Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Street network connectivity and local travel behaviour: assessing the relationship of travel outcomes… Hawkins, Christopher 2007

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
24-ubc_2008_spring_hawkins_christopher.pdf [ 9.7MB ]
Metadata
JSON: 24-1.0066199.json
JSON-LD: 24-1.0066199-ld.json
RDF/XML (Pretty): 24-1.0066199-rdf.xml
RDF/JSON: 24-1.0066199-rdf.json
Turtle: 24-1.0066199-turtle.txt
N-Triples: 24-1.0066199-rdf-ntriples.txt
Original Record: 24-1.0066199-source.json
Full Text
24-1.0066199-fulltext.txt
Citation
24-1.0066199.ris

Full Text

STREET NETWORK CONNECTIVITY AND LOCAL TRAVEL BEHAVIOUR Assessing the relationship of travel outcomes to disparate pedestrian and vehicular street network connectivity by Christopher C. Hawkins  B.A., Williams College, 1994  A THESIS SUBMITTED IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF ARTS in PLANNING in THE FACULTY OF GRADUATE STUDIES  THE UNIVERSITY OF BRITISH COLUMBIA  December 2007  © Christopher C. Hawkins 2007  Abstract This research investigated the association of street network connectivity differences across travel modes with travel behaviour – mode choice, distance traveled and number of trips. To date research on travel behaviour relationships with urban form has not developed empirical evidence on street designs as distinct networks for walking and driving. A street network having greater connectivity and continuity for the pedestrian mode of travel vis-à-vis the vehicular network, like the Fused Grid, will likely encourage more walking. This hypothesis was investigated using a quasi-experimental approach within a rational utility behavioural framework. Local travel behaviour is theorized to be affected by desire to access goods and services (broadly termed, ‘activities’) in the community where people live. Using inferential statistics, the research tested for relationships between measured street patterns and self-reported local travel by King County, WA households. The main variables were ratios (walking : driving) of network connectivity and density, in the vicinity of travel survey households. Demographics and household characteristics, as well as other behaviourally influential urban form factors (residential density, proximity of destinations, etc.), were included in regression models, allowing control for confounding factors. Findings suggest that street networks with connectivity that provides better routing for one mode of transportation over others encourage more travel by the favored mode. The regression model demonstrated that a change from a pure small-block grid to a modified grid (i.e. Fused Grid) can result in an 11.3% increase in odds of a home-based trip being walked. The modified street pattern like a Fused Grid is also associated with a 25.9% increase, over street patterns with equivalent route directness for walking and driving, in the odds a person will meet recommended levels of physical activity. Finally, the Fused Grid’s 10% increase in relative connectivity for pedestrians is associated with a 23% decrease in local vehicle travel distance (VMT), and its improved continuity is associated with increased walking trips and distance. Conclusions: Other factors being equal, residential street networks with either more direct routing for pedestrians or more pedestrian facilities relative to vehicular network are associated with improved odds of walking and reduced odds of driving.  ii  Table of Contents Abstract ............................................................................................................................................... ii Table of Contents .............................................................................................................................. .iii List of Tables ....................................................................................................................................... v List of Figures .................................................................................................................................... vii Preface ............................................................................................................................................... viii Acknowledgements.............................................................................................................................. xi Chapter 1. INTRODUCTION: Context and Historical Background ............................................ 1 I. Residential Streets Then and Now ................................................................................................. 2 II. Context and Definitions: Human Activity and the Built Environment .................................... 3 III. Historical Review: Urban Transportation and Residential Street Design .............................. 6 IV. Summary… ................................................................................................................................. 18 Chapter 2. LITERATURE REVIEW: Methodological Approach............................................... 19 I. Overview...................................................................................................................................... 20 II. Review of Urban Form-Travel Behaviour Research ............................................................... 20 III. Strengths and Limitations of Method and Project Scope ....................................................... 30 IV. Conclusion ................................................................................................................................... 32 Chapter 3. METHODS ..................................................................................................................... 33 I. Introduction to Proposed Methodological Framework ......................................................... 34 II. Data -- Sources & Collection..................................................................................................... 39 III. Sample Selection and Development of Dataset...................................................................... 44 IV. Analysis ........................................................................................................................................ 45 V. Findings & Limitations .............................................................................................................. 48 Chapter 4. DATA DEVELOPMENT .............................................................................................. 50 I. Introduction ............................................................................................................................... 51 II. Final Data ................................................................................................................................... 51 III. Urban Form Variable Development ......................................................................................... 55 IV. Characteristics of Persons & Travel Behaviour from PSRC Travel Survey ........................ 61 V. Summary ..................................................................................................................................... 62 Chapter 5. RESULTS A. Descriptive Analysis............................................................................... 63 I. Introduction ............................................................................................................................... 64 II. Descriptive Analysis Results ..................................................................................................... 64 III. Summary ..................................................................................................................................... 80 Chapter 6. RESULTS B. Inferential Statistics ............................................................................... 81 I. Introduction ............................................................................................................................... 82 II. Correlations ................................................................................................................................ 82 III. Travel Behaviour Regression Modeling ................................................................................... 84 IV. Interactions, Sensitivity Testing, and Neighbourhood Differences ....................................... 90 V. Summary: Relative Strength of Explanatory Factors ............................................................ 93 iii  Chapter 7. DISCUSSION ................................................................................................................. 95 I. Findings Interpretation ................................................................................................................. 96 II. Relation to Other Research on Street Networks and Travel ................................................ 100 III. Data & Methods, Limitations ................................................................................................... 101 Chapter 8. CONCLUSION ............................................................................................................. 105 I. Overview........................................................................................................................................ 106 II. Key Findings ................................................................................................................................ 106 III. Recommendations for Community Planning Policy & Practice......................................... 108 IV. Further Research.................................................................................................................. 110 V. Summary ..................................................................................................................................... 112  Bibliography ..................................................................................................................................... 114 Appendix A. Technical - Research Design, Methods, Data and Analysis ................................... 122 Appendix B. Technical - Results/Outputs ..................................................................................... 132 Appendix C. Other - Graphics & Images ..................................................................................... 164  iv  List of Tables Table 1-1. Problems of gridded streets ....................................................................................................... 7 Table 1-2. Problems of loop and culs-de-sac streets................................................................................. 10 Table 3-1. Street network types ................................................................................................................ 36 Table 3-2. Potential variable data for Fused Grid assessment .................................................................. 40 Table 3-3. Example street network types - locations in Seattle region database ...................................... 48 Table 4-1. Final variables for assessing street networks........................................................................... 52 Table 5-1. Descriptive Statistics – Household characteristics, all persons in sample .............................. 68 Table 5-2. Descriptive Statistics -- Demographic variables (person) ....................................................... 69 Table 5-3. Travel Behaviour – trip level................................................................................................... 70 Table 5-4. Travel distance by persons ...................................................................................................... 71 Table 5-5. Trip-making by persons........................................................................................................... 71 Table 5-6. Type of Travel: work and non-work trips .............................................................................. 72 Table 5-7. Fused Grid measurements (key network variables) ................................................................ 75 Table 5-8. Network measures for all persons ........................................................................................... 76 Table 5-9. Other Urban Form Descriptive Statistics – Household-level .................................................. 77 Table 5-10. Other Urban Form Descriptive Statistics – Persons with local travel .................................... 77 Table 5-11. Disparate connectivities and walking mode share.................................................................. 79 Table 5-12. Disparate connectivities and walking distance traveled ......................................................... 79 Table 6-1. Correlation - Walk mode share and all factors ........................................................................ 83 Table 6-2. Correlation – Driving distance and all factors ......................................................................... 84 Table 6-3. All Binary Logistic Regressions, model explanatory power ................................................... 86 Table 6-4. Walking versus not walking – trip & person-level Logistic regression .................................. 87 Table 6-5. Active Walking – person-level Logistic regression ................................................................ 88 Table 6-6. Driving vs. No-driving Travel – person-level logistic regression ........................................... 89 Table 6-7. Linear Regression Results Summary – person-level ............................................................... 89 Table 6-8. VMT -- person-level linear regression .................................................................................... 90 Table 6-9. Interaction Effects – crow-fly distance and connectivity to nearest commercial .................... 91 Table 6-10. Standardized Beta Coefficients – trip-level logistic regression.............................................. 93 Table 6-11. Standardized Beta Coefficients – person-level logistic regression ........................................ 93 Table A-1. Study area network file attribute table............................................................................. 124 Table A-2. Commercial Uses for Destination Selection ......................................................................... 124 Table B-1. Trip Distance – Descriptives ................................................................................................. 132 Table B-2. Correlations – Mix of Uses & Neigh. Retail ........................................................................ 133 Table B-3. Correlations – Trip-level Walk Trips..................................................................................... 133 Table B-4. Cross-tabulation & Chi-square 1 – Walk Share & Ratio of Route Directness ..................... 134 Table B-5. Cross-tabulation & Chi-square 2 – Walk Share & Ratio of Network Density ...................... 135 Table B-6. Walking vs. No-walking, Trip-level Logistic Regression – Full Model ............................... 136 Table B-7. Walking or not Walking – Person Level Logistic Regression – Full Model ......................... 137 Table B-8. Active Walking, Person Level Logistic Regression – Full Model ........................................ 138 Table B-9. Driving versus No-driving Travel – Person Level Logistic Regression ................................ 138 Table B-10. Walking versus Not Walking, Logistic Regression – Z-score Trip Level ...................... 140 Table B-11. Walking versus Not Walking, Logistic Regression – Z-score Person Level ........................ 140 Table B-12. Drive Trips – Linear Regression – Full Model .................................................................... 141 Table B-13. Total Trips – Linear Regression – Full Model..................................................................... 142 Table B-14. Walk Distance – Linear Regression – Full Model ............................................................... 144 Table B-15. Walk Trips – Linear Regression – Full Model .................................................................... 146 Table B-16. Interaction Term Crow-fly x Route Directness Ratio – Trip-Level Log. Regression ......... 147 Table B-17. Interaction, Walk vs. No-Walk -- Person-Level Logistic Regression ................................. 148 Table B-18. Neigh. Average Route Directness Walk or Non-walk Trip ................................................. 149 Table B-19. Impeded Network (20%) Walking or Not – Person-level Logistic Regression ................... 150 Table B-20. Intersection density model with Ratio of Route Directness................................................. 151 Table B-21. Active Walking - Route Directness to Park, Logistic Regression ....................................... 152  v  Table B-22. Table B-23. Table B-24. Table B-25. Table B-26.  Nonwork Trip Logistic Regression...................................................................................... 153 Work Trips Logistic Regression .......................................................................................... 154 Physical Activity – Full Logistic Regression, Fit Diagnostic (Hosmer & Lemeshow) ....... 155 Walking or Not – Person, Logistic Regression (Hosmer & Lemeshow) ............................. 160 Driving or Not - Person, Logistic Regression (Hosmer & Lemeshow) ............................... 162  vi  List of Figures Figure P-1. Route distance and network design........................................................................................ ix Figure 1-1. Typical Fused Grid street design.............................................................................................. 2 Figure 1-2. Conceptual model of travel behaviour and urban form............................................................ 5 Figure 1-3. Growth in Vehicle Miles Traveled........................................................................................... 9 Figure 1-4. Traffic calming devices: Diverter and Traffic Signal … ....................................................... 12 Figure 1-5. Examples of New Urbanist Street Design and Fused Grid Network … ................................ 17 Figure 3-1. Households, buffers, & networks. Seattle region … ............................................................. 36 Figure 3-2. Seattle Region Study Area – travel survey households and networks … .............................. 38 Figure 3-3. Street networks -differences in connectivity across modes … .............................................. 47 Figure 4-1. Disparate Modal Networks in Seattle, WA … ...................................................................... 53 Figure 4-2. Land use - Commercial, parks, & 1km veh. network buffer around household .................... 54 Figure 4-3. GIS process - adding sidewalk attributes to network file in GIS edit session........................ 56 Figure 4-4. Preparation of network - sidewalks to be snapped to network lines ...................................... 56 Figure 4-5. Route Directness Disparity .................................................................................................... 59 Figure 5-1. Study area – Seattle, Bellevue, Redmond .............................................................................. 64 Figure 5-2. Measured Travel Networks – Length..................................................................................... 65 Figure 5-3. Measured Travel Networks – Length, per capita ................................................................... 65 Figure 5-4 Measured street networks – Network density ........................................................................ 66 Figure 5-5. Capital Hill Modified Gridiron - approximating the Fused Grid ........................................... 66 Figure 5-6. Pedestrian Connection – Seattle modified gridiron ............................................................... 67 Figure 5-7. 20th Century street network patterns ...................................................................................... 67 Figure 5-8. Route Directness Examples ................................................................................................... 74 Figure 5-9. Contrasting Networks, Seattle region study area ................................................................... 75 Figure A-1. Line files to be merged .......................................................................................................... 123 Figure A-2. Seattle neighbourhood, showing crow-fly versus network distances.................................... 125 Figure A-3. Walk Mode Share Among Travel Survey Participants ......................................................... 129 Figure A-4. Binary Categorical Variable of Walking Mode Choice ........................................................ 130 Figure A-5. Continuous Variables for Linear Regression: DistDrive & WalkTrips ................................ 130 Figure B-1. Histogram – Household Income Categories .......................................................................... 132 Figure B-2. Trip Distance Frequency Histogram ..................................................................................... 133 Figure B-3. Standardized Residual Histogram ......................................................................................... 162 Figure B-4. Standardized Residual Normal P-P Plot ................................................................................ 163 Figure B-5. Residual v. Predicted Values Scatterplot............................................................................... 163 Figure C-1. Fused Grid Design ................................................................................................................. 164 Figure C-2. Un-sidewalked local streets, Seattle study area ..................................................................... 164 Figure C-3. Intersection Density ............................................................................................................... 165 Figure C-4. Distribution of Route Directness Disparity Across Study Region ........................................ 165 Figure C-5. Fused Grid Street Design Schematic ..................................................................................... 166 Figure C-6. Example of cul-de-sac with pedestrian-only connection through public park space ............ 166 Figure C-7. Seattle Street Network (Capitol Hill) .................................................................................... 167 Figure C-8. Seattle, I-90 Trail Neighbourhood Connection ..................................................................... 167  vii  Preface This research project aims to investigate the importance of street network design to travel patterns within the local area of residential neighbourhoods. Specifically it will describe how varying street network connectivity across modes (e.g., more or less direct routing for walking and driving) relates to travel behaviour outcomes such as distance traveled, number of local trips taken and mode share, the proportion of travel by different forms of transportation. The interest in understanding which aspects of urban form are most influential to the decision of whether to walk or to drive for travel, and how much of each to do, has become acute due to converging and associated trends in vehicle miles traveled and physical inactivity. If street pattern is associated with travel behaviour, as previous studies have shown, what type is most likely to produce the kinds of outcomes – reduced reliance on automobile transportation and increased active mobility like walking – called for by much of public policy related to urban transportation planning? Transportation networks are hotly debated, as municipalities and regional governments attempt to respond to resident demands for traffic control and safety even as they achieve public objectives of healthy activity, wise land use or avoided quality of life impacts while maintaining traffic flows. A prevalent concern in planning is making urban areas and new neighbourhoods more conducive to walking activity for its community benefits, yet opinion is varied on how best to achieve livability and walkability. Many neighbourhoods, including those with both loop-andculs-de-sac and gridiron streets, demand traffic calming, which is often a costly afterthought when new streets are built to current standards that specify street widths so wide as to induce high speeds. Street networks funnel large volumes of traffic onto collectors and major streets in an effort to keep through-traffic away from localized areas of pedestrian activity, inadvertently isolating residential areas from potential destinations for walking activity. There is, therefore, interest in knowing which residential street network characteristics will best support optimal achievement of multiple urban planning goals. While the notion that street design affects how people travel is intuitively obvious – i.e. one would expect a well-connected street pattern to encourage more trips to activities due to improved accessibility – studies to date have used measures that do not distinguish well between the distinct functional networks available to walking and driving modes of travel. A wide range of research has tested broader urban form patterns, including measures of street connectivity like block or intersection density, and has concluded that there is an empirical association with travel. viii  Yet these studies by and large have not been able to clarify how disparity of network connections (differences in connectivity) across travel modes relates to travel behaviour. This is a distinguishing characteristic among street networks that may be critical to understanding why people travel the way they do. Of great interest to community planners and transportation researchers are answers to the question of how the configuration of streets into patterns (gridiron, loop-and-cul-de-sac, etc.) can be done in such a way as to reduce automobile dependence and increase physically active transportation (walking & bicycling). How can residential area streets be planned to increase the likelihood of walking? Understanding this aspect of urban form-travel behaviour interactions can provide evidence on which to base changes to street design standards, the absence of which allows other factors, including conventional wisdom or development economic efficiency, to prevail despite less certain outcomes or problematic impacts to livability and environmental quality. Proposed solutions such as the Fused Grid residential street design, and public policy emphases on transportation efficiency, air quality and chronic disease prevention through increased levels of walking, call for further analysis to discover whether and to what extent street network design matters to transportation system performance and travel behaviour at the neighbourhood scale. Figure P-1. Route distance and network design  Orthophotos courtesy of MDA Corporation.  This quasi-experimental study will not only describe how travel behaviour (total vehicle and pedestrian distance traveled, mode choice, and number of trips) relates to street network ix  design but will also seek to explain the strength of the relationship and set the stage for being able to predict a variety of other outcomes likely to result from various street designs. It begins with a literature review of street design history and urban form analytical methods (Chapters 1 and 2) and continues with development of a methodological framework and database (Chapter 3 and 4) for the research. Results of the statistical analysis are summarized in Chapters 5 (descriptive) and 6 (inferential), before discussion and interpretation of the findings in Chapter 7. The final chapter provides conclusions and recommendations for both community planning and further research on street networks, non-motorized transportation, and travel behaviour.  x  Acknowledgments This researcher and UBC’s School of Community and Regional Planning (SCARP) and Active Transportation Collaboratory would like to thank the following organizations and people for their assistance in this research: Cities of Bellevue, City of Redmond and City of Seattle, together with Metro King County, generously provided data, and their GIS staff provided guidance about its use. MDA Corporation granted permission for use of digital orthophotography for the year 1999 at two of the three cities. This critical resource allowed for checking and correction of the street and pedestrian networks across the study region. LFC, Inc. provided household travel survey and urban form data from their LUTAQH and WSDOT studies. Canada Foundation for Innovation for their generous support in providing the necessary infrastructure to accomplish this research. Their “New Opportunities” grant, along with the BC Knowledge Fund, provided the basic funding for the development of the Active Transportation Collaboratory. Jim Chapman of LFC provided support on use of the urban form and travel databases. Josh Livni of Livni Consulting and Jack Horton of ESRI provided advice and technical support on GIS analysis. Mike Danilov, doctoral student and part of the UBC Department of Statistics Short-term Statistical Consulting program, provided advice on statistical methods. Sayre Hodgson lent her statistical and scientific technical writing skills to various phases of this thesis. She also provided support in all ways to the completion of the project. Dr. Lawrence D. Frank, J. Armand Bombardier Chair in Sustainable Transportation, secured the resources and laboratory facilities to enable transportation research at SCARP, and further established the connection with Fanis Grammenos, originator of the Fused Grid residential street design. Dr. Frank secured the funding and provided steady guidance for the completion of this particular research project. Dr. Stephanie Chang provided opportunities to learn about quantitative methods in the course of my completing SCARP’s core curriculum and classes. She provided invaluable assistance with the development of a sample and methods for the research, and served on the thesis committee. And finally, Fanis Grammenos was the inspiration for this research. His tireless pursuit of new knowledge about effective community design, in particular the development of better street systems for residential neighbourhoods, provided a steady supply of information, references and motivating stories about how new communities could benefit from this research. Without the support of Mr. Grammenos, and his agency the Canada Mortgage and Housing Corporation, this research would not have been possible. Mr. Grammenos was the external reviewer on the thesis committee.  xi  Chapter 1. INTRODUCTION: Context and Historical Background  1  I. Residential Streets Then and Now Residential streets are basic features of urban communities in the industrialized regions of the world built during the past century. These local street networks have been designed and built mainly by private developers in accordance with local government standards for new development. The standards arise from a history of street design in which more specialized and hierarchical street patterns have evolved driven, particularly in North America in the last hundred years, by rapid conversion to an automobile-focused transportation system. Increasing awareness of the impacts to the environment and quality of life from automobile dependence (Newman & Kenworthy 1989) has led to proposed design interventions. Key recommendations or studies in this effort to tame traffic, however, have either 1) reinforced some of the street network characteristics that contribute to auto-dependence or 2) retreated to past urban forms.1 In recent years local communities and municipalities have been searching for street configurations that will provide optimal community health and solutions to increasing traffic impacts or diminished walkability, but durable solutions have proven difficult to identify and implement. The Fused Grid street network design is a proposed answer to these challenges and prompts this inquiry into the history of residential street design and methods for evaluating street network configurations. This pattern of streets calls for provision of ample connecting paths for non-motorized forms of travel while creating less-direct loops and culs-de-sac for motorized vehicles. Figure 1-1 shows this pattern: dark areas are public plazas or open space, broader white lines are streets and smaller white lines indicate pedestrian-only pathways. Figure 1-1. Typical Fused Grid Street Design  Image Source: Grammenos, et al. 2005  1  See Marshall (2005) for discussion of Colin Buchanan’s Traffic in Towns, which proposed keeping pedestrians separate from automobile routes, thereby relegating ‘environmental areas’ to pockets without connection to the surrounding urban area. Also, see Boarnet & Crane (2001), Grammenos, et al. (2005), or Marshall (2005) about uncritical acceptance of New Urbanist design recommendations for gridiron street networks.  2  The research finds that street design has been affected by available or predominant transportation technologies, cultural preferences and ideologies of healthful urban form and safety, institutional arrangements, and most recently by concern about quality of life and the environment. Emerging evidence about environmental impacts, livability, and public health are setting the stage for new street network designs based in better understanding of environmental influences on travel behaviour. Past designs and retrofitting of newer streets, while part of the solution, may not provide the most satisfactory results. This chapter provides introductory context, definitions, and history that prompted this research before the report turns to methodologies for assessing the performance of different patterns of streets.  II. Context and Definitions: Human Activity and the Built Environment Streets and systems of roadways are a fundamental part of planning for development and urbanization. This has been true since the earliest urban settlements in human history, and these transportation networks have increasing importance in the present, hyper-mobile urban areas of North America. As one would expect from the parts of the environment that provide the critical functions of access and circulation for human settlements, streets constitute the bulk of public land in urban areas, as much as 1/4 to 1/3 or more of all land within cities. Streets provide the connections among the various locations of human activity within a given area. Because they are part of the public domain, streets are governed more directly and managed more easily by public agencies than other land; therefore, streets are features of the built environment where interventions that assist in the achievement of public goals, especially those related to access, mobility and public health, are more readily and immediately implemented by these agencies. How streets, and their overall configuration in street networks, are designed influences their use by people. Some dimensions influential to mode choice and other travel behaviour outcomes are: width of vehicular area (lanes), streetscape (extent and quality), access points, speed limit/design speed, surface quality, crossings/intersection design, and network type (Ewing, et al. 1996; Metro Regional Services 1997; Burden 1999). For example, a wider street with more lanes of traffic provides greater capacity for vehicular movement yet poorer accessibility and safety for pedestrians. This example shows that there are trade-offs in mode choice brought about by street design. Beyond travel, streets can serve a variety of other public  3  purposes described further below. They are central features not only to travel function but also to urban quality of life in a broader sense. Two main categories of urban form measurement that affect travel behaviour, and the access people are seeking to an urban area’s activities, are connectivity and proximity (Frank 2000; Frumkin, et al. 2004). Connectivity, the extent to which pathways for travel are linked, alongside other street design features, is considered to be the main transportation design element of the built environment’s influence on travel behaviour. It is also a main characteristic distinguishing different types of street network. Proximity, how close destinations are from the point of departure (called an ‘origin’ in transportation planning), is the other key environmental dimension affecting travel behaviour but derives more from land use attributes of density and mix (or diversity) of uses (Cervero & Kockelman 1997). This research recognizes that these factors co-vary (a fine-grained street network with high connectivity is known to increase walking activity and yet it also usually occurs where density is highest, meaning high proximity to activities), and so a main challenge for understanding the influence of street networks is to isolate connectivity from other urban design and land use factors. Greater access can be achieved through either means, high mobility through a connected network, or high proximity through more density and mixing of uses. The former means of improving access is difficult, not to mention highly resource-intensive, to implement for all modes at once. People can notice this influence when moving from place to place in an urban setting: some routes are available, others are not, based on the arrangement of objects (buildings) and spaces that may be used (parks, plazas, and streets); some activities are easy to reach on foot because they are close by, others may require a different mode of travel because they are distant. The focus of this research is on understanding the factor of connectivity, that is, the extent to which travel routes are linked together to form an interconnected system of routes that facilitates movement. Disparate connectivity across modes, a characteristic of the Fused Grid street design and the focus of many traffic management efforts, is of particular interest. Over the past twenty years, a wide array of research has attempted to uncover the effects of urban form on travel behaviour (Frank & Pivo 1995; Friedman, et al. 1994; Moudon, et al. 1997; Holtzclaw, et al. 2002; Frank & Engelke 2001) and various reviews have summarized the findings of these studies and defined a research agenda (Cervero & Kockelman 1997; Boarnet & Crane 2001; Ewing & Cervero 2001; Frank & Engelke 2001; Dannenberg, et al. 2003). Built environment has been shown to be most strongly related to distance and time traveled, whereas 4  socioeconomic factors are more strongly related to other travel behaviours (i.e. frequency of trips). Both sets of factors show some relationship to people’s choice of travel mode (Frumkin, et al. 2004). These factors are summarized here in a conceptual model (Figure 1-1). The measurable factors of the built environment are generally categorized into three main areas: residential or employment density, land use mix, and connectivity. Each of these built environment factors, often referred to as density, diversity and design, has been shown to have a significant relationship with travel behaviour outcomes. Figure 1-2. Conceptual Model of Travel Behaviour and Factors of Influence  More recently, researchers have begun to try to unravel the complex warp and weft of urban street fabric. Studies have supported New Urbanist principles, contended that there is not enough evidence to recommend one street pattern over another, or found that these principles overlook key preferences associated with conventional suburban form (and in the case of streets, the cul-de-sac). One analyst (Marshall 2005) has found that conventional network hierarchy is based on a faulty notion of an inverse relationship between access and mobility, creating a tendency to construct massive, free-flowing transportation infrastructure in places where access need is low which in turn gives rise to sprawling new development as the improved mobility generates demand for new access. 5  The following historical review will discuss key themes in the history of North American residential street networks, especially their design and the understanding of them, in response to cultural, ideological and technological developments over the past 150 years. It will continue in Chapter 2 with a review of the methods and scientific literature on analyzing the relationship between urban form, especially street networks, and travel behaviour. These reviews constitute important context for further investigation of the transportation performance of street network types and of the Fused Grid in particular.  III. Historical Review: Urban Transportation and Residential Street Design The Origins of Residential Streets Though street networks, and the gridiron in particular, have been built since the first cities were constructed (Grammenos & Pollard 2005), residential streets are a more recent and now ubiquitous phenomenon in urban transportation planning. This history, then, focuses on the past century and a half of changes in urban street networks. The history of metropolitan transportation development in North America runs through a series of distinctive eras (Muller 1995), each powerfully shaped by the emerging transportation technologies characteristic of the era: walking/horsecar (up to 1890), streetcar (1890-1920), automobile (1920-1945), and freeway (since 1945). The rectilinear grid street network prevailed in urban areas throughout the streetcar era as people’s primary transport mode continued to be pedestrian, necessitating a high level of walking accessibility to complement locally the pedestrian-dependent modes of longer-distance transit. The spreading use of automobiles began to reshape urban areas in the first two decades of the twentieth century, prompting some obvious changes to the physical design of streets – the accelerated improvement of road surfaces, the addition of traffic lights and eventually the removal of streetcar tracks. New Designs for Healthy Neighbourhoods – Garden Cities & the Curvilinear Street Even before these technological transformations, various street design innovations were proposed, with both aesthetic and social reform goals in mind. In fact, the appearance of urban streets as crowded and dark, due to their narrowness, the presence of buildings at the edge of the roadway, the mixing of various often incompatible uses (like factories), and, in older cities, organic forms derived from new streets accreted onto those constructed in prior eras, associated 6  them in public and professional perception with ill-health (Southworth & Ben-Joseph 1997; Frank, et al. 2003). Adding to this perception was the emergence of poor function (traffic congestion) and sanitation problems created by multiple modes of travel now competing for street space, including prevalent horse-drawn vehicles (Jacobs 1961). Early residential street plans by Olmstead and Vaux responded to this challenge with wide, curving and discontinuous street designs that included ample landscaping to make the street a park-like setting of fresh air and open space, complementing larger house lots with private yards. The emphasis in the emerging land use planning and street design fields on a suburban ideal of ample “land, light, and air” provided by residential development (Robinson 1916) coincided with increasing mobility in the form of motorized transportation. As automobile use became increasingly prevalent during the first quarter of the 20th century, and began to supplant walking as the most basic form of transport in the middle decades, higher volumes and speeds of vehicular traffic began to expose the weaknesses of the gridiron street pattern at the neighbourhood scale. At larger scales in the urban context, the new challenge of managing traffic, regardless of local street configuration, emerged. The problems of the gridiron stem from allowing ease of traffic through-movements (and attendant impacts) in all areas, its frequent, multi-conflict intersections, and its monotonous form (Southworth & BenJoseph 1997; Grammenos, et al. 2005). Some of the supposed drawbacks of the grid are summarized in the table below (Table 1-1). Thus, the move toward curvilinear, discontinuous street patterns was further encouraged. Table 1-1. Problems of the Gridded Streets (from Southworth & Ben-Joseph 1997; Kulash 2001; Grammenos, et al. 2005) A. Contributes to dispersion of vehicular travel to all streets, even for pass-through travel purposes, and hence increased traffic impacts on neighbourhoods, including: •  Pollutant emissions due to higher automobile volumes in close proximity to residences  •  Exposure to safety risks to children at play and pedestrians from increased volume of traffic possible on all streets  B. Frequent four-way intersections also increase traffic impacts: •  Congestion, and associated increase in air pollutant emissions from frequent stops and starts, (especially as traffic volumes increase)  •  Exposure to safety risk for both motorists and pedestrians due to complexity of turning movements  C. Monotony: if unaccompanied by design requirements, gridiron networks can be uninteresting in their regularity and discouraging to pedestrians 7  A main response by municipalities was to develop standards for street construction that initially ensured a more orderly, regular appearance to the street environment (e.g. Bye-law streets, see Southworth & Ben-Joseph 1997). With a cultural preference, at least in the United States, for suburban living together with public policy (incentives for new housing) and transportation technology (automobiles) which encouraged the spread of low-density, residentialonly districts (Kunstler 1993; Duany, et al. 2000), these problems pushed street design in the direction of Olmstead, Vaux and the successor Garden City suburban models of residential street design as well as toward modifications to existing grid networks that would allow more freely flowing motorized traffic (one-way restrictions, etc.). Clarence Perry’s concept of neighbourhood units comes from this historical period – partly the result of an effort to reduce the impacts of grime, noise and sound from traffic, and also to build an integral community that would be safer and more private in character. Curvilinear street configurations were the preferred form of circulation for districts that would have no continuous through streets, thereby, it was thought, avoiding the grid’s pernicious effects on neighbourhood quality of life (Grammenos, et al. 2005). The influential concepts coming from proponents of suburban form, particularly Clarence Stein and Henry Wright with their design for Radburn, NJ (1929), were readily adopted by developers as they required less total land devoted to streets than the gridiron and were highly marketable to Americans seeking houses with large yards and privacy while being within driving distance of urban employment. Loop and cul-de-sac street network design often increased buildable area – less land devoted to streets enables more saleable lots to fit in subdivision -- and reduced infrastructure costs per unit of residential development. The Radburn suburban street network, with its culs-de-sac and curving streets, provided the template for the street networks of most residential developments in the following half century, though usually only in pieces of the whole design concept by any given development and often in some distorted fashion – a) the ‘superblock’, b) complete separation of pedestrians from traffic, or c) large tracts of open space not well-integrated with the neighbourhood (Girling & Helphand 1994). More Traffic, More Traffic Impacts, and Street Standards – The Freeway Era The years following World War II unleashed a huge demand for new housing which was met in the United States by continued government support of suburban development in sprawling areas around the edges of existing development (Frumkin, et al. 2004). The federal funding of the Interstate Highway system beginning in the 1950s accelerated the process. While 40 years 8  since this housing boom has not matched its pace or national sweep, the economic incentive of cheap land to develop at the fringes of metropolitan areas has remained largely in effect as have conventional suburban designs. This is due to the emphasis in policy, from local government to federal, on serving supposed public expectations of particular housing types and meeting projected steady increases in vehicle usage, and the persistence of subdivision street standards that date from the early years of the Freeway Era. The results, especially increasing automobility, are evident in long term trends of vehicle use shown in Figure 1-3. Figure 1-3. Growth in Vehicle Miles of Travel (VMT) and VMT Per Capita, 1970 - 1993 Washington State  Source of graph data: CTED 1993.  As new development occurs currently in urbanizing areas, municipal governments require construction of (or contribution towards) streets according to adopted standards. These street standards specify the dimensions noted above and result in the form that streets and street networks take. (Chellman 1999; Kulash 2001) Street standards have evolved over time to meet changing demands for transportation system function. The streets required by standards generally shifted from highly connected to disconnected, and from non-hierarchical to hierarchical, and in residential districts from rectilinear to curvilinear. In the past one hundred years, block sizes have increased in response to a shift toward increasing use of higher speed 9  vehicles. (Southworth & Ben-Joseph 1997; Marshall 2005) Responding to increasing automobile use, trends in subdivision design moved in the direction of privacy, security and perceived child safety (Southworth & Owens 1993). These changes, in a positive feedback loop with land use and housing policy, have generally tended to create relatively greater ease and convenience of automobile travel while eroding that of non-motorized travel and transit. Conventional suburban street networks’ lack of connectivity, in single-use zones, combined with heavy vehicular traffic on frequent collector streets, make them inhospitable to walking. Problems with typical suburban street forms began to surface within the first generation of their predominant use and are described below in relation to their quintessential residential format as loop-and-cul de sac streets (Table 1-2). Table 1-2. Problems with Loop and Culs-de-Sac Streets (Kulash 1991; Duany, et al. 2000; Grammenos, et al. 2005) A. Contributing to route indirectness: •  Pedestrian routes are elongated due to dendritic (sparse) street pattern  •  Fewer choices of route for motorists  B. Increase in bicycle/pedestrian personal safety risk: •  Pedestrian pathways, if kept separate from streets in an effort to improve their directness, have reduced natural surveillance  •  Separate pathways frequently result in mixing of bicycle-pedestrian modes without appropriate design treatments, creating conflicts and higher likelihood of collision (especially when paths meet or cross a roadway)  C. Curvilinear streets are disorienting, and thus discouraging to walking activity D. Disconnected local access streets focus traffic onto larger classification streets closer to residences: •  Diminished accessibility for pedestrians (larger streets to cross with higher volumes of traffic) closer to home and often between home and shopping or other destinations.  •  More difficult left turns for motorists due to infrequency of intersections, larger collector streets, and higher traffic volumes on limited number of major streets (sparse network)  Community Uprisings – Tame the Traffic; Make it Safe to Walk Again Challenges to Orthodox Street Planning At the time of Jane Jacobs’ classic work (1961), various citizen movements re-asserted the importance of various local street features to the livability of neighbourhoods. Among the primary concerns were traffic volumes and the wholesale reconfiguration of urban 10  neighbourhoods to accommodate roadway widening and preferred alignments for highways and arterial streets. Jacobs articulated the principles of urban vitality underlying such discontents (exchange fostered by urban characteristics of density, mixing of uses, and ample pedestrianoriented pathways) that could counteract the “erosion of the city.” The street environment figured prominently in this analysis, and Jacobs described streets based on her experience of residing in a dense urban fabric that provided for many more uses than vehicular movement. To this way of thinking, how public spaces (and sidewalks and streets in particular) are designed is responsible for improved safety and either encouragement or discouragement of human activity. This is the classic urbanist message. “But multiplicity of choice and intensive city trading depend also on immense concentrations of people, and on intricate minglings of uses and complex interweaving of paths.” – Jane Jacobs, The Death and Life of Great American Cities, 1961, p. 340.  If streets are laid out in large blocks and with wide lanes to provide more vehicle capacity, then vehicles become the chief mode of movement. Conversely, if many uses are proximate to each other along streets with ample space for casual contact and interaction as well as frequent crossings, this facilitates more visual contact, pedestrian activity, and successful city development. Variants of Jacobs’ ‘eyes on the street’ observation of sidewalk safety were elaborated in later research (Newman 1972) indicating that streets provide natural surveillance and pointing out the problems with ‘superblocks’ (large development sites without streets connecting into or through them) from a public safety standpoint. Crime prevention through environmental design (CPTED), the codification of these theories, has become a major principle in cities’ urban design objectives. Lastly, Jacobs’ prescient analysis anticipated the advent of traffic calming as she discussed the means of accomplishing the attrition of cars by cities. Concurrent with these overarching shifts in ideas, and later the process, of good street design, citizens, communities and neighbourhoods have reacted to the ills of automobiledominated transportation systems. Calls for increased traffic management, in the form of traffic calming, or full closure of streets have occurred in neighbourhoods of many types, beginning during the middle decades of the 20th Century in major metropolitan areas such as London or the San Francisco Bay area (Appleyard 1981).  11  Traffic calming and management Some features of traffic calming known to work are street closures and diverters, which create interruptions in the street network thus blocking or redirecting through-traffic. These achieve the aims of reducing traffic speeds and volumes in the area where they are applied, and are similar to the assumed affects on traffic of the Fused Grid and other selectively closed street networks. Appleyard’s research focused on how neighbors interacted along various sizes of streets (primarily in terms of volume but also speeds), and found through interviewing that the size of street is inversely related to the number of interactions that occurred among residents. Newman likewise found personal safety enhanced by grid street networks closed to through traffic (see “Private Streets” in Newman 1996). This provided evidence to suggest that if neighborliness is a goal for urban development, designs and interventions may be needed which curtail either traffic speed or volume, or both, on local residential streets. These are exactly the kind of changes that Appleyard proposed for residential neighbourhoods to offer them more protection. These livability discoveries, coupled with the traffic safety research linking collision severity (in terms of injuries and deaths) with vehicle speeds (Daisa & Peers 1997), has led to more widespread use of traffic calming around North America in the last few decades. The devices are often applied over whole neighbourhoods or districts in an effort to manage the potential problem traffic diversion into adjacent areas, creating new volume issues there. Figure 1-4 shows two different devices, one (left) which redirects vehicles while allowing pedestrians and bicyclists to continue as on a gridded network, the other (right) which only slows vehicles. Figure 1-4. Traffic Calming/Management Devices: Diverter & Traffic Circle  Image Source: Adapted from Ewing 1999.  12  Another issue with calmed street design is the potential to delay emergency response or obstruct other service/utility vehicles. These public safety and service considerations have limited the extent to which residential streets can be narrowed or have devices or parking placed on them (Burden 1999; Burden & Zykovsky 2000). Nevertheless, traffic calming has become an increasingly standard practice across the United States as various neighbourhoods seek to prevent the impacts of motorized traffic (Ewing 1999). Other street design dimensions can be altered, for instance narrowing drivers’ field of vision with curving alignments or vertical streetscape elements to enclose the roadway (Dumbaugh 2005) and creating interruptions in the street network (‘T’ intersections or closure/diverter treatments) without significantly impeding emergency response or other services requiring large vehicle movements. Public Health, Livability Goals and Street Design Subsequent research (Moudon & Untermann 1987) and public policy (ISTEA 1991; Kulash 2001) have continued to assert the varied uses of streets and public goals of how streets are designed. As noted above, urban form and streets in particular have evolved in response to changing transportation technology; other research (Jackson 1985; Goddard 1994; Frank et al., 2003) has pointed to the importance of shifts in social thought and public policy in redirecting infrastructure investment and thereby favoring the development of particular environments and associated travel patterns. Through most of the 20th Century, North American public policy favored automobile-oriented development. Over the past four decades concern for social justice and increased attention to protection of the natural environment, alongside new understandings from public health which emphasize the benefits of walking in the home neighbourhood, has prompted changes in policy and planning, some of which have implications for street design. An outpouring of interest and research related to environmental impacts of transportation and sprawling land use patterns occurred during the late 1980s and continued throughout the 1990s to the present. The relationship between the built environment and physical health of the human population has been the target of investigation in the past ten years as an epidemic of obesity spreads across North America (see Frank, et al. 2003; Frumkin et al., 2004; Sallis, et al. 2004). The preponderant evidence from these studies suggests that urban form influences people’s travel behaviour, the amount of pollutant emissions from transportation sources, and people’s levels of physical activity. In addition, better understanding of non-point-source water pollution, especially stormwater runoff from streets and paved surfaces, has created a focus on 13  urban form in water quality management (Arnold & Gibbons 1996; Girling & Kellet 2005). These notions of physical and environmental health are related to other behavioural responses to public space, including streets, which have been studied in sociological and urban design literature – people may have better mental health, and build more social capital, in environments conducive to face-to-face interaction and walking (Whyte 1972; Gehl 1980; Frumkin, et al. 2004). Earlier notions of the influence of built environment, from Jacobs through Gehl, have been confirmed, but also to some extent confounded, through statistical scientific inquiries and methods that will be discussed in the literature review that follows. “Suburban sprawls generally make dull places to walk, and a large subdivision can become numbingly repetitious at three miles an hour instead of thirty of sixty. Many suburbs were designed with curving streets and cul-de-sacs that vastly expand distances… an example of an Irvine, California subdivision where in order to reach a destination a quarter mile away as the crow flies the traveler must walk or drive more than a mile.” – Rebecca Solnit, Wanderlust, a history of walking, 2000, p. 253.  Contemporary Responses and Barriers – Searching for Optimal Street Patterns The Freeway Era has gone through multiple stages, and the evolution and maturation of suburban bedroom communities as uniform sociospatial clusters into cities in their own right (an urban form that has come to be known as ‘edge cities’ after Garreau 1991) is a significant new phenomenon (Muller 1995). These mature suburban centers consist of a variety of services and employment locating near highway interchanges and thus within easier reach of what had been residential-only developments. Such infill shows the tendency of modern suburbs to evolve toward denser, mixed-use centers. Moreover, it indicates, in a period of diminishing infrastructure budgets for local governments, unstable energy supplies, and rising fuel prices, the need for solutions that redesign these emerging centers in ways that balance their transportation systems with enhanced travel choices and reduced automobile dependence (Randall 2002). The improved understanding of urban form and travel behaviour, together with municipalities’ objectives of providing multiple travel choices to achieve optimal transportation service, compels a search for better designs of streets and street network. A few of the key considerations (cost, compatibility with neighbourhood livability, feasibility of adoption) for implementing a shift in street design and traffic management are discussed below.  14  Cost Inefficiency Because traffic calming interventions are typically done as retrofits to existing streets, they may be more expensive means of achieving the livable neighbourhood and multiple-use, multi-modal residential street goals. A more cost-effective approach would seem to be designing streets well so that they perform as intended (keeping traffic slow and moderate in volume) from the moment they are opened for use. “…it is important to note that traffic calming is often necessary only because streets have been built the wrong way to begin with, unnecessarily wide and with too much distance between intersections.” Duany et al., Suburban Nation, 2000, p. 37.  Many municipalities are therefore turning to revisions of streets standards to accomplish the same effect as traffic calming. By adopting residential street standards for narrower widths, allowance of on-street parking on both sides of the street, and creating visual or physical friction (using landscaping areas and curb extensions), they are attempting to achieve new streets that are calm by design and therefore more encouraging of use for walking and other neighbourhood activities. Residential developers in turn, look for designs that allow flexibility and maximization of buildable area, seeking to minimize as much as possible the costly infrastructure of streets. Narrow streets would seem to be attractive then, as would disconnected designs. There are, however, practical limits as to how narrow and how disconnected a street network can be while still providing the desired level of access to a wide array of users. Capitalizing on improving proximity from infill development to achieve livability and walkability will require addressing the connectivity to these new nearby destinations. Compatibility with Neighbourhood Livability Tradeoffs in different approaches to the transportation network, how streets are arranged or relate to each other and how connected the system is, arise when designing streets to provide livability and balanced travel options. For example, cul-de-sac designs often leave residual open space that is poorly programmed for the neighbourhood and therefore relatively underused by residents (Duany, et al. 2000; Girling & Helphand 1994), whereas pure gridiron streets often leave little open space or remain insensitive to variations in topography or the environment more broadly (Southworth & Ben-Joseph 2004). Further, from a transportation and land use planning  15  perspective, the effects of the local network on travel patterns at the city and regional scales are issues on par with livability within neighbourhoods.2 Resistance to Adoption Under current street standards for new development, communities continue to have disputes about various dimensions of streets, including the extent to which the street network should be connected (Handy, et al. 2003) and how much traffic should be diffused through neighbourhoods versus channeled onto major streets (City of Olympia 2005). Into the 1990s the focus of urban design and planning had been on revitalizing the urban centers and protecting older neighbourhoods from traffic impacts. Such planning efforts are important, yet the predominant growth of metropolitan areas, in population and employment, has been on the edges, either as the maturation of former edges occurs (noted above) or as land is converted from rural uses to suburban ones. The solutions proposed often ignore the particular needs of edge communities, and into this design vacuum have come continued use of street patterns that allow the maximum number of buildable lots – loop and cul-de-sac street networks (Southworth & Owens 1993), designs whose results are seemingly well-known and therefore more acceptable to local governments. While the conventional suburban street design does address the desires for safety and livability for the residences situated at or near the ends of the culs-de-sac, the larger neighbourhood and city-wide context is one of auto-dependency, featuring a limited number of busier, often congested collector streets and poor environments for walking.3 The Debate - Walkable Grids vs. Neighbourly Culs-de-Sac In debates about design and planning of urban form, recent attention has been focused on street networks as a key feature of livable neighbourhoods and the achievement of more sustainable land use and balanced transportation systems (defined as a better range of mode choice and more even mode split). The contention of New Urbanists (Calthorpe 1993; Duany, et al. 2000) is that use of a gridiron street network is the basis for achieving the needed access and legibility to encourage and support pedestrian travel, though there has been some acknowledgement that the grid should be interrupted (Ewing, et al. 1996). Opposing this response to the street design problem is the notion that loop and cul-de-sac street patterns of the 2  A parallel instance of calming on major streets are the efforts to create road diets whereby a motor vehicle lane is converted into either pedestrian or bicycle space (see Burden & Lagerway 1999) 3 This in turn places more traffic on large streets leading to centers of employment, the impact of which is often felt in the older neighbourhoods closer to cores in the urban region or in exacerbating the poor linkages between residential and commercial areas, where most destinations in newer suburban developments lie.  16  past half century have been responsible for the benefits quieter, safer neighbourhoods due to restrictions on vehicular access that funnel most traffic away from neighbourhoods onto major streets which bound them (Southworth & Ben-Joseph 1997). This debate has continued in recent articles (Southworth & Ben-Joseph 2004; Grammenos & Pidgeon 2004) as a fusion of gridiron and loop-and-cul de sac street networks, the Fused Grid design (Grammenos, et al. 2005), has been proposed to retain the benefits of both street patterns while mitigating their several defects. Its contrast with a New Urbanist design is shown in Figure 1-5. This is not simply an academic debate – local planners and local governments recognize the importance of connected street networks for emergency response and reducing some traffic impacts but are also faced with the competing interest of retaining the neighbourhood quality provided by quiet dead-end streets (Handy, et al. 2003; Twaddell 2005). The debate marks the starting point for this research to develop an empirical basis for describing and analyzing the travel behaviour effects of street network design. The current research project aims to provide insights into the likely transportation and associated quality of life outcomes of implementing designs that provide a high degree of connectivity for non-motorized travel modes and relatively disconnected neighbourhood routes for motor vehicles. Figure 1-5. Examples of New Urbanist Street Design and Fused Grid Network  Laguna West New Urbanist Network  Fused Grid Residential Quadrant  Image source: Grammenos, et al. 2005.  The prescriptions of New Urbanists for areas like new towns, infill development, and edge cities are one attempt at correcting the problem of street networks that do not support a maturing suburb with a settled population. These areas increasingly demand more travel choices than driving the car and more livability than sprawling development along major roads. Traffic 17  calming and street re-designs like the Fused Grid may be used as interventions to improve quality of life for these maturing suburbs or as a way to revive older suburbs or even inner city or urban core districts with gridiron street networks. New development can be guided by street standards that support a more full range of travel choices and improved opportunities for healthy living through active transportation. What has been lacking is not only clarity about the problem and its structure being addressed (Marshall 2005) but also empirical evidence to back up design guidelines or standards.  IV. Summary A residential street design, in order to gain acceptance and come into use, must be proven to work for a variety of municipal objectives. Chief among these are people’s mobility (nonmotorized travel, transit access and vehicular traffic flow), pedestrian and overall traffic safety, air and water quality, community health, and neighbourhood livability. In light of their health and environmental outcomes, the automobile dependent designs of sparse street networks are in need of retrofit and correction and better street designs for new development are urgently needed. Local governments are caught between competing objectives in street design – improved connectivity for its transportation benefits versus closed, quiet streets for their benefits in improved safety and reduced neighbourhood traffic impacts. The Fused Grid design, an interrupted or modified gridiron pattern, offers a potential middle ground among street network types, and suggests the availability of more optimal solutions than currently exist in most municipal street standards. Those street standards typically do not specify a network configuration or connectivity as much as they set up rules that lead, given the economics of development, to status quo, dendritic and poorly connected street patterns. This research investigates street connectivity as a factor in travel and will provide additional information regarding the effectiveness of networks with the Fused Grid street network’s characteristics.  18  Chapter 2. LITERATURE REVIEW: Methodological Approach  19  I. Overview - Street Networks and Travel Behaviour The goal of this research is to provide improved understanding of the influence that street networks have on travel behaviour. A central, persistent question of planners and officials grappling with how to best plan and guide development, especially when considering new design proposals, is will the traffic work? (Kulash 1991) What are the implications of a particular street configuration for transportation system function? Because the subject of this assessment is a residential street network design, the research is interested in how a neighborhood’s street network affects the travel patterns of its residents. How does people’s travel, measured in miles traveled, number of trips taken, and modes chosen, differ among neighbourhoods with different patterns of street network connectivity? The salient characteristic that distinguishes among street network types is the kind of routing that they provide, usually evaluated by using a measurement of its circuitry or connectivity, because this dimension affects travel distance and thus the comparative costs of various travel mode choices. To date, the research on travel behaviour and urban form has not isolated the influence of variation in street network connectivity across different travel modes. Yet the degree of connectivity is a critical feature of new neighbourhood street design proposals and of related urban planning efforts such as traffic management. These techniques, as noted in the previous chapter, aim to reduce motor vehicle impacts while enhancing and encouraging walking or other non-motorized transportation. Many research projects that attempt to evaluate urban form such as street patterns have not utilized the econometric frameworks that are the basis of travel demand modeling. The scope of this research includes both descriptive and inferential analysis in the attempt to assess differing street network types for their association with individual travel behaviour (and related outcomes). It builds on the extensive literature of urban form and travel behaviour research, reviewed in this chapter, to offer improved understanding of the nature of these relationships.  II. Urban Form-Travel Behaviour Research: Street networks and the field of urban form research The measurement and evaluation of street networks, a science emerging over the last hundred years, has generally been focused on motor vehicle movement and level of service for 20  motorized transport. The main concern in the consideration given to non-motorized transportation in North America during this time has been how to keep these travel modes safe by separating them from motor vehicles, a bias stemming from such works as Buchanan’s Traffic in Towns (cited in Marshall 2005). Accessibility and mobility of pedestrians, and to a lesser degree bicyclists, have come to the fore as key topics of recent research, prompted in part by the shift in thinking discussed in Chapter 1 and by recognition in public policy of the desirability of walkable environments. Perhaps the most significant aspect of street design that affects the re-balancing of transportation systems to encourage more walking is the connectivity of the street both internally and to nearby destinations. This section will review current knowledge on the urban-form travel behaviour before outlining the research on the connectivity dimension relevant to this research. Various research has descriptively (comparing different regions or other large areas that have distinct characteristics – Newman & Kenworthy 1989; Friedman, et al. 1994) and experimentally (among the earliest being Handy 1993; Holtzclaw 1994; Frank & Pivo 1995) investigated the relationship of urban form to travel behaviour. The evidence points to significant effects of land use density and mix as well as neighbourhood design characteristics, particularly transit (Cervero & Kockelman 1997) or pedestrian access (Holtzclaw 1994). Mix of uses and density have continued to be investigated in numerous ways during this period (Cervero & Duncan 2006; Frank, et al. 2005; McCormack, et al. 2001). These studies, many of which were conducted at broad spatial scales such as a metropolitan region, have argued persuasively that variation of three main factors - density of residences or employment, mix of land uses, and urban design (including transportation system layout) – at a regional scale relates to differences in travel behaviour. Efforts to understand the impacts of urban design at a smaller scale have also been undertaken. Some research has pointed to the importance of distinguishing between different travel purposes (work vs. non-work) and different spatial scales of trip-making (local vs. regional) (Handy 1993; Rajamani, et al. 2003), and has also recognized the importance of site design (Moudon, et al. 1997). Several studies have tried to distinguish which urban form covariates of travel behaviour, and walking in particular, matter most (Cervero & Duncan 2006; Frank, et al. 2006; Saelens, et al. 2003; Lee & Moudon 2006). These findings have come under scrutiny because of their policy implications for land use and urban design. This is part of the larger debate, discussed in Chapter 1, about New Urbanist design including the claim that its recommended practices will result in reduced automobile 21  travel demand. The criticisms have ranged from inadequacy of methods, i.e. not isolating one aspect of urban form from others to know where intervention is warranted (Crane & Crepeau 1998), to poor methodological robustness, i.e., lack of an adequate explanatory theoretical framework (Boarnet & Crane 2001). These studies point out the multiple confounding factors, including urban form and non-environmental variables, that can affect travel behaviour and possibly result in various prescriptions having the opposite of intended travel outcomes (Crane 1996). For instance, requiring new street connections that improve route directness or constructing a new street network as a gridiron could benefit (i.e., improve access for) automobile travel as much as pedestrian activity, resulting in no meaningful change of behaviour or even enabling more vehicle trips. Studies (Boarnet & Greenwald 1999) have produced results that contradict findings that street connectivity (measured in terms of proportion of streets that are grid-like in pattern) influences travel. Another criticism leveled at urban form-travel behaviour research is that it fails to account for residential self-selection (i.e., a person choosing to live where they can travel most easily in their preferred manner), another plausible explanation of relationship between urban form and travel (Krizek 2003). Another challenge to finding conclusive evidence for links between urban form and travel is that many urban form characteristics vary together as location changes. Many of the characteristics thought to be more supportive of non-motorized activity or transit use occur together in space. For example, in a former streetcar suburb the density is higher than in a newer neighbourhood, but very often the mix of uses and continuity of sidewalks is also high. Such spatial covariation (Frank, et al. 2003) confounds easy conclusions about any one feature of the built environment because the same travel behaviour in a given area may be due to any number of several contributing local, built environment factors. One variable may be masking another that is a stronger correlate with travel outcomes. Some of the researchers noted above have pointed out that variations of urban form are really just more contributors to the cost of travel, which should be regarded as the real direct influential factor in individuals’ travel decision making. Finally, while travel surveys have been in use for decades and the practice is wellestablished for collecting travel activity information, data on non-motorized travel have only been systematically collected more recently. There are problems of underreporting on short trips, which are more likely to be traveled by non-motorized modes (Frank & Engelke 2001). The  22  difficulty and expense of collecting very localized data on the built environment means that usually it is unavailable – another complication in attempting this research. Summary of methodological challenges Efforts to understand the relationship between urban form and travel behaviour have produced significant, though inconsistent (Crane 2000) results and continue to face major methodological challenges: 1) Travel is a complex behaviour, and the research to date has lacked a strong enough conceptual framework or theory of behaviour. 2) Attempts to explain travel behaviour are confounded by the multitude of factors affecting individuals’ travel decisions. Many of these factors co-vary. 3) Causality is not possible to demonstrate without longitudinal methods, which require extensive time and resources to carry out. The data available, in the absence of an extensive long-term research program, is mainly cross-sectional (Frank & Engelke 2001). 4) Data is difficult to obtain, especially the kind of data needed to demonstrate more than a simple correlation or to explain behaviour more completely. The second and fourth point deserve a bit more discussion here. There are many dimensions of the built environment (Ewing & Cervero 2001), along with variation in the quality of transportation services, that have been shown to influence human behaviour and especially to encourage pedestrian activity (Moudon & Lee 2003). Though not all of these variables will be relevant to distinguishing among street networks, the methods used to test relationships must adequately consider these other variables and better yet control for them. Demographic and socio-economic characteristics of individuals or households are known to have strong relationships with travel behaviour, and these likewise can confound simple conclusions about urban form and much less the influence of a single factor like street design. Data to distinguish between the networks for driving and walking modes is particularly difficult to find because although street networks are quite well catalogued and measured, sidewalks and related walking facilities generally are not. Overcoming this latter problem, and using a methodological framework that allows control for the many factors that influence travel behaviour while acknowledging the limitations of this kind of analysis, is of central importance to the success of this research. 23  Framework and methods to address challenges Behavioural framework/theory & pedestrian travel As noted above, a framework or theory that has ample explanatory power in relation to travel activity is needed in order to interpret results from built environment - travel behaviour analysis. The most robust model for explaining travel behaviour is a rational choice framework (Boarnet & Crane 2001) that posits travel demand derived from the need to get to and from activities and decisions affected primarily by costs (in time and money). In this model, these costs, or the travel impedance, are theorized to determine the extent of travel, and, as the individual weighs the relative utility of different ways to travel by comparing their costs, his or her choice of travel mode. The rational choice framework, as the basis for standard traffic modeling, is the most widely used means of explaining and predicting travel behaviour for transportation and land use planning. While it appears to be focused on individual decision making, this framework is not inconsistent with socio-ecological models (Giles-Corti 2006; Stokols 1992) that suggest people’s behaviour is shaped by factors at a number of levels – personal, environmental and cultural. People respond to factors in their environment as one of many ‘costs’ that affect their travel decisions. This view of human activity, driven by utilitarian decision-making in a context of numerous cost factors ‘imposed’ by the environment and society, forms the basic theoretical framework for this research. Walking activity in particular has been shown to be influenced consistently by distance to destinations and the quality of the built environment (Moudon, et al. 1997; Frank & Engelke 2001; Berke, et al. 2007). Distance, then, becomes a main cost in this analysis. Walkability, the quality of an environment in terms of its supportiveness for walking, has been summarized in indices that attempt to capture the key dimensions of density, diversity and design. These factors attempt to encompass the whole of urban form, but the interest of this study is the effects of street configuration, and in order to capture streets’ associations with travel behaviour, this study must address some of the existing gaps in methods for measuring street networks. Gaps in Measuring the Urban Form – Street Networks The urban form measurement for this assessment is in the design realm (rather than land use intensity or mix), specifically transportation system design, and this analysis must be able to distinguish among street networks to provide a spectrum of possible network types which can 24  then be tested and compared. The chosen method should build from the experience of a number of studies that have attempted to assess the influence of street networks (Cervero & Kockelman 1997; Frank, et al. 2000; Pushkar, et al. 2000, noted in Ewing & Cervero 2001). The other influential urban form variables and design features should be accounted and controlled for as much as possible. Despite decades of concern about the effects of the transportation system on travel behaviour, and which if any interventions will support desirable transportation outcomes contributing to overall quality of life, appropriate solutions are still being sought. The relationship between various aspects of urban form, and more specifically street networks, and those travel outcomes is poorly understood or has undergone limited empirical investigation (Ewing & Cervero 2001). Furthermore, most research that has included consideration of street networks has not distinguished between motorized and non-motorized modes, though the networks for different modes may have varying degrees of connectivity (Handy, et al. 2003; Dill 2004). Frank, et al. (2000) found that street network, measured using census block density, has a significant association, distinct from other land use variables, on vehicle miles traveled and also modeled air pollutant emissions. Kitamura, et al. (1997) found that the presence of nonmotorized travel facilities (i.e. sidewalks, bike lanes) related to the frequency of travel by these modes. However, these studies did not focus on the relationship of street network design and mode choice as a travel behaviour. Moudon, et al. (1997) tested a variety of network variables (block size, sidewalk and street length, sidewalk completeness, and route directness) in relation to levels of walking in particular districts and found that none of them alone seemed to explain the variation in pedestrian trips. Later research using a different proxy measure (intersection density) has found that the street network alone did not have a significant influence (Lee & Moudon 2006) or that it was so highly correlated with other components of walkable urban form that a combined walkability index became the necessary measure to be tested against outcomes (Frank, et al. 2005). This study attempts to contrast the modal connectivities of street networks in order to answer the question of how the greater pedestrian connectivity of a modified version of the pure gridiron street network will function to shape travel patterns. Because the unit of analysis is household travel behaviour, the appropriate scale of study is the sub-neighbourhood level – urban form characteristics in the immediate vicinity of residences. The travel data to which the 25  urban form is being related should be local in nature as this is the kind of behavour likely to be influenced by the home neighbourhood. Additional research could be needed to build on this work and answer questions of how residential street networks might influence behaviour at different (regional versus local) spatial scales or to model transportation system function as a whole. Methods for Measuring Street Configurations Street network connectivity has been measured in a number of ways in the past two decades of urban form-travel behaviour research. Below is a listing and brief discussion of these methods. •  Number of street intersections, dead-ends or 4-way intersections (Cervero 1994; Ewing 1996) These are all early proxies for street network connectivity, used primarily in indicating whether pedestrians can easily reach transit stops or stations.  •  Block size, census block density (Frank, et al. 2000) measures capture the spatial relationships, or topology, of streets to buildings, and are again proxies for network connectivity.  •  Intersection density (Frank, et al. 2005; Schlossberg & Brown 2004) – More recently this measure has come into use for its ease of computation, and has been included as a component of walkability indices.  •  Another measure, coming into wider use by local governments, is the link-to-node ratio (Handy, et al. 2003), which like intersection density is focused just on the network itself and is an abstract measure that offers an indication of the number of dead-end segments (an increase of which makes the ratio lower, indicating poorer performance). Percentage of cul-de-sac streets has been used (Rajamani, et al. 2003), in combination with a link-tonode connectivity measure, and found to be associated with non-work travel. However, these methods are not able to capture cross-modal connectivity well particularly since distance along routes, a highly important consideration for non-motorized travel, is not specified.  •  Route Directness (Dill 2004; Gauthier 1999; Moudon, et al. 1997) Non-motorized modes of travel (bicycling and walking) are highly sensitive to distance, and thus a most appropriate way to measure the built environment vis-à-vis pedestrian movement is route directness. This measure also accounts to some extent for the continuity of the routes 26  because route directness is a measure of distance traveled to reach a destination, not the more abstracted spatial measures noted in the foregoing list of methods. •  Various network indices (Dill 2004; Schlossberg 2006; Tresridder 2005). The most recent research on bicycle and pedestrian networks has sought to measure streets using connectivity indices borrowed from geographic analysis or to introduce some degree of distinct measurement for the mode being analyzed (pedestrian catchment or effective walking areas). Space syntax modeling methods (Hillier, et al. 1993; Penn, et al. 1998) are coming to the  fore of methods to estimate pedestrian volumes and route choice on various travel networks, and are rooted in algorithms of route directness. These methods show how transportation networks co-determine, along with the activity patterns that they influence, the overall shape of urban space. While the first four sets of measures give an indication of the ‘grain’ of the street network they fall more into a class of variables that could be termed indicators. Each of these measures, while offering quick assessments of connectivity, are too coarse in their discernment of connectivity to capture differences in system function across modes unless each mode’s network is measured separately. Because they are street or block based, they may in fact miss much of the network that is available to pedestrians (Zhang & Yi 2006). Various other indices have been developed, including a Pedestrian Environment Factor comprised of crossing quality, continuity of sidewalks, street connectivity and topography (1000 Friends of Oregon 1993). While these are more complete and nuanced measures for assessing walkability, they are usually composite measures which, when integrated into a statistical model, pose challenges for interpretation of just one of the constituent elements. Also, a consistent measure that applies to both non-motorized and vehicular modes of travel is needed for this research, and vehicles, because they are not as sensitive to distance, may be more influenced by the continuity of routing than by distance per se. The most promising measure, because of its simulation of actual travel activity, is route directness, which expresses connectivity as a ratio of the distance one would have to travel on the street or travel network to the air-line (or crow-fly) distance. Space syntax, an extension of route directness measurement, is powerful for predicting pedestrian volumes and associated phenomena (Raford & Ragland 2005); however, this study of travel networks required a simple measurement of network connectivity across different modes rather than a means of predicting 27  pedestrians volumes. Route directness can be complemented with an area measure of facility extents (network density), in order to gauge both connectivity and continuity. If conducted for each mode’s available infrastructure, this second network measurement, similar to a network service area measurement such as effective walking area, allows the research to achieve a more nuanced measurement of the disparities of travel networks offered to each mode by different street configurations. Statistical analysis: Using quasi-experimental methods A growing volume of research has been using descriptive and inferential statistical methods (Cervero & Kockelman 1997; Frank, et al. 2000; Frank & Pivo 1995; Handy 1993; Rajamani, et al. 2003) to characterize and assess urban form - travel behaviour relationships. These methods allow the isolation of urban form variables of interest from other environmental factors and from demographic characteristics, provided that there is sufficient data available about these other factors. Strength of hypothetical relationships between urban form and travel behaviour, though not direction of influence, can be tested using case-control quasi-experimental methods (Frank et al. 2003). These methods have the advantage of using existing data sources. A newly built Fused Grid neighbourhood, or an existing neighbourhood that matches its characteristics, could be used for a study of the street network’s influence on travel behaviour over time, but this experimental method would require gathering extensive new data over a long time period, making it infeasible for the current project. The key for this analysis is to distinguish between each network’s connectivity by mode, in order to quantify the different street patterns in a way that relates to differences in the relative utility of the different travel modes. This quantity can be measured in a way that links it to the area where the travel behaviour is being reported through a process of spatial matching as has been conducted in the studies noted above. Statistical techniques, namely correlation and multivariate regression analysis, are used to analyze and isolate the influence of particular factors while controlling for other urban form variables and individuals’ sociodemographic characteristics. Care must be taken in specifying and processing variables in order to get valid analytical results. Chapter 3 provides further details on this methodology.  28  Types of Data Individual Household Behaviour Data Data from travel diaries, self-reported by individuals, offered the best opportunity to capture the actual travel behaviour of individuals in various residential locations for this research. The following chapters elaborate on the development of the particular dataset being used in this study. The data were taken within the same region, allowing this research to avoid possible geographic variations of travel behaviour among regions. Neighbourhood scale spatial information is crucial for distinguishing among different residential street patterns. Using the neighbourhood-scale analysis frame will allow very localized measurement of urban form variables and the eventual linking of individual’s reported travel behaviour in their local walking range and within what are the normal, largerscale aggregated zones of analysis in transport planning (Transportation Analysis Zones; see Chapter 3 below and Hess, Moudon & Logsdon 2001; Krizek 2003). This helps guard against the errors that stem from variation of street networks within zones and the edge effects from an adjacent zone on some nearby households (Frank & Engelke 2001). The characterizing of street networks is a complex undertaking, but recent work (Marshall 2005) provides the most systematic analysis to date for developing a typology of street patterns. The concept of street network connectivity in this present study encompasses dimensions that Marshall describes as connectivity and continuity. Connectivity is the extent to which pathways intersect at regular intervals, allowing for ease of movement (or connection) to desired destinations; continuity refers to whether a pathway is continuous, i.e. remains uninterrupted along its length. This project assumes that in order for a street network to be considered connected, routes through it must be continuous. Using the measure of route directness captures both connectivity and continuity of street networks as urban form because it is based on movement along the segments of path (both pedestrian and vehicular) that make up the network. Due to discontinuities typical of pedestrian networks (for example, each crossing of a street is a potential discontinuity, or interruption, of the pedestrian’s network of sidewalks and trails), assumptions are necessary in order to measure and compute route directness. A separate measurement of the networks’ distinct lengths within neighbourhood areas is also used as a complementary (though still a proxy) measure of continuity and a means of better specifying the continuity of the networks for each mode. By including consideration and separate measures of the distinct networks available 29  to two modes, walking and driving, the research can more effectively assess the effects on mode choice and travel behaviour more broadly.  III. Strengths and Limitations of Method and Project Scope The following lists some areas of strength and weakness in the present research, as well as its scope. The project can be seen as a discrete piece of the ongoing research effort to understand the connections between urban form and travel behaviour, especially active forms of travel such as walking. As with any research of limited duration, this project cannot be considered comprehensive or definitive. It seeks to add to our understanding in hopes that other research will continue to build on its findings. Strengths 1. Scale: individual level and neighbourhood scale, allows for greater power to model travel behaviour by individuals in response to their local street environments 2. Data and methods: control for a wide array of socio-demographic and urban form variables Limitations 3. Cross-sectional: description and strength of relationship not causation 4. Environmental determinism and self-selection Other Dimensions of Street Networks There are several other aspects of street network design that may be influential to how people travel and make other activity choices. This study does not investigate all of these directly but instead views then as correlates of connectivity. Defensible space is one of these aspects. Crime and personal safety factors can be influential to people’s decisions about how to more or travel, and the prevention of crime can be affected by the design of environments. This study did not include variables relating to public safety. As the degree of connectedness increases, more use of an area by people is likely to occur and natural surveillance, the perceived and actual safety of the environment (Newman 1972), would be expected to increase. In fact providing a connected street system (avoiding long blocks) is one of the recommendations in crime prevention through environment design practice (Newman 1996). 30  Real or perceived traffic safety risk can have a strong effect on behaviour. While studies have linked safety risk to speed of vehicles, and speed of vehicles to widths of roadways (Fitzpatrick & Carlson 2001, cited in King 2003; Leaf & Preusser 1999), and then related these concerns to different intersection types, only very few (Kulash 1991; Lovegrove & Sayed 2006) have attempted to investigate the link of street network patterns to traffic safety. This study will not delve into the safety question except to acknowledge that it deserves further research, in particular to evaluate the differences between separated and conjoint networks. If, as hypothesized, vehicular traffic is moderated by the street network, higher connectivity for this mode would result in greater disruptive impact to public space/neighborliness (Appleyard 1981), whereas lower vehicular connectivity, or design favoring non-motorized travel, would reduce vehicle use and concomitant impacts. Again, social impacts of street patterns were not assessed directly by this research, but do deserve further investigation. Finally, Marshall (2005) identifies a number of different dimensions of street networks in helping to decompose the problem of understanding how street networks function and how they can best be designed. Legibility, how easily one discerns routes in the pattern of street connections, may be an important qualitative factor of the built environment; the route directness measure of this assessment should capture some of this phenomenon.  Likewise, the space  syntax method noted above (see section II.) has been utilized to examine the relationship of street pattern and traffic safety (Hillier et al. 1993). These analysts have demonstrated that controlling urban design alone, especially the configuration of street networks and their width, can be used to manage better the interactions among modes of travel in urban environments and lead to more suitable overall urban form to achieve a variety of public policy or planning objectives.  IV. Conclusion Notwithstanding the challenges and limitations noted here and above, this research sets out to describe and test the relationship between the residential street connectivity and local travel behaviour. Municipalities should be aware of the potential for street network design to assist in attaining various transportation and livability objectives. Describing how street network configurations relate to travel behaviour is the goal of this research, and this review has 31  discussed relevant methods for analyzing the Fused Grid and other street network configurations. Measuring connectivity using pedestrian route directness, along with other metrics of network continuity, allows the study to capture differences in street network function across distinct modes of travel. These urban form characteristics can then be matched to individual travel behaviour and statistically analyzed. Finally, results from the descriptive and inferential analysis of street networks provide an assessment of the likely performance of the Fused Grid street design on several travel outcomes of interest to community development and urban transportation planning.  32  Chapter 3. METHODS  33  I. Introduction to Proposed Methodological Framework This research uses a quasi-experimental approach situated within an activity-based, rational choice framework (Boarnet & Crane 2001) to understand the travel implications of street patterns. The research statistically relates travel activity to urban form measured, using Geographic Information System (GIS), around each location (household) self-reporting travel in a two-day survey. Such methods have been used increasingly in recent years to develop better understandings of the interaction between urban form and travel behaviour and other outcomes. The methodological framework considers travel to be a derived demand; that is, people’s travel is driven by their desire or need to access a variety of activities in the urban area around them. It is an approach consistent with standard transportation planning – trip origins and destinations are identified and then travel costs (in terms of money and time) are assessed on various routes and across various modes in order to model travel behaviour (number of trips, distance traveled, mode choice, etc.; see Figure 1-1). People make decisions and act on the basis of routes and travel options that offer the greatest utility – lowest costs and highest benefits – in the process of accessing desired activities. The Fused Grid assessment responds to an identified need for empirical testing pointed out by studies that have described and conceptualized the importance of street network design (Ewing & Cervero 2001; Handy, et al. 2003; Moudon & Untermann 1987; Southworth & BenJoseph 1997) and by those that have proposed interventions to improve its performance (Appleyard 1981; Ewing, et al. 1996; Girling & Kellett 2005; Grammenos et al. 2005; Handy et al. 2003; Randall & Baetz 2001; Untermann 1984). It builds on research that has measured street design, and network connectivity in particular, as an urban form feature having the potential to affect travel behaviour (Cervero & Kockelman 1997; Dill 2004; Moudon, et al. 1997; Rajamani, et al. 2003) and related health or environmental outcomes (Frank, et al. 2000; Frank, et al. 2004; Frumkin, et al. 2004; Lee & Moudon 2006; Saelens et al., 2003). In assessing the Fused Grid and other street networks for their comparative travel and quality of life performance, the characteristic circuitry, or how connected and continuous are the various modal pathways provided by streets, is the distinguishing feature that is hypothesized to be influential to behaviour because of its relationship to distance to be traveled. The Fused Grid provides improved circulation for pedestrians relative to that for motorized transportation, thereby hypothetically shifting travel mode choice toward walking and away from driving. 34  Testing the Relationship - Street Networks & Travel The empirical approach of spatially matching and statistically relating the measured built environment to reported travel behaviour, used in several of the studies noted above, is the analytical method used in this research. Because of the interest in understanding street networks, this project focuses on isolating the effects of street connectivity and continuity as explanatory factors in travel behaviour. Though it is one of the primary design features of streets, network connectivity, especially the relationship of its differential supportiveness for various transport modes with travel behaviour, has not been adequately empirically tested using this experimental approach to date. Until now, empirical analysis of travel outcomes has not contrasted the levels of connectivity for the distinct pedestrian and vehicular modal networks. In previous analyses of urban form’s influence on travel behaviour, streets have been treated as single networks and their patterns measured using a variety of coarser methods. Some of the previous means of measuring connectivity are shown in Table 3-1, adapted from Southworth & Ben-Joseph (1997). Other measures of connectivity, particularly related to nonmotorized modes like bicycling and walking, include variations on the ones shown below (block size, link-to-node ratio, and various geographic indices), effective walking area, and pedestrian route directness (Dill 2004; Tresridder 2005). In this study, streets are considered to be a bundle of largely overlapping yet distinct modal networks, or separate systems of interconnecting lines in a GIS file. This research attempts to disaggregate the network patterns for vehicle and walking modes, measure them, and then compute ratios between the separate networks to reveal patterns that are either equivalent or disparate. Since a disparity in connectivity that favors walking over driving is precisely a quality that distinguishes the Fused Grid from many other network designs, and can also be the effect of traffic calming retrofits to existing neighbourhood streets, this is both a theoretical and immediately practical question. Table 3-1. Street Network Types, adapted from Southworth & Ben-Joseph 1997. Street Pattern Types Gridiron (c. 1900-1920) Fragmented Parallel (c. 1950) Warped Parallel (c. 1960) Loop and Cul-de-Sac (c. 1970)  Lineal Feet Of Streets  # of Blocks  # of Intersections  # of Access Points  # of Loops & Culs-de-Sac  20,800  28  26  19  0  19,000  19  22  10  1  16,500  14  14  7  2  15,300  12  12  6  8  35  Figure 3-1. Households, buffers, and networks. Seattle region  Seattle (top) and Bellevue (below) sample households (same scale). The larger, faint circle in each image is a 1km airline buffer whereas the ‘diamond’ and other oblong bright polygon are 1km network buffers. Households are shown as house symbols and the lines are various travel networks. Orthophotos courtesy of MDA Corporation.  To understand how variation in travel behaviour relates to changes in these ratios of street connectivity and continuity across modes, this study analyzes travel data with correlation and regression statistics. Using cross-sectional data, the statistical significance and strength of relationships between measured built environment and self-reported travel behaviour data can be assessed. Existing datasets (see below and the next chapter on data development) from the same 36  region enabled control for geographic variability, allowed keeping the data collection and processing manageable for this assessment project, and prompted a cross-sectional method (see also section V. “Findings & Limitations”). Due in large part to the efforts of previous studies of this same region, the datasets also include, and thus allowed this analysis to control for, other influential urban form data and sociodemographic information (Frank, et al. 2006) relating to the households. The strongest reason for the choice of the particular region of this study and its travel data, however, was the availability of detailed GIS data on pedestrian networks (including off-street connector pathways and trails) for three cities that had varied street patterns. Units of Analysis The unit of analysis was individual trip-making behaviour by persons in the households located within the study area (see Fig. 3-2). It was analyzed in an effort to identify likely outcomes from street configuration in terms of total trip length (distance walked or vehicle miles traveled) and frequency (number of trips) as well as mode choice (or the share of travel by walking or driving modes). Urban form characteristics were measured at the parcel level, using a GIS to relate these characteristics spatially (features within walking distance buffers -1km on the street network) to the exact locations of the households participating in the travel survey. Sampling The research first identified travel survey households within the cities for which good pedestrian network data was available. The three cities that constituted the study area are Bellevue, Redmond and Seattle in the State of Washington. Figure 3-2 below approximately depicts the 1500 households (green points), overlaid on the street networks located in these cities. The other urban form measures had already been computed for all of these households. The 1km buffers (purple polygon shapes), within which the urban form was measured, are also shown.  37  Figure 3-2. Seattle Region Study Area, travel survey households and networks  To describe and analyze effects of street network designs (independent, or predictor, variable) on personal travel activity and choices (dependent, or response, variable), a dataset of residential locations clustered into neighbourhoods that each represent particular street network types and have adequate sample size (at least 30-40) would be ideal. The available dataset, however, has scattered data points that are not sufficiently concentrated in any one neighbourhood to allow such a clustered sampling approach. Therefore this assessment focused on measuring and modeling a continuum of street network connectivities across all the households in the study region, which, as will be described in the first part of the results, correspond to a range of types of street pattern. The research was focused on the travel that was most likely to be influenced by the home neighbourhood – that is to say, home-based travel. To analyze all travel would have included trips taken in other cities or while away from home, and those where the travel choices are constrained (long commutes across the metropolitan region). The measurement of the predictor and response variables is described 38  next. The full dataset of travel was therefore screened to include only those trips that were home-based. The sample was also screened for outlying cases in terms of walk or driving distance, household characteristics, and all urban form variables.  II. Data – Sources & Collection Data Sources The primary data source used in this study is the 1999 Travel and Activity Survey conducted by Puget Sound Regional Council (NuStats 1999), August to November 1999. This same dataset was used in previous urban form travel behaviour research for the Seattle, WA metropolitan region, as noted in Frank et al., 2006, under the Land Use, Transportation, Air Quality, and Health (LUTAQH 2005) and WSDOT (2005) studies. The survey process included a two-day travel diary completed by individuals and a pre-diary questionnaire about household characteristics, thereby gathering information at the level of trips, individuals and households on their travel mode used, start and end time of the trip and the activity, origin and destination locations, and individual and household socio-demographic information (see Table 3-2). This study has used a distinct subset of the larger 4-county survey data, since the needed street network data was only available from the three city area shown in Figure 3-2 above. King County, three major local governments within it (the cities of Seattle, Bellevue and Redmond), and the LUTAQH/WSDOT studies are the sources of the GIS and other data on built environment for the analysis. This existing dataset, then, provided the basis for key outcome (dependent) variables as well as some of the needed control variables (eg., demographics. The other explanatory variable data was generated by the studies noted above and in Table 3-2, whereas the variables of interest were newly created using GIS and the region’s GIS data. Defining travel behaviour (dependent) variables Number of trips, travel distance (walking or vehicle miles of travel), mode choice, and amount of physically active transportation are the travel behaviour outcomes of interest in this research. This data is derived from self-reported travel survey responses, and processing of the data will select those trips most likely to be affected by the quality of the home neighbourhood’s urban form (home-based travel). Number of trips and mode choice have been used in combination with route measurements (i.e. shortest-path network distances between reported 39  origins and destinations) and reported travel times to model other outcomes of interest (air pollutant emissions, for example). These latter variables have been computed through the previous work of the LUTAQH and WSDOT projects by LFC. Table 3-2. Possible Variables for Street Network Experimental Analysis POTENTIAL VARIABLES BY TYPE  Network/ Urban Form Characteristics Source: GIS layers from Seattle metro region (King County; cites of Bellevue, Redmond and Seattle) street centerline data pedestrian networks parcel level land use (computed mix of uses; residential density) computed shortest path on separate networks from each household to nearest park and commercial destination total sidewalk length versus total street length within 1km buffer around households  Travel and Activity Patterns Source: PSRC Travel & Activity and NQLS surveys; computations from LUTAQH study # of trips (aggregated to household) trip origins and destinations  Vehicle Emissions  Household Characteristics & Demographics  Source: Modeled per sub- Source: PSRC trip link as part of Travel & Activity LUTAQH study Survey grams of Oxides of nitrogen / person household size grams of Carbon number of Monoxide / person vehicles  grams of Volatile Organic age computed # trips by mode Compounds / person  computed percent trips by grams of Carbon Dioxide mode / person income computed trip distance by mode  gender  Respondents to the survey were asked to provide relevant socio-demographic information, including household characteristics such as number of persons in the household and total vehicle ownership. The travel survey data includes relevant person-level data on age, income, gender, ethnicity, and educational attainment. These personal characteristics are also known to be influential to travel behaviour, and so having this data will allow control for these potentially confounding factors in subsequent analysis. Urban Form: measuring connectivity across modes (independent variable) To test the relationship of street network configuration to travel behaviour, this research required measurement of street networks that captured significant differences among various street pattern types likely to affect behaviour. As discussed above there are several different methods available to measure the connectivity of street networks. The connectivity measure for the assessment of a continuum of network configurations was selected based on its capability of distinguishing carefully between connectivity among different modes of travel (specifically, pedestrian vs. vehicular) and being defensible in terms of how it relates to the travel activity of the persons in the survey. More general measures such as block size, length or density, or 40  intersection density on the street network, though widely used, will not be adequate because they do not offer an indication of network differences among modes (i.e. presence of continuous, connected sidewalk routes). This is noted as a challenge by recent research on walkability (Schlossberg & Brown 2004): “Developing accurate data sets of sidewalks would enhance the walkability analysis by using actual walking paths as a primary data set, rather than using the street network as a proxy. However, the development of such layers is not without difficulty. For example, sidewalk layers often do not cross streets, making it difficult to model distance traveled along the network.” p. 41.  Pedestrian connectivity While the measurement of connectivity for bicyclists and pedestrians is a topic of ongoing research (Dill 2004; Schlossberg 2006), and municipalities have various means of assessing the connectivity of their networks (Handy et al. 2003), this research required measures which provide adequate indication of distance and directness of pedestrian routing as well as the continuity of the walking mode’s network, relative to equivalent measurements of the vehicular network. This entailed measurement at the finest scale possible – presence of sidewalk on street segments at the sub-block level. Distance is the ‘travel cost’ characteristic of networks that was captured and contrasted across modes in these measurements. The measurement method that offers one of the best estimations of connectivity for nonmotorized travel in this framework is pedestrian route directness (Moudon, et al. 1997; Gauthier 1999; Engelke, et al. 2000), in which a ratio of the network distance to the most direct, airline routing (“crow-fly”) is developed from each traveler’s trip origin (in this case, home). Nonmotorized modes are particularly sensitive to travel distance (Moudon, et al. 1997). Such a measurement further requires that a destination as well as an origin is specified for each household as an indicator of proximity to attractions. The research on travel behaviour commonly uses the shortest path distance from the home to various neighbourhood destinations. Among the most important features that attract trips in the local area around households are commercial (shopping, services, etc.) and recreational (parks, trailheads, community/recreation centers, etc.) destinations (Hurvitz 2005). This research therefore prepared two distinct GISbased origin-destination pairs – one from household to nearest commercial destination and one to nearest park – to develop the route directness measurements. As a first step, route directness for walking and driving was measured on each mode’s respective network but using the same denominator of crow-fly distance (i.e. the same origin41  destination pairs; see Equation 1 below). Because this measurement is based on the presence of a route for pedestrians (sidewalk) and considers both continuity and travel distance aspects of the network, it offers a more comprehensive indication of connectivity. However, it is also more computationally complex and demanding of accurate GIS data, and entails certain assumptions (see Chapter 4). In order to identify the range of pedestrian and vehicular connectivities, the research also assesses the contrasts in modal networks’ continuity more directly (i.e. relative continuousness of non-motorized versus motorized networks). Therefore, a second variable, again shown to be an important correlate with walking behaviour (Lee & Moudon 2006), and giving an indication of the relative completeness of the pedestrian infrastructure in the neighbourhoods around residences in the study area, was also computed: measurement of each mode’s respective total network length that allows estimation of the ratio of pedestrian network (total length of walking paths and sidewalks) to the overall street network (total length of all street segments) within a 1km buffer of the study’s households (for justification of this radius of activity, see below). Vehicular connectivity Connectivity of networks for motor vehicles is better understood and more easily measured than that for non-motorized travel. Street network data is readily available from most local and regional planning or public works departments. Though, as street design research points out (Marshall 2005), motorized modes can overcome route indirectness and lack of connectivity more easily than non-motorized modes, the distance that must be traveled, at least for local trips, is still a ‘cost’ factor in deciding whether to drive or not. This study used the same variables, route directness to nearest destinations and network segment length within the household buffer, measured for motorized vehicles in order to compare and contrast like quantities when creating the ratio variables. Pedestrian-to-Vehicular Route Directness Ratio Using the same connectivity measure for each mode, the study developed a ratio between the distinct modes’ network connectivities, dividing the walking route directness (PRD) by the motor vehicular (or on-street) route directness (VRD).  42  Equation 3-1. Ratio of Route Directness =  PRD DPedNet = VRD DCrowFly  DVehNet D = PedNet DCrowFly DVehNet  Where D = directness, P = pedestrian, R = route, and V = vehicle  The relative network continuities are similarly expressed in a ratio: Equation 3-2. Ratio of Network Density = total sidewalk (and trail) length in buffer / total street length in buffer These ratio variables express the pure relationship of the two networks’ connectivities and continuities. For example, in the ratio of route directness the proximity factor of crow-fly distance is cancelled out since it is the same origin-destination pair for both modes’ route directness measure. Likewise, for continuity, the quantities are comparable across households, despite buffer areas that differ in size due to geographic or other factors. Proximity to destinations as a factor in travel activity is captured by including crow-fly distance and other urban form variables in the final models (see below). For further details on the method for conducting this connectivity measurement and the process of developing the travel survey data, as well as the steps in conducting the analysis, please see Chapter 4 on database development. Other urban form variables The urban form data available from this metropolitan area included residential density, land use mix, and other influential factors of the built environment. A walkability index composed of mix and density of land use as well as street connectivity has already been constructed for the King County (LUTAQH) study area (Frank, et al. 2006). These measures have been computed for the 1km buffer around each household, a scale which avoids problems of relying on standard, coarser measurement frames (Hess, et al. 2001). This distance (1km = 0.6 miles) has been established by previous studies (Frank, et al, 2004; Moudon et al., 1997; Lee & Moudon 2006) as a walkable distance (10 to 20 minutes) from each household. It constitutes a smaller subunit of the area that is considered to be a home neighbourhood. The component urban form variables of this index, as well as others developed by this previous research, are measured in a consistent and comparable manner and thus can be controlled for in a model which isolates connectivity on behaviour. 43  III. Sample Selection and Development of Dataset The first task of this research was to develop a statistically valid sample on which to conduct analysis. The primary existing travel behaviour and urban form data sources available for this research were the PSRC Travel and Activity Survey (encompassing King, Pierce and Snohomish counties of Central Puget Sound, Washington State) and the LUTAQH (2005) study of King County, WA. However, full sets of GIS data on pedestrian networks (sidewalks and paths) were only available for the cities of Seattle, Bellevue, and Redmond. Therefore the sample of households consisted of only those respondents (approximately 1500 households) from these cities where network connectivity measurement, as discussed above, for the home neighbourhoods could be conducted. This research used 1km network buffers to serve as the measurement frame to approximate a neighbourhood’s urban form. This is one approach among many (Krizek 2003)4, but has been shown to be a highly accurate way to measure the local environment around a person’s residence (Frank, et al. 2005). In an effort to approximate better the overall neighbourhood street connectivity, this research also tested an aggregated measurement of the ratio of route directness in addition to the household-level variable. A description of the process for developing this variable is included in Chapter 4. However, other scales of measurement were not attempted for street connectivity, nor for other variables, though regional factors can show dominant associations with travel (Boarnet & Greenwald 1999). From this geographically-defined sample, the sample was further defined to include only those trips that were home-based in nature and occurred within the local neighbourhood. These local trips were those that occurred within 2.5 miles of the households reporting the travel. Next, only the trips that would truly have been a choice between driving and walking were included,  4  In the process of defining ‘neighbourhood’ in order to investigate how accessible it is – i.e. what sorts of travel it supports or encourages – specifying the units of analysis becomes a critical question. The phenomena that are the focus of this research are the street network pattern of a neighbourhood and travel behaviour of its residents, but ‘neighbourhood’ can be investigated in several ways – conventional boundaries as described in community planning (which may have historical or cultural as well as geographic or design origins), planimetric distinctions (i.e. the pattern of streets themselves when viewed orthogonally) or other differences derived from street pattern measurement, or other sampling frames established in past research (using census tracts, traffic analysis zones, or as in more recent literature, buffers or a smaller scale around each household). Some variables may be more sensitive to the choice of local area or scale of the analysis.  44  which meant excluding trips under 0.1 miles in length since these were likely to have been shorter than a distance people would reasonably drive. While the travel behaviour of persons is of central importance to the study, the analysis of trip-level data is also important because the quantity of trips generated from a particular area is a primary driver of development requirements vis-à-vis transportation infrastructure. The study thus analyzed both the trip-level sample as well as the aggregation of these trips to the personlevel sample. Finally, the samples were screened for outliers on all of the possible variables, dependent and independent. Those cases that were more than three standard deviations (> 3 sd’s) from the mean were excluded, as was done in previous research using the same data. This way of finalizing the sample is recommended in order to enhance the accuracy of estimates in the inferential analysis (Osborne & Overbay 2004). For example, if a trip was an outlier in terms of distance traveled by walking, or if a household had so many vehicles or a very high value on residential density, these cases were removed from the sample along with the persons who were associated with them.  IV. Analysis The goal of the analysis phase was to generate improved understanding of the relationship of street network design (connectivity) with patterns of travel behaviour through quantitative analysis. The computed ratio of route directness of the walking and driving travel modes will indicate the disparity in connectivity between the modes. This, and the comparative modal network density (continuity) measure were the continuous, independent variables of interest for the analysis. After this data preparation was complete, the analysis then proceeded through two stages, descriptive and inferential, methods which are discussed more below. More detailed description of the procedures for preparing data and analysis is included in Chapter 4 and technical appendices. As noted above, an approach involving descriptive and inferential statistics to explore associations and investigate hypotheses has yielded significant results in previous research of travel behaviour and urban form.  45  Descriptive assessment SPSS statistical software was used to conduct descriptive statistics on all cases and across all variables of the sample in order to organize the data better, explore relationships and variability in the data, and to diagnose potential problems or define direction for the inferential phase. Measures of central tendency (mean, median and range) and dispersion (standard deviation), along with frequencies, for the dependent and independent variables were generated. Second Level of Analysis – Inferences from Correlations and Regression Modeling One goal is to probe the hypothesis that the Fused Grid street network, with disparate connectivities (high for pedestrian (Ped) travel; low for motor vehicle (MV) travel) results in a shift of travel to more walking (and less driving) relative to other street networks characterized by more uniform (both modes) high or low connectivity, or networks where there is higher connectivity for vehicles than pedestrians. The research utilized multivariate statistical methods of analysis (Babbie 2003; Mertler & Vannatta 2005), with a procedure of regression analysis, to test the hypothetical relationship between the connectivity/continuity variables and the travel behaviour outcomes. A first step consisted of correlation analysis to gather preliminary indications of the significance/strength of relationships and the direction of influence (positive or negative) among variables. The outputs from bivariate correlations can help to show that an association exists, but not causality. The regression model for relating travel behaviour outcome variables to the two connectivity measures controlled for other urban form variables (mixed use, density) and other possible confounding factors (socio-demographic characteristics). The procedure followed was to enter socio-demographic covariates (the most influential, based on the previous research in this region, likely being vehicle ownership and household income) as a first block and then to add the other urban form variables as a second block to the model. Finally, the connectivity disparity (ped.-to-veh. route directness ratio), network continuity, and crow-fly distance variables were added, to ascertain whether there was significance in the outcomes’ relationship to these new variables, the magnitude of the association, and whether these variables added explanatory power as revealed by the significance at each step and changes in the coefficient of determination (r2) produced by adding the variables of interest (network connectivity and continuity). This group-wise regression was repeated for several travel outcome variables. 46  Because the mode share variable (whether as a proportion or as a dummy variable of whether a person chose a mode such as walking) is not continuous, a logistic regression technique was the primary analysis method. A linear regression was used to investigate relationships with continuous travel outcome variables such as total trips or distance traveled with a given mode of transportation. Relating street network configuration types to analytical findings The connectivity measures were used in tandem, along with previous measures such as intersection density, to recognize types of street networks within study area neighbourhoods (defined by where travel survey data is available). Type I: High Vehicle, Low Pedestrian Type II: High Vehicle, High Ped Type III: Low Vehicle, High Ped Type IV: Low Vehicle, Low Ped  Figure 3-3. Street Networks: differences in connectivity across modes  Descriptions of these different types of street network were developed, making use of the most current analytical description of street network characteristics (Southworth & Ben-Joseph 47  1997; Marshall 2005), for use in interpolating relationships between each distinct type and travel behaviour outcomes from the regression results. Table 3-3 shows where various street network types were represented geographically in the study area. Table 3-3. Example locations of four main network types in Seattle region database Type  Location  Street Network Classification  I. High veh,  (Bellevue, east; Seattle NE; Redmond edges) streets with discontinuous or absent sidewalks  Loop and cul-de-sac, warped parallel or modified grid  (Seattle, various – grid street neighbourhood, no diverters, few or no off-street pedestrian facilities)  Classic Gridiron  III. Low veh, high ped  (Seattle - Capitol Hill, Queen Anne) – trails and connector paths provide links across grid interruptions  Modified gridiron; proxy of Fused Grid  IV. Low veh, low ped  (Bellevue, east)  Superblock (longer major street segments), loop & cul-de-sac  low ped II. Hi veh, hi ped  Gridiron residential streets are well represented in the older core of Seattle, and the other end of the spectrum of street network types (large block, dendritic subdivisions often lacking sidewalks even on collector streets) can be found in neighbourhoods built since the mid-20th century. The data indicate where in the ranges of the variables these different street patterns lie. The Fused Grid street pattern was also measured, based on available designs, for these same connectivity parameters. This allowed comparison of Fused Grid’s particular pattern of network connectivity and continuity with the kinds of patterns occurring in the study area and the other general street network types. Conclusions about the travel outcomes related to a full range of street networks (conventional, existing streets and new models such as the Fused Grid) will be possible to make from the associations revealed by the analysis.  V. Findings & Limitations The research results and discussion (Chapters 6 and 7) will summarize findings concerning the central hypothesis that the Fused Grid network type encourages more walking mode share while reducing motor vehicle travel demand. A continuum of street network connectivity types, and the likely travel behaviour of changing connectivity along this gradient, will be discussed. The dataset includes measured urban form of three distinct cities built in 48  different eras and is expected to contain enough diversity and variation in street patterns, and thus network connectivity, that the Fused Grid’s pattern will be within the interval of values on the independent variables, allowing inferences to be made about its likely travel performance. The research is cross-sectional and involves correlation and regression, which can describe and contribute to understanding certain aspects of the relationship between built environment and travel behaviour, but cannot be used as a basis for asserting causation between these factors and travel. This can be useful for predicting some likely responses in behaviour and associated outcomes (air pollutants, etc.), but not specifying exact outcomes, from changes in urban form (street network design). The results likewise should be considered an indication about the specific geographic area of study, and caution should be exercised in generalizing from them. A longitudinal study would be needed to investigate causal questions, and further study using the same measures in other regions would be necessary to establish more generalizable understandings. Furthermore, perception and individual/cultural attitudes and how these might affect travel are quite difficult to control for, and the assessment cannot account adequately for residential self-selection (wherein people who are apt to walk more seek walkable neighbourhoods; see Krizek 2003). The effects of traffic and personal safety conditions, and perceptions thereof, are not possible to include, since such data from surveying of the individuals involved in this dataset would need to have been conducted concurrent with the travel and activity survey (in 1999).  49  Chapter 4. DATA DEVELOPMENT  50  I. Introduction The primary measures chosen, ratios of pedestrian to vehicular route directness and network density, allow the distinguishing of different networks according to features influential to travel behaviour because of their basis in distance, a primary constituent of the cost of travel. These measures, as well as other independent variables for which the experiment controlled, entailed substantial processing and computation to create the datasets built around the geographic locations of households participating in the travel survey. This chapter describes the development of the necessary data, which is stored in GIS and Statistical Program for the Social Sciences (SPSS) data files. Further details are provided in technical appendices.  II. Final Data There were two main data sources for the raw data that was processed to create the variables of interest to this research: local governments in the Seattle metropolitan region (GIS data) and the Puget Sound Regional Council’s (PSRC) Travel and Activity Survey of 1999. Subsequent studies conducted by Lawrence Frank and Company (LUTAQH and WSDOT) processed the data into usable variables of travel behaviour and urban form measured to the households’ immediate neighbourhoods (again using 1km network buffers).  51  Table 4-1. Variables for Fused Grid Assessment Independent Variables  Dependent Variables  Network Characteristics  Other Urban Form Characteristics & Factors  Travel and Activity Patterns  Source: GIS data/analysis, King County studies,CMHC project  Source: GIS analysis from King County studies  Source: Travel Diary data for the region from PSRC  Ratio_RD and RatRD_Com: ratio of trip network / airline distance, by mode (walk and vehicle) to two destinations (park/commercial)  Demographics: income, gender, age, ethnicity, and HH characteristics (eg. veh. ownership)  Rat_SW_Str: ratio of sidewalk and trail to street density  Household characteristics: vehicle ownership and hh size  interden_sqkm: intersections per square kilometer  rfa: retail floor area  Trip_SUM; Trips_MoVeh; Trips_Walk: trips total and by mode  Walk_Split; Drive_Split: percent trips by mode  VMT: total distance traveled by vehicle mode  DistWalk: total distance mix_fl_4: mix of uses measure  traveled by walking mode  busstop_SUM: # of transit stops in walking distance (w/in EWA) of household nrd_Census2000: residential density Shading indicates main variables of interest  GIS Data used to build the Network Measures and their Sources: •  Street Centerline and Network data (polylines): Bellevue, Redmond, Seattle, and King County  •  Parks: Bellevue, Redmond, and King County  •  Sidewalks: Bellevue, Redmond, and Seattle  •  Trails: Bellevue, Redmond, Seattle, and King County  •  Parcels: King County and Urban Sims (U. of Washington)  •  Households (generated through address geocoding from PSRC database)  •  Network Buffers: computed from street, sidewalk, and household data.  52  Figure 4-1. Disparate Modal Networks in Seattle, WA  Streets without sidewalks (black lines); Pedestrian connections/sidewalks, sometimes without full street (lighter lines). Orthophoto courtesy of MDA Corporation.  The vehicular network includes all streets that are considered driveable and includes sensitivity to dead-ends, and non-connected features (i.e. a limited access highway which passes over a local street network). The non-motorized network (for walking) includes sidewalks and off-street trails, and crossings of streets as well as stairs and connector paths to the extent possible. These other non-street facilities were included because they comprise walkable pathways which approximate the kind of walking environment that would be available for certain street patterns (like the Fused Grid) having high degree of connectivity for pedestrians, including off-street connections. ESRI’s GIS software (various versions) was used to integrate these data files into a single network with attributes that allow distinguishing between, and separate measurement of, pedestrian and vehicular networks. Data was either created or collected in, or corrected to, the year 1999 in order to match the year of the PSRC Travel and Activity Survey. This matching of measurement of urban form 53  to the same time that the measurement of the travel occurred is critical to the accurate analysis of relationships between built environment and human activity. The quality of this data varied greatly from jurisdiction to jurisdiction due to differing collection methods and lack of availability of particular items of data needed for this analysis - see discussion of limitations in the technical appendix and conclusions for this project. Those GIS data that were not available for this particular year, and which are therefore best estimates of that year’s built environment, include: •  Parcels (UrbanSim data from University of Washington 2000);  •  Streets (current year, but reflecting a street network that has not undergone substantial changes in the past 6+ years and was checked against 1999 orthophotos);  •  Sidewalks (Seattle – created using 1993 orthophotos, updated in 1996 and adjusted to 1999 by this project);  •  Trails (King County – only major trails included; no major trail additions have occurred since 1999 that would affect this network; Seattle - no major trail additions have occurred since 1999 that would affect this network). Figure 4-2. Land Use: commercial and parks, and 1km vehicular network buffer around household  Bright diamond shape highlights 1km network buffer. Orthophoto courtesy of MDA Corporation.  54  Digital orthophotography was used in a GIS to check and correct for accuracy various segments or point locations, particularly in developing the GIS data on the non-motorized network. Bellevue and Redmond were, for their street and sidewalk data, able to provide archived data from 1999. The trail locations in these cities were contained in the King County GIS data, and were verified by looking at the highly detailed (< 1ft-pixel) orthophotos. Connector paths were not necessarily fully inventoried for these two cities, but additional connections were identified in the review of the 1999 orthophotos; Seattle connectors had already been integrated because they were included in the street network database.  III. Urban Form Variable Development This research made use of variables that were developed in previous studies attempting to measure the built environment and travel, (see Table 4-1). These urban form measures of mix of uses and residential density are correlates in an index of walkability and have been built for the 1km buffer of each PSRC survey household. The statistical analysis in this study isolates the influence of transportation network patterns by controlling for these other urban form factors. Using ArcView GIS software, the research computed network connectivities (pedestrian and vehicular route directness) for each household travel survey respondent using the network analyst extension. The software employs a pathfinding algorithm developed originally by Dijkstra to calculate shortest network path from origins to destinations (Ormsby & Alvi 1999). Figures 4-3 and 4-4 show a step in this project’s method of network connectivity measurement. Pedestrian Connectivity The basis for the analysis of pedestrian networks is GIS data on sidewalks and trails. This data was obtained in the form of polyline shapefiles from the cities of Bellevue, Redmond and Seattle in King County, Washington. Additional features of the non-vehicular network, such as stairs and connector paths, were captured from other GIS datasets provided by these jurisdictions. The research then corrected the data as much as possible to be consistent with the time at which the travel survey was conducted (fall 1999), eliminating facilities and infrastructure that had been introduced since that time.  55  Figure 4-3. GIS Process: adding sidewalk attributes to network file in GIS edit session  Orthophoto courtesy of MDA Corporation.  Figure 4-4. Preparation of network: Sidewalks to be snapped to network lines  In order to run network analyst to assess pedestrian route directness, an effort to simulate the actual route that a person walking would take from home (origin) to the nearest recreation or 56  local retail/service (destinations), the data was transformed to be consistent across the three jurisdictions (Bellevue and Redmond local sidewalk data was converted into one line file, snapping the features to the street with which they were associated, rather than multiple lines for wherever sidewalks or walkways were physically separated. In effect, every street segment was attributed either a ‘yes’ or ‘no’ for presence of sidewalks and new segments were added for connections or trails where a footpath was present but not a full street. Notes/rules/assumptions – pedestrian network For the purpose of route measurement, walking was assumed to be dependent on presence of facilities (sidewalks, trails and other paths) that constitute the network, except on local access residential streets. In most cases this will be a good representation of actual pedestrian movement, as people walk from their house to a corner store despite lack of sidewalks on the local streets. Another assumption was that on major streets, as long as there was sidewalk on one side, the street was considered walkable (due to the complexity of modeling pedestrian movements from one side of the street to another. This research assumes street design details (presence of lighting, benches, interesting architecture, and other streetscape) to be relatively constant across the residential areas in this analysis, and not as influential to travel behaviour as distance (and thus our measures, relative route directness and network density). This is likely more true of walking for transportation (appropriate since the project relates the network measurements to outcome data from a travel survey) than for recreation (Lee & Moudon 2006). Urban parks were assumed to provide routing across open areas, unless a trail system was indicated by the trails in a separate GIS shapefile. In the latter case, the route would follow the most direct trail, if possible. Schools were treated differently, as their fenced perimeters mean school grounds provide limited public access; thus unless there is a park status accorded to the school’s playground, school properties were not considered feasible routes for pedestrians or vehicles just as private parcels (and their driveways and walkways) were not considered to be part of the travel network. An exception to this is the University of Washington’s campus, which has a large internal system of paths and streets which were included in the network for this study. Motor vehicle connectivity The street centerline GIS files provided the basis for creating the measureable network for vehicular travel. This project requested information on locations of dead-ends and traffic diversion devices (various kinds of traffic calming will restrict vehicle movements), and received 57  comprehensive data on street networks including one-way streets for the three cities in the study area. If a street is considered driveable, it is included in the routing computation for the same origin-destination pairs as described above. Notes/Rules/Assumptions – vehicle network Restrictions of vehicle movement were included only if they created a physical barrier to vehicle movement rather than simply causing it to slow. This included dead-ends and directionrestricted streets, since the interest is in distance rather than time of travel. Further, because local trips (nearest destinations) were being measured, some conventions of network analysis, such as hierarchy of roads, were not included, except that freeways were excluded from the network as limited access facilities would be used only for relatively distant (i.e. regional and well beyond the home neighbourhood area definition for this research) nearest destinations. Determination of the universe of possible destinations As noted above, an origin and destination pair needs to be specified in order to calculate route directness. The origins selected were the households that participated in the travel survey within the area where complete network data was available. These locations were converted from the travel survey database into a GIS point file using geocoded address coordinates, and then snapped to the network file in a second GIS process. Polygon GIS shapefiles of park and commercial land use parcels were processed to create the destination locations (nearest recreation and nearest shopping/activity location), and were similarly converted into files of points-on-the-network. Notes/Rules/Assumptions - destinations 1) Nearest park/recreation: in order to qualify as a neighbourhood park/recreation destination, the publicly accessible land needed to be developed in some way with recreational facilities for public use (presence of trails, equipment, or otherwise obvious public access). 2) Nearest commercial: in order to qualify as a nearby commercial destination, this project screened the GIS data for the following uses as likely attractors of both pedestrians and vehicle drivers: grocery stores/markets, restaurants/ eating/drinking places, banks, shopping centers, convenience stores, post offices, regional shopping centers, retail stores, and movie theaters. These types of commercial use have been shown to be attractive to walking trips (Lee & Moudon 2006; LUTAQH 2005) and have been used in previous analyses of this region. 58  Figure 4-5. Route Directness Disparity  Disparity in pedestrian to vehicular modal networks to nearest commercial and parks is shown. Orthophoto courtesy of MDA Corporation.  Route directness measurement The pedestrian route directness measure was conducted with the short list of rules noted above for each household. This was based on the assumption that for local, residential streets the presence of a street itself was adequate to ensure a walking route because of the ease of crossing and walking in the street on these low-traffic (and typically lower speed) roadways. A GIS spatial join function was used to generate a table of nearest destinations matched to each household (for both classes of destination). This list of origin-destination (O-D) pairs was then used to compute the network distance, using a “Multiple Closest Facilities” function in ArcView 3.2a to apply the algorithm to the O-D pairs in the table. After this initial analysis impedance factors for those street segments that lacked sidewalk were used to test the sensitivity of results to a more accurate modeling of the barriers and quality of walking environment in these locations. 59  The network distance outputs were then compared to the crow-fly distance (output of the spatial join function in ArcGIS 9.1) for the same O-D pairs to determine route directness for each mode. For example, pedestrian route directness = DPedNet DCrowFly The ratio of pedestrian connectivity to motor vehicle connectivity was then computed (see Equation 3.2) and became the independent variable of network connectivity to be used in the urban form - travel behaviour statistical models. The distinct pedestrian route directness and vehicular route directness measurements were compiled so that they could also be tested in the model. A route directness value of 1.0 would be perfect connectivity (the network distance being equal to the crow-fly distance). Values of route directness in the range of 1.2 to 1.5, typically measured just for the street as a single network (not necessarily accounting for the availability of pedestrian facilities, are considered to be most supportive for pedestrians (Allen 2006). Sidewalk network extent relative to motorized travel Another measure of the walking conditions in a neighbourhood that offers an indication of the comparative continuity of the two mode’s networks is the overall extent of the pedestrian network versus that of the vehicular network. This was measured by summing up the street segment lengths where at least one side had a sidewalk plus trail length and then dividing this result by the overall length of driveable street network.5 Other urban form variables A wide array of variables was made available from associated Central Puget Sound research through the generosity of King County and LFC (LUTAQH 2005; WSDOT, 2005) and the federal and state funding that supported these other studies. For example, the LUTAQH project provided neighbourhood retail variable, which was used in this study as a covariate control in the regression model because it was found to be highly significant in their study. In other studies, conducted more recently in the region and elsewhere in North America (Berke, et al. 2007; Cervero & Duncan 2006), this variable has continued to be a significant factor associated with levels of walking.  5  Alternatively, comparative network density could be computed by creating a ratio of the effective walking area (the 1km non-vehicular network buffer around households) to the local connectivity area (the 1km vehicular network buffer).  60  These previous research efforts have defined a neighbourhood, for the purposes of evaluating walkability, as the area in the immediate vicinity of the subject households. Krizek (2003) and others have warned about ecological fallacies that may arise from choosing too broad an area for neighbourhood. For instance, census tracts can include thousands of households and averaging the values across this many persons can obscure the real effect of the local area. Or using an artificial boundary can arbitrarily leave out some features because they fall just outside the boundary whereas a house nearby the boundary would logically be affected more by the features in its home range. The sources of urban form data used in this study have overcome this by measuring to a defined area (a buffer inscribed using the available street network to generate radii at what is considered to be a comfortable walking distance, usually ¼ mile up to 1 km) around the unit of the travel behaviour – in this case, households. Ideally, to offer a more accurate measurement of overall neighbourhood patterns, or even the pattern within a 1km household buffer, many more origin-destination pairs would be used to develop a mean measurement that captures the area’s characteristics rather than relying on a single route or pair of routes to indicate the overall pattern. Due to time and resource constraints (many households required manual measurement of route directness), this study conducted the measurement of two origin-destination pairs for each household. In order to gain a better indication of the wider neighbourhood area’s characteristic street pattern, neighbourhood averages were developed by combining the route directnesses of multiple households to arrive at an average score for various neighbourhoods wherein a contrast in route directness could be found. This alternative measurement was tested in the regression model only since correlation of the more limited set of averaged data points would not be valid.  IV. Characteristics of Persons & Travel Behaviour from PSRC Travel Survey The travel behaviour data available to this study was the same used in previous studies of the relationship of built environment walkability and reported travel activity in various Central Puget Sound, Washington studies. The databases and most of the needed variables were therefore already prepared prior to this research project. Selection from the full dataset was done to limit the sample of households to those that occurred within the study area (where sufficient non-motorized network data was available), namely the cities of Bellevue, Redmond and Seattle. The data required further processing to remove missing or outlier values from this new sample 61  and to check for normality and linearity of the distribution of dependent variables in order to be ready to perform the regression analysis. Demographic data on each individual survey participant was contained in a “Person” data file which could be linked to household characteristics in a separate “Household” file through a match variable “Household ID.” Likewise, the outcome data (an “Activity” file), total travel distance in automobile modes and mode split, was contained in a separate file but matched to the other two levels. This allowed data collected at one level to be aggregated or disaggregated as needed. Total number of trips, one of the variables of interest, was computed from the “Activity” file using MS Excel and SPSS. All data were merged into a single data table in SPSS. The new built environment measures critical to this research were thus matched to the other data using the household identification number as well. The measures could be spread to a table where each case was an individual trip (over 24,000 cases) or re-aggregated to the person (3080) or household (approximately 1500) levels for analysis, and then a final sample selected after outliers were removed at the appropriate levels.  V. Summary The databases resulting from this collection and processing provided the inputs for the statistical analysis used to characterize and test relationships between transportation network connectivity and travel outcomes. The descriptive statistics and results of correlation and regression analysis are reported in the following chapters. More detailed description of the procedures for preparing data and analysis are included in technical appendices.  62  Chapter 5. RESULTS A: Descriptive Analysis  63  I. Introduction This chapter conveys results of the first phase of analysis of street patterns in relation to travel. The main areas of interest in this research concern the influences of street network pattern on walking and driving travel behaviour. Analytical results were sought to gather information relevant to testing an hypothesis that more direct pedestrian routing (relative to vehicular routing) results in higher walking mode share (increase in % of trips by pedestrian mode) and less automobile use (lower vehicle miles of travel, or VMT, and fewer trips). The descriptive assessment reveals possible relationships between urban form and travel behaviour.  II. Descriptive Analysis Regional Characteristics The Seattle, Washington metropolitan region (Figure 5-1) was chosen as the area for conducting this research because of extensive available urban form and travel behaviour data. The cities of Bellevue (83 km2), Redmond (41 km2) and Seattle (216 km2) together encompass a land area of 340 square kilometers. The geography and climate are similar across the study area, with steep slopes occurring near the extensive shorelines (lake and Puget Sound), several hills interspersed across the urban area, and waterways creating both links and barriers between neighbourhoods and activity destinations. Figure 5-1. Seattle Metropolitan Region, Study Area  64  Transportation Networks – The street networks in these cities include over 3800 kilometers of roadway network (segments, not total lane length) with roughly 85% as much length of sidewalks, multi-use trails and other walking pathways. Seattle, though much larger in total land area, has a much greater transportation network density than its neighbouring cities for both streets and walking facilities (see Figures 5-2 through 5-4). Figure 5-2. Measured Travel Networks – Total Length  Figure 5-3. Measured Travel Networks – Network Lengths Per Capita  65  Figure 5-4. Measured Travel Networks – Pedestrian Network Densities  Topographical conditions and, less frequently, intentional transportation facility designs or retrofits of existing street networks create the sort of street pattern that modifies standard types and exhibits characteristics of the Fused Grid in various parts of the study region (Figures 5-5 and 5-6). Figure 5-5. Capital Hill Modified Gridiron – approximating the Fused Grid  Light lines or shading indicate connector paths for pedestrians despite interruptions in the grid of streets for motor vehicles.  66  Figure 5-6. Pedestrian Connection – Seattle modified gridiron  In other locations (south and east Bellevue), the street network was built to a loop and cul-de-sac standard, providing people walking with very indirect routes to neighbourhood destinations. There are also extensive areas of pure gridiron street pattern, usually built with complete sidewalk networks, especially in the pre-mid-20th Century residential areas. This contrast of street connectivity is essential to the process of testing the research hypothesis. The variation in street design, though having limited actual disparity between mode’s network connectivities across the region, runs from traditional gridiron to conventional post-World War II dendritic patterns ending in culs-de-sac (Figure 5-7). The City of Seattle and its suburbs were largely constructed within the past century, limiting variability in street design during its development, but more recent retrofits with traffic calming, pathway development, and other modifications have resulted in patterns that seek to overcome the flaws, e.g. of resulting high traffic volumes or speeds in neighbourhoods, of the various standards, or lack thereof, by which street networks were originally built. Figure 5-7. 20th Century Street Network Patterns Gridiron (ca. 1900)  Fragmented Parallel (ca. 1930)  Warped Parallel  Loop & Cul-de-sac  (ca. 1960)  (ca. 1970s to present)  Street Patterns  Source: Adapted from Grammenos et al., 2005; Southworth & Ben-Joseph 1997.  67  Route directness to commercial destinations was the main connectivity variable tested because the research literature consistently indicates that commercial destinations are more strongly associated with travel than parks and preliminary analysis confirmed this relationship. The ratio of route directness to parks was later tested to gauge model sensitivity. Sample Characteristics Household & Demographic Characteristics – The set of data on personal demographics and household characteristics used for this project was drawn from Puget Sound Regional Council’s Travel and Activity Survey of 1999. The Travel and Activity Survey was designed to generate a representative sample of persons and households across the Central Puget Sound region of Washington State and within its subregions. The descriptive statistics for the sample of households and persons selected from this population are listed in the Tables 5-1 and 5-2 below. Table 5-1. Descriptive Statistics – Person Level Household Characteristics  N  Total Mean Median Std. Deviation Range Minimum Maximum  Total vehicles in household  Household size  1402 1.70 2.00 .795 4 0 4  1402 2.70 3.00 1.206 4 1 5  Freq.  Income Valid  Missing Total  11 = < $10,000 12 = $10,000-14999 13 = $15,000- 24999 14= $25,000- 34,999 15 = $35,000- 44999 16 = $45,000- 54999 17 = $55,000- 74999 18 = $75,000 or more Total System  30 23 81 100 153 177 323 348 1235 167 1420  Pct. 2.1 1.6 5.8 7.1 10.9 12.6 23.0 24.8 88.1 11.9 1402  68  Categorical Household Variable & Frequencies for Income Variable Households There were a total of 1504 PSRC survey households in the three city study area. Of these, approximately 1468 remained after those with outlier urban form or vehicle / household size were excluded, and 1400 have valid connectivity measurements of the buffer areas around the household. Persons These households included 3080 persons, consisting of a representative range of ages and ethnicities for the metropolitan region. A bit under 1400 of these individuals were included in the sample because they reported local, home-based travel or activity during the travel survey.6 Table 5-2. Descriptive Statistics – Person Level Demographics. Age N Mean 1-18 19-34 35-64 65+ Maximum  Total  1365 38.7 23.3% 16.3% 46.7% 13.7% 95.0  Education N  Total  1= not high school grad.  23.7%  2= high school grad  7.4%  3=some college  14.4%  4=vocational/technical grad.  6  1371  2.2%  5=undergraduate/college degree  27.1%  6=Graduate/Post-graduate degree  25.2%  Ethnicity  n  Percentage  White  1227  88.5%  Black  36  2.6%  Asian/Pacific Islander  64  4.6%  Latino/Native American/Other  43  3.1%  Other cases were excluded from the regression modeling due to missing data on one or more covariates.  69  Travel Behaviour -Trips and Tours The persons in the study area households took a total of over 24,000 trips, with a total distance traveled of 113,000 miles. Of these trips, approximately 60 percent were part of nonwork-related tours, but only about half that number can be described as home-based (occurring within the neighbourhood around the household). Less than 7850 miles out of this total travel for the study area households was done within a walkable area (trips ≤ 2.5 miles or 4 km). The trips for local travel (any tour with all trips within in it being less than 2.5 miles (or 4km) in length, one way used in this research to select only the travel occurring in the home neighbourhood) numbered 7851. This amounts to nearly a third (32%) of all trips reported. When outliers of all non-categorical independent variables were removed, the total number of trips included in the analysis was reduced to 6545. Table 5.3 shows the measures of central tendency and dispersion for the trip distance variable and the proportions of trips either walked or driven. Table 5-3. Travel Behaviour (Dependent Variable) - Trip Level Descriptive Statistics  N  Total Trips  Proportion 0 (%) Proportion 1 (%)  Walk Behaviour (Walk_NoWalk; coded 0 = None; 1 = Yes)  Driving Behaviour (Drive_No Drive; coded 0 or 1)  Distance Walked (DistWalk)  Distance Driven (DistDrive)  6545  6545  6545  6545  .1215 .0000 .30616  .8025 .7000 .74910  2.47 .00 2.47  2.50 .00 2.50  77.5 22.5  31.2 68.8  Mean Median Std. Deviation Range Minimum Maximum  Table shows total trips before outlying trips (walk distance > 1.81 miles) were removed.  70  -Persons Of the 3080 persons in the study area, 754 (17%) reported some walking during the course of their 2 day travel diary, a total of 2794 trips, while only 125 (4%) reported sufficient walking distance (≥1.4 miles per day, or 2.25 km) to be considered physically active.7 A smaller sample of people (n = 1468) reported travel of a kind potentially influenced by their home neighbourhood (i.e. at least one of their tours of travel did not include any trips longer than a walking distance, 4 km (2.5 miles). The sampling also accounted for kinds of travel where car mode choice was still a likely option (i.e. trip distance ≥ 0.1 mile). 597 of these persons (37%) took at least one walk trip, all told walking 1066 miles and 2324 trips. Outliers on the dependent variables were removed to derive the final sample of 1387 persons. This was the subset used to analyze the association of variables in the inferential phase of the study. Table 5-4. Travel distance by persons Person-level Travel Distances N Mean Median Std. Deviation Range Minimum Maximum  Walk  Drive  Total  1387  1387  1387  .5108 .0000 1.17505 11.11 .00 11.11  3.7674 3.0300 3.61622 31.21 .00 31.21  4.2782 3.7000 3.53979 31.21 .00 31.21  -Driving In contrast, 63.4% of persons reported driving during the same 2-day survey. The vehicle miles of travel averaged 39.03 to work and 26.20 for non-work trips. Driving comprised almost 80% of all trips reported and a slightly lower proportion of local, home-based trips. Table 5-5. Trip-making by persons Trip – taking travel outcome – Person-level N Mean Median Std. Deviation Range Minimum Maximum  Walk  Drive  Total  1387 1.0224 0 2.00941 16 0 16  1387 3.2314 2 2.82852 23 0 23  1387 4.6626 4 2.99086 22 1.00 23  7  Recommendations for chronic disease prevention relating to obesity and diabetes generally specify at least 30 minutes per day of moderate physical activity like walking. This would translate into approximately a 1 - 1.5 mile walk each day or as estimated for this 2-day travel survey, 2.8 miles.  71  -Local Work and Non-work Travel Table 5-6. summarizes two main modes of travel by type of trip purpose for the final local travel sample to be analyzed statistically. Table 5-6. Type of Travel – work or non-work purpose Type of Travel  Walk  Proportion (%)  Drive  Proportion (%)  Other  Total  286 24.1 568 47.8 333 1189 984 18.4 3264 61.4 1072 5320 1270 19.5 3832 58.9 1407 6509 Note that this sample has excluded all non-home-based travel, all trips below .1 miles, and all outlying walk-trip distances (>1.81 miles), which affects the proportions shown.  Work-related Non-work Total  Descriptive Measurement and Typology of Street Networks Describing the urban form focus of this study, which is street or travel network connectivity across pedestrian and vehicular modes of travel, is a necessary basis for the later correlation and regression analyses. What follows is description of the dataset on the key variables that have been used to measure street patterns and their connectivity. Intersection density The number of intersections per square kilometer, or intersection density, was used as a measure of street connectivity in the LUTAQH analysis of urban form in the broader region in which this study is situated, Central Puget Sound. It was generated from analysis of street GIS. This variable was grouped by the current study according to intersection classes drawn from Southworth & Ben-Joseph (1997). A pattern emerges in both visual inspection of the spatial data in GIS and analysis in the literature on street networks. The gridiron street network type corresponds with the median of this study’s sample (73.08 median intersection density; mean 74.92; Std. Dev. 25.36). Thus the sample includes a high proportion of all neighbourhood street networks that would be characterized as medium to high intersection density classes in the Southworth & Ben-Joseph classification (see below). Southworth and Ben-Joseph (1997) categorize street network designs by number of intersections per 2000’ x 2000’ area (4,000,000 ft2 or 91.8 acres) – •  Loop and cul-de-sac streets = up to 12 intersections  •  Warped parallel and modified grid networks = >12 – 17 intersections  •  Fragmented parallel = >17 - 22 intersections  •  Modified gridiron and classic dense grid = >22 intersections  72  The number of intersections per 91.8 acres must be multiplied by 2.6962 (1/.37) to find the number per km2.8 The result is that the following street network types are matched to a level of intersection density in the PSRC dataset: Low: < 32.35 intersections/square kilometer (equivalent to Loop & culs-de-sac or lower density) Medium: 32.36 – 40.44 Int. / km2 (Warped Parallel); 40.44 – 53.92 Int. / km2 (Fragmented Parallel and other long block grid or curvilinear networks) High: 53.9 Int./ km2 or more (Neotraditional/Modified Grid to Classic central area Gridiron)  The Fused Grid street network has a range in intersection density from at least 72 / km2 (Barrhaven District, Nepean, Ontario) up to 82 / km2 (pedestrian quadrant) depending on the length of the block faces planned and the type of Fused Grid district configuration proposed, putting it on par with the mean of the street network of this trip sample. In terms of this intersection density measure, the Fused Grid appears to be well within the range of values that occur in the Seattle region sample. Measurement of Disparity: Ratios of Route Directness and Network Density – Street network patterns vary in the kinds of movement and modes of travel they support. One way to assess this variation is by measuring the relative route directness (itself a ratio of available network distance to crow-fly or airline distance) to a destination by mode on each mode’s respective network, expressed as a ratio of one mode’s directness to the other. Another way to assess the different types of street pattern is to assess relative levels of network density across modes. These two measurements were at the core of this research. A typical gridiron street, or any street with regular, continuous sidewalks on it but no street closures or diverters, pedestrian permeable park lands or plaza blocks, or mid-block foot passageways, will have a pedestrian to vehicle ratio of route directness of 1:1 (or 1.0). Other street networks may create a less direct pedestrian network (ratio >1.0) by omitting sidewalks or including large streets that may be crossed only at lengthy intervals (e.g., signalized intersections spaced far apart along a multilane arterial). Within a loop and cul-de-sac neighbourhood of the post-World War II era, such omissions of sidewalk, even on collector streets, were routine. Another type of street network ratio of route directness (ratio <1.0) occurs where additional walking facilities are provided, or barriers placed, that shorten the pedestrian (and often other non-motorized) paths relative to driveable routes. This is characteristic of the Fused Grid street 8  The conversion factor to the units that have been used in this UBC research project is # of acres * 0.00404685642 to derive the quantity in square kilometers. 91.8 acres = .37 km2. 1/x converts this number to 2.6962.  73  network, which is featured in the table below (Table 5-7). The range of connectivity patterns across the study region can be noted in images generated from the GIS analysis (Fig. 5-8). Figure 5-8. Route Directness Examples Queen Anne – Pedestrian Supportive Grid  East Bellevue – Auto Oriented Streets  Route directness ratios are shown as household (square) points, with a light-shaded star indicating higher pedestrian directness to two nearby destinations and a dark triangle higher vehicular directness. Clusters of these disparate modal route directnesses roughly correspond to types of street network: the former to modified grids, the latter to dendritic loop and culs-de-sac street networks.  These images give a sense of the range of network connectivity across the three cities. One would expect that as the crow-fly distance and pedestrian route directness variables increase (i.e. nearest destinations to walking becomes less direct relative to crow fly path), the amount of walking would diminish. The opposite would be true of intersection density, where an increase in density is associated with smaller, more walkable blocks, which, as previous studies have demonstrated statistically, should foster higher levels of walking. Finally, comparative network density, expressed in this study as ratio of sidewalk/trail to street is another way of assessing a street pattern’s connectivity and continuity for travel by different modes. This study hypothesizes that where this ratio is increased (as in a grid pattern with added shortcuts or vehicle-restricted segments), walking mode choice will increase. The Fused Grid street patterns’ measurement on these network variables is shown in Table 5-7.  74  Table 5-7. Fused Grid Measurement9 The table below summarizes measurements of the Fused Grid on the connectivity parameters, the one from previous work (intersection density) and the new variables (ratio of route directness to nearest commercial and ratio of lengths of sidewalk and major trail to street segments).  Fused Grid Connectivity & Continuity Intersection Density (# / sq. km)  Range 70 to 85 / km2  Ratio Route Directness 0.61 to 1.0  Ratio Sidewalk-to-Street 1.19 to >1.4  Mean  0.89  1.3  Figure 5-9. Contrasting Networks, Seattle Region Study Area  Seattle (Queen Anne)  Bellevue (East)  Redmond (South) Networks at same scale (1:5500). Bolder lines indicate sidewalk or trail. Shaded polygons are parks. Simple black lines are streets; bolded polylines are sidewalked streets; parks are light-shaded polygons. Each map is the same scale. 9  Measurements were obtained by hand measurement of Fused Grid designs for Barrhaven in Nepean and the basic pedestrian quadrant, averaging repeated measures of distinct 1 sq. km areas applied to the designs. Highways or limited access automobile-only roads were not included in the measurement of streets.  75  The measurements allow the investigators to test the relationship between urban form and travel behaviour at an enhanced level of detail and to match the travel behaviour of those neighbourhood street characteristics that most closely resemble the Fused Grid in their pattern and configuration. The characteristics of the Fused Grid are as follows: Intersection Density 70-85 per square kilometer (lower than many gridded networks in the study region, but higher than street networks for newer residential development, and within the interval of data). Route Directness Ratio Average of 0.89 ratio of pedestrian to vehicular routing, on the lower end of range of values for street patterns in the study area – about one standard deviation below the mean (see Table 5-8). The route directness of the Fused Grid, then, corresponds to the more pedestriandirect street networks in the study area. Comparative Network Density A wide range of values is possible on this measurement for the Fused Grid, but measurements taken for this research obtained values usually greater than 1.2, again associating the Fused Grid with the upper values in this study’s range of data. Table 5-8. Network Measures – Descriptive Statistics – Person  N  Valid Missing  Mean Median Std. Deviation Range Minimum Maximum  Ratio of Route directness to Commercial (Ped/Veh) 1420 0 .99 1.00 .10 1.16 .40 1.57  Ratio of Sidewalk/Trail length to Street length 1420 0 .82 .95 .25 .92 .21 1.13  Neighbourhood Patterns As noted in the methods chapter, the difficulty of using partial measures (i.e. measurements of just two routes from each household) of connectivity is that the overall neighbourhood pattern could be missed or poorly represented. An average value of the ratio of route directness (to park and commercial) was computed by aggregating to logical neighbourhood areas. The neighbourhood areas that most closely approximated the Fused Grid, were Capital Hill/Miller Park; Queen Anne; and Fremont/Wallingford. Neighbourhoods with contrasting street patterns and the opposite network connectivity (better vehicle than pedestrian 76  route directness) included north Seattle’s North Beach, Haller Lake, Olympic Hills and Wedgwood as well as south and east Bellevue. This provides another sample for analysis. Urban Form Covariates (Controls) This group of variables constituted an important set of controls for the experiment because they are known covariates in the relationship between urban form and travel behaviour. Intersection density was described in more detail above because of its close relation to the new network variables created in this research. Neighbourhood retail, or the number of locations with this kind of commercial use within the household buffer, has been shown to be associated with walking behaviour in previous studies using the same data (LUTAQH 2005). Likewise, residential density and mix of uses have been shown to be influential to the outcome variables both within this region and in similar studies of other metropolitan regions (Cervero & Kockelman 1997; Frank, et al. 2005). Descriptive statistics for these other variables can be found in Tables 5-9 and 5-10. Table 5-9. Other Urban Form Descriptive Statistics – Household Sample  N Mean Median Std. Deviation Range Minimum Maximum  Valid  Neighbourhood Retail (number in 1km buffer)  Net Residential Density (units/acre)  1271 29.39 21.00 28.129 121 0 121  1271 2.9939 2.2010 2.54556 23.41 .95 24.37  Intersection Density per sq. km 1271 74.9163 73.0770 25.35994 142.42 10.13 152.55  Crow Distance to Commercial 1271 1000.88 803.64 780.32 3765.57 .00 3765.57  Table 5-10. Other Urban Form Descriptive Statistics – Person Level  N Mean Median Std. Deviation Range Minimum Maximum  Valid  Neighbourhood Retail (number in 1km buffer) 1402  Net Residential Density (units/acre) 1402  28.67 22.00 26.762 121 0 121  2.74 2.07 2.29 23.38 .99 24.37  77  Intersection Density per Crow Distance sq. km to Commercial 1402 1402 75.25 74.60 24.48 149.48 13.13 162.61  1019.92 814.27 763.46 3746.46 .00 3746.46  Description of Walking Behaviour Associated with Network Connectivity An initial description of reported walking behaviour associated with high and low levels of connectivity across the two modes is presented in the 2 x 2 matrix tables below. These tables show the top two quartiles of connectivity (route directness by mode to nearest commercial) as high connectivity and the bottom two as low. There are comparatively few cases of disparity (low-high or high-low).10 The reported mean values, for share of trips by walk mode (Table 511) and total walking distance traveled (Table 5-12) are home-based travel from PSRC travel survey data aggregated to the person level. The neighbourhoods listed are examples. These exploratory findings are a preliminary indication of the relationships between walking and street patterns. The boxes in the tables above are cases (persons) grouped by pattern of street connectivity. Those people living in the neighbourhoods with high pedestrian and low vehicular connectivity (ratio of route directness < 1, or Type III in the typology shown in Figure 3-3) have the highest levels of walking for their local trips. Where pedestrians are offered a more direct route, as would be the case with the Fused Grid street design, there tends to be more travel on foot, both in mode share and total distance. These descriptive results offer hints about relationships that deserve further investigation, to draw statistically valid inferences while controlling for other influential factors, about travel behaviour relating to street design.  10  The ranges in each of these classifications of connectivity are as follows: Pedestrian (sidewalk, pathway and local streets) Low corresponds to Pedestrian Route Directness (PRD) = over 1.33 High corresponds to PRD = 1 to 1.33 Vehicular (street) Low corresponds to Vehicular Route Directness (VRD) = over 1.34 High corresponds to VRD = 1 to 1.34 Note: sample data was a subset of the PSRC travel survey including all persons (n = 3081) and all their home-based travel in the study area, not the final, outlier-excluded sample of travel used in the inferential analysis to follow.  78  Table 5-11. Walking Mode Share and Street Connectivity Disparate Street Pedestrian Connectivity  Connectivity and associated Walk Shares (by person to commercial)  Low  High  N and S Bellevue, N Seattle – Grid and major streets w/o sidewalks Type I  Downtown and Older Seattle Neighbourhoods – Gridiron Type II  Mean Mode Share: 10% walking  Mean Mode Share: 14% walking  n = 59 SE and Central Bellevue; SW Seattle – Loop and Culs-de-Sac Type IV  n = 966 Queen Anne, Capital Hill (Seattle) – Modified grid w/connectors Type III  Mean Mode Share: 10% walking  Mean Mode Share: 18% walking  High Vehicular Connectivity Low  n = 985  n = 66  Table 5-12. Walking Distance and Street Connectivity Disparate Street Connectivity Pedestrian Connectivity  and associated Distances Walked (by person, in home  Low  High  neighbourhood) N and S Bellevue, N Seattle – Grid and major streets w/o sidewalks Type I  Downtown and Older Seattle Neighbourhoods – Gridiron Type II  Mean Travel Distance: .32 miles walked  Mean Travel Distance: .66 miles walked  n = 41 SE and Central Bellevue; SW Seattle – Loop and Culs-de-Sac Type IV  n = 766 Queen Anne, Capital Hill (Seattle) – Modified grid w/connectors Type III  Mean Travel Distance: .49 miles walked  Mean Travel Distance: .74 miles walked  High  Vehicular Connectivity  Low  n = 714  79  n = 47  III. Summary The foregoing descriptive statistics provide the foundation for a more detailed exploration and analysis of street network design and travel behaviour. Enough variation in street network pattern occurs within the study area and travel data samples selected to draw conclusions in the final phase of inferential statistical analysis. The preliminary indication is that street networks’ variations in connectivity are associated with differences in travel behaviour. These conclusions may be applicable to other patterns of street network connectivity and generalizable to other regions, particularly when tested more thoroughly. The next chapters will summarize results from correlations and regression analysis, including assessment of likely travel behaviour outcomes associated with a variety of street network configurations, especially the Fused Grid.  80  Chapter 6. RESULTS B: Inferential Statistics  81  I. Introduction – Inferences Regarding the Relationship Between Street/Network Pattern and Travel Behaviour The descriptive assessment provided basic measures of central tendency and dispersion for the travel and urban form samples for this study. The salient conclusions from the descriptive analysis are that street networks do differ in how much connectivity they provide to various modes and that there appears to be a relationship between measured disparity in street connectivity and the travel of people living within these street networks. The following results provide additional statistical information that will allow conclusions as to how travel behaviour and street network pattern vary together (if at all) and how significant and generalizable these results are. The analysis models the relationship between street network patterns and travel outcomes in the presence of other known factors affecting travel, thus allowing isolation of travel and the measured street network connectivity. Moreover, the analysis provides details on the magnitude of the change in travel behaviour that could be expected from changes in street patterns.  II. Correlations Correlation statistics were used to explore relationships among variables. The correlation process does not assume a particular direction of influence in the relationship between the variables, but provides information about direction (positive or negative), form and degree of association. Pearson coefficient of correlation (r) results indicate whether an association is statistically significant (i.e. non-random) and how strong (i.e. how well fit) it is. Correlation results are shown in the tables that follow. Results of Bivariate Correlations Tables 6-1 and 6-2 show the results of correlations using the person-level data. The correlations between demographic and household variables and the travel behaviour data for the person-level sample showed moderate associations of number of walk trips and vehicle miles traveled with total vehicle ownership and household size. Weaker, yet still significant, was the relationship of walking distance travel outcomes with age, though VMT did not significantly associate with this same variable. The person-level walking travel data also showed significant and mainly stronger relationships to urban form variables. When present in a localized area around households, 82  neighbourhood retail and mix of use (which measure similar phenomena, proximity of attractive uses that would generate travel) associated moderately with walking behaviour. Because it was a stronger association and highly correlated with Mix of Uses (r > .65), the neighbourhood retail variable was used instead of ‘Mix’ in the regression models. Proximity to transit (Bus Stop Density) also had a moderate association, but was also strongly correlated with another variable of interest - relative network density (r > .42) - so it was not used in the final models except to test for sensitivity. Net residential unit density within the household buffers was weaker, yet still highly significant in association with walking and driving travel distance. Intersection density was highly correlated with the main continuity variable, relative network density (r >.61) and so was dropped from the regression models. The variables of interest for this research had the following correlated results: Ratio of route directness is negatively correlated with distance walked, as expected, but the strength of this correlation can be characterized as weak (Pearson r = -.044), and barely not significant at the 10% level (p = .104). The sidewalk to street ratio (relative network density) had a positive association with walking distance traveled, with highly significant (p=.000), moderate strength correlation (r = .186). As more sidewalks or other formal pathways are built within an area, i.e. pedestrian network density is increased relative to vehicular street density, walking travel would be expected to increase. Table 6-1. Correlation – Walk Distance Traveled and All Factors Demographic Measures Total vehicles in household Household Size Age Urban Form Measures Crow-fly Distance to Commercial Intersection Density Mix of Uses Neighbourhood Retail (number in 1km buffer) Net Residential Density (units/acre) Bus Stop Density (# / square km) Network Connectivity Measures Ratio of Route directness to Commercial (Ped/Veh) Ratio of Sidewalk/Trail length to Street length * = significant at the 90% level ** = significant at the 95% level *** = significant at the 99% or higher level  83  Pearson Correlation (r) -.229(***) -.128(***) -0.045(*) r -.149(***) .202(***) .196(***) .276(***) .098(***) .273(***) r -0.044 .186(**)  Sig. (p) 0.000 0.000 0.099 p 0.000 0.000 0.000 0.000 0.000 0.000 p 0.104 0.000  Similar, converse associations are revealed in the correlations for driving behaviour (Table 6-2). The signs on the coefficients of demographic and urban form correlates for driving distance are opposite that of the walking travel, as would be expected. The ratio of route directness was weak (r = .069) yet significant at the 1% significance level. Route directness ratio became more significant and gained in strength relative to how it associated with walking, while network density ratio decreased in strength. Table 6-2. Correlation - Distance Driven (VMT) and All Factors Demographics Measures Total vehicles in household Household Size Age Urban Form Measures Crow Distance to Commercial Intersection Density Mix of Uses Neighbourhood Retail (number in 1km buffer) Net Residential Density (units/acre) Bus Stop Density (# / square km) Network Connectivity Measures Ratio of Route directness to Commercial (Ped/Veh) Ratio of Sidewalk/Trail length to Street length  Pearson Correlation (r) .238(***) .202(***) 0.009 r .207(***) -.174(***) -.194(***) -.238(***) -.091(***) -.199(***) r .069(***) -.121(***)  Sig. (p) 0.000 0.000 0.731 p 0.000 0.000 0.000 0.000 0.001 0.000 p 0.010 0.000  Correlations of the trip level data produced similar results (see Technical Appendix – Results - Correlations): moderate, negative correlations of walk trips with total vehicles and household size. Weaker but expected associations result with demographic variables. The urban form and network connectivity measures held the same kind of relationship, with very similar correlation coefficients, to the person level results with the walking and driving variables, though with higher significance. For example, the ratio of route directness was significant at p < .05 in relation to walking distance.  III. Travel Behaviour Regression Modeling Several different regression models were tested in the course of this research, all with the aim of ascertaining more specifically the kind of relationship that exists among the hypothesized predictor (urban form) factors and response (travel activity) behavioural phenomena. The crosssectional research approach used here can provide greater clarity on the kind of relationship that likely exists and allows prediction about likely behavioural results from modifications of the 84  design of urban form and particularly street patterns. The use of regression statistics allows this study to test for the strength of the same variable relationships as emerged in the correlation results while at the same time controlling for other factors that may be influencing the reported travel response variables.11 As discussed in Chapter 3, a binary logistic regression model was developed to test the likelihood of someone walking or not walking for their neighbourhood travel in relation to various conditions. The mathematical expression of this regression is shown in Equation 6-1: Equation 6-1. logit( Y-walking versus not walking) = natural log(odds Y) = ln (π/1- π) = α + βX or in a multiple regression α + β1X1+ β2X2 + β3X3… Where X is independent variable(s) - demographic or urban form control or main variables of network connectivity.  This analysis assesses the change in the log-odds of walking from a change on X (with other independent variables being held constant). The coefficients, β1 etc., must be converted in order to obtain the odds ratio that is predicted from change in the independent variables. This is shown in the results tables below as Exp(B).12 Nagelkerke’s r-squared is the most direct parallel to, though generally lower than its equivalent, r-squared (coefficient of determination) from ordinary least squares linear regression statistical models.13  Linear regression modeling was  conducted as well where possible - on continuous travel outcome variables. The logistic regression models provided significant explanation of walking and driving behaviour. The final models ranged in Nagelkerke’s r-squared from .17 to .28 (and lower values in the linear regressions). These are moderate effects, considered to be within the range of values on explanatory power for drawing conclusions in social science. Table 6-3 summarizes the results of the logistic regression analyses.  11  Overall model significance is indicated by a coefficient of determination (R2), done at each stage in the modeling process so that separate variable subgroups (i.e. blocks of demographics and urban form) can be compared for their distinct contributions to the overall explanation of behaviour. R2 can range in value from 0 to 1, with 1 indicating that 100% of behaviour can be explained by the model. The Beta coefficients (B) provide measurement of the strength of each factor’s relationship to the travel outcome being regressed. Like correlation coefficients, they also indicate direction of relationship. B, as regression output, affords the ability to make predictions because the travel behaviour (Y - dependent) becomes a function of the various factors (X - independents), with the coefficient being the slope of the curve. 12 A value < 1 on the Exp(B) output for a variable indicates that it bears a negative association with the dependent. 13 More of the total variation in the probability of the dependent variable is explained by the independent variables in a logistic regression model than its Nagelkerke’s r-squared value would indicate.  85  Table 6-3. All Binary Logistic Regression Results – Model explanatory power Level of Travel Outcome Sample  Variables included  Model R-Square  Significant Factors  Street Pattern Factors Block – sig.?  Trip Walking vs. Not Total vehicles, Walking Household size,  Income, Age, Education, Gender, Ethnicity, Neigh. retail, Net resid. density, Crowfly dist. to commercial, Ratio of route directness, Ratio of network density  Person Walking vs. Not Walking same Driving vs. Not Driving  Active (Walking 30+ minutes or Not)  same  Total vehicles, Gender, Income, Education, Neigh. retail, Ratio of route directness, Ratio of Network density  0.186***  0.173***  0.281***  0.213***  Total vehicles, Household size, Income, Age, Education, Ethnicity, Neigh. retail, Crowfly dist. to commercial, Ratio of route directness, Ratio of network density Gender, Income, Neigh. Retail, Ratio of network density Total vehicles, Household size, Age, Neigh. retail, Net resid. Density, Crowfly distance Total vehicles, Gender, Neigh. retail, Ratio of route directness, Ratio of network density  Yes***  Yes**  No  Yes**  Walking – Logistic Regression The odds of walking (probability that a trip or person will be on foot for any given local travel divided by the probability that some other mode of travel will be used) showed consistent, statistically significant increase with increases in each of the following variables: Neighbourhood Retail Ratio of Network Density  Odds of walking decreased significantly with increases in14: Total vehicle ownership Household size  There was a highly significant, negative relationship of trip-level pedestrian travel with ratio of route directnesses across modes and an equally significant, positive relationship with  14  While the ratio of route directness (relative modal network connectivity across walking and driving) showed a significant relationship for the trip level sample, it was not significant at the person level – though other person-level analyses were significant and consistent with associations found in the correlation analysis.  86  ratio of network densities. Both the model and the step in which the network variables were added were significant, explaining over 18.5% of the odds of a trip being a walking trip. The relative strengths or importance of the independent variables in relation to the outcome variable was checked by examining the standardized logit coefficients (which are calculated using zscores of the input variables, see Technical Appendix - Z-score model runs). In the walking or not walking trip-level analysis, the ratio of route directness and ratio of network density coefficients using these standardized variables (-.198 and .161 respectively) rank about on par with demographic variables and other urban form, with the exceptions of the stronger results from total household vehicles (-.369) and neighbourhood retail (.428). Many of the same relationships were indicated when regressing person level travel in the same model (see Table 6-4 and Technical Appendix), with the neighbourhood retail (p = .000) and ratio of network densities (p = .015) remaining significant among the urban form and network characteristics that associate with walking. Income remained significant and gender became significant, but the ratio of route directness became non-significant as a factor in explaining the odds of walking (p = .650). The model as a whole continues to be highly significant, explaining at least 17% of the odds of a person walking. The step in which the network variables were added remained significant at the 5% significance level. In this model, the beta weights (see Technical Appendix – Regression Analysis) indicate increased importance of the ratio of network density relative to other urban form. Table 6-4. Walking vs. Non-Walking Trips/Persons Travel Outcome Trip – Walking vs. Not Walking  Person – Walking vs. Not  Variable Ratio of Route Directness Ratio of Network Density Ratio of Route Directness Ratio of Network Density  B Coefficients  Sig.  -1.073  0.008  0.635  0.002  .311  .696  .905  .014  Added model, rsquared from the street pattern measurement block  .005  .007  90% Confidence Interval Upper/Lower .155 .752  Exp(B) Odds Ratio .342  1.273  2.798  1.887  .369  5.046  1.364  1.351  4.521  2.472  Physical activity – Logistic Regression A separate logistic regression model was constructed around whether a person walked a 87  sufficient distance (about 1.4-1.5 miles (2.4 km) per day, or 2.8 miles in the two-day travel survey) to be considered physically active enough to meet recommendations for health.15 The results, shown in Tables 6-3 and 6-5, indicate that this regression model explained more of the variation in physically active travel behaviour (r-square = .213) than it did of whether any walking was done. Network connectivity was significant (the block of variables enters the model at a p < .05 level), and there were significant (p = .086 and .069, respectively) relationships of ratio of route directness and relative network densities to the outcome of odds of walking 30 minutes or more, with the expected signs on the coefficients.16 Table 6-5. Active Walking - Person level logistic regression Travel Outcome  Variable  Active Walking  Ratio of Route Directness Ratio of Network Density  B Coefficients  -2.306  Sig.  Added model, rsquared from the street pattern measurement block  .086  90% Confidence Interval Upper/Lower  Odds Ratio  .011  .907  .100  1.173  24.358  5.344  .016 1.676  .069  Driving – Logistic Regression Another major travel outcome variable is driving behaviour. This “Driving or nondriving” person dichotomous variable was regressed on the predictor variables using the same sample as was used to evaluate the ‘walking versus not walking’ outcome (see Table 6-6), though at least one predictor variable (education) was dropped in order to reduce the degrees of freedom to meet the 1:10 ratio test noted in footnote #16. While the model was significant and explained 28% of the variation in odds of driving, the block of network connectivity variables was not significant at the 10% significance level.  15  Walking 30 minutes or more per day achieves a level of physical activity that is recommended for reduced risk of chronic diseases stemming from obesity and diabetes. Measuring the influence of urban form on this kind of travel behaviour provides information for evaluating public health outcomes from changes to community design. 16 Fewer independent variables were included in this model because there were fewer cases in the smaller category of the dependent variable (only 73 people in the sample walked 30 minutes or more each day, limiting the maximum possible number of independent variables to seven in order to maintain the needed 1:10 ratio for valid analysis). In order to save on degrees of freedom, two demographic predictors were transformed into dummy variables: income (high and low) and education (college degree or higher or not); others showing less significance were dropped entirely. Similar recoding of variables into dummies was done to allow for their inclusion in the linear regression models.  88  Both network measures had expected signs on their coefficients, but only the ratio of network densities was close to being significant at the 10% level (p = .130). Table 6-6. Driving versus No-driving Travel – Person level logistic regression Travel Outcome  Variable  Driving vs. Not Driving  Ratio of Route Directness Ratio of Network Density  B Coefficients  .827  Sig.  Added model, rsquared from the street pattern measurement block  .408  90% Confidence Interval Upper/Lower  Odds Ratio  2.287  .322  16.245  .501  .205  1.226  .004 -.691  .130  Other Travel Outcomes: Linear regression of continuous response variables Continuous variables, such as numbers of trips or distances traveled, can be analyzed in a linear regression statistical model. This research tested for relationships of some key outcomes: number of local trips and miles of travel by mode. The linear regression results on all of these travel outcomes are shown in Table 6-7, and the full model output of distance driven by persons is shown in Table 6-8. The full model outputs for the other continuous travel behaviour variables can be found in the technical appendices (Technical Appendix – Regressions, Linear).  Table 6-7. Linear Regression Results Summary – Person-level B Coefficients Travel Ratio of Route Outcome Directness Distance -.422 walked Walk trips -.447 Distance 2.300** driven Vehicle trips .996 Total trips 1.627* * = significant at the 90% level ** = significant at the 95% level  Overall model, r-squared  Ratio of Network Density .311*  .136  .463* -.461  .154 .132  -.325 .229  .104 .044  89  Table 6-8. Vehicle Miles of Travel (VMT) – person-level linear regression Full Model - Variables  Unstandardized Coefficients  Standardized Coefficients  B -1.322  Std. Error 1.393  TotVeh  .462  .153  .101  Gender  .134  .202  Age  .016  .006  Household Size  .515 .057  (Constant)  White or non-white dummy of education variable (college educated or not) IncomeDummy (> or < $35k) Neighbourhood Retail (number in 1km buffer) Net Residential Density (units/acre) Crow Distance to Commercial Ratio of Route directness to Commercial (Ped/Veh) Ratio of Sidewalk/Trail length to Street length  Beta  t  Sig.  Zero-order -.949  Partial .343  3.027  .003  .018  .662  .508  .092  2.648  .008  .108 .328  .172 .005  4.777 .173  .000 .863  .455  .225  .062  2.017  .044  .198  .293  .021  .676  .499  -.016  .005  -.118  -3.416  .001  -.049  .045  -.031  -1.110  .267  .000  .000  .085  2.483  .013  2.300  1.108  .058  2.077  .038  -.461  .486  -.029  -.947  .344  These results show a significant relationship between ratio of route directness and vehicle miles of travel and overall number of trips. The positive association means that where vehicle routing is more direct relative to pedestrian routing, both more vehicle miles of travel and more total trips would be expected. The ratio of network densities (pedestrian-mode : vehicular-mode) had significant positive association with walking distance and number of walking trips, with the highest overall model fits (see Technical Appendix). The overall model R2 for VMT was .132 and the adjusted R2 (.113) was calculated using Stein’s equation (see technical appendix).  IV. Interactions, Sensitivity Testing, and Neighbourhood Differences Other work was done as part of this assessment to test various relationships, such as checking the effect of an interaction term, grouping households into clusters by neighbourhood connectivity pattern, and testing for sensitivity to certain data assumptions. Interaction of Distance and Connectivity Related to the main hypothesis, the notion that the association of ratio of route directness with travel behaviour could be affected by the total distance to the destination (i.e. certain 90  distances may be thresholds beyond which one or another mode is a far less likely choice) was tested using an interaction term, essentially including as another regression model variable the product of two variables thought to interact. The interaction term here is a product of ratio of route directness and crow-fly distance to account for the changing effect (non-linearity) of the ratio of route directness variable as crow-fly distance to the nearest commercial destination varies. For nearby distances (those less than 1/10 mile), it was assumed that there would be very few vehicle trips; conversely, at distances greater than ½ mile to nearest destinations, walking becomes much less likely. The lowest distances were already eliminated from the sample used in this analysis, but the upper limit of trip distances included in the sample was about 2.5 miles, to capture travel in an area that could be considered part of the home neighbourhood.17 The interaction term (Crow-fly distance x Ratio of Route Directness) used in the walk vs. no-walk logistic regression model at the trip level created a very small improvement on the significance of both connectivity variables and the full model’s r-square (+.001) . Used in the person level model, including the interaction term made the model stronger – improving model r-square by .014 and moving the ratio of route directness much closer to significance. The results of these tests are summarized in the table below (Table 6-9) and full model outputs can be found in the technical appendix. Table 6-9. Interaction Effects – crow-fly distance and connectivity to nearest commercial Interaction –Crow-fly and Connectivity Trip Person  Improvement of model R-squared .001  .014  Effect on Ratio of Route Directness sig. .008 to .003 .669 to .261  Effect on Ratio of Network Density sig. .002 to .001 .015 to .024  Sensitivity Testing of Sidewalk Availability/Use Assumptions and Connectivity Variables Route directness ratio was also measured and tested as separate modes’ route directnesses and with a few levels of impedance (10, 20 and 100%) to weight the unsidewalked local access streets, effectively accounting for the possible effects of discontinuous pedestrian facilities on pedestrian activity before deriving the measurement of route directness. The separate modes’ impeded route directnesses were significantly related to the outcomes variables in the correlation and regression models, but did not add explanatory power. The impeded route directness  17  Walking trip distances were further limited to 1.81 miles because trips longer than that were more than 3 standard deviations from the mean (i.e., these were outliers).  91  showed somewhat stronger association with the travel outcomes of interest (see Technical Appendix B – Sensitivity). However, the results using impeded route directness complicate the interpretation of the connectivity measure because continuity is imbedded in this measurement of network pattern. For one, multicollinearity becomes a more serious problem if both the impeded variable and the ratio of network density are included in the model, since the former uses the sidewalk length as a determinant of the impedance (the two variables, depending on the level of impedance, are correlated at about r=.45). The approach of using the ratio of network densities was selected as a means of including the completeness of the sidewalk network in the analysis, as a separate variable, with the main assumption for the connectivity (ratio of route directness) measurement being that local access streets could be walked even where they lacked sidewalks. Other impedances can be included on the pedestrian network (i.e. intersections/crossing; see Schlossberg, 2006), but this research was not able to measure these other dimensions of pedestrian environment for the time period that matched the available travel behaviour data. The network density measure was used as a proxy for continuity of the pedestrian route, thus allowing this research to capture some of the impedance that pedestrians would experience as a complement and control to the connectivity (route directness) of their local travel network. Intersection density is another measure of connectivity that has been useful and associated with changes in behaviour in previous studies. In this research, Intersection Density was highly correlated with one of the two variables of interest (r > .6 with relative network density). However, intersection density was significant and moderately correlated using the local travel sample, and so it was tested in the logistic regression model (by removing the ratio of network density variable). In the trip model of walking versus not walking, including ratio of route directness, it was not significant (see Technical Appendix - Sensitivity). Connectivity to Nearest Parks The research also conducted analysis on the ratio of route directness to nearest park locations. Interestingly, while route directness to nearest park did not have significant correlations with the walking and driving distance outcomes at the person level (as had the same connectivity measure to commercial destinations), it did have a more significant and opposite sign on the beta coefficient (positive) when analyzed in the trip level logistic regression (see technical appendix). There was not a significant association of park ratio of route directness with odds a person walked nor odds a person walked enough to be considered physically active. 92  Other Sensitivity Tests A few other sensitivity tests were conducted in the course of this analysis, including transit service proximity and clusters of similar neighbourhood-level street connectivity. These results did not seem to add substantially to the explanation of travel behaviour or, as above, involved correlated predictors or difficulties in interpretation because of similarities to other variables already in the model. These results are summarized in the technical appendix (“Other Sensitivities in Appendix B. – Sensitivity).  V. Summary: Relative Strength of Explanatory Factors The standardized version of Beta coefficients from the logistic regression allow the different explanatory variables to be compared and their relative importance to be assessed. In the logistic regression models, standardized coefficients were obtained by entering the z-scored variables directly into the model. The results for the walk/no-walk model (Tables 6-10 and 6-11) indicated that, as was true in the correlations, total vehicle ownership and proximity of neighbourhood retail are the two strongest variables with the network variables being less important factors, on par with the other demographic and other urban form factors in the model. At the person level, Ratio of Route Directness was insignificant. Table 6-10 & 6-11. Standardized Beta Coefficients –Walking logistic regression 6-10. Trip-level Step 1 Z Total Vehicle Z Household Size Z Age Z Neighbourhood Retail Z Net Resid. Density Z Crow-fly Distance Z Ratio of Route Dir. Z Ratio of Network Density Constant  B –Std. -.369 -.180 -.166 .428 .018 .031 -.198 .161 -.639  6-11. Person-level Step 1 Z Total Vehicles Z Household Size Z Neighbourhood Retail Z Crow-fly Distance Z Ratio of Route Dir. Z Ratio of Network Density Constant  B-Std. -.322 -.136 .475 .089 .034 .209 .465  93  In the linear regression models, the standardized beta coefficients were computed automatically in SPSS as part of the regression and included in the output table. For example, in the model of VMT (see Table 6-8) it appears that Total Household Vehicles, Household Size, Crow-fly Distance to nearest commercial (positive relationship) and neighbourhood retail (negative) had a stronger effect than the street network pattern in relation to the behaviour outcome. Ratio of route directness was on-par with various demographic variables in terms of the degree of association with the travel outcome in this model, and appears to have a stronger relationship than residential density and relative network density.  94  Chapter 7. DISCUSSION  95  I. Findings Interpretation The correlation and regression results presented above demonstrate that urban form measurements do associate with travel behaviour. The results also show that urban form adds significant explanatory power to travel behaviour models when introduced in blockwise fashion after demographic and household characteristics were entered. The street network connectivity and continuity variables, entered as a separate block, also usually enhanced the ability to explain travel outcomes. This is consistent with previous research that has conducted analysis of travel behaviour while controlling for the demographic and household characteristics of the people included in the research. The results indicate that there is a relationship between patterns of street connectivity and transportation outcomes such as walking and driving travel behaviour. The significant findings in relation to both ratio of route directness and ratio of network densities across walking and driving modes signal that one should accept the hypothesis that there is an association between street network patterns and travel behaviour, disparate modal connectivity or continuity on walking and driving levels in particular. The correlation results alone provide evidence for accepting the hypothesis that the way a street network is configured makes a difference in the pattern of travel for a residential neighbourhood. When a transportation network’s connectivity or continuity provides favorable conditions for one transport mode relative to others, travel increases on the mode with the improved relative utility. Changes in Odds of Walking Change in odds of walking is affected by numerous factors, including the covariates that constituted the independent variables in the foregoing analyses and possibly others as well that were not measured here (see ‘Limitations,’ below). Among the strongest factors contributing to differences in levels of walking, already demonstrated in previous studies, are the total vehicle ownership of the household (less walking) and the presence of proximate neighbourhood retail stores (more walking). The magnitude of change in walking behaviour associated with total vehicle ownership is a decrease of about 11% in the odds of walking from the addition of a single vehicle. This change is likely to be affected by thresholds such as increasing from zero cars to one, or one car to two. Adding one more neighbourhood retail destination within a 1km buffer around a person’s home is associated with a 1.8% increase in the odds a person will walk for some portion of their local travel. See Technical Appendix -Regression, Logistic.  96  Predictions from regression modeling The regression model was needed to test for the relative importance of these several factors and to offer predictions while controlling for all other variables. At the trip level in the inferential statistical analysis, this study’s variables relating to street network patterns, the ratio of route directness and ratio of network densities, were significant predictors of travel. The regression model allows us to make the following predictions: •  If route directness for pedestrians is increased 10% relative to motor vehicles (a change in the ratio from 1.00 (or 1:1) to .90, for example, which would be very similar to the change that would result from building a Fused Grid street design instead of a pure gridiron pattern) in a given area, the odds of a trip being walked rather than taken using another mode would increase by 11.3%, all else being equal.18  Likewise at the trip level, and also at the person level, increasing the density of sidewalk or other walking path relative to that of the motor vehicle network (driveable streets) is associated with increased odds of walking: •  At the trip level, a 10% increase in the pedestrian-to-vehicular network density ratio, the equivalent of adding less than a half-block length of pedestrian-only pathway to each block of network area), is associated with a 6.6% increase in odds of a trip being walked.  •  For persons, the same increase in relative pedestrian continuity associates with a 9.3% increase in odds of walking, all other factors remaining the same. There is more explanatory power in the person-level regression models for predicting  physically active walking behaviour and driving behaviour, even though the ratio of route directness and to some extent relative network density were less significant (though still significant at 10% significance level) predictors of these travel behaviours. In testing sufficient levels of walking to be considered physically active, the model indicates that: •  a 10% change on the ratio of route directness is associated with more than 25.9% increase in odds of walking 30 minutes per day or more, and  •  the same magnitude change on the ratio of network density in favor of pedestrian travel is associated with an 18.2% increase in odds of being active through walking.  18  The formula for calculating these estimated changes is: multiply the change in urban form (x) by the Beta coefficient, and then raise e to the power of this product term. So, e(-0.1)(-1.073) = 1.113 or an increase of 11.3%.  97  Odds Ratios The odds ratios from these logistic regressions, expressed as Exp(B) in the SPSS output tables, are another way to express these results. The values for Exp(B) from the logistic regressions indicate the relationship between walking or other binary travel outcome variables and the urban form factors. Where the upper and lower values on the confidence interval statistics are either both above or both below 1, the connectivity variables are associated with a significant change in odds of the dependent variable. For instance on physically active local travel, a person who lives in a much less direct pedestrian environment relative to that for vehicles (ratio of route directness 1.2) has a much lower odds of being physically active – the odds ratio (OR) is well below 1.0. 19 A person who lives in a neighbourhood with a denser pedestrian network than vehicle network, OR > 1 for increased active walking, is much more likely to have sufficient walking to be considered active. Finally, and perhaps most importantly for a variety of immediate environmental outcomes, the changes in ratio of route directness are linked with decreased vehicle miles of travel. The linear regression model indicates that: •  a 10% increase in the relative directness (to a nearest destination) of a street network for pedestrians (going from 1.0 to 0.9 on the ratio of route directness continuum) is associated with a 23% decrease in distance traveled by automobile, for local travel.  Increasing vehicular route directness relative to pedestrians is also associated with an increase in the total number of trips a person will take (10% change associated with 16.3% increase). Linear regressions indicate that increasing the pedestrian network density relative to that of vehicles relates to increases in both distance walked (+3.1% for a +10% network change) and number of walk trips (+4.6% for the same change in relative network continuity). While the association between route directness and odds of walking, or for other travel outcomes, in the logistic regression model at the person level is weak in one output (modeling walking versus no walking), the regression of active transportation (walking 30 minutes or more each day), and previous correlations, along with the trip-level regression analyses, certainly provide evidence that the odds of a person walking are associated with changes on the connectivity disparity measures. The association of walking and other travel behaviour with 19  A full unit change on these ratio measures (of route directness and network density) would be an unlikely scenario. A change of 1 on the ratio of route directness associates with a change of 1/10 odds ratio (since Exp(B) = .10), but this pattern would be an outlier far outside the interval of data (or, in the case of a retrofit, requiring wholesale neighbourhood street and path reconfiguration). Likewise for ratio of network density, a change of 1 would create an odds ratio of 5.34, but this is outside the interval of available data for the study region.  98  ratio of route directness (relative connectivity) appears in general to be weaker than their relationship with ratio of densities of pedestrian and vehicular travel facilities (relative network density), except for “walking a sufficient amount to be considered active” and “distance traveled by vehicle.” Street pattern associations with travel behaviour: Fused Grid and Levels of Walking The two connectivity measures are significant predictors of increased odds of walking. Based on the Fused Grid’s measurements (see Chapter 5, Table 5-7), the results of the correlation and regression analyses, and previous findings about the importance of connectivity (intersection density), this street design appears to be more walkable than other street networks which lack the Fused Grid’s distinct network pattern – more direct routing and greater density of network for pedestrians relative to motor vehicles. On the measurement of ratio of route directness, the Fused Grid street pattern design represents a change of 10% or more of increased walkability (relative directness or network density for pedestrians). This 10% route directness change corresponds, at the trip level, to an increase of 11.3% in odds of walking. On available length of pedestrian facilities (relative to street segment length) in a given area, the Fused Grid design represents a 20-40% increase over a street network with pedestrian facilities only along full streets in the form of sidewalks. This corresponds to a change of 13.5% to as much as 42.5% greater odds of a person walking for some of their local travel. It should be noted that these results are additive. That is to say, one could expect more shift in travel, such as increased walking levels, by the combined effects of changes to the network pattern which make it both more direct for walking and have increased length or continuity of pedestrian network relative to the vehicle network remains. Interpreting other relationships and interactions As in previous studies, intersection density was significantly correlated with travel behaviour, although it was so highly correlated with, and was an indicator of the same phenomenon as, the ratio of network density, that it was not included in the same multivariate regression model. Other urban form factors (presence of retail within 1km buffer and bus stop density) are more significantly related to the travel behaviour outcomes, and thus would seem to be more influential to broader travel patterns in a neighbourhood. The bus stop density variable, however, showed strong correlation with one or another of the variables of interest in this study (ratio of network density), complicating the ability to interpret it and resulting in it having to be excluded from the same regression models. 99  From the other modeling results (sensitivity testing, etc.), it is clear that there are some interactions that make a difference in the predictiveness of the model. For instance, even in a trip-distance constrained sample (home-based local trips only), the distance to nearest commercial destination appears to interact with the ratio of route directness to explain more of the behaviour involved in local travel. At the person level, inclusion of an interaction term improved the model and brought the ratio of route directness variable result closer to significant (at the 10% significance level) with a similar Beta coefficient to what it had at the trip level. Given this finding, the interaction of distance to destinations and directness of routing deserves further study.  II. Relation to Other Research on Street Networks and Travel This study has confirmed findings from previous research that street design and particularly its connectivity, measured in a variety of ways, matters in travel behaviour (Schlossberg 2006; Dill 2004; Handy 2003; Ewing & Cervero 2001; Engelke, et al. 2000). Ewing & Cervero (2001) note that there is a paucity of conclusive findings about street networks and travel behaviour. Frank, et al. (2000) found that vehicle miles traveled decreases with better connectivity (higher density of census blocks) and more recent studies have supported this and the link of street network design and availability of walking paths to various travel and activity outcomes (Moudon, et al. 2007; Pierce, et al. 2006). This study has extended that research by testing the microscale measurement of patterns of connectivity and continuity, contrasting the distinct networks available for walking and driving modes of travel. The number of trips and the share of those trips by mode is related to the relative connectivity provided to these two modes of transportation. That is to say, the more direct that pedestrian routes are relative to motor vehicles, the greater the odds that walking will be the mode choice, at least for the local travel that is most likely to be influenced by the neighbourhood’s urban form. This is consistent with, and again extends, findings of other Seattle region studies vis-à-vis street connectivity: Moudon et al. (1997) found significant associations of street connectivity and sidewalk density with household walk trip rates. LUTAQH (2005) found a moderate association of intersection density with walking behaviour. A quartile increase in number of intersections per square kilometer was associated with a 14% increase in the odds of walking. A similar result from an earlier report (WSDOT 2005) using a 100  multinomial logit model found an urban form-travel elasticity for walking of +2.8% from a 10% increase in home area intersection density. From the findings of this research, using the more constrained travel of home-based, local trips, one can begin to notice the importance of comparative connectivity of travel networks across modes. As in another recent study using the same dataset (WSDOT 2005), this research found significant association for total vehicle ownership and for several other demographic and household characteristics in the inferential analysis. The research also found significant relationships of neighbourhood retail and intersection density within the vicinity of households (1 km buffer) with walk mode share and other travel outcomes. Neighbourhood destinations, like the closest commercial land use or the number of retail locations in close proximity, are significant factors in local travel – a finding that corroborates previous analysis of this region (Hurvitz 2005; Lee & Moudon 2006; LUTAQH 2005; Moudon, et al. 1997). Net residential density is correlated with walking mode share in the absence of other variables, but it did not remain significant (except in relation to the logistic regression of driving) in a full regression model among other influential variables.  III. Data & Methods, Limitations Data Among the data sources that were important to this research, none has been more critically important than sidewalk and other pedestrian pathway locations and lengths. The detailed digital (GIS) pedestrian environment was only available in three cities of the study region, but more cities are beginning to develop inventories of current pedestrian infrastructure in a measureable format. The availability of this data is crucial for both travel behaviour research and for enhanced planning related to non-motorized transportation. This research has demonstrated an important application of this kind of information, and it is recommended that the various levels of government involved with transportation planning continue to improve the collection and availability of mode-specific network and spatial data. Some other important data to consider for future studies are crossing/intersection quality as well as more detailed micro-scale pedestrian environment factors – topography, streetscape, crossing quality, lighting and other factors affecting personal safety. Many of these features of the built environment have been tested in recent years (Lee & Moudon 2006). More 101  sophisticated modeling of behaviour is possible with the availability of these kinds of data because they can indicate where crossings are possible, or safest, and can more adequately estimate how a pedestrian’s comfort and convenience are affected by various routes. Quality of destinations is another factor in travel behaviour that can be assessed for future research on pedestrian behaviour that relies on origin-destination pairs. Having these kinds of data, again spatially linked for use with GIS analysis, would permit more accurate simulation or control for factors influential to travel behaviour, especially the decision of whether to walk. Methods The travel behaviour analysis framework used in this research, as described in Chapter 3, is consistent with other research on the association of travel with urban form. It builds not only on the substantial base of knowledge about Central Puget Sound urban form and travel behaviour relationships, but also on evidence from across North America. In this regard, it is not new. The measurement of a particular feature of street design is where this project has advanced methods and understanding of urban form. The disparity of travel network connectivity provided to different modes, measured using computed variables of ratios of separately measured modal connectivities (route directness on walking and driving networks), and how it relates to travel, has not been studied previously. This project has developed a means of measuring this phenomenon (ratio of route directness), used a second measure (relative network density) that parallels ones used in other studies (Moudon, et al. 1997; Schlossberg 2006), and employed each in an experiment to test relationships statistically. The ratio of route directness variable proved difficult to analyze in at least one respect: the urban form where the pedestrian network data was available consisted mainly of neighbourhoods with no detectable disparity in connectivity (1:1 ratios). This research complements other connectivity work being conducted – on perception and as a part of the experienced pedestrian environment and other travel behaviour – in urban design research. Dill (2004) and others, with results forthcoming in conjunction with Active Living Research (www.activelivingresearch.org), are investigating how to measure connectivity for bicycling and walking. The present study has advanced methods to a sub-link level – assessing sidewalk at increments of 100’ on the street network. This micro-scale of measured pedestrian network environments is something called for by reviewers of urban form-travel behaviour research (Ewing & Cervero 2001). It contributes to the development of methods and evidence that will in turn inform built environment policy development and aid in the setting of street 102  standards used for residential neighbourhoods. It also provides a foundation for continuing research on street networks (discussed in Chapter 8, Further Research), helping to answer questions concerning broader network patterns and how they might support public policy aimed at improving public health, transportation efficiency, and quality of life through reduced automobile dependence. Limitations As noted previously, this study is cross-sectional in design, providing an understanding of relationships between travel and urban form during one discrete period of time in this Seattle metropolitan region study area. This is useful for predicting some likely responses in behaviour and associated outcomes, but not proving causes of behaviour or attributing causation to the built environment factors. The results therefore should be viewed as indications of more general patterns in behaviour that are probable or could be expected in other locations. Further, some of the models’ diagnostics, notably the Hosmer-Lemeshow statistic on the trip-level logistic regression and P-P plots from linear regressions of distance walking and walk trips, indicate somewhat poor model fit, limiting the generalizability of these results. As well, the percent of correctly predicted cases of walking, while improved by the connectivity/continuity variables, remains low (see Technical Appendix). More accurate than the usual SPSS output of adjusted Rsquare, the Stein’s adjusted R-square (.113) for the Linear Regression model of vehicle miles of travel (distance driven) indicates that one could expect a loss of explanatory power when going from the sample (r-square =.132) to the population. A different study design would be needed to attempt to overcome such confounding factors as self-selection, and replication of this study in other regions would be necessary to validate theories for these relationships. The study made assumptions about travel when constructing the built environment measures, and this adds uncertainty to the analytical results. The main assumption of measuring route directness, that pedestrians would be able to use local access streets whether or not they have sidewalks, may reduce variability in the sample from what the real street networks are like in people’s experience of them. Sidewalk continuity can be an important element affecting the route a person walks, or whether they walk at all. The relative network density measure was a proxy for this aspect of urban form – person interaction, but a more direct measurement of continuity along actual routes would be more accurate (as was possible to do on larger streets in the ratio of route directness measurement). Furthermore, destinations, though the study attempted to model accurately and consistently where they were accessed from the street 103  networks and therefore accurately captured physical distances, were treated as equally attractive – which may or may not reduce the accuracy of the route directness measurement as an indication of connectivity. Each of these limitations was partially addressed through sensitivity testing (using an impeded pedestrian network which yielded results similar to using a different variable) and the inclusion of related variables (pedestrian/vehicular network density measure as a proxy of continuity and neighbourhood retail proximity factors) in the model. Due to the resource-intensive nature of collecting and refining data for use in this kind of analysis, the study also could not control for many additional factors that were potentially influential, as noted above. Personal attitudes, perception of environmental conditions and safety, and the details of routes’ suitability for various forms of transportation can be highly influential to travel decision making but were not collected on the individuals who participated in the travel survey in 1999 nor the routes they took. Other measurable aspects of the street environment – accessibility of design, streetscape, crossing quality - were not measured, and would indeed have been very difficult to reconstruct for the time period of the travel survey. Likewise site design can be a factor in how people travel. Thus there are several areas where greater control for built environment factors would be desirable (more details in Chapter 3) and may yet lead to better explanatory models and higher confidence in the results relating to street networks. Acquisition of enhanced data on these other factors should be part of future analysis of street pattern relationships with travel behaviour. Finally, the study did not find significant results for the ratio route directness to parks, which is a result consistent with recent findings in the same region (Berke, et al. 2007; Lee & Moudon 2006). The logistic regression model also did not explain non-work travel better than all travel including work-related tours. This may be due in part to the fact that the travel survey only gathered data during two weekdays, rather than including weekend days when a substantial portion of recreation-purpose and non-work travel, including much walk trip travel, occurs (Bhat & Lockwood 2004). Future research on the topic would therefore benefit from travel and activity survey data that captures more of the non-work travel that constitutes a majority of all trips in North America.  104  Chapter 8. CONCLUSION  105  I. Overview In conclusion, this study provides an assessment of residential street design vis-à-vis travel behaviour. It was prompted by the need for evidence as to likely outcomes from application of the Fused Grid, a street design that attempts to find a balance between the benefits of street connectivity for transportation efficiency with the use of street space to enhance neighbourhood quality of life for local residents. Residential street design is one element of ensuring healthier communities by design, and a debate continues in contemporary urban planning over whether connected streets such as traditional gridiron patterns or closed streets such as loops-and-culs-de-sac provide greater benefits locally and across a region. While the effects of various street designs have been or are being modeled for their effects on traffic flow and safety (IBI Group 2007; Lovegrove & Sayed 2006), this research has focused on empirical investigation of the outcomes, in terms of walking and driving behaviour, from varying levels and disparities of travel network connectivity and continuity by travel mode. The central hypothesis investigated here is that street network designs which enhance connectivity for pedestrians relative to vehicles, by including closures or diverters on the motor vehicle network (streets) or providing separate, more direct paths for pedestrians, result in increased share of neighbourhood travel by walking. Such changes to street design, like requests for traffic calming and standards for narrower streets to slow traffic, have been increasingly demanded in recent years by both local residents and nationally known experts in walkability (Burden 2000). Yet solid evidence, based in empirical scientific research, has been difficult to attain. This assessment begins to answer the question of whether new designs and standards, in addition to the retrofit programs of new pedestrian connections and traffic management, will result in the desired outcome of increased levels of walking and decreased reliance on automobiles. The study has demonstrated that there is a relationship between the urban form of local street networks and travel behaviour, and that modifications to street patterns which affect relative utility of different modes are associated with changes in levels of walking and driving for local travel.  II. Key Findings The two main network measures in this study, relative route directness and relative network density across walking and driving modes, have an association with odds of walking, 106  odds of driving, distance walked, distance traveled by vehicle, and number of trips. A ratio of route directness indicating a street pattern more direct for walking than for vehicles is associated with increased levels of walking and decreased driving. Likewise, where there is a higher density of walking facilities than driveable streets (ratio of sidewalks and trails to street segments) in a given residential area, increased walking would be expected. The implications of this primary finding are that street design that serves to increase the relative utility of walking by improving the directness of routes, relative to those enjoyed by other modes on their respective networks, associates with a shift in travel mode choice to more walking for trips in the home neighbourhood. The Fused Grid street design provides a transportation network pattern that achieves this kind of change in the relative utility of walking. The same would likely be true of certain traffic management interventions (i.e. creating pedestrian only malls within an existing street network, or other partial/full closures and diverters for motor vehicle traffic). A corresponding reduction in driving behaviour, at least for local travel, is also a likely outcome. The relative directness of routing across modes appears to matter more for local travel than route directness for one mode alone. This makes sense intuitively – as this research has hypothesized, greater route directness for one mode relative to another will enhance its utility relative to the other mode, thus encouraging a corresponding shift in travel behaviour to the more convenient or useful means of transportation. This appears to matter more for nearby commercial destinations than for nearby parks. Relative network density, a measure of continuity, exhibits a stronger relationship than relative route directness to travel behaviour correlated or regressed at the person level. This suggests that adding more pathways for pedestrians, whether or not they improve connectivity to destinations, increases levels of walking. The Fused Grid street design is characterized by a higher density of pedestrian pathways than vehicle pathways. Other factors held constant, this kind of street network would be associated with more walking and less driving. It bears repeating that these two factors, ratio of route directness and ratio of network density between the two modes, as modeled here are additive in their association, so even larger changes on the travel outcomes would be expected from designs or projects that create relative improvement for a particular travel mode on both network connectivity and continuity. The strength of association between the measure of relative modal network connectivity and travel behaviour appears to be somewhat less (smaller correlation coefficient) than the 107  relative density of facilities in a neighbourhood as well as for the sheer presence of nearby retail stores and total vehicle ownership. The regression model demonstrated that, in this study region, a change in a residential area from a pure small-block grid to a modified grid (with potentially the same intersection density but ratio of route directness around .90) providing more direct pedestrian connections, i.e. the characteristics of a Fused Grid, can result in an increase in odds of a trip being walked for home-based travel of 11.3%. This street pattern configuration is also associated with reduced vehicle travel and a decrease in overall number of trips. Similarly, when total sidewalk length relative to the length of street segments (relative network density) increases by 10% in the area around a household, odds of a person traveling by pedestrian mode increases by 9.3%. Both variables are associated with physical activity, with 10% change on the former associated with 25.9% increase in odds of walking a sufficient distance to be considered physically active, and on the latter, 18.2%.  III. Recommendations for Community Planning Policy & Practice This research was conducted in response to support and encouragement by the Canada Mortgage and Housing Corporation, a federal agency that works to create better community design by assisting local governments in Canada with their community development. There is interest in how various street designs, and the Fused Grid in particular, might perform on key transportation system outcomes such as walking mode share and vehicle miles traveled. Based on the results, this research project offers a set of policy and practice recommendations for community design and planning. Community design implications Both traffic management and street design standards that offer increased directness of routing for pedestrians relative to motor vehicles can be used to help achieve increased levels of walking, reduced motor vehicle use, or both. a) Increasing the extent of sidewalk or other pedestrian pathways, by adding them as stand-alone projects or including a higher density of them relative to street length in plans for new neighbourhoods, is likely to be useful in achieving the same walkability outcomes.  108  b) If promotion of neighbourhood walking for healthy physical activity is a primary goal, increasing the pedestrian network density relative to motor vehicles is the street environment change more strongly associated with walking the 30 minutes per day recommended for long-term health though improving relative pedestrian route directness would also likely provide some benefit. c) Creating designs and projects that result in both relative reduction in connectivity of the motor vehicle network and relative expansion of the walking network, will be even more effective in achieving these same outcomes and possibly reducing overall travel demand. While both land use densities and mix of uses are influential factors indicating proximity of destinations, it may be harder for policy to change these aspects of urban form, especially in the short term (LUTAQH 2005). Connectivity of streets is a feature of urban environments that falls more clearly within the purview of local governments to direct design and intervene with projects. This project has shown that there are potential benefits in public health and transportation to be gained from activities that modify street connectivity either through design guidelines for new development or retrofitting neighbourhood street networks with traffic management or enhanced pedestrian facilities. These results should not be construed as an endorsement of reduced street connectivity for motor vehicles without corresponding improvements to the pedestrian network. Such efforts to disconnect vehicle access may have unintended consequences such as longer distances per driving trip, and thus could counteract the benefits of reductions in vehicle miles of travel from the mode choice shift. In fact, in many situations, the lack of infrastructure to support pedestrians is a strong enough barrier that no mode shift is likely to occur if only the vehicular network is modified. Changes in street standards deserve careful attention to the broader systemwide effects that may result. Other resources Local governments seeking further information about street network design and connectivity can refer to resources at the Canada Mortgage and Housing Corporation (CMHC)20  20  CMHC assists local Canadian communities with design, planning and development. The Fused Grid street design was developed by this agency. See http://www.cmhc-schl.gc.ca/en/index.html and search for ‘Fused Grid’ or ‘street design’.  109  (in Canada) or various national (Local Government Commission21; American Planning Association22) or regional/state (Climate Solutions23 and Municipal Services and Research Center24 in Washington) level organizations in the United States.  IV. Further Research Continuing study of street networks is needed to validate these results, confirm their generalizability, and extend the empirical knowledge base regarding effective transportation system design. The influence of street design on travel behaviour is an area of research with great potential for policy and practice change implications. Empirical evidence of the impacts of various urban form features on desired outcomes is often necessary in order to demonstrate the desirability of change in practices, standards and investments in infrastructure. To this end, this research project has identified some key areas for continued research. Fused Grid and Street Networks, further assessment Efforts are under way to model the traffic implications of the Fused Grid – how well it will function for the movement of vehicles. This effort would be enhanced with the addition of modeling of pedestrian or bicycle movement on Fused Grid networks, perhaps using the improving techniques of space syntax or microsimulation to estimate demand. Regional travel implications from introduction of the Fused Grid to residential neighbourhood areas were not the focus of this study, but could be investigated by aggregating travel data at a larger scale. The enhancements noted below could also be applicable to continued study of the Fused Grid and other street networks. Enhanced urban form analysis Enhanced analysis of urban form is needed to account for other factors that are suspected and emerging as covariates with walking and overall travel behaviour. Several of these other factors were not possible to include in this assessment due to time constraints or the difficulty of ensuring accuracy for a time period that is now quite remote from the present (travel behaviour 21  LGC provides resources for livable community design. See http://www.lgc.org/community_design/street.html for resources on street design. 22 APA publishes guidance for local planners, including a report on street connectivity, by Susan Handy and coauthors, referenced in this research. See: http://www.planning.org/APAStore/Search/Default.aspx?p=2426. 23 Climate Solutions provides tools for local transportation planning around non-motorized modes. See: www.climatesolutions.org. 24 MSRC is a clearinghouse of information for Washington State’s local governments. See http://www.mrsc.org/Subjects/Transpo/efficientlanduse.aspx#street for more on street connectivity.  110  data was collected nearly seven years before the data development for this research began). Here is a partial listing of the kinds of data that would be required for an enhanced analysis using the same methodological framework: •  Quality of destinations –attractiveness of a park or commercial location and thus its tendency to generate travel. If destinations could be ranked on a scale of travel attractiveness, using a gravity model or other technique, the route directness measurement of connectivity would be enhanced.  •  More detailed qualitative analysis to build variables relating to streetscape and microscale pedestrian environment quality – as was done in a related UBC School of Community and Regional Planning research project in relation to school walking (Niece 2006) – measuring sidewalk width or surface quality, presence of street trees, vegetated buffers, lighting, or furniture.  •  Other key factors in pedestrian accessibility include slope (topography or whether stairs are part of pedestrian routes) and quality of crossings.  Each of these factors affects comfort and to some extent safety of the pedestrian. Of course, the study would also have benefited from having more than three cities’ pedestrian network data by which a wider variety of street networks could be measured on the key variables. With further updating of the respective modal networks, bringing them up to the current year and maintaining their accuracy as new pedestrian facilities or street improvements are constructed, the same urban form data would be available for analysis of relationships to behavioural data that has been collected since 1999. For example, in the past few years, the U.S. National Institutes of Health have funded surveying of physical activity behaviour in King County, Washington. Enhanced analysis using fuller travel dataset Various improvements to the research could be made with a more complete and thorough measurement of travel behaviour, including micro-scale details of the pedestrian environment. What follows is a listing of the possible enhancements in data: •  The present analysis also cannot account for perceptual (such as reported feeling of safety or security) or attitudinal factors. This sort of information would need to be gathered through surveys of the persons whose behaviour is being analyzed.  •  Bicycle facilities and travel -A paucity of data on bicycling facilities and travel did not allow this study to test the implications of network disparities for this transport mode, but 111  data sufficient for this kind of analysis is likely to be more available in more current municipal data and travel surveys as better GIS facility inventories are developed and as use of bicycles for transportation increases. •  Data on weekend travel behaviour would lead to a more robust assessment and conclusions about non-work or recreational travel, which may allow better analysis of another measure of route directness (to parks) or reveal new relationships of the variables that have already shown significant associations in this research. This would also allow further investigation on differences between work-related and non-work purposes in home-based, short-distance travel.  Longitudinal research In a broader sense, this study cannot, given its use of a discrete travel database from the past and cross-sectional research design, attribute causality or provide additional explanation that would overcome self-selection factors or the possibility that these connectivity factors are overwhelmed by others in the remaining seventy percent or more of the basis for travel behaviour that the models do not explain. This could be done by following a cohort of people living in a range of different street patterns, as was the case for the present cross-sectional study, and looking in detail at the pattern of their travel behaviour and attitudes. Even better, however, would be to compare the travel before and after of persons who recently moved to a Fused Grid neighbourhood from a variety of different street network types (and vice versa), which could offer a more definitive assessment of the travel implications of the Fused Grid  V. Summary Based on the results of this study, when demographic characteristics and other built environment conditions remain equal, residential street patterns that include improved pedestrian network connectivity or density relative to the network provided to motor vehicles would be expected to result in gains in walking mode share for local, home-based travel. The Fused Grid street network design exhibits these kinds of travel network connectivity and continuity patterns, and thus its application for residential neighbourhoods would likely be associated with higher levels of walking and reduced motor vehicle traffic. As well, this study indicates that transportation projects for retrofit situations, such as traffic calming diverters or street closures, that interrupt or make motor vehicle travel less-direct relative to walking travel, would likely 112  produce similar results. While further research is needed to confirm the generalizability of these results, there is significant evidence from this research that local streets can influence local travel patterns and that particular patterns of network connectivity disparity across modes of travel are associated with increases in travel by the favored mode.  113  BIBLIOGRAPHY Allen, Eliot. 2006. INDEX: software for community indicators. http://www.crit.com/documents/ESRIChapter.pdf (accessed 21 November 2006). Appleyard, Donald. 1981. Livable streets. Berkeley, CA: University of California Press. Arnold, Chester L., Jr. & C. James Gibbons. 1996. Impervious surface coverage: the emergence of a key environmental indicator. Journal of the American Planning Association. 62, 2: 243-258. Babbie, Earl R. 2001. Practice of social research. 9th Ed. Belmont, CA : Wadsworth Thomson Learning. Berke, E.M., T. D. Koepsell, A.V. Moudon, R.E. Hoskins, and E.B. Larson. 2007. Association of the built environment with physical activity and obesity in older persons. American Journal of Public Health. 97, 3: 486-492. Bhat, Chandra R. & Allison Lockwood. 2004. On distinguishing between physically active and physically passive episodes and between travel and activity episodes: an analysis of weekend recreational participation in the San Francisco Bay area. Paper # 150-03. http://www.ce.utexas.edu/prof/bhat/ABSTRACTS/Bhat_Lockwood_150-03_FINAL.doc (accessed 25 March 2007). Boarnet, Marlon & Randall Crane. 2001. The influence of land use on travel behavior: specification and estimation strategies. Transportation Research Part A., Policy and Practice, 35, 9: 823-845. Also summarized in the authors’ book, Travel by design: the influence of urban form on travel. New York: Oxford University Press. Boarnet, Marlon G. & Michael J. Greenwald, 1999. Land use, urban design, and non-work travel: reproducing for Portland, Oregon, empirical tests from other urban areas. Irvine, CA: U.C. Irvine Center for Activity Systems Analysis. http://repositories.cdlib.org/cgi/viewcontent.cgi?article=1034&context=itsirvine/casa (accessed 23 March 2007). Burden, Dan. 1999. Street design guidelines for healthy neighborhoods. Sacramento, CA: Local Government Commission. Burden, Dan. 2000. Streets and sidewalks, people and cars: the citizen’s guide to traffic calming. Sacramento, CA: Local Government Commission. Burden, Dan & Peter Lagerway. 1999. Road diets: fixing the big roads. Walkable Communities, Inc. http://www.walkable.org/download/rdiets.pdf (accessed 5 March 2006). Burden, Dan & Paul Zykofsky. 2000. Emergency response: traffic calming and traditional neighborhood streets. Sacramento, CA: Local Government Commission. Calthorpe, Peter. (1993). The next American metropolis: ecology, community and the American dream. New York: Princeton Architectural Press. Cervero, Robert. 1994. Transit-based housing in California: evidence on ridership impacts. Transportation Policy. 1, 3: 174-183. Described in Ewing & Cervero 2001.  114  Cervero, Robert & Michael Duncan. 2006. Which reduces vehicle travel more: jobs-housing balance or retail-housing mixing. Journal of the American Planning Association. 72, 4: 475-490. Cervero, Robert & Kockelman, Kara. 1997. Travel demand and the 3Ds: density, diversity, and design. Transportation Research Part D. 2, 3: 199-219. Chellman, Chester E. (Institute of Traffic Engineers). 1999. Traditional neighborhood development: street design guidelines. Washington, D.C.: ITE. City of Olympia. 2005. Street standards review: process to improve the city’s street standards. http://www.ci.olympia.wa.us/council/packets/pdf/20060123/AGENDA.122306.pdf (accessed 10  March 2007). Crane, Randall. 1996. On form versus function – will the new urbanism reduce traffic, or increase it? Journal of Planning Education and Research. 15, 2: 117-126. Crane, Randall. 2000. The influence of urban form on travel: an interpretive review. Journal of Planning Literature. 15, 1: 4-23. Crane, Randall & Richard Crepeau. 1998. Does neighborhood design influence travel? a behavioral analysis of travel diary and GIS data. Transportation Research Part D. 3, 4: 225-238. CTED, (1993). Washington state energy use profile: transportation sector. http://www.cted.wa.gov/energy/archive/FILES/PRFL/docs/key09/k09page.htm (accessed 4 June 2007). Daisa, J.M. & J.B. Peers 1997. Narrow residential streets: do they really slow down speeds? ITE conference paper. http://www.ite.org/traffic/documents/AHA97F46.pdf (accessed 6 February 2006). Dannenberg, A.L., R.J. Jackson, H. Frumkin, R.A. Schieber, M. Pratt, C. Kochtitsky, H.H. Tilson. 2003. The impact of community design and land use choices on public health: a research agenda. American Journal of Public Health. 93, 9: 1500-1508. Dill, Jennifer. 2004. Summary of research on bicycle and pedestrian connectivity. http://web.pdx.edu/~jdill/research.htm#Connectivity (accessed 3 February 2006). Duany, A., E. Plater-Zyberk, and J. Speck. 2000. Suburban nation: the rise of sprawl and the decline of the American dream. New York: North Point Press. Dumbaugh, Eric. 2005. Safe streets, livable streets. Journal of the American Planning Association. 71, 3: 283-300. Engelke, P.O., L.D. Frank, and J. Gauthier. 2000. An analysis of non-motorized travel and short trips in the Seattle area. Final report to Centers for Disease Control and Prevention. School of Community & Regional Planning (UBC) Active Transportation Laboratory collection (accessed 15 October 2005). Ewing, Reid. 1996. Appendix C. in Pedestrian- and transit-friendly design. Florida Department of Transportation. Described in Ewing & Cervero 2001. Ewing, Reid. 1999. Traffic calming: State of the practice. ITE/FHWA. http://www.ite.org/traffic/tcstate.htm#tcsop (accessed 23 March 2007).  115  Ewing, Reid & Robert Cervero. 2001. Travel and the built environment: a synthesis. Transportation Research Record. 1780: 87-122. Washington, D.C.: TRB, National Research Council. Ewing, Reid, M. DeAnna, C.C. Heflin, and D.R. Porter. 1996. Best development practices: doing the right thing and making money at the same time. Published in cooperation with the Urban Land Institute and the Florida Department of Community Affairs. Chicago: Planners Press (American Planning Association). Fitzpatrick, K. & Carlson, P. 2001. Design factors that affect driver speed on suburban streets. Paper presented at the TRB 80th Annual Meeting, Washington, DC: Transportation Research Board. Frank, Lawrence D. 2000. Land use and transportation interaction: implications on public health and quality of life. Journal of Planning Education and Research. 20: 6-22. Frank, Lawrence D. & Peter O. Engelke. 2001. The built environment and human activity patterns: exploring the impacts of urban form on public health. Journal of Planning Literature. 16: 202-18. Frank, L.D., P.O. Engelke, and T.L. Schmid. 2003. Health and community design: the impacts of the built environment on physical activity. Washington, D.C.: Island Press. Frank, Lawrence D. & Gary Pivo. 1995. Impacts of mixed use and density on utilization of three modes of travel: single-occupant vehicle, transit and walking. Transportation Research Record 1466: 44-52. Washington, D.C.: TRB, National Research Council. Frank, L.D., J.F. Sallis, B.E. Saelens, L. Leery, K. Cain, T. Conway, and P.E. Hess. 2004. A walkability index and its application to the trans-disciplinary neighborhood quality of life study. Paper submitted to Journal of the American Planning Association. Part of course packet for School of Community & Regional Planning (UBC) Course PLAN 580 (accessed 17 Oct. 2004). Frank, L.D., J. F. Sallis, T. L. Conway, J. E. Chapman, B. E. Saelens, and W. Bachman. 2006. Many pathways from land use to health: associations between neighborhood walkability and active transportation, body mass index, and air quality. Journal of the American Planning Association. 72, 1: 75-87. Frank, L.D., T.L. Schmid, J.F. Sallis, J. Chapman, and B.E. Saelens. 2005. Linking objectively measured physical activity with objectively measured urban form. American Journal of Preventive Medicine. 28(2S2): 117-125. Frank, L.D., B. Stone, Jr., and W. Bachmann. 2000. Linking land use with household vehicle emissions in the central Puget Sound. Transportation Research Part D. 5: 173-196. Friedman, B., S.P. Gordon, and J.B. Peers. 1994. Effect of neotraditional design on travel characteristics. Transportation Research Record 1466: 63-70. Washington, D.C.: TRB, National Research Council. Frumkin, H., L.D. Frank, and R. Jackson. 2004. Urban sprawl and public health. Washington, D.C.: Island Press. Garreau, Joel. 1991. Edge city: life on the new frontier. New York: Doubleday.  116  Gauthier, John. 1999. A procedure for the calculation of an index of route directness from geocoded travel survey data. Georgia Institute of Technology unpublished research paper. From School of Community & Regional Planning (UBC) Active Transportation Laboratory collection (accessed 10 February 2005). Gehl, Jan. 1980. Life between buildings: using public space. London: Van Nostrand Reinhold. Giles-Corti, Billie. 2006. People or places: what should be the target? Journal of Science & Medicine in Sports, 9: 357-366. Girling, Cynthia & Kenneth I. Helphand. 1994. Yard, street, park: the design of suburban open space. New York: John Wiley and Sons. Girling, Cynthia & Ronald Kellet. 2005. Skinny streets and green neighborhoods. Washington, DC: Island Press. Goddard, Stephen B. 1994. Getting there: the epic struggle between road and rail in the American century. Chicago: University of Chicago Press. Grammenos, Fanis & Chris Pidgeon. 2004. Fused grid planning in a Canadian city. Wharton Real Estate Review. 5: 17-25. Grammenos, F., S. Pogharian and J. Tasker-Brown. 2005. Residential street pattern design: a proposal. Revised. Ottawa: Canada Mortgage and Housing Corporation. Grammenos, Fanis & Douglas Pollard. 2005. Reevaluating the grid. http://www.greekworks.com/content/index.php/weblog/extended/reevaluating_the_grid/ (accessed 6 November 2005). Handy, Susan. 1993. Regional versus local accessibility: Implications for nonwork travel. Transportation Research Record 1400: 58-66. Washington, D.C.: TRB, National Research Council. Handy, S., R.G. Paterson and K. Butler. 2003. Planning for street connectivity: getting from here to there. Planning Advisory Service Report 515, American Planning Association. Hess, P.M., A.V. Moudon, and M.G. Logsdon. 2001. Measuring land use patterns for transportation research. Transportation Research Record. 1780: 17-24. Washington, D.C.: TRB, National Research Council. Hillier, B. A. Penn, J. Hanson, T. Grajewski, and J. Xu. 1993. Natural movement: or, configuration and attraction in urban pedestrian movement. Environment and Planning B: Planning and Design. 20: 29-66. Holtzclaw, John. 1994. Using residential patterns and transit to decrease auto dependence and costs. San Francisco: Natural Resources Defense Council. Holtzclaw, J., R. Clear, H. Dittmar, D. Goldstein and P. Haas. 2002. Location efficiency: neighborhood and socio-economic characteristics determine auto ownership and use - studies in Chicago, Los Angeles and San Francisco. Transportation Planning and Technology. 25, 1: 1–27.  117  Hurvitz, Phillip M. 2006. The Walkable-Bikeable Communities Analyst extension for ArcView GIS 3.x. http://gis.esri.com/library/userconf/proc05/papers/pap1040.pdf (accessed 16 January 2007). IBI Group. 2007. Assessment of the transportation impacts of current and Fused Grid layouts. Executive summary for Canada Mortgage and Housing Corporation. Ottawa, Ontario (sent by F. Grammenos of CMHC via email to researcher, 9 September 2007). ISTEA. 1991. Intermodal Surface Transportation Efficiency Act. U.S. Congressional Record, Public Law 102-240, 105 Stat. 1914 (1991). http://thomas.loc.gov/cgi-bin/bdquery/z?d102:HR02950: (accessed 11 March 2006). Jackson, Kenneth T. 1985. Crabgrass frontier: the suburbanization of the United States. New York: Oxford University Press. Jacobs, Jane. 1961. The death and life of great American cities. New York: Vintage Books. King, Michael. 2003. Pedestrian safety through a raised median and redesigned intersections. Paper presented at the TRB 82nd Annual Meeting. Washington, DC: Transportation Research Board. Kitamura, R., P.L. Mokhtarian, and L. Laidet. 1997. A micro-analysis of land use and travel in five neighborhoods in the San Francisco Bay Area. Transportation 24: 125–158 Krizek, Kevin J. 2003. Residential relocation and changes in urban travel: does neighborhood-scale urban form matter? Journal of the American Planning Association. 69, 3: 265-281. Kulash, Walter. (1991). Neotraditional town design: will the traffic work? Presentation to Eighth International Pedestrian Conference, Bellevue, WA. Kulash, Walter. 2001. Residential streets. Washington, DC: Urban Land Institute. Kunstler, James H. 1993. Geography of nowhere: the rise and decline of America's man-made landscape. New York: Simon and Schuster. Leaf, W.A. & D.F. Preusser, 1999. Literature review on vehicle travel speeds and pedestrian injuries. Washington, DC: NHTSA, US DOT Final Report 809 021. Lee, Chanam, & Anne Vernez Moudon. 2006. Correlates of walking for transportation or recreation purposes. Journal of Physical Activity and Health, 3, Suppl 1, S77-S98. Lovegrove, Gordon R. & Tarek Sayed. 2006. Macro-level collision prediction models for evaluating neighbourhood traffic safety. Canadian Journal of Civil Engineering. 33 (5): 609 – 621. LUTAQH. 2005. Travel behavior, environmental, and health impacts of community design & transportation investment: Final report. King County, prepared by LFC as part of the Land Use, Transportation, Air Quality and Health Study. http://www.metrokc.gov/healthscape/publications/LUTAQH_final_report.pdf (accessed 5 November 2006). Marshall, Stephen. 2005. Streets and patterns. London: Spon Press. McCormack, E., G.S. Rutherford, and M.G. Wilkinson. 2001. Travel impacts of mixed land use neighborhoods in Seattle, Washington. Transportation Research Record. 1780: 87-122. Washington, D.C.: TRB, National Research Council.  118  Mertler, Craig A. & Rachel A. Vannatta. 2005. Advanced and multivariate statistical methods: practical application and interpretation. 3rd Edition. Glendale, CA: Pyrczak Publishing. Metro Regional Services. 1997. Creating livable streets: street design guidelines for 2040. Portland, OR: Metro Regional Services. Moudon, Anne Vernez. 1987. Editor, Public streets for public use. New York: Van Nostrand Reinhold. Moudon, A.V., P. Hess, M. Snyder, and K. Stanilov. 1997. Effects of site design on pedestrian travel in mixed-use, medium-density environments. Transportation Research Record. 1578: 48-55. Washington, D.C.: TRB, National Research Council. Moudon, Anne Vernez & Channam Lee. 2003. Walking and bicycling: an evaluation of environmental audit instruments. American Journal of Health Promotion. 18, 1: 221-237. Moudon, A.V., C. Lee, A.D. Cheadle, C. Garvin, D.B. Johnson, T.L. Schmid, R.D. Weathers. 2007. Attributes of environments supporting walking. American Journal of Health Promotion. 21, 5: 448-459. Moudon, Anne V. & Richard K. Untermann. 1987. Grids revisited. In A.V. Moudon (Ed.), Public streets for public use. Ed. by Anne Vernez Moudon. New York: Van Nostrand Reinhold Co. Muller, Peter O. 1995. “Transportation and urban form: stages in the spatial evolution of the American metropolis.” In, S. Hanson (Ed.), The Geography of urban transportation. New York: Guilford Press. Newman, Oscar. 1972. Defensible space: crime prevention through urban design. New York: MacMillan Company. Newman, Oscar. 1996. Creating defensible space. Washington D.C.: U.S. Dept. of Housing and Urban Development. Newman, Peter W.G. and Jeffrey R. Kentworthy. (1989). Gasoline consumption and cities: a comparison of U.S. cities with a global survey. Journal of the American Planning Association. 55, 1: 24.37. Niece, Jennifer L. 2006. Children’s travel to school: the influence of built form and perceptions of safety. Vancouver, BC: Masters thesis, School of Community and Regional Planning, University of British Columbia. NuStats (PSRC). 1999. Puget Sound Household travel survey. Conducted for Puget Sound Regional Council, July – November, 1999 by NuStats Research and Consulting. 1000 Friends of Oregon (Parsons Brinckerhoff Quade and Douglas, Inc.). 1993. The pedestrian environment: making the land use transportation air quality connection. Volume 4a. From School of Community & Regional Planning (UBC) Active Transportation Laboratory collection (accessed 18 March 2005). Ormsby, Tim & Jonell Alvi. 1999. Extending ArcView GIS. Redlands, CA: ESRI Press. Osborne, Jason W. & Amy Overbay. 2004. The power of outliers (and why researchers should always check for them). Practical Assessment, Research & Evaluation, 9(6).  119  Penn, A., B. Hillier, D. Banister, and J. Xu. 1998. Configurational modeling of urban movement networks. Environment and Planning B: Planning and Design. 25: 59-84. Pierce, J.R., Jr., A.V. Denison, A.A. Arif, and J.E. Rohrer. 2006. Living near a trail is associated with increased odds of walking among patients using community clinics. Journal of Community Health. 31, 4: 289-302. Pushkar A. O., B. J. Hollingworth, and E. J. Miller. 2000. A multivariate regression model for estimating greenhouse gas emissions from alternative neighborhood designs. Presented at 79th Annual Meeting of the Transportation Research Board, Washington, D.C. Cited in Ewing & Cervero 2001. Raford, Noah, & David R. Ragland. 2005. Pedestrian volume modeling for traffic safety and exposure analysis: the case of Boston, Massachusetts. TRB 2006 Annual Meeting. http://repositories.cdlib.org/cgi/viewcontent.cgi?article=1028&context=its/tsc (accessed 11 November 2007). Rajamani, J., C.R. Bhat, S. Handy, G. Knapp, and Y. Song. 2003. Assessing the impact of urban form measures in nonwork trip mode choice after controlling for demographic and level-of-service effects. Transportation Research Record. http://www.ltrc.lsu.edu/TRB_82/TRB2003001392.pdf (accessed 3 March 2006). Randall, Todd A. 2002. Decision support for suburban retrofitting. PhD. Thesis, Department of Civil Engineering, McMaster University, Hamilton, Ontario. Randall, Todd A. & Brian W. Baetz. 2001. Evaluating pedestrian connectivity for suburban sustainability. Journal of Urban Planning and Development 127, 1: 1-15. Robinson, Charles M. 1916. City planning, with special reference to the planning of streets and lots. New York: G.P. Putnam’s Sons. Saelens, B., J. Sallis, and L. Frank. 2003. Environmental correlates of walking and cycling: findings from the transportation and urban design and planning literatures. Annals of Behavioral Medicine. 25: 80-91. Sallis, James F., L.D. Frank, B.E. Saelens, and M. K. Kraft. 2004. Active transportation and physical activity: opportunities for collaboration on transportation and public health research. Transportation Research Part A 38: 249–268. Schlossberg, Marc. 2006. From TIGER to audit instruments: measuring neighborhood walkability with street data based on Geographic Information Systems. Transportation Research Record. 1982: 48-56. Washington, D.C.: TRB, National Research Council. Schlossberg, Marc & Nathaniel Brown. 2004. Comparing transit oriented development sites by walkability indicators. Transportation Research Record. 1887: 34-42. Washington, D.C.: TRB, National Research Council. Solnit, Rebecca. 2000. Wanderlust: a history of walking. New York: Penguin Group. Southworth, Michael & Eran Ben-Joseph. 1997. Streets and the shaping of towns and cities. San Francisco, CA: McGraw-Hill.  120  Southworth, Michael & Eran Ben-Joseph. 2004. Reconsidering the cul-de-sac. Wharton Real Estate Review. 5: 5-16. Southworth, Michael & Peter M. Owens. 1993. The evolving metropolis: studies of community, neighborhood, and street form. Journal of the American Planning Association. 59, 3: 271-287. Stokols D. 1992. Establishing and maintaining healthy environment: toward a social ecology of health promotion. American Psychologist. 47, 1: 6-22. Tresridder, Mike. 2005. Using GIS to measure connectivity: an exploration of issues. Field area paper, School of Urban Studies and Planning, Portland State University. http://web.pdx.edu/~jdill/Tresidder_Using_GIS_to_Measure_Connectivity.pdf (accessed 8 March 2007). Twaddell, Hannah. 2005, Spring. Making the connection. Planning Commissioners Journal. 58: 1-2. Untermann, Richard K. 1984. Accommodating the pedestrian: adapting towns and neighborhoods for walking and bicycling. New York: Van Nostrand Reinhold Co. Untermann, Richard K. 1987. Changing street design standards for streets and roads. In A.V. Moudon (Ed.), Public streets for public use. New York: Van Nostrand Reinhold Co. Whyte, William H. 1980. Social life of small urban spaces. Washington, D.C.: The Conservation Foundation. WSDOT (Washington State Department of Transportation). 2005. Travel behavior, emissions, & land use: correlation analysis in the central Puget Sound. WA-RD 625.1 Final Research Report by LFC (Frank, Chapman, Bradley and Lawton). Zhang, Ming & Ching Yi. 2006. Cul-de-sac vs. grid: Comparing street connectivity and pedestrian accessibility of urban forms in the Houston metropolitan area. Presented to the Transportation Research Board, Washington, D.C.: TRB, National Research Council.  121  APPENDICES Appendix A. Technical – Research Design, Methods, Data and Analysis Data Development Procedures (referred to by thesis Chapter 4)  1. Methodology Researchers Chris Hawkins and Lawrence Frank recognized early in the process of developing the methodology for the Fused Grid Assessment project, which was at the centre of the research for this thesis, that existing cross-sectional travel data would be the kind of data available due to the duration of and resources available for this project (two years or less). Further, the travel data for the region where good urban form data about pedestrian networks was available was a representative sample across a metropolitan region rather than being clustered in local areas of distinct, homogeneous street patterns. Sufficient data on specific neighbourhoods were not gathered as part of the PSRC 1999 Travel and Activity Survey, and so statistically valid results would be available on a regional rather than neighborhood scale. Hence a research design that would be able to establish a continuum of measured connectivity values across the region was approach taken. The data were developed in the process outlined below. The process of data development involved assembling and then analyzing and computing network connectivity data for both pedestrian and vehicular modes of travel. These networks needed to match the year (1999) in which the travel survey was conducted to ascertain an accurate measurement of the built environment hypothesized to be influential to that travel. For the dependent cases, it was recognized that examining only home-based travel would be important in distinguishing the travel most directly influenced by the home neighbourhood’s environment. This eliminated some households (those that were traveling only outside of the area during their travel diary reporting) from the sample, and reduced the amount of other households’ valid travel for inclusion in the analysis. While initially it was hoped that both walking and bicycling could be examined in this research, from the outset it was determined that the distinct network for bicycling (on-street lanes and off-street multi-use paths) would be similarly difficult to model effectively and match to the year of the travel data. Thus, given time constraints and the even smaller subset of households that actually bicycled for travel, this aspect of non-motorized transportation was not examined.  122  2. Data Development The following is a more detailed description of the procedures used to gather and develop data in GIS and SPSS. a. Researchers began collecting GIS data from other jurisdictions – parks, trails and sidewalks in Bellevue and Redmond in addition to Seattle; some data was acquired from Metro King County; also approached other nearby cities to see if they had GIS data readily available (Kent and Kirkland held the most promise but did not have sidewalk in GIS) b. Researchers recognized that the street and sidewalk data would need to match to the network conditions in the year 1999, year of the PSRC travel survey. They found that while Bellevue and Redmond had datasets that represented the 1999 conditions well, Seattle and King County did not. Seattle’s sidewalk GIS file had been created in 1993, using orthophotos, and updated in 1996, but was only (in the words of the bicycle and pedestrian program manager) 70% accurate. Thus some additional correction of network data was needed to ensure accurate results. Fortunately, Seattle also had a highly accurate street network database that included connector paths as well as trails network data. c. The researcher worked with a GIS technician to define necessary data for the experiment and develop it from existing data sources: 1) Built network polyline shapefile to be analyzed, 2) Merged shapefiles of trails and sidewalks and snapped these lines to adjacent street network segments (see Fig. A-1).  Figure A-1. Line files to be merged (sidewalks, streets and trails)  3) Added attributes for street, sidewalk within 40’, and trail to new single network line file (see attribute table, Table A-1),  123  Table A–1. Study area network file attribute table. SIDEWALK 0 0 0 0 0 0 0 0 0 40 40  TRAIL STREET WALKABLE CLASS SWALK_CHK DRIVABLE Length Imped_A ImpedB Cost 0 1 0 1 210 1 143.93 158.33 172.72 287.87 0 1 0 1 210 1 250.00 275.00 300.00 500.00 0 1 0 1 210 1 250.00 275.00 300.00 500.00 0 1 0 1 210 1 250.00 275.00 300.00 500.00 0 1 0 1 210 1 250.00 275.00 300.00 500.00 0 1 98 1 98 1 69.56 76.52 83.47 139.12 0 1 98 1 98 1 250.00 275.00 300.00 500.00 0 1 1 1 3 1 11.04 11.04 11.04 11.04 0 1 98 1 98 1 84.29 92.72 101.15 168.58 0 1 98 1 98 1 50.00 50.00 50.00 50.00 0 1 0 1 98 1 20.10 20.10 20.10 20.10  Network features in this sample of the table are mainly streets without sidewalks. The “Cost” attribute represents a 100% impedance on the length of the segment (i.e. doubling the length to reflect a lack of sidewalk); the two other impedances are 10% (Imped_A) and 20% (ImpedB). The last two segments have sidewalk and are thus unimpeded.  4) Developed origin and destination point files (i.e. created points on the network adjacent to the parcels that represented the destination land uses.  Household (origin) parcels were used to generate points on the network file (after creating snap-to nodes by breaking up existing network data into segments 50’ in length), for both pedestrian and vehicular modes. The park point (destination 1) file was created by generating points, nearest to park parcel polygons, on the network at the nodes of network segments as identified above. The commercial point (destination 2) file was created in a similar snapping method to parks, after a subset of parcel polygons, representing those commercial locations likely to be attractive to pedestrian and vehicle travel, was selected from the full parcel file (see Table A-2). Table A-2. Commercial Uses for Destination Selection Commercial Uses in Parcel File to Serve as Commercial Destinations for Route Directness Measurement banks grocery stores/markets restaurants/ eating/drinking places shopping centers convenience stores post offices regional shopping centers retail stores (various) movie theaters d. The next stage began with data clean-up and checking of both travel survey dataset and GIS data:  124  1) Began process of acquiring orthophotos and obtained this resource (.5 foot-pixel color aerial photography in Mr. SID format) approved for use by MDA Corporation in late July and then supplied by the local government’s GIS departments to the researchers. 2) Manually checked sidewalk locations using ArcINFO GIS (version 9.2) software, as well as parks and trails (where there are entrances to these publicly accessible spaces), and some commercial locations. 3) The draft modified network was created and tested using “Near” command to identify nearest park and commercial destinations and Multiple Closest Facilities (MCF) algorithm to measure network distance to nearest destinations, using ArcView 3.2a software.  Figure A-2. Seattle neighbourhood, showing crow-fly versus network distances  Two measurements were needed to compute route directness – crow-fly distance to nearest destination (light blue arrow) and network distance (for walking – the line of Xs; for vehicles, the line of dots). House symbols represent travel survey household locations; lighter lines are sidewalks/pathways; black lines are streets. Orthophoto courtesy of MDA Corporation.  Ratio of Route Directness Computation: e. The project then began modeling of shortest path using MCF, which automates the search for a best route in terms of distance for each origin to a set of destinations using a specified network.  125  This was repeated for the two different classes of destination (parks and commercial), and for both modes of travel – thus, four model runs. f.  Further network corrections were made, mainly to pedestrian network/facilities, and so the MCF analysis was re-run.  g. Output distances were used to compute the main variable of interest: ratio of route directness. h. Linear Regression was attempted on a computed walk-share variable (percent of all trips that were by walk mode): 1) Constructed initial model, including blocks of previously significant demographic, household and urban form co-variates with travel behavior. 2) Ran initial model on new variables (route directness, route directness ratio, effective walking area, sidewalk exent w/in 1km buffer), and found no or very poor significance from the variable of interest (ratio of route directness).  Researchers discovered, unfortunately, that the MCF routine did not always identify the same destination point as the one identified in using a “Near” command script because the nearest destination on the network may not be the nearest via direct air-line distance (found by “Near”). This necessitated re-checking of all households’ nearest destination by crow-fly and on network, for both modes and afforded an opportunity to refine some of the essential parts of the dataset for the analysis. Smaller subsets of park and commercial destinations were identified to try to specify only points that represented true access to a park (rather than simply adjacency to what might be an undeveloped or otherwise un-useable green space) and a more attractive set of commercial for non-work travel (i.e. excluded offices and office parks) – see Table A-2. This reduced the number of destination points for both park and commercial destinations. It was hoped that this would not only be more behaviorally accurate in terms of the measure of accessibility (distance) used for this analysis, but would also reveal more differences in networks by resulting in a slightly longer route to nearest destinations (thus measuring over a greater extent of the local networks). The process was sped up by conducting a spatial join for each household point to each class of destination. This GIS process identified the nearest point in the joined data (which became a shorter list of points for analysis with MCF), generated a table that included the distance (which became the crow-fly measurement), and also assigned a unique identifier to the table of data. The unique identifier could then be used to check for a match with the output .dbf file from the output of the MCF process.  126  Because of a large number of unmatched destination points after the first MCF analysis (>300 out of the 1500 households), successive iterations of MCF analysis were needed to get matches for most of the cases (~100 unmatched). The remaining unmatched households were checked and measured manually. The networks were also reviewed again in this process of manually checking for matching crow-fly and shortest path O-D pairs. This resulted in the final set of measurements that could be used to compute route directness to these nearest destinations, and thereafter the ratio of route directness across walking and driving modes (see Fig. A-2).  Ratio of Network Densities Computation This other main variable was computed with the same corrected sidewalk network data noted above and by using a 1km buffer around each household as a measurement radius. The buffers were created in ArcGIS and then the network (line) shapefile was intersected with the buffer (polygon) shapefile. The output shapefile from this process was a polyline shapefile with all the lines within each household’s buffer given the unique identifier of that household. This file could then be aggregated and the lengths of both networks measured and then summed for each household using X-Tools (http://www.xtoolspro.com) software. The sum of lengths of sidewalks and trails as well as the sum of street segment lengths was then used to create the network density measures for each mode. These separate network densities were then divided (street into sidewalk/walking path) to arrive at the final, area-controlled ratio of network density for each household.  3. Data Quality, Limitations Accuracy of the network data used in this analysis was critical to validity of the later statistical analysis for associations between the built environment and travel behaviour. The researchers cannot be certain of 100% accuracy of the sidewalk or street data, likewise the parcel data, nor whether it matches the conditions at the exact moment of the travel survey. There are inherent limitations in using even high quality, detailed orthophotography: some segments of streets are obscured by vegetation or structures, especially where the photo was not perfectly orthogonal. We are confident, however, due to this painstaking review of the data using the aerial photographs that the built environment is approximated in the final GIS data used to measure urban form with better than 90% accuracy.  127  The research was not, due to time constraints, able to include other factors such as topography, intersection or turning functions for vehicular travel, or pedestrian crossing quality. This, and the assumptions noted in the thesis body above, mean that the routing measurement is necessarily an approximation of how pedestrians or drivers will actually move in the neighbourhoods of the study area. Nevertheless, this is as close to a solid estimation of the built environment as possible. A quick summary of the limitations: •  Possible data limitations (imperfect matching of built form to travel survey in terms of time when the data was recorded; uncertainties in the GIS data about exact sidewalk locations and possible interruptions to connectivity on one or both networks.  •  The study did not attempt to measure other possibly influential features of street design (green space, visual interest along the streets; intersection quality; etc.) While this latter aspect surely influences the decision of whether to walk (or allow one’s children to walk), the research assumed that there was a uniformly sparse distribution of intersections with excellent pedestrian treatments and that there was an uncontrolled intersection design that predominated in residential areas (except at certain major street crossings which typically lie at the edge of the residential neighbourhood units of interest).  •  Assumptions in measurement of pedestrian route directness (i.e. that sidewalk continuity matters little on low-volume, local access residential streets; major streets without sidewalks are considered unwalkable).  •  Analytical challenges: multi-collinearity, spatial covariation, and non-linearity in the relationships of urban form to behaviour.  •  Inability to capture all factors and interactions with available data (topography, for example).  •  While extensions of this research may encompass bicycle travel and overall physically active transportation, this initial research focused on the pedestrian and motor vehicular modes and just on the likelihood of walking.  4. Analysis As statistical analysis began on the project, advice was sought from the Short Term Statistical Consulting service of the Department of Statistics at the University of British Columbia. This advisor assisted with some research design as the data was being developed, appropriate statistical methods, data transformations, and diagnostics for the models used in the research.  128  While at first correlation and linear regression statistics were the main methods being employed, the results of descriptive assessment and initial linear regressions led to the advice of using a logistic regression to analyze the more discrete (not truly continuous) data of mode share (see Figure A-3). Mode share was converted into a binary variable (for example, walking or notwalking, see Figure A-4), so that a logistic regression could be conducted. This sort of statistical model made sense given the non-normal distribution of dependent variable for mode choice and the presence of several categorical independent variables. Figure A-3. Walk Mode Share Among Travel Survey Participants 2,000  Frequency  1,500  1,000  500  Mean Std N = 2,183  0 0  0.2  0.4  0.6  WalkTrip_ShrNw  129  0.8  1  Figure A-4. Binary Categorical Variable of Walking Mode Choice (1 = Walk; 0 = No Walk) 1,200  1,000  Frequency  800  600  400  200  0 -0.50  0.00  0.50  1.00  1.50  whether person walked or not  Linear regression models were still possible to use for other variables such as distance driving or number of walk trips (see Figure A-5). Figure A-5. Continuous Variables for Linear Regression: DistDrive & WalkTrips. 400  Frequency  300  200  100  0 0.00  10.00  20.00  DistDrive  130  30.00  40.00  1,000  Frequency  800  600  400  200  0 0.00  5.00  10.00  15.00  WalkTrips  The descriptive statistics also allowed the research to remove outliers (> 3 standard deviations) and ascertain ranges of data on the variables to be included in the regression model. Fused Grid street designs were then examined, measured using rulers, and averaged to determine whether the network connectivity of the FG would be within the interval of measured values of the networks in the study region. It was found that the Fused Grid did fit within the interval present in the available data and represented at least a 10% change on either the ratio of route directness or the ratio of network densities from a pure gridiron or loop-and-cul-de-sac street pattern with 100% sidewalk coverage (where the ratio variables would both be 1:1) on all streets.  5. Additional Calculations Used In addition to the equations noted in the text and standard statistical formulas, the following calculations were used: Stein’s equation for adjusted r-square (p. 84, 100) Stein’s adjusted R-square = 1 – [(n-1/n-k-1)(n-2/n-k-2)(n+1/n)]*(1-R-squared) n = 1181 k = 12 = 1 – [(1180/1168)*(1179/1167)*(1182/1181)]*(1-.132) = .113 Calculation of the length of a 10% addition of network length for a 1 block area. (p. 94) 300’x4 = 1200’ (street & sidewalk length in 1 block of grid street; 1:1 ped/veh netwk. density) So, a 10% increase would involve adding 120’ or less than a half block pedestrian only pathway. 131  Appendix B. Technical - Analysis Results/Outputs This appendix provides more detailed output tables of the results in correlation and regression analysis, including interaction terms and sensitivity testing as well as examples of regression model diagnostic tests conducted.  1. Additional Descriptive Statistics Household Income, Descriptive. Figure B-1. Histogram – Household Income Categories 400  Frequency  300  200  100  0 10  12  14  16  Table B-1. Trip Distance – Descriptives N  Valid Missing  Mean Median Std. Deviation Range  6681 0 1.05 .96 .652 2  Minimum  0.1  Maximum  2.5  132  18  20  Figure B-2. Trip Distance Frequency Histogram 300  Frequency  200  100  Mean =1.05 Std. Dev. =0.652 N =6,681 0 0  0.5  1  1.5  2  2.5  distance  2. Correlations and Cross Tabulation - Chi-square Results Mix – Neighbourhood Retail correlation table Table B-2. Correlations – Mix of Uses & Neigh. Retail Neighbourhood Retail (number in 1km buffer) Pearson Correlation ( r ) Neighbourhood Retail (number in 1km buffer) Mix  Sig. (2-tailed)  N  1 .656(**)  1,387 0.000  1,387  Trip-level correlations Table B-3. Correlations– Trip-level Walk Trips Walk_Trips total vehicles in household Walk_Trips Gender Ethnicity Age Education Neighbourhood Retail (number in 1km buffer) Net Residential Density (units/acre) Hhsize recoded Income (2 classes)  Pearson Correlation -.243(**) 1 -0.029 0.017 -0.033 .078(**) .315(**) .106(**) -.139(**) -.113(**)  **. Correlation is significant at the 0.01 level (2-tailed). *. Correlation is significant at the 0.05 level (2-tailed).  133  Sig. (2-tailed) 0.000 0.275 0.531 0.211 0.004 0.000 0.000 0.000 0.000  N 1,420 1,420 1,418 1,402 1,420 1,403 1,420 1,420 1,420 1,420  Cross Tabulation – Chi-square Results Table B-4. Cross-tabulation & Chi-square 1 – Walk Share & Ratio of Route Walk Share and Ratio of Route Directness – Person Level Ratio of Route Directness (to Commercial) 0 is more direct for veh.; 1 is more direct for pedestrians) .00 Walk_Share 0 = no walking 1= up to 50% walking 2= 51 – 100% walking  0  1.00  Expected Count  1  32.3  99.8  132.0  72.7%  100.0%  % within RecRatRDCom  83.7%  72.2%  75.0%  % of Total  20.5%  54.5%  75.0%  6  15  21  5.1  15.9  21.0  % within Walk_Share  28.6%  71.4%  100.0%  % within RecRatRDCom  14.0%  11.3%  11.9%  3.4%  8.5%  11.9%  % of Total Count  1  22  23  5.6  17.4  23.0  % within Walk_Share  4.3%  95.7%  100.0%  % within RecRatRDCom  2.3%  16.5%  13.1%  .6%  12.5%  13.1%  Expected Count  % of Total Total  Count  46  43  133  Expected Count  46  43.0  133.0  25.3%  24.4%  75.6%  100.0%  100.0%  100.0%  25.3%  24.4%  75.6%  % within Walk_Share % within RecRatRDCom % of Total Chi-Square Test Value 5.797(a) 7.669  Pearson Chi-Square Likelihood Ratio Linear-by-Linear Association N of Valid Cases  132  27.3%  Count  2  96  % within Walk_Share  Expected Count  4.312  2 2  Asymp. Sig. (2-sided) .055 .022  1  .038  df  176  Value Phi Cramer's V N of Valid Cases  .00  Count  36  Nominal by Nominal  Total  Approx. Sig.  .181  .055  .181  .055  176  134  Table B-5. Cross-tabulation & Chi-square 2 – Walk Share & Ratio of Network Density Walk Share & Ratio of Network Densities – Person Level Sidewalk to Street (recoded – 0 is below 1:1; 1 is above 1:1 .00 Walk_Share  .00  592 539.5  297.5  837.0  % within Walk_Share  70.7%  29.3%  100.0%  % within RecSWStr  79.0%  59.3%  72.0%  % of Total  50.9%  21.1%  72.0%  Count Expected Count  1.00  Count  74  153  54.4  153.0  % within Walk_Share  51.6%  48.4%  100.0%  % within RecSWStr  10.5%  17.9%  13.2%  6.8%  6.4%  13.2%  Count  78  94  172  110.9  61.1  172.0  % within Walk_Share  45.3%  54.7%  100.0%  % within RecSWStr  10.4%  22.8%  14.8%  6.7%  8.1%  14.8%  749  413  1162  749.0  413.0  1162  Expected Count  % of Total Count Expected Count % within Walk_Share % within RecSWStr % of Total  64.5%  35.5%  100.0%  100.0%  100.0%  100.0%  64.5%  35.5%  100.0%  Chi-Square Test -  Value Pearson Chi-Square Likelihood Ratio Linear-by-Linear Association N of Valid Cases  Nominal by Nominal  837  79  % of Total  Total  .00  98.6  Expected Count  2.00  Total  1.00 245  Asymp. Sig. (2-sided)  df  52.765(a) 51.387  2 2  0.000 0.000  50.561  1  0.000  1,162  Phi Cramer's V Contingency Coefficient N of Valid Cases  Value 0.213 0.213 0.208 1,162  Approx. Sig. 0.000 0.000 0.000  135  3. Regression Model Outputs. Trip Level: Table B-6. Walking vs. No-walking, Trip-level Logistic Regression – Full Model Omnibus Tests of Model Coefficients Step 1  Step  Chi-square 21.589  df 2  Sig. 0.000  Block  21.589  2  0.000  Model  729.074  26  0.000  Model Summary Step 1  -2 Log likelihood 5259.266  Cox & Snell R Square 0.122  Nagelkerke R Square 0.186 Classification Table  Observed  Predicted Walk Behaviour (0 = None; 1 = Yes) .00  Step 1  Walk Behaviour (0 = None; 1 = Yes)  .00 1.00  Percentage Correct  1.00  .00  4,195  120  97.2  966  304  23.9  Overall Percentage  80.6 Variables in the Equation B  Step 1(a)  Total Vehicles (household) Household Size  S.E.  Wald 62.860  1  0.000  0.631  -0.146  0.038  14.452  1  0.000  0.864  49.312  7  0.000  -0.008  0.003  -0.019  0.070  Ethnicity (cat.) Neighbourhood Retail Net Residential Density Crowfly distance to commercial Ratio of Route Directness Ratio of Network Densities Constant  Exp(B)  0.058  Education (cat.) Gender (cat.)  Sig.  -0.461  Income (categorical) Age  df  0.015  0.001  8.094  1  0.004  22.689  5  0.000  0.070  1  0.792  14.649  5  0.012  105.340  1  0.000  .823  0.982 1.015  0.015  0.285  1  0.593  1.008  0.000  0.000  0.379  1  0.538  1.000  -1.073  0.402  7.124  1  0.008  0.342  0.635  0.201  10.000  1  0.002  1.887  1.700  1  0.192  2.315  136  .570  Upper  0.992  0.008  0.839 0.644 a Variable(s) entered on step 1: Rat_RDCom, Ratio_SW_Str.  Lower  1.013  1.023  .955  1.053  1.000  1.000  .369  5.046  1.351  4.521  Person Level: Table B-7. Walking or not Walking – Person Level Logistic Regression – Full Model Omnibus Tests of Model Coefficients Step 1  Step  Chi-square 6.145  Df 2  Sig. 0.046  Block  6.145  2  0.046  Model  154.908  20  0.000  Model Summary Step 1  -2 Log likelihood  Cox & Snell R Square  Nagelkerke R Square  0.123  0.173  1304.750  Classification Table(a) Observed  Predicted Percentage Correct  Did person walk .00 Step 1  Did person walk  1.00  .00  .00  755  63  92.3  1.00  266  98  26.9  Overall Percentage  72.2 Variables in the Equation B  Step 1(a)  Totat Vehicles  S.E.  Wald  df  Sig.  Exp(B)  -0.113  0.075  2.278  1  0.131  0.894  0.164  0.195  0.708  1  0.400  1.178  Gender (categorical)  -0.407  0.111  13.331  1  0.000  0.666  Ethnicity (cat.)  -0.130  0.137  0.897  1  0.343  0.878  3.246  5  0.662  Household Size  Age (cat.) Education (cat.)  -0.008  0.005  Income (cat.)  2.408  1  0.121  20.674  5  0.001  0.992  Neigh. Retail  0.018  0.003  32.594  1  0.000  1.018  Net Resid. Density  0.001  0.030  0.001  1  0.970  1.001  0.000  0.000  0.925  1  0.336  1.000  0.359  0.792  0.206  1  0.650  1.432  0.885  0.364  5.907  1  0.015  2.423  -0.290  1.243  0.054  1  0.815  0.748  Crowfly distance (commercial) Ratio of Route Directness Ratio of Network Density Constant  a Variable(s) entered on step 1: Rat_RDCom, Ratio_SW_Str.  137  Table B-8. Active Walking, Person Level Logistic Regression– Full Model Omnibus Tests of Model Coefficients Chi-square Step 1  Step  df  Sig.  7.583  2  .023  Block  7.583  2  .023  Model  97.891  7  .000  Model Summary Step 1  -2 Log likelihood  Cox & Snell R Square  Nagelkerke R Square  .078  .213  452.946(a)  Classification Table(a) Observed  Predicted whether someone walked 30 minutes or not .00  Step 1  whether someone walked 30 minutes or not  .00  1.00  .00  1130  2  68  5  1.00  Percentage Correct  99.8 6.8  Overall Percentage  94.2 Variables in the Equation B  Step 1(a)  Total Vehicles Gender Income (dummy) Education (dummy) Neighbourhood Retail Ratio of Route Directness Ratio of Network Density Constant  S.E.  Wald  df  Sig.  -1.003  .205  23.856  1  .000  .367  .577  .263  4.797  1  .029  1.780  -.119  .315  .144  1  .705  .887  .409  .271  2.267  1  .132  1.505  .016  .004  14.347  1  .000  1.016  -2.306  1.343  2.949  1  .086  .100  1.676  .922  3.303  1  .069  5.344  -1.822  1.696  1.154  1  .283  .162  Driving – Person Level Logistic Regression Table B-9. Driving versus No-driving Travel – Person Level Logistic Regression Omnibus Tests of Model Coefficients Chi-square df Sig. 3.546  2  .170  Step 1 Block 3.546  2  .170  Step  Model 230.725  Exp(B)  21 .000  138  Model Summary Step -2 Log likelihood Cox & Snell R Square Nagelkerke R Square 1  948.845(a)  .175  .280 Classification Table Predicted  Observed  whether person drove or not Percentage Correct .00  Step 1  whether person drove or not  1.00  .00  .00  80  153  34.3  1.00  29  934  97.0 84.8  Overall Percentage  Variables in the Equation B  S.E.  Wald  df  Sig.  Exp(B)  90.0% C.I.for EXP(B) Lower  Step 1(a)  Total Vehicles Household Size Gender Age  Upper  .742  .143  26.851  1  .000  2.100  1.659  2.658  .180  .091  3.882  1  .049  1.197  1.030  1.391  -.182  .167  1.190  1  .275  .833  .633  1.097  .013  .005  1.013  1.005  1.021  7.497  1  .006  Income  10.628  7  .156  Ethnicity  7.929  5  .160  Neigh. Retail  -.012  .003  11.761  1  .001  .988  .983  .994  Net Resid. Density  -.081  .031  6.689  1  .010  .922  .876  .971  .000  .000  4.014  1  .045  1.000  1.000  1.001  .827  1.000  .684  1  .408  2.287  .441  11.853  -.691  .456  2.291  1  .130  .501  .237  1.062  19.459  13862.56  .000  1  .999  282407526.4  Crowfly Dist. to Commercial Ratio of Route Directness Ratio of Network Density Constant  a Variable(s) entered on step 1: Rat_RDCom, Ratio_SW_Str.  139  Z-scored Factors Model Runs Table B-10. Walking versus Not Walking, Logistic Regression -- Trip Level Variables in the Equation B Step 1(a)  ZTotVehicle  S.E.  Wald  df  Sig.  95.0% C.I.for EXP(B)  Exp(B)  -.369  .046  62.860  1  .000  .692  .631  .758  -.180  .047  14.452  1  .000  .835  .761  .917  49.312  7  .000  Income(7)  -.272  .101  7.282  1  .007  .762  .625  .928  gender(1)  -.019  .070  .982  .855  1.127  .847  .756  .950  ZHhsize Income  Ethnicity Zage  .070  1  .792  14.649  5  .012  -.166  .058  8.094  1  .004  22.689  5  .000  Znr  .428  .042  105.340  1  .000  1.534  1.414  1.665  ZNRD_2000_Capped  .018  .033  .285  1  .593  1.018  .954  1.085  ZCrow_DistCom  .031  .050  .379  1  .538  1.031  .935  1.138  -.198  .074  7.124  1  .008  .821  .710  .949  .161  .051  10.000  1  .002  1.174  1.063  1.297  -.639  .367  3.027  1  .082  .528  Education  ZRat_RDCom ZRatio_SW_Str Constant  a Variable(s) entered on step 1: ZCrow_DistCom, ZRat_RDCom, ZRatio_SW_Str.  Table B-11. Walking versus Not Walking, Logistic Regression – Person Level (Z-score) Variables in the Equation B Step 1(a)  ZTotVeh gender(1)  S.E.  -.331  .084  -.151  .136  Wald  df  Sig.  95.0% C.I.for EXP(B)  15.579  1  .000  .718  .609  .846  .860  .658  1.123  1.225  1  .268  21.061  5  .001  Income  9.519  7  .217  ethnicity  2.812  5  .729  education  Exp(B)  Znr  .477  .083  32.993  1  .000  1.611  1.369  1.896  ZNRD_2000_Capped  .014  .068  .040  1  .842  1.014  .888  1.157  ZCrow_DistCom  .070  .092  .575  1  .448  1.072  .895  1.284  ZRat_RDCom  .026  .075  .124  1  .725  1.027  .887  1.189  ZRatio_SW_Str  .216  .087  6.226  1  .013  1.242  1.048  1.472  1  .758  1.264  Constant  .234 .760 .095 a Variable(s) entered on step 1: ZRat_RDCom, ZRatio_SW_Str.  140  Linear regression model outputs for other outcomes besides distance driven Table B-12. Drive Trips – Linear Regression – Full Model Model Summary Model 1  R .291(a)  R Square .085  Adjusted R Square .079  Std. Error of the Estimate 2.72082  2  .319(b)  .102  .094  2.69851  3  .322(c) .104 .095 2.69796 a Predictors: (Constant), TotVeh, dummy of education variable, gender, white or non-white, age, IncomeDummy, Hhsize b Predictors: (Constant), TotVeh, dummy of education variable, gender, white or non-white, age, IncomeDummy, Hhsize, Net Residential Density (units/acre), Crow Distance to Commercial, Neighbourhood Retail (number in 1km buffer) c Predictors: (Constant), TotVeh, dummy of education variable, gender, white or non-white, age, IncomeDummy, Hhsize, Net Residential Density (units/acre), Crow Distance to Commercial, Neighbourhood Retail (number in 1km buffer), Ratio of Route directness to Commercial (Ped/Veh), Ratio of Sidewalk/Trail length to Street length ANOVA(d) Model 1  2  3  Regression  Sum of Squares 804.285  Residual Total  7  Mean Square 114.898  8690.950  1174  7.403  9495.235  1181  Regression  df  968.063  10  96.806  Residual  8527.172  1171  7.282  Total  9495.235  1181  Regression  986.091  12  82.174  Residual  8509.144  1169  7.279  Total  9495.235  1181  F 15.521  Sig. .000(a)  13.294  .000(b)  11.289  .000(c)  a Predictors: (Constant), TotVeh, dummy of education variable, gender, white or non-white, age, IncomeDummy, Hhsize b Predictors: (Constant), TotVeh, dummy of education variable, gender, white or non-white, age, IncomeDummy, Hhsize, Net Residential Density (units/acre), Crow Distance to Commercial, Neighbourhood Retail (number in 1km buffer) c Predictors: (Constant), TotVeh, dummy of education variable, gender, white or non-white, age, IncomeDummy, Hhsize, Net Residential Density (units/acre), Crow Distance to Commercial, Neighbourhood Retail (number in 1km buffer), Ratio of Route directness to Commercial (Ped/Veh), Ratio of Sidewalk/Trail length to Street length d Dependent Variable: VehTrips  Unstandardized Coefficients Std. B Error  Model Step 1  (Constant)  Coefficients(a) Standardized Coefficients t Beta  Sig.  Tolerance  -.463  .458  Hhsize  .560  .084  .241  IncomeDummy  .223  .230  gender  .178  .160  age  .020  dummy of education variable  VIF  Collinearity Statistics Std. B Error  -1.011  .312  6.682  .000  .598  1.672  .031  .968  .333  .754  1.326  .031  1.117  .264  .989  1.011  .005  .153  4.408  .000  .644  1.552  .089  .176  .016  .508  .611  .812  1.232  white or non-white  .324  .260  .035  1.247  .213  .973  1.028  TotVeh  .360  .118  .101  3.041  .002  .702  1.425  141  2  (Constant) Hhsize IncomeDummy gender age dummy of education variable white or non-white  .411 .486 .079 .130 .017  .538 .085 .230 .159 .005  .209 .011 .023 .125  .763 5.728 .344 .816 3.578  .445 .000 .731 .415 .000  .574 .740 .983 .624  1.742 1.351 1.017 1.603  .157  .175  .028  .897  .370  .804  1.244  .282  .258  .031  1.092  .275  .970  1.030  TotVeh  .247  .120  .069  2.054  .040  .671  1.491  -.008  .004  -.077  -2.265  .024  .655  1.526  -.065  .035  -.053  -1.861  .063  .941  1.062  .000  .000  .072  2.118  .034  .661  1.513  -.277  1.098  -.252  .801  .479  .085  .206  5.642  .000  .572  1.747  .094  .231  .013  .409  .683  .739  1.354  .135  .159  .024  .851  .395  .981  1.020  .016  .005  .123  3.494  .000  .620  1.614  .189  .178  .033  1.063  .288  .783  1.277  .258 .244  .259 .120  .028 .069  .998 2.032  .319 .042  .967 .670  1.035 1.493  -.007  .004  -.072  -2.048  .041  .623  1.606  -.061  .035  -.050  -1.730  .084  .935  1.070  .000  .000  .066  1.898  .058  .640  1.563  .996  .873  .032  1.141  .254  .949  1.054  -.325  .383  -.027  -.847  .397  .771  1.297  Neighbourhood Retail (number in 1km buffer) Net Residential Density (units/acre) Crow Distance to Commercial 3  (Constant) Hhsize IncomeDummy gender age dummy of education variable white or non-white TotVeh  Neighbourhood Retail (number in 1km buffer) Net Residential Density (units/acre) Crow Distance to Commercial Ratio of Route directness to Commercial (Ped/Veh) Ratio of Sidewalk/Trail length to Street length a Dependent Variable: VehTrips  Table B-13. Total Trips – Linear Regression – Full Model Model Summary Model 1  R .144(a)  R Square .021  Adjusted R Square .015  Std. Error of the Estimate 2.93915  2  .204(b)  .041  .033  2.91176  3  .210(c)  .044  .034  2.91047  a Predictors: (Constant), TotVeh, dummy of education variable, gender, white or non-white, age, IncomeDummy, Hhsize b Predictors: (Constant), TotVeh, dummy of education variable, gender, white or non-white, age, IncomeDummy, Hhsize, Net Residential Density (units/acre), Crow Distance to Commercial, Neighbourhood Retail (number in 1km buffer) c Predictors: (Constant), TotVeh, dummy of education variable, gender, white or non-white, age,  142  IncomeDummy, Hhsize, Net Residential Density (units/acre), Crow Distance to Commercial, Neighbourhood Retail (number in 1km buffer), Ratio of Route directness to Commercial (Ped/Veh), Ratio of Sidewalk/Trail length to Street length ANOVA(d) Model 1  2  Sum of Squares Regression  Mean Square 7  30.879  Residual  10141.752  1174  8.639  Total  10357.905  1181  429.783  10  42.978 8.478  Regression Residual Total  3  df  216.153  9928.122  1171  10357.905  1181  455.527  12  37.961  9902.379  1169  8.471  Regression Residual  F  Sig.  3.575  .001(a)  5.069  .000(b)  4.481  .000(c)  Total  10357.905 1181 a Predictors: (Constant), TotVeh, dummy of education variable, gender, white or non-white, age, IncomeDummy, Hhsize b Predictors: (Constant), TotVeh, dummy of education variable, gender, white or non-white, age, IncomeDummy, Hhsize, Net Residential Density (units/acre), Crow Distance to Commercial, Neighbourhood Retail (number in 1km buffer) c Predictors: (Constant), TotVeh, dummy of education variable, gender, white or non-white, age, IncomeDummy, Hhsize, Net Residential Density (units/acre), Crow Distance to Commercial, Neighbourhood Retail (number in 1km buffer), Ratio of Route directness to Commercial (Ped/Veh), Ratio of Sidewalk/Trail length to Street length d Dependent Variable: TotalTrips  Model Step 1  (Constant) Hhsize IncomeDummy  .339  .091  Coefficients(a) Standardized Coefficients t Beta .140  Sig.  Collinearity Statistics Std. B Error  Tolerance 8.436  VIF .000  3.745  .000  .598  1.672  .224  .249  .030  .902  .367  .754  1.326  -.095  .172  -.016  -.553  .581  .989  1.011  age  .001  .005  .009  .242  .809  .644  1.552  dummy of education variable  .311  .190  .053  1.638  .102  .812  1.232  white or non-white  .196  .281  .020  .696  .487  .973  1.028  -.482 2.805 .379 .308 -.037 .003  .128 .581 .092 .249 .171 .005  -.130  .000 .000 .000 .216 .828 .589  .702  1.425  .156 .041 -.006 .020  -3.768 4.831 4.142 1.239 -.217 .541  .574 .740 .983 .624  1.742 1.351 1.017 1.603  .274  .189  .046  1.450  .147  .804  1.244  .237  .279  .025  .850  .395  .970  1.030  -.466  .130  -.125  -3.590  .000  .671  1.491  .019  .004  .173  4.899  .000  .655  1.526  .003  .038  .003  .090  .928  .941  1.062  .000  .000  .112  3.195  .001  .661  1.513  gender  TotVeh 2  Unstandardized Coefficients Std. B Error 4.174 .495  (Constant) Hhsize IncomeDummy gender age dummy of education variable white or non-white TotVeh Neighbourhood Retail (number in 1km buffer) Net Residential Density (units/acre) Crow Distance to Commercial  143  3  (Constant) Hhsize IncomeDummy gender age dummy of education variable white or non-white TotVeh Neighbourhood Retail (number in 1km buffer) Net Residential Density (units/acre) Crow Distance to Commercial Ratio of Route directness to Commercial (Ped/Veh) Ratio of Sidewalk/Trail length to Street length  1.042  1.185  .879  .379  .371  .092  .153  4.051  .000  .572  1.747  .314  .249  .042  1.263  .207  .739  1.354  -.043  .172  -.007  -.250  .802  .981  1.020  .003  .005  .021  .580  .562  .620  1.614  .269  .192  .045  1.402  .161  .783  1.277  .210 -.463  .279 .130  .022 -.125  .754 -3.570  .451 .000  .967 .670  1.035 1.493  .018  .004  .167  4.620  .000  .623  1.606  .005  .038  .004  .143  .887  .935  1.070  .000  .000  .114  3.197  .001  .640  1.563  1.627  .942  .051  1.728  .084  .949  1.054  .229  .413  .018  .555  .579  .771  1.297  a Dependent Variable: TotalTrips  Table B-14. Walk Distance – Linear Regression – Full Model Model Summary R  R Square  Adjusted R Square  Std. Error of the Estimate  Model 1  .296(a)  .088  .082  1.17465  2  .363(b)  .131  .124  1.14765  3  .369(c)  .136  .127  1.14570  a Predictors: (Constant), IncomeDummy, gender, white or non-white, dummy of education variable, Hhsize, TotVeh, age b Predictors: (Constant), IncomeDummy, gender, white or non-white, dummy of education variable, Hhsize, TotVeh, age, Net Residential Density (units/acre), Crow Distance to Commercial, Neighbourhood Retail (number in 1km buffer) c Predictors: (Constant), IncomeDummy, gender, white or non-white, dummy of education variable, Hhsize, TotVeh, age, Net Residential Density (units/acre), Crow Distance to Commercial, Neighbourhood Retail (number in 1km buffer), Ratio of Route directness to Commercial (Ped/Veh), Ratio of Sidewalk/Trail length to Street length ANOVA(d) Model 1  2  3  Sum of Squares Regression  df  Mean Square  155.801  7  22.257  Residual  1619.888  1174  1.380  Total  1775.689  1181  Regression  233.355  10  23.336  Residual  1542.334  1171  1.317  Total  1775.689  1181  Regression Residual  241.231  12  20.103  1534.458  1169  1.313  Total  F  Sig.  16.131  .000(a)  17.717  .000(b)  15.315  .000(c)  1775.689 1181 a Predictors: (Constant), IncomeDummy, gender, white or non-white, dummy of education variable, Hhsize, TotVeh, age  144  b Predictors: (Constant), IncomeDummy, gender, white or non-white, dummy of education variable, Hhsize, TotVeh, age, Net Residential Density (units/acre), Crow Distance to Commercial, Neighbourhood Retail (# in 1km buffer) c Predictors: (Constant), IncomeDummy, gender, white or non-white, dummy of education variable, Hhsize, TotVeh, age, Net Residential Density (units/acre), Crow Distance to Commercial, Neighbourhood Retail (number in 1km buffer), Ratio of Route directness to Commercial (Ped/Veh), Ratio of Sidewalk/Trail length to Street length d Dependent Variable: DistWalk  Model 1  Coefficients(a) Unstandardized Coefficients B Std. Error  Sig. Std. Error  1.923  .198  9.727  .000  TotVeh  -.318  .051  -.207  -6.213  .000  gender  -.137  .069  -.056  -1.983  .048  age  -.008  .002  -.142  -4.075  .000  Hhsize  -.105  .036  -.105  -2.914  .004  white or non-white  -.046  .112  -.011  -.406  .685  .215  .076  .087  2.828  .005  -.137 1.074 -.267 -.095 -.006 -.063 -.015  .099 .229 .051 .068 .002 .036 .110  -.044 -.174 -.039 -.106 -.062 -.004  -1.374 4.695 -5.219 -1.412 -3.064 -1.737 -.133  .170 .000 .000 .158 .002 .083 .894  .173  .075  .070  2.313  .021  -.048  .098  -.015  -.487  .627  .010  .002  .234  6.957  .000  .020  .015  .038  1.365  .172  6.15E-005  .000  .038  1.146  .252  1.219  .466  2.614  .009  -.264  .051  -.172  -5.174  .000  -.101  .068  -.041  -1.499  .134  -.006  .002  -.100  -2.898  .004  -.059  .036  -.059  -1.640  .101  -.001  .110  .000  -.013  .989  .146  .075  .060  1.938  .053  IncomeDummy (Constant) TotVeh gender age Hhsize white or non-white dummy of education variable IncomeDummy  3  t B  (Constant)  dummy of education variable 2  Standardized Coefficients Beta  Neighbourhood Retail (number in 1km buffer) Net Residential Density (units/acre) Crow Distance to Commercial (Constant) TotVeh gender age Hhsize white or non-white dummy of education variable IncomeDummy  -.058  .098  -.019  -.595  .552  Neighbourhood Retail  .010  .002  .220  6.400  .000  Net Residential Density  .017  .015  .033  1.158  .247  Crow Distance to Com.  8.22E-005  .000  .051  1.510  .131  -.422  .371  -.032  -1.138  .255  .311  .163  .059  1.913  .056  Ratio of Route directness to Commercial (Ped/Veh) Ratio of Sidewalk/Trail length to Street length a Dependent Variable: DistWalk  145  Table B-15. Walk Trips – Linear Regression – Full Model Model Summary R  R Square  Adjusted R Square  Std. Error of the Estimate  Model 1  .293(a)  .086  .081  1.95223  2  .389(b)  .151  .144  1.88357  3  .393(c)  .154  .146  1.88182  a Predictors: (Constant), TotVeh, dummy of education variable, gender, white or non-white, age, IncomeDummy, Hhsize b Predictors: (Constant), TotVeh, dummy of education variable, gender, white or non-white, age, IncomeDummy, Hhsize, Net Residential Density (units/acre), Crow Distance to Commercial, Neighbourhood Retail (number in 1km buffer) c Predictors: (Constant), TotVeh, dummy of education variable, gender, white or non-white, age, IncomeDummy, Hhsize, Net Residential Density (units/acre), Crow Distance to Commercial, Neighbourhood Retail (number in 1km buffer), Ratio of Route directness to Commercial (Ped/Veh), Ratio of Sidewalk/Trail length to Street length ANOVA(d) Model 1  2  3  Regression  Sum of Squares 421.268  Residual Total  7  Mean Square 60.181  4474.346  1174  3.811  4895.614  1181  Regression  df  741.119  10  74.112  Residual  4154.495  1171  3.548  Total  4895.614  1181  Regression  755.909  12  62.992  Residual  4139.706  1169  3.541  Total  4895.614  1181  F 15.791  Sig. .000(a)  20.889  .000(b)  17.788  .000(c)  a Predictors: (Constant), TotVeh, dummy of education variable, gender, white or non-white, age, IncomeDummy, Hhsize b Predictors: (Constant), TotVeh, dummy of education variable, gender, white or non-white, age, IncomeDummy, Hhsize, Net Residential Density (units/acre), Crow Distance to Commercial, Neighbourhood Retail (number in 1km buffer) c Predictors: (Constant), TotVeh, dummy of education variable, gender, white or non-white, age, IncomeDummy, Hhsize, Net Residential Density (units/acre), Crow Distance to Commercial, Neighbourhood Retail (number in 1km buffer), Ratio of Route directness to Commercial (Ped/Veh), Ratio of Sidewalk/Trail length to Street length d Dependent Variable: WalkTrips  Model  1  (Constant)  Unstandardize d Coefficients Std. B Error 3.184 .329  Coefficients(a) Standardized Coefficients t Beta  Sig.  Tolerance 9.689  VIF .000  Collinearity Statistics Std. B Error  Hhsize  -.151  .060  -.090  -2.504  .012  .598  1.672  IncomeDummy  -.228  .165  -.044  -1.381  .168  .754  1.326  gender  -.146  .115  -.036  -1.277  .202  .989  1.011  age  -.013  .003  -.140  -4.039  .000  .644  1.552  .361  .126  .089  2.860  .004  .812  1.232  white or non-white  -.060  .187  -.009  -.320  .749  .973  1.028  TotVeh  -.552  .085  -.216  -6.491  .000  .702  1.425  dummy of education variable  146  2  (Constant) Hhsize IncomeDummy gender age dummy of education variable white or non-white  1.482 -.072 -.056 -.063 -.009  .376 .059 .161 .111 .003  -.043 -.011 -.015 -.099  3.946 -1.212 -.348 -.569 -2.904  .000 .226 .728 .569 .004  .574 .740 .983 .624  1.742 1.351 1.017 1.603  .277  .122  .068  2.263  .024  .804  1.244  -.001  .180  .000  -.005  .996  .970  1.030  TotVeh  -.459  .084  -.180  -5.472  .000  .671  1.491  .022  .002  .299  8.980  .000  .655  1.526  .022  .024  .025  .893  .372  .941  1.062  .000  .000  .071  2.141  .033  .661  1.513  (Constant)  1.523  .766  1.988  .047  Hhsize  -.068  .059  -.041  -1.140  .255  .572  1.747  -.070  .161  -.014  -.437  .662  .739  1.354  -.072  .111  -.018  -.648  .517  .981  1.020  -.009  .003  -.094  -2.752  .006  .620  1.614  .239  .124  .059  1.929  .054  .783  1.277  .015 -.455  .180 .084  .002 -.178  .085 -5.427  .933 .000  .967 .670  1.035 1.493  .021  .003  .286  8.400  .000  .623  1.606  .018  .024  .020  .721  .471  .935  1.070  .000  .000  .082  2.444  .015  .640  1.563  -.447  .609  -.020  -.734  .463  .949  1.054  .463  .267  .053  1.733  .083  .771  1.297  Neighbourhood Retail (number in 1km buffer) Net Residential Density (units/acre) Crow Distance to Commercial 3  IncomeDummy gender age dummy of education variable white or non-white TotVeh Neighbourhood Retail (number in 1km buffer) Net Residential Density (units/acre) Crow Distance to Commercial Ratio of Route directness to Commercial (Ped/Veh) Ratio of Sidewalk/Trail length to Street length a Dependent Variable: WalkTrips  4. Sensitivity testing of various travel outcomes on impeded route directness Table B-16. Interaction Term Crow-fly x Route Directness Ratio – Trip-Level Logistic Regression Model Summary Step 1  -2 Log likelihood  Cox & Snell R Square  Nagelkerke R Square  .123  .187  5255.928  Variables in the Equation B Step 1(a)  Total Vehicle Household Size  S.E.  Wald  Sig.  Exp(B )  -.459  .058  62.476  1  .000  .632  -.150  .038  15.318  1  .000  .860  49.657  7  .000  -.017  .070  .056  1  .813  15.354  5  .009  Income Gender  df  Ethnicity  147  .983  Age  -.008  .003  Education Neigh. Retail  7.831  1  .005  22.728  5  .000  .992  104.373  1  .000  1.015  .015  .001  Net Resid. Density  .006  .015  .176  1  .675  1.006  Crow-fly Distance  -.001  .001  3.249  1  .071  .999  -2.386  .808  8.714  1  .003  .092  .654  .202  10.528  1  .001  1.923  .001  .001  3.607  1  .058  1.001  2.136  .946  5.096  1  .024  8.465  Ratio of Route Directness Ratio of Network Density Crow-fly by Rat_RDCom (Interaction) Constant  Table B-17. Interaction, Walk vs. No-Walk -- Person-Level Logistic Regression Model Summary Step 1  -2 Log likelihood  Cox & Snell R Square  Nagelkerke R Square  .130  .183  1295.228  Variables in the Equation B Step 1(a)  Total Vehicles Household Size  S.E.  Wald  Exp(B)  .113  12.712  1  .000  .667  -.104  .076  1.895  1  .169  .901  7.756  7  .355  -.007  .005  Education gender(1)  Sig.  -.404  Income Age  df  1.731  1  .188  20.329  5  .001  .993  -.134  .138  .940  1  .332  3.000  5  .700  .018  .003  31.124  1  .000  1.018  Net Resid. Density  -.002  .030  .005  1  .941  .998  Crow_DistCom Crow_DistCom by Rat_RDCom  -.001  .001  1.889  1  .169  .999  .001  .001  2.249  1  .134  1.001  -1.811  1.611  1.263  1  .261  .163  .835  .371  5.074  1  .024  2.305  1.998  1.872  1.140  1  .286  7.377  Ethnicity Neigh. Retail  Rat_RDCom Ratio_SW_Str Constant  .875  Inclusion of a Transit Proximity Variable. The research also tested a variable of proximity to transit. The logistic regression at the trip level found this to be a significant addition to the model, but only improved its R-square by .001. Due to its high correlation with Ratio of Network Density noted above, this variable was not included in the other regression analysis, but does appear to a be an important factor to consider in examining walking behaviour. 148  Neighbourhood Average Ratio of Route Directness A neighbourhood average route directness measurement was conducted for several neighbourhoods (n = 13) where there was a detectable disparity in route directness between the two modes. This was done by grouping according to the clusters apparent in Figure 5-9. The results of the logistic regression model (odds of walking – person level) using this measure are shown in Table 6-14. The neighbourhood average ratio of route directness measure did not enter as significant factor in this trip-level model, though it was close to the 10% level of significance and had double the beta coefficient of the non-averaged trip-level analysis. Table B-18. Neigh. Average Route Directness Walk or Non-walk – Person-level Log. Regression Omnibus Tests of Model Coefficients Step 1  Chi-square 12.322  Step  df 2  Sig. .002  Block  12.322  2  .002  Model  592.457  26  .000  Model Summary Step 1  -2 Log likelihood  Cox & Snell R Square  Nagelkerke R Square  .198  .286  2573.608  Variables in the Equation B Step 1(a)  Total Vehicles Household Size  S.E.  1  .000  .679  -.202  .056  12.970  1  .000  .817  68.163  7  .000  -.002  .004  .194  1  .660  4.410  5  .492  .087  Net Resid. Density Crow-fly Distance Avg. Ratio of Route Dir. Ratio of Network Density Constant  Exp(B)  20.794  .100  Ethnicity Neighbourhood Retail  Sig.  .085  Education Gender  df  -.387  Income Age  Wald  .743  1  .389  26.906  5  .000  .998 1.090  .017  .002  67.116  1  .000  1.017  .023 .000 -2.007 .510 -18.803  .026 .000 1.341 .352 28427.211  .826 4.400 2.241 2.100 .000  1 1 1 1 1  .363 .036 .134 .147 .999  1.024 1.000 .134 1.665 .000  Table B-19. Impeded Network (20%) Walking or Not – Person-level Logistic Regression Omnibus Tests of Model Coefficients Chisquare df Step 1 Step 25.316 2 Block 25.316 2 Model  785.867  Sig. .000 .000  26  .000  Model Summary  149  St e p 1  -2 Log Cox & Nagelker likeliho Snell R ke R od Square Square 5483.0 .125 .191 38(a) a Estimation terminated at iteration number 5 because parameter estimates changed by less than .001. Classification Table(a) Predicted Percenta ge Correct  Walk Behavior (0 = None; 1 = Yes) Observed Walk Behavior (0 = None; 1 = Yes) Overall Percentage  St ep 1  .00  .00 4429  1.00 120  .00 97.4  1.00  1005  319  24.1 80.8  a The cut value is .500 Variables in the Equation B Step 1(a)  TotVehicle Hhsize  S.E.  Sig.  Exp(B)  95.0% C.I.for EXP(B) Lower Upper  .058  70.826  1  .000  .615  .549  .689  -.131  .035  13.626  1  .000  .877  .818  .940  50.410  7  .000  -.006  .003  5.636  1  .018  .994  .989  .999  19.975  5  .001  -.060  .069  .741  1  .389  .942  .823  1.079  14.337  5  .014  education gender(1)  df  -.486  Income age  Wald  ethnicity .014  .001  91.354  1  .000  1.014  1.011  1.017  NRD_2000_Capped  .035  .018  3.815  1  .051  1.035  1.000  1.072  Crow_DistCom Impeded (20%) RatRD  .000  .000  .797  1  .372  1.000  1.000  1.000  -.759  .299  6.447  1  .011  .468  .260  .841  Ratio_SW_Str  .586  .209  7.847  1  .005  1.797  1.192  2.707  .610 1.671 Constant .788 a Variable(s) entered on step 1: ImpB_RatRD, Ratio_SW_Str.  1  .196  2.199  nr  150  Table B-20. Intersection density model with Ratio of Route Directness Omnibus Tests of Model Coefficients Chisquare Df Step 1 Step 11.587 2 Block 11.587 2 Model 715.455 26  Sig. .003 .003 .000  Model Summary Step 1  Step 1  -2 Log likelihood 5375.8 23(a)  Cox & Snell R Square  Nagelkerke R Square  .120  .181  Hosmer and Lemeshow Test Chisquare df Sig. 62.297 8 .000 Classification Table(a) Predicted Percenta ge Correct  Walk Behavior (0 = None; 1 = Yes) Step 1  Observed Walk Behavior (0 = None; 1 = Yes) Overall Percentage  .00  .00 4191  1.00 124  .00 97.1  1.00  1004  301  23.1 79.9  a The cut value is .500 Variables in the Equation B Step 1(a)  S.E.  Wald  df  Sig.  TotVehicle  -.476  .058  67.597  1  .000  Hhsize  -.152  .038  16.070  1  .000  50.528  7  .000  -.007  .003  7.614  1  .006  24.457  5  .000  Income age education gender(1)  -.032  .069  ethnicity nr  .208  95.0% C.I.for EXP(B)  Exp(B) .621  Lower .554  Upper .696  .859  .797  .925  .993  .987  .998  .969  .846  1.110  1  .649  15.767  5  .008  96.591  1  .000  1.015  1.012  1.018  .015  .001  .006  .015  .165  1  .685  1.006  .977  1.037  .000  .000  .014  1  .906  1.000  1.000  1.000  .002  .002  1.844  1  .174  1.002  .999  1.006  -1.139  .394  8.371  1  .004  .148  .692  1.490 .600 6.161 a Variable(s) entered on step 1: InterDen, Rat_RDCom.  1  .013  NRD_2000_Ca pped Crow_DistCom InterDen Rat_RDCom Constant  151  .320 4.436  Route directness to park model Table B-21. Active Walking - Route Directness to Park, Logistic Regression Omnibus Tests of Model Coefficients Chi-square Step 1  df  Sig.  Step  6.357  2  .042  Block  6.357  2  .042  Model  121.419  24  .000  Model Summary -2 Log likelihood  Step 1  Cox & Snell R Square  Nagelkerke R Square  407.806(a) .103 .273 a Estimation terminated at iteration number 20 because maximum iterations has been reached. Final solution cannot be found. Hosmer and Lemeshow Test Step 1  Chi-square  df  15.284  Sig. 8  .054  Contingency Table for Hosmer and Lemeshow Test Active = .00 Step 1  Active = 1.00  1  Observed 112  Expected 112.862  Observed 1  2  112  111.451  3  110  110.929  4  109  5 6  Total  Expected .138  Observed 113  0  .549  112  2  1.071  112  111.189  4  1.811  113  109  109.319  3  2.681  112  110  108.175  2  3.825  112  7  106  106.536  6  5.464  112  8  103  104.140  9  7.860  112  9  106  99.061  6  12.939  112  10  73  76.337  38  34.663  111  Classification Table(a) Observed  Predicted Percentage Correct  Active .00 Step 1  Active  .00 1.00  Overall Percentage  1.00  .00  1045  5  64  7  99.5 9.9 93.8  a The cut value is .500  152  Variables in the Equation B  S.E.  Wald  df  Sig.  Exp(B)  95.0% C.I.for EXP(B) Lower  Step 1(a)  TotVeh  -1.018  .229  19.676  1  .000  .361  .231  .567  .482  .274  1.620  .947  2.770  gender(1)  3.103  1  .078  education  8.226  5  .144  Income  8.150  7  .320  ethnicity  3.752  5  .586  11.994  1  .001  1.017  1.007  1.027  .056  .062  1  .803  1.014  .909  1.131  .000  1.755  1  .185  1.000  .999  1.000  1.678  2.022  .689  1  .407  5.355  .102  281.858  2.528  1.163  4.722  1  .030  12.527  1.281  122.475  -4.040  2.826  2.045  1  .153  .018  nr  .017  .005  NRD_2000_Capped  .014  Crow_Dist  .000  Ratio_RD Ratio_SW_Str Constant  Upper  a Variable(s) entered on step 1: Ratio_RD, Ratio_SW_Str.  Work versus Non-work Local Travel As noted in the thesis body, most of the trips in the final sample were for non-workrelated purposes, which is to be expected given the focus on local, within-neighbourhood travel for this study. A logistic regression was run on two separate subsets (work and non-work travel) of the final travel data sample, and the model generated better explanatory power for workrelated (r-square = .332) than for non-work (r-square = .171) travel. Table B-22. Nonwork Trip Logistic Regression Omnibus Tests of Model Coefficients Chi-square Step 1  Step 1  Step  Df  Sig.  8.107  2  .017  Block  8.107  2  .017  Model  520.183  22  .000  -2 Log likelihood  Model Summary Cox & Snell R Nagelkerke R Square Square  4254.346(a) .108 .166 a Estimation terminated at iteration number 5 because parameter estimates changed by less than .001. Hosmer and Lemeshow Test Step 1  Chi-square 19.303  df  Sig. 8  .013  153  Classification Table(a) Observed  Predicted Walk Behavior (0 = None; 1 = Yes) .00  Step 1  Walk Behavior (0 = None; 1 = Yes)  .00  Percentage Correct  1.00  .00  3484  73  97.9  790  203  20.4  1.00 Overall Percentage  81.0  a The cut value is .500 Variables in the Equation B  S.E.  Wald  df  Sig.  Exp(B)  95.0% C.I.for EXP(B) Lower  Step 1(a)  TotVehicle Hhsize  -.372  .064  34.040  1  .000  .689  .608  .781  -.201  .043  22.078  1  .000  .818  .752  .889  47.187  7  .000 .620  .380  1.013  Income Income(1)  -.478  .251  education gender(1) EthnDummy(1)  Upper  -.125  .079  3.636  1  .057  30.213  5  .000  2.532  1  .112  .882  .756  1.029  .171  .127  1.808  1  .179  1.187  .925  1.522  -.011  .003  14.118  1  .000  .989  .983  .995  nr  .013  .002  64.459  1  .000  1.013  1.010  1.017  NRD_2000_Capped  .055  .019  7.884  1  .005  1.056  1.017  1.097  Crow_DistCom  .000  .000  .234  1  .629  1.000  1.000  1.000  age  Rat_RDCom  -.508  .433  1.373  1  .241  .602  .257  1.407  Ratio_SW_Str  .492  .214  5.297  1  .021  1.635  1.076  2.485  Constant  .154  .566  .074  1  .786  1.166  a Variable(s) entered on step 1: Rat_RDCom, Ratio_SW_Str.  Table B-23. Work Trips Logistic Regression Omnibus Tests of Model Coefficients Chi-square Step 1  df  Sig.  Step  18.754  2  .000  Block  18.754  2  .000  Model  225.298  19  .000  Model Summary Step 1  -2 Log likelihood 691.303(a)  Cox & Snell R Square .232  Nagelkerke R Square .353  a Estimation terminated at iteration number 6 because parameter estimates changed by less than .001. Hosmer and Lemeshow Test Step 1  Chi-square 12.040  df 8  Sig. .149  154  Classification Table(a) Observed  Predicted Walk Behavior (0 = None; 1 = Yes) .00  Step 1  Walk Behavior (0 = None; 1 = Yes)  .00 1.00  1.00  Percentage Correct .00  623  34  94.8  128  67  34.4  Overall Percentage  81.0  a The cut value is .500 Variables in the Equation B  S.E.  Wald  df  Sig.  Exp(B)  95.0% C.I.for EXP(B) Lower  Step 1(a)  TotVehicle Hhsize  -1.431  .184  .317  .109  Income education gender(1) EthnDummy(1) nr Rat_RDCom  Upper  60.741  1  .000  .239  .167  .343  1.373  1.109  1.700  8.483  1  .004  17.849  7  .013  19.905  5  .001  .473  .200  5.621  1  .018  1.606  1.086  2.375  -.907  .341  7.094  1  .008  .404  .207  .787  .016  .003  21.289  1  .000  1.016  1.009  1.023  -2.525  1.206  4.381  1  .036  .080  .008  .852  2.362  46.696  Ratio_SW_Str  2.352  .761  9.542  1  .002  10.502  Constant  -.986  1.499  .433  1  .511  .373  a Variable(s) entered on step 1: Rat_RDCom, Ratio_SW_Str.  5. Hosmer-Lemeshow and other diagnostics on regression models DiagnosticsBoth logistic and linear regression models were diagnosed for how well they fit the data. While there were outlier residuals in both sets of models, the outlying cases did not appear to exert a high degree of influence (based on Cook’s and Leverage statistic values for these models). Table B-24. Physical Activity – Full Logistic Regression, Fit Diagnostic (Hosmer & Lemeshow) Case Processing Summary Unweighted Cases(a) Selected Cases Included in Analysis Missing Cases Total Unselected Cases Total  N 1205  Percent 86.9  182  13.1  1387  100.0  0  .0  1387  100.0  a If weight is in effect, see classification table for the total number of cases.  155  Dependent Variable Encoding Original Value .00  Internal Value 0  1.00  1 Categorical Variables Codings\  IncomeDummy dummy of education variable gender  Frequency  Parameter coding  (1)  (1)  .00  233  1.000  1.00  972  .000  578  1.000  1.00  627  .000  1.00  554  1.000  2.00  651  .000  .00  Beginning Block, PA – Logistic Regression Classification Table(a,b) Observed  Predicted whether someone walked 30 minutes or not .00  Step 0  whether someone walked 30 minutes or not  .00  1.00  Percentage Correct .00  1132  0  100.0  73  0  .0  1.00 Overall Percentage  93.9  a Constant is included in the model. b The cut value is .500 Variables in the Equation B Step 0  Constant  S.E.  Lower -2.741  Upper .121  Wald Lower 515.335  df  Sig.  Upper  Exp(B)  Lower .000  1  Variables not in the Equation Step 0  Variables  TotVeh  df 1  Sig. .000  gender(1)  6.397  1  .011  EdDummy(1)  2.876  1  .090  13.205  1  .000  63.193  4  .000  IncomeDummy(1) Overall Statistics  Score 52.598  156  Upper .064  Block 1: Method = Enter, PA Logistic Regression Omnibus Tests of Model Coefficients Chi-square Step 1  df  Sig.  Step  68.466  4  .000  Block  68.466  4  .000  Model  68.466  4  .000  Model Summary Step 1  -2 Log likelihood 482.370(a)  Cox & Snell R Square .055  Nagelkerke R Square .151  a Estimation terminated at iteration number 6 because parameter estimates changed by less than .001. Hosmer and Lemeshow Test Step 1  Chi-square  df  8.582  Sig. 7  .284  Contingency Table for Hosmer and Lemeshow Test whether someone whether someone walked 30 minutes or walked 30 minutes or not = .00 not = 1.00 Observed Step 1  Expected  Observed  Expected  Total Observed  1  109  108.434  0  .566  109  2  163  162.620  2  2.380  165  3  162  166.079  8  3.921  170  4  130  128.144  2  3.856  132  5  145  144.961  7  7.039  152  6  113  114.797  8  6.203  121  7  109  107.425  8  9.575  117  8  139  134.429  15  19.571  154  9  62  65.112  23  19.888  85  Classification Table(a) Observed  Predicted whether someone walked 30 minutes or not .00  Step 1  whether someone walked 30 minutes or not  .00 1.00  Overall Percentage  1.00  Percentage Correct .00  1132  0  73  0  100.0 .0 93.9  a The cut value is .500  157  Variables in the Equation B  S.E.  Wald  df  Sig.  Exp(B)  95.0% C.I.for EXP(B) Lower  Step 1(a)  TotVeh  Upper  -1.273  .201  40.282  1  .000  .280  .189  .415  .718  .256  7.869  1  .005  2.051  1.242  3.387  -.504  .264  3.638  1  .056  .604  .360  1.014  .062  .306  .041  1  .839  1.064  .584  1.939  -1.165 .347 11.260 1 .001 a Variable(s) entered on step 1: TotVeh, gender, EdDummy, IncomeDummy.  .312  gender(1) EdDummy(1) IncomeDummy(1) Constant  Block 2: Method = Enter, PA Logistic Regression Omnibus Tests of Model Coefficients Step 1  Chi-square 21.841  Step  df 1  Sig. .000  Block  21.841  1  .000  Model  90.308  5  .000  Model Summary -2 Log likelihood  Step 1  Cox & Snell R Square  Nagelkerke R Square  460.529(a) .072 .197 a Estimation terminated at iteration number 6 because parameter estimates changed by less than .001. Hosmer and Lemeshow Test Step 1  Chi-square 5.881  df 8  Sig. .661  Contingency Table for Hosmer and Lemeshow Test whether someone whether someone walked 30 minutes or walked 30 minutes or not = .00 not = 1.00  Step 1  Total  1  Observed 124  Expected 123.251  Observed 0  Expected .749  Observed 124  2  118  119.470  3  1.530  121  3  118  118.888  3  2.112  121  4  119  119.158  3  2.842  122  5  119  118.362  3  3.638  122  6  117  117.286  5  4.714  122  7  112  114.911  9  6.089  121  8  117  113.470  5  8.530  122  9  109  107.767  12  13.233  121  10  79  79.438  30  29.562  109  158  Classification Table(a) Observed  Predicted whether someone walked 30 minutes or not .00  Step 1  whether someone walked 30 minutes or not  .00 1.00  Percentage Correct  1.00  .00  1130  2  69  4  99.8 5.5  Overall Percentage  94.1  a The cut value is .500 Variables in the Equation B  S.E.  Wald  df  Sig.  Exp(B)  95.0% C.I.for EXP(B) Lower  Step 1(a)  TotVeh  Upper  -1.031  .204  25.488  1  .000  .357  .239  .532  gender(1)  .582  .262  4.916  1  .027  1.789  1.070  2.992  EdDummy(1)  -.489  .269  3.309  1  .069  .613  .362  1.039  IncomeDummy(1)  -.110  .314  .123  1  .726  .896  .484  1.658  .019  .004  22.293  1  .000  1.019  1.011  1.027  -2.116  .414  26.158  1  .000  .120  nr Constant  a Variable(s) entered on step 1: nr.  Block 3: Method = Enter, PA Logistic Regression Omnibus Tests of Model Coefficients Step 1  Step  Chi-square 7.583  Df 2  Sig. .023  Block  7.583  2  .023  Model  97.891  7  .000  Model Summary Step 1  -2 Log likelihood 452.946(a)  Cox & Snell R Square .078  Nagelkerke R Square .213  a Estimation terminated at iteration number 7 because parameter estimates changed by less than .001. Hosmer and Lemeshow Test Step 1  Chi-square 9.381  df 8  Sig. .311  159  Contingency Table for Hosmer and Lemeshow Test whether someone whether someone walked 30 minutes or walked 30 minutes or not = .00 not = 1.00  Step 1  Total  1  Observed 121  Expected 120.532  Observed 0  Expected .468  Observed 121  2  119  119.915  2  1.085  121  3  121  119.170  0  1.830  121  4  116  118.460  5  2.540  121  5  115  117.605  6  3.395  121  6  117  116.493  4  4.507  121  7  116  115.073  5  5.927  121  8  114  112.704  7  8.296  121  9  111  107.429  10  13.571  121  10  82  84.620  34  31.380  116  Classification Table(a) Observed  Predicted whether someone walked 30 minutes or not .00  Step 1  whether someone walked 30 minutes or not  .00 1.00  1.00  Percentage Correct .00  1130  2  68  5  99.8 6.8  Overall Percentage  94.2  a The cut value is .500 Variables in the Equation B  S.E.  Wald  df  Sig.  Exp(B)  95.0% C.I.for EXP(B) Lower  Step 1(a)  TotVeh gender(1)  -1.003  .205  23.856  1  .000  .367  .245  .549  .577  .263  4.797  1  .029  1.780  1.062  2.982  EdDummy(1)  -.409  .271  2.267  1  .132  .665  .390  1.131  IncomeDummy(1)  -.119  .315  .144  1  .705  .887  .479  1.645  .016  .004  14.347  1  .000  1.016  1.008  1.025  -2.306  1.343  2.949  1  .086  .100  .007  1.385  1.676  .922  3.303  1  .069  5.344  .877  32.572  -1.413  1.699  .692  1  .406  .243  nr Rat_RDCom Ratio_SW_Str Constant  a Variable(s) entered on step 1: Rat_RDCom, Ratio_SW_Str.  Other Hosmer & Lemeshow Results: Table B-25. Walking or Not – Person, Logistic Regression Omnibus Tests of Model Coefficients Chi-square Step 1  Upper  Step  6.145  df  Sig. 2  .046  Block  6.145  2  .046  Model  154.908  20  .000  160  -2 Log likelihood  Step 1  Model Summary Cox & Snell R Nagelkerke R Square Square  1304.750(a) .123 .173 a Estimation terminated at iteration number 5 because parameter estimates changed by less than .001. Hosmer and Lemeshow Test Step 1  Chi-square 4.634  df  Sig. .796  8  Contingency Table for Hosmer and Lemeshow Test whether person walked whether person walked or not = .00 or not = 1.00 Observed Step 1  Expected  Observed  Expected  Total Observed  1  110  107.323  8  10.677  118  2  97  100.875  21  17.125  118  3  99  96.134  19  21.866  118  4  87  92.043  31  25.957  118  5  85  87.803  33  30.197  118  6  83  82.907  35  35.093  118  7  81  77.939  37  40.061  118  8  75  71.490  43  46.510  118  9  60  60.572  58  57.428  118  10  41  40.915  79  79.085  120  Classification Table(a) Observed  Step 1  whether person walked or not  Predicted whether person walked or not .00 1.00 .00 1.00  Percentage Correct .00  755  63  92.3  266  98  26.9 72.2  Overall Percentage a The cut value is .500 Variables in the Equation B Step 1  Hhsize  S.E.  Wald  df  Sig.  Exp(B)  90.0% C.I.for EXP(B) Lower Upper  -.113  .075  2.278  1  .131  .894  .790  1.010  .164  .195  .708  1  .400  1.178  .855  1.624  TotVeh  -.407  .111  13.331  1  .000  .666  .554  .800  gender(1)  -.130  .137  1.100  .992  .983  1.000  .018 .001 .000  .003 .030 .000  .343 .662 .121 .001 .000 .970 .336  .701  .005  1 5 1 5 1 1 1  .878  -.008  .897 3.246 2.408 20.674 32.594 .001 .925  1.018 1.001 1.000  1.013 .954 1.000  1.023 1.051 1.000  .359  .792  .206  1  .650  1.432  .389  5.272  .885 .364 5.907 Constant -.290 1.243 .054 a Variable(s) entered on step 1: Rat_RDCom, Ratio_SW_Str.  1 1  .015 .815  2.423 .748  1.331  4.410  IncomeDummy(1)  ethnicity age education nr NRD_2000_Capped Crow_DistCom Rat_RDCom Ratio_SW_Str  161  Table B-26. Driving or Not - Person, Logistic Regression Omnibus Tests of Model Coefficients Chi-square df Sig. Step Step 1 Block Model  3.546  2 .170  3.546  2 .170  230.725 21 .000 Model Summary  Step -2 Log likelihood Cox & Snell R Square Nagelkerke R Square 948.845(a)  1  .175  .280  Hosmer and Lemeshow Test Step Chi-square df Sig. 13.268  1  8 .103  Linear Regression Diagnostics: The following is an example of the diagnostics conducted for the linear regression models. Figure B-3. Standardized Residual Histogram  Dependent Variable: DistWalk 400  Frequency  300  200  100  0 -2  0  2  4  6  Regression Standardized Residual  162  8  Figure B-4. Standardized Residual Normal P-P Plot  Dependent Variable: DistWalk  Expected Cum Prob  1.0  0.8  0.6  0.4  0.2  0.0 0.0  0.2  0.4  0.6  0.8  1.0  Observed Cum Prob  Figure B-5. Residual v. Predicted Values Scatterplot  Dependent Variable: DistWalk  Regression Standardized Residual  8  6  4  2  0  -2 -2  0  2  Regression Standardized Predicted Value  163  4  Appendix C. Other - Graphics & Images This appendix includes miscellaneous graphics and images generated over the course of the research that were not included in the final report/thesis body. Figure C-1. Fused Grid Design  Image courtesy of CMHC/Grammenos, et al. 2005  Figure C-2. Un-sidewalked Local Streets, Seattle study area  Light gray indicates presence of unsidewalked streets.  164  Figure C-3. Intersection Density  Seattle/King County region: Darker points indicate higher # of intersections per sq km around households; lighter points indicate lower intersection density.  Figure C-4. Distribution of Route Directness Disparity Across Study Region  Light circles indicate higher pedestrian route directness to two nearby destinations and dark circles a higher vehicular directness.  165  Figure C-5. Fused Grid Street Design Schematic: disparate pedestrian and vehicular networks  Source of Fused Grid image: Grammenos, et al. 2005  Figure C-6. Example of cul-de-sac with pedestrian-only connection through public park space  Jackson neighbourhood, Seattle. Photo by Chris Hawkins  166  Figure C-7. Seattle Street Network (Capitol Hill)  GIS schematic highlighting areas (circled intersections) of higher relative pedestrian directness. Dots are destinations on the network.  Figure C-8. Seattle, I-90 Trail Neighborhood Connection  167  

Cite

Citation Scheme:

        

Citations by CSL (citeproc-js)

Usage Statistics

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.24.1-0066199/manifest

Comment

Related Items