Lp

Aa

Ub

a

ARRA

KTLMMSS

I

hidttKznru2

phile2

(

0h

Land Use Policy 30 (2013) 458– 469

Contents lists available at SciVerse ScienceDirect

Land Use Policy

jou rn al h om epa ge: www.elsev ier .com/ locate / landusepol

and use and transit ridership connections: Implications for state-levellanning agencies

rnab Chakrabortya,∗, Sabyasachee Mishrab

Department of Urban and Regional Planning, University of Illinois at Urbana-Champaign, 611 E. Lorado Taft Drive, M230 Temple Buell Hall (MC-619), Champaign, IL 61821,nited StatesNational Center for Smart Growth Research and Education, 054 Preinkert Hall (1112J), University of Maryland, College Park, MD 20742, United States

r t i c l e i n f o

rticle history:eceived 9 September 2011eceived in revised form 18 April 2012ccepted 19 April 2012

eywords:

a b s t r a c t

In this article we attempt to establish the connections between transit ridership and land use and socio-economic variables, and project future ridership under different scenarios. We subdivided the state ofMaryland, USA into 1151 Statewide Modeling Zones and developed a set of variables for the base year(2000). We estimated multiple models of transit ridership – using ordinary least squares and spatial errormodeling approaches – for the entire state. We also test for the determinants of ridership within urban,

ransit ridershipand useodelaryland

tatewidecenarios

suburban and rural typologies. We find that land use type, transit accessibility, income, and density arestrongly significant and robust predictors of transit ridership for the statewide and urban areas datasets.We also find that the determinants and their coefficients vary across urban, suburban and rural areas. Nextwe used a suite of econometric, land use and other models to generate two sets of future transit ridershipscenarios under conditions of – (a) business as usual and (b) high energy price – for a 30-year horizon.We analyze these scenarios to demonstrate the value of our approach for state-level decision-making.

© 2012 Elsevier Ltd. All rights reserved.

ntroduction

Planning for land use and transit in the United States oftenappens independent of one another. While land use planning

s primarily done at the local level, the scale of transit planningepends on factors such as, the spatial extent of served areas,ype and number of operating modes, and financing structure ofhe agency (Cervero and Landis, 1997; Badoe and Miller, 2000;itamura et al., 1997). This disconnect is evident in cases where, say,oning restricts high-density development near transit stations. Aumber of studies implicate this lack of coordination to be in partesponsible for the American auto-oriented landscapes and manynderused transit systems (Volinski and Page, 2006; Shaheen et al.,009).

Despite the potential to influence local decisions, larger scalelanning agencies, such as state departments of transportation,ave historically made few systematic efforts in harnessing the

nterdependencies between land use and transit. Consequently, the

iterature on frameworks of such coordination under existing gov-rnance structures remains underdeveloped (Garrett and Taylor,003).

∗ Corresponding author. Tel.: +1 2172448728; fax: +1 2172441717.E-mail addresses: arnab@illinois.edu (A. Chakraborty), mishra@umd.edu

S. Mishra).

264-8377/$ – see front matter © 2012 Elsevier Ltd. All rights reserved.ttp://dx.doi.org/10.1016/j.landusepol.2012.04.017

Transit has been argued to be a catalyst to refocus devel-opments in dense, mixed-use, and mixed-income communities(Badoe and Miller, 2000; Boarnet and Crane, 2001; Cervero, 1996;Cervero and Landis, 1997; Kitamura et al., 1997). According tothese studies, proliferation of such communities can be associatedwith lower consumption of natural resources, residential and trans-portation energy savings, reduction in government expendituresin infrastructure and service provisions, and better resiliency touncertainties in future energy prices. At the same time, higher den-sities and accessibility to transit is associated with better transitservices and making higher ridership more viable (Tong and Wong,1997; Messenger and Ewing, 1996; Moudon et al., 1997; Levineand Inam, 2004; Levine et al., 2005; Boarnet and Greenwald, 2000).Accordingly, researchers and practitioners have attempted to iden-tify factors that encourage and sustain higher densities and transituse, such as design principles for new subdivisions, accessibilityto stations and regional urban form, all factors influenced by localland use policy (Ewing and Cervero, 2001; Krizek, 2003; Miller et al.,1999; Heath et al., 2006). Advocates have promoted the incorpo-ration of these ideas in urban plans and ordinances, and in transitsiting decisions.

Many of these principles have been adopted at certain scales

or in a piecemeal fashion. For example, land use change modelsconsider transit services in determining development potential andattractiveness of land (Deal, 2001; Landis, 1994, 1995; Landis andZhang, 1998; Waddell, 2002; Deal and Schunk, 2004), which could

and Us

tdeau

tfmacaBmFrisirsvscdlbbts

rictr(rpbweaSrtrsra

gztois

fTecip

A. Chakraborty, S. Mishra / L

hen be taken into account when planning for, say, transit-orientedevelopments or municipal zoning change (Cervero, 1994; Cerverot al., 1993; Taylor et al., 2004). Similarly, land use characteristicsre routinely included in travel demand models that are sometimessed by transit planners.1

While the above approaches serve specific short-term, opera-ional purposes, they have several planning limitations. Designed tourther local objectives or resolve narrowly defined local concerns,

ost analytical approaches take factors that are beyond a localgency’s control as givens. For example, land use patterns are oftenonsidered as exogenous inputs in a travel demand model (Boarnetnd Crane, 2001; Boarnet and Sarmiento, 1998; Kockelman, 1997).y failing to consider the implications of variations in these factorsore carefully, such approaches forego potentially richer analysis.

urther, analyses that are limited to local scales fail to consideregional implications of local decisions and, by extension, theirnteraction with broader uncertainties. Land use patterns and tran-it provisions often have spatial, financial and other spillovers andt is the responsibility of larger scale agencies to balance negativeegional externalities. For example, multiple transit agencies in atate may be looking to fund expansions of their own systems for aariety of reasons. Each may advocate in its own interest for stateupport. They might even compete for funds with yet other agen-ies, e.g. a municipality looking to support high-density land useevelopment. However, a state agency with influence over both

and use and transit ought to evaluate the overall outcome on itsroader jurisdiction. The higher-level agency can harness possi-le interdependencies in making its own decisions by looking forrade-offs without regard to local interests and biases. Such analy-is, however, needs a suitable analytical framework.

In this article, we develop such a framework. Using transitidership as the key measure, we test how different outcomesn different futures may present a state agency with specifichoices regarding planning. We chose the state of Maryland inhe United States as our study area. Using a number of crite-ia, we subdivide the state into 1151 statewide modeling zonesSMZs) and, for each SMZ for the base year (2000), estimate aange of variables, including developed land under different uses,opulation and employment densities, developed land densitiesy industry category, auto-ownership, household income density,orkers per household, free-flow and congested speeds of the

xisting transportation infrastructure, current transport capacities,nd accessibility to different transport modes. Using the statewideMZ dataset, we estimated ordinary least square and spatial erroregression models for the base year data. We also model the rela-ionships on subsets of SMZs representing urban, suburban andural typologies. We find that characteristics of land use, tran-it accessibility, income, and density are strongly significant andobust predictors of transit ridership for the statewide and urbanreas datasets.

We then used a suite of econometric and land use models toenerate two sets of development outcomes for a planning hori-on year (2030) – one under ‘business as usual’ conditions andhe other under ‘high-energy prices’. We use these conditions and

ur key ridership model to generate two distinct sets of scenar-os for future transit ridership. The scenarios are constructed toeparate the choices over which the decision-makers can exercise

1 The most common of these is the Four-step Travel Demand Model (TDM). Usedor decades to determine both highway and transit demand, developing reliableDMs can be expensive and time-consuming and require extensive computationalffort. Modeling transit ridership component is particularly complex, as it requiresreating a virtual transit network, conducting ridership surveys, and incorporat-ng routes, stops, headways, and fare-matrix returns. As a result, only the transitlanning agencies that have considerable resources use these practices.

e Policy 30 (2013) 458– 469 459

some control, from uncertainties or regional forces that are beyondtheir control. Drawing upon their differences, we discuss how suchanalysis can inform decision-making.

We proceed as follows. In the next section, we establish the con-nections between transit ridership and land use through a reviewof modeling practices to derive and frame the key planning ques-tions. In the following section, we discuss the datasets, the rationalebehind the choice of our study area, and the modeling frameworkfor our empirical analysis; and in the next two sections, we presentfindings of this analysis and apply our model to develop scenariosfor the horizon year and discuss implication for state level deci-sions, respectively. We offer concluding remarks in the final sectionfollowed by caveats and future scope of research.

Existing research on transit ridership modeling anddecision making

Ridership is a commonly used measure to capture the effect ofclustered development, diversity, density, transit supply, systemefficiency, and surrounding land uses on transit use. Studies aboundthat attempt to model ridership on a variety of factors (Kockelman,1997; Boarnet and Crane, 2001; Du and Mulley, 2007; Lin and Gau,2006). However, for specific considerations on how ridership varieswith land use and what that means for state level choices, we orga-nize the available literature in the following streams: (1) land useconsiderations in transit models; (2) the lessons from, and limita-tion of such models in large-scale decision making; and (3) ongoingpractices in large-scale decision-making.

Ridership estimation models are frequently studied in publictransit and have been reviewed multiple times (see, for example,Kain and Liu, 1999; Abdel-Aty, 2001; Wang and Skinner, 1984;Horowitz, 1984; Taylor et al., 2004; Ben-Akiva and Morikawa,2002). Not surprisingly, these studies are framed for transit agencyrelated questions and purposes. Taylor et al. (2009) group ridershipdetermination factors into two categories from a transit agency per-spective: external and internal. External factors include population,economic conditions, auto ownership levels, and urban density; allfactors over which agency managers have no control. Internal fac-tors, in contrast, allow transit agency managers to exercise somecontrol. They include the amount of service the agency provides,the reliability of service, service amenities, and fare. Taylor et al.(2009) show that understanding the influence of these factors isimportant to transportation system investments, pricing, timing,and deployment of transit services.

Studies about the influence of external factors on ridershiphave employed a variety of methodological approaches, includingcase studies, interviews, surveys, statistical analyses of character-istics of a transit district or region, and cross-sectional statisticalanalyses. These studies find that transit ridership varies depend-ing upon a number of factors, such as (i) regional geography (e.g.total population, population density, total employment, employ-ment density, geographic land area, and regional location) (Ongand Blumenberg, 1998; Kuby et al., 2004; Hsiao et al., 1997;Wu and Murray, 2005; Zhao et al., 1997; Polzin et al., 2002;Peng and Dueker, 1995), (ii) metropolitan economy (e.g. medianhousehold income, income distribution) (Ingram, 1998; Cohn andCanada, 1999; Frisken, 1991; Thompson and Brown, 2006; Fujiiand Hartshorni, 1995; Yoh et al., 2003; Hirsch et al., 2000; Kyteet al., 1988; Cervero et al., 1993), (iii) population characteristics (e.g.percent of captive and choice riders, or household with zero cars)(Cohn and Canada, 1999; Polzin et al., 2000; Ewing, 2008; Davies,

1976), and (iv) auto/highway system characteristics (specificallynon-transit/non-single occupancy vehicle trips, including commut-ing via carpools) (Cervero, 2007; Lisco, 1968; Holtzclaw et al., 1994;Taylor and Fink, 2002; Gómez-Ibánez and Fauth, 1980). They also

4 and Us

cwtietre2

Sftdttb(dMaeowstt

dmtbTithsia

aeeslvtipAatf

ptsi

aueat(e

fic sheds of major roads, and employment activity centers, anda frequent grouping of adjacent TAZs, where they existed. SMZsalso delineate areas with good accessibility to Metro rail stations

60 A. Chakraborty, S. Mishra / L

onfirm that availability of public transit is strongly correlatedith the urbanization of the area. Overall, the relative impor-

ance of external factors, the interaction between them, and theirmpact on the transit ridership continues to be debated in the lit-rature. To address this, a number of studies focus on examininghe relative causal influences of internal and external factors onidership. They generally find external factors more important inxplaining ridership change (Kain and Liu, 1999; Taylor and Fink,002).

A number of studies, including some listed above (Pucher, 1980;mith and Gihring, 2006; Pickrell, 1989; Puentes, 2000), note thatew states base transit financing decisions on land use or other long-erm considerations, or employ ridership modeling. Instead, theirecisions are often driven by operational considerations and main-aining the existing fleet system. A number of past research showshat some states have mechanisms to allocate resources for a num-er of rehabilitation, remanufacturing, and replacement optionsKhasnabis et al., 2007; Mishra et al., 2010; Mathew et al., 2010), buto not focus upon modeling future transit demand. For example,ishra et al. (2010) discuss the resource allocation among 93 transit

gencies in Michigan in a multi-year planning period, but only mod-led a constant transit fleet system, which is insensitive to changesver time. Land uses and intensities however, change over time andill result in varying transit demand. In sum, the aforementioned

tudies do not provide adequate mechanisms for estimating futureransit demand that can augment decision making for statewideransit planning agencies.

The state of Maryland can be considered as another example toemonstrate the need for a statewide transit ridership estimationodel. The state of Maryland is adjacent to Washington, DC, and

he Commonwealth of Virginia and parts of Maryland are servedy regional transit service such as the Washington Metropolitanransit Authority (WMATA). In addition, a number of counties andndependent cities operate their own bus and feeder transit sys-em. Currently, the state level financing of these myriad agenciesappen without assessment of current and future interregionalpillovers and shifting demands, nor is there any framework todentify decisions that can help achieve state level parity in servicesnd expenditure.

As many of the above studies note, common modelingpproaches have several limitations. First, the use of transit rid-rship estimation models are geographically limited as they areither too cumbersome to build and operate, or unavailable out-ide of large metropolitan areas, or are overlooked. They are alsoimited to one or two transit modes at a time and consider other ser-ices and variation in critical factors such as land use as exogenouso the model. Second, the internal/external separation of factorsn the literature is usually framed from a regional transit agencyerspective and is not directly translatable to, say, a state agency.

number of factors, such as urban densities, that are external to transit agency may indeed be within the sphere of influence ofhe state. Conversely, specific transit agency choices such a servicerequency and fares are not.

This suggests that to address transit ridership questions from theerspective of higher-level agencies, planners should – (1) considerransit interdependencies with a broader range of transportationervices and regional urban form characteristics; and (2) reframenternal/external factors for the specific decision-making entity.

Firstly, the regional land use and transportation system inter-ctions have been studied extensively in the regional planning andrban modeling literature. It cites commuting costs, housing andmployment locations, housing prices, institutional considerations,

nd history of investment in transportation systems among factorshat have reinforcing effects on transit accessibility and ridershipMurray, 2003; Murray and Wu, 2003; Litman, 2004; Beimbornt al., 2003). Lacking institutional frameworks for regional planning

e Policy 30 (2013) 458– 469

in US, the policy efficacy studies are fewer in US (e.g. Portland, OR)and draw heavily upon European experiences in integrated spatialplanning (Badoe and Miller, 2000; Boarnet and Greenwald, 2000;Kain and Liu, 1999; Cervero and Kockelman, 1997; Newman andKenworthy, 1996). This literature has also illuminated the ben-efits of transit in providing energy use and air quality benefits,and in promoting more equitable urban form in the future underhigher energy price fluctuations. This is an important motivationfor our work and in our study area, which includes a high varia-tion in transit services and densities. As we discuss later in detail,such considerations can help fine tune policies to specific spatialtypologies.

Secondly, internal/external framing of factors is important andapply to state level analysis as well, albeit in an adjusted for-mat. Internal/external frame has been used often in the modelingliterature to separate controllable “choices” (internal) from uncon-trollable “forces” (external) or uncertainties (Chakraborty et al.,2011). It allows modelers to generate scenarios that reflect dif-ferent combinations of choices and uncertainties. The scenariosreflect alternative futures, not as a result of choices alone but howthey interact with external conditions. Comparing scenarios canthen identify decisions that are (1) robust under most likely setof external conditions, and/or (2) useful for unlikely yet impor-tant contingencies, or (3) resilient over a range of uncertainties.As the internal and external factors can be adjusted to suit specificdecision-makers, it can help organize the institutional complexitiesin a region with layers of overlapping jurisdictions and influences.As we explain later in this paper, considering land use as an inter-nal factor from a state agency perspective can be useful in makingdecisions.

Developing a statewide ridership model

Data

Our objective was to develop a transit ridership model for theentire state of Maryland and demonstrate its usefulness in state-level decision-making, especially a more integrated land use andtransit planning. In this section we focus on the first part usingdatasets for the year 2000.2 The state of Maryland consists of 23counties and one independent city and had a total population of5.8 million and a total employment of 3.4 million in the year 2010(BEA, 2011; Census, 2010). It also has seventeen types of publictransportation systems including metro rail, commuter rail, localbus and long distance buses. To develop our dataset, we subdi-vided the state into 1151 Statewide Modeling Zones (SMZs). TheSMZ development went through an iterative process including sev-eral reviews by the State Highway Administration and was part ofa larger modeling project. We identified the broader study areausing 2000 Census Transportation Planning Package (CTPP) data toencompass the bulk of labor flows in and out of Maryland. Withinthis larger boundary, six regions were identified for SMZ formation.The outline of the state and the broader region with its sub-regionalcomponents is shown in Fig. 1. The transit ridership model devel-oped here is restricted to the SMZs within Maryland.

The main criteria for SMZ delineation included, conforming tocensus geographies and nesting within Counties, separating traf-

2 This remains the latest time period for which this data is contiguously and con-sistently available. Some beta releases are coming out for 2007 data but are notexpected to be available for our entire region in the near future. We also believethat this does not affect the important purposes of our analysis.

A. Chakraborty, S. Mishra / Land Use Policy 30 (2013) 458– 469 461

elop sS

adtatdc

thsu

htahhS

2rMsdahswso

files, and Maryland Department of Transportation (MDOT) datasetsare used to determine the total freeway distance, average freeflow speed, average congested speed, and presence of a bus stop.

Fig. 1. Regions used to devource: Maryland Statewide Transportation Model, State Highway Administration.

nd, to the extent possible, distinguish rural from urban/suburbanevelopment and zoning boundaries. The variations in land use pat-erns across the state are characterized in Fig. 2. Both householdnd employment density maps show the concentrated growth inhe central portion of the state, while there is relatively less denseevelopment in counties east of the Chesapeake Bay and in westernounties of Maryland.

To account for variations in relationship between land use pat-erns and ridership across the state, we used a combination ofousehold and employment densities to classify SMZs under threepatial typologies – urban, suburban, and rural. The defined valuesnder each category are presented in Table 1.

The classification ranges were based on composite measure ofousehold and employment densities, approach consistent withhe Maryland State Highway’s long range planning. For example, anrea is classified Rural when the household density is less than 0.15ouseholds/acre and employment density is less than 0.20 house-olds/acre. Table 2 shows the classification map and the number ofMZ units under each typology.

Next we built an extensive dataset at the SMZ level for year000. Our key dependent variable, transit ridership, is based onidership data provided to the State Highway Administration byPOs and other local agencies. Though data by individual transit

ervices are available, we chose to combine them to create a totalaily transit ridership value for each SMZ that includes all boardingnd alighting. In our dataset, there are 1151 SMZs, 900 of whichave non-zero transit ridership, and 240 of which have a transit

top (access point). This means that there are a large number of SMZhere transit riders originate though they do not access a transit

ervice in that SMZ. Transit ridership data is based on the point ofrigin of a trip; for example, a home-based park-and-ride trip will

tatewide modeling zones.

be counted in transit ridership value of the SMZ where the riderresides, not where s/he board a transit service.

The accessibility to transit ridership is a dummy variable is basedon whether an SMZ has a transit access point. These include lightrail and heavy rail stations, and bus stops. Fig. 3 below shows rid-ership distribution in the central Maryland region and locationsof transit stops classified by bus, local rail (Metro), regional rail(MARC), and long-distance rail (AMTRAK). Bus stops from through-out the state are included in the analysis, including regional servicessuch Washington Metropolitan Transit Authority (WMTA), Mary-land Transit Authority (MTA), Montgomery County’s Ride On, andPrince George County’s The Bus. In addition, stops from 18 separatebus services provided by several cities and counties, and bus stopsfrom major universities such as University of Maryland CollegePark, John Hopkins University are also included.

We develop a number of land use and network characteristicsfrom a variety of datasets and through GIS analysis. MPO regionalemployment data is used for the MPO regions and Quarterly CensusEmployment and Wages (QCEW3) data is used for employment inthe non-MPO covered areas. The MPO and QCEW data are aggre-gated to determine the employment by categories such as retail,office, industrial, and other. Household income data is collectedfrom MPOs and Census for the MPO and non-MPO region, respec-tively. The transportation network datasets from Census TIGER

3 QCEW (formerly known as ES202) is an employment data source prepared bythe Department of Labor, Licensing and Regulations (DLLR).

462 A. Chakraborty, S. Mishra / Land Use Policy 30 (2013) 458– 469

nsity

Micwss

E

bssln

TD

ac

Fig. 2. Household and employment de

aryland Department of Planning’s (MDP) Property View datasets used to determine specific square footages land used for healthare, housing, shopping, industry, office, recreation, dining, andarehouse, and other commercial establishments. The descriptive

tatistics for the key datasets discussed in the empirical analysisection below are presented in Table 3.

mpirical analysis

We regressed the daily transit ridership in an SMZ with a num-er of explanatory variables using two approaches: ordinary least

quares (OLS) and spatial error model (SEM). OLS is a strong thoughimple linear modeling approach that we apply here and draw uponater in the paper to demonstrate our proof-of principle in sce-ario analysis and decision-making application. SEM is a stronger

able 1istribution of densities (in households/acre) by spatial typology.

Employment density ≤0.20

Household density ≤0.15 Rural

Household density >0.15 and ≤2.0 Rural

Household density >2.0a Suburban

a While net urban densities tend to be higher, we chose 2 HH/Acres as our cut-off. Ths a denominator in both household and employment density calculations. Also, our visuharacterization.

ranges (in households/acre) for SMZs.

approach that is adopted when there is a high likelihood of spa-tial autocorrelation. Together, these models provide comparisonsacross different specifications, tests for robustness, and allow us tocapture the importance of spatial interactions and transit interde-pendencies.

Table 4 presents the results for a number of alternative specifi-cations for the statewide dataset. Models I and II are based on theentire state’s SMZs where as Models I-A, I-B and II-A and II-B includeonly those SMZs that have non-zero transit ridership. Also, the spa-tial error models include a contiguity weight factor (Lambda) thataccount for the effects due to the characteristics of surrounding

SMZs. This matrix is separately estimated for the statewide and thenon-zero transit ridership datasets.

Overall, the results follow a priori expectations and are robustacross specifications. Among the OLS models, Model I, based on

Employment density >0.20 and ≤3.0 Employment density >3.0

Rural SuburbanSuburban UrbanUrban Urban

is is because we estimate gross densities that include the whole area of the SMZal inspection of aerial imagery under SMZs with different typologies confirms this

A. Chakraborty, S. Mishra / Land Use Policy 30 (2013) 458– 469 463

Table 2Area types and counts for typology subsets.

Typology SMZ count

Rural 414

Swlut

SuburbanUrban

Total

MZs for the whole state, shows that transit ridership increases

ith household and employment density, is higher in areas with

ower income and lower car ownership. This is consistent withrban economic theory and confirms findings from past studieshat were previously limited to metro areas.

Fig. 3. Transit access points and transit ridership

312425

1151

More specifically, the results of Model I reflect the fact that

majority of employment is located in the urbanized areas andis concentrated around transportation networks. Also, locationdecisions for siting employment centers often take transit intoconsideration and vice versa. Transit ridership increases with

distribution in the central Maryland region.

464 A. Chakraborty, S. Mishra / Land Use Policy 30 (2013) 458– 469

Table 3Descriptive statistics.

Variables Mean S.D. Min Max

Daily transit ridership 5612.48 9032.29 0 62012.50Household density 1246.44 2698.96 0 36331.25Population density 3289.24 7009.19 0 98837.50Household workers density 1498.58 3025.23 0 39700.00Total employment density 1398.50 3594.03 0 46770.00Retail employment density 246.55 1567.29 0 50340.00Office employment density 759.27 2009.43 0 24980.30Industrial employment density 133.93 412.72 0 6792.00Other employment density 618.59 1756.10 0 35680.00Drive alone density 6542.70 5108.36 0 57233.51Household with 0 carsa 193.35 360.70 0 3027.00Income less than 20,000 268.52 377.32 0 3550.00Income between 20,000 and 40,000 362.36 384.75 0 2813.00Income between 40,000 and 60,000 336.82 313.93 0 2975.00Income between 60,000 and 100,000 441.22 394.25 0 3680.00Income over 100,000 312.06 336.81 0 2587.00Total school enrollment 706.08 988.00 0 7132.00Total free way distance 7.05 10.90 0 91.80Average free flow speed 42.92 8.86 10 70.00Accessibility to transit (dummy variable) 0.19 – 0 1.00Dinning square feet 2431.37 10305.17 0 197750.00Healthcare square feet 8521.98 71774.91 0 1990004.00Housing square feet 26719.70 146540.25 0 2758645.00Industry square feet 2637.49 28409.03 0 706565.00Office square feet 20086.47 112786.73 0 2279896.00Recreation square feet 2866.03 14314.23 0 246000.00Shopping square feet 20237.54 83697.81 0 1157052.00Warehouse square feet 9833.56 55494.20 0 862137.00

a e areao e can

p

dmu

vuhWagimsotra

tsiefrs

nOsii

(ns

We deliberately chose this over normalized values. One factor in transit servicr driving distance in a suburban/rural area. Given that and the fact more distancroblematic.

ecreasing auto-ownership and decreases with amount of freewaysiles in an SMZ and drive alone density, or number of drivers per

nit of land area in the SMZ, both consistent with expectations.The effect of a number of subcategories of land uses in Model I

iz. healthcare, housing, and recreation are also significant, thoughnderstandably smaller in magnitude. For example, presence ofealthcare centers is negatively correlated with transit ridership.hile this may reflect greater accessibility by emergency vehicles

nd personal automobiles, a good thing in the event of an emer-ency, the lower ridership may also reflect lower service suggestingnequities in service to those without automobile for routine treat-

ents and visiting patients. The other variables show expectedigns as ridership increases with increases in total square footagef housing and decreases with recreation square footage. Overall,he results (and R-square) from Model I show that SMZ level transitidership model for the entire state is viable and can explain a largemount of variation in ridership across a number of transit systems.

The OLS models for SMZs with non-zero transit ridership only,hough generally consistent, show some differences from thetatewide model. For example, while the household density andncome coefficients show a higher magnitude, the magnitude ofmployment density coefficient is lower. These relationships areurther strengthened when the transit stop accessibility variable isemoved, though the overall explanatory power of the model fallslightly.

The spatial error models for the statewide SMZ dataset and theon-zero ridership subset show very similar relationships with theLS model outputs, though the coefficients of most variables get

maller. The weight matrix variable (Lambda) is strongly signif-cant in all these models and reflects the importance of spatialnteractions and transit interdependencies.

We find that the standard OLS approach reported in Table 4Model I) and its variations – re-estimating the relationships foron-zero ridership SMZs and comparing OLS model outputs withpatial error model – add to the depth of our analysis. While there

determination is households within 15 min of walking distance in an urban areabe covered in suburban areas than urban in 15 min, areas-based normalization is

are some differences in the magnitude of coefficients, general scaleof the effect, directionality of the relationships and significance ofthe variables, are all consistent across these models. In the nextsection, we chose one of these models – Model I – for the pur-pose of developing transit ridership in two future scenarios. Thisis discussed later in more detail, though we should note here thatbased on the robustness of our key variables across different mod-els and the proof-of-principle nature of our state planning analysisdemonstration, we find it appropriate to use the coefficients of thestatewide OLS model.

We also ran another set of OLS models where one model is testedfor each typology subset (Models III, IV and V, for SMZs classified asUrban, Suburban and Rural, respectively). The results are presentedin Table 5.

The Models III, IV and V attempts to look at typology level asso-ciations for urban, suburban and rural areas, respectively (Table 5).The directionality and magnitudes of the effects of significant vari-ables in these models are generally consistent with the findingsof Model I. To synthesize the results, the following can be said:the constant is positive and significant for all the models, but itsdecreasing magnitude from Model III to V reflects the commonlyknown lower overall ridership difference between urban to ruralareas. Lower income and higher transit accessibilities are positivelycorrelated and strongly significant across all models. Health carerelated developments remain negatively correlated with ridershipand its effect increases in rural areas – lending credence to the logicthat access to such services is even more difficult for householdswithout automobiles in rural areas.

There are also some key differences across these models. Whilehousehold and employment densities are both positively correlatedwith ridership, household density is not a significant determinant of

ridership in suburban areas (Model IV) and employment density isnot a significant determinant in rural areas (Model V). The SMZ withhigher school enrollment is expected to produce more transit trips.The square footage for different land use types is differently related

A. Chakraborty, S. Mishra / Land Use Policy 30 (2013) 458– 469 465

Table 4Regression results for alternative specifications using the statewide SMZ dataset.

Independent variables Ordinary least squares Spatial error model

Model I Model I-A+ Model I-B+ Model II Model II-A+ Model II-B+

Constant 631.4** 4835.13*** 10094.94*** 1421.74*** 4300.65*** 4483.74***

(2.73) (4.22) (11.19) (3.01) (3.59) (3.99)Household density 1.480*** 2.15*** 2.06*** 0.33 0.79** 0.66**

(4.40) (6.07) (5.70) (1.172) (2.22) (1.85)Employment density 0.7892*** 0.45*** 0.44*** 0.31*** 0.27*** 0.26**

(9.68) (6.59) (6.22) (4.51) (4.16) (4.03)Drive alone density −1.835*** −0.78*** −0.76*** −1.02*** −0.30** −0.26**

(−10.53) (−6.53) (−6.26) (−6.87) (−2.46) (−2.08)Household without cars 14.62*** 11.72*** 12.04*** 10.896*** 10.62*** 10.34***

(19.01) (14.28) (14.43)) (14.01) (12.12) (11.78)Household workers density 3.4188*** 2.38***

(8.14) (6.92)Income less than 60,000 1.793*** 3.16*** 3.24*** 2.35*** 3.15*** 3.34***

(7.21) (10.55) (11.36) (9.78) (10.90) (12.20)Number of school enrollment 0.115

(0.64) (0.65)Total freeway distance −73.07*** −75.71** −15.19 13.54

(−5.57) (−2.28) (−1.02) (0.44)Average free flow speed −91.50*** −203.69*** −80.84*** −74.78***

(−3.44) (−9.96) (−3.04) (−3.21)Accessibility to transit stop (0,1) 5325***

3150.13*** 185.384 1488.41***

(14.58) (7.48) (0.407) (2.94)Health care square feet −0.0046* −0.006*** −0.006*** −0.005*** −0.006*** −0.006***

(−2.38) (−3.03) (−2.82) (−3.43) (−3.71) (−3.76)Housing square feet 0.0042*** 0.003*** 0.004*** 0.001 0.003*** 0.003***

(4.35) (3.21) (3.60) (1.54) (3.23) (3.19)Recreation square feet −0.0362*** −0.038*** −0.035*** −0.032*** −0.03*** −0.03***

(−3.66) (−3.54) (−3.31) (−4.24) (−3.21) (−2.98)Lambda 0.751 0.673 0.713

(32.92) (13.55) (15.46)Sample size (n) 1151 900 900 1151 900 900R2 0.7383 0.7574 0.7407 0.8266 0.794 0.794Adjusted R2 0.7358 0.7536 0.7378

Dependent variable: total daily ridership; T-statistics (in OLS) and Z-values (in SEM) are in parentheses. +Model I-A, I-B, II-A, II-B include only those SMZ that have non-zerotransit ridership.

* Significant at 90%.

tsiedestmerar

talpd

P

ppt2

** Significant at 95%.*** Significant at 99%.

o ridership. While beyond the scope of this analysis, this could betudied in greater detail. Also, a number of variables strongly signif-cant in Model I lose their significance in subset-based models. Forxample, household density, employment density and drive aloneensity are not significant in Model III. A closer look at the data mayxplain why. The relationships may not be clear due to the relativeimilarity of urban form and transit services among SMZs within aypology, or lower variance among the explanatory variables. This

ay also explain why the coefficient of determination (R2) is high-st for Model III and least for Model V. The lower magnitude ofidership might be one of the reasons of lower (R2) for Model IVnd Model V. On the contrary, Model III has the highest (R2) as theidership for urban area is the highest among all.

Irrespective of these differences, however, our analysis confirmshe following: (1) overall, land use and other neighborhood char-cteristics are useful predictors of transit ridership at a statewideevel and; (2) the variation in relationships by subarea typologiesresent a useful framework for fine-tuning policies and investmentecisions.

lanning application: horizon year ridership scenarios

Having developed a model for statewide transit ridership, we

resent a general framework for applying it in decision-making,articularly at large scales by agencies such as state DOTs. To dohis we first develop two sets of input variables for the horizon year030. Then we use Model I from the previous section to generate

two transit ridership scenarios. We use this as a stylized case forassessing state level decision choices.

To illustrate our application, we drew upon the work of Mary-land Scenario Project (MSP), a large-scale visioning exercise led bythe National Center for Smart Growth (NCSG) at the University ofMaryland. For more on MSP, please refer to the past publications(Chakraborty, 2010, 2011). The growth principles of the scenarioswere developed by the Scenario Advisory Group, an MSP-affiliatedgroup of nearly 40 land use and transportation planning profession-als. The Group identified a number of important yet uncertain setsof conditions that may affect development of the region. The mostrelevant among these for our purposes were growth rates of energyprices and federal expenditures. Building on these, we character-ized one set of year 2030 conditions as (a) Business as Usual (BAU),where the past relationships between sectors, investment patterns,demographics, etc., continue unimpeded; and the other set of vari-ables as (b) High Energy Prices (HEP), an alteration of energy priceswhere some past historical relationships will continue, and changesin energy prices reverberate throughout the economy and changeeconomic, land use, and transportation outcomes. Under HEP, realcrude oil prices rise faster than BAU at 1 percent above the projectedinflation rate. In BAU, oil prices roughly follow the Energy Informa-tion Administration’s short-term projections. In addition, three key

parameters are considered in the HEP scenario: (1) increase in fed-eral defense spending, (2) increase in employment in professionalservice, and (3) increase in agriculture commodity price. Thesefactors were selected by a scenario advisory committee and the

466 A. Chakraborty, S. Mishra / Land Use Policy 30 (2013) 458– 469

Table 5OLS models for each typology.

Independent variables Model III (urban) Model IV (suburban) Model V (rural)

Constant 6647*** 1537.931*** 212.198***

(2.907) (5.583) (5.045)Household density 1.073 0.209*

(1.364) (1.899)Employment density 0.192 0.344***

(1.375) (5.437)Drive alone density −0.329 −0.139*** −0.036

(−1.128) (−3.869) (−1.298)Household without cars 11.72***

(9.849)Income less than 60,000 2.675*** 0.848***

(5.846) (3.364)Number of school enrollment 0.361** 0.078*

(2.129) (1.96)Total freeway distance −52.893**

(−2.496)Average free flow speed −106.8** −4.124**

(−2.215) (−2.121)Accessibility to transit stop (0,1) 4824*** 3117.371*** 828.929***

(5.255) (7.094) (6.955)Health care square feet −0.007** −0.027**

(−2.526) (−2.019)Housing square feet 0.003**

(2.445)Recreation square feet −0.059***

(−4.117)Dinning square feet −0.037 −0.03

(−1.188) (−1.595)R2 0.699 0.267 0.171Adjusted R2 0.692 0.248 0.158Dependent variable: total daily ridership; T-statistics are in parenthesis

* Significant at 90%.

rttnw

tetvTlmfr(

emnatTcleg2wras

** Significant at 95%.*** Significant at 99%.

ationale was to identify exogenous trends that would provide clus-ered urban development, more employment and housing close toransit stations, lesser development on green infrastructure, lesserew impervious surfaces, and lesser vehicle miles travelled (VMT)ithout major changes in government policy.

Next we leveraged a set of models developed for MSP to opera-ionalize these conditions into specific variables for each SMZ underach scenario. The national economic model is based on the Long-erm Interindustry Forecasting Tool (LIFT), forecasts outcomes ofarious macroeconomic policies and changes such as energy prices.he outcomes of this model, in turn, influence the regional andocal demographic model, which determines households’ employ-

ent in various county and SMZ-level sectors. Since this paper isocused on the adequacy and accuracy of these models, we refer theeader to model documentation and published material elsewhereChakraborty et al., 2011; Mishra et al., 2011).

The aggregate socioeconomic data obtained from the MPO coop-rative forecasts was converted using econometric and land useodels into two sets of values under BAU and HEP scenarios for a

umber of variables such as, income less than $60,000, householdnd employment densities. The network related data was used fromhe 2030 network developed at the NCSG and the Maryland Stateransportation Model. These are leveraged to generate two sets ofonditions for average free flow speed and drive alone density. Theand use data such as the square feet under different land uses arestimated by extrapolating the trends in different land use cate-ories from last five years (using MD Property View data) into year030. Finally, we assume that other variables such as average free-

ay distance, school enrollment, and some other land use variables

emain same as these data is not available from public and privategencies in the state and it is very difficult to predict for a HEPcenarios. This assumption is reasonable because these variables’

contributions are lower compared to other more significant vari-ables. From the two sets variables representing 2030 conditions –BAU and HEP – we then develop two ridership scenarios using coef-ficients from Model I in Table 4. We also estimated a summary mapof the difference between them as depicted in Fig. 4.

The map compares total ridership under each scenario in 2030.Dark gray-colored SMZs are those that have high ridership irrespec-tive of the changes in external conditions. This is expected as, beingurbanized areas, they already had high ridership and high transitservices in year 2000. The gray areas are those that have consider-ably higher ridership in the BAU scenario (than HEP) and hatchedareas are those that have considerably higher ridership in the HEPscenario (than BAU). This reflects a key outcome of explicitly con-sidering different sets of external conditions. For example, sincehigh-energy prices make the inner SMZs more attractive to devel-opment (due to many reasons, including lower commute times,and proximity to multiple employment scenarios), which is thenreflected in higher growth in these areas, leading to greater demandfor transit ridership. In the case of BAU-scenarios, where energyprices increase at a lower rate, the trend of higher development inexurban areas lead to more growth away from the urban centersand an increase in transit ridership demands in those areas.

These findings have several implications. For example, our anal-ysis shows that simple assumptions in a future-oriented analysiscan lead to significantly different outcomes and, as a result, theyshould be carefully considered and analyzed. While a large num-ber of SMZs will continue to require transit services under bothscenarios, a number of them will require additional services only

under HEP or under BAU scenario. How this information is usedin decision-making will depend on the agency and the decision inquestion. For example, a transit agency overseeing a SMZ that mayhave high ridership demand in one scenario but little in another,

A. Chakraborty, S. Mishra / Land Use Policy 30 (2013) 458– 469 467

en HE

mimap

Fig. 4. Change betwe

ight want to track the likelihood of external conditions (since

t cannot directly influence them), and make any new invest-

ent decisions only if there is a high likelihood of HEP. A stategency, however, might use the same information for differenturposes.

Fig. 5. Transit ridership in Maryland coun

P and BAU scenarios.

Fig. 5 shows, as expected, statewide transit ridership is higher in

the high-energy price scenario. Further, it shows that some countieswill receive a higher share of this growth than other. Such differ-ences may play a role in state level decisions, including land usepolicies and future transit subsidies. For example, promoting new

ties under BAU and HEP scenarios.

4 and Us

dbiksiaef

wbidcct

C

loStcn

aIiptgiwiaf

prtlcsavfi

issasaei

maMtfib

68 A. Chakraborty, S. Mishra / L

evelopment in Baltimore City or Prince George’s County seems toe one way to encourage transit ridership. Also, if steep increase

n energy prices becomes a likely scenario, it might be useful tonow that it might have a spatially varied impact and can informtate financing of new projects. Furthermore, if a state level agencys interested in (and capable of) coordinating urban developmentnd transit investment it may look at development patterns and rid-rship across counties, projected trends in development and otheractors in making land use and transit related policies.

In an actual planning situation, of course, more future scenariosould have to be formulated and tested, and the final decision will

e the outcome of a political process. This framework, nevertheless,s important as it lends itself to both collaborative and independentecision-making process. In a collaborative process, a larger set ofonditions will be “internal” to the planning agencies whereas inase of independent decision-making, agencies will need to identifyheir internal choices and external givens or uncertainties.

onclusions and discussion

Frequent under-utilization of transit networks combined withack of transit services in many areas points to a mismatch in devel-pment preferences and limitation of public transportation modes.uch disconnect raises economic and equity concerns that are fur-her heightened under steep increases in energy prices and housingosts. The challenges are exacerbated by fragmented nature of plan-ing agencies across sectors and scales.

Research has long shown that efficacies of public transitnd high-density land use developments are interdependent.ncreasing sprawl, residential segregation, and income inequal-ty, decreasing share of transit use and uncertainties in gasolinerices make it imperative that planning agencies take advantage ofhese interdependencies. However, the literature provides limiteduidance on modeling transit use at a large scale, thereby limit-ng the potential for coordinated land use and transit planning. As

e have discussed, this may be due to several reasons including,nstitutional barriers to agency functions, models that take limiteddvantage of the notion of uncertainty, or simply a lack of data andrameworks for analyzing the future.

In this paper, we show that a higher-level agency can harnessossible interdependencies in making its own decisions withoutegard to local interests and biases. To do this, we developed aransit ridership model for the whole State of Maryland that usesand use and other neighborhood level variables. We found thatharacteristics of land use, transit accessibility, income, and den-ity are strongly significant and robust for the statewide and urbanreas datasets. We also find that determinants and their coefficientsary across urban, suburban and rural areas suggesting the need fornely tuned policy.

Development of travel demand models can be expensive, requir-ng extensive data collection, and many states does not havetatewide travel demand models. In the absence of functional four-tep travel demand model to predict transit ridership, planninggencies often need to have an alternate measure of determiningtrategies for investment in transit. This framework could be usefuls a way to inform service provision decisions in such places and tonhance the use of transit in rural regions by incorporating changesn land use characteristics.

DOTs seldom succeed at conducting planning for the statewideultimodal network; typically transit is missing, added separately

t key stages based on planning by transit agencies or sometimes

POs, but seldom with the ability to reflect transit’s statewide con-

ribution and potential. This paper takes an important step towardlling that void. Further, in using a stylized case of two scenarios –usiness as usual and high energy prices – we demonstrated how

e Policy 30 (2013) 458– 469

such analysis could lead to multiple choices that a state level agencycan consider in making its decisions. Estimating transit ridershipunder multiple scenarios shows how demand could vary by partsof the state and demonstrates the model’s value in assessing tran-sit and land use planning decisions. Given the State Plan and strongcounty planning in Maryland, this paper describes a useful support-ive analysis and modeling effort that can likely be incorporated inMaryland statewide planning activities and project analyses (andelsewhere).

We, however, acknowledge that there are several limitationsto this study. While our statewide and subarea models are robustthey are based on several estimated variables, many of which couldbe fine-tuned with additional resources. The treatment of differ-ent transit modes separately may also affect our results. Finally,as we noted earlier, the scenario analysis is highly stylized and ismeant for the purpose of demonstrating the framework and is notintended to recommend policy. That being said, we believe thatthere are unique opportunities in considering state level questionsas they not only consider interdependencies but also non-urbanregions in the analysis and decision-making choices for higher lev-els of governments.

References

Abdel-Aty, M.A., 2001. Using ordered probit modeling to study the effect of ATIS ontransit ridership. Transportation Research Part C: Emerging Technologies 9 (4),265–277.

Badoe, D.A., Miller, E.J., 2000. Transportation-land-use interaction: empirical find-ings in North America, and their implications for modeling. TransportationResearch Part D: Transport and Environment 5 (4), 235–263.

BEA, 2011. Bureau of Economic Analysis: Bureau of Economic Analysis:SA25N. TotalFull-time and Part-time Employment by NAICS Industry. http://www.bea.gov/(accessed May 2011).

Beimborn, E.A., Greenwald, M.J., Jin, X., 2003. Accessibility, connectivity, and cap-tivity: impacts on transit choice. Transportation Research Record: Journal of theTransportation Research Board 1835 (-1), 1–9.

Ben-Akiva, M., Morikawa, T., 2002. Comparing ridership attraction of rail and bus.Transport Policy 9 (2), 107–116.

Boarnet, M.G., Greenwald, M.J., 2000. Land use, urban design, and nonwork travel:reproducing other urban areas’ empirical test results in Portland, Oregon. Trans-portation Research Record: Journal of the Transportation Research Board 1722(-1), 27–37.

Boarnet, M.G., Sarmiento, S., 1998. Can land-use policy really affect travel behaviour?A study of the link between non-work travel and land-use characteristics. UrbanStudies 35 (7), 1155.

Boarnet, M., Crane, R., 2001. The influence of land use on travel behavior: speci-fication and estimation strategies. Transportation Research Part A: Policy andPractice 35 (9), 823–845.

Census, 2010. Census Summary Files. http://www.census.gov/ (accessed, May 2011).Cervero, R., 1994. Transit-based housing in California: evidence on ridership impacts.

Transport Policy 1 (3), 174–183.Cervero, R., 1996. Mixed land-uses and commuting: evidence from the American

Housing Survey. Transportation Research Part A: Policy and Practice 30 (5),361–377.

Cervero, R., 2007. Transit-oriented development’s ridership bonus: a product of self-selection and public policies. Environment and Planning A 39 (9), 2068–2085.

Cervero, R., Kockelman, K., 1997. Travel demand and the 3Ds: density, diversity,and design. Transportation Research Part D: Transport and Environment 2 (3),199–219.

Cervero, R., Landis, J., 1997. Twenty years of the Bay Area Rapid Transit System:land use and development impacts. Transportation Research Part A: Policy andPractice 31 (4), 309–333.

Cervero, R., Carroll, M., Munkres, J., Ketelsen, J., 1993. Ridership Impacts of Transit-Focused Development in California. University of California at Berkeley, Instituteof Urban and Regional Development.

Chakraborty, A., Kaza, N., Knaap, G.J., Deal, B., 2011. Robust plans and contingentplans. Journal of the American Planning Association 99999 (1), 1-1.

Chakraborty, A., 2010. Scenario planning for effective regional governance:promises and limitations. State and Local Government Review 42 (2), 156–167,http://dx.doi.org/10.1177/0160323X10377344.

Chakraborty, A., 2011. Enhancing the role of participatory scenario planning pro-cesses: lessons from reality check exercises. Futures 43 (May (4)), 387–399,http://dx.doi.org/10.1016/j.futures.2011.01.004.

Cohn, H.M., Statistics Canada, 1999. Factors Affecting Urban Transit Ridership. Statis-

tics Canada.

Davies, G.W., 1976. The effect of a subway on the spatial distribution of population.Journal of Transport Economics and Policy, 126–136.

Deal, B., 2001. Ecological urban dynamics: the convergence of spatial modelling andsustainability. Building Research and Information 29 (5), 381–393.

and Us

D

D

E

E

F

F

G

G

H

H

H

H

H

I

K

K

K

K

K

K

K

L

L

L

L

L

L

LL

M

case studies of successful transit systems in the 1990. Transportation Research

A. Chakraborty, S. Mishra / L

eal, B., Schunk, D., 2004. Spatial dynamic modeling and urban land use transfor-mation: a simulation approach to assessing the costs of urban sprawl. EcologicalEconomics 51 (1–2), 79–95.

u, H., Mulley, C., 2007. The short-term land value impacts of urban rail transit:quantitative evidence from Sunderland, UK. Land Use Policy 24 (January (1)),223–233, http://dx.doi.org/10.1016/j.landusepol.2005.12.003.

wing, R., Cervero, R., 2001. Travel and the built environment: a synthesis. Trans-portation Research Record: Journal of the Transportation Research Board 1780(-1), 87–114.

wing, R.H., 2008. Characteristics, causes, and effects of sprawl: a literature review.Urban Ecology, 519–535.

risken, F., 1991. The contributions of metropolitan government to the success ofToronto’s public transit system. Urban Affairs Review 27 (2), 268.

ujii, T., Hartshorni, T.A., 1995. The changing metropolitan structure of Atlanta,Georgia: locations of functions and regional structure in a multinucleated urbanarea. Urban Geography 16 (8), 680–707.

arrett, M., Taylor, B., 2003. Reconsidering social equity in public transit. In: SocialResearch in Transport (SORT) Clearinghouse, p. 193.

ómez-Ibánez, J.A., Fauth, G.R., 1980. Downtown auto restraint policies: the costsand benefits for Boston. Journal of Transport Economics and Policy, 133–153.

eath, G., Brownson, W.R.C., Kruger, J., Miles, R., Powell, K.E., Ramsey, L.T., et al.,2006. The effectiveness of urban design and land use and transport policies andpractices to increase physical activity: a systematic review. Journal of PhysicalActivity & Health 3, 55.

irsch, L.R., Jordan, J.D., Hickey, R.L., Cravo, V., 2000. Effects of fare incentives onNew York City Transit ridership. Transportation Research Record: Journal of theTransportation Research Board 1735 (-1), 147–157.

oltzclaw, J., Natural Resources Defense Council, California Home Energy EfficiencyRating Systems, 1994. Using Residential Patterns and Transit to Decrease AutoDependence and Costs, vol. 11. Natural Resources Defense Council.

orowitz, A.J., 1984. Simplifications for single-route transit-ridership forecastingmodels. Transportation 12 (3), 261–275.

siao, S., Lu, J., Sterling, J., Weatherford, M., 1997. Use of geographic informationsystem for analysis of transit pedestrian access. Transportation Research Record:Journal of the Transportation Research Board 1604 (-1), 50–59.

ngram, G.K., 1998. Patterns of metropolitan development: what have we learned?Urban Studies 35 (7), 1019.

ain, J.F., Liu, Z., 1999. Secrets of success: assessing the large increases in transitridership achieved by Houston and San Diego transit providers. TransportationResearch Part A: Policy and Practice 33 (7–8), 601–624.

hasnabis, S., Mathew, T.V., Mishra, S., 2007. A single-stage mixed integer program-ming model for transit fleet resource allocation. In: Transportation ResearchBoard 86th Annual Meeting.

itamura, R., Mokhtarian, P.L., Laidet, L., 1997. A micro-analysis of land use andtravel in five neighborhoods in the San Francisco Bay Area. Transportation 24(2), 125–158.

ockelman, K.M., 1997. Travel behavior as function of accessibility, land use mixing,and land use balance: evidence from San Francisco Bay Area. Transporta-tion Research Record: Journal of the Transportation Research Board 1607 (-1),116–125.

rizek, K.J., 2003. Operationalizing neighborhood accessibility for land use-travelbehavior research and regional modeling. Journal of Planning Education andResearch 22 (3), 270.

uby, M., Barranda, A., Upchurch, C., 2004. Factors influencing light-rail stationboardings in the United States. Transportation Research Part A: Policy and Prac-tice 38 (3), 223–247.

yte, M., Stoner, J., Cryer, J., 1988. A time-series analysis of public transit ridership inPortland, Oregon, 1971–1982. Transportation Research Part A: General 22 (5),345–359.

andis, J.D., 1994. The California Urban Futures Model: a new generation ofmetropolitan simulation models. Environment and Planning B 21, 399.

andis, J.D., 1995. Imagining Land Use Futures: Applying the California FuturesModel.

andis, J., Zhang, M., 1998. The second generation of the California urban futuresmodel. Part 2: specification and calibration results of the land-use change sub-model. Environment and Planning B 25, 795–824.

evine, J., Inam, A., 2004. The market for transportation-land use integration: dodevelopers want smarter growth than regulations allow? Transportation 31 (4),409–427.

evine, J., Inam, A., Torng, G.W., 2005. A choice-based rationale for land use andtransportation alternatives. Journal of Planning Education and Research 24 (3),317.

in, J.J., Gau, C.C., 2006. A TOD planning model to review the regulation of allowabledevelopment densities around subway stations. Land Use Policy 23 (July (3)),353–360, http://dx.doi.org/10.1016/j.landusepol.2004.11.003.

isco, T.E., 1968. Value of Commuters Travel Time – A Study in Urban Transportation.

itman, T., 2004. Transit price elasticities and cross-elasticities. Journal of Public

Transportation, 37–58.athew, T.V., Khasnabis, S., Mishra, S., 2010. Optimal resource allocation among

transit agencies for fleet management. Transportation Research Part A: Policyand Practice 44 (6), 418–432.

e Policy 30 (2013) 458– 469 469

Messenger, T., Ewing, R., 1996. Transit-oriented development in the sun belt. Trans-portation Research Record: Journal of the Transportation Research Board 1552(-1), 145–153.

Miller, E., Kriger, J.D.S., Hunt, J.D., 1999. Integrated Urban Models for Simulationof Transit and Land Use Policies: Guidelines for Implementation and Use. 48.Transportation Research Board.

Mishra, S., Ye, X., Ducca, F., Knaap, G.J., 2011. A functional integrated land use-transportation model for analyzing transportation impacts in the Maryland –Washington, DC region. Journal of Sustainability Science Policy and Practice.

Mishra, S., Mathew, T.V., Khasnabis, S., et al., 2010. Single-stage integer programmingmodel for long-term transit fleet resource allocation. Journal of TransportationEngineering 136, 281.

Moudon, A.V., Hess, P.M., Snyder, M.C., Stanilov, K., 1997. Effects of site designon pedestrian travel in mixed-use, medium-density environments. Transporta-tion Research Record: Journal of the Transportation Research Board 1578 (-1),48–55.

Murray, A.T., 2003. A coverage model for improving public transit systemaccessibility and expanding access. Annals of Operations Research 123 (1),143–156.

Murray, A.T., Wu, X., 2003. Accessibility tradeoffs in public transit planning. Journalof Geographical Systems 5 (1), 93–107.

Newman, P.W.G., Kenworthy, J.R., 1996. The land use – transport connection: anoverview. Land Use Policy 13 (1), 1–22.

Ong, P., Blumenberg, E., 1998. Job access, commute and travel burden among welfarerecipients. Urban Studies 35 (1), 77.

Peng, Z., Dueker, K.J., 1995. Spatial data integration in route-level transit demandmodeling. Seattle City Light’s Electrical Usage 7 (1), 26.

Pickrell, D.H., 1989. Urban Rail Transit Projects: Forecast versus Actual Ridership andCosts. Final Report. Transportation Systems Center, Cambridge, MA (USA).

Polzin, S.E., Chu, X., Rey, J.R., 2000. Density and captivity in public transit success:observations from the 1995 nationwide personal transportation study. Trans-portation Research Record: Journal of the Transportation Research Board 1735(-1), 10–18.

Polzin, S.E., Pendyala, R.M., Navari, S., 2002. Development of time-of-day-based tran-sit accessibility analysis tool. Transportation Research Record: Journal of theTransportation Research Board 1799 (-1), 35–41.

Pucher, J., 1980. Transit financing trends in large US metropolitan areas: 1973–1978.Transportation Research Record, 759.

Puentes, R., 2000. Flexible Funding for Transit: Who Uses It? Brookings Institution,Center on Urban and Metropolitan Policy.

Shaheen, S., Allen, A.D., Liu, J., 2009. Public transit training: mechanism to increaseridership among older adults. In: Transportation Research Board 88th AnnualMeeting (CD-ROM), Transportation Research Board.

Smith, J.J., Gihring, T.A., 2006. Financing transit systems through value capture.American Journal of Economics and Sociology 65 (3), 751–786.

Taylor, B.D., Fink, C.N.Y., 2002. The Factors Influencing Transit Ridership: A Reviewand Analysis of the Ridership Literature. UCLA Institute of Transportation Stud-ies.

Taylor, B., Miller, D.D., Iseki, H., Fink, C., 2004. Analyzing the determinants of transitridership using a two-stage least squares regression on a national sample ofurbanized areas. In: 83rd Annual Meeting of the Transportation Research Board,Washington, DC.

Taylor, B., Miller, D.D., Iseki, H., Fink, C., 2009. Nature And/or Nurture? Analyzing theDeterminants of Transit Ridership Across US Urbanized Areas. TransportationResearch Part A: Policy and Practice 43 (1), 60–77.

Thompson, G.L., Brown, J.R., 2006. Explaining variation in transit ridership in USmetropolitan areas between 1990 and 2000: multivariate analysis. Transporta-tion Research Record: Journal of the Transportation Research Board 1986 (-1),172–181.

Tong, C.O., Wong, S.C., 1997. The advantages of a high density, mixed land use, linearurban development. Transportation 24 (3), 295–307.

Volinski, J., Page, O., 2006. Part 1: intermodal transfer facilities and ferry trans-portation: developing bus transfer facilities for maximum transit agency andcommunity benefit. Transportation Research Record: Journal of the Transporta-tion Research Board 1955 (-1), 2–7.

Waddell, P., 2002. UrbanSim: modeling urban development for land use, transporta-tion, and environmental planning. Journal of American Planning Association 68(3), 297–314.

Wang, G.H.K., Skinner, D., 1984. The impact of fare and gasoline price changes onmonthly transit ridership: empirical evidence from seven US transit authorities*1. Transportation Research Part B: Methodological 18 (1), 29–41.

Wu, C., Murray, A.T., 2005. Optimizing public transit quality and system access:the multiple-route, maximal covering/shortest-path problem. Environment andPlanning B: Planning and Design 32 (2), 163–178.

Yoh, A.C., Haas, P.J., Taylor, B.D., 2003. Understanding transit ridership growth:

Record: Journal of the Transportation Research Board 1835 (-1), 111–120.Zhao, F., Wang, L., Elbadrawi, H., Shen, L.D., 1997. Temporal geographic information

system and its application to transportation. Transportation Research Record:Journal of the Transportation Research Board 1593 (-1), 47–54.