This item has the following downloads:
1 A STATEWIDE ANALYSIS OF THE INTERACTION BETWEEN THE BUILT ENVIRONMENT AND TRAVEL BEHAVIOR USING GEOGRAPHICALLY WEIGHTED REGRESSION By RUSSELL ERIC PROVOST A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF ARTS IN URBAN AND REGIONAL PLANNING UNIVERSITY OF FLORIDA 2013
2 2013 Russell Eric Provost
3 ACKNOWLEDGMENTS I thank my family and friends who continued to encourage me to fini sh my thesis. I also thank Florida Department of Transportation for providing such a rich dataset for analysis. I especially thank the University of Florida c ommunity for six un forgettable years in Gainesville.
4 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 3 LIST OF TABLES ................................ ................................ ................................ ............ 6 LIST OF FIGURES ................................ ................................ ................................ .......... 7 LIST OF ABBREVIATIONS ................................ ................................ ............................. 8 ABSTRACT ................................ ................................ ................................ ..................... 9 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 11 2 LITERA TURE REVIEW ................................ ................................ .......................... 14 Trends ................................ ................................ ................................ ..................... 14 VMT ................................ ................................ ................................ .................. 15 Funding and Costs ................................ ................................ ........................... 16 Emer ging Greenhouse Gas Emissions Legislation ................................ .......... 17 The Built Environment and Travel Behavior ................................ ............................ 19 Definitions ................................ ................................ ................................ ......... 19 Theoretical Framewor ks ................................ ................................ ................... 21 Empirical Studies ................................ ................................ .............................. 24 Aggregate studies ................................ ................................ ...................... 25 Disaggregate studies ................................ ................................ ................. 27 3 METHODOLOGY ................................ ................................ ................................ ... 36 Model Development ................................ ................................ ................................ 36 Data and Variables ................................ ................................ ................................ 38 4 RESULTS ................................ ................................ ................................ ............... 49 Linear Linear Global R egression Model Summary ................................ ................. 49 Log Linear Global Regression Model Summary ................................ ..................... 51 Log Log Global Regression Model Summary ................................ ......................... 54 Linear Linear GWR Regression Model Summary ................................ ................... 56 5 DISCUSSION ................................ ................................ ................................ ......... 70 6 CONCLUSION ................................ ................................ ................................ ........ 74 LIST OF REFERENCES ................................ ................................ ............................... 76
5 BIOGRAPHICAL SKETCH ................................ ................................ ............................ 82
6 LIST OF TABLES Table page 2 1 Common operationalizations of the built environment ................................ ........ 35 3 1 Mo del parameters ................................ ................................ ............................... 44 4 1 Descriptive Statistics of Study Sample ................................ ............................... 61 4 2 OLS Model Outputs ................................ ................................ ............................ 62
7 LIST OF FIGURES Figure page 3 1 Geocoded households from the NHTS add on ................................ ................... 46 3 2 Accessibility search area based on the average NHTS shopping trip length. ..... 47 3 3 Flow of employees between census tracts. ................................ ........................ 48 4 1 Location of examples describing la nd use coefficients. ................................ ...... 64 4 2 Localized R Squared values from the GWR model. ................................ ........... 65 4 3 Accessibility coefficient surface. ................................ ................................ ......... 66 4 4 Density coefficient surface. ................................ ................................ ................. 67 4 5 Minimum Commute coefficient surface. ................................ .............................. 68 4 6 Travel Delay coefficient surface ................................ ................................ .......... 69
8 LIST OF ABBREVIATION S y i Yearly VMT at household i x iSES A vector of socioeconomic variables at household i SES yearly VMT from a one unit change in the ve ctor of socioeconomic variables x iTA The travel attitude of household i TA yearly VMT fr om the travel attitude variable x iBE A vector of built environment variables thought to impact the quality of travel, quantity of travel needed, and the cost of vehicle travel BE yearly VMT from a one unit change in the vector of built environment variables x iC An additional travel cost v ariable C yearly VMT from a one unit change in the travel cost variable i Unobserved fact ors impact household yearly VMT GIS G eographic I nformation S ystems GWR G eographically W eighted R egression GHG Greenhouse Gas NHTS National Household Transportation Survey NTD N eo T raditional D evelopment TMC T heoretical M inimum C ommute VMT Vehicle Miles Traveled
9 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master s of Arts in Urban and Regional Planning A STATEWIDE ANALYSIS OF THE INTERACTION BETWEEN THE BUILT ENVIRONMENT AND TRAVEL BEHAVIOR USING GEOGRAPHICALLY WEIGHTED REGRESSION By Russell Eric Provost August 2013 Chair: Ruth Steiner Cochair: Paul Zwick Major: Urban and Regional Planning Concerns about traffic congestion, air pollution, climate change, transportation revenue shortfalls, and obesity rates are coalescing to form a perfect storm that is challenging the way development and transportation investments occur. Confronting these challenges will require a paradigm shift in the way politicians and citizens understand the synergies betwe en the built environment and travel behavior. Justifying policy changes to address th ese issues require a thorough understanding of the interactions between the built environment and travel behavior. Current and past research has predominantly found that factors in the built environment including density, diversity, and design influence travel behavior. These models, however, assume that these relationships do not vary across geographical space ignoring spatial non stationarity Spatial non stationarity i s the phenomenon when r elationships between the dependent and independent variables vary across geographic space. This research affirms that there is a relationship between the built environment and travel behavior after controlling for socioeconomic variables but that the directionality and magnitudes of these relationships often vary across the state of
10 Florida. I t is also demonstrated that the proposed model explains a greater amount of variability in south Florida and the Tampa area than in north Florida. Regional measures of the bui lt environment were found to have the strongest influence at reducing VMT, partic ularly in urban areas. This research demonstrates that non stationarity is an important consideration in the study of transportation behavior.
11 CHAPTER 1 INTRODUCTION Concerns about traffic congestion, air pollution, climate change, transportation revenue shortfalls, and obesity rates are coalescing to form a perfect storm that is challenging the way development and transportation investments occur. Confronting these challenges will require a paradigm shift in the way politicians and citizens think about and understand the synergies between the built environment and travel behavior. Justifying policy changes to address these issues requires a thorough understanding of the interactions between the built environment and travel behavior. Th e interaction between the built environment and travel behavior is potentially the most studied subject in the field of transportation and land use planning. Researchers have conducted a plethora of studies ranging in research designs, variables, and results. The subject has inspired meta analyses ( e.g., Leck, 2006; Ewing & Cervero, 2001; Ewing & Cervero, 2010 ) congressional commissioned reports ( e.g., National Research Council, 2009) books, and numerous journal articles. Despi te the existence of this rich body of literature areas of research related to these relationships still need to be explored Thanks to the growing availability of built environment and transportation spatial data, opportunities exist to explore and add to the existing body of literature regarding the complex synergies between the built environment and travel behavior. A majority of the research published regarding the built envi ronment and travel behavior relies on global models to infer statistical relat ionships between the independent and dependent variables Global models assume a constant relationship between the response and explanatory variables and ignore spatial non stationarity
12 Spatial non stationarity is the phenomenon with which modeled relatio nships vary across geographic space. These spatial variations are hidden in traditional regression modeling, masking any regional differences in the behavior of regression coefficients and model performance This significant hole in the literature leads to an important question; does the interaction between the built environment and travel behavior vary across the rural, suburban, and urban development gradients? By taking advantage of improved analytical techniques using geographic information systems (GIS ), this thesis examines how the relationships between factors in the built environment and travel behavior vary across the state of Florida. This research aims to answer two fundamental questions. The first step is to determine if there are associations between factors in the built environment and travel behavior while controlling for attitudes, socioeconomic variables, and travel costs throughout the entire state of Florida. The second step examines if these associations are consistent throughout the sta te of Florida. It is hypothesized that doubling the density, accessibility, and diversity in downtown Miami (in urban environment) will not have the same impact on travel behavior as doubling the density in Pahokee (a rural environment). To examine thes e two research questions, three global OLS models are developed that model household vehicle miles traveled (VMT) while controlling for attitudes, travel costs, and socioeconomic factors. Once a viable global model is determined, two logarithmic transforma tions are undertaken for interpretability. Finally, to determine the presence of non stationarity in the study area, a geographically weighted regression (GWR) model is developed and interpreted. Coefficients developed
13 by the GW R model will be compared ac ross rural, suburban, and urban municipalities to explore the differences in how the built environment impacts travel behavior throughout the State of Florida.
14 CHAPTER 2 LITERATURE REVIEW Trends This section examines the current trends that illustrate the importance of understanding the potential impacts of the build environment on travel behavior. The passage of the National Defense and Interstate Highway Act of 1956 signaled that subsidizing private automobile travel would be a priority of the federal government. With billion to be expended between 1957 and 1969 ( Boarnet, 2011, p.198). This massive infrastructure investment, coupled with cheap energy, rising incomes, mass produced automobiles, and a cultural love affair with the private vehicle has created a car dependent populous. Although more recent federal legi slation including the 2005 the Safe Accountable, Flexible, Efficient Transportation Equity Act: A legacy for Users (SAFETEA LU) has allocated $ 52.6 billion to transit, the private automobile has maintained its hegemony over personal travel (Federal Transit Administration ( FTA ) 2012) The most recent federal transportation legislation, the Moving Ahead for Progress in the 21st Century Act (MAP 21), allocates a meager 2% of the authorized States Department of Transportation (DOT), 2013 a ). This indicates that the Federal Government is intent at continuing its bias towards investments that favor private single occupancy vehicles. Recently, however, the costs to society of our dependence on t he private automobile have become increasingly apparent. Rev en ue streams cannot keep up with needed investments in surface transportation infrastructure. Despite attempts to alleviate congestion through increased federal spending annual hours of delay con tinues to
15 grow throughout America ( Shoup & Lang 2011). Aside from an aging and ever more expensive surface transportation network, concerns over climate change and greenhouse gas emissions continue to spark debate. The Federal Surface Transportation Poli cy and Planning Act of 2011 explicitly states two objective s ; reduce the national per capita motor VMT annually and reduce carbon dioxide levels by 40% by 2030 (S. 326, 2011). Although this language was not incorporated into the MAP 21, signed into law by President Obama on July 6, 2012, congestion reduction and environmental sustainability are core elements of the latest transportation legislation ( DOT, 2013 b ). VMT The growth rate of VMT in America far outpaces population growth. Between 1 982 and 2007 it is estimated that VM T increased by 189% nationally (National Research Council The U.S. Department % from 2005 to 2030 l outpacing population growth by 23% (as cited in Ewing, Bartholomew, Winkerman, Walters, & Chen, 2007, p.52). Although the recent recession has managed to quell the annual growth in VMT for the first time since 1980, a recovery in the economy is likely t o bring about a return to VMT growth (Puentes & Tomer, 2008). Florida, perhaps more than any other state, has not been immune to the growing dependency on the private automobile. For example, a ccording to the American Society for Civil Engineers, travel seventeen years between 1990 and 2007 (2008). This unprecedented growth in VMT continues to put a strain on federal and state resources as infrastructure projects helplessly attempt to keep up with demand. Mor e VMT also means more congestion
16 which manifests itself as a household cost through wasted travel time and additional gas expenditures Funding and Costs S hrinking revenues for new transportation projects and maintenance of existing surface transportation infrastructure throughout the County is a growing concern The surface transportation systems cost households and businesses nearly $130 billion (2011, p. 1). These cost s mainly include vehicle operating costs and travel time delays. 2005 (as cited in Blanco, Steiner, Peng, Shmaltsuyev, and Wang, 2010). Unfortunately the Interstat e 35 Bridge collapse in Minneapolis in 2007, although an extreme case, is infrastructure. According to the Center for Urban Transportation Research (CUTR) after examining th e long range transportation plans of each Metropolitan Planning Organization in the State of Florida, 20 year unfunded transportation needs stand at $74.3 million statewide in 2008 (2012). States and the federal government rely primarily on receipts from fuel taxes to fund transportation projects. In Florida to increases in fuel efficiency, it is projected that fuel tax revenue in Florida will grow at only 8% from fiscal year 1999/00 to fiscal year 2019/20 lagging well behind the projected 82% increase in VMT during the same time period (CUTR, 2012). T he emerging trend of increasing transportation project costs and dwindling transportation revenues s pells trouble for the future of surface transportation
17 infrastructure throughout the country. These estimates, however, assume the continuation of the status quo appeasement of meeting capacity driven metrics that Since the advent of the mass produced car, drivers have been able to ignore the marginal social cost of the private vehicle and the burning of fossil fuels. Emerging climate change legislation, however, is attempting to address the negative externalities of fossil fuel usage. Emerging Greenhouse Gas Emissions Legislation The United States decision not to ratify the Kyoto Protocol in December of 1997 has not shielded it from the changing political landscape regarding climate change. On April 2, 2007 the Supreme Court in Massachusetts v. United States Environmental Protection Agency (EPA) 549 U.S. 497, declared that greenhouse gasses (GHGs) are pollutants, and therefore regulated under the Clean Air Act (CAA) (EPA, 2009). The Supreme Court directed the EP A to determine the contribution of GHGs from new motor vehicles to air pollution therefore endangering the public welfare. Nearly two years after door to GHG regulation at the federal lev el (EPA, 2009). Some states, however, have already begun regulating GHG emissions. first global warming legislation, in 2006 has appeared to have caused reverberations through out the political landscape in this Country. AB 32 requires the state to reduce GHG emissions by 27 % in 2020. Preceding this legislation, however, Governor Schwarzenegger issued Executive Order 3
18 emissions to 200 0 levels by 2010, reducing emissions to 1990 levels by 2020, and reducing emissions to 80 % Senate Bill 375 was signed into law by Governor Schwarzenegger in 2008 which specifically targets emission stand ards in the transportation sector and procedures to meet those standards. According to the legislation, even after considering increases in necessary to achieve si gnificant additional [GHG] reductions from changed land use patterns and improved transportation ( Sustainable Communities and Climate Protection Act of 2008 2008). The bill r equires each region within the S tate to incorporate reduction targets into their regional transportation plan culminating in a sustainable communities strategy (Shaheen, et al., 2009). The bill goes on to state that planning models used to asses transportation infrastructure decisions must be updated to oices, such as residential development patterns, Sustainable Communities and Climate Protection Act of 2008 2008). da signed three executive orders aimed at curbing climate change. Executive order 07 127 r by 80 % 2009). In 2008 the Florida legislature strengthened and showed support for the greenhouse gases and energy efficiency in local comprehensive plans (Florida Department of Commu
19 Energy and Climate Change (2008 ) there are 21 states, including Florida and California, that have climate action plans underway or completed. The transportation sector is responsible for 28 % of the United States GHG emissions (Ewing, et al.,2007). To date, particularly at the federal level, GHG reduction strategies in the transportation sector have been aimed at reducing the reliance of carbon intensive fuels, and improving the fuel efficien cy of vehicles. Future growth in vehicle miles of travel, however, is anticipated to outweigh any reduction in GHG emissions created by such efforts. For example, despite C passage of legislation mandating tougher fuel economy standards to 35 mile s per gallon ( MPG ) by cars and light trucks would be 40 % above the 1990 level in 2030 even if these standards were adopted nati onwide (Ewing, et, al., 2007). The Built Env ironment and Travel Behavior Many in the urban planning and transportation field have suggested that the built environment can serve as a means to reduce demand of the private vehicle and help address the greenhouse gas and funding issues raised in the pr ior sections. The first section attempts to clarify what is meant by the built environment and travel behavior. Then, theoretical frameworks are examined that attempt to understand and conceptualize travel behavior. Finally specific studies are reviewed th at e mploy various forms of the theoretical frameworks and concepts discussed. Definitions In order to move forward in the discussion of the built environment and travel behavior, one should have a clear understanding of what underlying concepts are being conveyed when using the terms
20 general concepts can be discussed, particularly regarding the built environment, methods of operationalizing these concepts vary significantly as will be seen in later s ections. It is still useful, however, to begin the discussion with general understanding of the two main variables in this study. The built environment is a somewhat nebulous concept that invokes many different feelings for different people and hence the difficult y of objectively measuring its components. Some definitions are more abstract than others. For example Carmona, Heath, Oc, & Tiesdell describe d the built environment as being p.134). They go on to describe urban design and the built environment consisting of six dimensions: morphological, perceptual, social, visual, functional, and temporal (Carmone, et al., 2003). Many of these dimensions are abstract by nature and good quan titative research demands more objective concepts. Handy took a more objective approach and conceptualize d the built environment using three components ; land use patterns the spatial distribution of human ) the transportation system ( ysical infrastructure and the services that ), and design the aesthetic qualities the environment) (2005 p.5 ). Cervero and Kockelman took a similar, but slightly more precise approach by conceptualizing the built en vironment using three dimensions: density, diversity, and design (1997). Later destination accessibility and distance to transit were added to capture more dimensions of the built envir onment (Ewing & Cervero, 2010) and the totality of the built environme nt descriptors were
21 pite gaining traction, research ers continued to operationalize individual s in number of ways ( Table 2 1). Travel behavior also has many components. Mode choice, trip distance, vehicle miles t raveled ( VMT) ( a composite of vehicle trip length and vehicle trip frequency), and trip chaining/complexity have all been the subject to investigation. For the purposes of the National Household Travel Survey, trip chaining between two anchors (we call this a tour, such as between home and work) that is direct or has The individual or household is frequently the subject of such travel behaviors. It is important t o note that factors in the built environment may impact each of these behaviors differently. Theoretical Frameworks Travel behavior is a complex phenomenon which requires knowledge of an array of research fields and concepts. A robust conceptual model, as Acker, Van Wee, & Witlox describe d involve combining and linking theories stemming from not only (2010, p.2). Early frameworks were very aggregate in nature and lac ked behavioral content and interdisciplinary frameworks and almost exclusively toward analysis of long term, capital intensive expansion of the transportation system, primarily in the form of highwa 2006, p.285). After decades of accommodating the vehicle at all costs, including displacing huge numbers of inner city residents, the externalities of single occupancy vehicle dependency began to mount. Beginning in the 1980s forecasting travel to influencing travel, and, as such, it became necessary to
22 conceptualize travel as the outcome of a host of decisions made by the traveler (Boarnet, 2011, p.199). With the growing popularity of planning concepts such as transit oriented development, smart growth, new urbanism, and neo traditional design, planners began to justify particular planning arrangements as a method to reduce automobile dependency. In order to justify these claims, more decentralized beha vioral based models had to be devised. Boarnet and Crane invoked the theory of microeconomics and the derived consumer demand theory that follows the logic that travel is a derived mize a environment, therefore, is thought to influence the price of travel and therefore impact the perceived disutility of particular forms of travel (Boarnet & Crane, 2000). An extension of the utility maximizing theory is the activi ty based approach. Handy stated based approach takes its starting point that the demand for travel is deriv he utility the relevance of uncertainty, the importance of habit and thresholds, the role of constraints, and the influence of levels of adequate information and knowledge & Hensher, as cited in Handy, 2005, p.11). With activity b ased frameworks, the focus shifted from attempting to understand in a vacuum to utility maximizing framework, the built environment is thought to influenc e the costs of various modes of transportation and the disutility of each.
23 Chatman surmised that there are two underlying assumptions when studying the built environment and travel behavior ; household preferences for travel that vary with socioeconomic characteristics and urban ndividuals maximize utility over trips subject to a time budget constraint ( 2005, p.9). Both approaches h ave their share of weaknesses with the first omitting trip time o r money cost often neglects variances in direct utility due to built environment variation, because speed and distance are assumed to be the mechanism through which any influ ences of the built environment on travel behavior would occur (Chatman, 2005, p.10). The question then becomes influential mechanism in influencing travel behavior its impact on travel costs (direct utility) or through th e quality of travel (indirect utility) Chatman argued that the built types of travel characteristics; the qualities of travel that directly affect the utility of the travel experience; the quantity of travel inputs needed to produce out of home activities; and the per unit prices of 2005, p.15). All of the aforementioned conceptual models assume that travel is an indirect demand. Some research has shown however that ce rtain travel may serve as a positive utility, giving credence to the adage that life is a journey, not a destination For example, M ed more than 1900 residents in the San Francisco Bay Area and quarters of th e sample reported sometimes or may offer positive utility adds complicat
24 minimizations principle that underlies a great deal of policy making as well as virtually all regional travel demand forecasting models ( Mokhtarian & Chen, 2003 p.1). The theory of planned behavior offers insight into human action and travel behavior According to the theory of planned behavior y three kinds of considerations; beliefs about likely consequences of the behavior (behavioral beliefs), beliefs about the normative expectations o f others (normative beliefs), and beliefs about the presence of factors that may further hinder performance of the Schmidt (p.175). Although this theory has performed fai rly well in empirical studies, the past more accurately predicts its future fre quency of performance than does stated Garling, 1998, p.131). These theories, although not exhaustive, provide for a solid foundation when interpreting empirical studies examining travel behavior. Theory aids in deciphering theoretical framework deconstructed the role of the built environment on travel behavior into three components: the qualities of travel, the quantity of tra vel needed, and the per unit price of travel. Using this framework one can conjecture how density may impact each of the components The following section reviews empirical studies examining travel behavior across various disciplines. Empirical Studies Researchers have relied on s everal typologies over the past two decades when investigating the built environment and travel behavior. According to Crane, some of the
25 more relevant variations include the travel purpose under study, the nature and level of d etail in the data, and the how the characterization of the built environment is operationalized (as cited in Leck, 2006). The lack of a standardized method for defining dependent and dependent variables has made it difficult to compare results across the m yriad of empirical studies dealing with the role of the built environment on travel behavior. For example, in a meta analysis conducted by Ewing and Cervero, o f the 31, 23, 22 and 22 studies reviewed that examined VMT with respect to density, diversity, design, and destination accessibility respectively, there were 9 methods of operationalizing density, 11 methods of o perationalizing diversity, 14 methods of operationalizing design, and 8 methods of operationalizing accessibility (2010). Aggregate s tudie s A renewed interest in the impact of the built environment and travel behavior surfaced in the early 1990s with the foundation of the Congress for New Urbanism. The movement, conceived by architects Peter Calthorpe Andres Duany, Elizabeth Plater Zyberk and others, promulgated transit oriented development and neo traditional development (NTD) as a means to reduce demand for the private vehicle (Calthorpe & Fulton, 2001). Earlier studies were fairly aggregate and crafted to examine the impact of these generalized concepts on travel behavior. Often neighborhoods were aggregate studies found a relatively strong relationship between neighborhood type and travel behavior. Quasi experimental designs typically find the strongest relationships between neighborhood design and distance traveled (Ewing & Cervero, 2010). Guiding principles such as gridded street networks, mixed use neighborhood centers, and pede strian
26 friendly environments provide a higher percentage of mode split, and higher internal capture rates. Levinson and Kumar (1997) suggest ed that density may be used as a substitute for city size. They analyze d 38 US cities to investigate the effec ts of residential density on travel behavior. Their regression analysis showed that distance and time are negatively related with density while auto travel time seems to have a threshold density at 10,000 people per square mile. Once density exceeds 10,000, auto travel time shows positive associations with density. They argue that beyond a certain density level automobile travel is less attractive because of increasing traffic congestion. The works of Handy (1996 ), Ewing et al. (1994), and Rutherford (1996) sugg est ed neighborhoods designed with NTD principles produce both shorter trips and fewer trips when compared to conventional suburban subdivisions. Shay & Khattak (2005) found that people in NTDs compared to residents of a conventional cookie cutter suburban neighborhood (p. 10). No differences were found, however in auto ownership. Cervero and Radisch (1996) discovered that residents of a pre s more likely to go to a store or other non work destination b war suburban neighborhood counterpart ( p. 122). Interestingly, the number of non work trips taken was statistically equal between the two neighborhoods but residents of the u rban neighborhood substituted many of their potential driving trips with walking trips. Other studies suggest, however, that residents of NTD neighborhoods make more trips than their conventional neighborhood counterparts. A study commissioned by the Oreg on Transportation Research and Education Consortium (2011) found residents of NTD s made more trips than non NTD residents, however, no difference in VMT was
27 found, suggesting NTD residents make more internal and non motorized trips Although the neighborho od type dummy variable was the best predictor of total trip making, the motorized trips when the built walking and cycling tri ps might be promoted through improved street connectivity and Treating neighborhood type as a dummy variable allowed the researches to Findings as these suggest that quasi experimental designs make no effort to isolate the effects of specific land use features and indeed factors like density and mixed u uses are accepted as co G eographical information systems have made it possible for a myriad of disaggregate studies to be published with varying degrees of sophistication, methods, and variables. Disa ggregate s tudies Disaggregate studies tend to find that the built environment h as a modest impact on travel behavior. According to Ewing and Cervero (2010) over 200 studies examine the relationship between the built environment and travel behavior that employ various levels of sophistication, variables, controls, and data sources. Fa ced with limitations in creating true experimental designs (you cannot randomly select and relocate a group of households into a rural or urban area and observe their changes in travel behavior), establishing causality between the built environment and tra vel behavior has been difficult Nevertheless the majority of studies find a correlation between built environment variables and travel behavior.
28 Many studies find accessibility (operationalized in a number of ways) to be the most influential built envir onment variable in regards influencing travel behavior. Kockelman analyzed VMT per household using only socioeconomic variables (household size, auto ownership, and income) in a base model and then added built environment variables to ascertain if the mod included the built environment variables (accessibility, and land use mix) improved the u Kockelman 1991, p.27). Accessibility was found to be the strong est influence on household VMT with an elasticity of .31 (a 100 % increase in accessibility reduces VMT by 31 % ) A similar conclusion was made by Ewing and Ce r vero in their meta analysis of 50 disaggregate studies that examined travel and the built environm ent. The weighted elasticities of VMT with respect to density, diversity, design, destination accessibility, and distance to transit estimated in the study reveal a very modest impact of each variable on VMT. Destination accessibility (o perationalized as distance to downtown) incurred that largest impact with a weighted average elasticity of VMT of .22. Neither density, diversity, nor distance to transit reached a weighted average elasticity of greater than .09. Ewing and Cervero conclude d however, that travel could be quite large (Ewing & Cervero, 2010, p.275). Bento and C ropper found that features in the built environment such as city shape, road density, population centrality, distribution of employment, and transit road density, rail supply (for rai l cities) and jobs housing balance is to change average annual miles driven by at most .7% for each 5 p. 475 ).
29 There appears to be a synergistic relationship between the variables, however, and changing each variable in conce rt with one another can have a relatively large impact on travel behavior. Bento and C ropper estimated a 25 % reduction in household VMT when a hypothetical family moves from Atlanta, Georgia to Boston, Massachusetts (Bento & Cropper, 2005 ) Many research ers have discovered that the spatial resolution in which the built environment is measured is an important consideration when estimating travel demand models. Steiner et al. (2010) used highly disaggregate data to determine the influences of the built env ironment on trip lengths. Findings suggest ed that the same built environment variables at the parcel, neighborhood, and regional scales can impact trip lengths differently depending on if the trip is being produced or attracted to the particular location. For example, non work trips were found to be shorter if produced at a location closer to a regional activity center (a hub of commercial activity within the region). A location, however, that is closer to a regional activity center will attract longer non work trips. This research demonstrated not only the importance of examining the built environment at both the trip origin and destination, but also the interrelationships between the parcel, neighborhood, and context within the region. Susan Handy (1993) also recognized the importance of differentiating the influences of various spatial scales on travel behavior. Using a conventional exponential form of the gravity model to calculate local and regional accessibility, Susan H andy estimated the relative impo rtance of both measures when analyzi ng travel behavior. She defined 5 ). Regional accessibility is defined as
30 centers, such as suburban shopping malls or downtown that although both measures impact travel behavior, regional accessibility appears to be more influential. Many resear chers, however, raised concerns with the issue of self selection. Bhat a nd Guo (2007) admonish ed that locate themselves in neighborhoods and then, based on neighborhood attributes, determine their travel beha This is an important tend to be more affluent, have more cars, live in a larger household, be more auto oriented and prefer larger space than their counterparts in urban areas, a result of residential self If residents self se lect into neighborhoods that suit their travel needs, results from studies could be biased. The research suggests that attitudes do have an impact on travel behavior. An experiment that used K means clustering to group over 600 individuals into six discernible groups based solely on attitudes found that the average VMT of each group was significantly different from o ne another. Of the socioeconomic variables measured, Although attitudes can impact one the built environment can curtail or exasperate travel behavior manifestations of these attitudes.
31 Studies suggest that even after controlling for attitudes, the built environment plays an important ro randomly selected individual moves from an inner ring suburb to an exurb, we expect selection plays a negligible role however, if the comparison is made between households located in the inner ring suburb and urban area. In this s cenario, self selection accounted for 50% of the increase in VMT of inner suburbanites compared to urbanites. This suggests that attitudes have a bigger role to play with regards to travel behavior for individuals living in urban or semi urban environments. Zhou and Kockelman conduct ed a similar study that attempts to simulate a treated/untreated approach to test the impact of a hypothetical move from a suburban area to a rural ar ea. Findings suggest ed selected household is expect ed to increase its daily VMT by 17 miles when living in a rural or suburban neighborhood, as compared to living in the CBD of urban neighborhood (p.10). Researchers estimate d s election accounts for 42% of observed VMT differences across Austin households in suburban or rural versus CBD Another attempt to isolate the importance of attitudes and self selection on travel behavior is a study conducted by Sch anen and Mokhtarian (2005) what degree a lack of congruence between physical neighborhood structure and
32 (p.127). Using principal component analysis with data obtained from a 14 page questionnaire targeting attitudes, the authors identified four basic traveler types to compare their travel habits; true urbanites, dis sonant urban dwellers, true suburb anites, and dissonant suburbanites. Researchers conclude structure appears to exert a stronger influence on distance traveled than do preferences In a thorough review of the literature, Cao, Mokhtarian, and Handy analysis of 38 empirical studies found that found a statistically significant influence of the built environment remained after accounting for self selection (2006). In sum, previous literature suggests that travel behavior can be moderated by factors in the bui lt environment. Although the review presented here is certainly not exhaustive and concepts such as internal capture, trip chaining and tour complexity have gone unmentioned, certain generalizations can be made. One is the importance of understanding the nexus between the built environment and travel behavior in light of several disturbing trends; climate change, a gei ng infrastructure, and shrinking revenue. With climate change legislation becoming reality, local policy makers will need to invoke a balance d approach to meet greenhouse gas reduction targets. Justifying the built environment as one part of the solution requires unambiguous results from the research community. In reviewing the literature presented here, a few themes crystallize. First is the lack of standard definitions when operationalizing built environment variables. Although it appears accessibilit y is the most effective at reducing VMT, it is difficult to separate the importance of the concept with how it is being measured It is also im portant to
33 consider spa tial scale. The built environment can be operationalized at the parcel, neighborhood, and regional scales. The interdependency and different impacts of these scales are important to consider. Finally the nature of the phenomenon under study makes it difficult to conclusively make a causal link between the built environment and travel behavior. Despite the limitations, advances in the field have attempted to mimic a true experimental design through the use of propensity score matching. Results from these methodologies suggest a causal link can be implied. The absence of any acknowledgement of the potential existence of spatial non stationarity within the body of research presented leaves a void in the otherwise robust set of literature examining travel behavior and the built environment Spatial non stationarity space or that there might be some problem with the specification of the model from which the relationships are being measured and which manifests itself in terms of p.282). Every piece of literature reviewed assumes a global model when iso lating the effects of the built environment on travel behavior. For example, the elasticities estimated by Ewing and others suggest that doubling the density, accessibility, and diversity would have the same impact on travel behavior in rural Florida as it would in downtown Miami. Ali, Patridge, & Olfert (2007) conclude d that global models such as (p.301). Brunsdon, Fotheringham, & Charlton (1996) propose d a technique call geographically weighted regression (GWR) to address spatial non stationarity. This
34 paper attempts to utilize the GWR method to augment the existing built environment/travel behavior global models. Chapter 2 demonstrates that reducing VMT nationally is an important endeavor for many reasons. The growing acknowledgment that an increased reliance on the private automobile will continue to put pressure on budgets and the environment has sparked a myriad of studies investigating potential methods to reduce VMT Many of these empirical studies have focused on the role of the built environment at achieving a reduction in demand for the private automobile. In many instances, evidence supports lling for self selection. These studies, however, assume these relationships are consistent across geographical space, ignoring spatial non stationarity.
35 Table 2 1. Common operationalizations of the built environment Variable Operationalization Density Net residential density (number of residential units/residential area) Gross residential density (number of residential units/total area) Population density(total population/total area) Employment density(total jobs/total area) Diversity Design Destination Accessibility Entropy (see Cervero & Kockelman 1997) Dissimilarity index ( see Cervero & Kockelman 1997) Proportion of each land use type Connected node ratio Percentage of 4 way intersections Block length Pedestrian route directness Intersection density Job accessibility by auto Job acessilbibilty by transit Various gravity model iterations (see Bhat et al., 2000 for literature review)
36 CHAPTER 3 METHODOLOGY T his m ethodology seeks to determine the correlation between the built environment and travel behavior. The built environment is assumed to impact travel that directly affect the utility of the tr avel experience, the quantity of the travel inputs needed to produce out of home activities, and the per unit prices of travel by different ment of some of these mechanisms, it is argued the available resources provide enough information to formulate a methodology that reflects this conceptual model. As stated earlier, this methodology also seeks to determine if the relationship between the built environment and travel behavior via these mechanisms, is constant across Florida For interpretation purposes, three global models and one GWR model are developed. Model Development In this study, household VMT is first modeled using a global OLS linear regression structure. The residents of the household must have maintained tenure for at least a year Also, the house had to have been geocoded to at least the intersection (as opposed to the zip code as some had been). After applying these restrictions,9985 households were av ailable for analysis. The base model is as follows; y i = 0 + SES x iSES + TA x iTA + B E x iBE + C x i C + i (see the list of abbreviations for definitions). Two additional variations of the b ase model are also developed for further coefficients. A log linear model is developed using the base model in which the natural logarithm of household yearly VMT is the dependent variable;
37 Ln(y i )= 0 + SES x iSES + TA x iTA + BE x iBE + C x iC + i. According to UCLA Statistical Consulting Group (2013) the percentage increase in the outcome variable from a one unit increase in a independent variable can b e estimated by calculating the exponentiated value of the coefficient (exp( )) in a log linear regression model The second variation of the base model is the log log transformation The log log model is developed using the base model in which the natural logarithm of the dependent variable (yearly household VMT) and independent variables are taken; Ln(y i )= 0 + SES Ln( x iSES ) + TA x iTA + BE Ln( x iBE ) + C Ln( x iC ) + i. According to UCLA: Statistical Consulting Group (2013) the percentage change in the outcome variable from a corresponding percentage change in a predictor variable can be calculated by raising the percentage change in the independent variable by the (( x i BE2/x i BE1)^ BE ). T hese interpretive concepts will be applied to the VMT model in the subsequent results section. A GWR model is developed that allow s cients to vary regionally (Mitchell, 2005, p.219) in order to determine how relationships between factors in the built environment and travel behavior vary across the rural, suburban, and urban gradients of Florida. GWR reduces the sphere of influence wh en determining model outputs to a local and or regional scale dependent upon a prescribed kernel. GWR is employed and takes the following form; y i (g) = 0 + SES (g)x iSES + TA (g)x iTA + BE (g)x iBE + C (g)x iC + i
38 w here the model parameters are the same as previously described for the base global regression model with the additional g parameter and coefficients are for a single geographic location ( Mitchell, 2005, p.219) Data and Variables The 2009 National Household Transportation Survey (NHTS) provided a rich set of variables for this study including the dependent variable, yearly household VMT. The NHTS was initiated in 1969 (formerly known as the National Persona l Transportation Survey) and collected every five to seven years throughout the Country. The NHTS is need comprehensive data on travel and transportation patterns in the United States rips taken in a 24 hour period. ( Federal Highway Administration, 2013 ). States and Metropolitan Planning Organizations (MPOs) have the opportunity to purchase and participate in the add on program making larger and more complete samples available that allo w for more accurate modeling. Florida Department of Transportation (FDOT) decided to participate in the 2009 NHTS add on program. The result was a geographic stratified sample of over 14,000 households throughout the state of Florida ( Figure 3.1). The add on includes the travel diary, household and personal socioeconomic data, information regarding perceptions and attitudes, vehicle data, and the locations of each household, workplace, origin and destination. The purpose of this study is to isolate the imp acts of the built environment on travel behavior. To isolate these impacts, several socioeconomic variables are incorporated into the global and local models ( Table 3 1). Total Household income was reported via 18 categories. The first category represented an income of $5,000 or less
39 with each subsequent category increasing by $4,999 ($5,000 $9,999, $10,000 $14,999, etc.) until a category of $100,000 or more is reach ed. The methodology derives a semi continuous variable by taking the midpoint of each income category. The values $5,000 and $100,000 are utilized for the bottom and top categories respectively. Additional control variables in the global and local models a re household size, the number of workers, head of household retirement status, the number of drivers, the number of kids, and the number of commercial vehicles owned by the household. The global and local models attempt to control for self selection. As p art of the interview process, the NHTS survey asked each head of household hat is the most important reason you c Survey, 2009, p.5)? The respondent could choose from a set of predefined answers or respond in an open ended format. The attitude dummy variable was coded one if the signifying that the household may have self selected into the neighborhood due to travel preferences and a ttitudes A ll other answers including the cost/price of the home, the school system, and home or lot size were coded with a zero The model includes two variables that are believed to influence the quantity of the travel needed to satisfy the desired out of home activities of each household. 004 p. 105). Accessibility in the global and local model s repre sents the ease of travel to shopping and office establishments f rom an individual household. To calculate the accessibility index for each household in the NHTS survey, a 2010 statewide parcel dataset was
40 collected for the state of Florida. Retail and office parcels were extracted from the larger database leaving 177,865 records for analysis. Each parcel record contains the total conditioned square footage space on the property. An origin destination (OD) matrix was calculated for each NHTS household with the household serving as the origin and each retail/office parcel centroid serving as the destination. Due to the unmanageable size of each OD matrix, a network search distance of 6.3 miles which represents the average shopping trip distance in the N HTS, was applied ( Florida Department of Transportation 2010). A routable transportation network that contains the travel cost in minutes to traverse each link was used to calculate the amount of time it takes to travel by automobile to each retail/office parcel within 6.3 miles of the NHTS household ( Figure 3 2 ). Once the cumulative opportunities (square footage) and travel costs (minutes) were extracted from the datasets, the conventional Hansen accessibility formula based on the gravity model was applie d to each NHTS household ( Table 3 1). The accessibility variable described above provides insight into the character of the immediate neighborhood around e ach NHTS household. It does not however capture the spatial structure of the region. The theoretica l minimum commute (TMC) represents the find a job as close to home as possible under the assumption that actual residential locations and job locations are maintained and the total distance travell ed (by all workers together) is housing ratio which is insensitive to its context within the region, the TMC is a proxy of the urban structure at the regional scale (Horner, 2 006).
41 To calculate the TMC ( Table 3 Longitudinal Employer Household Dynamics (LEHD) were extracted for the state of Florida from OnTheMap ( http://onthemap.ces.census. gov/ ) application. The data contained the number of employees and employers in each income category (less than $1250/month, between $1250 $3333/mo nth, and more than $3333/month) residing w ithin each census block in 2009 for the entire state of Florida. Using a customized tool in ArcGIS Desktop, the optimum allocation of employees between each census block was determined for each income category. The customized tool relied upon the Network Analyst extension and a linear optimization algorithm to assign ea ch employee to an employer in such a way that minimized the total distance traveled statewide while satisfying the total supply of employees ( Figure 3 3). The destination census tract in the example provided in Figure 3 3 contains over 33,000 jobs making between $1250 and $3333 per month. The lines indicate the flow of employees falling within that income category into the census tract to meet th e demand of employers in the destination census tract. Due to the unmanageable size of each OD matrix using cens us blocks, the data was aggregated to the census t ract. Intrazonal trips were assigned a distance of SQRT(census area/PI) following the procedure undertaken by Frost, et al., 1998. The total distance for each census tract was added together and divided by the number of employees to obtain the MTC. This process was repeated three times for each income category. This guaranteed that employees earning a particular wage were assigned a job who paid a similar wage. The TMC for each income category was then inte rpolated using the Natural N eighbor interpolation technique within the Spatial Analyst Extension for ArcGIS
42 Desktop. This technique finds the closest subset of input samples to a query point and applies weights to them based on proportionate areas to inter polate a value (Sibson, 1981).This interpolation method is very local in nature and produced lower root mean square prediction errors than the Inverse Distance Weighted and Kriging interpolation methods. Once the continuous TMC surfaces were interpolated, the values were extracted to the NHTS households where the three TMC values for each income category were averaged for the final TMC variable for the global and local regression models. Cost is an important mechanism through which the built environment c an impact travel behavior. Although variables such as density can serve as proxies for cost (denser areas are thought to be more prohibitive on the private vehicle and more accommodating to transit), this methodology attempts to directly measure a trip del ay ratio. Within the NHTS, respondents were asked to report the total time each recorded optimal time. The optimal travel time was calculated using a statewide routable network that includes th e time to traverse each link. Each recorded trip in the NHTS includes the geocoded origin and destination. The optimal travel time for each trip was calculated using the Network Analyst Extension in ArcGIS Desktop. The reported travel time was then divided by the optimal travel time. This travel delay ratio is less than one when the reported travel time is less than the optimal travel time and greater than one when the reported travel time is larger than the optimal travel time. Larger ratios imply a greater cost associated with the
43 particular trip. The travel delay ratio was calculated for each trip in the NHTS and averaged for each household and included in the local and global regression models. Density was calculated using 20 10 census blocks. The number of residential units in each census block was divided by the total area to obtain the gross residential density The vector polygons were then converted to a raster with 100 square meter cells. To avoid a hard edge between cens us blocks, focal statistics was applied to calculate the mean value of each cell within a rectangle neighborhood of three cells. The values were then extracted to the NHTS households for inclusion in the global and local regression models. The dependent v ariable, total household VMT, was derived from the vehicle database included in the NHTS The vehicle database included information about each of the s vehicles. Vehicle owners were aske d to record the VMT during the past year for each of th e vehicle s owned. Vehicles owned for less than a year were extrapolated Households who moved to the current location within the past year of the NHTS were removed from the analysis. Household VMT was calculated from T belonging to the same household.
44 Table 3 1. Model parameters Variable Model Parameter Calculation Source Household Income x iSES Midpoint of recorded income range NHTS Household size x iSES Number of household members NHTS Number of household workers x iSES Number of employed household members NHTS Household retirement status x iSES Binary variable (1,0) coded 1 if head of household reported as retired NHTS Number of household children x iSES Number of children 18 years of age or younger NHTS Number of household commercial vehicles x iSES The number of reported commercial vehicles owned by the household NHTS Household travel attitude x iTA Binary variable (1,0) coded 1 if the head of household indicated proximity to destinations was the main reason for staying in/purchasing the house NHTS Accessibility x iBE where, = Accessibility index for household j. = parcel attractiveness (square feet) t ij = the network travel time to reach parcel j from household i Derived from parcels
45 Table 3 1. Continued Variable Model Parameter Calculation Source Theoretical minimum commute x iBE Minimize H = g iven: where, H = total distance traveled within the state of Florida to match workers and jobs within the same income category. n = number of census tracts O i = number of workers in census tract i D j = number of jobs in census tract j d ij = network distance between centroids of census tract i and census tract j t ij = number of trips between census tract i and census tract j Derived from Census LEHD data Travel Delay Ratio x i C ( )/ t i where, rt ij = the reported travel time for trip k between origin i and destination j ot ij = the calculated optimal travel time for trip k between origin i and destination j t ij = the number of trips reported for household i Derived from NHTS Density x iBE The number of household units per square mile Derived from 2010 Census blocks
46 Figure 3 1. Geocoded households from the NHTS add on
47 Fi gure 3 2 Accessibility search area based on the average NHTS shopping trip length
48 Fi gure 3 3 Flow of employees between census tracts
49 CHAPTER 4 RESULTS For comparative and interoperability purposes, findings from the three global regression models are first discussed followed by the results from the GWR model. As with the prevailing body of literature, findings suggest a correlation between the built environment and travel behavior, but global models may inhibit a robust interpretation of the interaction between the two. The built environment variables are highly skewed throughout the State as indicated by the standard deviation ( Table 4 1). Overall, GWR models perform slightly better due to a reduction in unexplained variation when predicting household VMT. The GWR model, however performs best in Southeast Florida and the Tampa area. The GWR models indicate ther e is variation across the Stat e of Florida with regards to model coefficient s, and in some instances, change directionality All coefficients are deemed significant at the 95 % confidence interval and are non standardized Linear Linear Global Regression Model Summary Overall model per formance is indicated by the adjusted r squared value, and the Joint Wald Statistic. The adjusted r squared value of the linear linear OLS model is .43 indicating the model explains 43 % of the variability among household yearly VMT ( Table 4 2 ). The Joint % confidence level probability is zero indicating the model is statistically significant. Additional model statistics are the Koenker's studentized Bruesch Pagan (BP) statistic and the Jarque Bera statistic. The BP statistic assess if non stationary (model coefficients vary across space) and/or heteroscedasticity exists (the relationship between the dependent and independent variable is not consistent throughout the dataset). The BP % confidence
50 level probability is zero indicating non stationary and/or heteroscedasticity exists Finally, the Jarque Bera statistic assesses model bias, and when significant, indicates the residuals are not normally distributed signifying model misspecification. The Jarque Bera statistic 95 % confidence level probability is zero indicating the results should be interpreted with caution because one or more key variables are missing from the model. In a traditional OLS re gression model, coefficients are interpreted as indicating the expected change in the dependent variable (household VMT) from a one unit change in the independent variable when all other covariates are held constant. All of the variables are statistically significant with the exception of the attitude dummy variable. All coeff icients indicate the expected directionality. An increase in a higher household VMT. For example, for every additional worker, a household can expect an increase in yearly VMT of vehicle count by one will increase the yearly VMT by 3,368 miles. If a head of household is retired, that household is estimated to reduce its yearly VMT by 2,010 miles. An increase in one unit of ac cessibility (a unitless variable) reduces yearly household VMT by .002 miles. To put this in perspective, a NHTS household in downtown Miami was measured to have an accessibility index of 1,418,125.5. A NHTS household in rural Belle Glade was measured to h ave an accessibility index of 212,929 resulting in a difference of 1,205,196.5 ( Figure 4 1) Holding all other variables constant, less than the Belle Glade household. An increase in one residential unit per acre
51 reduces yearly VMT by 39.2 miles. To put this in perspective, a NHTS household in downtown Tampa was estimated to have a gross residential density of 9.15 units per acre. A NHTS household in rural Greenville was estimated to have a gross residential density of .65 acres, a difference of 8 .5 units per acre ( Figure 4 1) Due to this difference in densities, the household in Greenville can expect to incur 3 33.2 VMT more per year, all other things being equal. A one mile increase in the minimum commute, an indicator of the job housing balance throughout the region, increases yearly household VMT by 75.7 miles. For example, a NHTS household in Miami Beach was estimated to have a minimum commute distance of .36 miles. A NHTS household in rural Madison was estimated to have a minimum commute of 48.8 miles, a difference of 48.44 miles ( Figure 4 1) All other things being equal, the better regional jobs housing balance in the Miami Beach area affords a savings of 3,666.9 m iles a year compared to a household located in the Madison area. Finally, an increase in one unit in the travel delay ratio, an indicator of travel cost, decreases yearly VMT by only .56 miles. A household with a travel delay ratio of five, which would ind icate a household experiences on average travel delays equaling five times the amount of time during optimal conditions, would incur 2.8 VMT l ess per year than a household who experiences no delays. Although seemingly ionality and statistical significance affirms the hypothesis that higher costs, as measured in travel delay, reduces travel Log Linear Global Regression Model Summary In the log linear model the dependent variable, household yearly VMT, has been log transformed. O verall model performance is reduced slightly with the adjusted r squared value falling to .38 indicating the model explains 38% of the variability in the log
52 of household VMT ( Table 4 2 ). % confidence level probabi lity is zero % confidence level probability is zero indicating non stationary and/or heteroscedasticity exists. The Jarque % confidence level probability is zero indicating the results should be interpreted with caution because one or more key variables are missing from the model. When the outcome variable is log transformed binary (dummy) variables can be interpreted as the ratio of the geometric means for the t wo groups by taking the exponentiated value of the coefficient. For continuous variables, the exponentiated value of the coefficient can be interpreted as the percentage increase in the dependent variable from a one unit increase in the independent variabl e. In the log linear model, a ll of the variables are statistically significant with the exception of the attitude dummy variable and the trip delay ratio All coefficients indicate the expected directionality. An kers, vehicles, and commercial vehicles correlates with higher household VMT. For example, a household whose head is retired is expected to have a yearly VMT geometric mean 15 % less than a household with a non retired head of household ( ). Incre rly VMT by 27 % ( F or every additional worker, expected to increase 17 % Similar interpretations can be made for all of the variables The accessibility coefficient is rounded to six decimal places and therefore is reported to be zero despite being statistically significant. Statistically significant
53 variables d o not often have a coefficient of ze ro, however, its interpretation makes sense in the log linear regression framework. Accessibility is a unitless measure with each increment increase in its measurement representing little change. As noted above, the accessibility index of the rural town of Belle Glade, Florida is 212,929, The accessib ility index of Downtown Miami is 1,205,196.5. Therefore a one unit increase in the accessibility index is inconsequential, and should not be expected to reduce the percentage of yearly VMT by a measurable amount. Therefore, a one unit increase in accessibi lity can be expected to decrease yearly VMT by 0% An increase in o ne residential unit per acre reduces yearly VMT by .5 % Returning to the previous density example, a NHTS household in downtown Tampa was estimated to have a gross residential density of 9.15 units per acre. A NHTS household in rural Greenville was estimated to have a gross residential density o f .65 acres, a difference of 8. 5 units per acre ( Figure 4 1). According to the log linear model, c eteris paribus y VMT located in Tampa Bay is estimated to be 4.25 % less than the household located outside of Lake City due to the difference in density s a .5 % Returning to the previous example a NHTS household in Miami Beach was estimated to have a minimum commute distance of .36 miles. A NHTS household in rural Madison was estimated to have a minimum commute of 48.8 miles, a difference of 48.44 miles According to the log linear model, ceteris 25 % less
54 than the household located outside of Lake City due to the difference in the average minimum commute Log Log Global Regression Model Summary In the log log model the dependent variable and the continuous independent variables have been log transformed. T he adjusted r squared value is .40 indicating the model explains 40% of the variability in the log of household VMT ( Table 4 2 ) .The Joint % confidence level probability is zero, indicating the model is % confidence level probability is zero, indicating non stationary and/or heteroscedasticity exists. The Jarque Ber 95 % confidence level probability is zero, indicating the results should be interpreted with caution because one or more key variables are missing from the model. In a log log model, coefficients of log transformed independent variables can b e percentage change in an independent variable. This ratio is often referred to as an elasticity and is used extensively in economics. In the log log model, discrete variables s uch as household size, household workers, household vehicles, household commercial vehicles, and household children were not log transformed. Discrete variables were omitted from being log transformed because of the difficulty with dealing with zero values Continuous variables such as accessibility, density, minimum commute, and trip delay ratio were transformed and can be interpreted as elasticities. All variables demonstrated the expected directionality, but unlike the previous models, the average minimu m commute distance was not statistically significant. Since the control variables and dummy variables are discrete, and therefore not log transformed, the results are the same from the log linear model discussed above.
55 The income variable was transformed, however, and can be interpreted as an elasticity. A 10% increase in household income is associated with a 2.9 % increase in household yearly VMT Household income had the largest elasticity among the variables that were log transformed. A 10% increase in accessibility reduces yearly household VMT by only .2 % Although seemingly small, % larger than Belle Glade log linear model, ceteris paribus the 466 % increase in accessibility reduces household yearly VMT by 9.8 % ( .021*466). A 10% increase in gross residential density reduces yearly VMT by .4 % Although this is a highly inelastic relationship, context is needed. Returning to the example given in the l inear linear model summary, a household in Tampa, Florida was estimated to have a gross residential density of 9.15 units per acre. A household in rural Greenville, Florida was estimated to have a gross residential density of only .65 units per acre. The d ensity of the Tampa household is 1308 % higher than the Greenville household corresponding to a 52 % decrease in yearly household VMT ( .038 1380). Although insignificant in the log log model, the minimum commute results are discussed. A 10% increase in t he average minimum commute increases yearly VMT by .145 % Although these findings may seem insignificant, context is needed to interpret the results. Returning to the example examined in the linear linear model summary, a household in Miami Beach was estim ated to have an average minimum commute of .36 miles and a household in rural Madison County in Northern Florida was estimated to have an average minimum commute of 48.8 miles. minimum commute is 13,456 % more than the minimum commute for the Miami Beach
56 household corresponding to a 195 % increase in yearly VMT, all other things being equal. Finally, a 10% increase in the average trip delay ratio decreases yearly VMT by .7 % To put this in perspective, a household in downtown Miami was estimated to have a travel delay ratio 266 % greater than a household in High Springs (7.5 versus 2.3). According to the log log model, ceteris paribus the Miami household y early VMT is 18.6 % less than the household in High Springs due to the d ifferences in the travel delay ratio, a surrogate for travel costs. Linear Linear GWR Regression Model Summary The previous models ignore non stationarity, the phenomenon where regression coefficients vary across geographic space. To assess if this is occurs in the NHTS dataset, a GWR model is developed. Only variables found to be significant in the linear linear OLS model were incorporated into the two GWR models. The adjusted r squared value for the overall linear linear GWR model is .44, indicating t he model explains 44 % yearly VMT. This is a very slight improvement over squared value of .43. One useful way of comparing two or more regression models is the Akaike's Information Criterion (AIC). AIC is a relative measurement with smaller values representing a better goodness of fit. The AIC value for the linear linear OLS model is 209,023. The AIC value for the linear linear GWR model is 208,937 indicating it is superior to the global OLS model. In a GWR analysis, a local model is developed for every observation in the dataset. Model coefficients and outputs are calculated based on a kernel that imposes an extent around the unit under analysis. The kernels for this GWR model were designed to minimize the AIC while being dynamic in nature. From each of the local
57 regression models a local r squared value is calculated ( Figure 4 2). The points in dark green represent households whose r squared values are larger than .5 (the r squared value fo r the linear linear OLS model was .43) Generally speaking, the model fits better south of Orlando including in large urban areas of Tampa, and Southeast Florida. Interestingly, the one major exception is an area just north of Miami including Miami Beach, Hialeah, Miramar, and Hollywood. The model largely underperforms north of Orlando including Jacksonville, Tallahassee, and Pensacola. While examining the following coefficient surfaces, please keep in mind that orange and red are a s typically represent rela tionships not expected or that are not found in the literature. Examining the accessibility coefficient surface, an interesting pattern arises ( Figure 4 3). The coefficient for the accessibility index in the linear linear OLS was .02; a one unit increase in the accessibility index reduces yearly VMT by .02 miles. Areas in yellow illustrate where accessibility reduces VMT less than what was estimated in the global model. Areas in red delineate neighborhoods where there is a positive relationship between VM T and accessibility, that is, neighborhoods with greater with VMT is greater than the estimated relationship in the global OLS model in the areas delineated by the two shades of green. For the most part, positive relationships between accessibility and VMT are concentrated around major urban areas throughout south Florida and the rural/suburban area of south central Florida including Sebring, Lake Placid, and Avon Park. A positive relationship implies that an increase in neighborhood accessibility, and indicator of access to office and commercial activity, increases household VMT. This
58 relationship is contrary to what is stated in the literature and estimated in the globa l OLS models. Neighborhood accessibility appears to play a key role in reducing VMT in the Daytona Beach, Port Charlotte and Pensacola areas. The density coefficient surface is also intriguing. The coefficient for gross residential density in the linear linear OLS model was 39.2; a one unit increase in the residential density reduces yearly household VMT by 39.2 miles. In the GWR model, however, there is great variability in the density coefficient across the state of Florida ( Figure 4 4). Areas in red i ndicate a positive relationship between density and VMT,that is an increase in residential density increases VMT. This primarily occurs in Southeast Florida and around Lakeland. Areas in burnt orange exhibit a negative relationship between density and VMT but at a rate less than the OLS estimated This primarily occurs around Venice in Southwest Florida. The remainder of the colors signifies a stronger negative relationship between density and VMT than was estimated in the OLS model. Density appears to be an important mechanism to reduce VMT in the Tallahassee area with coefficients reaching 1,595 indicating a one unit increase in density reduces household VMT by 1,595 miles per year. Density also plays a k ey role in reducing VMT around the Sebring, Lake Placid, and Avon Park areas. Interestingly, this is the same area where there was a positive relationship between accessibility and VMT. The average minimum commute, a regional indicator of jobs housing bal ance, also varies across Florida ( Figure 4 5). The minimum commute coefficient in the linear linear OLS model was 75.7; a one mile increase in the average minimum commute increases household VMT by 75.7 miles. The larger the minimum commute, the more
59 uneve nly jobs and housing within the same income category occur throughout the region. Areas in red indicate a negative relationship between the average minimum commute and VMT. This primarily occurs within the interior of south Florida, a primarily rural area. Areas in orange indicate a positive relationship between the avera ge minimum commute and VMT but at a smaller rate than indicated by the OLS model. Of particular note are the areas in green which represent a highly positive relationship between average minimum commute and VMT. Major urban areas including Orlando, Tampa, and most of Southeast Florida indicate that the minimum average commute plays a major role in determining household VMT. Dark green areas indicate that a one mile increase in the average minimum commute increases yearly VMT by up to 826.7 miles. This seems to indicate that regional indicators may better model VMT in large urban areas than neighborhood oriented statistics. Finally, the travel delay ratio a surrogate for trip cost, coeffic ient surface appears to be more random ( Figure 4 6) The travel delay ratio coefficient in the linear linear OLS model was .56, indicating a one unit increase in the travel delay reduces VMT by .56 miles. Areas in orange and red indicate a very small nega tive relationship between the travel delay ratio and VMT ( .56 0) or a positive relationship. Areas in yellow and green signify a negative relationship greater than that estimated by the global OLS model. No clear pattern arises from the coefficient sur face. Of particular interest is Pinellas County just east of Tampa. Pinellas County is estimated to have a strong negative relationship between the travel delay ratio and VMT, indicating travel cost, such as congestion, is an important determent of VMT. Ot her urban areas estimated to
60 have a strong negative relationship is Miami and Stuart. This also occurs in relatively rural areas, however, including Levy Dixie, and Eastern Marion Counties.
61 Table 4 1 Descriptive Statistics of Study Sample Variable Mean Standard Deviation Household VMT 16,884 .2 11,242 .5 Household Income 55,743 30,484 Household size 2 .2 1 .1 Workers per household .88 .85 Children per household .3 4 79 Number of household commercial vehicles .03 18 Accessibility 168008 295352.6 Theoretical minimum commute 9. 4 10.5 Travel Delay Ratio 6.6 191.3 Density 3.1 4.8
62 Table 4 2 OLS Model Outputs Output Linear Linear Log Linear Log Log Intercept 1031.06 4 8.2 59 5.81 0 A ccessibility Coefficient (t statistic) 0.002 ( 2.884 ) 0 ( 5.838 ) 0.02 1 ( 5.350 ) Density Coefficient (t statistic) 39.15 0 ( 2.06 7) 0.00 5 ( 3.105 ) 0.03 8 ( 5.290 ) Average Minimum Commute Coefficient (t statistic) 75.662 (8.666) 0.00 5 ( 6.484 ) 0.014 ( 1.808 ) Travel Delay Ratio Coefficient (t statistic) 0.55 7 ( 1.25 2) 0.000063 ( 1.73 5) 0.069( 6.477 ) Household Income Coefficient (t statistic) 0.07 3 ( 23.017 ) 0 ( 6.484 ) 0.2 90 ( 26.54 6) Household Size Coefficient (t statistic) 1776.56 7 ( 10.91 ) 0.147 ( 11.062 ) 0.147 (11.175) Household Workers Coefficient (t statistic) 2460.934 ( 16.4 1) 0.156 ( 12.753 ) 0.15 3 ( 12.59 ) Household Vehicles Coefficient (t statistic) 3367.92 ( 28.95 6) 0.236 (24.8) 0.226 ( 24.044 ) Household Kids Coefficient (t statistic) 705.360 ( 3.35) 0.092 ( 5.357 ) 0.091 ( 5.35 5) Household Commercial Vehicles Coefficient (t statistic) 2032.65 8 ( 4.243 ) 0.10 2 ( 2. 60) 0.09 7 ( 2.507 )
63 Table 4 2 Continued Output Linear Linear Log Linear Log Log Retired Dummy Coefficient (t statistic) 2010.13 6 ( 8.28 ) 0.168 ( 8.46 9) 0.168 ( 8. 532 ) Attitude Dummy Coefficient (t statistic) 250.56 1 ( 1.205 ) 0.015 ( 0.888 ) 0.01 2 ( 0.708 ) R Squared Value .4 3 .38 .40 Significant at the 95% Confidence Interval N = 9985
64 Figure 4 1. Location of examples describing land use coefficients
65 Figure 4 2 Localized R Squared values from the GWR model
66 Figure 4 3. Accessibility coefficient surface
67 Figure 4 4 Density coefficient surface
68 Figure 4 5 Minimum Commute coefficient surface
69 Figure 4 6 Travel Delay coefficient surface
70 CHAPTER 5 DISCUSSION Florida is a peninsula which provides an excellent geographic space to conduct spatial analysis because it lacks anthropogenic influence along a majority of its borders. The three global OLS models developed demonstrated and reaffirmed that, after controll ing for socioeconomic variables, the built environment impacts travel behavior. The methodology failed however, to construct an adequate measure of attitudes in an from answers to a question in the NHTS designed to capture the main reason the head of household moved to their current residence. There could be a number of explanations for this failure; one being the question only reflected the head of Al expected directionality according to the literature. For each of the global models, an example provided context for the implications ples of the built environment on opposite ends of the rural/urban spectrum in Florida. In all cases the average minimum commute, an indicator of regional accessibility and / or job housing balance appeared to be the most influential at reducing VMT. accessibility was a better predictor of travel behavior than neighborhood accessibility (Handy, 1993) The magnitudes of the elasticities derived from the log log OLS model were small with none reaching the gre atest magnitude found by Ewing of .39 (Ewing & Cervero, 2010). The density elasticity with respect to VMT was .04, exactly what was significantly smaller that others est imated.
71 Just relying on the absolute magnitude of the coefficients and/or elasticities, however, is meaningless. Despite the small overall magnitude of the coefficients, their he difference in density between urban Tampa and rural Greenville reduces yearly VMT by 52 % all other factors being the same. Although this is an extreme example of the urban/rural must be noted that density is often seen as a surrogate for other hard to measure variables such as walkability, parking costs, and transit availability. These vari ables were not directly measured in this study but are assumed to vary with density. Alth ough the global models demonstrate the expected directionalities from all of the built environment variables, a closer examination using GWR reveal some interesting trends. The coefficient surfaces clearly demonstrate that non stationarity exists in the mo del. Also, by mapping the localized r squared values, it is clear that the model proposed in this research fits better in southern Florida and the Tampa area. The model fits the least in the major college towns of Gainesville and Tallahassee, and also the military town of Jacksonville. It is possible that these unique populations necessitate different models to explain travel behavior. In some instances in the coefficient surfaces, n eighborhood variables such as densit y and accessibility to shopping impact VMT in the opposite direction that was estimated in the global OLS models and found in the literature. Large portions of Orlando, Tampa, and Southeast Florida have a positive relationship between accessibility and VMT suggesting an increase shopping oppor tunities within six miles of the household increases VMT. There is also a positive relationship between density and
72 VMT in Southeast Florida. Perhaps within these areas, accessibility and density are largely uniform and other factors at measuring urban for m must be devised. Many of these same areas, however, have a highly positive relationship between minimum commute and VMT suggesting that regional job accessibility is an important mechanism to reduce VMT in urban areas. Rural areas provide a less amount of discernable patterns One generalized pattern is that minimum commute often does not have a positive relationship with VMT This suggests that the regional measure of job housing is not a viable mechanism to reduce VMT in rural areas. Again, these find ings could be from a lack of variability in the jobs housing balance in rural areas to decipher a viable relationship. Density and accessibility, for the most part, demonstrated the expected directionality in rural areas with varying degrees of magnitude. Density was particularly useful at reducing VMT in the Tallahassee and Sebring/Lake Placid areas. If density is in fact an intermediate variable as the literature suggests, other factors such as transit, parking supply, and crime may be important determina nts of VMT in these areas. Although the GWR analysis demonstrates coefficients can reverse their directionality, a closer examination of the model outputs indicate that this should be interpreted with caution. Using the standard error of the coefficients in the GWR output to calculate the regression coefficient 95% confidence interval reveals that many of the confidence intervals are quite large and include positive and negative values. For example an observation with a density coefficient of 281.6, an unexpected positive relationship, has a 95% confidence interval of by this very large confidence interval, we could still expect the coefficient for this
73 observation to be negative. Another caveat of interpreting GW R is the possibility of variables, although found to be significant in the global models, are not statistically significant in the local model. By utilizing GWR, researchers and policy analysts can differentiate between areas where models are accurate an d need adjustment. GWR provides a starting point to uncover unique relationships that would otherwise be overlooked using global regression techniques. In this study, it was demonstrated that traditional neighborhood built environment variables may not cap ture the necessary information in built out urban areas. Other variables may need to be devised. Modeling travel behavior is difficult due to the unique travel situations of the respondents. Answers to cross sectional studies regarding travel behavior can also be suspect especially when respondents are asked to recall certain travel information like the amount of miles driven within the past year for each vehicle owned. Nevertheless, this exercise provided insight into the varying relationship between the b uilt environment and travel behavior across Florida.
74 CHAPTER 5 CONCLUSION This thesis utilizes a statewide travel survey to conduct statistical analysis of the relationship between the built environment and travel behavior. First, current trends that demonstrate the saliency of the subject were discussed. Second, conceptual frameworks were examined that outlined the building blocks for an in depth examination of the subject. Third, empirical studies were reviewed to summarize the prevailing trends found in the literature. Fourth, the methodology section outlined the construction of three global OLS models and one GWR model. Attitudes, travel costs, and socioeconomic variables were controlled for in the models The built enviro nment was presumed to influence travel behavior through qualities of travel that directly affect the utility of the travel experience, the quantity of the travel inputs needed to produce out of home activities, and the per unit prices of travel by di The gl obal models, which assume constant relationships between the independent and dependent variables across space, affirmed this hypothesis. It was also hypothesized that these relationships would vary across the state of Flo rida. By utilizing GWR, it was clearly demonstrated that the relationships between factors in the built environment and travel behavior incur non stationarity. Future research should not ignore non stationarity and more work needs to concentrate on the var ying degrees of model goodness of fit across geographic space. If research is to inform policy decisions, a one size fits all model cannot be assumed. Also, as this study demonstrates, larger regions should be the focus of empirical research so that contex t can be given and model implications can clearly be articulated. This research provides the one of the few statewide analysis of the relationship between
75 the built environment and travel behavior that maps out and clearly identifies non stationarity in re gression analysis for this subject It is urged that future research on this subject matter utilizes the growing spatial statistical tools now available. Only then can more precise models be developed. This research is just the first step in developing a r obust travel demand model. Future research will focus on determining the mechanisms that make the proposed model fit better in south Florida and Tampa than north Florida. Other built environment variables should be developed for urban areas that capture su btleness that perhaps density and neighborhood accessibility cannot. Research that utilizes GWR is adaptive in nature, and requires the analyst to adjust the models as results dictate. This research is step one in that process.
76 LIST OF REFERENC ES Ali, K., Partridge, M., & Olfert, R. (2007). Can Geographically Weighted Regressions Improve Regional Analysis and Policy Making? International Regional Science Review 30 (3), 300 329. American Society of Civil Engineers. (2008). Report Card for America Infrastructure Retrieved from http://www.infrastructurereportcard.org/state page/florida American Society of Civil Engineers. (2011). Failure to Act: The E conomic I mpact of C urrent Investment Trends in S urface Transportation Infrastructure Reston, VA. Retrieved from http://www.asce.org/uploadedFiles/Infrastructure/Report_Card /ASCE FailureToActFinal.pdf Anable, J. Identifying Travel Behavior Segments Using Attitude Theory. Transport Policy 12, 65 78. Bamberg, S., Ajzen, I., & Schmidt, P. (2010). Choice of Travel Mode in the Theory of Planned Behavior: The Roles of Past Behavior, Habit, and Reasoned Action. Basic and Applied Social Psychology 25 (3), 175 187. Retrieved from http://dx.doi.org/10.1207/S153 24834BASP2503_01 Baussauw, K., Derudder, B., & Witlox, F. (2011). Measuring Spatial Separation Proces ses Through the Minimum Commute: the Case of Flanders. European Journal of Transport and Infrastructure Research 11(1), 42 60. Bento, A., Cropper, M., Mobarak, A., & Vinha, K. (2005). The Impact of Urban Spatial Structure on Travel Demand in the United States. The Review of Economics and Statistics 87, 466 478. Bhat, C., & Guo, J. (2007). A Comprehensive Analysis of Built Environment Characteristics on Household Residential Choice and Auto Ownership Levels. Transportation Research Part B 41, 506 526. Bhat, C., Kockelman, K., Chen, Q., Handy, S., Mahmassani, H., & Weston, L. (2000). Urban Accessibility Index: Literature Review, (pp. 1 84). Austin, Texas. Blanco, A., Steiner, R. L., Peng, Z. R., Shmaltsuyev, M., & Wang, R. (2010). The Economic Cost of Traffic Congestion in Florida (p. 217). Gainesville, FL. Retrieved from http://www.dot.state.fl.us/research center/Completed_Proj/Summary_OP/FDOT_BDK75_977 19_rpt.pdf
77 Boarnet, M. (2011). A broader C ontext for L and U se and T ravel B ehavior, and a R esearch A genda. Journal of the American Planning Association 77 (3), 197 213. Boarnet, M., & Crane, R. (2000). The Influence of L and use on T ravel B ehavior: S pecification and E stimation S trategies. Transportation Research Part A 35, 823 845. Brundson, C., Fotheringham, S., & Charlton, M. (1996). Geographically Weighted Regression: A Method for Exploring Spatial Nonstationarity. Geographical Analysis 28 (4), 281 298. Calthorpe, P., & Fulton, W. (2001). The Regional City Washington, D C: Island Press. Cao, X., Mokhtarian, P., & Handy, S. (2008). Examining the Impacts of Residential Self Selection on Travel Behavior: Methodologies and Empirical Findings (pp. 1 56). Davis, California. Cao, X., Xu, Z., & Fan, Y. (2010). Exploring the Conne ctions Among Residential Location, Self Selection, and Driving: Propensity Score Matching with Multiple Treatments Transportation Research Part A 44, 797 805. Carmona, M., Heath, T., Oc, T., & Tiesdell, S. (2003). Public Places Urban Spaces (1st ed., p. 312). Oxford: Architectural Press. Center for Urban Transportation Research. (2012). Florida MPOAC Transportation Revenue Study (p. 123). Tampa, FL. Retrieved from http://www.mpoac.org/rev enuestudy/RS_Final_Report.pdf Cervero, R. (2003). The Built Environment and Travel: Evidence from the United States. European Journal of Transport and Infrastructure Research 3 (2), 119 137. Cervero, R. (2006). Alternative Approaches to Modeling the Travel Demand Impacts of Smart Growth. Journal of the American Planning Association 72 (3), 285 295. Cervero, R., & Kockelman, K. (1997). Travel Demand and the 3Ds: Density, Diversity and Design. Transportation Research Part D: Transport and Environment 2 (3), 1 99 219. Retrieved from http://www.sciencedirect.com/science/article/pii/S1361920997000096 Cervero, R., & Radisch, C. (1996). Travel choices in pedestrian versus automobile oriented neighborhoods. Transport Policy 3 (3), 127 141. Chatman, D. (2005). How the Built Environment Influences Non Work Travel: Theoretical and Empirical Essays University of California. Retrieved from http://policy.rutgers.edu/faculty/chatman/Howenvironm enttravel.pdf
78 Ewing, R., Bartholomeu, K., Winkelman, S., Walters, J., & Chen, D. (2007). Growing Cooler: The Evidence on Urban Development and Climate Change (p. 170). Washington, DC: Urban Land Institute. Retrieved from postcarboncities.net/files/SGA_GrowingCooler9 18 07small.pdf Ewing, R., & Cervero, R. (2001). Travel and the Built Environment: A Synthesis. Transportation Research Record: Journal of the Transportation Research Boar d 1780 (1), 87 114. Retrieved from http://trb.metapress.com/content/a1w1712rw1225372/ Ewing, R., & Cervero, R. (2010). Travel and the Built Environment. Journal of the American Planning Association 76 (3), 265 294. Ewing, R., Haliyur, P., & Page, G. (1994) Getting A round a T raditional C ity, S uburban P lanned U nit D evelopment and E verything in B etween. Transportation Research Record 1466, 53 62. Federal Surface Transportation Policy and Planning Act of 2011. (2011), S. 326, 112 th Congress. Federal Transit Administration (2012). SAFETEA LU Implementation. Federal Transit Administration Retrieved from http://www.fta.dot.gov/legislation_law/12916_4696.html Florida Department of Community Af fairs (2009). Transportation Planning Retrieved from http://www.dca.state.fl.us/fdcp/DCP/transportation/CurrentTopics.cfm#Greenhouse Florida Department of Environmental Protection. (2009). Executive Orders and Partnership Agreements. Retrieved from http://www.dep.state.fl.us/climatechange/eo.htm Florida Department of Transportation. (2010). 2009 N ational Household Travel Survey User Guide (pp. 1 21). Tallahassee, Florida. Frost, M., Linneker, B., & Spence, N. (1998). Excess or Wasteful Commuting in a Selection of British Cities. Transportation Research Part A 32 529 538. Garling, T., Gillholm, R. & Garling, A. (1998). Reintroducing A ttitude T heory in T ravel B ehavior R esearch. The validity of an I nteractive I nterview P rocedure to P redict C ar use. Transportation 25 (2), 129 146. Retrieved from http://link.springer.com/article/10.1023/A:1005004311776?LI=true (2008). Documents Retrieved from http://www.flclimatechange.us/documents.cfm
79 Handy, S. (1993). Regional Versus Local Accessibility: Implications for Nonwork Travel. Transportation Research Record 1400 58 66. Handy, S. (1996). Understanding the Link Between Urban Form and Nonwork Travel Behavior. Journal of Planning Education and Research 15 (3), 183 198. Handy, S. (2005). Critical Assessment of the Literature on the Relationships Among Transportation, Land Use, and Physical Activity (p. 102). Washington, DC Retrieved from http://onlinepubs.trb.org/onlinepubs/archive/downloads/sr282papers/sr282Handy.p df Horner, M. (2007). A Multi Scale Analysis of Urban and Commut ing Chang in a Small Metropolitan Area (1990 200). The Annals of Regional Science 41 315 332. Kockelman, K. (1991). Travel Behavior as a function of Accessibility, Land Use Mixing, and Land Use Balance: Evidence from the San Francisco Bay Area University of California, Berkeley. Leck, E. (2006). The Impact of Urban Form on Travel Behavior: A Meta Analysis. Berkeley Planning Journal 19 (1), 37 58. Retrieved from http://escholarship. org/uc/item/20s78772#page 22 Levinson, D., & Kumar, A. (1997). Density and Journey to Work. Growth and Change 28 (2), 147 172. Retrieved from http://nexus.umn.edu/papers/density.pdf Liu, S., & Zhu, X. ( 2004). Accessibility Analyst: An Integrated GIS Tool for Accessibility Analysis in Urban Transportation Planning. Environment and Planning B: Planning and Design 31 105 124. Mokhtarian, P., & Chen, C. (2003). TTB or Not TTB, T hat is the Question: A Revie w and Analysis of the Empirical Literature on Travel Time (and Money) Budgets. Transportation Research Part A 38 (9), 643 675. Retrieved from http://www.escholarship.org/uc/item/3kr185ts Mokhtarian P., & Salomon, I. (2001). How Derived is the Demand for Travel? Some Conceptual and Measurement Considerations. Transportation Research Part A 35 695 719. Retrieved from http://escholarship.org/uc/ item/1z26n1r8 National Household Travel Survey (2001). FAQ What is trip chaining? Retrieved from http://nhts.ornl.gov/2001/pub/tripchaining.pdf National Research Council. (2009). Driving and t he Built Environment: The Effects of Compact Development on Motorized Travel, Energy Use, and CO2 Emissions -Special Report 298 Transportation Research Washington, DC: The National
80 Academies Press. Retrieved from http://www.nap.edu/catalog.php?record_id=12747#orgs Oregon Transportation Research and Education Consortium. (2011). Travel Behavior, Residential Preference, and Urban Design: A Multi Discip linary National Analysis (pp. 1 63). Portland, Oregon. Puentes, R., & Tomer, A. (2008). Miles Traveled Trends in the U.S. Washington, DC. Retrieved from http://www.brookings.edu/research/reports/2008/12/16 transportation tomer puentes Rutherford, S., McCormack, E., & Wilkinson, M. (1996). Travel impacts of urban form: Implications from an analysis of two Seattle area travel diaries. TMIP Conference on Urban Design Telecommuting and Travel Behavior 95 167. Schwanen, T., & Mokhtarian, P. (2005). What if You Live in the Wrong Neighborhood? The Impact of Residential Neighborhood Type Dissonance on Distance Traveled. Tran sportation Research Part D: Transport and Environment 10 127 151. Shaheen, S., Bengamin Chung, J., Allen, D., & Howe Steiger, L. (2009). Achieving Under AB 32: An Exploration of Pot ential Policy Processes and Mechanisms (p. 95). Retrieved from http://tsrc.berkeley.edu/sites/tsrc.berkeley.edu/files/achieving use.pdf Shay, E., & Khattak, A. (200 5 ). Auto Ownership and Use in Neo Traditional and Conventional Neighborhoods. Transportation Research Record 1902 18 25. Shoup, L., & Lang, M. (2011). Transportation 101:An Introduction to Federal Transportation Polic y (pp. 1 84). Washington, DC. Steiner, R. L., Srinivasan, S., Provost, R. E., Mackey, J., Arafat, A., Anderson, N., &DeLarco, L. (2010). VMT Based Traffic Impact Assessment: Development of a Trip Length Model (pp. 1 56). Gainesville, Florida. Sustainable Communities and Climate Protection Act of 2008 65080(b)(2)(A) (2008). UCLA: Statistical Consulting Group (2013). How do I interpret a regression model when some variables are log transformed? Retrieved February 2, 2013, from http://www.ats.ucla.edu/stat/mult_pkg/faq/general/log_transformed_regression.htm U.S. Department of Transportation (2013) a Highway Authorizations: Moving Ahead for Progr ess in the 21st Century Act (MAP 21). MAP 21 Retrieved from http://www.fhwa.dot.gov/map21/ha.cfm
81 U.S. Department of Transportation. (2013) b MAP 21: Performance Management. MAP 21 Retrieved from http://www.fhwa.dot.gov/map21/pm.cfm United States Environmental Protection Agency (EPA). (2009). Fact Sheet -Proposed Rule: Prevention of Significant Deterioration and Title V Greenhouse Gas Tailoring Rul e. Retrieved from http://www.epa.gov/NSR/fs20090930action.html Van Acker, V., Van Wee, B., & Witlox, F. (2010). When Transport Geography Meets Social Psychology: Toward a Conceptual Model of Travel Behavior Transport Reviews 30 (2), 219 240. Zhou, B., & Kockleman, K. (2008). Self Selection in Home Choice: Use of Treatment Effects in Evaluating the Relationship Between the Built Environment and Travel Behavior. Transportation Research Record 2077 54 61.
82 BIOGRAPHICAL SKETCH Urban Affairs in 2005 where he became accustomed to GIS applications. After graduating, Mr. Provost participated in the AmeriCorps Program coordinating a local effort to develo p ecotourism opportunities in rural Oregon. While in Oregon, he also leader in the community in various leadership skills. Mr. Provost held two research positions while attending the experience in GIS analysis in a variety of planning contexts including ecotourism, watershed planning, suitability modeling, and long range planning. His research interests include GIS applications in environmental and sustainability planning.