribution and frequency

1Corresponding author. Tel. ++390815475314 Fax ++390815474314 **PHD in progress. E-mail address: [email protected] 2 of the routes (Senn L. and Percoco M mixed loose-knit network, with a few 2003). The main aim of this paper is to fast roads connecting to the main study how to evaluate the effects of national road system (Low N., and possible improvements to the public Gleeson B., 2003). A few high speed transport network in medium and small roads transversal connections leading to towns. A key element in this context is the industrial and port areas will form a the transport demand knowledge. great axis roads system .The paper will Which instruments should be used focus on a specific corridor connecting modelling realities where the public centre of Lecce city with sub-urban area transport alternative is inadequate? The of about 9 km distance. In this context, method proposed focuses on the the connection is by road with car and assessment of demand , based on the with two bus transit lines: the first is theory of the preferences of the traveller between peripheral zones and the second (Domencich T. A. e McFadden D. is a connector to the city's main roads, 1975),. The objective is to determine with a elevated number of stops (14’ for usefulness in a different choice context, the part between towns, 10’ for the through a more suitable calibration and urban part with a resident population validation of the parameters in question. outside Lecce of 21,455 inhabitants. The By a SP survey on a sample of corridor has the features of an individuals, different logit model will integration of the three areas, given the simulate the effects of a reserved lane gradual union of their population centres along a suburban corridor ( Oum T., ( Figure 2). 1979). The area chosen is a municipality The population density along the of Lecce a medium town in southern corridor stretch, its form protending Italy, with a population of around towards the south-east, as well as 200,000 inhabitants. significant population growth in the municipality, lead us to conclude that it 2. Urbanisation and corridor design may be chosen to assess the effects of the first application of the town planning More than a conurbation the urban concept of “the environmental corridor”. area develops along a radial directrix It is one of the more recent examples of determining same agglomeration effects urban replanning involving the smaller of disperse satellite zones( Figure 1). municipalities of the province. The Their limited accessibility slows down corridor has such strong characteristics the economic development process that of a sustainable model of development can be noted in larger centres. The supported by a suitable dedicated district has a resident population of infrastructure (Boyer K.D., 1977) . The 221,960 inhabitants, with a population aim of the present study is to envisage 2 density of 338 inhabitants per km , the creation of anefficient public remarkably higher than that of the transport system there and to assess the 2 province as a whole (286 inhab/Km ). impact on the mobility demand among The whole area has the widely dispersed the various forms of transport, currently morphological characteristics which dominated by private means. The tend to favour the use of private means service could equally well be carried out of transport. Conurbation growth could by any of these: thus take place through a strengthening  A traditional bus service but of the public transportation network using urban methane powered connections in order to capture latent vehicles to reduce the impact of demand thresholds (Chiffi 2005). This the pollution on the atmosphere; structure should be developed as a 3

 A new tram service , on tracks town and parking, environmental powered by overhead on conditions, the image of the city, protected lanes bearing in mind travel habits, the perception of the town centre's needs. public transport; Alternatively, a tram system but  Transport data revealed on rubber tyres, such as the Iris preferences as mode of transport bus Civis, Translohr or used; purpose and frequency of Bombardier TVR vehicles. travel; attributes of service The more convenient choice can be chosen - including parking costs made by applying a discrete choice (car only), the two ways model using a database suitably obtained journeys; through interviews (a survey) carried out  Transport data stated in the area involved. preferences as comparative answers (games) based on 3. The investigation into journeys and different attributes as times survey design journey costs (parking travel by car and only the travel/ticket cost The model presented the results of for public transport), agent effect an investigation (survey) carried out and comfort. This last is a especially in the area of interest on a psychometric variable and sample of 111 individuals. The aim was includes a set of attributes as air to obtain RP/SP survey. The RP conditioning, spaciousness, (revealed preferences) information on seating, radio, visual controlling transport choices actually preferred by etc. those interviewed and SP (stated The questionnaire was first tested through a series of interviews preferences) on preferences concerning administered to persons of diverse age, hypothetical choice contexts, where the economic conditions and travel habits; public option is included, which in our 12 games were proposed to each case is the “alternative” of the project. respondent and a total of 1,332 answers The questionnaire also contained were obtained (repeated choices). subjective judgements on aspects able to The test was useful for assessing the latently influence the respondents’ order of the questions and for observing choice of the means of transport. These the possible presence of dominant are attributes that concern the town alternatives in the SP games. The directly and not only the single means respondents were chosen at random from among the residents in the area of of transport under discussion, with the study. The investigation was carried out aim of showing the likelihood of the over two working days and two holidays town to accept metropolitan kinds of in the month of May 2005. Over the transport systems. holidays we interviewed the people who The following information was gathered normally move about by car/motorbike in detail: for study/work purposes along the  Socio-economic data as age, sex, corridor. As for the public transport number of family members, users (town buses), the interviews were members over 14 years, carried out on board the buses occupation, income, driving themselves. They were interviewed licence, possession of a vehicle; face-to-face at the entrance/exits at the  Psychometric data as comfort of bus-stops relevant to the area under the journey, accessibility of the consideration and the different parts of 4 the questionnaire were explained, and survey, a subset of scenarios consisting support was given as to how to answer, of only 12 different combinations without this influencing their choices. administered to each respondent was created. The car and the bus have very 4. City attitude market research similar journey time values. In part this is due to the fact that the two means of The database obtained from the transport share the same roads and that investigation was used for calibration of the bus does not have its own protected a choice model, in which alongside the route. The traffic and the number of traditional logit model a latent variable stops along the route, determines for the was included, called City Attitude, bus on centre/outskirts run at from 20 to expressing the judgements supplied 30 minutes. The case of the tram regarding traffic conditions, impact on however will involve less waiting time the environment, the image of the town, and shorter journey times due to the convenience of travel and the travel privileged routes. The dedicated lane habits of each respondent. In order to could extend along the whole route avoid the so-called “tiredness effect” or covered by the tram service, or could be errors such as spelling mistakes, and so partial, with a few crossings (reduced to preserve the same meaningfulness of section, cross-roads, proximity to the the data, we adopted a compact form centre ). with not only the essential information, This consideration is not been but also subjective opinions on a few underestimated in a scenario where there indicators considered able to express the is still a strong preference for the private latent aspects of city attitude. To car due to not taking into consideration facilitate comprehension of the journeys the costs inherent in travelling by car and the levels concerning public (insurance, road tax, wear and tear, transport (bus and tram), synoptic tables maintenance, etc). In effect the scenarios were used to illustrate the different proposed are more similar to what the forms of vehicles and the differences in respondents are able to perceive going the levels of comfort, and the rules of by their experience, and so the circulation. private/public comparison can come out Even though there are two means of more favourable attributing realistic public transport within the context of levels “for the individual” to the games choice (bus and tram), in the survey we (Table 1). intended to attribute very different meanings to these alternatives. The bus 5. Behavioural approach as MNL, alternative represents a traditional MIXED LOGIT and HIERAR- automobile service, carried out with CHICAL LOGIT models. standard 12-metre vehicles, running alongside private traffic, with longer Random utility theory is based on the journey times due to the more tortuous hypothesis that every individual is a itineraries and the number of stops. The rational decision maker ,maximising tram, on the other hand, will be carried utility relative to his/her choices. It is out in a totally protected environment not usually possible to predict with lane, with shorter running times, not certainty the alternative that the generic influenced by the traffic conditions decision-maker will select. However, it found on the normal roads. is possible to express the probability of The final games model was obtained selecting alternative j conditional on from the choice of a subset of scenarios his/her choice set I i , as the probability from Complete Factorial Plane. For the that the perceived utility of alternative j 5 is greater than that of all the other the maximum among those relating to available alternatives (McFadden D. all the alternatives available. In fact, in 1975 and Cascetta E 2001) : the hypotheses made, the probability of

i i  i i i  choosing alternative j from among those p j / I  Pr obU j U k k  j, k  I  ( 1)   available belonging to the set of choices i (1,2,……….m) I can be expressed in The perceived utility U j can be closed form as: expressed by the sum of the systematic exp(V / ) i j uitility V j ,which represents the mean p j  m (5) or the expected value of the utilities exp(Vi / ) perceived by all the decision-makers i1 having the same choice contest as Expression (5) defines the Multinomial decision- makers i and a random Logit model which is one of the most i used random utility models. residual  j which is the (unknown) The other model proposed is the single- deviation of the utility perceived by the level hierarchical logit model. This user I from this value: model allows to overcame partially the U i  V i   i j  I i (2) assumption of independent random j j j residuals underlying the Multinomial The Multinomial Logit (MNL) is based Logit model. At the same time it retains on the hypothesis that the random a closed analytical expression. Suppose residuals εj relating to the different that the decision –makers choice set I alternatives are independently and is subdivided into non-overlapping identically distributed (i.i.d.) according subsets I , I ,…, I ,… so that the utility to a Gumbel random variable (r.v.) of 1 2 k zero mean and parameter θ. The function of the generic alternative j marginal probability distribution ,belonging to the subset I k can be3 function of each random residual is expressed as : given by (McFadden D. 2000 and Cascetta E. 2001) : U j  V j   j  V j  j  j / k j  I k k (6) F (x)  Pr  x  exp exp x     j j (3) E  E   E  0 (7) Where ф is the eulero constant. In j k j / k particular, mean and variance of the Cov  ,  Cov ,    k h   k j  Gumbel variable expressed by (3)  h   Cov ,   0  j i   k k  E j  0 j (4)

2 The expression of the overall 2  2 choice Var j       j 6 pro bability of the generic alternative Cov j , h   0 j,h  I ( 5) p j is obtained as the product of the p j  Stability compared with maximisation probability  k  of choosing means that the Gumbel variable is a elementary alternative j within hypothesis particularly convenient for group k (lower level), multiplied by the distribution of the uncertain residue the probability pk of choosing in the models just described, as these express the probability of choice of an group k (upper level).The probability alternative to the probability that the structure is : utility perceived for that alternative is 6

p j  p j / k pk (8) where g (.) is the multivariate normal density. This unconditional probability, Bypassing overall demonstration we will called the mixed logit choice have : probability, is the product of logits mixed over a density of parameters expV j   expY  pj p j / k pk   k (Train K, 2003). For Bayesian analysis, expV   expY  (9)  iI k i  h i h it is necessary to specify the prior on the that define choice probability of model parameters. The prior on b is alternative j . specified as normal with sufficiently The last application is one utilizes the large variance that it is effectively flat mixed logit ( kernel logit) model. The from a numerical perspective. The mixed logit model is well defined in the density is denoted IW (  K,KI ). literature. This is one of the most With these priors, the joint posterior on important complete models developed is by Daniel McFadden (1996) and  nn ,b,  successively by Train and McFadden, (n ,n ,b,Y )(L( yn n g( b,)IW ( K, KI ) (13) (2000), Ben Akiva and Wolker (2002). n Generally the Bayesian procedure The code elaborated by Train (Train establishes (Bhat C.2000, and Revelt D. K,2003) considers a transformation of and Train K. 1998) that, under certain normals, defined parameters of conditions the mean of the Bayesian passengers n as cn, which is a vector posterior is a classical estimator with the same length n ; successively asymptotically equivalent to the the parameters defined by cn, = T( n ), maximum likelihood estimator (Train K, where T is a transformation that depends 1998). Bayesian procedure considers only on  and is weakly monotonic passengers choices among three n k different alternatives in t periods c (such that n  0 for each element k, (Allenby G, 1997). The perceived utility k  n from each alternative is (Train K,  -Sonnier G, 2003): of cn, and n ). ' The distribution of cn, is determined by U   n x   njt njt njt (10) the transformation. Normally distributed where  njt  iid extreme value and n n are drawn as before but then

 N (b, ). The vectors of variables transformed to cn, when enter utility.

xnjt and parameters n . Conditional on Utility is specified as: U njt  T ( n )' xnjt   njt (14) n the probability sequence of choices is the product of standard logit formulas 6. The calibration of the models (Train K, 2003):  ' x t e 'n nynt L y   The structure of the model is represented n n   ' x t 'n nynt (11) t  e by the classical utility functions that j considers the attributes present in the SP The unconditional probability is the survey and the presence of the latent integral of L yn n over all values of variable expressing the City Attitude.  weighted by the density of  : The hierarchical model include also n n attributes for branch choice equation as Pn yn b,)  private and public mode ( labelled pri, (12) pub).For the estimation of the  L(y  )g  b, d  n n n parameters of MNL and Hierarchical Logit model we used the Nlogit 3.0 7

Econometric Software Inc. which evidences a clear preference for tram supplied the following results (Table 2). 52%, 28% for bus and 20 % for car. This The statistic indicators evidences a good result should be read in the light of the fit of the models. The maximum territorial context in question. Parking simulated likelihood evidences a better costs, traffic congestion and modest value of Hierarchical logit as regards the travelling time reduction to cover a others (Agha , M. and Branker, D.S. relatively short journey don’t modify 1997). individuals perceived preferred The mixed logit application by Train alternative. The quota attracted to the code (Apithec Corporation, Gauss 8.0) tram, based on “stated” preferences and evidences versatility of the application thus not effective, is perhaps ( Train K and Sonnier G 2003) also if overestimated. The same reasons prevail overall models present statistics al good for the prediction of the bus share. standard level of fitness. Individuals are not willing to change A synthesis of calibrations we learn that: their habits if a “weak” service from the  hierarchical model seems to have performance point of view is proposed, a better performance especially in terms of comfort and travel  mixed logit reports a better value times, even though it is cheaper than of latent variable coefficients using one's own vehicle. But when the  the value of simulated Log characteristics of the service are Likelihood for hierarchical “strong”, as in the case of a public model is acceptable (-1015.182) service on a reserved lane, with precise  the signs of the cost and time timetabling and times of arrival at coefficient are usual. destination, as well as high impact with The relationships between the regard to the image and comfort of the coefficients of the level of service vehicles( air conditioned tram), the attributes are also meaningful, and are probability that a most significant share also known as reciprocal substitution “might pass over” to public transport is relationships. Among these, the ones remarkably higher. This substantial relating to monetary cost are particularly difference, found also in other important (willingness to pay). The assessments made at a European level, relationship between the journey time changes the profile of the demand coefficients and the journey cost estimates potentially satisfied by the coefficients supplies the perceived value new transport infrastructure. time as expected (vot – value of time), that generally must have a positive 7. Conclusions value. Furthermore the research Results of calibration evidence a net confirming the pre-eminence of preference for public transport. The commuter journeys connected with design of new line with a dedicated lane work(Watson P.L. 1974). This implies could count on a substantial potential that reductions in journey time are daily demand sufficient to justify its important from the point of view of creation (850 passengers per hour per demand, with benefits for the means of direction of travel). This value should transport able to cover the journey in the influence choices concerning the shortest time and cost. The prediction frequency of the new service and ability of the model is fairly good (the passenger capacity of the vehicles to be model predicted correctly 884 cases out chosen. The transit system adopted must of 1,332) with around 66% of the be regulated in order to satisfy demand estimated choices corresponding to the in the peak hours to avoid overcrowding reality. The calculated probability 8 on board vehicles. It needs to use BOYER K.D., (1977) Minimum rate regulation, vehicles holding at least 140 persons modal split sensitivities and the railroad 2 problem, Journal of Political Economy, 85 (seated + standing with 4 people/m ) BROWNSTONE, D. AND K. TRAIN (1999), fully satisfy demand in peak hours. ‘Forecasting new product penetration with These reasons and demand model split flexible substitution patterns’, Journal of results, lead us to identify the tram as the Econometrics 89, 109-129. best vehicle for this service. The CAMAGNI R. (1992), Le reti di città: verso una philosophy of using shorter, but more teorizzazione e una tassonomia in ROSINI R (1992) (a cura di), L’urbanistica delle frequent vehicles has been adopted by a aree metropolitane, Alinea, Firenze, pp. 87- lot of public transport companies in 114 Europe but we disagreement. Even in CAMAGNI R. (1997) (a cura di), La the large cities, even if only on some pianificazione sostenibile delle aree routes, the length of the single vehicles perturbane, Il Mulino, Bologna CASCETTA E., (2001) Transportation system can exceed 20 metres in length. engineering :theory and methods. Kluver The time-saving for the individuals Academic Publishers, The Netherlands will consent us to quantify the public ,pp.101-106,137-151,192-197. interest to new line and expressed it in CASELLA, G. AND E. GEORGE (1992), financial terms. To this advantage we ‘Explaining the gibbs sampler’, The American Statistician 46, 167-174. should add the positive effects on the CATALANI M. (2004), Analisi dei sistemi di environment of the reduction of the trasporto in ambiti territoriali complessi, number of cars running along the Giappichelli, Torino connection, in addition to the economic CERTU (1998), Méthodes d’observation des saving compared with the low use of effets sur l’urbanisme et le cadre de vie, Certu, Lyon private means (fuel, parking, etc.). At CERTU (2003), Les Transport Publics Urbains last the service will guarantees greater en France – Organisation institutionnelle, sustainability of the demand flow, Certu, Lyon increasing attractiveness for external CHIFFI A (2005) Sustainable transport on residential area decongesting urban medium town. Dissertation thesis Univ. Parthenepe traffic. DOMENCICH T. A. e McFADDEN D. (1975), Urban Travel Demand: a behavioral Acknowledgment analisys, North-Holland PC, Amsterdam - Oxford Many thanks to professor R. HILEN B. e PHAROAH T. (2002), Making Jayakrishnan for useful counsels. Tracks – Light rail in England and France, Swedish National Road and Transport References Research Institute, Linköping Sweden LOW N., and GLEESON B., (2003), Making Urban Transport Sustainable, Palgrave AGHA, M. AND BRANKER, D.S. (1997) Macmillian, Basingstoke Hampshire – UK Maximum likelihood estimation and MC FADDEN, D. (1973) Conditional logit goodness of it tests for mixtures of analysis of quantitative choice behavior. distribution, Journal of the Royal Statistical Front in econometrics. Academic Press Society, volume 46, n. 3, London. New York ALLEMBY .G.(1997), ‘An introduction to MCFADDEN D., (2000), “Disaggregate hierarchical Bayesian modelling’, Tutorial Behavioral Travel Demand’s RUM Side’’ – notes, Advanced Research Techniques A 30 Year Retrospective, International Forum, American Marketing Association. Association for Travel Behavior (IATBR), BEN AKIVA, M., LERMAN S. (1985). Discrete Gold Coast, Queensland, Australia, 2 – 7 choice analysis. The MIT Press July 2000 Massachusetts Institute of Technology MCFADDEN D., TRAIN K.,(2000) Mixed Cambridge, Massachusetts 02142. NML models for discrete response. Journal BHAT, C. (2000), ‘Incorporating observed and of Applied Econometrics, John Wiley & unobserved heterogeneity in urban work Sons, New York mode choice modelling’, Transportation Science 34, 228-238. 9

MCFADDEN, D. (1975) ‘’Aggregate travel demand forecasting from disaggregated behavioral models’’ Transportation Research Record: Travel behavior and values, No 354,24-37, OUM T., (1979) A Warning on the use of linear logit models in transport mode choice studies, Bell Journal of Economics 10, pp 374-378. REVELT, D. AND TRAIN (1998), ‘Mixed logit with repeated choice’, Review of Economics and Statistics 80, 647-657. SENN L. and PERCOCO M. (a cura di) (2003), Trasporti e sostenibilità ambientale: analisi economica dei rapporti tra infrastrutture, mobilità e ambiente, EGEA, Milano TRAIN K G .SONNIER (2003) Mixed logit with bounded distribution of parthworths. University of California, Bertkeley 2003 [email protected] TRAIN, K (2001) A comparison of hierarchical Bayes and maximum simulated likelihood for mixed logit.Working paper, Depatment of Economics; University of California, Berkley, TRAIN, K (2003) Discrete Choices Methods with Simulation. Cambridge university Press, New York. TRAIN, K. (1998), ‘Recreation demand models with taste variation’, Land Economics 74, 230-239. BEN AKIVA AND WALKER J.L. (2002), Generalised Random Utility Model. Mathematical Social sciences 43 MIT, Boston WATSON P.L., (1974) The Value of Time: Behavioral Models of Modal Choice, D.C. Heath & Co., 10

Table. 1 General framework of variables and factor utilised variables Generic Description Factors Levels (nr.3) name

Quantitativ Twalk Timing to get to stop, for bus and TwB, TwT e tram Bus 1, 2, 3 minutes variables Tram 3, 4, 5 minutes

Ttravel Travel time for public transport TtA, TtB, including waiting time at the stop. TtT Car 12, 18, 25 min Bus 12, 18, 30 min Tram 10, 15, 20 min

Ctravel CtA, CTB,

Travel cost also including parking CtT Car € 5.40 – 4.50 – and car running costs 3.60

Bus € 1.40 – 1.00 – 0.60 Tram € 1.95 – 1.50 – 1.05

Qualitative Comfort CMA, Car high, medium, variables Comfort for cars includes CMB, low amenities such as radio, air CMT Bus high, medium, conditioning, size of car, driver low comfort, etc- for public transport Tram high, medium, such as seating, air conditioning on low board, information about next stop, vehicle size, visual system of passengers, control on board bus.

Agent AE Agent effect as percentage AEA, Effect variation due to the cost of travel AEB, AET Car +/- 20% per mode of transport Bus +/- 40% Tram +/- 30%

Table 2 Multinomial ,mixed and hierarchical logit models estimates MNL Mixed logit Hierarchical Logit

Variable β1 Coefficients T-Statistics Coefficients T-statistics Coefficients T-statistics Car spec. constant 0.859 0.847 0.146 2.357 -2.321 -1.749 Tram spec. constant 0.571 5.429 0.681 9.841 - - Bus spec. constant - - - - -1.422 2.183 COST -0.863 -5.738 0.936 11.18 -1.974 -9.158 TIME -0.124 -13.09 -0.157 -2.521 -0.456 -10.35 LATENT -0.086 -3.915 0.158 3.655 1.412 1.849 COMFORT -0.459 -4.235 0.658 10.57 - 0.324 2.169 INCOME 0.673 5.322 0.463 7.160 - - AGENT 1.619 6.383 - 0.851 -10.01 1.487 1.389 AGE -1.437 - 2.956 -0.527 -5.512 -0.175 -0.418 HINCC - - - - 1.129 4.751 HINCB - - - - -0.671 -2.456 PRIV - - - - 0.531 7.817 PUB 0.193 1.511 Simulated Log-Likelihood - 1471.3248 -1588.3949 -1015.182 12

Figure 1

Fig. 1. Urban town system :zoning and boundary limits Source: Lecce Urban Traffic Plan. 13

Figure 2

Fig. 2. Study area with main routes and corridor identification Source: Lecce Urban Traffic Plan