Florida State University Libraries

Electronic Theses, Treatises and Dissertations The Graduate School

2018 Multi-Scalar Assessment of Built- Environment and Bus Networks Influence on Rapid-Transit Patronage: The Case of Metropolitan Transit Network Luis Enrique Ramos Santiago

Follow this and additional works at the DigiNole: FSU's Digital Repository. For more information, please contact [email protected]

FLORIDA STATE UNIVERSITY

COLLEGE OF SOCIAL SCIENCES AND PUBLIC POLICY

MULTI-SCALAR ASSESSMENT OF BUILT-ENVIRONMENT AND BUS NETWORKS

INFLUENCE ON RAPID-TRANSIT PATRONAGE

- THE CASE OF LOS ANGELES METROPOLITAN TRANSIT NETWORK -

By

LUIS ENRIQUE RAMOS SANTIAGO

A Dissertation submitted to the Department of Urban and Regional Planning in partial fulfillment of the requirements for the degree of Doctor of Philosophy

2018 Luis Enrique Ramos Santiago defended this dissertation on June 19, 2018. The members of the supervisory committee were:

Jeffrey Brown Professor Directing Dissertation

Mark Horner University Representative

Michael Duncan Committee Member

John Felkner Committee Member

The Graduate School has verified and approved the above-named committee members, and certifies that the dissertation has been approved in accordance with university requirements.

ii

I dedicate this doctoral dissertation to a group of special people who have supported me in a variety of ways throughout my doctoral studies.

To my mother, Carmen Luz Santiago de Ramos, for her unrelenting love and support of my personal, professional, and intellectual pursuits.

To my wife, Yasha N. Rodríguez Meléndez, for her continuous encouragement, patience, and work ethic. You have been and exemplary model to our daughter and myself as we make progress towards better life conditions from fast changing and uncertain economic circumstances at our homeland.

To my parents-in-law, Don Angel Rodríguez and Doña Carmen Meléndez, for their continuous and unwavering support to my family and myself during the years of doctoral studies as well as preceding ones when our well-being and that of many other young Puerto Rican families was (and still is) challenged by exceptional economic and environmental circumstances, and where grandparents emerged as stable foundations from which a new generation of Puerto Ricans begin to rebuild a centuries-old noble, resilient, and unique nation.

To my daughter, Mariela Alexandra Ramos Rodríguez. Your happiness, courage, hard work, musicality, intelligence, and exceptional adaptation skills to new landscapes, people, language, and cultures has been an inspiration to me. You have brought joy to this family since the day you were born.

- In memory of my father, Dr. Luis Ramos González (R.I.P.) -

iii ACKNOWLEDGMENTS

I want to recognize and celebrate the help of several individuals who have been instrumental for achieving completion and approval of this doctoral dissertation. First of all, my dissertation director Dr. Jeffrey Brown. His patient and wise counseling, shared knowledge, and overall guidance since my first year in the doctoral program and during my doctoral dissertation investigation was exceptional and I feel very fortunate to have had him as my supervisor during all these years at FSU-DURP. I really appreciate the opportunity to collaborate and co-publish with him and with other researchers as these experiences served as foundation for this dissertation work and for future academic endeavors.

Equally important was the support of an excellent dissertation committee composed of experts in fields pertinent to my investigation and their support throughout the research process. They are Dr. Michael Duncan, Dr. John Felkner, and Dr. Mark Horner. Their direct counseling, research, and nuanced knowledge of transportation/land-use interactions, spatial analysis, transit planning, and statistical/GIS methods was instrumental in completion and results of my investigative efforts.

I also want to acknowledge and recognize the financial support of Florida State University- Graduate School Dissertation Grants program without which I would not had been able to collect and process the necessary data for this investigation. Likewise I am very appreciative of FSU’s Undergraduate Research Opportunity Program (UROP) through which I served two years as research mentor to Alannah Harrington (FSU 2nd year-School of Engineering) and Dominique Turnbull (FSU 3rd year-International Studies and Planning). These two undergraduate students assisted in the collection and organization of data and calculation of some general descriptive statistics, tables, and charts in Chapter 1 and Chapter 2.

I am also very grateful and acknowledge the contributions of then graduate student Karissa Moffett (FSU M.Sc. ‘17-Geography; currently transportation planner at Jacksonville, Florida) who assisted in the early stages of bus schedules data collection, standardization, and file organization; Orlando N. Santaella Cruz (M.Sc. Geospatial Science and Technology, Polytechnic University of Puerto Rico) who assisted in bus and rail historic GTFS files and agency data collection, GIS protocols, custom GIS model-builder development, and development of a new integrated bus network geodatabase for a sub-set of multi-agency bus routes in Los Angeles regional transit system; and Luis R. Ortiz Sanchez (M.A. Geography, State University of New York at Binghamton; currently Planner II/Data Analyst at Central Midlands Regional Transit Authority, Columbia, South Carolina) who assisted in data collection and organization from the National Transit Database (FTA-NTD), in addition to gathering bus route schedule and vehicle data from agencies that did not have or provide GTFS files.

The support of ‘Walk Score® Professional’ data services (www.walkscore.com) who contributed at no cost to this investigation more than 5400 Walk Score rankings and associated measurement sub-components for both bus-stop and rapid-transit station portal locations was indispensable for completion and evaluation of key issues in this investigation and I am very grateful and indebted to them (data provided by Redfin Real Estate; https://www.redfin.com).

iv TABLE OF CONTENTS

List of Tables ...... vi List of Figures ...... viii Abstract ...... x

1. INTRODUCTION ...... 1

2. CASE STUDY SELECTION AND DESCRIPTION ...... 12

3. AN ASSESSMENT OF THE INFLUENCE OF BUS NETWORKS ON RAPID-TRANSIT STATION-LEVEL BOARDINGS ...... 31

4. AN ASSESSMENT OF THE INFLUENCE OF BUILT-ENVIRONMENT ATTRIBUTES AROUND BUS-STOPS ON RAPID-TRANSIT STATION BOARDINGS ……………………71

5. CONCLUSIONS ...... 126

APPENDICES ...... 129

A. DETAILED SURVEY DESCRIPTION AND DATABASE PREPARATION FOR ACCESS MODE POPULATION PROPORTION ESTIMATION ...... 129

B. FIGURE 1.1 AND FIGURE 1.2 SOURCES ...... 131

C. DISTRIBUTION OF EUCLIDEAN WALKING DISTANCES (WEIGHTED FREQUENCIES VS. MILES) FOR VARIOUS TRIP SEGMENTS ...... 133

D. MODEL 1A, 1B, AND 1C PREDICTED VS. OBSERVED OUTCOME PLOTS ...... 134

References ...... 135

Biographical Sketch ...... 149

v LIST OF TABLES

Table 2.1 Estimated System-wide Distribution of Access Mode - LA -Transit Services ...... 19

Table 2.2 LA Metro Average Weekday Access Mode Count and Bus Access Share Estimates By Rapid-Transit Line (2011-2012)...... 20

Table 2.3 LA Metro Station Annual Average Weekday Boardings1, Bus Access Mode Share and Count Estimates, and Confidence Intervals (2011)...... 23

Table 3.1 International Review of Rapid-Transit Station-Level Ridership Determinants (partial table): Research Methods and Demographic / Socioeconomic Factors 1996-2013 ...... 49

Table 3.2 International Review of Rapid-Transit Station-Level Ridership Determinants (partial table): Research Methods and Demographic / Socioeconomic Factors 2014-2017 ...... 50

Table 3.3 International Review of Rapid-Transit Station-Level Ridership Determinants (partial table): Land-Use / Urban-Design Factors 1996-2013 ...... 51

Table 3.4 International Review of Rapid-Transit Station-Level Ridership Determinants (partial table): Land-Use / Urban-Design Factors 2014-2017 ...... 52

Table 3.5 International Review of Rapid-Transit Station-Level Ridership Determinants (partial table): Transit Service Quality 1996-2013 ...... 53

Table 3.6 International Review of Rapid-Transit Station-Level Ridership Determinants (partial table): Transit Service Quality 2014-2017 ...... 54

Table 3.7 Descriptive Statistics for Model Variables ...... 60

Table 3.8 Regression Model Results and Model Fit ...... 67

Table 4.1 Summary of Associations and Weighted Elasticities of VMT*, 1, 2, Walking, and Transit with Respect to Build-Environment Variables ...... 82

Table 4.2 From 3Ds to 5Ds to 7Ds: Definitions and Measurements of Typical Variable Measures Used in Land-Use / Travel-Behavior Studies ...... 83

vi Table 4.3 Generalized Associations between Socioeconomic Characteristics, Land-Use and Built-Environment Attributes, and Travel Behavior...... 84

Table 4.4 Key Built Landscape Attributes Associated with Active Transportation: Walking and Bicycling...... 88

Table 4.5 GSEM Measures Descriptive Statistics...... 107

Table 4.6 GSEM Models AIC and BIC Criterion...... 121

vii LIST OF FIGURES

Figure 1.1 System-wide Access Mode Distribution and Metropolitan/Urban Sprawl Correlation Factors for Eight Large Rapid-Transit Systems in the U.S...... 10

Figure 1.2 Bus Access Share (%) in Nine Large Rapid-Transit Metropolitan Systems in the United States...... 11

Figure 2.1 Los Angeles Rapid-Transit and Freeway Network - 2011 ...... 13

Figure 2.2 LA Metro Rapid-Transit Network and SCAG Multi-Agency Bus Network ...... 16

Figure 2.3 Estimated Number of Bus Routes That Serviced Patrons Accessing LA Metro Rapid-Transit Services - By Transit Agency ...... 17

Figure 2.4 Estimated LA Metro Rapid-Transit Service Average Weekday Bus Transfers (Access) By Transit Agency ...... 18

Figure 2.5 a) Distribution of Access Mode: LA Metro Rail Services b) Estimated Distribution of Access Mode: LA Metro Rapid-Transit Services (Including Bus Access Mode) 21

Figure 2.6 Station-Level Average Weekday Boardings vs. Estimated Average Weekday Bus Access Count Scatterplot (2012) ...... 22

Figure 2.7 LA Metro Annual Average Station Weekday Boardings (FY 2011-2012) and Access Mode Share Estimates ...... 25

Figure 2.8 Estimated Annual Mean (Gray) and Median (Red) Household Income by Access Mode to Rapid-Transit Service (2011) ...... 26

Figure 2.9 Estimated General Socioeconomic Attributes of Rapid-Transit Patrons that Access Via Bus (2011) ...... 28

Figure 2.10 Estimated Activity Distribution Based on One-Way Person-Trip Origin Vs. Destination Place for Patrons that Accessed Rapid-Transit via Bus (2011) ...... 29

Figure 2.11 Estimated Home-Based Trips vs. Work-Based Trips Destination Activity Distribution of Patrons that Accessed Rapid-Transit via Bus (2011) ...... 30

Figure 4.1 LA Metro Rapid-Transit Network, SCAG Multi-Agency Bus Network, and Bus Stops That Registered at Least One Boarding Associated with Access to a Rapid- Transit Station (Partial Map of SCAG Region; Year 2011) ...... 97

Figure 4.2 Estimated Bus Stop Average Weekday Boardings Associated with Access to Rapid- Transit Stations (2011; excluding extreme outliers [SD>3]) ...... 98

viii

Figure 4.3 GSEM Model.0 ...... 109

Figure 4.4 GSEM Model.1 ...... 111

Figure 4.5 GSEM Model.2 ...... 113

Figure 4.6 GSEM Model.3 ...... 115

Figure 4.7 GSEM Model.4 ...... 116

Figure 4.8 GSEM Model.5 ...... 122

Figure 4.9 Los Angeles Rapid-Transit Network and Non-LA Metro Express Bus Routes .....123

Figure 4.10 Los Angeles Rapid-Transit Network and LA Metro ‘Rapid’ Bus Routes ...... 124

ix ABSTRACT

The advent of accelerated global warming and volatile climate change has prompted the need for a better understanding of what factors and policies might contribute to mitigate these events as well as increase the resilience of communities. Transit systems’ effectiveness and efficiency in increasingly disperse, car-dependent, and poly-centric urban agglomerations is one such factor, including the search for strategies to increase transit patronage and decrease car-dependence. Improving access to rapid-transit systems is one key area as it has the potential to expand the system’s influence beyond station’s immediate pedestrian service areas into larger and less developed suburban areas, and/or serve more disperse employment. Precedent studies and most on-board surveys have focused on a variety of access modes to reach rapid-transit services, including automobile, walking, and bicycle. Bus access, despite representing on average a non- trivial 19.3% of all access trips at national level, more than 30% at some large poly-centric cities in the U.S., and close to 50% of access trips for some rapid-transit lines (out-sizing the share of pedestrian access) has not received as much attention as other access modes. Predictive models for bus access mode report notably lower explanatory power as compared to other modes and the account of bus access events is often conflated with that of walk access in many technical reports and surveys for reasons yet to be understood. Ignoring, overlooking and/or misrepresenting this mode of access may lead to misunderstanding of multi-modal transit travel behavior and its spatial extent, possibly misguiding planners and policy-makers’ decision-making and resulting in system-wide ineffectiveness and/or inefficiency. This investigation documents bus access share for one exemplary case study and clarifies built-environment and bus networks’ influence on rapid-transit patronage within descriptive and inferential quantitative methodologies. This study seeks to answer two guiding research questions: 1- How important are bus networks to rapid- transit ridership in large, dispersed, poly-centric metropolitan regions in the U.S.? and 2- Do land-use and built-environment attributes around feeder bus-stops influence rapid-transit boardings? Because of diverse geographical scales and service levels experienced by a rider on a chained bus / rapid-transit trip this study focuses on two distinct yet linked geographies for analysis: 1-rapid-transit station; and 2- bus-stop. Research design is based on a single-case study in the United States (Los Angeles metropolitan multi-modal transit system). The first study focuses on quantifying the share of bus access trips at station-level and gaging its influence on

x total boardings within a multivariate generalized regression framework. Several socio-economic, service-level, built-environment, and network attributes are taken into consideration as informed by travel behavior theory and literature review. A strong positive association between bus network’s service and connectivity levels with rapid-transit station boardings registers high statistical confidence levels with boardings across all specified models. The mutual dependence of rapid-transit and bus networks evinced in the case of Los Angeles argues for a full multi- modal transit planning and operations paradigm for advancing a more effective, equitable, and sustainable transit system if it is to compete with ubiquitous automobile travel and its underpinning policy, fiscal, infrastructural, and cultural support. For Los Angeles, rapid-transit bus access represents an estimated 33.5% of all access events at a system-wide level, 20% - 49% at line-level, and a notably wider range at station-level (0% - 86%). The second study in this investigation focuses in assessing bus-stop pedestrian service areas built-environment and land- use attributes’ potential influence on rapid-transit station boardings, whilst controlling for both known and hypothesized control factors at bus-stop and station-level. By simultaneously focusing on bus-stop level attributes and higher-level rapid-transit stations’ attributes this part of the investigation fills a gap in the extant land-use / travel-behavior literature that more often focuses on pedestrian service areas adjacent to rapid-transit stations and ignores those around feeder bus-stops. Results evince a highly significant statistical relationship between bus-stop service area built-environment characteristics and the number of boardings associated with access trips to rapid-transit stations. However, the absolute effect relative to bus service levels and to automobile availability is notably smaller. Taken together as a multi-scalar study of bus and rapid-transit network interactions this investigation points to the importance of bus / rapid- transit network connectivity and service integration for maintaining and increasing rapid-transit patronage and the potential of synergistic contributions of built-environment interventions at feeder bus stops that seek to improve walkability and shorter walking distances. As a general conclusion, Los Angeles County Metropolitan Transit Authority and its associated MPO policy emphasis on TOD development as a strategy to increase transit ridership is limited. A more comprehensive policy approach based on ‘integrated public transportation’ and a more extensive station access policy that incorporates improvements around feeder bus stops, not only around stations, is the recommended course.

xi CHAPTER 1

INTRODUCTION

Bus Access to Rapid-Transit Service as a Key Component For Sustainable Urban Transportation Systems

The Social-Ecological Imperative for Sustainable Development

Car-dependent urban transportation systems generate significant greenhouse gas emissions and contaminants that contribute to environmental degradation, increased health risks, global warming and concomitant accelerated climate change, among other social and economic negative externalities at local, regional, and global scales (IPCC, 2014; Hanson & Giuliano, 2004; Bae 2004). Given the weak political support for fiscal policies geared to transfer social costs to automobile drivers, many urban and transportation scholars in the U.S. and beyond have proposed combined land-use/transport policies as a second-best option to advance more sustainable human settlements. In general, these policies are oriented towards three main goals: 1- promote higher shares of trips on transit systems and/or non-motorized modes; 2- reduce dependence on private automobile; and 3-develop compact mixed-use developments in support of the two previous goals and in support of other environmental and social benefits (Campbell 1996, Vuchic 2000, Newman & Kenworthy 1999, Cervero 1998, Mees 2010, CNU Charter 2000, APA 2002).

Combined land-use/transportation strategies to advance these ends entail implementation and improvement of transit systems and service levels, investment in non-motorized travel infrastructures, and transforming existing developments and/or creating new ones that are more pedestrian-friendly, auto-deterrent, and transit-supportive (i.e. TOD’s; Hanson & Giuliano, 2004; Campbell 1996, Vuchic 2000, Newman & Kenworthy 1999, Cervero 1998, Mees 2010, CNU Charter 2000, APA 2002).

1 It is expected by the proponents of said policies that the combined effect of these measures will hinder automobile availability and/or use and increase transit patronage. This would result in reduction of per-capita greenhouse gas production, assuming sufficient transit patronage is achieved; decrease in consumption of non-renewable resources; and help mitigate other multi-sectorial negative consequences associated with car-dependence, including interrelated contamination-health issues, injury or death as result of car collisions, global warming, and concomitant accelerated climate change with both its known and unknown consequences (IPCC, 2014).

Despite a general academic and professional convergence on the benefits of increased transit travel share and more pedestrian-friendly environments, and after implementation of several mass transit projects and associated Transit Oriented Development (TOD) programs across U.S. cities, transit share has remained weak and stagnant in the U.S. since the 1990’s. It averaged 1.9% of total trips in year 2009 as compared to less sustainable automobile travel share of 83.4% (NTHS 2009) for which average per capita trip distance still registers an increasing trend (Millard-Ball & Schipper 2011; US-EPA 2014). Nevertheless, absolute (net) transit ridership in the U.S. registers long-term positive trends since the 1970’s and some of the growth has been ascribed to investments in public transportation and renewed interest in central city living resulting in growing investments in transit accessible areas (APTA 2016).

Still in the period between years 2015-2016 the vast majority of cities in the continental U.S., with only three exceptions, experienced a fall in overall transit ridership. Most of this decline appears to have occurred in bus services rather than fixed-rail transit (FDOT-Urban iNTDS, accessed 11/17). It is interesting to note, however, that two of the only three cities that report an increase in transit ridership have recently upgraded and/or restructured their bus services and networks (Houston and Seattle; Bliss, L. Feb 24, 2017; FDOT-Urban iNTDS, accessed 11/17). As evinced in previous transit research (Brown et al. 2014, 2012a,b; Concas & DeSalvo 2014; Mees 2010) this observation suggests that policy geared to restructuring bus networks to better match emerging metropolitan structures and poly-centricity is likely effective and supports the aforementioned urban sustainability goals.

2 Some possible factors mentioned in association to this most recent nationwide transit patronage decline are economic expansion and falling oil prices, increase in vehicle purchases and driving, continuation of suburbanization trends, and a shift in immigrants’ behavior towards residential relocation in suburbs from the central city together with increased vehicle purchases (Blumenberg & Evans 2010). Reliability and overcrowding in some large transit systems, as well as service cuts in others are also mentioned as potential internal factors (Bliss 2017). Forefront on these trends is the convenience and flexibility afforded by the automobile in auto-friendly contexts (i.e. car-oriented suburban and urban landscapes that dominate U.S. metropolitan areas), which renders transit service an inferior good in the urban mobility market (McLeod et al. 1991).

Given the context of long-term stagnant transit trip share in the U.S., pervasive and un- sustainable car-dependent metropolitan landscapes, and the recent nationwide decline in transit ridership in a majority of U.S. cities it is pertinent and necessary to better understand which factors influence mass transit patronage in this 21st century. Moreover, it is appropriate to explore feasible ways to make it more attractive and competitive in the face of still dominant automobile use, ongoing global warming, and accelerated climate change.

Integrated (multimodal) Public Transport and Access to Rapid-Transit

Operational, financial, planning, built-environment, psychological, network design and topological aspects have been explored and studied in pursuit of better understanding and performance improvements of transit systems. Access to rapid-transit service is one such component considered indispensable for achieving rapid-transit cost-efficient patronage, access to jobs equity, and potentially increasing overall transit share in metropolitan regions (Guan et al. 2007; Kim et al. 2007; Brons et al. 2009; Sivakumaran, K. 2012; Chandra et al. 2013; Brands et al. 2014; Mees 2014; Chowdhury & Ceder 2016; Tal et al. 2016; Boarnet et al. 2017; Tabassum et al. 2017). No transit trip can transpire without a functional, safe, and convenient link from trip origin to transit service whether by walking, driving, bicycling, or riding a bus to a trunk-line station. According to a study by Beimborn et al. (2003) the ability of choice riders to walk to transit is more important than the difference in travel times between automobile and transit modes (as related to mode choice for travel to work). The same study also revealed that transfer-

3 time and wait-time are key factors for choice-rider market, whilst automobile availability is determinant to the transit-dependent segment.

Mees (2014) and Chowdhury and Ceder (2016) note the importance of effective integrated multimodal transit systems that provide users multiplicity of destinations, strategically located transfer nodes, and ‘seamless’ transfers for maintaining patronage and attracting automobile drivers to switch to public transportation. In reviewing integrated multimodal transit research Chowdhury and Ceder (2016) also note that more attention needs to be paid to factors related to travelers’ willingness to make transfers. In discussing Singapore’s and Honk Kong’s high transit trip share Luk and Olszewski (2003) also note the importance of ‘integrated public transport’ as a key element for successfully curtailing car usage. Whilst recognizing the distinct high population densities in these Asian cities the authors also highlight the importance of the high bus access rate to mass rapid transit and the strategic role of both local and bus feeder routes. These latter authors underscore five key elements for successful multimodal public transit: physical integration, network integration, fare integration, information integration, and institutional integration.

In assessing the role of accessibility in overall rail journey satisfaction in the Netherlands Brons et al. (2009) noted that access to rail service receives less attention as compared to rail service levels in patronage studies. Yet these authors concluded in their study that accessibility is a very important determinant of the overall rail journey satisfaction based on users’ survey data and suggest that improving access to railway stations can be a more cost-efficient strategy than expanding rail network services for increasing rail ridership. In a study of access mode-choice for high-speed rail in Taiwan Wen et al. (2012) identified that most riders are cost-sensitive to access modes as compared to reduced access time. However, these preceding studies focus on national rail services and not on urban rapid-transit services. Literature on determinants of access mode-choice for rapid-transit service has been more difficult to find, particularly studies that focus on the United States.

Tabassum et al. (2017) evaluated existing informal feeder modes for a BRT system in the city of Lahore (Pakistan) and conclude that accessibility by a ‘regular feeder service’ would be

4 significant better as compared to other conventional public modes, noting that improvements in accessibility to line-haul service eventually will raise ridership by shifting commuters from their automobiles. Guan et al. (2007) concluded that the effectiveness of urban railway system in Beijing, China, requires effective connectivity with other transportation modes to increase accessibility. Their results indicate that access time, access cost, and access distance affect railway access choices as well as users income and vehicle ownership.

Research related rapid-transit access mode-choice in the U.S. has been sporadic. In the past ‘urban transportation analyses ignored or dealt casually with cost of travel between the home and the line-haul system’ and thus neglected more ‘accurate estimation of costs associated with additional parking or feeder bus operations’ aimed at providing access to/from stations (Meyer et al. 1969). Korf and Demetsky (1981) called to attention the extensive research and application of travel demand models yet scant attention to the choice of mode of access. More recent rapid-transit TOD and direct-demand forecasting studies tends to focus their discussion on walk and bicycle access, with some attention to park-n-ride facilities (Coffel, K. et al. 2012; Brands et al. 2014; Mees 2014).

Bus access to rapid-transit service is less frequently studied despite representing on average a non-trivial 19.10% system-wide access mode share in large U.S. transit systems (Figure 1.1, 1.2). In disperse, large, poly-centric urban agglomerations bus access share for rapid-transit services is relatively higher and greater than 30% (i.e. Los Angeles, Atlanta, and Miami; Figure 1.2) and can register a highly variable range at station-level (0% - 86%; Table 2.3, Figure 2.7). For some rapid-transit lines, particularly those whose alignments and stations locate mostly within freeways right-of-ways, bus access share can be the dominant mode, registering shares close to 50% and greater than walking share (Table 2.2).

Bus access potential in supporting rapid-transit patronage, particularly in sprawling, disperse, poly-centric urban agglomerations warrants further examination (Coffel et al. 2012; Mees 2014). Linear predictive models for other access modes register notably higher explanatory power. A lack of sufficient bus services information and statistics has been noted as one 2 2 2 2 plausible cause (푅푏푢푠= 0.373, 푅푐푎푟= 0.821, 푅푤푎푙푘= 0.717, 푅푏푖푐푦푐푙푒= 0.771; Coffel, K. et al.

5 2012). Perhaps the more complex multi-modal nature of bus access to rapid-transit stations makes more difficult the gathering of pertinent statistics and its analysis. Yet, modern APC (Automated Passenger Counting) technology should facilitate comprehensive recordings of counts, transfers, and waypoints for this type of trip in the future.

Several large transit agency on-board surveys and technical reports mischaracterize bus access mode shares (Figure 2.5a), which are often invisibilized and conflated with walk access for reasons yet to be understood. When documented, bus access shares are reported at system- wide and/or line-level but often ignored at the station-level. This is problematic as it is at station- level where both metropolitan (i.e. rapid-transit) and local (i.e. bus) transit supply and demand forces meet, thus crucial for planning, operations, and research purposes. The misrepresentation of bus access trips could also mislead planners and decision-makers on the spatial extent and composition of multi-modal transit trips, and on modal access distribution. Under this scenario walking (pedestrian) access to rapid-transit stations may be over-valued and bus access under- valued and could result in un-balanced land-use/transportation policies and lower overall transit system effectiveness and efficiency.

As registered in Figure 1.1 (bottom), a strong correlation exist between various spatial measures of metropolitan structure, such as decentralization, density gradient, urban area population, and accessibility with the share of bus access trips to rapid-transit services. If long- term trends of urban decentralization and suburbanization in the United States continue into the future (Ewing & Hamidi 2014; Lee 2007) it is plausible to expect that the share of bus access trips to rapid-transit service will increase as well, all else equal. Thus, it is important and pertinent to gain a better understanding of this particular travel behavior, as well as identify associated factors that might contribute to higher patronage of rapid-transit services.

Among several factors that have been identified as significant for rapid-transit patronage, particularly at the station-level, are population levels, bus network connectivity levels, and pedestrian-friendly land-use / built-environment characteristics within stations’ service areas (Cervero 2001; Kim et al. 2007; Guerra & Cervero 2011; Chakraborty & Mishra 2013; Ramos & Brown 2016; Balcombe et al. 2004; Taylor & Fink 2003; Brown & Thompson 2008a,b;

6 Balcombe et al. 2004; Gutierrez et al. 2011; Dill et al. 2013; Zhao et al. 2014; Durning & Townsend 2015; Chen & Zegras 2016; Renee et al. 2017; Mees 2014).

The focus on TOD policies and programs in part reflects the importance given to walk access, population density and other local characteristics, such as mix of uses and pedestrian friendly environments often characterized by smaller block lengths and/or higher intersection densities. These specifics and emphasis on the built-environment respond in part to key studies and publications that have emerged since the 1970s that evince a link between the aforementioned attributes and transit patronage at line-level and oriented towards CBD destinations (Pushkarev and Zupan 1977; Parsons Brinckerhoff et al. 1996; among others).

Most if not all station-level rapid-transit patronage studies that have considered land-use and built-environment attributes (LU.BE) have fixated in areas immediately surrounding stations (Kuby et al. 2004; Gutierrez et al. 2011; Dill et al. 2013; Zhao et al. 2014; Mees 2014; Durning & Townsend 2015; Ramos and Brown 2016; Chen & Zegras 2016; Renee et al. 2017). Yet they have ignored the potential influence of LU.BE attributes around feeder bus-stops. This might be relevant for systems that register high rates of bus access, such as those that operate within large, poly-centric urban agglomerations. Hence, two important aspects of multi-modal transit systems in the U.S., bus access to rapid-transit service and the potential influence of LU.BE factors around feeder bus-stops remain to be explored and better understood. This dissertation aims to shed light on these two gaps in the hope of advancing more effective and sustainable multi- modal urban transportation systems.

The Importance of Bus Access to Rapid-Transit Service for Lower-Income Transit- Dependent Communities

Lower-income households and adults, and transit-dependent riders are particularly reliant on bus services (PTV NuStats-LACMTA 2012; Brown et al. 2012 a,b); register higher frequency of multi-modal trips (Blumenberg and Evans 2010, Blumenberg and Pierce 2014); and tend to rely more on bus services to access rapid-transit service (Ramos, working paper).

7 In addition to mobility limitations due to lower automobile availability and/or lower driving license tenure, housing market factors can further increase their dependence on transit and on bus access to rapid-transit service. Some studies that focus on high-capacity transit effects on real-estate values near stations note that land, housing, and/or commercial property values within walking distance of stations tend to be relatively higher for some rapid-transit lines in Los Angeles County (Cervero, R. & Duncan, M. 2002). As such, it is plausible to think that higher housing costs (rental and/or sales price) near rapid-transit stations might exclude lower-income and transit-dependent riders.

These combined socio-economic and housing market forces in part may co-determine lower-income households’ peripheral housing locations relative to rapid-transit stations and hence their accentuated dependence on bus mode for access to rapid-transit service (Coffel, K. et al. 2012; PTV NuStats-LACMTA 2012). As such this line of research and focus on bus mode as feeder-service has an important social equity dimension.

Dissertation Outline

The rest of this manuscript is organized in four additional chapters. The upcoming Chapter 2 presents and describes the case study’s multi-modal transit system network infrastructure and service characteristics, as well as general socioeconomic attributes and trip patterns of patrons that rely on bus service to access rapid-transit stations. This information serves as a shared platform for the more focused studies in the subsequent two chapters.

Chapter 3 presents and discusses the literature review, research design, and findings related to bus networks influence on rapid-transit station boardings, as well as a more nuanced exploration of bus network connectivity and supply measures in pursuit of generalized linear regression demand models predictive power improvements. Both descriptive and inferential statistical methods are implemented. The guiding research question in this first study is:

RQ-1 How important are bus networks to rapid-transit ridership in large, dispersed, poly- centric metropolitan regions in the U.S.?

8 Chapter 4 shifts the scale and geographical foci of this inquiry from the station to feeder bus- stops in trying to better understand bus access-mode determinants in a more disaggregate level, and in assessing the potential role that land-use and/or built-environment attributes (LU.BE) around feeder bus-stops may have on the number of boardings at station-level. Literature review, research design, and results discussion sections organize this second study where a general structural equations model with a latent variable construct is implemented. The guiding research question in this second study is:

RQ-2 Do land-use and built-environment attributes around feeder bus-stops influence rapid-transit boardings?

At the end of Chapter 3 and Chapter 4 standardized regression coefficients and predictive marginal effects are presented for key variables of interest. These are compared and discussed for potential policy analyses and/or theoretical discussion. Research limitations in both study-1 and study-2 and future areas of inquiry are also discussed in both chapters.

The final Chapter 5 summarizes and synthesizes the findings of the two previous studies and presents recommended policies for bus / rapid-transit multimodal transit networks in pursuit of a better understanding of travel behavior in multimodal contexts, increasing higher transit patronage, and promoting a more effective and convenient urban transit system as a viable alternative to on-going car-dependency in the Los Angeles region as well as in other dispersed car-dependent metropolises. The findings are also contrasted with the on-going policy focus on Transit Oriented Development (TOD) in areas adjacent to rapid-transit stations.

9 100% 100

90% 90 80%

70% 80

60% 70 50% 60 40%

30% 50

20% 33.5* 32.0 30.7 40 10% 22.0 21.0 16.0 8.0 4.7 0% 30

Bus Walk Car Bicycle Other Accessibility Index (1992)11

Correlation Factor with Factor Geography Bus Access Share Factor Data Source

Accessibility Index metropolitan 0.9457 Allen et al. (1993) Density Gradient urban 0.8774 Ewing & Hamidi (2014a) Centeredness urban (-0.7451) Ewing & Hamidi (2014b) Urban Area Population urban 0.7333 Ewing & Hamidi (2014a)

Figure 1.1 System-wide Access Mode Distribution and Metropolitan/Urban Sprawl Correlation Factors for Eight Large Rapid-Transit Systems in the U.S. Sources: see Appendix B * Estimated by author using STATA v.13 ‘svyset’ command and ‘subpopulation’ calibration on PTV NuStats-LACMTA (2012) On-board survey digital database.

10 40

35 Los Angeles * Miami Atlanta 30 Boston

25 San Francisco Washington DC

20

19.10 MEAN Chicago 15

10

5 Philadelphia San Juan

0

1985 1990 1995 2000 2005 2010 2015 2020

Figure 1.2 Bus Access Share (%) in Nine Large Rapid-Transit Metropolitan Systems in the United States Sources: see Appendix B * Estimated by author using STATA v.13 ‘svyset’ command and ‘subpopulation’ calibration on PTV NusStats LACMTA (2012) On-board survey digital database.

11 CHAPTER 2

CASE STUDY SELECTION AND DESCRIPTION

Los Angeles Metropolitan Transit System

This chapter presents and describes the selected case study for this investigation, including selection criteria, institutional framework, network and service characteristics, and a general socioeconomic profile of rapid-transit users that accessed LA Metro’s rapid-transit stations via a variety of bus services. This information serves as background and reference for the two subsequent and more focused studies on bus / rapid-transit network interactions contained in Chapters 3 and 4.

Overall Research Design, Case Selection, Description, and General Definitions

Criteria for selection of a case that would facilitate the exploration of bus feeder networks influence on station-level ridership, including potential influence of LU.BE attributes, ideally considers a metropolitan area in the U.S. with a poly-centric structure; a spread-out metropolitan footprint that contains a variety of urban and suburban landscapes; disperse employment; and a multimodal transit network consisting of a variety rapid-transit services and bus networks. In addition, relevant ridership and boarding statistics at a variety of network scales, person-trip socioeconomic and spatial data, multimodal transit network GIS database, transit service characteristics, and land-use/built-environment data should be available.

Based on the above criteria this study is based on Los Angeles metropolitan rapid-transit system and its associated multi-agency, multi-city, and multi-county bus network. Los Angeles is considered an archetype of modern poly-centric metropolitan structures (Dear 2002), or a ‘dispersed mega-region’ as characterized by other scholars that challenge the poly-centricity thesis (Gordon and Richardson 1996). It is listed as one of a limited set of elite ‘global’ and ‘alpha’ cities due to its population size, spatial extent, economic output, and global cultural influence (Taylor 2005), but especially due to its global advanced-producer service-industry network connectivity (Taylor 2011; http://www.lboro.ac.uk/gawc/world2016t.html).

12 The ‘Los Angeles-Long Beach-Anaheim Metropolitan Statistical Area’ (LA-MSA) reports a large and very diverse multi-ethnic population of 13,154,457 residents in year 2015 (US Census ACS 2015). As an agglomeration of several cities and counties that gradually coalesced in time, mostly and rapidly during early 20th century (Soja et al. 1983), Los Angeles metropolitan urban and suburban landscapes also exhibit a great variety of built-environments ideal for this investigation. Furthermore Los Angeles is also considered the epitome of modern automobile- oriented city planning and design (Wachs 1993) and contains an extensive network of freeways. This dominant automobile orientation has been challenged and juxtaposed with an aggressive rapid-transit network investment program that started with the Blue Line in year 1990, nonetheless with limited success (Wachs 1993; Lee et al. 2017; Figure 2.1).

Figure 2.1 Los Angeles Rapid-Transit1 and Freeway2 Networks - 2011 Sources: Metro ‘Bus & Rail GIS Data (2011)’1 (http://developer.metro.net:80/introduction/gis-data/download-gis-data/; retrieved December2015 from ‘Internet Archive – Wayback Machine’; https://web.archive.org/) and OpenStreetMap ‘California’ shapefile2 (california-latest-free.shp.zip; retrieved Dec. 2015) 13 Los Angeles multi-modal transit system is purposefully selected to address the research questions and spatial context posed for this dissertation; to inform statistical analyses; to be relevant to large poly-centric urban agglomerations in the U.S. context; and for which pertinent person-trip survey information and associated spatial observations, socioeconomic, land-use, network, and multimodal transit service data are available.

Definitions. In this study rapid-transit refers to high-capacity, high-speed, and high service (10-15min frequency or less) public urban passenger transportation that mostly operates on exclusive right-of-ways, excluding commuter rail. Rapid-transit modes in the Los Angeles metropolitan region include heavy-rail (HRT), light-rail (LRT), and bus rapid transit (BRT) lines, all of which are operated by Los Angeles County Metropolitan Transportation Authority (LACMTA, a.k.a. LA Metro). In this study bus feeder services include all bus routes across a wide service spectrum (i.e. local, community circulators, limited-stop, Rapid, Express, and commute) that have or could potentially carry passengers to a rapid-transit portal, whether because it is physically and operationally linked to a rapid-transit station (i.e. bus stop coincides with location or is adjacent to rapid-transit portal), or because the bus routes and associated stops are located within a reasonable walking distance to a rapid-transit portal. The ‘reasonable walking distance’ parameter was empirically determined based on PTV NuStats-LACMTA’s (2012) on-board person-trip survey spatial data where the maximum Euclidean walk distance between the alighting bus-stop and first rapid-transit portal is used as the rapid-transit portal’s feeder bus route catchment area (r=.34mile, after excluding extreme outliers SD>2.5).

Institutional Framework. The Los Angeles multi-agency regional transit system is under the purview of the ‘Southern California Association of Governments’ (SCAG), which is a federally mandated and funded transportation policy-making Metropolitan Planning Organization (MPO). SCAG, founded in 1965, is the largest MPO in the U.S. and encompasses six counties (Imperial, Los Angeles, Orange, Riverside, San Bernardino and Ventura), six county transportation commissions, 191 cities, and the region’s Native American tribes (SCAG 2012).

The ‘Los Angeles County Metropolitan Transportation Authority’ (LACMTA, a.k.a. LA Metro), which is the largest multimodal transit agency operating within SCAG, focuses its

14 policies and activities in providing an efficient and effective transportation system in a sustainable manner for the Los Angeles County (Metro 2011a). It performs transportation planning, coordination, design, building, and operations in one of the most populated counties in the United States, which in year 2011 registered more than 9.6 million individuals (close to 1/3 of California’s residents). It also services 1,433 square miles of transit service area (Metro 2011a).

Transit Network Description. The transit system in the six-county SCAG Region is comprised of an extensive network of services provided by dozens of operators (SCAG 2015). The network includes fixed-route local bus, community circulators, express bus, limited-stop, rapid, bus rapid transit (BRT), demand response, commuter rail, heavy rail (HRT), and light rail (LRT; Figure 2.2). LA Metro operations interact with several other city and county transit services under SCAG’s administrative umbrella. In year 2012, which is the base year for this investigation, LA Metro operated 183 bus routes (local, limited-stop, express, circulators, and Rapid), 3 light-rail lines (LRT: Blue, Green, Gold), 2 heavy-rail lines (HRT: Red, Purple), and 2 bus rapid transit lines (BRT: Orange, Silver; LA Metro 2012; Figures 2.3 and 2.4)

According to this author’s estimates using LA Metro’s on-board person-trip survey database (PTV NuStats-LACMTA 2012), the majority of bus routes associated with access trips to rapid-transit stations are operated by non-LA Metro agencies (196 vs.178; Figures 2.3) and suggests a significant degree of inter-county interactions and inter-agency transfers. Still the majority of persons that access LA Metro’s rapid-transit services using bus rely on LA Metro’s own bus routes (63,928 average weekday person-trips (71.2%) vs. 12,956 non-LA Metro average weekday person-trips (14.4%) Figure 2.4). ‘’, ‘Foothill’, and ‘LADOT’ (Los Angeles City - Department of Transportation) are three non-LA Metro bus agencies that feed the largest number of riders to LA Metro’s rapid-transit services.

15

Figure 2.2 LA Metro Rapid-Transit Network1 and SCAG Multi-Agency Bus Network2 Sources: Metro ‘Bus & Rail GIS Data (2011)’1 (http://developer.metro.net:80/introduction/gis- data/download-gis-data/; retrieved December2015 from ‘Internet Archive – Wayback Machine’; https://web.archive.org/ ) and SCAG ‘Bus Routes (2007)’ shapefile (retrieved December 2015; http://gisdata.scag.ca.gov/Pages/GIS-Library.aspx)

16 Figure 2.3 Estimated Number of Bus Routes That Serviced Patrons Accessing LA Metro Rapid-Transit Services - By Transit Agency Source: ‘Metro 2011 On-board Survey’ digital database (PTV NuStats-LACMTA (2012); estimated by author in STATA v.12 using ‘svyset’ command and ‘subpopulation’ calibration.

17

Figure 2.4 Estimated LA Metro Rapid-Transit Service Average Weekday Bus Transfers (Access) By Transit Agency Source: ‘Metro 2011 On-board Survey’ digital database (PTV NuStats-LACMTA 2012; estimated by author in STATA v.12 using ‘svyset’ command and ‘subpopulation’ calibration).

18 Bus Access Share Estimations at Multiple Network Scales. As registered in Table 2.1, Figure 2.5a, and Figure 2.5b system-wide bus access to LA Metro rapid-transit service is estimated at 33.5% of all access trips, second to that for walking at 54.2%. Bus access to rapid- transit service in Los Angeles also registers the largest bus access share of the nine large US transit agencies documented in this study. Car access reports the third largest access share at 9.6% and bicycle access represents a relatively small share of all access trips at 2.3%.

Compared to the original and typical mode access statistics representation in LA Metro’s on-board survey report, as well as in that of many other transit agencies where bus access mode category is often ignored and/or confounded with walk access mode (Figure 2.5a), the revised version presented in this manuscript offers a more nuanced and arguably more valid representation where the relative importance of bus access vis-a-vis walk, car, bicycle, and other access options is evinced (Figure 2.5b). Not including bus access category in system-wide, line- and station-level reports misrepresents the magnitude and importance of multimodal transit trips, users travel behavior, and trips spatial extent. This shortcoming in mode access (and egress) statistic representations might bias planners, policy-makers, and the general public by over- representing walk access events and under-representing (or ignoring) bus access events. This distortion in measurement might affect which polices are advanced or not (i.e. TOD land-use development around stations vs. improvements in bus-rail connectivity levels, bus service levels, and/or land-use and urban design improvements along bus feeder corridors).

Table 2.1 Estimated System-wide Distribution of Access Mode - LA Metro Rapid- Transit Services Source: ‘Metro 2011 On-board Survey’ digital database (PTV NuStats-LACMTA (2012); estimated by author in STATA v.12 using ‘svyset’ command and ‘subpopulation’ calibration.

Percent Proportion Std. Err. LB 90th CL UB 90th CL

BUS 33.53% 0.3353 0.0366 0.2777 0.3983 WALK 54.24% 0.5424 0.0334 0.4867 0.5971 CAR 9.61% 0.0961 0.0102 0.0805 0.1143 BICYCLE 2.28% 0.0228 0.0038 0.0173 0.0299 OTHER 0.23% 0.0023 0.0009 0.0012 0.0045

19 Line-level access mode share statistics register a notable variability in bus access rates between 19.86% at the lowest (LRT Gold Line) and 48.47% at the highest (BRT Silver Line; Table 2.2). Because the Silver Line is a mixed limited-stop / BRT service that operates for the most part in a shared-busway along Freeway 110 and 10 it is not surprising that it registers the highest bus access rate and the lowest average weekday boardings at station level (2180 vs. 8547.). The context surrounding busway stations along the Silver Line, with the exception of street-level stops downtown, is dedicated for the most part to motorways and undeveloped, which makes this line-haul service more dependent on bus access rather than walk access. This contextual condition probably places Silver Line freeway stations in disadvantage as compared to other rapid-transit stations located in more develop, accessible and dense environments. Furthermore the relatively unpleasant environmental conditions at stations located in Freeway medians, mainly noise and pollution from nearby speeding automobiles, might be a considerable factor that may discourage use of the line.

Table 2.2 LA Metro Average Weekday Access Mode Count and Bus Access Share Estimates By Rapid-Transit Line (2011-2012) Source: ‘Metro 2011 On-board Survey’ digital database (PTV NuStats-LACMTA 2012; estimated by author in STATA v.12 using ‘svyset’ command and ‘subpopulation’ calibration).

ESTIMATED ACCESS MODE COUNT

Bus ROUTE sub- ROUTE MODE Bus Walk Car Bicycle Other Access NAME total Share MT-801 Light Rail Blue Line 26,914 47,155 8,272 2,895 252 85,488 30.32%

MT-802 Heavy Rail Red Line 46,805 68,684 8,261 1,856 154 125,760 34.76%

MT-803 Light Rail Green Line 18,278 15,675 6,857 1,043 90 41,943 43.99%

MT-804 Light Rail Gold Line 8,855 22,115 5,238 1,340 471 38,019 19.86%

MT-805 Heavy Rail Purple Line 10,223 19,102 1,730 700 - 31,755 31.34% Bus Rapid Orange MT-901 11,617 12,731 4,160 381 - 28,890 38.96% Transit Line Bus Rapid MT-910 Silver Line 5,815 4,963 1,644 141 - 12,562 48.47% Transit sub-total= 128,506 190,426 36,163 8,355 966 364,417

20

a)

100% 90% 80% 70% 54.24% 60% 50% 40% 33.53% 30%

20% 9.61% b) 10% 2.28% 0.23% 0%

Figure 2.5 a) Distribution of Access Mode: LA Metro Rail Services b) Estimated Distribution of Access Mode: LA Metro Rapid-Transit Services (Including Bus Access Mode) Sources: a) PTV NuStats-LACMTA ‘Metro On-board Bus & Rail Survey’ (2012; p.46); b) ‘Metro 2011 On-board Survey’ digital database (PTV NuStats-LACMTA (2012); estimated by author in STATA v.12 using ‘svyset’ command and ‘subpopulation’ calibration).

21 At the station-level, which is one of the two units of analysis in this study, considerable variability in bus access mode share exist (1.40% - 87.00%; N=105, Table 2.3, Figure 2.6). As noted in Figure 2.6, there is a positive linear association between the number of bus-access counts and total boardings at station-level, which registers a strong positive correlation factor (0.9131).

40000

30000

20000

10000

Rapid-TransitStation WeekdayAverage Boardings

0

0 5000 10000 15000 20000 Bus Transfers (Access) at Rapid-Transit Station

Figure 2.6 Station-Level Average Weekday Boardings1 vs. Estimated Average Weekday Bus Access Count2 Scatterplot (2012) Source: LA Metro1 (2015) and ‘Metro 2011 On-board Survey’2 (PTV NuStats-LACMTA2 (2012); estimated by author in Stata v.13 using ‘svyset’ command and ‘subpopulation’ calibration).

22 Table 2.3 LA Metro Station Annual Average Weekday Boardings1, Bus Access Mode Share and Count Estimates, and Confidence Intervals (2011).

Avg. Bus Bus Access Standard Weekday Access n RAPID TRANSIT STATION Rate Error Lower 90th CL Upper 90th CL Boarding Count 1 1ST / HILL 0.695 0.112 0.488 0.845 791 549 2 1ST STREET STATION 0.000 0.000 0.000 0.000 460 0 3 5TH STREET STATION 0.100 0.056 0.038 0.238 996 100 4 7TH STREET/METRO CENTER STATION 0.266 0.062 0.176 0.380 38,665 10,266 5 0.046 0.029 0.016 0.128 1,431 66 6 ANAHEIM STATION 0.174 0.024 0.138 0.218 2,961 515 7 0.675 0.032 0.620 0.727 3,743 2,528 8 ARTESIA TRANSIT CENTER 0.740 0.054 0.642 0.818 1,836 1,358 9 ATLANTIC STATION 0.258 0.075 0.153 0.400 1,881 485 10 AVALON STATION 0.473 0.161 0.235 0.724 2,678 1,268 11 AVIATION LAX STATION 0.400 0.020 0.368 0.433 3,475 1,390 12 0.099 0.086 0.021 0.353 1,634 161 13 CAL STATE LA BUSWAY STATION 0.020 0.016 0.005 0.073 2,057 41 14 0.067 0.030 0.032 0.137 1,349 91 15 CHINATOWN STATION 0.222 0.068 0.129 0.354 1,688 374 16 CIVIC CENTER STATION 0.442 0.080 0.317 0.575 5,635 2,493 17 0.103 0.026 0.067 0.155 4,209 433 18 0.818 0.036 0.751 0.870 3,171 2,593 19 DE 0.224 0.180 0.049 0.616 819 183 20 0.156 0.024 0.121 0.200 3,166 494 21 0.061 0.015 0.040 0.090 1,442 87 22 DOUGLAS STATION 0.140 0.027 0.101 0.192 761 107 23 DOWNTOWN LONG BEACH STATION 0.388 0.096 0.246 0.553 7,129 2,768 24 EAST LA CIVIC CENTER STATION 0.023 0.009 0.012 0.042 751 17 25 EL MONTE BUSWAY / ALAMEDA UNION STATION 0.826 0.089 0.631 0.930 908 750 26 0.541 0.127 0.335 0.734 2,759 1,493 27 0.000 0.000 0.000 0.000 1,033 0 28 FIGUEROA / 7TH 0.195 0.044 0.132 0.279 658 128 29 FIGUEROA / ADAMS 0.344 0.210 0.101 0.711 50 17 30 FIGUEROA / EXPOSITION 0.225 0.179 0.050 0.614 197 44 31 FIGUEROA / OLYMPIC 1.000 0.000 0.000 0.000 71 71 32 FIGUEROA / PICO 0.873 0.106 0.586 0.971 214 187 33 FIGUEROA / WASHINGTON 0.306 0.225 0.071 0.717 206 63 34 0.038 0.041 0.006 0.204 1,590 61 35 0.478 0.117 0.297 0.666 3,183 1,522 36 FLORENCE STATION 0.618 0.132 0.390 0.804 4,283 2,648 37 FLOWER / 5TH 0.066 0.008 0.054 0.081 186 12 38 FLOWER / 7TH 0.275 0.119 0.124 0.505 720 198 39 FLOWER / OLYMPIC 0.360 0.078 0.244 0.496 119 43 40 GRAND / 3RD 1.000 0.000 0.000 0.000 83 83 41 GRAND STATION 0.246 0.077 0.141 0.392 4,722 1,161 42 0.543 0.086 0.402 0.678 3,887 2,111 43 HARBOR TRANSITWAY / ROSECRANS 0.639 0.281 0.191 0.930 184 118 44 HARBOR TRANSITWAY / SLAUSON 0.798 0.059 0.682 0.879 232 185 45 HAWTHORNE STATION 0.609 0.048 0.528 0.685 2,987 1,820 46 0.033 0.022 0.011 0.098 840 28 47 HIGHLAND PARK STATION 0.314 0.122 0.152 0.539 2,663 836 48 HOLLYWOOD/HIGHLAND STATION 0.217 0.064 0.129 0.341 8,160 1,770 49 HOLLYWOOD/VINE STATION 0.212 0.037 0.157 0.280 6,268 1,329 50 HOLLYWOOD/WESTERN STATION 0.304 0.093 0.174 0.475 5,303 1,610 51 IMPERIAL WILMINGTON STATION 0.271 0.073 0.168 0.407 20,116 5,459 52 INDIANA 1ST STATION 0.131 0.041 0.076 0.216 1,875 245 53 LAKE STATION 0.109 0.065 0.039 0.271 1,658 182 54 LAKEWOOD STATION 0.522 0.044 0.449 0.593 2,862 1,493 55 LAUREL CANYON STATION 0.178 0.109 0.059 0.428 1,441 256 56 LINCOLN AV/CYPRESS STATION 0.348 0.080 0.230 0.488 1,326 461 57 LITTLE TOKYO STATION 0.142 0.043 0.084 0.230 2,160 307 58 LONG BEACH STATION 0.497 0.134 0.288 0.706 3,433 1,705 59 MARAVILLA STATION 0.072 0.030 0.036 0.139 434 31 60 STATION 0.014 0.001 0.011 0.016 854 12 61 MARIPOSA STATION 0.214 0.055 0.137 0.318 1,263 270 62 0.242 0.091 0.123 0.421 2,212 536 63 MISSION STATION 0.059 0.018 0.036 0.096 1,633 97 64 0.413 0.106 0.254 0.593 16,115 6,661 65 NORTH HOLLYWOOD STATION LAYOVER 0.453 0.090 0.312 0.603 8,155 3,695 66 NORWALK STATION 0.362 0.093 0.226 0.525 4,866 1,762 67 OLIVE / 5TH 0.212 0.151 0.057 0.547 491 104 68 OLIVE / GENERAL THADDEUS KOSCIUSZKO 0.000 0.000 0.000 0.000 109 0 23 Table 2.3-continued

Avg. Bus Bus Access Standard Weekday Access n RAPID TRANSIT STATION Rate Error Lower 90th CL Upper 90th CL Boarding Count 69 PACIFIC COAST HWY STATION 0.268 0.016 0.241 0.295 2,769 741 70 PACIFIC STATION 0.075 0.012 0.057 0.098 1,312 98 71 0.338 0.043 0.270 0.413 10,812 3,650 72 PICO ALISO STATION 0.002 0.001 0.002 0.003 886 2 73 0.467 0.111 0.295 0.646 2,801 1,307 74 0.107 0.024 0.073 0.153 1,939 207 75 0.037 0.015 0.019 0.071 1,110 41 76 0.593 0.063 0.485 0.692 2,723 1,614 77 SAN PEDRO STATION 0.300 0.081 0.185 0.447 2,253 675 78 0.851 0.102 0.601 0.956 1,939 1,650 79 SIERRA MADRE VILLA STATION 0.185 0.064 0.102 0.314 2,841 527 80 0.440 0.108 0.276 0.619 2,567 1,130 81 SOTO STATION 0.352 0.040 0.289 0.420 1,818 640 82 0.115 0.059 0.047 0.254 915 105 83 SPRING / 1ST - CH 0.729 0.000- 0.057 1.000 390 284 84 0.176 0.123 0.050 0.467 668 118 85 UNION STATION 0.640 0.048 0.557 0.715 32,011 20,471 86 UNIVERSAL CITY STATION 0.404 0.104 0.249 0.582 7,621 3,080 87 USC MEDICAL CTR BUSWAY STATION 0.184 0.081 0.085 0.356 259 48 88 0.047 0.031 0.016 0.132 1,517 71 89 VAN NUYS STATION 0.572 0.129 0.359 0.762 3,869 2,214 90 VERMONT STATION 0.432 0.049 0.353 0.513 3,518 1,518 91 VERMONT/BEVERLY STATION 0.271 0.054 0.191 0.369 4,624 1,253 92 VERMONT/SANTA MONICA STATION 0.209 0.069 0.117 0.346 5,983 1,251 93 VERMONT/SUNSET STATION 0.463 0.224 0.162 0.793 4,884 2,262 94 0.273 0.094 0.146 0.451 2,880 785 95 0.124 0.057 0.056 0.251 1,633 202 96 WARNER CENTER TRANSIT HUB 0.213 0.017 0.187 0.243 934 199 97 WARNER CENTER TRANSIT HUB LAYOVER 0.217 0.098 0.096 0.420 91 20 98 WASHINGTON STATION 0.077 0.021 0.049 0.118 1,720 132 99 WESTLAKE/MACARTHUR PARK STATION 0.278 0.100 0.144 0.468 9,280 2,581 100 WILLOW STATION 0.149 0.026 0.110 0.198 4,145 618 101 WILSHIRE/NORMANDIE STATION 0.067 0.029 0.032 0.135 3,516 234 102 WILSHIRE/VERMONT STATION 0.335 0.062 0.241 0.444 12,865 4,306 103 WILSHIRE/WESTERN STATION 0.449 0.074 0.332 0.571 5,337 2,395 104 WOODLEY STATION 0.378 0.061 0.284 0.484 870 329 105 0.820 0.063 0.691 0.903 1,012 830

Source: LA Metro1 (2015) ; ‘Metro 2011 On-board Survey’ digital database (PTV NuStats-LACMTA (2012); estimated by author using STATA v.12 ‘svyset’ command and ‘subpopulation’ calibration; see Appendix A)

Notes: 1-‘LA-Metro Fiscal Year 2011-2012 (Jun 2011 – July 2012) Fixed Guideways Line and Station Boardings Report (average day and annual)’; via email (pers. comm. John Stesney, Transportation Planning Manager III, Systems Analysis & Research, Los Angeles County Metropolitan Transportation Authority; via email July 27, 2015) 2-Estimated from ‘Los Angeles County Metropolitan Transportation Authority (Metro) System-Wide On-Board Origin-Destination Study: Draft Final Report (December 2011)’; weighting and expansion protocol for station-level estimates followed survey’s protocol and methodological adjustments based on agency’s consultant ‘NuStats’ (Austin, TX) ‘Memo’ via email (pers. comm., Sujin Hong, Senior Data Analyst, and Ryan McCutchan, Project Manager at PTV NuStats, Austin, TX via email; November 02, 2015). Survey database cleanup, merging of several existing variables from ‘Summary” and “Waypoint” files, and creation of new fields was required to perform station-level calculations. All access mode distributions and estimated counts were estimated using Stata v.12 ‘svyset’ command for handling complex survey designs and ‘subpop’ calibration (see Appendix A). 3- In year 2012 there were a total of 113 rapid-transit stations in the Los Angeles County. However, the 2011 survey did not register person-trips nor boarding observations for some street-level stops/stations associated with bus rapid transit service ‘Silver Line’ in or near downtown LA and some stations associated with BRT Orange Line were not operational during the period the survey was conducted (Orange Line extension from Canoga station to ) . Also four (4)additional adjacent stations were consolidated into two (2)stations as each pair operates and function as a single station (Spring St. / 1st Street and Spring St./ City-Hall; Indiana 1st St. and Indiana Station). As such 105 rapid-transit stations are included in this table and in the study’s database for calculations.

24

Figure 2.7 LA Metro Annual Average Station Weekday Boardings1 (FY 2011-2012) and Access Mode Share Estimates2 Source: LA Metro1 (2015) and L.A. Metro On-board Survey digital database2 (PTV NuStats- LACMTA (2012); estimated by author using STATA v.12 ‘svyset’ command and ‘subpopulation’ calibration; see Appendix A) 25 General Socioeconomic Profile of Rapid-Transit Patrons that Access via Bus

An abundance of literature related to travel behavior in the United States evince a strong link between lower-income levels, racial and/or ethnic minority status, lower mobility levels, and transit travel, particularly in bus mode (Pucher and Renne 2003). It is not clear however if a similar association exists for patrons that access rapid-transit service via bus, as compared to other access modes (car, walk, and bicycle) or to bus-only riders. Based on estimates from Los Angeles on-board person-trip survey data (PTV NuStats-LACMTA 2012) key socio-economic attributes of riders were estimated, compared, and hereby discussed in order to delineate a general profile of this transit market segment.

$100,000

$90,000 $86,270.00

$80,000

$70,000

$60,000

$47,025.09 $43,602.03 $50,000

$40,000 $29,707.08

$28,405.32 $25,482.44

$30,000 $17,289.55

$20,000 $12,499.50

$10,000

$-

Figure 2.8 Estimated Annual Mean (Gray) and Median (Red) Household Income By Access Mode to Rapid-Transit Service (2011). Sources: ‘Metro 2011 On-board Survey’ digital database (PTV NuStats-LACMTA (2012); estimated by author in STATA v.12 using ‘svyset’ command and ‘subpopulation’ calibration; LA- MSA (Los Angeles Metropolitan Statistical Area) values based on US Census ACS 2011 5yr data. 26 As registered in Figure 2.8, patrons that use bus services to access rapid-transit stations report the lowest mean annual household income as compared to patrons that walk, bike, or drive to stations, and the lowest median value as compared to bus-only riders and general metropolitan population (LA-MSA). Yet their mean value is higher than that of bus-only riders ($25,482 vs. $17,289). In regards to driving license tenure, which is an important mobility factor in an automobile-oriented region, the survey registers a high proportion (65.3%) that do not have tenure; in terms of race/ethnicity a majority self-identify of Hispanic ethnicity at a rate larger than that of their share for the entire metropolitan population; and patrons are mostly Non-White (92.7%). The employment rate, whether full- or partially-employed, is relatively high at 68.9% with students registering the second largest share with 16.4% (Figure 2.9).

Based on these general traits the majority of rapid-transit riders that access via bus can be characterized as working-class and transit-dependent, with relatively lower mobility levels and lower household incomes as compared to those that access via other modes and as compared to the rest of the metropolitan population, with the exception of those that only travel by bus. It is hereby evinced that multimodal transit service is a critical component for their well-being and sustenance given the high rate of driving license non-tenure. An analysis of trips purpose by origin-place vs. destination-place suggests the relative importance of commuting to/from work for this important segment of the rapid-transit market (Figure 2.10).

A more nuanced ‘Home-Based’ vs. ‘Work-Based’ trip origin-destination distribution analysis reveals a greater variety of destinations for home-based trips that register a notable 40.28% of non-work related destinations versus 59.72% of work and work-related destinations (Figure 2.11). Conversely the surveyed one-way work-based trips reflect a more constrained activity destination pattern where the vast majority of trips end at home (91%). These patterns most likely reflect typical weekday job commutes and the fact that most surveyed work-based trips took place in peak-afternoon time, which is when most workers return home from their jobs.

27

Employment Status Race / Ethnicity

80,000 80,000 3.0% 0.7% 70,000 9.9%, Retired Native 70,000 Unemployed American 60,000 60,000 20.6% 16.4%, Black 50,000 Student 50,000 8.3% White 40,000 21.6%, 40,000 Employed part-time 30,000 30,000 60.3% Hispanic 20,000 20,000 47.2%, Employed 10,000 full-time 10,000 8.7% Asian 0 0

Driving License Tenure

80,000

70,000

60,000

50,000 No 40,000

30,000

20,000

10,000 Yes

0

Figure 2.9 Estimated General Socioeconomic Attributes of Rapid-Transit Patrons that Access Via Bus (2011). Source: ‘Metro 2011 On-board Survey’ digital database (PTV NuStats-LACMTA (2012); estimated by author in STATA v.12 using ‘svyset’ command and ‘subpopulation’ calibraiton).

28 College or Shopping Origin Place Pick up / University 2.30% Medical Drop Off (Student) 2.12% Someone 6.68% (Post- Coded) 0.07% Hotel (guest only) 0.19% Work or Work Related Airport (Airline Civic, 34.00% Passenger Religious, or Only) Personal 0.08% Business 0.62% School (K- 12) (Student) 2.33%

Social or My home recreational 46.67% 4.86%

Pick up / College or Destination Place Drop off University Shopping Medical Someone (Student) 3.87% 2.09% (Post- 5.60% Coded) 0.08% Hotel (guest only) 0.51% Work or Work Related Civic, 29.81% Airport (Airline Religious, or Passenger Personal Only) Business 0.03% 1.77%

School (K- 12) (Student) 2.39%

My home 45.18% Social or recreational 8.67%

Figure 2.10 Estimated Activity Distribution Based on One-Way Person-Trip Origin Vs. Destination Place for Patrons that Accessed Rapid-Transit via Bus (2011). Source: ‘Metro 2011 On-board Survey’ database (PTV NuStats-LACMTA (2012); estimated by author in STATA v.12 using ‘svyset’ command and ‘subpopulation’ calibration). 29

Home-Based Trips | Destinations

School (K-12) (Student) 4.64%

Social or recreational 12.85%

Work Hotel (guest only) or 0.96% Work-related Other College or University 59.72% 40.28% (Student) 10.53% Shopping 4.66% Medical 3.92% Civic, Religious, or Personal Business 2.58%

Work-Based Trips | Destinations

Work or Work Related 2.22% School (K-12) (Student) 0.31% Social or recreational My home Other 2.70% 91% 9% College or University (Student) 1.01% Shopping 2.06% Civic, Religious, or Personal Business 0.51%

Figure 2.11 Estimated Home-Based Trips vs. Work-Based Trips Destination Activity Distribution of Patrons that Accessed Rapid-Transit via Bus (2011). Source: ‘Metro 2011 On-board Survey’ digital database (PTV NuStats-LACMTA 2012; estimated by author in STATA v.12 using ‘svyset’ command and ‘subpopulation’ calibration).

30 CHAPTER 3

AN ASSESSMENT OF THE INFLUENCE OF BUS NETWORKS ON RAPID-TRANSIT STATION-LEVEL BOARDINGS

Introduction

As noted in Chapter 1 (‘Introduction’, p.3) access to rapid-transit service is an important and indispensable component for achieving rapid-transit cost-efficient patronage, access to jobs equity, and potentially increasing overall transit share in metropolitan regions (Sivakumaran, K. 2012; Kim et al. 2007; Chandra et al. 2013; Mees 2014; Tal et al. 2016; Boarnet et al. 2017).

Research on modal access to rapid-transit service is mostly framed within discrete choice models, and some do not include bus mode (Korf and Demetsky 1981; Mahmoud 2015; Wen et al. 2012; Chie-Hua 2012; Guan et al. 2007; Park et al. 2014). More recent studies have implemented direct-demand modeling (DDM) with limited success for bus mode (Coffel, K. et al. 2012). As noted earlier in this manuscript bus access to rapid-transit service represents on average a non-trivial 19.10% system-wide access mode share in large U.S. systems and registers a highly variable range at both line- and station-level (Figures 1.1, 1.2, 2.5b, 2.7; Tables 2.1, 2.2, 2.3). Bus access potential in supporting multi-modal rapid-transit networks, particularly in sprawling, disperse, and/or poly-centric urban agglomerations warrants further examination (Coffel, K. et al. 2012).

In the past century many U.S. cities have evolved from dense monocentric to lower- density polycentric structures, although traditional centrally-located employment centers (i.e. CBD’s) still exert significant influence in the new metropolitan landscapes (Ihlandfeldt 1995) and most small and medium-sized cities retain monocentric structures (Arribas-Bel & Sanz- Gracia, 2014). Yet a recent longitudinal study of urban sprawl in large urbanized areas in the U.S. reports that from year 2000 to 2010 compactness decreased and sprawl increased slightly, with some exceptions documented for some areas (Hamidi & Ewing. 2014). Emerging metropolitan poly-centricity and on-going sprawl are particularly relevant phenomena to the issues that motivate this study as these two emerging spatial and land-use patterns could increase the potential demand for bus-access to rapid-transit stations in the future, and possibly increase 31 the transfer rate within multi-modal transit systems as well. As suggested in Figure 2.6, strong correlation between this particular travel behavior and metropolitan spatial dimensions of sprawl, dispersion, and poly-centricity measures is evident and merits further investigation.

Both job decentralization (poly-centricity and/or dispersal) and sprawl trends have characterize many U.S. settlements in past decades (Ewing & Hamidi 2014; Lee 2007). It would be expected that bus access share to rapid-transit networks would be significant and notable in contemporary urban agglomerations that exhibit these spatial characteristics. Surprisingly large transit agency reports, on-board rapid-transit surveys, and other transit planning technical reports are scant or silent about bus access to rapid-transit service, particularly at the station-level. A few exceptions exist. Boston’s MBTA and San Francisco’s BART periodical access mode station- level reports are exemplary. Yet bus access share in these two cases are notably low. Consequently station-level statistics are for the most part lacking or difficult to obtain for a majority of large ply-centric cities’ multimodal transit systems.

A general picture of system-wide bus access trends for nine large systems in the United States was constructed from the limited data this author was able to obtain (Figure 1.2, Appendix B). Long-term trends of bus access share to rapid-transit services in the United States reveal a mixed pattern where some cities exhibit drastic decline (San Francisco and Boston), others report sustained increase (Chicago), whilst others register stable trajectories within the period for which data was available (Washington DC; Figure 1.2). For other cities long-term access-mode distribution data was not available or non-existent, in particular and ironically for those that register the highest system-wide bus transfer rates (Los Angeles, Miami, and Atlanta).

Several factors could be influencing these modal access trends, including socio- economic, private automobile availability, increase in use of access modes other than bus, infrastructure investments, or and/or changes in service and quality of bus transit. Differential metropolitan structures, sprawl, and/or decentralization trends could also be influential yet data and research on these topics is limited or non-existent.

32 In rapid-transit patronage studies that rely on direct-demand models (DDM) the potential influence of bus network connectivity typically relies on an often-used basic supply-side count of bus routes that could interact and potentially bring riders to rapid-transit stations. Although this control variable typically yields positive and statistically significant results the measure neglects individual bus route service levels that could result in diverse aggregate station connectivity amongst stations that register an equal number of bus route connections. Further to this, no previous rapid-transit ridership study has incorporated demand-side bus feeder measures (i.e. bus access transfer counts at rapid-transit stations) that may represent a more direct and valid measure of bus feeder network influence and its effects on rapid-transit ridership.

This study aims to shed light on some of these issues by incorporating more nuanced supply-side bus network connectivity instruments, and survey-based estimates of bus access counts at station-level as a key demand-side factor within multivariate regression models of average weekday station boardings. Also, a multi-dimensional built-environment and urban design measure for areas surrounding rapid-transit stations will be implemented in the models as an exploratory exercise.

Research Questions

Three research questions guide this first study: 1) How important are bus feeder networks to rapid-transit ridership in large, dispersed, and decentralized metropolitan regions in the U.S.?; 2) Do supply-side measures of bus network connectivity that incorporate distinct route service levels improve the predictive power of station-level direct-demand ridership models?; and 3) Do demand-side bus transfer counts as a measure of bus network connectivity improve direct- demand station-level model explanatory power?

This remainder of this chapter is structured in four sections that explain the research design, discuss results, and considers policy implications. Section 1 presents a literature review of transit ridership determinants and station-level direct-demand ridership models pertinent to this study and unit of analysis; Section 2 discusses research design as well as research methods; Section 3 presents model specifications and results; Section 4 presents conclusions, policy

33 implications, and future areas of inquiry based on results and analyses; the final Section 5 addresses research limitations.

Literature Review

Literature related to transit ridership determinants and specific to rapid-transit patronage is discussed within the emerging research methodology based on station-level direct-demand modeling (DDM). These fields capture key theoretical and methodological aspects that inform this investigation, its quantitative methods, and inform model specifications for both key and control variables. It also helps in identifying potential statistical remediation protocols for previously identified methodological shortcomings in transit ridership modeling and land- use/travel-behavior (LU/TB) research.

Determinants of Transit Ridership. An extensive literature on the factors that influence transit ridership exist, attending to different modes, and reflecting the importance that scholars, policymakers, and planners place on this topic (Ramos et al., 2015a,b). Taylor and Fink (2003) organize transit ridership factors in two categories: external factors that are beyond the direct control of transit planners and managers; and internal factors, which are susceptible to influence by transit planners and managers. Most transit scholarship has emphasized the overwhelming importance of socioeconomic factors (i.e. income levels, automobile availability) and built environment factors (i.e. land-use and urban design) as key external influences on ridership while acknowledging the critical roles played by internal factors such as fare and service level decisions made by transit officials (Balcombe et al., 2004; Cervero 2002; Mees 2010; Walker 2011). Transit ridership is thus susceptible to multiple external and internal factors that should be simultaneously considered when trying to understand and/or clarify ridership determinants, patronage, and assess combined land-use/transportation policy options.

Research has consistently found that higher population and employment densities, more mixing of land use types, more walkable environments, lower levels of vehicle ownership, and lower unemployment rates (more employed residents) are associated with higher numbers of transit trips (Cervero 2001; Kim et al. 2007; Guerra and Cervero 2011; Chakraborty and Mishra 2013; Ramos and Brown 2016). Researchers have also pointed to the roles of lower fares

34 (Balcombe et al. 2004; Guerra and Cervero 2011; Taylor and Fink 2003), more frequent service (Brown and Thompson 2008a,b; Guerra and Cervero 2011; Balcombe et al. 2004; Ramos and Brown 2016), and better service coordination (Brown and Thompson 2008; Currie et al. 2011; Mees 2010; Walker 2011) in promoting more transit usage.

These results have been found at a variety of geographic scales (international, national, statewide, metropolitan, system, route, station, and/or stop) and using a variety of methodological approaches such as case study, multinomial logit, geographically weighted regression, multi-level regression, negative binomial regression, and/or bootstrapped OLS regression, among others, in locations throughout North America, Europe, and Australasia (Tables 3.1 - 3.6).

Direct-Demand Models. Studies that focus on station and/or stop-level boardings are of special interest in this investigation. It is at the station-level where both local and metropolitan supply and demand forces meet. This makes transit portals (i.e. rapid-transit station and/or bus stops) an ideal spatial unit for the analysis of transit ridership factors and potential effects of multimodal network attributes, such as bus networks connectivity at stations; land-use and built- environment characteristics; service area socio-economic attributes, among other key factors. Bus access to rapid-transit service is one such behavior where the users experience a diversity of factors at various spatial scales during his/her multi-modal trip, including local scale characteristics, metropolitan scale transit network attributes, and distinct transit service levels at both local and metropolitan scales.

Most recent station-level studies rely on direct-demand multivariate regression using cross-sectional data. This flexible statistical model is considered an effective, more appropriate, and less costly approach to ridership prediction and policy analyses vis-a-vis the more complex, timely, and costly 4-step model (Cervero 2006; Cervero et al. 2010; Kuby et al. 2004; Gutierrez et al. 2011; Duduta 2013; Zhao et al. 2014; Durning & Townsend 2015; Chen & Zegras 2016).

As noted by Cervero (2006) the 4-step demand forecasting model was not intended to assess finer-grain neighborhood level and/or land-use attributes on travel behavior, but “rather to

35 guide regional highway and transit investments”. Traffic Analysis Zones (TAZ) are the 4-step’s model primary unit of analysis. These range in size from census block groups to census tracts that are better suited for macro- and meso-scale studies (i.e. corridors, sub-regions, metropolitan areas, and states) rather than the finer grain of transit portals and their associated pedestrian service areas (Ped Shed).

Other relative weaknesses associated to the traditional 4-step travel demand forecast model have been noted by several scholars that include matters related to relative accuracy, travel input data (i.e. use of old surveys and few transit trips), sensitivity to land-use (regional focus), institutional barriers and high expense (Gutiérrez et al. 2011; Zhao et al. 2013;); more emphasis on traffic counts rather than transit loadings (Gutiérrez et al. 2011); among others (for a comprehensive review and discussion see Richmond 2005 and Cervero 2006).

Furthermore several scholars highlight advantages of direct-demand models vis-à-vis 4- step protocols (Kuby et al. 2004; Gutierrez et al. 2011; Cervero 2006; Zhao et al. 2014; Durning and Townsend 2015). DDM is a simplified approach to forecast demand in one direct equation that incorporates aspects of ‘trip demand’, ‘mode choice’, ‘trip distribution’, and ‘traffic assignment’ features; it has the ability to simultaneously evaluate effects of a large number of factors that may include both supply and demand variables, and the possibility to incorporate finer-grained land-use and urban design attributes around stations (Durning and Townsend 2015). In addition, DDM models are flexible and easier to understand to wider audience, can handle both numeric and binary (0-1 dummy) conditions, and allow evaluation of variations in parking, feeder bus service, station spacing, and transit speed and frequency for alternative route- alignment/stop-location analyses (Walters and Cervero 2003; Gutierrez et al. 2011). DDM estimated parameters’ elasticities can also inform urban planning policies related to land-use densities and design (i.e. TOD: Transit Oriented Development; Cervero 2006; Zhao et al. 2014); among others.

It is also clear that DDM cannot replace the 4-step regional scale model, but they can complement each other to account for both macro- and micro-level factors in multi-scalar contexts and problems (Walters and Cervero 2003; Cervero 2006 ; Gutierrez et al. 2011). As

36 noted by Cervero (2006) and Duduta (2013), DDR models typically rely on small sample sizes as they rely on stations as the unit of analysis. This results in ‘degree of freedom’ constraints that restrict the number of explanatory variables that could be specified and ‘may preclude the inclusion of useful interactive terms’ (Cervero 2006). As result they are recommended as ‘sketch-planning’ tools that provide quick order of magnitude information (Cervero 2006; Duduta 2013). However, recent land-use/travel-behavior studies that focus on station-level analysis compensate the relative low number of observations with bootstrapped (resampling) protocols that provide better confidence intervals on estimated parameters and associated confidence levels (see Durning and Townsend 2015; Chen and Zegras 2016).

In more recent DDM models the dependent variable has been gradually diversifying from monthly boarding averages to daily averages, average weekday boarding’s, peak- and non-peak periods, weekend boarding’s, station-to-station O-D flows by time of day, among others. Their predictive potential have also been increasing as researchers continue to explore and improve on methodological protocols, statistical models, and theoretical frameworks (Tables 3.1 – 3.6).

DDM in Rapid-Transit Ridership Studies. The most commonly cited station-level ridership studies are those by Parsons Brinckerhoff et al. (1996) and Kuby et al. (2004). Both extend Pushkarev and Zupan (1977) seminal work on transit ridership, land-use characteristics, and cost-effectiveness in a CBD-oriented rail-transit corridor level study; and on Cervero et al.’s (1995) study on rail access modes and catchment areas for San Francisco’s BART system. The first two studies focus on boardings for U.S. light-rail (LRT) system stations, with the former study focused on 261 non-central business district (CBD) stations in 11 U.S. cities and the latter study considering 268 stations, both inside and outside CBDs, in nine U.S. cities. These studies find positive relationships between boardings and employment density within 800 meters of a station, the presence of park and ride facilities at a station, the number of connecting bus lines at a station, and the station’s network function as either a terminal or transfer station.

Kuby et al. (2004) also found positive relationships between station level boardings, population, and airports located within 800 meters of the station, as well as links to international borders, and the non-significance of CBD-located station dummy. It is also important to

37 recognize context-specific situations related to special activities generators in proximity to transit service as noted by Folleta et al. (2013) and Ramos & Brown (2016). Kuby’s study also introduced a methodological improvement for stations service area boundary definitions based on a shortest network distance raster-approach (for a detailed discussion see Upchurch et al. (2004)).

Other scholars that have studied rapid-transit systems around the world that include heavy rail/metro (HRT), light-rail systems (LRT), and bus rapid transit (BRT) usually find results that are consistent with those found in the works cited above. Sohn and Shim’s (2010) study of heavy-rail transit (metro) ridership at 251 stations in Seoul and both Gutierrez et al. (2011) and Cardozo et al. (2012) studies of metro ridership at more than 150 stations in Madrid each found positive relationships between boardings and both employment levels within 800 meters of the station and the number of connecting bus routes at the station. A study by Ozbil et al. (2009) on ridership at 219 heavy rail and LRT stations in Atlanta, Chicago, and Dallas also found positive relationships between boardings and the number of bus and rail connections that were available at a station, the population density, the walkability of the street system, and the availability of park and ride facilities.

The study by Gutierrez et al. (2011) on Madrid’s metro stations also extends and presents methodological improvements related to the spatial definition of station service areas and potential demand calculations by introducing distance-decay functions in a system of regression models. This approach recognizes and captures the “tendency of patronage to decay with walking distance to station” within the delimited pedestrian service areas. Their study also presents quantitative evidence of greater validity of network distance-based station buffer delimitations vis-à-vis Euclidean distance buffer generation protocols previously used in this line of research. In their study the authors also introduced a nodal accessibility measure based on Hansen (1959) potential model to capture stations’ network accessibility levels, which register the largest relative magnitude and significance of all their final model predictors.

Both Kuby et al. (2004) and Gutierrez et al. (2011) studies used the generally accepted 1/2mile buffer distance parameter across the entire set of stations and do not consider the

38 possibility of distinct service area boundaries across stations and/or across rapid-transit modes which might reflect variance according to local context-specific conditions and/or service levels as noted by Jiang et al. (2012).

More recent rapid-transit patronage studies (2013-2017) that utilize DDM have expanded the list of transit modes, explanatory vectors, and quantitative methods in explaining ridership at the station-level. Some studies have looked into heavy-rail (metro/subway; Guerra et al. 2012; Duduta 2013; Zhao et al. 2014; Durning and Townsend 2015; Chen and Zegras 2016; Renee et al. 2017) , light-rail (Guerra et al. 2012; Dill et al. 2013; Ramos & Brown 2016; Durning and Townsend 2015; Chen and Zegras 2016; Renee et al. 2017), and bus rapid transit (BRT; Guerra et al. 2012; Duduta 2013; Renee et al. 2017). DDM and probit choice models have also been used in analyses of slower ‘street-transit’ technologies as well, focusing on bus (Dill et al. 2013; Chakour & Eluru 2016) and streetcar (tram) stops (Foletta et al. 2013; Ramos & Brown 2016).

Rapid-transit and bus systems included in these more recent studies feature new case- study cities in Asia and the Americas (Tables 3.1 – 3.6). Quantitative methods have also evolved from OLS regression to more nuanced and complex generalized linear regression and probit models, including more robust negative binomial linear regressions that account for non-zero and non-normality of dependent count measures and boarding count distribution (Ramos & Brown 2016), multiplicative models for analyzing inter-station flows (Zhao et al. 2014), multi-level regression to account for clustering effects around distinct geographical sampling areas, and ordered response probit using a composite marginal likelihood (CML) inference approach (Chakour & Eluru 2016).

Resampling strategies such as bootstrapping protocols have also been implemented in various studies to address the relatively low number of observations and buttress confidence intervals and inferential analyses (Durning and Townsend 2015; Chen and Zegras 2016). Instrumental variables in Two-Stage least squares (2SLS) methods have also been applied in DDM rapid-transit studies to address potential endogenous relationships among variables, often related to transit supply and demand co-dependence, and improve estimates (Taylor et al. 2009; Ramos & Brown 2016).

39 Dependent variables have also diversified to explore temporal changes in patronage to include AM and PM travel analyses as well as differences in origin vs. destination locations (Zhao et al. 2014), peak and non-peak patronage levels (Chen & Zegras 2016), and stratified models of stops based on ridership level ranking (i.e. high, medium, low; Chakour & Eluru 2016). These have yielded more nuanced understandings of ridership factors along temporal, spatial, demographic, socioeconomic, service levels, network connectivity, and LU/BE vectors, among others. All these DDM station- and/or stop-level studies are based on cross-sectional data given the difficulty of obtaining longitudinal data, and mostly focus on geographies surrounding station portals. Yet none have addressed the potential influence of LU/BE factors in peripheral geographies associated with feeder bus stops nor explored more nuanced measures of bus network connectivity beyond the basic count of connecting routes.

The original set of three vectors of explanatory ridership factors in station-level DDM models, which include socioeconomics (SE), transit service levels (TS), and land-use/built- environment (LU/BE) attributes within stations’ pedestrian catchment areas has been reinforced and extended with new ‘network structure’ and ‘destination accessibility’ vectors (NT) as related to stations’ network function (Parsons Brinckerhoff 1996; Kuby et al. 2004, Ozbil et al. 2009; Sohn and Shim 2010; Gutierrez et al. 2011; Guerra et al. 2011; Cardozo et al. 2012; Foletta et al. 2013; Duduta 2013; Dill et al. 2013; Ramos & Brown 2016; Zhao et al. 2014; Durning & Townsend 2015;Chen & Zegras 2016), stations’ relative network destination accessibility measures (Kuby et al. 2004; Chen & Zegras 2016; Renee et al. 2016), and special activity generators within 400m radius of station portals (Foletta et al. 2013; Ramos & Brown 2015; Zhao et al. 2014).

The importance of inter-station spacing has also resurfaced in recent studies (Parsons Brinckerhoff et al. 1996; Foletta et al. 2013; Ramos & Brown 2015) as well as mode based differences where heavy-rail usually averages more patronage as compared to other modes after controlling for other relevant factors (Guerra et al. 2011; Chen & Zegras 2016; Renee et al. 2017).

40 These more recent studies further buttress previous evidence on key station-level ridership factors (Tables 3.1 – 3.6): population levels, employment levels, combined population and employment levels, and their respective densities around stations (Parsons Brinckerhoff 1996; Kuby et al. 2004, Ozbil et al. 2009; Sohn and Shim 2010; Gutierrez et al. 2011; Guerra et al. 2011; Cardozo et al. 2012; Foletta et al. 2013; Duduta 2013; Dill et al. 2013; Ramos & Brown 2016; Zhao et al. 2014; Durning & Townsend 2015; Chen & Zegras 2016).

In addition transit network connectivity measures and functional categorizations, represented by the number of bus and/or rail lines connections per station and by transfer or terminal station categorical status consistently register highly significant and positive association with boardings (Parsons Brinckerhoff 1996; Kuby et al. 2004, Ozbil et al. 2009; Sohn and Shim 2010; Gutierrez et al. 2011; Guerra et al. 2011; Cardozo et al. 2012; Foletta et al. 2013; Duduta 2013; Dill et al. 2013; Ramos & Brown 2016; Zhao et al. 2014; Durning & Townsend 2015; Chen & Zegras 2016).

Other consistent supply-side factors that emerge as highly significant and positively related to ridership are car park & ride facilities, usually captured as a dummy variable or number of parking spaces (Parsons Brinckerhoff 1996; Kuby et al. 2004, Ozbil et al. 2009; Sohn and Shim 2010; Gutierrez et al. 2011; Guerra et al. 2011; Cardozo et al. 2012; Foletta et al. 2013; Duduta 2013; Dill et al. 2013; Ramos & Brown 2016; Zhao et al. 2014; Durning & Townsend 2015; Chen & Zegras 2016); and bicycle park & ride facilities (Zhao et al. 2004). Location of station within a CBD district has emerged as significant in some studies (Duduta 2013; Dill et al. 2013; Zhao et al. 2014) but not in others (Kuby et al. 2004). This inconsistency may reflect the effect of distinct context-based sprawl and/or poly-centricity levels.

In regards to the potential influence of LU.BE (land-use and built-environment) factors on station ridership, which has been an area of research receiving notable attention, previous studies reveal that various measures of land-use mix register statistically significant and positive relationships (Gutierrez et al. 2011; Dill et al. 2013; Zhao et al. 2014; Durning & Townsend 2015; Chen & Zegras 2016; Renee et al. 2017). Entropy measures, share of commercial areas, and job-population balance index are some of the operationalizations used to capture the

41 potential effect of this particular land-use attribute. The other LU.BE attribute that frequently emerges as significant in station-level DDM ridership models is intersection density, usually considering the share of 4-way intersections within pedestrian catchment areas that serves as a measure of pedestrian-friendliness (Durning & Townsend 2015; Chen & Zegras 2016; Renee et al. 2017 ). Average block size has also been implemented as a reasonable proxy for pedestrian- friendliness in a few studies with statistical significant outcomes (i.e. Ramos & Brown 2016).

More recent studies incorporate more nuanced multi-dimensional walkability indices in their ridership models, such as the Walk Index computed by Chen & Zegras (2016) and the Walk Score® walkability measure incorporated in Renee et al. (2017) ridership model, both of which emerged as significant and positively associated with ridership outcomes at the station-level and stop-level. These two similar measures address the quality of pedestrian environment around stations and incorporate similar dimensions related to land-use mix, share and/or accessibility to commercial areas and/or employment, and intersection density (Chen & Zegras 2016; Renee et al. 2017). In the case of the Walk Index it incorporates residential density as well (Chen & Zegras 2016).

However, neither of the two previous measures incorporates one key “D” variable that is strongly associated with walking to transit: distance to transit (De Bourdeaudhuij et al. 2003; Schlossberg et al. 2007; Agrawal et al. 2008; Ewing & Cervero 2010; Park et al. 2014) either as a component of the multi-dimensional walkability indexes nor as an independent explanatory variable in their models. As noted in the more general LU.BE / TB literature review, a statistically significant yet relative weak effect of land-use and built-environment factors relative to network attributes, socio-economics, and transit service levels (Ewing & Cervero 2010; Stevens, M. 2017) is also found in most station-level DDM studies reviewed for this manuscript as related to boarding levels.

This literature review also reveals important significant factors that are negatively associated with station-level ridership and that register relatively large coefficient estimates. From the socioeconomic vector, average household auto availability emerges as a significant deterrent of transit ridership levels (Dill et al. 2013; Chen & Zegras 2016; Ramos & Brown

42 2016). In the U.S. context this result should not be surprising as transit travel has been described as an inferior good in the urban mobility market (McLeod Jr. et al. 1991).

Fare, when incorporated in models, also results as expected in lower ridership levels (Ramos et al. 2015a; Ramos & Brown 2016). Time related measures, including headway, transfer time, and travel time have also resulted in significant effects associated with lower propensity for transit ridership when implemented in choice models (Dill et al. 2013; Zhao et al. 2004). These latter service-level and direct-cost (monetary) factors are consistent with utilitarian travel behavior theoretical frameworks where they operate as cost components of the generalized cost of travel (GCT).

In sum, rapid-transit station-level DDM models have increased in number, explanatory power, and sophistication. Among the various consistent determinants found to be positively associated with ridership is bus network connectivity. However a rather blunt measure based on the number of bus routes is still being used, and discussion of the importance of bus-rail interactions in regards to rapid-transit ridership and performance lags behind those focused on land-use and built-environment (urban design) factors around rapid-transit stations.

In large poly-centric urban agglomerations, such as Los Angeles, a diversity of bus services exist that interact with rapid-transit services. This great variety of bus service levels and operating windows are not captured in simple bus route count measures of network connectivity. That is, rapid-transit stations with equal number of connecting bus routes might not enjoy equal levels of bus network connectivity. Hence a more valid measure of bus network connectivity is warranted that might yield better understanding of bus-rail interactions and improve DDM models predictive power.

Furthermore, to this author’s knowledge no demand-side measure for bus feeder networks’ influence has been implemented in station-level DDM studies (i.e. bus transfer (access) counts at station level). The limited availability of data related to bus-rail transfers in part might explain this lack of attention. This study aims to fill these gaps by developing two distinct measures of bus network connectivity and incorporating them in multivariate models: 1-

43 an aggregate supply-side measure of bus service levels (weekday vehicles per day) at the station level based on multiple bus agencies’ GTFS data and route schedules in a new integrated GIS geodatabase; 2- a demand-side measure based on estimated bus-rail transfer counts (access) using bus-rail onboard person-trip survey information.

Research Design & Methods

The research design for this first investigation is based on a single case-study of the Los Angeles metropolitan rapid-transit system and its associated interacting multi-agency bus networks. As noted in the previous chapter it is a purposefully selected case geared to inform statistical analyses and to address the three research questions posed for this investigation; to be relevant to large poly-centric urban agglomerations in the U.S. context; and for which pertinent person-trip and associated spatial observations, socioeconomic, land-use, network, and multimodal transit service data are available.

Definitions. See Chapter 1 (p.14, ‘Definitions’) for a detailed discussion of key terms and general GIS service area delimitation protocols.

Method. Methods are quantitative, inferential, and rely on a set of DDM multivariate generalized linear equations whose specifications are informed by travel theory and the preceding literature review. LA Metro’s rapid-transit stations along HRT, LRT, and BRT lines that were operational during year 2011 (N=105), which is the year that the on-board bus-rail survey was conducted, serve as the unit of analysis. Survey-based person-trip observations and mode access shares were adjusted on fiscal year 2011-2012 station-level ridership as facilitated by LA Metro (2015); as such the multi-agency bus and rail service data and GTFS feeder bus route data that inform stations bus connectivity levels was collected based on year 2012 (or closest year available). Likewise year 2012 (or closest available) land-use, social-economic, and employment data inform stations service area attributes that are part of the explanatory set of variables.

Rapid-transit station’s portal-based pedestrian catchment areas follow a mutually- exclusive clipped Thiessen-polygon delimiting protocol implemented in ArcGIS software. Buffer

44 radius specification is based on the 95th percentile Euclidean walk distance value from trip origin to first boarding rapid-transit station based on PTV NuStats-LACMTA’s (2012) bus-rail on- board survey spatial data, excluding extreme outliers (>2.5 SD). The remaining percentile values (95th to 99th) are considered outliers in this study. This protocol yields a standard buffer distance parameter for rapid-transit stations of 0.74mile, slightly larger than the typical 0.50mile (800mt) buffer used in a many preceding DDM rapid-transit studies but consistent with recent research that indicates more appropriate and larger buffer areas (Jiang et al. 2012; El-Geneidy et al. 2014).

Previous research on walking distance to rapid-transit stations and delimitation of station pedestrian catchment areas have documented that different transit service levels, transit technologies (modes), local built-environment context, and/or cultural factors yield statistically significant distinct station pedestrian catchment areas (Jiang et al. 2012; El-Geneidy et al. 2014). Nonparametric test of median and 95th percentile walk distance values result in statistically significant differences between HRT and LRT/BRT rapid-transit modes in Los Angeles rapid- transit system. As such, two distinct network distance buffers were implemented in the final model for HRT (0.61miles) and LRT/BRT stations (.76miles). These values are also larger than the typical .50mile parameter, which confirms previous empirical studies that indicate longer walking distances (Appendix C). Whenever multiple lines of different transit technologies coincide at a station the larger buffer value is used.

Estimation of Station-Level Bus Transfer (Access) Counts and Confidence Intervals Based on Survey Person-Trip Data. One key demand-side explanatory variable of interest in this study, the count of bus transfers (access) at rapid-transit stations, is not reported in PTV NuStats-LACMTA’s ‘On-board Bus and Rail O-D Survey Report’ (2012). Yet the survey’s rich person-trip spatial and route/stop sequence composition, sampling weights, and appropriate weight-expansion and adjustment protocol was used to identify and estimate simple population proportions and corresponding confidence intervals for each rapid-transit station (Table 2.3; see Appendix A for a detailed explanation). Station average weekday boarding figures provided by LA Metro for fiscal year 2011-2012 (based on APC data) allowed for calculation of station-level expansion factors, which are absent in Metro’s survey database as it is focused on line-level statistics and more general O-D patterns.

45 Logical parameters scripting in STATA was used to identify pertinent rapid-transit trips in the database (n=10,535) and to calculate the sample access mode proportions (share of bus, walk, car, and bicycle access events per station). Survey’s weighting and expansion protocol (as per PTV NuStats memo, 2016) was implemented with the new station-level expansion factors for total boardings and for bus transfer (access) count estimations. PTV NuStats-LACMTA’s (2012) complex two-stage survey design is accounted for in the calculation of population sub-group estimates and confidence intervals, using Stata’s svyset command and subpopulation (subpop) calibration, which considers survey design, sample clustering, and stratification as specified in the survey’s report.

Descriptive statistics for the relationship between bus access events and rapid-transit boardings in LA Metro’s transit system suggest a strong and significant positive association (Figure 2.6, p.22). However as noted in the literature review several factors have been associated with station-level rapid-transit boardings. Thus a more nuanced multivariate regression approach is implemented to more accurately assess bus feeder networks’ influence on rapid-transit station boardings whilst controlling for other known and hypothesized factors as identified in the literature and informed by the particular case context of Los Angeles metropolitan transit network and associated built-environments.

DDM Models. This investigation is based on quantitative inferential methods to assess the influence of various explanatory factors on rapid-transit station annual average weekday boardings (AWB), with a focus on bus feeder networks’ potential influence. A direct-demand modeling approach (DDM) to station-level ridership forecasting is implemented whose specification and form is influenced by the preceding literature review, data characteristics, and the particular context of the case.

Based on applied statistics literature (Cameron and Trivedi 2010; Hilbe 2011) and on the dependent variable’s non-negative integer and highly eschewed logarithmic distribution (see Figure 2.7) negative-binomial multivariable generalized linear regressions (NBREG) are specified, fitted, and tested in this study. As compared to the more traditional ordinary linear regression approach (OLS), which in this line of study often needs and relies on distortion of the

46 dependent variable to satisfy basic model assumptions, NBREG is considered a more robust model able to provide better unbiased parameter estimates, improve on confidence interval estimations for significance interpretations, and control for presence of heteroscedasticity without having to rely on outcome variable transformations (Cameron and Trivedi 2010; Hilbe 2011). It also allows for relatively easy estimation of elasticities for analysis and planning purposes.

The model result analyses rely on estimates significance, direction of association, and relative order of magnitude of key variables of interest as compared to controls in the final model. Because of the relative low number of observations in the final model (n=103) bootstrapped standard errors are specified for a more conservative assessment of significance results.

Models General Form, Specification, and Variables Description. The base model is informed by literature review of rapid-transit ridership determinants (Tables 3.1 - 3.6), DDM station-level research, and travel theory. Following and expanding on previous DDM studies four main vectors inform the set of explanatory variables (Eq. 0): 1-socioeconomic attributes of households and employment levels within stations’ pedestrian service areas (SE); 2-land- use/built-environment characteristics of stations’ service areas (LU.BE); 3-transit service levels (TS) of both line-haul and bus feeder services; and 4-bus and rapid-transit network attributes (NT). The outcome variable in this study is annual average weekday boardings at rapid-transit stations (AWB; FY 2011-2012, LA Metro).

AWB station = f (SE, LU.BE, TS, NT) Eq. (0)

The study analysis strategy relies on comparison of magnitude, direction, and significance levels of key explanatory variables of interest, controls, and models’ explanatory power between three sub-models (Model 1A, 1B, 1C). Each model incorporates a distinct bus feeder network connectivity measure. Model 1A incorporates the basic measure of bus network connectivity based on the count of bus routes associated with each station. Model 1B incorporates a measure of aggregate bus route service levels for each station based on a set of

47 captured routes from a custom geodatabase, and Model 1C incorporates a demand-side estimate of bus-rail transfer (access) counts for each station based on LA Metro’s 2011 on-board survey database.

The models negative-binomial multiplicative form (Eq. 1) and corresponding specified variables for alternative models (Eq. 1A, 1B, 1C) are presented below with a full set of documented dependent and independent variables. Their definitions, descriptive statistics, sources, and hypothesized signs are registered in Table 3.7. Model results and measures of fit including pseudo-R2, Akaike Information Criteria (AIC), and Bayesian Information Criteria (BIC) are reported in Table 3.8 and these in turn inform evaluation and comparison between models as per bus-network connectivity measurement hypotheses.

station β1 β2 β3 β4 βm AWB = X1 * X2 * X3 * X4 *… X n Eq. ( 1 )

station β1 β2 β3 β4 AWB = POP_101 * JOB_112 * Veh_HU3 * Perc_Hisp4 Eq. ( 1 [A, B, C] )

β5 β6 β6 * Wlk_SCR5 * AvgDIST_mt6 * SpeGEN6

β7 β8 β9 β9 β9 * RTv_pDAY7 *n_PKG8 * [ (A) Bus_Line9 , (B) Bus_Freq9 , (C) Bus_ACCe9 ]

β10 β11 β12 β13 * Commtr_L10 * Rail_K11 * Terminal12 * TrnfHub13

β14 β15 β16 βm * Fwy_ROW14 * OneWay15 * BRT_Silver16 * … Xn * e

A conservative approach to standard errors estimation via a bootstrapped re-sampling approach is implemented for the assessment of variables significance levels and policy implications due to a relative low number of observations at station-level vis-à-vis number of predictors.

48 Table 3.1 International Review of Rapid-Transit Station-Level Ridership Determinants (partial table): Research Methods and Demographic / Socioeconomic Factors 1996-2013

Parsons Kuby, Ozbil, Sohn and Gutierrez, Guerra, Cardozo, Foletta, Dill, Brinckerhoff et al. et al. Shim et al. et al. et al. et al. Duduta et al. Authors: 1996 2004 2009 2010 2011 2011 2012 2013 2013 2013

Modes of Interest Heavy Heavy Light- Rail, Heavy- Streetcar BRT, Light-Rail, Light-Rail Heavy-Rail Rail, LRT, Hevay-Rail Rail Light- Rail ( Tram ) Metro Bus BRT Rail Location United United United States States: States: Mexico: Portland, United Korea: Spain: United Spain: United States Chicago, Seattle, Mexico Rogue States Seoul Madrid States Madrid Dallas, Portland, City Valley, and Atlanta Tacoma Lane County, OR Number of Cases 11 9 3 1 1 20 1 3 1 3 supercolumn Dependent log DB AWB AWLB MB AWBA MB AWB AWB AWB Variable (ADB) 125 Number of 261 268 219 251 158 1449 190 67 Metro 8964 Observations 51 BRT Time Period of not September not not specified 2000 2007 Nov 2004 Nov 2004 2010 Fall 2008 Analysis specified 2009 specified CS-MV CS-MV CS-MV Methodology CS-MV CS-MV CS-MV CS-MV CS-MV CS-MV CS-MV SEM GWR GWR Station Buffer

Delimitation Power of Model (R2 for OLS; 1/df .51 to Pearson for 0.536 0.727 0.35 0.634 0.753 .75 to .80 0.57 .77 to .79 .53 to .69 .54 NBREG; pseudoR for MLM )

Population ✚ ✚ ✚ ✚ ✚ ✚ ✚ ✚ ✚ Population Density ✚ ✚ n.s. ✚ Employment ✚ ✚ ✚ ✚ ✚ ✚ ✚ ✚

Employment

Density Population + Employment Density % Nonwhite Workers ✚ College Enrollment n.s. ✚ ✚ PMSA Population n.s. Percent of PMSA

Employment ✚

DEMOGRAPHIC / DEMOGRAPHIC SOCIOECONOMIC Renters ✚ Household Income ✚ Household Average Vehicle ( ‒ ) Access

LEGEND: ✚ Positive and significant (‒) Negative and significant n.s. Not significant o Origin d Destination Not specified in final model

49 Table 3.2 International Review of Rapid-Transit Station-Level Ridership Determinants (partial table): Research Methods and Demographic / Socioeconomic Factors 2014-2017

Durning & Renee et Ramos & Brown Zhao, et al. Townsend Chen & Zegras al. Authors: 2014 2014 2015 2016 2017

Modes of Fixed Interest Guideway; Heavy- commuter Streetcar Light- Heavy-Rail Rail, Heavy-Rail, Light-Rail rail, ( Tram ) Rail Light-Rail heavy-rail, light-rail, BRT, ferry Location Canada: Calgary, Edmonton, United United China: United States: United Montreal, States States Nanjing Boston States Toronto, Vancouver

Number of 7 14 1 5 1 Cases supercolumn DTVL DTVL AWB station station 4 √AWB 4 √AWB 4 √AWB log ( MS- Dependent log 2 √AWB AWB AWB station- - to - - to - 4 √AWB AM Off- PM- TC ) Variable (AWB) Weekend level station station Peak Peak Peak Commute* AM PM Number of 2970 2970 475 432 55 342 120 120 120 120 66 4400 Observations pairs pairs 2009- 2009- 2009- 2009- 2009- Time Period 2012 2012 2011 2011 2012 2012 2010 2010 2010 2010 2010 2000-2010 of Analysis (var.) (var.) (var.) (var.) (var.)

CS-MV CS-MV CS-MV CS-MV CS-MV CS-MV CS-MV CS-MV CS-MV CS-MV log-log Methodology CS-MV NBREG NBREG ∏ ∏ BST BST BST BST BST BST CS-MLM

800mt Station 800mt (urban) .5mile .5mile .5mile .5mile .5mile 400mt 400mt 800mt .5mile Buffer Euclide 1000mt Network Network Network Network Network Euclidean Euclidean Euclidean Euclidean Delimitation an (suburban) exclusive exclusive exclusive exclusive exclusive Network Power of Model (R2 for OLS; 1/df .811 .803 to [ 1.12 ] [ 1.09 ] .958 .829 PM .803 .782 .796 .791 .804 .597 Pearson for AM .816 NBREG; pR for MLM )

Population ✚ ✚ ✚ ✚ ✚ ✚ ✚ ✚ ✚ ✚ Population

Density ✚ Employment n.s. ✚ ✚ ✚.d ✚.o ✚ Employment

Density n.s. ✚ Population + Employment ✚ Density % Nonwhite ✚ Workers College

Enrollment PMSA

Population Percent of PMSA Employment Renters Household

Income ✚ n.s. ( ‒ ) Household Average

Vehicle ( ‒ ) n.s. ( ‒ ) Access

50 Table 3.3 International Review of Rapid-Transit Station-Level Ridership Determinants (partial table): Land-Use / Urban-Design Factors 1996-2013

Sohn Parsons Kuby, Ozbil, and Gutierrez, Guerra, Cardozo, Foletta, Dill, Brinckerhoff et al. et al. Shim et al. et al. et al. et al. Duduta et al. Authors: 1996 2004 2009 2010 2011 2011 2012 2013 2013 2013

Land Use Mix.1

(entropy) n.s. ✚ ✚ Land Use Mix.2 ( %

manf, % com ) ✚ Land Use Mix.3 ( %

business/office ) Land Use Mix.4 ( %

residential ) Land Use Mix.5 ( jobs/population balance index ) Walkability Index.1 (Entropy + Res. Density

+ Comm. Site + Intersection density ) Walk Score® ( land mix,retail emp.,4-way int. density ) Special Activity

Generators ✚ Average Road Width Road Denisty ( linear mt.

w/n PCA ) Sidewalk Density Road Network.1 ( intersection density; % 4-

LAND USE LAND / DESIGN URBAN way ) Road Network.2 (

connectivity ) ✚ n.s. ✚ Block Size Floor Area ✚ Retail Employment

Density Sprawl Index (regional) Airports ✚ ✚ International Borders ✚ CBD (dummy) n.s. ✚ ✚ Normalized accessibility ( ‒ ) ✚ n.s. n.s. ✚ ✚ ( ‒ )

LEGEND: ✚ Positive and significant (‒) Negative and significant n.s. Not significant o Origin d Destination Not specified in final model

51 Table 3.4 International Review of Rapid-Transit Station-Level Ridership Determinants (partial table): Land-Use / Urban-Design Factors 2014-2017

Durning & Renee Ramos & Brown Zhao, et al. Townsend Chen & Zegras et al. Authors: 2014 2014 2015 2016 2017

Land Use Mix.1 (entropy) n.s. Land Use Mix.2 ( % manf, %

com ) ✚ Land Use Mix.3 ( %

business/office ) ✚ ✚.d ✚.o Land Use Mix.4 ( % residential ) ✚.o ✚.d ✚ Land Use Mix.5 (

jobs/population balance index ) ✚ Walkability Index.1 (Entropy + Res. Density + Comm. Site + ✚ n.s. Intersection density ) Walk Score® ( land mix,retail

emp.,4-way int. density ) ✚ Special Activity Generators ✚ ✚ ✚ ✚.d ✚.o-d educ/ent/shop educ educ-ent/shop Average Road Width ✚ Road Density ( linear mt. w/n

PCA ) ✚ ✚.o-d ( ‒ ) Sidewalk Density ✚ Road Network.1 ( intersection

density; % 4-way ) ✚ ✚ n.s. n.s. ✚ LAND USE LAND / DESIGN URBAN Road Network.2 ( connectivity ) Block Size ( ‒ ) n.s. Floor Area Retail Employment Density ✚ ✚ Sprawl Index (regional) n.s. Airports International Borders CBD (dummy) ✚ ✚.d ✚.o Normalized accessibility

LEGEND: ✚ Positive and significant (‒) Negative and significant n.s. Not significant o Origin d Destination Not specified in final model

52 Table 3.5 International Review of Rapid-Transit Station-Level Ridership Determinants (partial table): Transit Service Quality 1996-2013

Sohn Parsons Kuby, Ozbil, and Gutierrez, Guerra, Cardozo, Foletta, Dill, Brinckerhoff et al. et al. Shim et al. et al. et al. et al. Duduta et al. Authors: 1996 2004 2009 2010 2011 2011 2012 2013 2013 2013

Service Frequency ✚ Peak-Only Service Headway ( ‒ ) Travel Time Transfer Time Bi-Directional stop service Bus connections (# of

routes) ✚ ✚ ✚ ✚ ✚ ✚ ✚ ✚ ✚ ✚ Rail connections n.s. ✚ n.s. ✚ ✚ ✚ ✚ ✚ Connectedness measures

(Hwy) n.s.

Terminal Station (dummy) ✚ ✚ ✚ ✚ Transfer Station (dummy) ✚ ✚ ✚ ✚ ✚ ✚ ✚ ✚ ✚ Elevated Station (dummy) Number of Bus Stops w/n

Buffer Transfers (#) ( ‒ ) Relative Location (accessibility: gravity-model w/ time impedance) Regional Accessibility TRANSIT SERVICE QUALITY TRANSIT ('network effect': share of regional jobs + population within 1/2mile buffer of all stations in region) Distance to CBD Interstation Spacing ✚ n.s. n.s. ✚ Mode-based dummy ( HRT

/ LRT / BRT ) ✚ Fare Free station ✚ Fare Bicycle Park and Ride ( #

spaces or dummy ) Car Park and Ride ( #

spaces or dummy ) ✚ ✚ ✚ ✚ ✚ ✚ ✚ ✚ ✚ ✚

City-specific dummy

variable ✚ ✚

Congestion (annual traffic

delay per commuter) Line-specific dummy

OTHER variable(s) Heating & Cooling Degree

Days ✚

LEGEND: ✚ Positive and significant (‒) Negative and significant n.s. Not significant o Origin d Destination Not specified in final model

53 Table 3.6 International Review of Rapid-Transit Station-Level Ridership Determinants (partial table): Transit Service Quality 2014-2017

Ramos & Durning & Brown Zhao, et al. Townsend Chen & Zegras Renee et al. Authors: 2014 2014 2015 2016 2017

Service Frequency n.s. ✚ ✚ Peak-Only Service ✚ Headway Travel Time ( ‒ ) ( ‒ ) Transfer Time ( ‒ ) ( ‒ ) Bi-Directional stop service ✚ Bus connections (# of routes) ✚ ✚ ✚ ✚ ✚ ✚ ✚ ✚ ✚ ✚ ✚ Rail connections Connectedness measures

(Hwy)

Terminal Station (dummy) ✚ ✚ n.s. n.s. ✚ ✚ ✚ Transfer Station (dummy) ✚ ✚ ✚ ✚ ✚ ✚ n.s. Elevated Station (dummy) n.s. Number of Bus Stops w/n

Buffer Transfers (#) Relative Location (accessibility: gravity-model ✚ w/ time impedance) Regional Accessibility TRANSIT SERVICE QUALITY TRANSIT ('network effect': share of regional jobs + population ✚ within 1/2mile buffer of all stations in region) o.‒ Distance to CBD ✚.d ✚ (‒) (-) n.s. d.‒ Interstation Spacing ✚ ✚.o ✚ ✚ ✚ ✚ ✚ Mode-based dummy ( HRT /

LRT / BRT ) ✚.HRT ✚.HRT/LRT Fare Free station Fare (‒) n.s. Bicycle Park and Ride ( #

spaces or dummy ) ✚ ✚.o-d ✚.o-d Car Park and Ride ( # spaces

or dummy ) ✚ ✚ ✚ ✚ ✚ ✚ ✚

City-specific dummy variable ✚

Congestion (annual traffic

delay per commuter) ✚ Line-specific dummy

OTHER variable(s) ✚ ✚ ✚ ✚ ✚ Heating & Cooling Degree

Days

\LEGEND: ✚ Positive and significant (‒) Negative and significant n.s. Not significant o Origin d Destination Not specified in final model

54 Bus Network and Rapid-Transit Network Attributes Vector [NT]:

Bus_Line (bus feeder network connectivity measure - Model 1A; counts of bus routes within station’s bus feeder influence radius): Bus network connectivity at rapid- transit stations consistently registers significant and positive associations with higher ridership levels in preceding DDM studies. Bus routes that interact with rapid-transit stations extend and tap into potential demand in areas beyond the station’s immediate pedestrian service area. The supply of bus feeder service, measured as the number of bus routes across the full range of bus services (local, express, rapid, commuter, circulator, etc.) that operate within a station’s feeder- bus capture radius (.36mile; as per LA Metro 2011 Survey GIS spatial person-trip empirical data) represents a basic measure for bus network connectivity in most preceding DDM studies, and is likewise used and hypothesized in this study to have significant and positive relationship with station boardings. It also serves as benchmark for two additional exploratory bus connectivity measures specified for Model 1B and Model 1C (below).

Bus_Freq (bus feeder network connectivity measure - Model 1B; aggregate measure of bus routes weekday vehicle frequency): For exploratory purposes a more nuanced aggregate bus feeder connectivity measure is implemented in an alternative regression model. It is hypothesized as an improvement in validity in capturing bus feeder network connectivity at station level as it most closely captures bus transit supply levels. Conceptually and dimensionally it also pairs the measure for rapid-transit service supply levels as a weekday vehicles-per-day measure. This is done by using agency GTFS files and schedules integrated in a new GIS geodatabase, where each station is linked to a sub-set of feeding bus lines. The supply capacity of each bus route (vehicles per day) is calculated and aggregated across all bus service types. It is hypothesized to reflect a significant and positive association to boardings and improve on the explanatory power of the model as compared to the simpler ‘Bus_Line’ measure in Model 1A and in various preceding DDM studies.

Bus_ACCe (bus transfer (access) estimated count - Model 1C): this is a key explanatory variable of interest in this study, which captures the average weekday bus transfer events (access) at each rapid-transit station and serves as a proxy for demand-side bus feeder network influence on rapid transit ridership. It is hypothesized that in large, disperse, and decentralized urban agglomerations, such as Los Angeles metropolitan area, bus network connectivity is critical for overall rapid-transit ridership and performance and is expected to have a highly significant and large magnitude estimate. Actual observations and/or survey-based estimates of bus access events, when available, should provide a better instrument for capturing this bus network connectivity effect. This model is also expected to result in a higher explanatory power as compared to Model 1A and Model 1B.

55 Commtr_L (count of rail commuter lines; connectivity measure): station connections to commuter rail lines have greater accessibility to larger travel markets and are expected to register additional boardings as compared to stations with no commuter line connections, all else equal. In the case of Los Angeles ‘Metrolink’ operates five commuter lines that mostly interact with Union Station.

Centrality (node centrality topological measure; number of links): In lieu of providing a gravity-based destination accessibility measure, for which data was not available in this study, a basic network topological measure of node centrality is provided as control. It is hypothesized that a higher node centrality measure is associated with higher boardings due to higher accessibility levels.

Terminal (terminal station dummy): stations designated as a line terminal should register higher boardings as result of larger service areas, their function as transit hub, and/or presence of larger Park & Ride facilities.

TrnfHUB (transfer station dummy): stations that allow for transfer should register higher boardings due to aggregate supply levels, higher accessibility, and higher transfer rates than other rapid-transit stations.

AvgDIST_mt (average centroid-to-centroid distance with adjacent stations): the average Euclidean distance with adjacent stations influences their service area size and associated trip demand as stations compete for travel market share. This measure aims to capture differences, if any, in stations’ service area size beyond the standard station pedestrian catchment area delimitations. Higher average distance between adjacent stations is expected to reflect larger market area and higher boarding counts, all else equal.

Socio-Economic Vector [SE]:

POP_10 (aggregate population level): The number of residents within a station’s pedestrian service area serves a proxy for potential person-trip generation. The higher the population level (i.e. density) the higher the expected number of boardings.

JOB_11 (aggregate employment level): Employment levels within station’s service area serves a proxy for a variety of socio-economic activities with potential for both trip attraction and generation. In this study higher employment levels is expected to register higher boardings at station-level for a variety of trip purposes.

Veh_HU (average vehicle availability in occupied households within stations’ pedestrian service areas): Within the U.S. urban mobility market, and particularly in car- oriented urban agglomerations, transit travel is generally considered an inferior good. As such, higher levels of household vehicle availability would tend to

56 reflect on lower transit patronage and lower boardings at station-level. This relationship has been highly significant and large in magnitude in previous rapid- transit ridership studies in the U.S. when specified in models and is hypothesized to have a similar effect in this model.

Perc_Hisp (Percent residents of Hispanic ethnicity): In Los Angeles the Hispanic community is the largest patron of transit as compared to other ethnic communities and at a larger proportion to their share of the general metropolitan population. This observation, based on LA Metro bus and rail on-board survey socio-economic statistics (PTV NuStats-LAMCTA 2012) applies to all transit modes as well as access modes, particularly for bus transit and bus access to rapid-transit service. Hispanic ethnicity in Los Angeles is also correlated with lower income levels, lower vehicle availability, and lower rates of driving license tenure, all of which contribute to higher dependence on transit for mobility. Although transit literature typically associates rail travel market with higher-income choice riders, in this study this demographic attribute is hypothesized to have a positive relationship with boardings at rapid-transit station-level.

Land-Use and Built Environment Attributes Vector [LU.BE]:

Wlk_SCR (Walk Score®; multi-dimensional walkability rank): land-use/travel-behavior research points to the multi-dimensional nature of walkability (pedestrian friendliness) as a key factor in encouraging more walking and transit patronage. Key components that inform this ranking index are population density, land-use mix, and intersection density as a proxy for higher street network connectivity. Previous station-level DDM ridership studies have found significant, positive, yet relatively weak associations between various walkability measures, including Walk Score®. It is thus hypothesized that this measure, captured for rapid-transit portals and averaged for station locations, will also have a significant and positive effect on station-level ridership levels as higher scores would likely increase the propensity of potential riders to walk and use transit. For a more detailed explanation of Walk Score® ranking of pedestrian-friendliness please refer to page 89 in Chapter 4.

Spe_Gen (special activity generators): Previous transit ridership and station-level transit DDM studies have documented that the presence of one or several special trip generator/attractor (i.e. university campuses, health-care and hospital campuses, active convention centers, sports arenas, tourism attractions and districts, regional commercial centers, etc.) within stations’ service areas generate more riders as compared to the average rapid-transit station. This author hypothesizes a similar effect in Los Angeles metropolitan area. Several sources were used to identify these special land-use features including city and tourism agency information related to key destinations in their cities, Los Angeles County assessor’s GIS parcel database (year 2011) that contains specific land-use descriptors, and

57 qualitative assessment of year 2011 aerial imagery that helped identify large assembly-occupancy building types in proximity to stations. A higher count of special generators is hypothesized to generate additional boardings.

Transit Service Level Vector [TS]:

RTv_pDAY (rapid-transit line(s) weekday vehicle frequency; trains per day): higher service frequency (reduced headways and/or longer service windows) has been positively associated with higher ridership levels at several network scales. It increases transit supply and capacity levels and as such the potential to accommodate greater demand and boardings at stations. It is expected to register a significant and positive effect on station average weekday boardings in this study. n_PKG (free and paid Park and Ride facilities; number of parking spaces): the presence of park and ride facilities, whether private, public, or in partnership has registered significant and positive relationships with rapid-transit patronage, particularly for stations located in suburban contexts and terminals. The aggregate number of free public parking spaces, whether operated by LA Metro, City, and/or County, as well as paid private parking storage within one block distance of a station is expected to have a positive association with station’s average weekday boardings.

Other Control and Exploratory Variables:

BRT_Silver (BRT Silver Line dummy; MT-910): LA Metro’s rapid-transit Silver Line is distinct in that it combines limited-stop street-level bus service in downtown LA, and BRT rapid-transit service along larger segments of the line. The stations associated to the Silver Line also register notably lower average boarding counts as compared to the average for LA Metro rapid-transit stations. To control for these unique service and performance characteristics, and unique stations spatial context a dummy variable identifies Silver Line stations as control, which is hypothesized to register relative lower boarding counts, all else equal.

Fwy_ROW (location of station within freeway right-of way; dummy variable): some stations associated with LA Metro Green line and LA Metro Silver line operate for the most part in an interstitial context within the central right-of-way of freeways. These locations could be considered and perceived as a lower service experience by some patrons due to undesirable environmental and aesthetic conditions (i.e. higher noise levels and air contamination from nearby freeway traffic, lack of urban amenities, safety, etc.). A dummy variable controls for this situation that is expected to be negatively associated with boardings.

OneWay (station or stop providing one-way service only): some stations and stops provide service in only one direction (i.e. terminals, some BRT stops downtown). They

58 thus provide lower accessibility, all else equal, as compared to the typical station or stop that provides two or more directional travel options. Previous DDM studies have also registered statistically significant negative associations between one-way stops and boardings, relative to other stations. This negative association is also expected in this study. bRapid_ln (LA Metro Rapid bus service dummy): a qualitative visual assessment in GIS platform of LA Metro ‘Rapid’ bus network, rapid-transit station boarding levels, and bus access-mode shares suggests a positive association between proximity and / or coincidence of ‘Rapid’ bus lines, rapid-transit station bus-access share and total boarding levels. It is expected that rapid-transit stations that benefit from connectivity with this faster and higher frequency bus service will register higher boarding counts, all else equal.

Comm_xBus (Non-LA Metro commute express routes dummy): peripheral counties that offer express commute bus service to Los Angeles downtown and other sub-centers tend to partially operate lines in parallel to LA Metro rapid-transit lines. Complementarity or competitiveness of these routes vis-à-vis Metro’s rapid- transit lines is not clear. A dummy variable will register their presence and accessibility (boarding and alighting of commute route is possible at or near rapid-transit stations). Significance and directionality of dummy estimate will inform assessment of complementarity or competitiveness as related to rapid- transit station boardings. No explicit hypothesis is posited as this aspect of the investigation is exploratory and inductive.

Table 3.7 reports outcome and explanatory variable definitions, descriptive statistics, and hypothesized signs. Efforts to compile data for the same year as the on-board survey (2011), or closest year available for all datasets were successful for the most part. Walk Score® does not archive historical data and the closest year available is 2016. However alternative distance-based and composite walkability (pedestrian-friendliness) Indeces are possible to compute for the base year if data and resources are available, including GIS-Transport software applications such as TransCAD or ArcMAP and appropriate land-use, activity, and street/pedestrian network databases.

59 Table 3.7 Descriptive Statistics for Model Variables

Expected Variables Description Source Year Mean Std. Dev. Min Max Sign Average weekday FY RT_AWB boardings N/A LA Metro 2011- 3443.33 5515.29 50.00 38665.33 (dependent) 2012 LA Metro / SCAG Count of connecting transit agencies BUS_Line ( + ) 2011 17.13 18.85 1.00 70.00 bus lines schedules and GTFS files LA Metro / SCAG Aggregate number of transit agencies BUS_Freq connecting bus lines ( + ) 2011 1539.11 1890.53 44.00 7506.00 schedules and GTFS vehicles per day files LACMTA BUS_ACCe Bus access counts ( + ) 2011 1214.49 2424.36 0.00 20471.42 (estimated by author) LACMTA BUS_ACCr Bus access rate ( + ) (estimated by author 2011 0.32 0.25 0.00 1.00 w/ Stata v.13) Population levels within station US Census 2010 POP_10 ( + ) 2010 17951.23 10146.24 183 55385 pedestrian service block-level area Total number of jobs US Dept. Labor within station 'OnTheMap’ JOB_11 ( + ) 2011 15749.85 25994.75 742 154439 pedestrian service point feature data area US Census 2010 Population and block-level employment levels & 2010 / 9478. PopJob ( + ) 33380.11 27180.56 173233.00 within station service US Dept. Labor 2011 00 area 'OnTheMap' point feature data Average number of US Census vehicles per occupied Veh_HU ( - ) ACS 2012 5YR 2012 1.27 0.43 0.29 2.06 housing unit within block group station service area Percent Hispanic US Census Perc_Hisp population within ( + ) ACS 2012 5YR 2012 50.08 24.16 0.21 99.00 station service area block group-level

Walk Score®; Wlk_SCR multidimensional ( + ) Walk Score® 2016 76.7194 18.37445 25 99 index for walkability

Average distance 119.5 AvgDIST_mt between adjacent ( + ) LA Metro shapefiles 2011 1802.24 1522.46 11882.43 2 stations (meters) LA Metro / Rapid-transit vehicles SCAG Transit 2011 / RTv_pDAY ( + ) 224.93 108.94 80.00 667.00 per day agencies GTFS files 2012 and schedules Number of free n_PKGfree ( + ) LA Metro 2011 221.69 414.62 0.00 2050.00 parking spaces Number of nearby n_PKGpay paid parking spaces ( + ) Parkopedia.com 2015 111.14 434.87 0.00 3030.00 (one-block buffer) Total number of LA Metro / 2011 / n_PKG parking spaces (free ( + ) 332.83 567.00 0.00 3030.00 Parkopedia.com 2015 & paid) Number of Commtr_L connecting commuter ( + ) LA Metro / Metrolink 2011 0.06 0.50 0.00 5.00 lines (MetroLink) Terminal station (0-1 Terminal ( + ) LA Metro 2011 0.11 0.32 0.00 1.00 dummy) Transfer hub (0-1 TrnfHUB ( + ) LA Metro 2011 0.10 0.29 0.00 1.00 dummy)

60 Table 3.7-continued

Expected Variables Description Source Year Mean Std. Dev. Min Max Sign (0-1 dummy) Fwy_ROW Station located in ( - ) LA Metro 2011 0.13 0.34 0.00 1.00 freeway ROW

(0-1 dummy) OneWay One way direction ( - ) LA Metro 2011 0.24 0.43 0.00 1.00 service

(0-1 dummy) BRT_Silver ( - ) LA Metro shapefiles 2011 0.20_ 0.40_ 0.00 1.00 Identifies Silver Line stations Los Angeles County Assessor’s Office Special activity Parcel Data / SpeGEN ( + ) 2011 4.17 4.37 0.00 32.00 generators Los Angeles City / LA Metro / Google Map (0-1 dummy) TwoWay Two-way direction ( + ) LA Metro 2011 0.76 0.43 0.00 1.00 service

Network topological Centrality centrality measure ( + ) LA Metro 2011 1.92 0.69 1.00 5.00 (num. links per node)

(0-1 dummy) Street-level stop Street_Srvc ( - ) LA Metro 2011 0.12 0.33 0.00 1.00 (Silver Line)

Number of RT_Line connecting ( + ) LA Metro 2011 1.09 0.37 1.00 3.00 rapid-transit lines

Number of TotR_Line connecting rail and ( + ) LA Metro / Metrolink 2011 1.14 0.75 1.00 8.00 rapid-transit lines LA Metro / SCAG Number of limited- transit agencies bLimtd_ln stop ( + ) 2011 0.04 0.19 0.00 1.00 schedules and GTFS bus line files LA Metro / SCAG Number of rapid bus transit agencies bRapid_ln ( + ) 2011 1.50 2.20 0.00 8.00 service lines schedules and GTFS files LA Metro / SCAG Number of express transit agencies bXpress_ln ( + ) 2011 4.84 7.05 0.00 22.00 bus service lines schedules and GTFS files LA Metro / SCAG Number of commuter transit agencies bCommtr ( + ) 2011 1.76 3.05 0.00 9.00 bus service lines schedules and GTFS files (0-1 dummy) Los Angeles County Station adjacent and Assessor’s Office Airport ( + ) 2011 0.01 0.10 0.00 1.00 linked to international Parcel Data / LA airport Metro

61 Results and Discussion

The vector of explanatory variables was regressed onto average weekday boardings at station level for three negative-binomial models (NBREG 1A, NBREG 1B, and NBREG 1C; Table 3.8). The data was fitted in Stata (version 13; data processing and statistical software). Standard errors were bootstrapped in all three NBREG models to address the relative low number of observations in the final models (N=103). Model results for the most part are consistent with previous station-level DDM studies and literature review (see Tables 3.1 - 3.6), although a few explanatory variables did not enter the final models due to multicollinearity issues and other variables needed transformation to address residuals non-linearity. For example the variable that accounted for special activity generators (SpeGEN) is very highly correlated with the variable for number of jobs (JOB_11) and caused nonsensical results in the original model. Its removal allowed for a better model fit. Also the JOB_11 variable was log-transformed to address residuals non-linearity. The original variable that measured rapid-transit supply in the form of aggregate vehicles per day (RTv_pDAY) also reported residuals funnel-shape non- linearity and counter-intuitive parameter results. Thus, a squared component was added to the model to correct this issue (RTv_pDAY^2).

It is possible that endogeneity between transit demand and supply factors (average weekday boardings (RT_AWB) and rapid-transit vehicles per day (RTv_pDAY)) caused unexpected non-significance and negative relationship of the original rapid-transit supply factor, which typically registers a positive and significant relationship in previous transit ridership studies. Although endogeneity is not addressed in this present study, the squared term did improve model fit and residuals linearity, and the resulting parameter follows expected directionality based on theory and literature review, yet it remains non-significant. One observation (‘Union Station’) registers high-leverage characteristics for the original RTv_pDAY supply measure and is a potential influential data point. However, its removal notably affected model performance and remains in the dataset. Overall the aforementioned corrections produced a better model fit in all three sub-models (see Table 3.3; Models 1A, 1B, 1C).

62 Key explanatory variables of interest in this study, which measure the degree and influence of bus network connectivity at rapid-transit stations, register highly significant 99th percent confidence levels in the first two sub-models (Model 1A, ‘BUS_line’; Model 1B, ‘BUS_Freq’) and at 95th percent confidence level in the third model (Model 1C, ‘BUS_ACCe’). The ‘BUS_Line’ variable captures the number of bus lines associated with each station and the ‘BUS_Freq’ variable is the aggregate bus route service measure (number of vehicles per day). These two variables also report positive inelastic relationships with boardings, comparable to that for population levels and consistent with theory and literature. The ‘BUS_ACCe’ variable incorporated in Model 1C, which captures demand-side estimated bus-rail transfer counts (access), registers a lower magnitude of influence (elasticity) and confidence level.

The results of overall model fit diagnostics for the three sub-models (Table 3.8, bottom) partially confirm the hypothesis that a measure of bus network connectivity based on the aggregate supply of vehicles per day from a sub-set of interacting feeder routes (‘BUS_Freq’; Model 1B) better captures bus network connectivity as compared to the simpler count of routes (‘BUS_Line’; Model 1A) and improves overall model explanatory power. Cragg & Uhler’s pseudo-R2 increases from 0.899 to 0.909 between Model 1A and Model 1B. AIC and BIC model fit statistics also register an improvement from Model 1A to Model 1B (16.352 vs. 16.243; 1731.7 vs. 1720.4, respectively). However negative-binomial regressions ‘Pearson (1/df)’ statistic is considered a better measure of model fit as compared to pseudo-R2 measures (Hilbe 2011). It is a geometrically-based comparison of observed and predicted outcomes and its interpretation is based on how close the statistic is to value ‘1’; the closer to a value of ‘1’ the better the fit. As such the reduction in value of the Pearson (1/df) statistic from Model 1A to Model 1B (0.2223, 0.2055, respectively) presents a contradiction when compared to the AIC, BIC, and Cragg & Uhler’s pseudo-R2 statistic results and most likely represents under-dispersion in the new model among other potential explanations that should be addressed in future research extensions.

It is also hypothesized in this study that a demand-side measure of bus network connectivity derived from survey-based estimates of bus-rail transfers (BUS_ACCe) at the station-level would present a more accurate and valid measure of bus network connectivity as

63 compared to the previous two measures implemented in Model 1A and Model 1B. As such it was expected to reflect as an additional improvement in model explanatory potential. However, the results on model fit statistics are also mixed for this measure. This particular instrument presents a significant positive association yet lower confidence level (95th percent) and elasticity than the previous two instruments implemented in Model 1A and Model 1C. On the other hand Model 1C presents the Pearson (1/df) statistic that is closer to the value of “1”, thus could be considered the better overall model based solely on this fit measure.

Consideration to the other three model fit statistics, Crag & Uhler’s R2, AIC, and BIC, report weaker overall model fit for Model 1C and thus equally renders an inconclusive analysis. Perhaps more attention to station-level bus transfer events in on-board surveys (i.e. larger sampling) could allow for more accurate estimates and higher confidence levels that could render more conclusive analyses. Integration of bus and rail fare cards together with APC (Automated Passenger Count) technology could provide more accurate and periodical statistics of bus-rail transfer events in the future and reduce uncertainty on the magnitude of this and other travel behaviors.

Consistent with previous DDM station ridership studies population levels (POP_10) and the log-transformation of total number of jobs (logJOB_11) exhibit positive and significant associations with boardings, with population level registering a stronger relationship at the 99th percent confidence level in two of the three sub-models (Model 1A, Model 1B). The average Euclidean distance between adjacent stations (AvgDIST_mt) also exhibits a highly significant and positive relationship with boardings for the first two models, and a positive inelastic relationship.

Other strong and statistically significant factors associated with rapid-transit station boardings in Los Angeles are average vehicle availability per occupied housing (Veh_HU), which is as expected negatively associated to boardings and significant at the 90th percent confidence level in Model 1A and Model 1B, and at a higher 95th percent confidence level for Model 1C. It also reports a relatively large inelastic relationship with the outcome variable.

64 Categorical dummy variables that account for rapid-transit network structure, such as terminal stations (Terminal) and transfer stations (TrnfHUB) are highly significant and positive in all three sub-models, and consistent with previous findings. Contextual and operational variables such as one-way service (OneWay) also registers highly significant and negative association in all models, which is also consistent with previous findings and likely reflects congested street-level operating conditions. The largest reductive multiplier effect results register for the ‘BRT_Silver’ dummy variable. This variable was specified to capture the distinct combined limited-stop and BRT service of line MT-910 that is hypothesized to represent a lower service-level and thus lower boarding counts. Although not statistically significant, the contextual ‘Fwy_ROW’ dummy (location of stations within freeways right-of-way; LRT-Green and BRT-Silver lines) also reports an expected negative association with boardings and points toward the importance of spatial and environmental factors in rapid-transit ridership.

The multi-dimensional measure for pedestrian-friendliness, Walk Score®, which was implemented in the model as a proxy for built-environment TOD attributes (intersection density, land-use mix, and population density) did not register a significant relationship to boardings and resulted in a negative association with the outcome variable. This counter-intuitive result is likely due to a high negative correlation with another covariate in the model (Veh_HU) and/or the fact that it does not incorporates the most important spatial factor associated with the specific travel behavior of walking-to-transit (distance; Agrawal et al. 2008; Tal et al. 2016). As such it may not be the most valid instrument for this type of investigation and may require modification in future studies.

In regards to the first research question that guides this study, how important are bus feeder networks to rapid-transit ridership in large, dispersed, and decentralized metropolitan regions in the U.S.?; the descriptive and inferential statistic results confirm their importance and critical contribution to station-level ridership, representing 20% - 50% of total boardings at line- level and reporting high confidence levels relatively larger elasticities (on average and range) than those reported for urban design attributes in other studies as related to transit use. In the case of Los Angeles metropolitan transit system the resulting bus networks connectivity elasticities

65 register a range of [0.28 - 0.42] as compared to other studies’ range of [0.23 - 0.29] for intersection density (see Ewing & Cervero 2010; Stevens, M. 2017).

Do supply-side measures of bus network connectivity that incorporate distinct route service levels (‘BUS_Freq’) improve the predictive power of station-level direct-demand ridership models?, three out of four fit statistics for Model 1B confirm this hypothesis. Do demand-side estimates of bus transfer (access) events (‘BUS-ACCe’) as a measure of bus network connectivity at station-level improve direct-demand model explanatory power?, one out of four fit statistics supports this hypothesis. As such this latter aspect of the investigation remains inconclusive but suggestive that Model 1B better fits the empirical data (see Appendix D, Observed vs. Predicted Outcomes plots). In the future better bus-rail survey statistics (i.e. more station-level sampling for narrower confidence intervals) or actual counts of bus-rail transfers (access and egress) might provide more robust data to elucidate this issue, in addition to an assessment and correction, if necessary, of potential endogeneity in the model.

Conclusions

The principal research questions that guided this investigation focus on assessing the influence of bus feeder networks on rapid-transit ridership in the special case of a large decentralized urban agglomeration; and on exploring the usefulness of more nuanced bus feeder network measures at the station-level for improving station-level DDM model performance, and a better understanding of travel behavior and multi-modal interactions. The introductory descriptive statistics and correlation analyses of system-wide bus access share to rapid-transit systems for nine large cities in the United States already highlight the non-trivial nature of this access mode and its plausible association with large, decentralized, and poly-centric metropolitan structures. Given on-going sprawl and decentralization trends in many U.S. cities it is pertinent to address, document, and study this travel behavior as it may become more important in the future.

66 Table 3.8 Regression Model Results and Model Fit

Negative binomial regression: Model 1A Model 1B Model 1C Model 1A Model 1B Model 1C (bootstrapped SE) b/se b/se b/se Number of obs 103 103 103 Wald chi2(16) 1125.97 1340.99 638.56 Dispersion = mean;Prob > chi2 0.000 0.000 0.000 Log pseudolikelihood -824.123 818.502 -825.454 Rapid-Transit Station RT_AWB [DV] : IRR Elasticity IRR Elasticity IRR Elasticity Annual Average Weekday Boardings

BUS_Line 0.024202*** 1.02450 0.418 -0.00491 BUS_Freq 0.000270*** 1.00027 0.419 -0.00004 BUS_ACCe 0.000227** 1.00023 0.278 -0.00008 POP_10 0.000025*** 0.000021*** 0.000016** 1.00003 0.446 1.00002 0.379 1.00002 0.282 -0.00001 -0.00001 -0.00001 logJOB_11 0.131202* 0.104712* 0.169188** 1.14020 1.181 1.11039 0.943 1.18434 1.523 -0.06476 -0.05175 -0.06539 Veh_HU -0.486146* -0.416627* -0.597091** 0.61499 -0.613 0.65927 -0.526 0.55041 -0.753 -0.24208 -0.18079 -0.21375 Perc_Hisp -0.001315 -0.001257 -0.001115 -0.00198 -0.00235 -0.00201 Wlk_SCR -0.005944 -0.005342 -0.004134 -0.00436 -0.00344 -0.00365 AvgDIST_mt 0.000305*** 0.000308*** 0.000266* 1.00031 0.552 1.00031 0.557 1.00027 0.480 -0.00008 -0.00007 -0.00011 RTv_pDAY -0.003839 -0.003781 0.000463 -0.00392 -0.00313 -0.0032 RTv_pDAY2 0.000002 0.000001 -0.000005 -0.00001 -0.00001 -0.00001 n_PKG 0.000238 0.000309* 0.000300* 1.00031 0.102 1.00030 0.099 -0.00015 -0.00013 -0.00012 Commtr_L 0.350535 0.392581 -0.177487 -0.51258 -0.52052 -0.9354 Terminal 0.926684* 0.988540*** 0.275916 2.52612 2.68731 -0.36698 -0.26825 -0.30772 TrnfHUB 1.345977*** 1.404471*** 0.852957** 3.84194 4.07337 2.34658 -0.17645 -0.21715 -0.29218 Fwy_ROW -0.106801 -0.101178 -0.157584 -0.23816 -0.19189 -0.20258 OneWay -1.349967*** -1.466069*** -0.634398 0.25925 0.23083 -0.31377 -0.23996 -0.33261 BRT_Silver -2.965471*** -2.948325*** -2.061625*** 0.05154 0.05243 0.12725 -0.25301 -0.19251 -0.21652 constant 7.395224*** 7.553634*** 6.650466*** -0.95349 -0.79939 -0.95742

/lnalpha -1.786858 -1.890515 -1.766436 alpha 0.1674856 0.1509941 0.1709412

Model 1A Model 1B Model 1C model fit model fit model fit 2 Cragg & Uhler's R 0.899 0.909 0.896 (1/df) Pearson 0.2223 0.2055 0.2354 AIC 16.352 16.243 16.378 BIC 1731.7 1720.4 1734.3

67 The detailed assessment of Los Angeles metropolitan multi-modal transit system reveals it has the largest bus access mode share of the nine document systems. The multivariate inferential statistics assessment also reveals that bus network connectivity at the rapid-transit station level, measured with three alternative instruments, is highly significant and reports relatively large elasticity values comparable to those of population levels. Likewise, the study shows that more nuanced aggregate supply measures of bus feeder connectivity yields better model results and improved explanatory power as compared to the often used and more basic count of bus routes. This aggregate measure is relatively easy to calculate based on readily available GTFS files and route schedules and only require intermediate level experience with GIS technology in producing customized geodatabases.

A general review of regression results indicates that policies directed at increasing population and/or employment levels in areas surrounding rapid-transit stations in Los Angeles would yield higher transit patronage, all else equal. Higher bus network connectivity and service improvements should also yield higher boarding counts. In assessing potential marginal effects based on the best fitting Model 1B results, a 20% increase in bus service frequency would yield 14 additional station boardings, all else equal; a 20% increase in population levels would yield 269 additional station boardings, holding other covariates at mean; and a 20% increase in log- jobs would yield 68 additional boardings, all else equal. The feasibility and the net benefits or costs of each approach, or those of a combined land-use/transit-service approach, warrants a detailed analysis beyond the scope of this investigation. However, data and insights provided in this study could contribute in such efforts.

Statistics of bus transfer counts at various network levels, but particularly at the station- level, are scant, absent, and/or difficult to obtain. Ironically, those cities that register higher system-wide bus access share are the ones with more data limitations (i.e. Los Angeles, Atlanta, Miami), whilst those systems with more centralized urban structures and lower bus access shares periodically document bus transfer events at multiple network scales, including at station level (i.e. see Boston’s MBTA or Philadelphia’s SEPTA station survey reports).

68 It is clear from this and previous transit studies that an integrated multi-modal bus / rapid- transit ‘network’ approach to planning and operations in the United States is warranted. Better data collection, on-board travel survey design and statistics that take into account the importance of bus / rapid-transit transfers, and development of better access predictive models for bus mode at station-level is warranted if transit is to maintain and ideally improve its share in the urban travel market, preferably in combination to other land-use strategies such as TOD, and/or travel demand-management policies like those geared to manage parking supply and reduce automobile travel.

Limitations and Future Extensions

As a single case-study of a distinctly large poly-centric urban agglomeration, typically described as a megalopolis in urban and transportation literature, the findings may not be generalizable to other global regions or smaller cities and transit systems. However, many regions are exhibiting similar conurbation and decentralization processes, and increasing motorization that may result in conditions similar to those of Los Angeles.

The probable presence of endogeneity in the models is a particular weakness that should be addressed in the future possibly via 2SLS (two-stage last square) system of equations, assuming pertinent instrumental variables are identified and data is accessible. Likewise an improvement of the instrument(s) aimed at capturing the degree of pedestrian-friendliness around and near rapid-transit stations should be developed and incorporated in future models, ideally integrating a distance to transit portal component, either independent or combined with other LU.BE measures. This could be achieved by calculating a ‘directness’ index for the pedestrian service area around each station where the ratio of average street network distance from parcel centroids to transit portal is divided over the average Euclidean distance. Otherwise an empirically derived shortest-path protocol along the pedestrian network from each person-trip origin to station could serve a similar purpose, based on survey person-trip spatial data.

Methodological improvements related to delimitation of stations pedestrian service areas, which are used to capture areal socioeconomic and spatial attributes could be accommodated in

69 the future if pertinent data and resources are available. For this study clipped Thiessen polygons were implemented in ArcGIS (v.13) software, which address the likely mutually-exclusive conditions as stations compete for market share. The buffer radiuses used as template for splitting the polygons were empirically determined based on survey person-trip spatial data (average Euclidean distance between trip origin and fist boarded station; Appendix B). However, studies on pedestrian service area delimitation methods have revealed that network distance- based protocols are more accurate and valid definitions than Euclidean distance-based buffers as the latter overestimate service areas and population-employment levels (Upchurch et al. 2004; Gutiérrez& Palomares 2008). Future methodological improvements would incorporate network distance-based protocols assuming pedestrian network data, specialized software (i.e. TransCad and/or ArcGIS network-based tools), and resources required to process these attributes are available.

It is also important to recognize inherent limitations of cross-sectional studies such as this one, as they cannot address causality nor time-related dynamics that might affect explanatory factors and their associations with the outcome variable. On-going implementation of APC technology across a variety of transit systems, which can gather transit boarding statistics for a variety of time periods, together with periodical institutional collection of socio-economic and built-environment data such as Census ACS 5-year program and yearly City assessor’s parcel updates would help in building pertinent panel data for future longitudinal studies.

Finally, a pressing issue in regards to the phenomenon of interest in this study is understanding why such an important travel behavior, bus access to rapid-transit service, is being overlooked in many on-board surveys and transit systems reports, particularly at station-level and in those cities where bus access share for rapid-transit systems are the highest in the nation. Plausible research approaches and theories to understand this phenomenon might involve political-economic frameworks and qualitative methods such as surveys, document reviews, and/or interviews with key decision-makers and transit planners.

70 CHAPTER 4

AN ASSESSMENT OF THE INFLUENCE OF BUILT-ENVIRONMENT ATTRIBUTES AROUND BUS-STOPS ON RAPID-TRANSIT STATION BOARDINGS

Introduction

Increasing the share of transit trips is one key strategy for advancing more sustainable cities in an era of global warming and rapid climate change. Numerous studies related to rapid- transit systems have focused on the potential influence of land-use and built-environment (LU.BE) features as key determinants or mediators on transit system’s patronage. Some find statistically significant but relatively weak associations as compared to socio-economic, transit network connectivity, regional accessibility, and transit service level factors. Most if not all recent LU.BE / transit-patronage studies, particularly those that implement direct-demand modeling (DDM) focus their attention on areas adjacent to rapid-transit stations (i.e. TOD’s pedestrian catchment areas) yet ignore areas surrounding feeder bus stops. These peripheral areas could be relevant in cities where the share of bus access to rapid-transit service is relatively high. For example large, more dispersed and poly-centric urban agglomerations in the United States register higher proportions of bus access share in their rapid-transit systems (i.e. Los Angeles, Miami, and Atlanta; > 30%) as compared to older and more center-focused cities (i.e. Chicago, Boston, Philadelphia). In addition, rapid-transit lines that mostly operate within freeway right-of- ways register bus access rates near 50%, higher than their share for pedestrian access.

Given long-term trends toward sprawl, poly-centricity, and its correlation with higher bus access shares in large U.S. urban agglomerations it is pertinent and logical to assess the yet unexplored LU.BE / rapid-transit patronage relationships in areas surrounding feeder bus stops. This study looks at Los Angeles multi-modal metropolitan transit system and its diverse built- environments to understand how LU.BE features around feeder line bus-stops might affect boardings at rapid-transit stations. A multilevel structural equations model is implemented and fitted with survey-based bus-stop boarding estimates, GIS-based person-trip spatial data, and LU.BE measures for both bus stops and linked rapid-transit stations. The goal is to gain a better understanding of LU.BE influence on rapid-transit access by bus in the special context of large

71 multi-modal transit systems, and ideally inform their planning and combined transportation/land- use policies in pursuit of increasing patronage and advancing less car-dependent and more sustainable urban agglomerations.

The main research questions that guide this study are: 1- Do land-use and built- environment attributes around feeder line bus-stops influence the number of bus transfers (access) counts at rapid-transit stations?; 2- Do rapid-transit station attributes associated with destination accessibility and service levels influence the number of boardings at bus-stop level?; 3- Do land-use and built-environment attributes around rapid-transit stations influence the number of boardings at bus-stop level?

The rest of this second study is organized in five sections. Section 1 presents a literature review of overlapping branches of literature related to general transit and more specific bus ridership determinants, as well as LU.BE / TB relationships pertinent to this study and scale; Section 2 discusses case study selection, case description, as well as research method; Section 3 presents and compares model results, models fit, and discussion; Section 4 presents conclusions and policy implications based on model results and analyses. A final Section 5 discusses research limitations and possible future extensions.

Literature Review

The focus of this second investigation requires spanning various sub-branches in transportation literature. Specifically, bus and rapid-transit transit ridership determinants; land- use/built-environment and travel behavior interactions (LU.BE / TB); walkability; and walking- to-transit. These fields capture key theoretical and methodological aspects that inform this study, its quantitative measures and methods, and model specifications for key and control variables. It also helps in identifying appropriate modeling approaches and remediation protocols for previously identified statistical shortcomings. These literature fields are discussed below in turn, and conclude with a summary section.

72 Transit Ridership Determinants. An extensive literature on the factors that influence transit ridership exist, attending to different modes, units of analysis, and reflecting the importance that scholars, policymakers, and planners place on this topic (Ramos et al., 2015a,b). For a detailed discussion please refer to pages 34-35 in Chapter 3.

Bus Ridership Determinants at Stop Level. Some transit patronage studies have focused on ridership factors at the bus-stop level, which is the unit of analysis in this study and considered the most appropriate level of bus demand analysis (Kimpel et al. 2007; Peng and Dueker 1995; Furth et al. 2003; Chu 2004). A variety of factors have been associated with bus stop-level boardings, including socio-demographics in pedestrian catchment areas, transit level of service, street level built-environment quality, stop amenities, accessibility to employment and population, interaction with other modes, and competition with other modes (Chu 2004; Pulugurtha and Agurla 2012; Chakour and Eluru 2013). Distinct research methods have also been implemented in studying stop-level boardings, including both quantitative inferential and qualitative/ethnographic (Chu 2004; Loukaitou-Sideris 2007; Kimpel et al. 2007; Peng and Dueker 1995; Furth et al. 2003).

As related to weekday boardings Chu (2004) found a series of consistent predictors using a Poisson-count model. These include median household income, number of jobs, zero-vehicle households, share of persons under 18, share of persons 18-64, share of persons female, share of persons Hispanic, share of persons white; transit level of service; pedestrian factor, population up- and down-stream in a 1hr period; jobs up- an down-stream in a 1hr period. Pulugurtha and Agurla (2012) also assessed stop-level transit ridership models using similar demographic, socioeconomic, land use characteristics around each selected bus stop, and on-network characteristics. These latter authors also implemented a count-model and compared a distance- weighted and a proximity-based buffer approach. As noted by the authors the distance-decay function did not perform better than standard .25mile distance buffer. Significant explanatory variables were organized in three categories: Demographic and socio-economic that include population (by gender, ethnicity, and age); mean household income; auto ownership; and total employment. On-network variables found significant were road speed-limit; presence of median; One- or Two-way street; and number of lanes. Land-use measures found to be significant were

73 .25 buffer areas of residential, institutional, light-commercial, heavy-commercial, light- industrial, and heavy-industrial uses.

Using the city of Montreal as case study Chakour and Eluru (2013, 2016) framed their investigation from the perspective of the transit provider, considering the influence of transit system operational characteristics, transportation system infrastructure attributes, and built environment attributes on disaggregate stop level boardings and alightings, at various time periods and across three categories of ridership (high, medium, low). Their empirical analysis focused and involved estimating the effect of the built environment and urban design on ridership at a stop level using an ordered regression model. Stop-level operational variables found significant were headway (negative and highly significant); presence of rapid-transit, metro stations, and commuter trains during AM period within a 200mt buffer (positive and significant). The authors conclude that this pattern is indicative of significant inter- and intra- modal transit transfer activities.

In the same study Built-environment and infrastructure variables found significant were the presence of major roads around the stop (positive and significant) for off-peak night period and morning alightings; length of highway (negative and significant) only at off-peak night boardings; distance from CBD registers fewer alightings during off-peak night period. Land-use measures report that the area of parks has a negative and significant effect on boardings and alightings, but the number of parks (smaller in area) have an opposite positive and significant effect on boardings; the number of commerce also has a positive an significant effect whilst government and institutional activities report positive effects only in the AM peak-period. Residential areas have opposite and significant effects based on AM-PM periods that most likely reflect commute patterns; employment density at the TAZ level has negative effects on boardings and positive effects for alightings; resource and industrial areas have significant negative effects on ridership, particularly for boardings. Based on elasticity analyses the authors conclude that that ‘the most effective way to increase ridership is to increase public transport service and accessibility, whereas enhancements to land use have a smaller effect on ridership’ (Chakour and Eluru 2016).

74 Chakrabarti (2015) cross-sectional study of variation in average estimated stop-level line boardings across a set of time periods focused on service reliability in Los Angeles Metro bus system. His findings indicate that higher average service punctuality (schedule adherence) and lower variation in schedule deviation in time are associated with greater ridership, all else equal. Chakrabarti also reports that the effect of reliability on peak-period ridership is moderated by headway. That is, the demand for reliability is higher for lines with longer headways. Stop-level control variables used in this study were organized in four general groups: stop neighborhood built environment factors; stop neighborhood socioeconomic and demographic factors; bus line service quality factors; and other factors.

The set of explanatory variables specified in Chakraborti’s linear negative-binomial models, whose specification was informed in part by Chu’s (2004) study and by several transit- related works of other scholars, were average on-time performance, standard deviation of schedule deviation, population density, employment density, median income, transit availability (stops/stations within 0.5mile radius), line population density, employment accessibility, stops per mile, line type (service), headway, transit alternatives, and position of stop with respect to line. The indicator variable for ‘Rapid’ service bus lines was not significant in any of the time period models.

Carpio-Pineda (2014) explored the potential influence of alternative built- environment/spatial-integration measures that incorporate perceptual dimensions, space-syntax, on urban bus-stop ridership forecasting models. His study on Madrid’s urban bus network finds that spatial-integration measures improve predictive model explanatory power and confirms previous studies statistical significant findings related to population jobs, population reach, presence of retail and services, and transport network. Other explanatory factors such as crime and environmental attributes (Loukaitou-Sideris 2007), distance-decay accessibility-weighted dwelling units (Kimpel et al. 2007), and space-syntax indicators (Carpio-Pinedo 2014) have also been explored and found to be significantly associated to bus-stop boardings at different degrees of magnitudes.

75 In sum, bus-stop ridership factors are similar to those identified in general transit and rapid-transit patronage studies. The factors tend to organize around four general categories of explanatory variables related to socio-demographics, built-environment, transit service, and transit network intra- and inter-modal connectivity. Dependent variable measures tend to focus on survey-based estimates or actual counts of weekday boardings where APC technology and data are available. Recent studies further disaggregate this measure along distinct time-of-day and peak/off-peak time periods, and perform separate analyses for boardings and alightings.

Cross-sectional studies and linear multivariate regression models dominate this field of research, and more recent studies rely on negative-binomial distributions using log-link function. In regards to the problem of transit supply-demand endogeneity identified in previous rapid- transit ridership studies that have been conducted at the metropolitan, line, and station levels, Chu (2004) argues that bus service planning is focused on line-level and not on stop-level. Moreover, bus-stops exhibit high variability in terms of demand and thus argues that endogeneity is not a problem at this scale and in this line of research. Other bus stop-level studies presented in this review do not address endogeneity in their models. However, it appears that multi- collinearity issues have emerged in several studies as related to their explanatory variables and must be addressed in future bus stop-level studies.

Land-Use / Built-Environment and Travel Behavior Literature (LU.BE / TB). Numerous LU.BE / TB investigations have been conducted in the past few decades, registering more than 50 empirical studies, addressing a variety of travel outcomes and modes, and implementing a wide variety of methods of increasing sophistication (Ewing and Cervero 2010; Stevens, M. 2017; Handy, S. 2017; Nelson, A. 2017; Knaap et al. 2017; Manville, M. 2017; Renee et al. 2017). In addition various summaries and syntheses, and more recently meta- analysis and meta-regressions have been produced in an attempt to clarify a voluminous literature that has not been exempt from contradictory and/or non-conclusive results (Ewing and Cervero 2010; Boarnet & Crane 2001; Stevens, M. 2017).

In general mounting LU.BE / TB evidence points to statistically significant yet relative weak inelastic influence of individual land-use and built-environment measures as compared to

76 other socio-economic, transportation service levels, regional and destination accessibility, and network attributes and structure with respect to transit and other types of travel. Yet some scholars maintain that their aggregate effect is considerable and thus relevant for advancing sustainable development and transportation policy objectives (Ewing & Cervero 2010; Handy et al. 2007; Newman et al. 1995; Newman & Kenworthy 1999, Hong et al. 2014, among others).

Other scholars however call for a re-focus and re-prioritization in transportation, transit planning and operations, and LU.BE / TB research towards what they believe are more effective policy options related to service levels, fiscal disincentives and incentives, and/or transit network restructuring to better match existing decentralized, suburban metropolitan structures (Mees 2010; Thompson & Matoff 2003; Brown & Thompson 2008b, 2012a,b; Brown et al. 2014; Concas & DeSalvo 2014, among others).

Theoretical Frameworks. LU.BE / TB research often remits to travel behavior theory, which is informed in large part by micro-economic consumer choice theory and a general understanding that land-use and built-environment attributes affect travel behavior as mediators of the price of travel (see Boarnet & Crane 2001, Cervero 2002, Weinberger and Sweet 2012). Grounded in microeconomics, travel utility theory states that “individuals make rational travel choices to maximize their benefit (and minimize costs) on the basis of full knowledge of alternatives for any given travel behavior decision” (Weinberger and Sweet 2012; Ben-Akiva & Lerman 1985).

Other distinct theories are also being used to frame and study travel behavior, such as Alonso’s (1964) residential location theory based on trade-offs between transportation and housing costs within a limited budget constrain, applied more recently on contemporary poly- centric settlements (Concas & DeSalvo 2014). This approach has been used for both transit use and trip-chaining studies. Psychological, attitudinal, perceptual, life-style, among other qualitative factors have also been incorporated in travel research and have registered effects on travel behavior in addition to those documented for built-environment dimensions (Hong et al. 2014; Dill et al. 2014; Van Acker and Witlox 2007 ; Lindelöw et al. 2014). Cognitive psychology, transport geography, social psychology, social-ecological system frameworks, and

77 other behavioral sciences and theories are thus being integrated in a still growing and diversifying LU.BE / TB theory field (Van Acker and Witlox 2007 ; Lindelöw et al. 2014). These diverse perspectives have been recently categorized in four general theoretical frameworks for travel behavior research: (a) utility-maximizing theory, (b) activity-based approach, (c) theory of planned behavior, and (4) social-cognitive theory (Van Acker and Witlox, 2007).

From 3, to 5, to 7 Ds. In a seminal and influential study Cervero & Kockelman (1997) hypothesized and analyzed three set of explanatory vectors for their potential influence on vehicle miles traveled (VMT) and mode choice: 1-socio-demographics of trip maker; 2- household characteristics (size, vehicle ownership, income, tenure); and 3-built environment characteristics. Several land-use and built environment (urban design) attributes were condensed under three rubrics via a factor analysis that yielded the now familiar “3Ds”: 1- Density: (i.e population/acre, employment/acre, accessibility to jobs) ; 2- Diversity: (i.e land-use dissimilarity index, entropy index, etc.); and 3- Design: (i.e. street pattern [ intersections, freeway rate, arterial speeds, street widths]; pedestrian and cycling provisions [ proportion of blocks w/ sidewalk, trees, planting strip, mid-block crossings, intersection with signalized controls, bicycle lanes/acre ]; and site design [ commercial/service parcels with off-street parking, on-street parking, drive- ins, drive-through ]. In their study both discrete-choice and linear regression models that included built-environment variables yielded slightly higher explanatory power.

Ewing & Cervero (2010) updated and expanded Cervero & Kokelman’s (1997) original 3Ds scheme using a meta-analysis approach. Informed by multiple LU.BE / TB empirical studies the authors produced a summary on associations between built-environment attributes and three modes of travel (automobile, walking, and transit). Weighted elasticities were estimated from these studies and in turn updated the original framework for characterizing the built environment from the original 3Ds (Density, Diversity, and Design) into a five attribute set (5Ds) that takes into consideration larger-scale conditions at destination and distance to transit. As such a new updated rubric emerges: 1- Density; 2- Diversity; 3- Design; 4- Destination accessibility (i.e. typically based on gravity models that incorporate number of jobs as opportunities and time- distance as impedances); 5- Distance to transit (i.e. average of shortest street routes to transit or bus stop/station; or transit route density, distance between stops, or number of stations per area).

78 More recently Stevens (2017) updated these associations and elasticities using a more robust meta-regression protocol (Table 4.1).

Their results in general indicate Density measures based on population and/or employment, and commercial floor area are positively associated with walking and transit trips, yet report small and inelastic estimates. Diversity measures, such as land-use mix, report slightly higher elasticities for walk and transit modes and are also positively associated with more walking and transit patronage. Design aspects, in particular intersection density, registers the highest elasticity (0.39) for walk mode and is also positively associated with transit trips at a lower elasticity level (0.23). The more recent ‘D’ factors, Destination accessibility and Distance (inverse) to transit also report as significant yet relatively inelastic, and estimates for the inverse of distance to transit is stronger than that for walking (0.29 vs. 0.15 respectively).

LU.BE, Walking, and Transit Use. A variety of instruments and measures have been used in capturing the 5Ds and multi-dimensional indices have also been developed in an attempt to capture hypothesized synergistic effects (Table 4.2). Ewing et al. (2014) also provide a useful set of generalized relationships of built-environment and land-use attributes with a variety of travel behaviors based on a meta-analysis of multiple LU / TB studies (Table 4.3). Of particular interest in this investigation is transit use and walking, which appear to be influenced by land-use and urban design factors, and by socioeconomics in terms of frequency (Ewing et al. 2014; Litman, T. 2016).

In relation to transit patronage, LU.BE / TB research points to greater influence of larger- scale regional destination accessibility levels, service levels (i.e. frequencies, travel time), network connectivity, and special activity generators as compared to local land-use and built- environment features around stations (Chakour & Eluru 2016). Still local neighborhood-scale LU.BE factors register significant but relatively weaker magnitudes as related to transit use, for which some researchers describe them more as amenities than effective policy levers for increasing patronage and other policy goals (Brown & Thompson 2012a; Mees 2010, Concas & DeSalvo 2014 ). Out of the variety of measures that have been developed to capture their potential effect, the three built-environment factors that register the largest elasticities, as per

79 Stevens’ (2007) recent meta-regression, are distance to transit (0.29), intersection density (0.23), and share of four-way intersections (0.29; Table 4.1).

These often encountered results for land-use and built-environment factors in transit literature have prompted several scholars to call ‘attention from transit and land-use planners and policymakers to internal transit service factors and improvements in service quality’, which could include better route alignments, higher frequencies of service and speed, improved multimodal connectivity and convenience in transfers, among other factors (Brown & Thompson 2012a,b; Hong et al. 2014; Concas & DeSalvo 2014). These internal initiatives are mostly under the purview of transit agency and planners and thus could be implemented together with/or independent of other LU.BE policies and programs that seek to promote pedestrian and transit- friendly contexts such as New Urbanist schemes and TOD developments.

As noted by Concas & DeSalvo (2014):

“…the debate on the relationship between urban form and transit travel has shifted from the need to determine minimum density thresholds that support transit to the need to provide reliable information to guide decision makers about what mix of land-use policies would better promote transit use in an increasingly suburban environment and decentralized employment context. Land-use policies can be successful in increasing transit patronage; while population density is a factor in determining demand, targeting land-use policies affecting residential location decisions and development in suburban areas can be more effective”.

Also, in critiquing Ewing & Cervero’s (2001) recent meta-analysis Nelson & Niles (‘comment’ section, 2017) note:

“Public policy intervention to change regional travel patterns in North America typically begins by constructing new transit infrastructure, often a dedicated guideway with limited geographic reach. This investment is leveraged with government efforts to put more housing and shopping in a specific, limited set of

80 station areas. Meanwhile, the regional economy continues operating regionally. The “typical elasticities of travel with respect to the built environment” at best apply to a small fraction of an urban region. Success is likely to be limited by the large costs required to achieve significant change. Thus, as skeptics of the transit-oriented development paradigm, we find Ewing and Cervero’s work not disappointing, but confirming.”

Given these relatively recent criticisms and debates several urban and transportation scholars call for research and policy re-direction in LU.BE / TB studies, which potentially includes new theoretical frameworks, more focus on longitudinal studies, more attention to factors other than LU.BE features, inclusion of more perceptual studies, among other recommendations (Handy 2017; Manville 2017; Concas and DeSalvo 2016).

Yet this author believes there are still overlooked geographies and potential LU.BE / TB associations that remain unexplored, particularly related to inter-modal transit interactions that occur in large poly-centric agglomerations. This new focus might shed light and contribute to research and policy debates.

Many transit-focused LU.BE / TB studies have fixated in areas adjacent to transit stations (i.e. station pedestrian catchment area). To this author’s knowledge there has been no study that addresses the potential influence of LU.BE attributes around feeder bus stops on rapid-transit station boardings. As some US cities continue to grow in mostly suburban formats and/or decentralize (Ewing & Hamidi 2014) it is plausible to expect that intra- and inter-modal transfer events and bus access to rapid-transit service would increase as well, thus playing a larger role in the systems’ overall effectiveness and performance. Given these circumstances it is logical and useful to raise our gaze and look at these peripheral locations not only for their potential influence on line-haul rapid transit patronage, but also to recognize the important role of intermodal network connectivity in advancing more sustainable urban transportation systems and cities.

81 Table 4.1 Summary of Associations and Weighted Elasticities of VMT*, 1, 2, Walking, and Transit with Respect to Build-Environment Variables. Sources: Ewing & Cervero (2010); Stevens, M. (2017) 1, 2

Land-Use and Built- Number of studies with Environment controls for self- Attributes Total number of studies selection Weighted average elasticity (e)

5 D's : Measures VMT Walking Transit total VMT Walking Transit total VMT* VMT 1 VMT 2 Walking Transit Density Household/population 9 10 1 1 0 0 -0.04 -0.22 -0.10 0.07 0.07 density Job density 6 6 6 1 0 0 0.00 -0.07 -0.01 0.04 0.01

Commercial floor area 0 3 0 na 0 na 0.07 Diversity Land use mix (entropy 10 8 6 0 1 0 -0.09 0.11 -0.03 0.15 0.12 index)

Jobs-housing balance 4 4 0 0 0 na -0.02 0.00b 0.19

Distance to store 0 5 0 na 3 na 0.25 Design Intersection/street 6 7 4 0 2 0 -0.12 -0.14a 0.39 0.23 [urban] density

% 4-way intersections 3 5 5 1 1 2 -0.12 -0.06a -0.06 0.29 Destination Job accesibility by auto 5 0 0 0 na na -0.20 -0.20a accesibility

Job accesibility by transit 3 0 0 0 na na -0.05 0.00b

Distance to downtown 3 0 0 1 na na -0.22 -0.63 -0.34

Job within one mile 0 3 0 na 0 na 0.15

Distance to Distance to nearest 6 3 3 1 2 1 -0.05 -0.05a 0.15 0.29 Transit transit stop

sub-totals 55 54 25 134 5 9 3 17 percentages 41% 40% 19% 100% 9% 17% 12% 13%

Notes: * VMT = vehicle miles travelled 1. VMT elasticities estimated from studies that controlled for residential self-selection and that reported statistically significant differences in size from elasticities estimated from studies that did not controlled for self-selection. 2. VMT elasticities estimated from studies that do not controlled for residential self-selection.

82 Table 4.2 From 3Ds to 5Ds to 7Ds: Definitions and Measurement of Typical Variable Measures Used in Land-Use / Travel-Behavior Studies Sources: Ewing & Cervero (2010); Ewing et al. (2015); Litman, T. (2017a,b)

“D” Variables Definition Phenomena of Interest

1 Density Measured as the variable of Population, dwelling units, employment, interest per unit of area. The building floor area, or something else. area can be gross or net. Population and Employment are sometimes summed to compute overall “activity density” per areal unit.

2 Diversity Measures pertain to the number Entropy measures of diversity, wherein low of different land uses in given values indicate single-use environments and area and the degree to which higher values more varied uses; jobs-to- they are represented in land housing or jobs-to-population ratios are less area, floor area, or employment. frequently used.

3 Design Includes street network Average block size, proportion of four-way characteristics within an area, intersections, and number of intersections per which vary from dense urban square-mile; sidewalk coverage (share of block grids of highly interconnected, faces with sidewalks), or number of pedestrian straight streets to sparse crossings, street trees, or other physical suburban networks of curving variables that differentiate pedestrian-oriented streets forming loops and environments from auto-oriented ones. lollipops.

4 Destination accessibility Measures of ease of access to Distance to CBD, number of jobs or other trip attractions. It may be attractions reachable within a given time, the regional or local (Handy, 1993). gravity model of trip attraction.

5 Distance to Transit Measured as the average of the Alternative measures include transit route shortest street routes from density, distance between transit stops or the residences or workplaces to the number of stations per unit area. nearest rail station or bus stop.

6 Demographics Controls for confounding influences in travel studies.

7 Demand Management Mobility Management. Policies and programs to reduce automobile travel and encourage use of alternative modes. For example: flextime, bicycle improvements, bike/transit integration, park & ride, traffic calming, car-sharing, transit improvements, parking supply and cost, etc.

83 Table 4.3 Generalized Associations between Socioeconomic Characteristics, Land-Use and Built-Environment Attributes, and Travel Behavior. Sources: Ewing et al. (2014); Litman, T. (2013, 2017a,b)

Travel Behavior Association(s)

Primarily a function of socioeconomic characteristics of Trip frequency travelers and secondarily a function of the built environment.

Trip length Primarily a function of the built environment and secondarily

socioeconomic characteristics.

Shorter at locations that enjoy higher accessibility, exhibit

higher densities and feature mixed uses. This observation

applies to both home-based trips from different neighborhoods

well as to trips to non-home destinations in different activity

centers.

Destination accessibility is the dominant environmental

influence.

Mode choice Depends on both, socioeconomic and built-environment, though probably more on socioeconomics.

VMT and VHT Also depend on both socioeconomic and built environment characteristics.

Transit use Varies primarily with local densities and secondarily with the degree of land-use mixing.

Some of the density effects is in part due to better walking conditions, shorter distances to transit service, and less free parking.

Walking Varies with both the degree of land-use mixing and with land densities.

General Design has a more ambiguous relationship with travel behavior and any effect is likely to be a collective phenomenon involving multiple design features, in addition to possible interactions with other ‘D’ variables.

84 Walkability and Walking-To-Transit Research. The underlying hypothesis in this second study posits that feeder bus-stops located in more pedestrian-friendly contexts should generate on average higher number of boardings associated to rapid-transit station access trips, all else equal. Within a utilitarian model of travel behavior a more pedestrian-friendly environment would represent lower travel costs (i.e. more comfort, safety from cars and crime, reduced travel time, or reduced access trip distance) to pedestrians that would otherwise consider other access modes to reach rapid-transit service (i.e. car, bicycle), choose another line-haul mode to reach their desired destination (i.e. automobile trip), or decide not to make the trip at all. On the other hand, less pedestrian-friendly environments within feeder bus-stop pedestrian catchment areas would represent higher impedance (lower utility) and would reflect as lower bus access trips to rapid-transit stations. Hence walking, and walking-to-transit are key concepts in this investigation and further discussed in this section.

Walking. Considered by some scholars the most fundamental transport mode in accessible and sustainable cities (Lindelöw et al. 2014; Litman, T. 2016) citizens of all ages and socio-economic backgrounds walk to a multitude of destinations for a variety of purposes as they engage in day-to-day sustenance, recreational and/or social activities. In addition environments suitable for walking constitute a vital part of a successful public transport system and potentially an attractive urban environment as well (Lindelöw et al. 2014).

Litman (2011, 2013) defines walkability as the existence of walking facilities and the quality of walking condition (safety, comfort, and convenience); criticizes the characterization of walkability as a ‘minor’ mode in engineering and transport planning circles; and argues quantitatively and theoretically in favor of walking mode. Litman also notes that although walking represents less than 3% of total miles travelled it accounts for 18% of total travel time and 25% of total trips in a survey conducted in the United Kingdom. In addition as compared to in-vehicle-time (IVT) walking as access mode to bus-stop or rapid-transit station is weighted on average by a factor of 2 in terms of IVT time (Balcombe et al. 2004). In sum, walking is a critical component of the transport system and improvements can provide significant benefits to society (Litman, T. 2016). As such, Litman argues for walking to receive an appropriate share of

85 transport resources and that it be accounted for in transportation surveys, in addition to transit and automobile modes.

Every trip usually begins by walking (Mees 2010; Litman, T. 2016), including access to rapid-transit portals and access to local feeder bus-stops. Walking is the dominant mode of access to bus service and to a majority of rapid-transit systems (LACMTA 2012; Litman, T. 2016; Coffel et al. 2012). However, as noted in the introduction to this manuscript, certain spatial, socioeconomic, and development characteristics can create conditions that allow for other access modes such as bus, car, and/or bicycle to play an equal or bigger role than walking in accessing rapid-transit stations.

Knowledge about how land-use and built-environment features might affect travel behavior, and walking in particular, is thus central to this investigation. This section discusses scientific literature related to walking and walking-to-transit. To recapitulate, the main hypothesis in this study states that key land-use and built-environment attributes, or an integrated multi-dimensional measure of such, influence walkability levels within pedestrian catchment areas of transit portals, whether at bus feeder stops or at rapid-transit stations. These walkability levels in turn potentially affect the propensity of riders to walk to feeder service and/or rapid- transit portals, thus potentially affecting levels of single mode trips and/or chained multi-modal trips such as the one examined in this investigation.

Walkability. As noted by Weinberger and Sweet (2012) walkability measures the opportunity to walk, rather than actual walking behavior. Several definitions have been posed in the LU.BE / TB literature, and these vary based on research purpose, walking contexts, unit of analysis, or simply on available data and methods. For example Schlossberg & Brown (2004) define it as a simple dichotomy of pedestrian-friendly or pedestrian-hostile network, and incorporate a variety of factors and measures related to population density, land-use mix, roadway connectivity, and building design. Rodriguez & Joo (2004) interpret walkability within a mode choice model as an objective measurement of built environment attributes related to topography, sidewalk availability, residential density, and presence of walking and cycling paths.

86 Iacono et al. (2010) define walkability as non-motorized accessibility using a gravity model while Diao and Ferreira (2010) define it as a pedestrian environment that can reduce travel costs of walking evaluated within a general hedonic price model of housing. In this latter study a linear combination of objective factors that include average sidewalk width, percent of roads with curbs, percent of roads with sidewalks, and a gravity-based measure of accessibility to jobs and other eight non-work destinations are distilled in a factor analysis protocol derived from twenty-seven built environment variables.

Capp and Maghelal (2011) offer a simplified definition: “… [area] more or less suitable to walk” and focus their research in identifying and categorizing walkability constructs. They identify two main categories of factors: 1- socio-psychological, and 2- environmental. For environmental factors (i.e. land-use and built-environment), which is the focus of this study, they define a rubric of three sub-categories for walkability measures: objective (quantifiable with standard methods and replicable), subjective (quantifiable but may or may not be replicable), and distinctive (non-standard methods and not replicable). They also identify and organize a large number of indicators that include distance to destinations, presence and characteristics of sidewalks, type and characteristics of roads, number and type of intersections, number and speed of vehicles, lateral separation of pedestrians from roadway, demographics, land-use attributes, as well as safety, comfort, and convenience considerations.

Perhaps the most comprehensive yet challenging definition to operationalize is Southworth’s (2005): “…the extent to which the built environment supports and encourages walking by providing for pedestrians comfort and safety, connecting people with varied destinations within reasonable time and effort, and offering visual interest in journeys throughout the network”. This comprehensive definition incorporates physical, perceptual, spatial, and aesthetic/phenomenological dimensions and points towards the need for multi-dimensional instruments or heuristics in order to better operationalize walkability in research.

In a recent literature review Weinberger and Sweet (2012) note that several databases describe walkability using concepts such as ‘density of amenities’ within a specific area, land-use mix, together with other features of the built environment and point to the availability of a

87 variety of walkability metrics that facilitates comparison of areas. The authors also note that walking behavior studies has mostly been framed within the 5Ds taxonomy of density, diversity, design, destination accessibility, and distance to transit originally postulated by Cervero and Kockelman, later extended by Ewing. This framework has been employed in several LU.BE / TB studies where significant yet relatively weak correlations were identified with travel outcomes, including active transportation such as walking and bicycling (Table 4.4).

Table 4.4 Key Built Landscape Attributes Associated with Active Transportation: Walking and Bicycling

Attribute Study Built Environment:

Density of Built Environment Craig et al. (2002); Saelens et al. (2003); Saelens & Handy (2006); Weinberger and Sweet (2012); Ewing & Cervero (2010), among others. Diversity of Land Use Craig et al. (2002); Saelens et al. (2003); Saelens & Handy (2006); Weinberger and Sweet (2012); Ewing & Cervero (2010), among others Distance to Transit De Bourdeaudhuij et al. (2003); Schlossberg et al. (2007); Ewing & Cervero (2010) Design of Street Connectivity and Built Environment Saelens et al. (2003); Weinberger and Sweet (2012); Ewing & Cervero (2010) Destination Accessibility Greenwald & Boarnet (2001); Hoehner et al. (2005); De Bourdeaudhuij et al. (2003); Ewing & Cervero (2010); among others.

Other Non-built: Demographics Frank & Pivo (1994); Forsyth, A. (2007); Forsyth & Southworth (2008); Bagley & Mokhtarian (2002) Crime Craig et al. (2002); Loukaitou-Sideris, A. (2007) Travel Demand Management Shoup (2004); Gillen (1977) Individual Preference Bagley & Mokhtarian (2002); Mokhtarian & Cao (2008) Space-Time Constrains (feasibility) Lindelöw et al. (2014)

Walkability Measures. Weinberger and Sweet (2012) compared a set of walkability metrics on the basis of their ability to predict walking behavior (walk mode share for a variety of purposes). Their results indicate that Walk Score®, which is widely available open-source metric, outperformed other objective measures in their study for variety of trip purposes and concluded that it is a reasonable heuristic to assist in assessing potential trip impacts. This result is supported by other recent studies. Hirsch et al. (2013) found that higher scores of neighborhood

88 Walk Score®, and Transit Scores® are associated to lower odds of not walking to transport and more minutes/week of transport walking; Duncan & Aldstadt (2013) compared Walk Score®, and Transit Score® to several objective GIS measures of neighborhood walkability and transit availability based on 400-800m buffers. The authors found positive and significant correlations for all measures at the 800m level and for most measures at the 400m level, indicating that Walk Score®, is a good and convenient indicator that captures certain aspects of walkability and Transit Score® a good indicator of transit availability. Yet the authors clarify that Walk Score®, work best at larger scales.

A general explanation of both Walk Score®, and Transit Score® methodology follows:

“Walk Score measures the walkability of any address using a patented system. For each address, Walk Score analyzes hundreds of walking routes to nearby amenities. Points are awarded based on the distance to amenities in each category. Amenities within a 5 minute walk (.25 miles) are given maximum points. A decay function is used to give points to more distant amenities, with no points given after a 30 minute walk. Walk Score also measures pedestrian friendliness by analyzing population density and road metrics such as block length and intersection density. Data sources include Google, Education.com, Open Street Map, the U.S. Census, Localeze, and places added by the Walk Score user community.”

“Transit Score is a patented measure of how well a location is served by public transit. Transit Score is based on data released in a standard format by public transit agencies (GTFS format). To calculate a Transit Score, we assign a "usefulness" value to nearby transit routes based on the frequency, type of route (rail, bus, etc.), and distance to the nearest stop on the route. The "usefulness" of all nearby routes is summed and normalized to a score between 0 – 100.”

Walk Score®, (https://www.walkscore.com/transit-score-methodology.shtml; accessed 11/19/2017)

Similar to Litman (2011, 2013) Tal and Handy (2012) define walkability along two parallel branches: 1- as the quality of the environment, including safety, comfort, and pleasure experienced by pedestrians; and 2- echoing Iacono et al. (2010) the ability of pedestrians to access their destinations as a function of: a- proximity to destinations; b- directness of route (network connectivity); and c- the resulting potential for interaction measured as the cost of

89 reaching potential destinations using a Hansen model (i.e. travel distance or number of destinations within a distance threshold).

These authors identify two main factors in their analysis of walkability. One is pedestrian accessibility and the other network connectivity using a variety of objective GIS measures. They conclude that higher walkability is associated with healthier communities as result of higher levels of physical exercise, with walking as a mode of transportation and access to transit, and emphasize that walkability is a major component in urban planning efforts to reduce dependence on automobiles.

Dill et al. (2014) relied on structural equation modeling and operationalized a construct of walkability that incorporates psychological (perceptual) endogenous factors and built environment exogenous factors. Multiple measures were organized in five themes related to build environment, socio-demographics, attitudes, social norms, and perceived behavioral control. The authors conclude that attitudes may be as important as infrastructure and built- environment conditions to increase walking and biking. They also conclude that both built- environment attributes and demographics are important influences in behavior, largely because they influence people’s perceived behavioral control (PBC) and attitudes, which in turn help predict how often they walk or bike home. That is, PBC is the most important factor in predicting walking, and also the most influenced by the built environment. As such these authors contribute a new causal and theoretical framework in walkability studies and conclude that changes to the physical environment in the form of street network design, land use patterns, and pedestrian facilities may be enough to encourage walking.

In sum, walkability is an important multi-dimensional phenomena that is influenced by built-environment, demographic, and psychological, perceptual, and attitudinal factors. Research points to population density, network connectivity, and density of activity opportunities as key built-environment influences, together with non-built-environment factors associated with safety (from crime and traffic) and individual psychological/attitudinal factors. Publicly available multi-dimensional walkability indices exist and have been utilized in scientific research with significant results as compared to other bespoke GIS-based objective measures of walkability.

90 Yet, in the case of Walk Score®, it must be noted that the measure does not addresses a key “D” factor, distance-to-transit, nor attitudinal/psychological factors.

Walking to Transit. Recent research that focus on walking-to-transit (i.e. walking to rapid-transit stations and/or bus stops) point to minimizing distance and time as the most important factors in determining access routes for pedestrians, whilst safety from traffic and crime, attractiveness of route, sidewalk quality, and reduced wait at traffic lights are secondary but relevant factors (Agrawal et al. 2008; Park et al. 2014; Chalermpong and Wibowo 2007; Yang et al. 2013). Using a system of simultaneous equations to account for simultaneity of transit supply and demand, and factor analysis to capture built environment characteristics within a manageable number of variables Estupiñan et al. (2008) found that environmental supports of walking (i.e. built environment characteristics) and personal and environmental barriers to automobile use were strongly associated to BRT ridership at the stop level. These authors recommend interventions to the built environment in support of pedestrian access to BRT service as a practical and viable policy.

Crowley et al. (2009) focused on examining how variations in walking distance to transit relate to mode choice, vehicle ownership, and auto use and looked at temporal changes in built environment characteristics associated to TOD developments and how it affects transit patronage. Specific factors documented were walkability to transit at both ends of the trip (distance), built-environment characteristics (TOD characteristics: density, land-use mix), socioeconomic characteristics of trip makers (auto ownership, possession of driver’s license), and elements of transportation supply (parking availability and out-of-pocket expenses). They found that built environment characteristics influence accessibility and perceived convenience of walking, and the most important single factor in explaining variations in transit use is automobile availability, a result that corroborates other LU.BE / TB station-level ridership studies.

Crowley et al. (2009) also evaluated the 3Ds in their study areas (density, diversity, and pedestrian oriented design), primarily focusing on local densities and secondarily at mixed land uses, network of streets and walkways (whether they were direct, well-connected, and safe), and at physical, psychological, and symbolic barriers to walking. Based on their results the authors

91 emphasize that good-quality transit (fast, frequent, reliable, direct, and inexpensive service) is necessary but not sufficient to achieve mode shift from auto to transit and state that for transit to be relevant and competitive, and to achieve mode shifts good-quality transit must be easily accessible at both trip ends. The authors suggest to minimize walking distance at trip origins and destinations. They also noted modal differences in walking behavior as subway users are willing to walk further than to buses.

Crowley et al. (2009) main conclusion is that maximizing subway ridership requires that the development be more concentrated within a convenient walking distance of transit (400m ideally). They also noted that even in suburbs once recognized as auto-oriented neighborhoods, careful design and implementation of TOD in terms of density and transit accessibility can lead to higher transit mode shares.

However, none of the previously referenced studies address the potential influence of LU.BE attributes around feeder bus-stops. In studying distance-based interventions to support walking to light-rail transit stations Maghelal et al. (2011) found that relevant constructs of the built environment vary based on distance of walking. The authors documented several variables in relation to the built environment surrounding light rail stations. Thirty (30) independent variables in total were documented under 11 distinct categories: sidewalk, roads, intersection, vehicle (traffic), pleasantness, density, safety, destination density, street lateral separations, land- use, and station infrastructure. Based on an exploratory principal component analysis four specific factors were identified; 1-vehicle oriented design, 2-density, 3-diversity, and 4-walking- oriented design. Interestingly they found that increase in density does not necessarily increase walking share to transit as there is a mediating effect of parking at transit stations on population density.

Park et al. (2014) aimed to identify the determinants that influence walking and biking to transit stations using two separate binomial logit models. The study was based on a survey of 277 pedestrians and 280 bicyclists using forty travel, socio-economic, and built-environment variables and both aggregate (neighborhood) and disaggregate (individual traveler) data. The variables were trip distance, car availability, proximity to auto-friendly streets (>35mph), trip

92 purpose, race, and intersection density. Their main findings are that travel distance is a significant deterrent of walking; a conclusion that supports compact TOD development and concurs with that of Crowley et al. (2009) and Estupiñan et al. (2008).

Park et al. also emphasize the need for micro-level data to improve the explanatory power of the tested built-environment variables and that living near a street with fast-moving vehicles might influence mode choice in two ways: as a barrier and thus discourage walking, and by providing more easy driving access to stations and thus encouraging automobile use an discouraging walking access. They also note that some travel and socio-economic variables have few policy implications (i.e. trip purpose, gender, race) and that built-environment attributes can be improved via public investment and may be more important for policy makers. They also suggest that the influence of built-environment variables seems relatively low and posit that the inability of large-scale aggregate measures in capturing quality of street-level walkability as a possible explanation.

Jiang et al. (2012) explored BRT walk access patterns in rapidly urbanizing China, and its relationship to station context, corridor context, and the distance people walk. Their main conclusions are that people walk farther to BRT stations when the walking environment has certain features: median transit-way station location, shaded corridors (i.e. presence of trees), and busy and interesting streetscapes (land-use mix and active streets). On the other hand, they find that trip and trip-maker characteristics have a relative minor role in defining BRT walk access distance. Based on these findings the authors conclude that there is a need for flexible station pedestrian catchment area definitions in identifying TOD opportunities and estimating system demand.

Jiang et al. also notice that walking distances varies across cities and countries. As such, individual station catchment areas may be contextual and culture-specific. Based on their finding that highly walkable station surroundings will increase walking distance, all else equal, the authors suggest that the expected distance that people will walk to access/egress the system will increase by reducing the real and/or perceived time (disutility) of the walk. As such a feasible

93 policy for transit investment decision-makers would be “providing a few critical pedestrian access routes to stations as a cost effective way to enlarge catchment areas”.

El Geneidy et al. (2014) further explore some of the issues brought forward by Jiang et al. First they recognize the importance of the percentage of population served by a transit system in a metropolitan area as a key performance measure; and how heavily it depends on the definition of service area. El Geneidy et al. also affirm that not all transit portals are alike and therefore each has its own distinct service area. They further assert that the definition of service area should be related to the type of service being offered, its frequency, and its reliability, in addition to attributes of people and places served.

Based on the results of their multi-level multivariate analyses they conclude that pedestrians seek to minimize time and distance to transit stop as a primary factor, followed by individual characteristics, area characteristics (built-environment), transit route features, and weather-temperature. Household income and blue-collar neighborhoods, as well as vehicle availability are negatively associated with propensity to walk to transit, a result that concurs with other LU.BE / TB research on rapid-transit station patronage. Population and dwelling density, as well as education level are positively associated with propensity to walk, but not distance. Also worth noting is that El Geneidy et al. empirically determine that the 85th percentile of transit users walk 524m to bus service for home-based trip origins and 1259m for home-based commuter rail trip origins. As noted by the authors, these figures are clearly larger than the traditional 400m-800m buffers used in many academic research and professional work. In sum, minimizing walking time and distance to transit portals, whether for reaching a bus-stop or rapid- transit station are two key factors to consider in future research addressing access to transit.

Literature Review Summary. External factors such as population and employment levels, and the presence of special activity centers and/or land-use mix near bus stops influence bus stop-level ridership, as are some socioeconomic characteristics of the local population such as household-level vehicle accessibility, household income, non-white racial/ethnic status, among others. It is also clear that internal factors such as bus line service frequencies, fare levels, and transit network connectivity (i.e. number connecting bus routes and/or proximity with metro

94 stations or other bus tops) are also important contributors to stop-level ridership. Furthermore, other stop amenities (i.e. shelter, benches, real-time information), line position, crime levels, among others also have been shown to have significant effects, both positive and negative on the number of boardings.

However, none of these studies have addressed potential associations between bus-stop LU.BE factors and the number of boardings associated with access trips to rapid-transit stations. It is not clear either whether local bus-stop factors alone or in combination with more distant factors associated with rapid-transit stations service levels or destination accessibility levels influence patron’s behavior. To this author’s knowledge no previous study has focused on these inter-modal associations nor on the potential combined effects of LU.BE attributes around bus- stops and rapid-transit stations’ service and network attributes. This would be particularly relevant for rapid-transit systems that register a high share of bus access events, which tends to occur in large, poly-centric urban/suburban agglomerations in the United States. Likewise results from this study can help in the development of predictive models for bus access events at rapid- transit stations, which tend to be understudied and under-performing as compared to other access modes (i.e. walk, bicycle, automobile) and whose predictive models yield relatively low explanatory powers (Coffel et al. 2012).

Unit of Analysis and Direct-Demand Models (DDM). Studies that focus on station- or stop-level ridership are of particular interest in this investigation. It is at the station-level and bus-stop level where both local transportation supply and travel demand forces meet. It is hypothesized in this investigation that walkability levels around bus-stops influence the number of boardings associated with access to rapid-transit service. That is, higher walkability levels at bus-stops will result in higher number of boardings, all else equal, at rapid-transit stations.

In addition it is also hypothesized that rapid-transit station service levels and stations’ destination accessibility levels also influence travel decisions of patrons that engage in multimodal bus / rapid-transit travel, as is the behavior of interest in this investigation. Higher relative destination accessibility and service levels at a station should also yield higher boardings

95 at linked bus stops, all else equal. Hence bus-stop/rapid-transit station pairs constitute the basic unit of observation in this investigation.

Coupled bus-stop/rapid-transit station pairs are ideal units for the analysis of inter-modal access interactions and for analyzing the potential effects of LU.BE attributes at each location (Figure 4.1). Bus access to rapid-transit service is one such behavior where the riders experience in sequence distinct services and built landscapes during his/her multi-modal chained trip, and theoretically would consider these variegated factors in their travel decisions (assuming a utilitarian model of travel decision-making where patrons seek to maximize their utility and minimize cost).

Direct-Demand Models (DDM). See Chapter 3 (p.35) for a detailed discussion.

Research Design and Methods

Research Design. The research design for this study is structured around a single case- study: Los Angeles multi-modal and multi-agency metropolitan transit system. For a detailed discussion please refer to Chapter 2 (pp.12-14).

Definitions. In this study rapid-transit refers to high-capacity, high-speed, and high service (10-15min frequency or less) public urban passenger transportation that mostly operates on exclusive right-of-ways, and excludes commuter rail. Feeder bus-stops include all stops across a wide bus service spectrum (i.e. local, limited-stop, express, rapid, and commuter) that registered at least one boarding associated with access to a rapid-transit station as per LA Metro on-board survey (n=5,550; PTV NuStats-LACMTA 2011;). Feeder bus-stops pedestrian catchment areas were defined by mutually-exclusive Thiessen polygons in ArcMAP with a 0.25mile buffer clip.

Case Selection and Description. Please refer to Chapter 2 “CASE STUDY SELECTION AND DESCRIPTION”.

96

Figure 4.1 LA Metro Rapid-Transit Network, SCAG Multi-Agency Bus Network, and Bus Stops That Registered at Least One Boarding Associated with Access to a Rapid-Transit Station (Partial Map of SCAG Region; Year 2011) Sources: Bus stops x-y coordinates extracted from ‘Metro 2011 On-board Survey’ digital database (PTV NuStats LACMTA (2012) using STATA v.12 and displayed in ArcMAP GIS v.10.3.1; Metro ‘Bus & Rail GIS Data (2011)’ (http://developer.metro.net:80/introduction/gis- data/download-gis-data/; retrieved December2015 from ‘Internet Archive – Wayback Machine’; https://web.archive.org/ ); and SCAG ‘Bus Routes (2007)’ shapefile (retrieved December 2015; http://gisdata.scag.ca.gov/Pages/GIS-Library.aspx). 97 At the bus-stop level, which is the main geography of interest in this second study, substantial variability exists in the number of estimated boardings associated with access to rapid-transit service. Survey-based estimates report a mean of 42 average weekday boardings, a minimum of 1, and maximum of 1432 (SD=72.5). As reflected in Figure 4.2 the distribution of this outcome measure reflects a highly skewed logarithmic distribution typical of count non-zero integer measures.

.04

.03

.02

Density

.01

0 0 50 100 150 200

Figure 4.2 Bus Stop Average Weekday Boardings Associated with Access to Rapid- Transit Stations (2011; excluding extreme outliers [SD>3]) Source: ‘Metro 2011 On-board Survey’ digital database (PTV NuStats-LACMTA 2012; estimated by author in STATA v.12 using ‘svyset’ command and ‘subpopulation’ calibration).

Methods and Hypotheses. The methods in this second study are quantitative and rely on a multi-level generalized linear equations set within a generalized structural equation modeling framework (ML-GSEM-NBREG). Model specification relies on travel theory, literature review, and this author hypotheses. Because the phenomenon of interest clusters around specific transit network nodes (rapid-transit stations), and because this author posits that significant variance 98 exists among these, the data is analyzed within a multilevel structure. The first-level unit of analysis is the feeder bus-stop/station pair (observation level; n=3,498); and the second-level unit of analysis is the rapid-transit station (n=102). The outcome variable of interested is the survey- based estimate of bus-stop boardings associated with access trips to rapid-transit stations. The main explanatory variable of interest is Walk Score® and alternatively sub-components. It directly addresses the main research question that guides this second study and represents a multi-dimensional measure of walkability around feeder bus-stops.

Hypotheses. The key hypothesis in this investigation regards the potential influence of LU.BE attributes around feeder bus stops on the number of access trips to rapid-transit stations. Following previous studies’ finding of significant associations between LU.BE attributes and higher number of boardings at the rapid-transit station level this author posits that a similar association exists as related to areas surrounding bus stops. In addition it is also hypothesized that the particular travel behavior under study (bus access portion of a longer multi-modal trip) is also influenced by factors nested at the station-level. Specifically the destination accessibility level provided at the station.

The interactions within bus-stops and rapid-transit station pairs in the access portion of a longer multi-modal transit trip are hereby conceptualized to be similar to those based on the first two steps of the 4-step Urban Transportation Modeling System (UTMS), where the potential for trip production in zones i (P i) and trip attractions in destination zones j (A j) are calculated based on population and employment levels, respectively, and the potential flow between ‘ij’ pairs is estimated on a modified Newtonian gravity model. Similarly bus-stop attributes associated with boardings (trip production) and factors at rapid-transit stations associated with trip attraction (destination accessibility) are considered in this study within a direct-demand multivariate regression context in explaining the number of average weekday access trips within bus- stop/station pairs (AWT b.s -> rt.s ; Equation 2).

P i  A j ⸫ [ AWT b.s -> rt.s = f ( [ bus_stop ] i , [ rt.station ] j ) ] GSEM Eq. (2)

99 Where […] GSEM stands for a ‘generalized structural equation modeling’ overarching framework, as discussed below.

An ideal measure of destination accessibility for rapid-transit stations, which has been incorporated in recent DDM studies is a variant of Hansen’s gravity-model (1959) where both transit network and land-use attributes are considered with a mediating non-linear impedance factor (i.e. distance or time). However, the necessary data for constructing this measure was not available for this study. Instead a latent construct for stations’ relative destination-accessibility level is proposed, thus the use of SEM as an overarching modeling framework.

A set of stations’ network, service level, and spatial context measures partially informs the latent ‘STATION’ level-2 variable within the SEM model. Because the response variable exhibits a highly eschewed logarithmic distribution, typical for count measures such as boardings, and because of the nested multi-level data structure a Multilevel Generalized Outcome Model (GSEM-ML) is implemented with a negative-binomial distribution family and log-link function regression component (NBREG; Hilbe 2011).

GSEM-ML-NBREG Model Specifications and Variables Description. Based on the discussed literature review related to bus-stop and station-level ridership determinants, LU.BE / TB studies and walkability literature, and this author’s hypothesis of the combined multi-scale influence of station- and stop-level factors on the outcome of interest a set of explanatory variables and latent variable components are specified and described below for each level of analysis (Table 4.5).

Level-1 Bus-Stop Level Response and Explanatory Variables. Selection of stop-level ridership factors is informed by the four main vectors of explanatory variables identified in the literature review for bus transit ridership determinants and from stop-level demand forecasting studies: 1-socioeconomic (SE), 2-land-use/built-environment (LU.BE), 3-transit service (TS), and 4-network attributes (NT).

Outcome Variable:

100 BS0_AWBi Bus-stop average weekday boardings associated with access to rapid- transit station.

Socio-Economic (SE): yPOP_10 Population levels within bus-stop pedestrian service area. Higher population levels are expected to generate more trips as it is an essential trip generation component in travel. yJobs_11 Number of jobs within bus-stop pedestrian service area. Higher number of jobs is also expected to generate and/or attract more trips as it is a proxy for potential socio-economic activities in the city’s plexus. yVeh_HU Average number of vehicles per occupied housing unit within pedestrian service area. In the U.S. higher levels of automobile availability are associated with lower rates of transit travel given that the built- environment is more automobile-oriented and transit is usually considered an inferior good in the travel market. Higher rates of vehicles per housing unit is expected to be negatively associated with the number of bus-stop boardings.

Tot_0veh Total number of occupied housing units with zero vehicles within bus-stop pedestrian service area. Households with zero vehicle availability are more dependent on transit service and thus expected to be positively associated with higher bus-stop boardings.

Perc_Hisp Percentage of resident population that self-identify of Hispanic ethnicity within bus-stop pedestrian service area. In the Los Angeles region Hispanics register higher rates of transit patronage relative to other ethnic groups, as well as lower socio-economic indicators and driving license tenure. These circumstances characterize this group as being more transit dependent than others and thus expected to register a positive association with bus-stop boardings.

Land-Use and Built-Environment (LU.BE) Index and Sub-components:

Walk_Score Walk Score®, multi-dimensional walkability rank for bus-stop locations. For a detailed explanation refer to page 89 in this Chapter 4.

101 Walk Score®, sub-components:

Avg_block Average block length (meters) within bus-stop pedestrian service area.1; also serves as proxy for distance to bus-stop in some models. Intersecti Intersection density within bus stop pedestrian service area.1 Culture_Sc Score based on walking distance from bus stop to cultural destinations.1 Dining_and Score based on walking distance from bus stop to dining and drinking destinations.1 Errands_Sc Score based on walking distance from bus stop to service-type destinations.1 Grocery_Sc Score based on walking distance from bus stop to grocery destinations.1 Parks_Sc Score based on walking distance from bus stop location to parks.1 Schools_Sc Score based on walking distance from bus stop location to schools.1 Shopping_S Score based on walking distance from bus stop location to shopping destinations.1

SpeGen_LA Number of special activity generators within bus-stop pedestrian service areas that locate within the political boundaries of Los Angeles County (i.e. Schools, University Campus Hospitals, Sports/Entertainment arenas, among others). The presence of special activity generators near rapid- transit stations has been associated with higher boardings and is also hypothesized in this study to be positively associated with higher bus-stop boardings.

Bus Stop Service Level and Network Attributes (TS & NT):

B0_nRTE07 Number of bus lines that serve a bus stop2; as a basic measure of transit supply it is expected to be positively associated with a higher service level and boardings.

B0_frq07 Sum of one-way service vehicles-per-day of all bus lines servicing a bus stop 2; as a more nuanced measure of transit supply that takes into account distinct service levels of Local, Limited-Stop, Rapid Line, and Express bus line services it is also expected to be positively associated with higher boardings and higher service levels. bRapid_cnt Count of ‘Rapid’ bus lines servicing bus-stop; because of the faster and more frequent service provided by this service it is expected that a higher

102 number of rapid lines would result in higher bus-stop boardings associated with access trips to rapid-transit stations.

bXpres_cnt Count of ‘Express’ bus lines servicing bus stop 2; this specialized bus service focuses on providing access to downtown and other sub-centers, and operates the line-haul portion of the trip on freeways. It is expected to be negatively associated with bus-stop boardings related to access to rapid-transit stations as it may compete for a similar travel market.

bLimtd_cnt Count of ‘Limited-stop’ bus lines servicing bus stops 2 ; operates at slightly better speed as compared to regular local bus service and the count is expected to be positively associated with higher bus-top boardings.

lcl_RTE07 Count of ‘local’ bus routes servicing bus stop. A higher number of local lines servicing a bus-stop is expected to generate more boardings at the stop-level.

B0_RT1mtrs Euclidean distance (meters) between bus-stop and linked rapid-transit station. Distance (or time) represents an impedance factor that increases the cost of travel, and makes the trip less useful to potential travelers. It is expected to be negatively associated with the number of boardings at bus- stop.

Level-2 Rapid Transit Station Latent Construct Components (‘STATION’ [Destination-Accessibility]). Destination accessibility, which is understood in this study as the level of accessibility from a rapid-transit station to other stations and to other metropolitan destinations along the rapid-transit network is distinct from the level of accessibility to the station, which is a measure of the ease of access to a rapid-transit station via a variety of modes (i.e. walk, bicycle, bus, and automobile). The relative destination accessibility level of a rapid- transit station is considered a key significant factor that influences in part the number of access trips from linked bus-stops, in combination with bus-stop areal socio-economic characteristics, LU.BE attributes, network attributes, and bus service levels. Thus it is an important control variable to consider in the model.

In general accessibility can be defined as the relative ease of access to opportunities in a given built-environment given the possibilities and characteristics afforded by its transportation network, the associated land-use structure it serves, and individual/activites time (schedule)

103 constraints. In this study this unobserved phenomenon is represented via a construct (‘STATION’) that is implemented as a stand-alone latent variable that captures the variance associated with station-level destination accessibility in the response variable. This latent variable is also informed in an alternate model by observed (measured) variables hypothesized to capture key aspects associated with destination accessibility.

Some of these components refer to rapid-transit station network structure and functions, while others relate to rapid-transit service levels, demand levels, and LU.BE factors within stations’ pedestrian service areas. A description and justification for each follows.

‘STATION’ Destination Accessibility Components:

logRTAWB Log transformation of rapid-transit station annual average weekday boardings. Previous transit ridership studies at station-level have found significant associations and correlation between total boarding counts and the relative level of destination accessibility provided by the station, usually measured as a variant of a gravity-based model using travel time or distance as a non-linear impedance factor. A higher number of boardings at station-level is hypothesized as being reflective of a higher destination accessibility level.

Centrality Centrality is a basic topological attribute of the rapid-transit network and consist of the sum of links (one-directional lines) associated with each node (stations). A higher centrality value indicates higher connectivity and accessibility to a larger number of destinations and opportunities in the region. For example, stations that function as transfer hubs, or stations centrally located in a hub-and-spoke configuration would register a higher centrality value and thus higher destination accessibility level.

RTv_pDAY Rapid-transit vehicles per day: represents an aggregate transit-supply measure associated with higher station-level boardings, but is also posited in this study to capture a key component in the destination accessibility construct as it reflects higher frequency service and thus potentially lower wait/transfer time at station, higher average service speed, and/or the presence of multiple lines servicing a station. All these possibilities would reduce total travel time to destination and/or increase the number of reachable opportunities and thus lower total trip cost (increase destination accessibility).

104 Other Station-Level Local and Network Attributes:

Wlk_SCR Average Walk Score®, of rapid-transit station portals: some transit patrons that access rapid-transit services via bus need to walk from the alighting bus-stop to the first rapid-transit station in their one-way trip. As such they also experience the LU.BE characteristics around rapid-transit stations as part of their overall trip. An exception would be when a bus stop is physically adjacent or within a rapid-transit station immediate boundaries. It is hypothesized that the pedestrian friendliness around rapid-transit stations also influences travel behavior of the particular set of users that access via bus. More pedestrian-friendly rapid-transit stations are expected to attract more access trips from linked bus-stops as compared to stations that feature less pedestrian-friendly environments.

TrnfHUB A station’s function as a transfer hub is associated with a higher accessibility level to more destinations and higher service levels. Thus it is hypothesized that it would attract more riders via a variety of access modes, including bus. It is similar to the Centrality measure but operationalized as a dummy variable [0, 1]. Compared to the Centrality measure, which tends to be biased toward downtown nodes, the ‘TranfHUB’ dummy is a more neutral measure for stations in poly-centric agglomerations, particularly for suburban stations. It is expected to be positively associated with higher boardings from linked bus-stops and serves as an alternate measure to Centrality.

Model Results, Fit, and Discussion

The hypothesized relationships were evaluated using Stata 13 software. The two-level nested data was fitted within a multilevel generalized linear equation of the negative-binomial family (NBREG), log-link function, to account for the count response non-zero integer nature and highly skewed logarithmic distribution. Logarithmic transformation of some variables (Jobs_11, RT_AWB, RTv_pDAY) allowed for correction of residuals non-linearity issues and all models processed within the overarching Generalized Structural Equation Modeling framework (GSEM-MLM-NBREG) achieved convergence.

A set of five (5) consecutive models of increased complexity were fitted in order to evaluate the leading research question as well as other sub-hypotheses that emerge from the

105 discussion of LU.BE / TB literature review. To recapitulate, this study aims to evaluate if a significant relationship exists between LU.BE attributes around feeder bus-stops and the count of annual average weekday boardings that are associated with trips to rapid-transit stations. The models also explore whether a multi-dimensional index of pedestrian-friendliness (Walk Score®), or individual dimensions of the built-environment are a more appropriate instrument to capture the phenomenon of interest in this study, which relates more closely to ‘walking-to- transit’ research than to the more general LU.BE / TB literature.

The overall analysis strategy relies on comparison of key and control variables significance levels, directionality, magnitudes, as well as two model fit criteria: AIC (Akaike Information Criterion); BIC (Bayesian Information Criterion). Each of the five GSEM models are discussed in sequence and followed by a discussion of results, general and particular conclusions, and a final section that discusses research limitations, methodological improvements, and potential future research extensions.

The final model from which conclusions and final discussion will based on will be that which registers the best fit based on both AIC and BIC criteria (lowest values) as well as best estimated parameters match based on theoretical foundations and literature review of ridership factors. That is, the measured variables within the negative-binomial regression component of the overall GSEM-MLM-NBREG model should ideally follow expected directionality and significance (see Table 4.5), whilst the relative magnitude of the estimated parameters, particularly those associated with walkability, LU.BE characteristics around bus-stops, and bus network connectivity and service levels will receive special attention in discussion.

Predictive as well as conditional margins are also estimated for key variables of interest based on the best fit model. These are discussed for potential stand-alone or combined LU.BE / transit-service improvements policy considerations as related to the goal of increasing rapid- transit ridership and improving the effectiveness of multi-modal (walk | bus | rapid-transit) transit networks.

106 Table 4.5 GSEM Measures Descriptive Statistics

Expected Std. Variable Description Sign Source Year Obs Mean Dev. Min Max Outcome Variable: Bus stop average weekday PTV NuStats-LACMTA BS0_AWBi boardings assoc. w/ access trips to N/A 2011 On-board Survey 2011 5208 42.32 72.47 0.00 1432.00 rapid-transit stations (outcome) Database Observed Variables: Population levels within bus-stop US Census 2010 yPOP_10 ( + ) 2010 3350 3942.49 2737.77 0.00 23409.00 pedestrian service area Block-Level US Dept. Labor Total number of jobs within bus- logJob_11 ( + ) 'OnTheMap’ 2011 3301 5.85 1.43 0.00 11.22 stop pedestrian service area point feature data US Census Percent Hispanic population within ACS 2012 5YR yPerc_Hisp ( + ) 2012 3350 52.20 28.30 0.00 100.00 station service area block group-level (w/ ratio rule) Euclidean distance between first PTV NuStats-LACMTA yB0_RT1mtrs boarding bus-stop and first boarding ( - ) 2011 On-board Survey 2011 3350 7418.55 10531.19 250.53 181961.90 station (meters) Database Count of 'Express' bus lines within SCAG Bus Routes bXpres_cnt ( - ) 2007 5179 0.81 2.00 0.00 42.00 bus-stop feeder bus buffer shapefile Sum of one-way service bus SCAG Bus Routes B0_frq07 ( + ) 2007 5179 568.08 604.20 1.00 6628.00 vehicles per day shapefile Count of 'Rapid' bus lines within SCAG Bus Routes bRapid_cnt ( + ) 2007 5179 0.35 0.56 0.00 3.00 bus-stop feeder bus buffer shapefile Count of 'Limited' bus lines within SCAG Bus Routes bLimtd_cnt ( + ) 2007 5179 0.50 0.76 0.00 7.00 bus-stop feeder bus buffer shapefile Multidimensional index for Walk_Score ( + ) Walk Score® 2016 5208 74.67 15.51 0.00 99.00 walkability Multidimensional index for Wlk_SCR walkability ( + ) Walk Score® 2016 5157 78.35 19.15 25.00 99.00 (average of station portals) Avg__block Average block length (meters) ( - ) Walk Score® 2016 5208 158.57 40.55 72.00 420.00 Number of dining and drinking Dining_and destinations within bus-stop ( + ) Walk Score® 2016 5208 79.38 14.64 0.00 99.98 pedestrian service area County of Los Angeles 'Locations/Points of Nov- ySpeGen_LA Count of special activity generators ( + ) 3350 2.37 4.11 0.00 59.00 Interest (LMS Data)' ‘11 geodatabase US Census Average number of vehicles per ACS 2012 5YR yVeh_HU occupied housing unit within bus- ( - ) 2012 3350 1.59 0.42 0.00 2.70 block group-level (w/ stop service area ratio rule) US Census Total number of households with ACS 2012 5YR yTot_0veh zero vehicles within bus-stop ( + ) 2012 5208 334.42 583.52 0.00 5195.00 block group-level (w/ pedestrian service area ratio rule) log transformation of station LA Metro 'destination accessibility' construct logDest_ACC ( + ) AWB / Line Schedule / 2012 5157 7.36 0.98 3.51 9.24 based on 3-component factor Rail and Bus Shapefiles analysis LA Metro Basic topological measure of nodal Centrality ( + ) ' Rail Line' and 'Rail 2012 5157 2.26 1.12 1.00 5.00 connectivity station' Shapefiles LA Metro Dummy variable identifying station TrnfHUB ( + ) ' Rail Line' and 'Rail 2012 5157 0.27 0.44 0.00 1.00 function as a transfer hub station' Shapefiles log transformation of station average logRTAWB ( + ) La Metro 2012 5157 8.64 1.03 3.91 10.56 weekday boardings

107 MODEL 0. This model serves as baseline for evaluation of the results and fit of subsequent models and purposefully ignores the expected variance component of geographically clustered observations around rapid-transit stations (Figure 4.3). Although fitted within a GSEM framework it is essentially a generalized negative-binomial regression. A set of measures informed by the four vectors of transit determinants (as discussed in the literature review: 1- socioeconomics / demographic; 2-land use and built-environment; 3- transit service level; 4- network attributes) inform the model, where the key variable of interest is the ‘Walk_Score’ that serves as proxy for the LU.BE attributes around bus stops.

Despite the model’s misspecification and shortcoming most of the estimated parameters report expected directionality. The number of jobs (log), percent of Hispanic population, bus daily aggregate vehicle frequency, number of Rapid bus lines, and ‘Walk_Score’ all register a positive relationship with boardings. Also expected in their negative relationship with boardings are the distance between bus-stop and linked rapid-transit station (yB0_RT1mtrs), the number of Express bus lines, and the average number of vehicles per occupied housing units. A few unexpected parameter directions exist, such as the negative coefficient for population level and special generators. These could be result of multicollinearity with the Walk_Score index that takes into account population density and accessibility to a diversity of destinations that may include special generators as well.

Of the thirteen measures included in ‘Model.0’ three report statistically significant associations with the response measure. The number of jobs (log) and the count of Rapid service bus lines positively associate with higher boardings at (p=0.000) confidence level. The percent of Hispanic population within bus stops pedestrian service areas also registers a positive a significant association with boardings at (p=0.037) confidence level. The variable of interest (‘Walk_Score’) however, does not registers a significant confidence level (p=0.207). Also, the variable that represents Walk Score® ranking at the station-level (‘Wlk_SCR’), although positive in directionality does not reports a significant confidence level (p=0.160). Yet these results are non-conclusive given the previously discussed shortcoming in the models’ specification.

108

exp(b) Coef. Std. Err. z P>z [95% Conf. Interval] BS0_AWBi <- yPOP_10 1.000 -0.000006 0.000010 -0.58 0.564 -0.000026 0.000014 logJob_11 1.063 0.061178 0.015529 3.94 0.000 0.030741 0.091615 yPerc_Hisp 1.002 0.001701 0.000816 2.09 0.037 0.000103 0.003299 yB0_RT1mtrs 1.000 -0.000003 0.000002 -1.30 0.194 -0.000007 0.000001 B0_frq07 1.000 0.000042 0.000048 0.86 0.389 -0.000053 0.000136 bRapid_cnt 1.333 0.287442 0.045471 6.32 0.000 0.198320 0.376563 bXpres_cnt 0.981 -0.018885 0.011491 -1.64 0.100 -0.041406 0.003637 Walk_Score 1.002 0.002323 0.001839 1.26 0.207 -0.001282 0.005927 ySpeGen_LA 0.996 -0.004275 0.006885 -0.62 0.535 -0.017770 0.009219 bLimtd_cnt 0.992 -0.007540 0.036263 -0.21 0.835 -0.078614 0.063535 yVeh_HU 0.902 -0.102743 0.070304 -1.46 0.144 -0.240537 0.035051 yTot_0veh 1.000 0.000013 0.000054 0.24 0.808 -0.000093 0.000120 Wlk_SCR 1.002 0.001655 0.001178 1.40 0.160 -0.000654 0.003964 _cons 9.165 2.215390 0.241599 9.17 0.000 1.741865 2.688915 BS0_AWBi /lnalpha 0.0581082 0.0262623 2.21 0.027 0.006635 0.1095814

Model Obs ll(null) ll(model) df AIC BIC M.0 2645 . -10201.05 15 20432.10 20520.31

Figure 4.3 GSEM Model.0

109 MODEL 1. This model incorporates a latent variance component at Level-2 that aims to capture un-observed factors at station-level (‘STATION’) in a multilevel regression framework (see Figure 4.4). All other measured variables at Level-1 (bus-stop) remain the same as specified in ‘Model.0’. Changes in significance occur in the exogenous measures aimed at explaining the number of boardings at bus-stops that associate with access to rapid-transit stations, and these correspond with transit ridership literature and theory.

The estimated parameter for number of jobs (log) is still positive but no longer significant (p=0.115). The percent of population of Hispanic ethnicity (yPerc_Hisp) remains significant and positively related with boardings at a (p=0.037) confidence level as well as the count of Rapid bus lines at a high confidence level (p=0.000). The count of Rapid bus lines also registers the second largest standardized coefficient (1.349; see “exp(b)” column) after that of the station- level variance component (2.718), which indicates the high influence of un-observed factors nested at Level-2.

Three new stop-level measures arise as significant in this ‘Model.1’. Euclidean distance between bus-stop and linked rapid-transit station (yB0_RT1mtrs) exhibits the expected negative directionality and is significant at the (p=0.080) level. Also negatively related to the response variable, as expected, is the ratio of vehicles per occupied housing units (p=0.081) that also registers the third largest effect as compared to that of the variance component (M3[RT_STATION]) and count of Rapid service lines (bRapid_cnt) standardized coefficients (‘exp(b)’ column).

The key variable of interest, ‘Walk_Score’, emerges as significant in this ‘Model.1’ at a (p=0.060) confidence level and both measures of model fit (AIC and BIC) report lower values as compared to ‘Model.0’, thus a better fit registers between the model and empirical data variance matrix (Table 4.6).

110 exp(b) Coef. Std.Err. z P>z [95% Conf. Interval] BS0_AWBi <- yPOP_10 1.000 0.000005 0.000011 0.47 0.640 -0.000016 0.000025 logJob_11 1.025 0.024935 0.015830 1.58 0.115 -0.006092 0.055961 yPerc_Hisp 1.002 0.001866 0.000892 2.09 0.037 0.000117 0.003614 yB0_RT1mtrs 1.000 -0.000004 0.000002 -1.75 0.080 -0.000009 0.000000 bXpres_cnt 0.985 -0.014675 0.011245 -1.31 0.192 -0.036714 0.007364 B0_frq07 1.000 0.000067 0.000048 1.38 0.168 -0.000028 0.000161 bRapid_cnt 1.349 0.299329 0.048893 6.12 0.000 0.203501 0.395157 Walk_Score 1.003 0.003463 0.001844 1.88 0.060 -0.000150 0.007076 ySpeGen_LA 1.002 0.002094 0.006760 0.31 0.757 -0.011155 0.015343 bLimtd_cnt 1.007 0.006784 0.037884 0.18 0.858 -0.067467 0.081036 yVeh_HU 0.882 -0.125293 0.071809 -1.74 0.081 -0.266036 0.015450 yTot_0veh 1.000 -0.000014 0.000054 -0.25 0.800 -0.000119 0.000092 Wlk_SCR 0.999 -0.000979 0.003043 -0.32 0.748 -0.006943 0.004985 M3[RT_STATION] 2.718 1 (constrained) _cons 10.615 2.362289 0.324444 7.28 0.000 1.726389 2.998188 BS0_AWBi /lnalpha -0.06735 0.027402 -2.46 0.014 -0.1210564 -0.01364 var(M3[RT_STATION]) 0.222545 0.048211 0.1455522 0.340264

Model Obs ll(null) ll(model) df AIC BIC

M1 2645 . -10088.9 16 20209.87 20303.96

Figure 4.4 GSEM Model.1

111 MODEL 2. The literature review of ‘walking-to-transit’ indicates that what transit patrons value the most in regards to LU.BE factors is shorter distances (reduced walking time) between trip origin and transit stop or station. In order to test this proposition a sub-component of the Walk Score® replaces the multidimensional walkability index in the model specification. ‘Avg__block’ measures average block length (meters) within a specific location pedestrian service area and serves as a proxy for average walking distance to transit bus-stop (see Figure 4.5). That is, a more granular street network offers more and potentially shorter paths between trip origins and destinations. All other explanatory and latent variables remain as specified in the previous ‘Model.1’.

Substituting ‘Avg__block’ for ‘Walk_Score’ yields a higher significance level (p=0.009) than that for ‘Walk_Score’ (p=0.060), and the effect of this built-environment feature is negative in direction relative to the response variable, as expected. The introduction of this sub- component of the built-environment also results in the re-emergence of the number of jobs (log) as a relevant measure with a high significance level (p=0.017), and could be attributed to a plausible correlation with the removed ‘Walk_Score’ measure from the model. All other measures at the bus-stop level remain with identical directionality relative to the response variable and only one explanatory variable losses statistical significance (yB0_RT1mtrs; p=0.100) yet maintains its negative relationship with the response variable. The largest standardized coefficient (‘exp(b)’ column) still registers for the Level-2 ‘STATION’ variance component (ME[RT_STATION]), followed by the count of ‘Rapid’ bus lines in second place, and ratio of vehicles per occupied housing units in third place (yVeh_HU). This model also improves AIC and BIC model fit indices relative to the previous models (Table 4.6).

MODEL 3. The review of the more general LU.BE / TB literature reports two prominent factors associated with higher transit travel: (1) land-use mix; and (2) intersection density (i.e. number of 3- and 4-way intersections). Having incorporated the average block length in the previous model (which is highly correlated with intersection density) for this ‘Model.3’ several land-use sub-components of the Walk Score® ranking were tested for significance, including a new aggregate variable of all land-use sub-components and each sub-component by itself. All other measured and latent variables remain the same as in ‘Model.2’ (see Figure 4.6).

112 exp(b) Coef. Std. Err. z P>z [95% Conf. Interval] BS0_AWBi <- yPOP_10 1.00001 0.000005 0.000010 0.49 0.627 -0.000015 0.000026 logJob_11 1.03641 0.035763 0.015033 2.38 0.017 0.006299 0.065228 yPerc_Hisp 1.00188 0.001882 0.000891 2.11 0.035 0.000135 0.003628 yB0_RT1mtrs 1.00000 -0.000004 0.000002 -1.64 0.100 -0.000008 0.000001 bXpres_cnt 0.98483 -0.015288 0.011199 -1.37 0.172 -0.037237 0.006662 B0_frq07 1.00006 0.000064 0.000048 1.33 0.185 -0.000031 0.000159 bRapid_cnt 1.36255 0.309359 0.048563 6.37 0.000 0.214176 0.404541 Avg__block 0.99826 -0.001742 0.000666 -2.62 0.009 -0.003046 -0.000437 ySpeGen_LA 1.00176 0.001761 0.006772 0.26 0.795 -0.011512 0.015035 bLimtd_cnt 1.01177 0.011702 0.037742 0.31 0.757 -0.062272 0.085675 yVeh_HU 0.87830 -0.129766 0.069964 -1.85 0.064 -0.266893 0.007360 yTot_0veh 0.99999 -0.000005 0.000054 -0.09 0.925 -0.000111 0.000101 Wlk_SCR 0.99894 -0.001065 0.003016 -0.35 0.724 -0.006976 0.004846 M3[RT_STATION] 2.718282 1 (constrained) _cons 16.98637 2.832411 0.31715 8.93 0 2.21081 3.454013 BS0_AWBi /lnalpha n/a -0.06793 0.027405 -2.48 0.013 -0.12165 -0.01422 var(M3[RT_STATION]) n/a 0.217697 0.047352 n/a n/a 0.142137 0.333425

Model Obs ll(null) ll(model) df AIC BIC M2 2645 . -10087.34 16 20206.67 20300.76

Figure 4.5 GSEM Model.2 113 Only one land-use sub-component registered a statistically significant association with the response variable (‘Dining_and’: number of dining and drinking destinations within pedestrian service area; p=0.029). The incorporation of this measure in ‘Model.3’ specification results in model fit improvement (lower AIC and BIC values) however the number of jobs variable (logJob_11) loses significance. This could be result of strong autocorrelation between the number of jobs variable and the ‘Dining_and’ variable. The percent of Hispanic residents, the count of Rapid service lines, and the average block length measures remain statistically significant at relatively high levels of confidence (p=0.032; p=0.000; p=0.031; respectively).

MODEL 4. This model explicitly incorporates exogenous measures at Level-2 (station- level) that, as hypothesized by this author, capture aspects of relative destination accessibility that would increase the attractiveness of stations to potential patrons (see Figure 4.7). The three measures are: 1- transfer hub (‘TrnHUB’; nodal network function that is also correlated with higher service frequency and boardings); 2- Centrality (topological measure of network connectivity); 3- total average weekday boardings (‘(log) RTAWB’; correlates with higher destination accessibility in previous transit studies). All other variables remain the same as in the previous ‘Model.3’.

Introducing these explicit measures at Level-2 improves model fit (lower AIC value) and significance is maintained in all previous stop-level measures. Two of the specified measures at Level-2, ‘Centrality’ and ‘logRTAWB’ report significant associations (p=0.051; p=0.003, respectively), and the third measure ‘TrnfHUB’ is not significant yet positively associated and reports the largest coefficient of the three.

An unexpected result is the negative directionality of the ‘Centrality’ measure. A plausible explanation could be autocorrelation with the ‘TrnfHUB’ measure, which measures similar aspects of network topology, or that the centrality measure is biased towards older central nodes (i.e. downtown LA) that tend to accommodate more connected ‘legacy’ nodes (i.e. Union Station) whilst most of the trips under study may be occurring in more peripheral areas and associated with cross-town travel rather than downtown-oriented trips.

114 exp(b) Coef. Std. Err. z P>z [95% Conf. Interval] BS0_AWBi <- yPOP_10 1.000003 0.000003 0.000010 0.30 0.766 -0.000017 0.000023 logJob_11 1.022584 0.022333 0.016250 1.37 0.169 -0.009517 0.054183 yPerc_Hisp 1.001905 0.001903 0.000890 2.14 0.032 0.000160 0.003647 yB0_RT1mtrs 0.999996 -0.000004 0.000002 -1.58 0.114 -0.000008 0.000001 bXpres_cnt 0.986428 -0.013665 0.011220 -1.22 0.223 -0.035655 0.008326 B0_frq07 1.000061 0.000061 0.000048 1.25 0.210 -0.000034 0.000155 bRapid_cnt 1.354053 0.303103 0.047430 6.39 0.000 0.210143 0.396063 Avg__block 0.998541 -0.001460 0.000676 -2.16 0.031 -0.002786 -0.000135 bLimtd_cnt 1.005311 0.005297 0.037821 0.14 0.889 -0.068831 0.079425 yVeh_HU 0.924338 -0.078677 0.073086 -1.08 0.282 -0.221923 0.064568 yTot_0veh 1.000003 0.000003 0.000050 0.05 0.959 -0.000095 0.000100 Wlk_SCR 0.998904 -0.001097 0.003019 -0.36 0.716 -0.007015 0.004821 Dining_and 1.004020 0.004012 0.001842 2.18 0.029 0.000403 0.007621 M3[RT_STATION] n/a 1 (constrained) _cons 2.718282 2.483386 0.3545537 7 0 1.788473 3.178298 BS0_AWBi /lnalpha 11.9818 -0.0696549 0.0274134 -2.54 0.011 -0.123384 -0.015926 var(M3[RT_STATION]) n/a 0.2187709 0.0474306 0.143036 0.334606

Model Obs ll(null) ll(model) df AIC BIC M3 2645 . -10085.02 16 20202.05 20296.13

Figure 4.6 GSEM Model.3

115 exp(b) Coef. Std. Err. z P>z [95% Conf. Interval] BS0_AWBi <- yPOP_10 1.000003 0.000003 0.000010 0.30 0.766 -0.000017 0.000023 logJob_11 1.021050 0.020832 0.016263 1.28 0.200 -0.011043 0.052706 yPerc_Hisp 1.002001 0.001999 0.000887 2.25 0.024 0.000261 0.003738 yB0_RT1mtrs 0.999996 -0.000004 0.000002 -1.62 0.106 -0.000008 0.000001 bXpres_cnt 0.987064 -0.013020 0.011212 -1.16 0.246 -0.034996 0.008955 B0_frq07 1.000056 0.000056 0.000048 1.15 0.250 -0.000039 0.000150 bRapid_cnt 1.356093 0.304608 0.047365 6.43 0.000 0.211775 0.397441 Avg__block 0.998522 -0.001479 0.000675 -2.19 0.029 -0.002802 -0.000155 bLimtd_cnt 1.010009 0.009959 0.037817 0.26 0.792 -0.064161 0.084079 yVeh_HU 0.922283 -0.080903 0.073072 -1.11 0.268 -0.224120 0.062315 yTot_0veh 1.000004 0.000004 0.000050 0.08 0.932 -0.000093 0.000101 Wlk_SCR 0.997889 -0.002113 0.002889 -0.73 0.465 -0.007776 0.003550 Dining_and 1.003951 0.003943 0.001840 2.14 0.032 0.000338 0.007548 M3[RT_STATION] 1 (constrained) _cons 2.718282 1.669006 0.5420871 3.08 0.002 0.606535 2.731477 M3[RT_STATION] <- Centrality 5.306891 -0.1849697 0.0946699 -1.95 0.051 -0.3705194 0.0005799 logRTAWB 0.1583801 0.0536913 2.95 0.003 0.053147 0.2636132 TrnfHUB 0.2949647 0.2185367 1.35 0.177 -0.1333592 0.7232887 BS0_AWBi /lnalpha -0.0713421 0.0273793 -2.61 0.009 -0.1250045 -0.0176797 var(e.M3[RT_STATION]) 0.1917471 0.0408963 0.1262358 0.2912561

Model Obs ll(null) ll(model) df AIC BIC

M4 2645 . -10078.62 19 20195.24 20306.97

Figure 4.7 GSEM Model.4 116 The overall impact of these measures on the station-level variance component (e.M3[RT_STATION]) is evident but small (0.192 [Model.4] vs. 0.219 [Model.3]). Thus, an important Level-2 unobserved dimension remains unidentified as captured in the error term. All previously identified Level-1 bus-stop measures that registered high significance confidence levels in the previous ‘Model.3’ maintain their status (percent Hispanic residents, count of Rapid bus lines, average block length, and the number of dining and drinking destinations). The final model, discussed next, aims to define a simpler parsimonious set of variables, improve on Level- 2 exogenous variables explanatory power, and improve overall model fit.

MODEL 5 (Final). This final model aims for parsimony by eliminating non-significant measures that register high correlation with others, improvement of explanatory power of Level- 2 components, and improvement of overall model fit for final evaluation of results and conclusion statements (see Figure 4.8).

Attempts to define a new Level-2 exogenous latent variable for destination accessibility as an integral part of the GSEM model structure, using the three aforementioned measurements (as well as others), did not reach convergence. Instead, a factor analysis component was calculated outside of the GSEM model in Stata software using a set of three highly correlated measures hypothesized to capture aspects of the destination accessibility construct at station level: ‘RTv_pDAY’ (supply-side aggregate number of vehicles per day), ‘RT_AWB’ (demand- side average weekday boardings per day), and ‘TrnfHUB’ (network attribute; station function as transfer hub (dummy)). A new variable (‘Dest_ACC1’) was defined and calculated using single- factor scores from the three aforementioned measures and incorporated in the GSEM model at Level-2 (Station-level) as an exogenous measured variable. This approach is considered more conceptually consistent and a more valid measure of the phenomenon of interest as it only captures the hypothesized shared influence of the three measures and not the full information contained in each measure (as implemented in ‘Model.4’).

At Level-1 (bus-stop) the following measures were removed to minimize redundancy and autocorrelation: ‘Dining_and’ which is highly correlated with number of jobs (‘logJob_11’); ‘yTot_0veh’ (number of zero-car households) which is highly correlated with ‘yVeh_HU’

117 (average number of vehicles per occupied housing unit); and ‘bLimtd_cnt’ was removed because it was consistently non-significant in all previous models and does not contribute significantly to overall model fit. These model adjustments result in better model fit as reflected in lower AIC and BIC values as compared to all previous models (see Table 4.6, Appendix D).

After these model adjustments all except one explanatory variable report high significance levels and their directionality follows theoretical expectation and previous literature findings. The hypothesized destination accessibility construct coefficient at Level-2 (‘logDest_ACC’; 0.164) explains almost as much variance as the latent variable coefficient for the error term (‘e.M3[RT_STATION]’; 0.199).

The only Level-1 measure that does not registers statistical significance is the count of ‘Express’ service lines (‘bXpres_cnt’). For illustrative purposes the author maintains its result in the final model as it is suggestive and corresponds in its negative directionality with the hypothesized association with the response variable. That is, it appears that this specialized express commuter bus service competes with rapid-transit services for specific downtown and other sub-center commute trip market (Figure 4.9). In contrast, ‘Rapid’ bus line services (bRapid_cnt) whose mostly perpendicular alignments with respect to rapid-transit lines suggest a feeder-oriented functionality register a highly significant (p=0.000) and positive association with bus-stop boardings associated with access trips to rapid-transit stations (Figure 4.10).

The following bus-stop level measures registered positive and significant associations with the response variable (average weekday boardings associated with access trips to rapid- transit stations): number of jobs (‘logJob_11’), percent Hispanic residents (‘yPerc_Hisp’), aggregate bus vehicle frequency per day (‘B0_frq07’), and count of Rapid bus lines (‘bRapid_cnt’). The following measures register significant negative associations with the outcome variable: Euclidean distance between bus-stop and linked rapid-transit station (‘yB0_RT1mtrs’), average block length (‘Avg__block’) and the ratio of vehicles per occupied housing units (‘yVeh_HU’). All measures follow expected directionality and significance as per theory and previous literature findings.

118 The only unexpected result in this investigation is the non-significance of population levels around bus-stops (‘yPOP_10’ and its log-transformation). It is possible that the ratio of Hispanic population living near bus stops serves as proxy for overall population levels given medium correlation levels between the two measures, or that socioeconomic factors override population levels for this particular multimodal trip sub-market as is suggested in Figure 2.9 (i.e. high rate of driving license non-tenure and Hispanic ethnicity).

The key variable of interest, ‘Walk_Score’, when incorporated in the model registers positive and statistical significant association with the response variable (p=0.049). However the more basic average block length measure (‘Avg_block’) reports a higher significance level (p=0.008) and the overall model reports better fit values. The relative effect magnitude of the standardized coefficient for the LU.BE variables is relatively low as compared to the effect magnitude of other significant measures, and this is also consistent with previous LU.BE / TB literature (‘Walk_Score’ exp(b)=1.003 ; ‘Avg_block’ exp(b)=0.998). That is, a one unit score increase in Walk Score® results in a 0.3% increase in bus-stop boardings, and a one unit increase in block length results in a 0.1% decrease in bus-stop boardings, all else equal. Because residential self-selection is not accounted for in this study it is possible that effects are weaker.

For comparison purposes the standardized coefficients of the two top bus-stop variables are exp(b) = 1.35 for ‘bRapid_cnt’ (count of ‘Rapid’ bus lines), and exp(b) = 0.878 for ‘yVeh_HU’ (average number of vehicles per occupied housing units). That is, a one unit increase in the count of ‘Rapid’ bus lines results in a 35% increase in bus-stop boardings associated with access to rapid-transit stations, all else equal; and a one unit increase in the ratio of vehicles per occupied housing unit results in a 12% decrease in boardings at bus-stop level, all else equal. In regards to the variable of interest (Avg_block), a one unit (1meter) reduction in the average block length within a bus-stop’s pedestrian service area would result in a 0.1% increase in boardings associated with access trips to rapid-transit stations.

In more practical terms, based on predictive margins from Model.5 that only account for direct effects (as per Stata v.13 ‘margins’ command, i.e. Level-2 latent variable values=0), allowing a pedestrian path to bisect 200mt long suburban or urban blocks within a bus-stop

119 pedestrian service areas, assuming the new path is laid out in direction of a bus-stop, could increase boardings associated with access trips to rapid-transit stations by 1.60 per bus-stop, holding all other covariates at mean. Depending on the number of inbound bus-stops along the bus route the aggregate sum of boardings at the rapid-transit station could be significant. Based on predictive margins, if this built-environment policy is combined with provision of a new ‘Rapid’ bus route the combined effect could yield 2 additional boardings per bus-stop, for a total of 3.6 additional boardings per bus-stop (1.60 + 2.0).

Conclusions and Policy Implications

The guiding research question in this investigation asks whether land-use and built- environment attributes (LU.BE) around feeder bus-stops influence the number of boardings associated with access to rapid-transit stations. Given the statistically significant results in the previous models it can be concluded that there is a significant association, yet the relative impact is smaller as compared to the positive and large influence associated with provision of faster and higher frequency bus services (Rapid lines) and as compared to the notable negative effect of levels of automobile availability per occupied housing unit. However, combined initiatives related to bus service improvements (i.e. reducing headways, adding new ‘Rapid’ lines, and/or increasing bus average speeds), together with modifications to pedestrian networks geared to reduce walking distance to bus stops and/or policies aimed at reducing automobile use and/or availability could synergistically encourage more transit use and access to rapid-transit services via bus.

The contributions of this study to LU.BE / TB and rapid-transit ridership in multimodal contexts expand the geographical extent and discussion from areas surrounding rapid-transit stations, which have received much research and policy attention in the past decades (i.e. TOD programs) to more peripheral yet still important areas around feeder bus-stops. In addition, results from this study evince and confirm the interdependent relationship between bus networks and rapid-transit services in Los Angeles metropolitan transit system, and plausibly in other large, disperse poly-centric agglomerations in the United States and beyond.

120 It can also be concluded from the results of this investigation that both a multi- dimensional index of walkability such as Walk Score® or a more specific built-environment areal measure such as average block length capture significant LU.BE effects on the behavior of interest, yet for specific travel behavior research that focus on ‘walking-to-transit’ it appears that distance-based measures better capture effects based on significance level results as compared to the multidimensional walkability index. It can also be concluded from this study that significant observed and un-observed factors nested at rapid-transit station level influence boardings at feeder bus-stop locations. Better instruments are needed to capture the unexplained portion of this influence. The net costs or benefits associated with LU.BE approaches around bus-stops to increase rapid-transit ridership are not contemplated in this study, nor are the costs and benefits associated with a transit service improvement or automobile demand-management approach. It is plausible that a combined LU.BE and transit service level approach along feeder bus corridors will contribute in a synergistic manner to increase rapid-transit boardings. Future research could focus on the feasibility of interventions related to LU.BE and/or transit service improvements in the pursuit of higher transit ridership, more sustainable urban transit systems, and lower automobile dependence within a Benefit-Cost analytical framework.

Table 4.6 GSEM Models AIC and BIC Criterion

Model Obs ll(null) ll(model) df AIC BIC

M.0 2645 . -10201.05 15 20432.10 20520.31 M.1 2645 . -10088.94 16 20209.87 20303.96 M.2 2645 . -10087.34 16 20206.67 20300.76 M.3 2645 . -10085.02 16 20202.04 20296.13 M.4 2645 . -10078.62 19 20195.24 20306.97 M.5 2645 . -10083.24 12 20190.49 20261.05

121 e(b) Coef. Std. Err. z P>z [95% Conf. Interval] BS0_AWBi <- logJob_11 1.037082 0.036411 0.014900 2.44 0.015 0.007208 0.065615 yPerc_Hisp 1.002036 0.002034 0.000868 2.34 0.019 0.000333 0.003736 yB0_RT1mtrs 0.999996 -0.000004 0.000002 -1.92 0.055 -0.000009 0.000000 bXpres_cnt 0.984646 -0.015473 0.011094 -1.39 0.163 -0.037217 0.006271 B0_frq07 1.000070 0.000070 0.000040 1.77 0.077 -0.000008 0.000147 bRapid_cnt 1.357070 0.305328 0.046541 6.56 0.000 0.214110 0.396546 Avg__block 0.998250 -0.001751 0.000661 -2.65 0.008 -0.003046 -0.000456 yVeh_HU 0.878372 -0.129686 0.061404 -2.11 0.035 -0.250036 -0.009335 M3[RT_STATION] 1 (constrained) _cons 2.718282 1.679925 0.408651 4.11 0.000 0.878984 2.480866 M3[RT_STATION] <- logDest_ACC 5.365153 0.164417 0.054792 3 0.003 0.057026 0.271808 BS0_AWBi /lnalpha -0.069129 0.027377 -2.53 0.012 -0.122788 -0.015470 var(e.M3[RT_STATION]) 0.199876 0.042622 0.131597 0.303580

Model Obs ll(null) ll(model) df AIC BIC M.5 2645 . -10083.24 12 20190.49 20261.05

Figure 4.8 GSEM Model.5

122

Figure 4.9 Los Angeles Rapid-Transit Network and Non-LA Metro Express Bus Routes Source: Metro ‘Bus & Rail GIS Data (2011)’1 (http://developer.metro.net:80/introduction/gis- data/download-gis-data/; retrieved December2015 from ‘Internet Archive – Wayback Machine’; https://web.archive.org/) and SCAG ‘Bus Routes (2007)’ shapefile (retrieved December 2015; http://gisdata.scag.ca.gov/Pages/GIS-Library.aspx) 123

Figure 4.10 Los Angeles Rapid-Transit Network and LA Metro ‘Rapid’ Bus Routes Source: Metro ‘Bus & Rail GIS Data (2011)’1 (http://developer.metro.net:80/introduction/gis- data/download-gis-data/; retrieved December2015 from ‘Internet Archive – Wayback Machine’; https://web.archive.org/) and SCAG ‘Bus Routes (2007)’ shapefile (retrieved December 2015; http://gisdata.scag.ca.gov/Pages/GIS-Library.aspx) 124 Research Limitations and Future Extensions

As a single case study based of a uniquely large car-dependent poly-centric metropolis the results from this study may not be generalizable to other urban agglomerations in the U.S. or beyond, nor establish causality based on cross-sectional data. If decentralization, growth, and sprawl trends, however, continue in cities within the United States and beyond it, is plausible to consider that bus access shares will also increase, all else equal. Thus lessons and insights from this study might be useful in the future.

Considerable effort was invested in building a multi-agency bus service database for the same year that the on-board bus and rail survey was conducted (2011), from which bus-stop boardings estimates were derived. However, the completion of this task was not feasible at the time of this study and a comprehensive second-best database from year 2007 (SCAG) was used instead. Hence, results must be evaluated with caution as service levels most likely have changed in the 4-year period between the base years of the bus network database to the year the on-board survey was conducted.

More precise and valid bus-stop areal ‘distance to transit’ measures can be obtained based on combined land-use GIS and transportation network planning software applications, such as a directionality-based index in which network-based shortest-path protocol between parcels centroids and bus-stops is implemented and compared to Euclidean-based distances as a ratio.

The possibility of residential self-selection in assessing the influence of LU.BE factors is not addressed is study. In the absence of surveys and personal qualitative data some scholars have argued and used a rich set of socio-economic variables within a set of simultaneous equations in a GSEM framework (Mitra et al. 2017). In the future the existing set of socio- economic variables could be expanded and a similar protocol implemented within a GSEM framework.

125 CHAPTER 5

CONCLUSION

This investigation had two main objectives and one overarching goal. This first objective was to quantify and clarify the importance of bus network services to rapid-transit patronage in the special case of a large, dispersed, poly-centric urban/suburban agglomeration. This is explained and discussed in Chapter 2 and Chapter 3. The second objective was to asses if land- use and built-environment attributes in areas surrounding feeder bus-stops influence the number of boardings associated with access trips to rapid-transit stations. This is explained and discussed in Chapter 4. Each of the two investigations was conducted at a different network scale in tune with the specific phenomena of interest and research questions: 1) station-level; and 2) bus-stop level.

The overarching goal was to gain a better understanding of multimodal transit travel and multi-scalar interactions associated with bus access to rapid-transit services. This author asserts that this particular travel behavior and access mode has been under-studied in the literature related to TOD, rapid-transit services, and land-use / travel-behavior research. The findings of this investigation shed light on these issues and contribute to a better understanding of bus and rapid-transit networks interaction in the context of Los Angeles metropolitan region and provides answers to the question of land-use and built-environment potential influence on areas surrounding feeder bus-stops.

Quantitative results from this study evinces that bus and rapid-transit networks in Los Angeles metropolitan transit system are an interdependent system based on relatively high rate of estimated bus access events (>30%) and on the high statistical significance and relative coefficient magnitude of bus connectivity measures at station-level as related to boardings (Table 3.3). The bus-stop level study also evinces a high statistical significance of land-use and built- environment attributes (LU.BE) around feeder bus-stops on boardings associated with access to rapid-transit stations, yet the effect magnitude is relatively small when compared to the large positive effect of bus service levels (particularly ‘Rapid’ service) and notable negative effects of

126 automobile availability. These results are consistent with travel theory and literature and confirm the hypotheses posited in each study.

Whilst methodological shortcomings are recognized, specifically validity issues related to multidimensional walkability measures in ‘walking-to-transit’ research; pedestrian service area delimitation protocols; station-level destination accessibility instrumentation; and omission of controls for potential self-selection influence on LU.BE factors assessment, the evidence and results gathered in this investigation points to an undeniable importance of bus transit as access mode to rapid-transit services in the case of large, dispersed, poly-centric urban/suburban agglomerations. More attention and research should be paid to multimodal service improvements, their coordination in pursuit of better service, and potential synergistic relationship with land-use and built-environment characteristics at both stations and feeder bus stops. Equally important is understanding and correcting for why such an important travel behavior is often un-documented in travel surveys and official technical reports.

Future research extensions and improvements would include addressing the aforementioned methodological shortcomings and the development and improvement of DDM models for forecasting bus transfers (access) at rapid-transit station. This would aid multimodal transit planning and operations, as well as inform rapid-transit station spatial programming and design. Findings and insights from this study will contribute to these ends.

The advent of telecommuting, shared mobility companies (i.e. TNCs, cars, bicycles, electric scooter, Segway) and automated vehicles presents novel technological contexts and possible disruptive competition that might challenge the current importance of bus mode as access to (and egress from) rapid-transit stations in the near future. Detailed and periodic access mode share statistics at all network levels should allow for more informed understanding on these trends, and facilitate a better understanding of its implications and future policy formulations.

To reiterate, approximately more than one-third of rapid-transit patrons in the Los Angeles region access via a variety of bus services, and bus access to rapid-transit station is the

127 dominant access mode in stations located within freeway right-of-ways (i.e. BRT Silver Line and LRT Green Line). Furthermore this particular travel market is represented in large part by working, lower-income, transit-dependent users and mostly members of the Hispanic community of Los Angeles. From both an equity and system-performance perspectives it is important to pay more attention to these users, their travel behavior, and to multimodal transit service coordination.

Documentation, surveys, and statistics associated with access modes and modal access share should include bus mode as an independent category at a variety of scales, including system-wide, line-level, and station-level for planning, operations, and research purposes. This will allow for a better understanding of multimodal bus/rapid-transit travel, its relative importance and longitudinal trends as compared to other access modes, and the geographical extent of multimodal transit trips in large, disperse, poly-centric urban/suburban agglomerations in the United States.

Based on this study’s results SCAG and LA Metro current policy emphasis on TOD programs as a strategy to maintain and/or increase transit patronage is limited. A more comprehensive policy approach based on ‘integrated public transportation’ and a more extensive station access policy that incorporates improvements in bus service levels and LU.BE attributes around feeder bus stops, not only around stations, is the recommended course.

128 APPENDIX A

DETAILED SURVEY DESCRIPTION AND DATABASE PREPARATION FOR ACCESS MODE POPULATION PROPORTION ESTIMATION

Detailed Survey Description and Database Preparation for Bus, Walk, Car, Bicycle, and ‘Other’ Access Population Proportion Estimation. The original LA Metro comprehensive on-board survey design and strategy relied on a standard two-stage random sampling with no-replacement of bus and rail users’ one-way weekday person-trips. This survey was conducted from January 11 through June 24, 2011 (bus data) and from September 26 through November 1, 2011 (rail data). The survey consultant (PTV NuStats) implemented and tested a hybrid paper/computer-assisted telephone interview method. The sampling strategy clustered around 166 Metro routes and sample stratification occurred along three key vectors: 1- time of day; 2- directionality; and 3- route. Weights and expansion factors account for sampling design (clustering and stratification) and a standard two-stage sampling protocol based on rider and boarded vehicle characteristics. Weights for line-level, system-wide, and stop (station)-level estimates and analyses are provided in the digital dataset. As noted in Metro’s survey report, a total of 33,782 usable bus and rail questionnaires were collected for this study; the final bus data files incorporated a total of 113,380 eligible boardings for a response rate of 24 percent; rail data set was comprised of 28,875 eligible boardings for a response rate of 23 percent. Total valid responses represent 2% of system-wide daily boardings based on year 2009 figures, and higher percentages for individual station-level analyses (2% - 24%). For the sub-population of interest in this study (rapid- transit person-trips using HRT, LRT, and/or BRT services) a total of 10,535 responses were identified in the survey.

LA Metros’ digital survey database consists of two associated files, the ‘SUMMARY’ file and the ‘WAYPOINT’ file. The SUMMARY file gathers general person-trip data, socioeconomics, vehicle/route information, weights based on route, time of day, and directionality; and on system-wide, line-, and stop (station)-level, among other trip and person characteristics. The WAYPOINT file contains detailed person-trip one-way stop sequence, stop names, route identification and information, and spatial x-y coordinates for each travel point from origin to destination. A common unique identifier for each person-trip observation (‘SAMPN’) exists in both files and is used for data merging protocols between the two files. These two files were originally facilitated to this author by the agency together with a code dictionary and a copy of the official survey report (pers. comm. John Stesney, Transportation Planning Manager III, Systems Analysis & Research, Los Angeles County Metropolitan Transportation Authority; via email July 27, 2015). In addition, this author contacted the private consultant in charge of the original survey design and implementation (PTV NuStats, Houston TX) for clarification and consultancy related to database structure, weighting and expansion protocol, and feasibility of constructing station-level access mode estimates based on existing line-level and stop-level survey data. Complementary information and clarifications were exchanged via email communication with PTV NuStats project manager and senior statistician and a detailed weighting-expansion protocol memo was shared with this author for implementation (pers. comm., Sujin Hong, Senior Data Analyst, and Ryan McCutchan, Project Manager at PTV NuStats, Austin, TX via email; November 02, 2015).

Key survey issues addressed by this author relate to calculation of station-level expansion factors using annual FY 2011-2012 station annual average weekday boardings as per APC records. The construction of appropriate confidence intervals for the calculated sample proportion was managed with the svyset command in Stata software that accounts for the survey’s complex design and subpop calibration option for appropriate calculation of confidence intervals. The process for estimating bus access events per station began by the author first familiarizing with the files content and structure, and screening for any data and typo errors, extreme outliers, and other nonsensical results following evaluation of standard data description and summary queries in STATA software. A substantial number of errors related to rapid-transit and bus stop name misspellings (>450) were identified and corrected. Several dummy variables were created in the SUMMARY and WAYPOINT files for identifying rapid-transit trips and routes by technology and route identifiers (HRT, LRT, BRT vs. bus routes); survey responses associated with rapid-transit use; their respective access mode (bus, walking, car, bicycle, and other); and spatial x-y coordinates for origin, first bus stop, last bus stop before boarding a rapid-transit station, and first rapid-transit station boarded, together with their respective names using conditional logic scripting in STATA software. These spatial x-y coordinates and complementary information were later transferred and merged with associated person-trip observations from the WAYPOINT file to the SUMMARY files using unique identifiers. Key person-trip spatial points facilitated Euclidean

129 distance calculations between key points of interest for statistical analyses and delimitation of pertinent station buffers for implementation in GIS (see Appendix C).

Weighting and Expansion Protocol for Survey’s One-Way Weekday Person-Trips (data source file:SUMMARY.sav):

Steps for Weighting (‘Weight Factor’):

I. [ Vehicle Factor (corrects for trip sampling rates along three strata: 1-route; 2-time of day; 3- direction) ] X [ Response Factor (corrects for non-response at the bus stop (and rail station) level for both boardings and alighting) ]

Steps for Sample Expansion (adjusts the ‘weighted sample’ to the total trips at the system-wide level):

II. [ Expansion Factor (adjust expected ridership based on observations based on average daily ridership at route level or station level ]: Population Average Daily Ridership / Weighted Ridership III. [ Final Expansion Weight = Weight Factor X Expansion Factor ] , at this stage the weighted data represents the population boarding to alighting (One-way Person Trip)

Steps for Estimating Station-Level Boardings and Mode of Access Proportions (Bus, Walk, Car. Bicycle, Other):

IV. Identification and indexing of rapid-transit lines by route name and transit technology (BRT, LRT, and HRT) using logical conditional parameters in Stata software. V. Identification and indexing of person-trips that used rapid-transit services (HRT, LRT, BRT) during their one-way trip using logical conditional parameters in Stata software. VI. Identification and indexing of access mode (Walk, Bus, Car, Bicycle, Other) for each rapid-transit person- trip using trip stop-sequence and spatial data contained in “WAYPOINT” file; logical conditional parameters in Stata software. VII. Aggregation of rapid-transit person-trips (using linked ‘Final Expansion Weight’) by station to estimate total annual average weekday boardings. VIII. Aggregation of ‘weighted-expanded’ person-trips by mode of access for each station. IX. Calculation of access mode proportions by station (sum of weighted person-trip by mode/total person-trips per station). X. Person-trip expansion adjustment to station-level year 2012 average weekday boardings (the original survey expansion factors were based on year 2010 data). Access mode proportions are maintained and assumed to be the same for the year 2012 estimates.

Note: All logical scripting, calculations, and estimations were performed in Stata v.13 software. The ‘svyset’ command for handling complex survey design data and subgroup estimations and calculations of appropriate confidence intervals, such as the one in this study, was implemented as per LACMTA 2012 survey description for a 2-stage, random sampling with no-replacement, clustered (n=133 routes) and stratified survey (3-strata integrated in the Vehicle Factor).

130 APPENDIX B

FIGURE 1.1 AND FIGURE 1.2 SOURCES

CITY / AGENCY REGION ABBRV. AGENCY NAME SOURCE

United States APTA American Public American Public Transportation Association. (May 2007). A Profile of Public (Multi- Transportation Transportation Passenger Demographics and Travel Characteristics agency) Association Reported in On-Board Surveys. Washington D.C.: APTA. Atlanta MARTA Metropolitan Atlanta Regional Commission (ARC). 2010. Regional On-board transit Survey-FINAL Atlanta Rapid REPORT. Atlanta, GA: ARC. Transit Authority CUBIC systemwide Transfer Report FY2014, (email pers.comm w/ Robert H Thomas, Manager, Transit Analysis, MARTA); 4/29/2015 2014 Regional Breeze Transfers, (email pers.comm. w/ Aaron Fowler, AICP, Senior Transit Planner, Mobility Services Division) Metropolitan Washington Council of Governments - National Capital Region Transportation Planning Board. (July 2012). Metropolitan Station Access Alternatives Study - Final Report. Washington D.C.: MWGC. Baltimore MTA Maryland Transit Baltimore Metropolitan Council. 2010. 2007 On-board Transit Survey (BMC Analysis). Administration Baltimore, MA: BMC. Boston MBTA Massachusetts Bay Central Transportation Planning Staff (2010). MBTA Systemwide Passenger Survey- Transit Authority Rapid Transit 2008-2009: Blue Line. Boston: MBTA.; Heavy-Rail Central Transportation Planning Staff (2010). MBTA Systemwide Passenger Survey- Rapid Transit 2008-2009: Green Line. Boston: MBTA.; Light-Rail Central Transportation Planning Staff (2010). MBTA Systemwide Passenger Survey- Rapid Transit 2008-2009: Orange Line. Boston: MBTA.; Heavy-Rail Central Transportation Planning Staff (2010). MBTA Systemwide Passenger Survey- Rapid Transit 2008-2009: Red Line. Boston: MBTA.; Heavy-Rail Central Transportation Planning Staff (2010). MBTA Systemwide Passenger Survey- Rapid Transit 2008-2009: SILVER LINE. Boston: MBTA.; BRT (Blue, Green, Orange, Red Lines average) Metropolitan Washington Council of Governments - National Capital Region Transportation Planning Board. (July 2012). Metropolitan Station Access Alternatives Study - Final Report. Washington D.C.: MWGC. Chicago CTA Chicago Transit SPD-CTA.1989. Results of CTA Household Travel Market Survey. Chicago: CTA. Authority Metropolitan Washington Council of Governments - National Capital Region Transportation Planning Board. (July 2012). Metropolitan Station Access Alternatives Study - Final Report. Washington D.C.: MWGC. Cleveland RTA Greater Cleveland RED LINE TRANSFERS.xls - 2013 On-board Survey (email pers.comm. w/ Valerie - Regional Transit Shea, PE; Planning Team Leader, Programming and Planning Dept.; GCRTA); 4/23/2015 Authority Los Angeles LA Los Angeles Los Angeles County Metropolitan Transportation Authority. (2002).FY 2002 On-Board METRO County Bus Weekday Survey Report Metropolitan Volume I. Los Angeles: LACMTA. Transportation Authority Metropolitan Transportation Authority. (2004). 2004 METRO RAIL ON-BOARD SURVEY REPORT. Los Angeles: LACMTA. Los Angeles County Metropolitan Transportation Authority. (Dec. 2011).System-Wide On-Board Origin-Destinaiton Study - Draft Final Report. Los Angeles: LACMTA. access_MODEshares.xls'. Source: 'EXPORTABLE_LAMTA_OD_GROUPED_WAYPOINTS_012512.sav' and 'EXPORTABLE_OD_SUMMARY_WEIGHTED_012512.sav' SPSS files (2011-2012 System-wide On-board Origin-Destination Study Digital Database); email pers.comm. w/ John Stesney, Transportation Planning Manager III, Systems Analysis & Research, Los Angeles County Metropolitan Transportation Authority; 07/27/2015 Miami Miami- Miami-Dade Miami-Dade Metropolitan Planning Organization. (2010). Compare Transit survey results Dade Metropolitan to Actual Count Data. Miami: MDT. Transit Planning (MDT) Organization

131 CITY / AGENCY REGION ABBRV. AGENCY NAME SOURCE

Philadlephia SEPTA South Eastern Delaware Valley Regional Planning Commission. (Jun. 1992). 11 RAIL PASSENGER Pennsylvania SURVEY-I-95 Intermodal Mobility Project: Heading for the Twenty-First Century. Transportation Philadelphia: DVRPC. Authority Delaware Valley Regional Planning Commission [DVRCP]. (March 2015). Philadelphia Regional ON-BOARD TRANSIT SURVEY. Philadelphia: DVRPC. Estimated by author using : Delaware Valley Regional Planning Commission [DVRCP], DVRPC_OnBoard_Transit_Survey_2012_PublicVersion(Access Database *.accdb file); http://www.dvrpc.org/Transportation/Modeling/Data/zip/DVRPC_OnBoard_Transit_Surv ey_2012_PublicVersion.zip San BART Bay Area Rapid BART marketing and Reearch Department. (2012). 2012 BART Customer Satisfaction Francisco Transit Study. San Francisco: BART. /Oakland BART marketing and Reearch Department. (2008). 2008 BART Station Profile Study San Francisco: BART. BART marketing and Reearch Department. (2016). 2015 Station Profile Survey Preliminary Results. San Francisco: BART. Metropolitan Washington Council of Governments - National Capital Region Transportation Planning Board. (July 2012). Metropolitan Station Access Alternatives Study - Final Report. Washington D.C.: MWGC. San Juan ATI Autoridad de Puerto Rico Highway and Transportation Authority.(June 2013). Puerto Rico 2040 (Tren Transporte Islandwide Long Range Transportation Plan (Long Range Multimodal Plan): Appendix F Urbano) Integrado de - Transit Onboard Survey. San Juan:PRHTA. Puerto Rico Graterole, A.(2009). Pedestrian Accessibility and Residential Density around the Tren Urbano Rail Transit System, San Juan Metropolitan Region. Department fo Geography and Planning - University of Akron. Akron: University of Akron. Washington WMATA Washington Metropolitan Washington Council of Governments - National Capital Region D.C. Metropolitan Area Transportation Planning Board. (July 2012). Metropolitan Station Access Alternatives Transit Authority Study - Final Report. Washington D.C.: MWGC.

132 APPENDIX C

DISTRIBUTION OF EUCLIDEAN WALKING DISTANCES (WEIGHTED FREQUENCIES VS. MILES) FOR VARIOUS TRIP SEGMENTS

40,000

30,000

20,000 Trips One-Way 10,000

Euclidean walking distance between trip

origins and first boarding rapid-transit 0 station. 0 .2 .4 .6 .8 1 Euclidean Walking Distance (Miles) Between Trip Origin and First Rapid-Transit Station Frequency 95th percentile

40,000

30,000

20,000

Trips One-Way

Euclidean walking distance between trip 10,000 origins and first bus-stop in access portion of multimodal trip. 0 0 .2 .4 .6 .8 1 Euclidean Walking Distance (Miles) Between Trip Origin and First Bus Stop Frequency 90th percentile

40,000

30,000

20,000

One-Way Trips One-Way Euclidean walking distance between 10,000 alighting bus stop and first rapid-transit station in access portion of multimodal trip.

0 0 .2 .4 .6 .8 1 Euclidean Walking Distance (Miles) Between Last Bus Stop and First Rapid-Transit Station

Frequency

Max. Dist. excluding outliers

133 APPENDIX D

MODEL 1A, 1B, AND 1C PREDICTED VS. OBSERVED OUTCOME PLOTS

40000

30000

20000

aRT_AWB

10000 Model 1A

0 0 10000 20000 30000 40000 Predicted number of events

observed = fitted

40000

30000

20000 aRT_AWB

10000 Model 1B 0 0 10000 20000 30000 40000 Predicted number of events observed = fitted

50000

40000

30000

aRT_AWB

20000

Model 1C 10000 0 0 10000 20000 30000 40000 50000 Predicted number of events observed = fitted

134 REFERENCES

Agrawal, A. W., Schlossberg, M., & Irvin, K. (2008). How far, by which route and why? A spatial analysis of pedestrian preference. Journal of Urban Design, 13(1), 81-98.

Allen, W. B., Liu, D., & Singer, S. (1993). Accessibility measures of US metropolitan areas. Transportation Research Part B: Methodological, 27(6), 439-449.

Alonso, W. (1964). Location and land use. Toward a general theory of land rent.

American Planning Association (APA). (2002). Policy guide on smart growth. Chicago: American Planning Association.

APTA 2016 Public Transportation Fact Book. https://web.archive.org/web/20170328020735/http://www.apta.com/resources/statistics/pages/tra nsitstats.aspx (Accessed Dec/2016).

Arribas-Bel, D., & Sanz-Gracia, F. (2014). The validity of the monocentric city model in a polycentric age: US metropolitan areas in 1990, 2000 and 2010. Urban Geography, 35(7), 980- 997.

Bae, C.H. (2004). Transportation and the Environment, Chapter 13, In Giuliano, G. and Hanson, S. (Eds.), The Geography of Urban Transportation (pp.356-381), New York-London, The Guildford Press.

Bagley, M. N., & Mokhtarian, P. L. (2002). The impact of residential neighborhood type on travel behavior: A structural equations modeling approach. The Annals of regional science, 36(2), 279-297.

Balcombe, R., Mackett, R., Paulley, N., Preston, J., Shires, J., Titheridge, H., & White, P. (2004). The demand for public transport: a practical guide.

BART marketing and Research Department. (2008). 2008 BART Station Profile Study San Francisco: BART, San Francisco, CA.

Beimborn, E., Greenwald, M., & Jin, X. (2003). Accessibility, connectivity, and captivity: impacts on transit choice. Transportation Research Record: Journal of the Transportation Research Board, (1835), 1-9.

Ben-Akiva, M., and S. R. Lerman (1985). Discrete Choice Analysis: Theory and Application to Travel Demand. MIT Press, Cambridge, Mass.

Bliss, L. (2017, Feb 24). What's Behind Declining Transit Ridership Nationwide?. Retrieved from https://www.citylab.com/transportation/2017/02/whats-behind-declining-transit-ridership- nationwide/517701/.

135 Blumenberg, E. & Evans, A. E. (2010). Planning for Demographic Diversity: The Case of Immigrants and Public Transit. Journal of Public Transportation, 13 (2): 23-45.

Blumenberg, E. and Gregory Pierce (2014). “Multimodal Travel and the Poor: Evidence from the 2009 National Household Travel Survey,” Transportation Letters. 6(1), January.

Boarnet, M. & Crane, R. (2001). Travel by Design: The Influence of Urban Form on Travel. New York, N.Y.: Oxford University Press.

Boarnet, M. G., Giuliano, G., Hou, Y., & Shin, E. J. (2017). First/last mile transit access as an equity planning issue. Transportation Research Part A: Policy and Practice, 103, 296-310.

Brands, T., de Romph, E., Veitch, T., & Cook, J. (2014). Modelling public transport route choice, with multiple access and egress modes. Transportation research procedia, 1(1), 12-23.

Brons, M., Givoni, M., & Rietveld, P. (2009). Access to railway stations and its potential in increasing rail use. Transportation Research Part A: Policy and Practice, 43(2), 136-149.

Brown, J., & Thompson, G. (2008a). Service orientation, bus—rail service integration, and transit performance: Examination of 45 US metropolitan areas. Transportation Research Record: Journal of the Transportation Research Board, (2042), 82-89.

Brown, J. R., & Thompson, G. L. (2008b). Examining the influence of multidestination service orientation on transit service productivity: a multivariate analysis. Transportation, 35(2), 237- 252.

Brown, J. R., Thompson, G. L., Bhattacharya, T., & Jaroszynski, M. (2014). Understanding Transit Ridership Demand for the Multi-Destination, Multi-Modal Transit Network in Atlanta, Georgia: Lessons for Increasing Rail Transit Choice Ridership While Maintaining Transit- Dependent Bus Ridership. Urban Studies 51(5): 938-958

Brown, J., & Thompson, G. (2012a). What Really Matters for Increasing Transit Ridership: Understanding the Determinants of Transit Ridership Demand in Broward County, Florida. Urban Studies, 49 (15).

Brown, J., & Thompson, G. (2012b). Should Transit Serve the CBD or a Diverse Array of Destinations?: A Case Study Comparison of Two Transit Systems. Journal of Public Transportation, 15 (1), 1-18.

Brown, J., Nixon, H., & Ramos, E. (2015). The Purpose, Function, and Performance of Streetcar Transit in the Modern US City: A Multiple-Case-Study Investigation (No. CA-MTI-14- 1201).

Brueckner, J. (2000). Urban Sprawl: Diagnosis and Remedies. International Regional Science Review 23(2): 160-171.

136 Cameron, A.C. and Trivedi, P.K. (2010). Microeconometrics Using Stata (Revised Edition). College Station, TX: Stata Press.

Campbell, S. (1996). Green Cities, Growing Cities, Just Cities?: Urban Planning and the Contradictions of Sustainable Development, APA Journal, Summer: p. 296-309.

Cardozo, O. D., García-Palomares, J. C., & Gutiérrez, J. (2012). Application of geographically weighted regression to the direct forecasting of transit ridership at station-level. Applied Geography, 34, 548-558.

Carpio Pinedo, J. (2014). Dinámicas urbanas y multi-accesibilidad metropolitana: comercio urbano y demanda de autobús en la ciudad de Madrid.

Campbell, S. (1996). Green Cities, Growing Cities, Just Cities?: Urban Planning and the Contradictions of Sustainable Development, APA Journal, Summer: p. 296-309.

Central Transportation Planning Staff. (2010). MBTA Systemwide Passenger Survey-Rapid Transit 2008-2009: (Blue, Green, Orange, Red, and Silver Lines). Boston: MBTA; (Heavy-Rail, Light-Rail, BRT)

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. (1998). The Transit Metropolis: A Global Inquiry. Covelo, CA: Island Press.

Cervero, R. (2001). Walk-and-Ride: Factors Influencing Pedestrian Access to Transit, Journal of Public Transportation, Vol. 3, No. 4: pp. 1-23.

Cervero, R. (2006). Alternative approaches to modeling the travel-demand impacts of smart growth. Journal of the American Planning Association, 72(3): 285-295.

Cervero, R., & Duncan, M. (2002). Land value impacts of rail transit services in Los Angeles County. Report prepared for National Association of Realtors Urban Land Institute.

Cervero, R., Murakami, J., & Miller, M. (2010). Direct ridership model of bus rapid transit in Los Angeles County, California. Transportation Research Record: Journal of the Transportation Research Board, (2145), 1-7.

Cervero, R., Guerra, E. (2011). To T or Not to T: A Ballpark Assessment of the Costs and Benefits of Urban Rail Transportation, Public Works management & Policy, (16) 2: 111-128.

Cervero, R., & Guerra, E. (2011). Urban densities and transit: A multi-dimensional perspective. Institute of Transportation Studies. University of California, Berkeley.

137 Chakour, V., & Eluru, N. (2013). Examining the Influence of Urban form and Land Use on Bus Ridership in Montreal. Procedia - Social And Behavioral Sciences, 104(2nd Conference of Transportation Research Group of India (2nd CTRG), 875-884. doi:10.1016/j.sbspro.2013.11.182.

Chakour, V., & Eluru, N. (2016). Examining the influence of stop level infrastructure and built environment on bus ridership in Montreal. Journal of Transport Geography, 51, 205-217.

Chakraborty, A., & Mishra, S. (2013). Land use and transit ridership connections: Implications for state-level planning agencies. Land Use Policy, 30(1), 458-469.

Chalermpong, S., & Wibowo, S. S. (2007). Transit station access trips and factors affecting propensity to walk to transit stations in Bangkok, Thailand. In Proceedings of the Eastern Asia Society for Transportation Studies The 7th International Conference of Eastern Asia Society for Transportation Studies, 2007 (pp. 232-232). Eastern Asia Society for Transportation Studies.

Chandra, S., Bari, M. E., Devarasetty, P. C., & Vadali, S. (2013). Accessibility evaluations of feeder transit services. Transportation Research Part A, 5247-63.

Chen, W. (2007). Analysis of Rail Transit Project selection Bias with and Incentive Approach, Planning Theory, (6) 1: 69-94.

Chen, S. H., & Zegras, C. (2016). Rail Transit Ridership: Station-Area Analysis of Boston’s Massachusetts Bay Transportation Authority. Transportation Research Record: Journal of the Transportation Research Board, (2544), 110-122. Chowdhury, S., & Ceder, A. A. (2016). Users’ willingness to ride an integrated public-transport service: A literature review. Transport Policy, 48, 183-195.

Chu, X. (2004). Ridership models at the stop level. Tampa, FL: Center for Urban Transportation Research, University of South Florida, Report No. BC137-31

Coffel, K., Parks, J., Semler, C., Ryus, P., Sampson, D., Kachadoorian, C., & Schofer, J. (2012). TCRP Report 153: Guidelines for Providing Access to Public Transportation Stations. Transportation Research Board of the National Academies, Washington, DC.

Concas, S. & DeSalvo, J. S. (2014). The Effect of Density and Trip-Chaining on the Interaction between Urban Form and Transit Demand. Journal of Public Transportation, 17 (3): 16-38.

CNU (2018). The Charter of New Urbanism, https://www.cnu.org/who-we-are/charter-new- urbanism (Accessed 5/14/2018).

Craig, C. L., Brownson, R. C., Cragg, S. E., & Dunn, A. L. (2002). Exploring the effect of the environment on physical activity: a study examining walking to work. American journal of preventive medicine, 23(2), 36-43.

138 Creswell, J.W. (2014). Research Design: Qualitative, Quantitative, and Mixed Methods Approaches. Sage, Los Angeles, London, New Dheli, Singapore, Washington DC.

Crowley, D., Shalaby, A., & Zarei, H. (2009). Access walking distance, transit use, and transit- oriented development in North York City Center, Toronto, Canada. Transportation Research Record: Journal of the Transportation Research Board, (2110), 96-105.

Currie, G., & Delbosc, A. (2011). Understanding bus rapid transit route ridership drivers: An empirical study of Australian BRT systems. Transport Policy, 18(5), 755-764.

Dear, M. (March 2002). Los Angeles and the Chicago School: Invitation to a Debate, City & Community; Vol. 1 (1): p5-32.

De Bourdeaudhuij, I., Sallis, J. F., & Saelens, B. E. (2003). Environmental correlates of physical activity in a sample of Belgian adults. American journal of health promotion, 18(1), 83-92.

Debrezion, G., Pels, E., & Rietveld, P. (2009). Modelling the joint access mode and railway station choice. Transportation Research Part E: logistics and transportation review, 45(1), 270- 283.

Diao, M., & Ferreira Jr, J. (2010). Residential property values and the built environment: Empirical study in the Boston, Massachusetts, metropolitan area. Transportation Research Record: Journal of the Transportation Research Board, (2174), 138-147.

Dill, J., Schlossberg, M., Ma, L., & Meyer, C. (2013, January). Predicting transit ridership at the stop level: The role of service and urban form. In 92nd annual meeting of the Transportation Research Board, Washington, DC.

Dill, J., McNeil, N., Broach, J., & Ma, L. (2014). Bicycle boulevards and changes in physical activity and active transportation: Findings from a natural experiment. Preventive medicine, 69, S74-S78.

Duduta, N. (2013). Direct ridership model of Mexico City’s BRT and metro systems. Transp. Res. Rec, 2394(1), 93-99.

Durning, M., & Townsend, C. (2015). Direct Ridership Model of Rail Rapid Transit Systems in Canada. Transportation Research Record: Journal of the Transportation Research Board, (2537), 96-102.

Duncan, D. T., Aldstadt, J., Whalen, J., & Melly, S. J. (2013). Validation of Walk Scores and Transit Scores for estimating neighborhood walkability and transit availability: a small-area analysis. GeoJournal, 78(2), 407-416.

El-Geneidy, A., Grimsrud, M., Wasfi, R., Tétreault, P., & Surprenant-Legault, J. (2014). New evidence on walking distances to transit stops: identifying redundancies and gaps using variable service areas. Transportation, 41(1), 193-210.

139 Estupiñán, N., & Rodríguez, D. A. (2008). The relationship between urban form and station boardings for Bogota’s BRT. Transportation Research Part A: Policy and Practice, 42(2), 296- 306.

Ewing, R., Tian, G., Goates, J. P., Zhang, M., Greenwald, M. J., Joyce, A., & Greene, W. (2015). Varying influences of the built environment on household travel in 15 diverse regions of the United States. Urban Studies, 52(13), 2330-2348.

Ewing, R. H., & Hamidi, S. (2014a). Measuring Sprawl 2014. Smart Growth America.

Ewing, R., & Hamidi, S. (2014b). Measuring urban sprawl and validating sprawl measures. Washington, DC: National Institutes of Health and Smart Growth America; ‘Sprawl Indeces 2010-Urbanized Areas.xlx’; https://gis.cancer.gov/tools/urban-sprawl/

Ewing, R. & Cervero, R. (2010). Travel and the Built Environment: A Meta-Analysis, Journal of the American Planning Association, (76) 3: 265-294.

Foletta N, Vanderwaak V and Grandy B. (2013). Factors that influence urban streetcar ridership in the United States. Transportation Research Record, 2353: 92–99.

Forsyth, A., Oakes, J. M., Lee, B., & Schmitz, K. H. (2009). The built environment, walking, and physical activity: Is the environment more important to some people than others?. Transportation research part D: transport and environment, 14(1), 42-49.

Forsyth, A., and Southworth, M. (2008) Cities Afoot—Pedestrians, Walkability and Urban Design, Journal of Urban Design, 13:1, 1-3, DOI: 10.1080/13574800701816896

Forsyth, A., Oakes, J. M., Schmitz, K. H., & Hearst, M. (2007). Does residential density increase walking and other physical activity?. Urban Studies, 44(4), 679-697.

FTA (2008). Before and After Studies of New Starts Projects (Report to Congress). Federal Transit Administration – Office of Planning and Environment, US Department of Transportation. Washington D.C., USA.

Furth, P. G., B. Hemily, T. H. Mueller, and J. G. Strathman. (2003). Uses of archived AVL-APC data to improve transit performance and management: review and potential. Washington, D.C.: Transportation Research Board. TRCP Report No. 23 (Project H-28), http://gulliver.trb.org/publications/tcrp/tcrp_webdoc_23.pdf.

Gillen, D. W. (1977). Estimation and specification of the effects of parking costs on urban transport mode choice. Journal of Urban Economics, 4(2), 186-199.

Giuliano, G. et al. (2010). Transportation Policy: Public Transit, Settlement Patterns, and Equity in the United States, Chapter 28, Handbook on Urban Economics and Planning; Khuno-Nector Conference on Transport Economics.

140 Givoni, M., & Rietveld, P. (2014). Do cities deserve more railway stations? The choice of a departure railway station in a multiple-station region. Journal of Transport Geography, 36, 89-97.

Gordon, P., & Richardson, H. W. (1996). Beyond polycentricity: the dispersed metropolis, Los Angeles, 1970-1990. Journal of the American Planning Association, 62(3), 289-295.

Greenwald, M., & Boarnet, M. (2001). Built environment as determinant of walking behavior: Analyzing non-work pedestrian travel in Portland, Oregon. Transportation Research Record: Journal of the Transportation Research Board, (1780), 33-41.

Guan, H., Yin, Y., Yan, H., Han, Y., & Qin, H. (2007). Urban railway accessibility. Tsinghua Science & Technology, 12(2), 192-197.

Guerra, E. & Cervero, R. (2011). Cost of a Ride: The Effects of Densities on Fixed-Guideway Transit Ridership and Costs, Journal of the American Planning Association, (77) 3: 267-290.

Gutiérrez, J., & García-Palomares, J. C. (2008). Distance-measure impacts on the calculation of transport service areas using GIS. Environment and Planning B: Planning and Design, 35(3), 480-503.

Gutiérrez, J., Cardozo, O. D., & García-Palomares, J. C. (2011). Transit ridership forecasting at station level: an approach based on distance-decay weighted regression. Journal of Transport Geography, 19(6), 1081-1092.

Hamidi, S., & Ewing, R. (2014). A longitudinal study of changes in urban sprawl between 2000 and 2010 in the United States. Landscape and Urban Planning, 128, 72-82.

Hamidi, S., Ewing, R., Preuss, I., & Dodds, A. (2015). Measuring Sprawl and Its Impacts: An Update. Journal of Planning Education and Research, 35(1): 35-50. DOI: 10.1177/0739456X14565247

Hancock, G. R. (2003). Fortune cookies, measurement error, and experimental design. Journal of Modern Applied Statistical Methods, 2(2), 3.

Handy, S. (2017). Thoughts on the Meaning of Mark Stevens’s Meta-Analysis, Journal of the American Planning Association, 83:1, 26-28, DOI: 10.1080/01944363.2016.1246379.

Handy, S., Xinyu, C., and Mokhtarian, P.L. (2007). Self-Selection in the Relationship between the Built Environment and Walking: Empirical Evidence from Northern California, Journal of the American Planning Association, 72:1, 55-74, DOI: 10.1080/01944360608976724.

Hansen, W.G. (1959). How accessibility shapes land use. Journal of American Institute of Planners, 25 (1): 73–76.

Hanson, S., & Giuliano, G. (Eds.). (2004). The geography of urban transportation. Guilford Press.

141 Hilbe, J.M. (2011). Negative Binomial Regression (2 ed.), Cambridge University Press, Cambridge, U.K.

Hirsch, J. A., Moore, K. A., Evenson, K. R., Rodriguez, D. A., & Roux, A. V. D. (2013). Walk Score® and Transit Score® and walking in the multi-ethnic study of atherosclerosis. American journal of preventive medicine, 45(2), 158-166.

Hoehner, C. M., Ramirez, L. K. B., Elliott, M. B., Handy, S. L., & Brownson, R. C. (2005). Perceived and objective environmental measures and physical activity among urban adults. American journal of preventive medicine, 28(2), 105-116.

Hong, J., Shen, Q., & Zhang, L. (2014). How do built-environment factors affect travel behavior? A spatial analysis at different geographic scales. Transportation, 41(3), 419-440.

Huber, C. (2013, November). Generalized structural equation modeling in Stata. In Italian Stata Users' Group Meetings 2013(No. 06). Stata Users Group.

Iacono, M., Krizek, K., El-Geneidy, A. (2010). Measuring non-motorized accessibility: issues, alternatives, and execution. Journal of Transport Geography, 18, pp. 133-140.

Ihlanfeldt, R. K. (1995). The Importance of the Central City to the Regional and National Economy: A Review of the Arguments and Empirical Evidence. Cityscape: A Journal of Policy Development and Research. 1.

Jiang, Y., Zegras, P. C., & Mehndiratta, S. (2012). Walk the line: station context, corridor type and bus rapid transit walk access in Jinan, China. Journal of Transport Geography, 20 (1): 1-14.

Kaplan, David (2007). Structural Equation Modeling. Sage. pp. 1089–1093. doi:10.4135/9781412950589. ISBN 9781412950589.

Kim, S., Ulfarsson, G. F., & Hennessy, J. T. (2007). Analysis of light rail rider travel behavior: impacts of individual, built environment, and crime characteristics on transit access. Transportation Research Part A: Policy and Practice, 41(6): 511-522.

Kimpel, T. J., Dueker, K. J., & El-Geneidy, A. M. (2007). Using GIS to measure the effect of overlapping service areas on passenger boardings at bus stops. Urban and Regional Information Systems Association Journal, 19(1).

Knaap, G. J. (2017). An inquiry into the promise and prospects of smart growth. In Towards Sustainable Cities (pp. 61-79). Routledge.

Korf, J. L., & Demetsky, M. J. (1981). Analysis of rapid transit access mode choice. Transportation research record, (817).

Kuby, M., Barranda, A., & Upchurch, C. (2004). Factors influencing light-rail station boardings in the United States. Transportation Research Part A: Policy and Practice, 38(3): 223-247.

142 Lee, B. (2007). “Edge” or “edgeless” cities? Urban spatial structure in US metropolitan areas, 1980 to 2000. Journal of Regional Science, 47(3), 479-515.

Lee, J., Boarnet, M., Houston, D., Nixon, H., & Spears, S. (2017). Changes in Service and Associated Ridership Impacts near a New Light Rail Transit Line. Sustainability, 9(10), 1827.

Lindelöw, D., Svensson, Å., Sternudd, C., & Johansson, M. (2014). What limits the pedestrian? Exploring perceptions of walking in the built environment and in the context of every-day life. Journal of Transport & Health, 1(4), 223-231.

Litman, T. (2017a). Land use impacts on transport. Victoria Transport Policy Institute.

Litman, T. (2017b). Economic Value of Walking. In Walking: Connecting Sustainable Transport with Health (pp. 81-98). Emerald Publishing Limited.

Litman, T. (2016). Well Measured Developing Indicators for Sustainable and Livable Transport Planning. Victoria Transport Policy Institute.

Litman, T. (2013). The new transportation planning paradigm. Institute of Transportation Engineers. ITE Journal, 83(6), 20.

Litman, T. (2013). Evaluating active transport benefits and costs: guide to valuing walking and cycling improvements and encouragement programs. Victoria Transport Policy Institute.

Loukaitou-Sideris, A. (2007). Hot Spots of Bus Stop Crime, Journal of the American Planning Association, 65:4, 395-411,DOI: 10.1080/01944369908976070.

Luk, J., & Olszewski, P. (2003). Integrated public transport in Singapore and Hong Kong. Road & Transport Research, 12(4), 41.

Maghelal, P.K. and Capp, C.J. (2011). Walkability: A Review of Existing Pedestrian Indeces, URISA journal, Vol 23: 2; pp. 5-19

Mahmoud, M. S. (2015). Demand Modelling of Cross-Regional Commuting Trips in Multimodal Networks (Doctoral dissertation, University of Toronto (Canada)).

Manville, M. (2017). Travel and the built environment: time for change. Journal of the American Planning Association, 83(1), 29-32.

McLeod Jr, M. S., Flannelly, K. J., Flannelly, L., & Behnke, R. W. (1991). MULTIVARIATE TIME-SERIES MODEL OF TRANSIT RIDERSHIP BASED ON HISTORICAL, AGGREGATE DATA: THE PAST, PRESENT AND FUTURE OF HONOLULU (WITH DISCUSSION AND CLOSURE) (No. 1297).

Mees, P. (2010). Transport for Suburbia: Beyond the Automobile Age. London, U.K.: Earthscan.

143 Mees, P. (2014). TOD and multi-modal public transport. Planning Practice & Research, 29(5), 461-470.

Metro (2011). About Us, http://www.metro.net/about/ (Accessed December 2017).

Meyer, J. R., Kain, J. F., & Wohl, M. (1969). The Urban Transportation Problem. Cambridge, Massachusetts: Harvard University Press.

Mitra, S. K., & Saphores, J. D. M. (2017). Carless in California: green choice or misery?. Journal of transport geography, 65, 1-12.

Millard‐Ball, A., & Schipper, L. (2011). Are we reaching peak travel? Trends in passenger transport in eight industrialized countries. Transport reviews, 31(3), 357-378.

Mokhtarian, P. L., & Cao, X. (2008). Examining the impacts of residential self-selection on travel behavior: A focus on methodologies. Transportation Research Part B: Methodological, 42(3), 204-228.

Nelson, A. C. (2017). Compact development reduces VMT: Evidence and application for planners—Comment on “does compact development make people drive less?”. Journal of the American Planning Association, 83(1), 36-41.

Newman, P. & Kenworthy, J. (1999). Sustainability and Cities: Overcoming Automobile Dependence. Covelo, CA: Island Press.

Newman, P., Kenworthy, J., Vintila, P. (1995). Can We Overcome Automobile Dependence?: Physical Planning in the Age of Urban Cynicism, Cities, 12 (1): 53-65.

Ozbil, A., Peponis, J., & Bafna, S. (2009, June). The effects of street configuration on transit ridership. In 7th International Space Syntax Symposium. Stockholm: KTH School of Architecture and the Built Environment.

Pachauri, R. K., Allen, M. R., Barros, V. R., Broome, J., Cramer, W., Christ, R., ... & Dubash, N. K. (2014). Climate change 2014: synthesis report. Contribution of Working Groups I, II and III to the fifth assessment report of the Intergovernmental Panel on Climate Change (p. 151). IPCC.

Park, S., Kang, J., & Choi, K. (2014). Finding determinants of transit users’ walking and biking access trips to the station: A pilot case study. KSCE Journal of Civil Engineering, 18(2), 651- 658.

Parsons Brinckerhoff, Quade and Douglas, Inc. (1996). Transit and urban form. Report for the TCRP. Report no. 16. Washington, DC: National Academy Press.

Peng, Z., and K. J. Dueker. (1995). Spatial data integration in route level transit demand modeling. URISA Journal, 7: 26-37

144 PTV NuStats-LACMTA (2012). Los Angeles County Metropolitan Transportation Authority (LACMTA, a.k.a. Metro) System-Wide On-Board Origin-Destination Study, Draft Final Report (December 2011).

Pucher, J., & Renne, J. L. (2003). Socioeconomics of urban travel: evidence from the 2001 NHTS. Transportation Quarterly, 57(3), 49-77.

Pulugurtha, S. S & Agurla, M. (2012). Assessment of Models to Estimate Bus-Stop Level Transit Ridership using Spatial Modeling Methods. Journal of Public Transportation, 15 (1): 33- 52. DOI: http://doi.org/10.5038/2375-0901.15.1.3

Pushkarev, B. & Zupan, J. (1977). Public Transportation & Land Use Policy. Bloomington & London: Indiana University Press.

Ramos, L.E. (working paper; 2018). An Assessment of Bus Feeder Networks Influence on Rapid-Transit Station Boardings: The Case of Los Angeles Metropolitan Multi-modal System.

Ramos, L. E., Brown, J.R., & Nixon, H. (2015a). The Transit Performance of Modern-Era Streetcars: A Consideration of Five U.S. Cities. Transportation Research Record 2534: 57-67. doi:10.3141/2534-08

Ramos, L. E., Brown, J.R., & Nixon, H. (2015b). A Cautionary Tale of Two Streetcars: Little Rock's River Rail and Tampa's TECO Line. Journal of Public Transportation, 18 (1): 1-17.

Ramos, L. E., & Brown, J. R. (2016). A Comparative Assessment of the Factors Associated with Station-Level Streetcar versus Light Rail Transit Ridership in the United States. Urban Studies, 53 (5): 915-935.

Renne, J. L., Hamidi, S., & Ewing, R. (2017). Transit commuting, the network accessibility effect, and the built environment in station areas across the United States. Research in Transportation Economics, 60, 35-43.

Richmond, J. (2005). Transport of Delight: The Mythical Conception of Rail Transit in Los Angeles. Akron, OH: The University of Akron Press.

Rodrı́guez, D. A., & Joo, J. (2004). The relationship between non-motorized mode choice and the local physical environment. Transportation Research Part D: Transport and Environment, 9(2), 151-173.

Saelens, B. E., Sallis, J. F., & Frank, L. D. (2003). Environmental correlates of walking and cycling: findings from the transportation, urban design, and planning literatures. Annals of behavioral medicine, 25(2), 80-91.

Saelens, B. E., & Handy, S. L. (2008). Built environment correlates of walking: a review. Medicine and science in sports and exercise, 40(7 Suppl), S550.

145 Schlossberg, M., & Brown, N. (2004). Comparing transit-oriented development sites by walkability indicators. Transportation Research Record: Journal of the transportation research board, (1887), 34-42.

Schlossberg, M., Agrawal, A. W., Irvin, K., Bekkouche, V.L. (2007). How far, by which route and why? A spatial analysis of pedestrian preference. MTI Report 06-06.

Semler, C., & Hale, C. (2010). Rail Station Access–an assessment of options.

Shoup, D. C. (2004). The ideal source of local public revenue. Regional Science and Urban Economics, 34(6), 753-784.

Sivakumaran, k., Li, Y., Cassidy, M.J., & Madanat, S. (2012). Cost-saving properties of Schedule coordination in a simple trunk-and-feeder transit system. Transportation Research Part A: Policy and Practice 46(1): 131-139.

Sohn, K., & Shim, H. (2010). Factors generating boardings at Metro stations in the Seoul metropolitan area. Cities, 27(5):358-368.

Soja, E., Morales, R., & Wolff, G. (1983). Urban restructuring: an analysis of social and spatial change in Los Angeles. Economic Geography, 59(2), 195-230.

Southworth, M. (2005). Designing the walkable city. Journal of urban planning and development, 131(4), 246-257.

Stevens, M. (2017). Does Compact Development Make People Drive Less?, Journal of the American Planning Association, 83:1, 7-18, DOI: 10.1080/01944363.2016.1240044

Tabassum, S., Tanaka, S., Nakamura, F., & Ryo, A. (2017). Feeder Network Design for Mass Transit System in Developing Countries (Case study of Lahore, Pakistan). Transportation research procedia, 25, 3129-3146.

Tal, G., Handy, S., & Boarnet, M. G. (2016). Policy Brief on the Impacts of Transit Access (Distance to Transit) Based on a Review of the Empirical Literature. Sacramento, CA: California Air Resources Board.

Taylor, P. J. (2005). Leading world cities: empirical evaluations of urban nodes in multiple networks. Urban Studies, 42(9), 1593-1608.

Taylor, P. (2011). Measuring the world city network: new results and developments. In ICTs for Mobile and Ubiquitous Urban Infrastructures: Surveillance, Locative Media and Global Networks (pp. 15-23). IGI Global.

Taylor, B. D., & Fink, C. N. (2003). The factors influencing transit ridership: A review and analysis of the ridership literature.

146 Taylor, B. D., Miller, D., Iseki, H., & Fink, C. (2009). Nature and/or nurture? Analyzing the determinants of transit ridership across US urbanized areas. Transportation Research Part A: Policy and Practice, 43(1): 60-77.

Thompson, G. L., & Matoff, T. G. (2003). Keeping up with Joneses: Radical vs. multidestinational transit in decentralizing regions. American Planning Association. Journal of the American Planning Association, 69(3), 296.

U.S. Census Bureau (2010). 2010 US Decennial Census Geodatabase, Housing & Population- California Shapefile, Block level geography (machine readable data files. https://www.census.gov/geo/maps-data/data/tiger-data.html

U.S. Census Bureau (2017). American Community Survey 5-Year Estimates - Geodatabase Format, 2011-2015 Detailed Tables, Block Group level geography, California (machine readable data files). https://www.census.gov/geo/maps-data/data/tiger-data.html

U.S. Department of Transportation, Federal Highway Administration, 2009 National Household Travel Survey. URL: https://nhts.ornl.gov.

US-FHWA (2016). UNDERSTANDING TRAVEL BEHAVIOR – Research Scan. PL17-025.

Upchurch, C., Kuby, M., Zoldak, M., & Barranda, A. (2004). Using GIS to generate mutually exclusive service areas linking travel on and off a network. Journal of Transport Geography, 12(1), 23-33.

Walker, J. (2011) Human Transit: How Clearer Thinking about Public Transit Can Enrich Our Communities and Our Lives. Washington, DC: Island Press.

Walters, G., & Cervero, R. (2003). Forecasting transit demand in a fast growing corridor: The direct-ridership model approach. Fehrs and Peers Associates.

Van Acker, V., & Witlox, F. (2007). Explaining travel behavior: Towards a new conceptual framework and modeling technique. In Proceedings of the 11th World Conference on Transport Research.

Vuchic, V. (2000). Transportation for Livable Cities. New Brunswick, N.J.: Center for Urban Policy Research.

Wachs, M. (2004). Reflections on the Planning Process, Chapter 6, In Giuliano, G. and Hanson, S. (Eds), The Geography of Urban Transportation (pp.356-381), New York-London, The Guildford Press.

Walters, G. & Cervero, R., (2003). Forecasting Transit Demand in a Fast Growing Corridor: The Direct-Ridership Model Approach. Technical Memorandum Prepared for the Bay Area Rapid Transit District. Fehr and Peers, Lafayette, CA.

147 Weinberger, R., & Sweet, M. (2012). Integrating walkability into planning practice. Transportation Research Record: Journal of the Transportation Research Board, (2322), 20-30.

Wen, C. H., Wang, W. C., & Fu, C. (2012). Latent class nested logit model for analyzing high- speed rail access mode choice. Transportation Research Part E: Logistics and Transportation Review, 48(2), 545-554.

Xi, Y., Saxe, S., & Miller, E. (2016). Accessing the Subway in Toronto, Canada: Access Mode and Catchment Areas. Transportation Research Record: Journal of the Transportation Research Board, (2543), 52-61.

Yang, R., Yan, H., Xiong, W., & Liu, T. (2013). The Study of Pedestrian Accessibility to Rail Transit Stations Based on KLP Model. Procedia-Social and Behavioral Sciences, 96, 714-722.

Zhao, J., Deng, W., Song, Y., & Zhu, Y. (2014). Analysis of Metro ridership at station level and station-to-station level in Nanjing: an approach based on direct demand models. Transportation, 41(1): 133-155.

148 BIOGRAPHICAL SKETCH

Architect/Planner Luis Enrique Ramos Santiago is a native from the island of Puerto Rico and holds a Masters in Architecture from Tulane University (New Orleans, Louisiana (USA)) and a Masters in Urban and Regional Planning from the Graduate School of Planning at the University of Puerto Rico-Rio Piedras Campus (San Juan, Puerto Rico). Enrique has more than 16 years of professional experiences in architecture, urban design, and comprehensive municipal land-use planning in Puerto Rico and North Florida region (USA). Before commencing doctoral studies at Florida State University Enrique participated in scientific social-ecological studies and published as lead author in peer-reviewed journals as well as professional and cultural journals in topics related to architecture, neighborhood planning and evolution, and urban ecology.

During his doctoral studies at the Department of Urban and Regional Planning Enrique served as instructor for upper-level undergraduate classes related to urban planning and transportation, worked as research assistant, and published as lead- and co-author in several transport related peer-reviewed manuscripts and reports together with his doctoral supervisor, Dr. Jeffrey Brown. The resurgence, performance, and predictive models of ridership for modern-era streetcars (trams) in the United States were some of the topics He explored while at FSU. His current research interests lie at the intersection of travel behavior and built-environment attributes, focusing on urban and suburban contexts, sustainable non-motorized modes, and multimodal transit networks (walking, bicycle, streetcar, bus, light-rail, and heavy-rail).

Enrique is also interested in tropical architectural design and sustainable pedestrian-friendly urban design, and has a keen interest in studying early- and mid- 20th century neighborhood archetypes and their social-economic-physical/spatial-environmental evolution within a coupled social-ecological systems approach. He is married to Yasha Nicole Rodríguez Meléndez (Ph.D. Cornell ’04) and has one daughter, Mariela Alexandra Ramos Rodríguez. He is a firm believer in sustainable transportation and is a utilitarian bicyclist that registers more that 60% of his daily trips on bicycle mode. He will begin teaching and conducting research as Assistant Professor at Clemson University (South Carolina, USA), Department of City Planning and Real Estate Development in Fall 2018.

149