OREGON DOT .edu



Transferability & forecasting of the Pedestrian Index Environment (PIE) for modeling applicationsFinal ReportNITC-RR-###byKelly J. Clifton, PhD1 (PI)Jaime P. Orrego-Onate1Patrick A. Singleton, PhD2Robert J. Schneider, PhD31Portland State University2Utah State University3University of Wisconsin – Milwaukeefor National Institute for Transportation and Communities (NITC)P.O. Box 751Portland, OR 97207March 2018Technical Report Documentation Page1. Report No.NITC-RR-###2. Government Accession No.3. Recipient’s Catalog No.4. Title and SubtitleTransferability & Forecasting of the Pedestrian Index of the Environment (PIE) for Modeling Applications5. Report DateMarch 20196. Performing Organization Code7. Author(s)Kelly J. Clifton; Jaime P. Orrego-O?ate; Patrick A. Singleton; Robert J. Schneider8. Performing Organization Report No.9. Performing Organization Name and AddressPortland State University, Department of Civil & Environmental EngineeringPO Box 751 - CEEPortland, OR 97207-075110. Work Unit No. (trais)11. Contract or Grant No.12. Sponsoring Agency Name and AddressNational Institute for Transportation and Communities (NITC)P.O. Box 751 Portland, Oregon 9720713. Type of Report and Period Covered14. Sponsoring Agency Code15. Supplementary Notes16. AbstractThis project focuses on making our measures, models, and methods more transferable to other locations. Specifically, we re-evaluate, compare and test our pedestrian index of the environment (PIE) measure using data resources more commonly available to planning agencies across the country. Next, we test the results of PIE and its input data in models of pedestrian mode choice for stability of estimation results within a region (intraregional) and between regions (interregional). This research is the next logical step in the MoPeD’s enhancement and is critical to enabling its utility beyond the Portland region.The results of this project show that population density and pedestrian connectivity had the most consistent and strong relationship to walk mode choice across all of our regions, which echoed the long literature on this topic. However, the other components of the built environment included in PIE had more variability in their ability to explain walk mode choice. Employment density and its subset urban living infrastructure (ULI), intended to capture retail and service access, had less explanatory power and stability in the cities tested. Based upon these findings, we provide several guidelines for the construct of walkability indices, including variables and spatial scales. Our findings raise questions about the relationship between walking and the built environment within a region and thus, the intraregional transferability of one walkability index is suspect. Estimation results suggest that there may be different responses to the built environment in lower-density vs. higher density regimes and that these relationships may be nonlinear. However, smaller sample sizes of travel survey data in high density areas in all of the US cities tested pose limitations to drawing more confident conclusions from these results. The interregional comparisons of PIE and walk mode share between Los Angeles and Portland showed promise for the use of the index in different regions. In these two regions, model results showed a similar walk mode share for the same values of PIE constructed at the block group level. This provides initial support that the PIEbg construct may be transferrable between metropolitan regions, in part, due to population density's prominent role in PIE.17. Key Words18. Distribution StatementNo restrictions. Copies available from NITC:nitc-19. Security Classification (of this report)Unclassified20. Security Classification (of this page)Unclassified21. No. of Pages22. PriceacknowledgmentsThis project was funded by the National Institute for Transportation and Communities (NITC)and the City of Tigard, Oregon. The authors thank colleagues from Metro, Portland State University, University of Wisconsin, Milwaukee, Polytechnique Montréal, Technical University of Munich, and the Metropolitan Council, St. Paul, MN, for their insights and interest in this topic. DisclaimerThe contents of this report reflect the views of the authors, who are solely responsible for the facts and the accuracy of the material and information presented herein. This document is disseminated under the sponsorship of the U.S. Department of Transportation University Transportation Centers Program and Portland State University in the interest of information exchange. The U.S. Government and Portland State University assumes no liability for the contents or use thereof. The contents do not necessarily reflect the official views of the U.S. Government and Portland State University. This report does not constitute a standard, specification, or regulation.RECOMMENDED CITATIONClifton, Kelly J; Orrego-O?ate, Jaime; Singleton, Patrick; and Schneider, Robert. Transferability & Forecasting of the Pedestrian Index of the Environment (PIE) for Modeling Applications. NITC-RR-###. Portland, OR: Transportation Research and Education Center (TREC), 2017.table of contents TOC \o "1-5" ExEcutive Summary PAGEREF _Toc2003596 \h 11.0Introduction PAGEREF _Toc2003597 \h 22.0Background PAGEREF _Toc2003598 \h 53.0PEdestrian Index of the Environment (PIE) PAGEREF _Toc2003599 \h 64.0Data and MEthods for NEW PIE (PIEbg) PAGEREF _Toc2003600 \h 95.0Testing the transferability of PIEbg PAGEREF _Toc2003601 \h 125.1Comparing PIEbg and PIE0 PAGEREF _Toc2003602 \h 125.2Intraregional measure transferability PAGEREF _Toc2003603 \h 185.3Interregional measure transferability PAGEREF _Toc2003604 \h 216.0Discussion and conclusions PAGEREF _Toc2003605 \h 276.1Transferability of models and measures PAGEREF _Toc2003606 \h 276.2Representing the built environment PAGEREF _Toc2003607 \h 296.3Future work PAGEREF _Toc2003608 \h 307.0References PAGEREF _Toc2003609 \h 31List of tables TOC \t "Table Title,1" \c "Table" Table 1 Binary Logit Models for PIE0 PAGEREF _Toc2003590 \h 8Table 2 PIEbg estimation results: binary logit models of walk mode share PAGEREF _Toc2003591 \h 14Table 3 Estimated weights for each built environment measure for PIE0 and PIEbg PAGEREF _Toc2003592 \h 14Table 4 Pooled vs. Urban/Suburban Walk Mode Share Models PAGEREF _Toc2003593 \h 20Table 5 Trip information and built environment measures in each metropolitan area PAGEREF _Toc2003594 \h 22Table 6 Unscaled PIEbg coefficients estimated from Portland data PAGEREF _Toc2003595 \h 25List of figures TOC \t "Table Title,1" \c "Figure" Figure 1 Regional Map of the Built Environment Index PIE0 PAGEREF _Toc2003578 \h 9Figure 2 Comparison between block groups (bold line) and PAZ (grid) in downtown Portland, OR PAGEREF _Toc2003579 \h 11Figure 3 Distribution of the scaled attributes of the built environment for PIEbg in Portland, OR PAGEREF _Toc2003580 \h 13Figure 4 Frequency distribution of PIEbg scores by block group PAGEREF _Toc2003581 \h 15Figure 5 Land percentage by PIEbg score PAGEREF _Toc2003582 \h 15Figure 6 PIEbg distribution in Portland PAGEREF _Toc2003583 \h 16Figure 7 PIE0 versus PIEbg PAGEREF _Toc2003584 \h 17Figure 8 PIEbg aggregated by deciles vs walking share PAGEREF _Toc2003585 \h 18Figure 9 Predicted walk probabilities PAGEREF _Toc2003586 \h 21Figure 10 Standardized coefficients for different built environment measures PAGEREF _Toc2003587 \h 23Figure 11 Unscaled PIEbg distribution in Los Angeles and Portland PAGEREF _Toc2003588 \h 26Figure 12 The role of walkability gradients PAGEREF _Toc2003589 \h 28ExEcutive SummaryThere have been important advances in non-motorized planning tools in recent years, including the development of the MoPeD pedestrian demand model (Clifton et al., 2013, 2015). This tool and others are increasingly requested by governments and agencies seeking to increase walking activity and create more walkable places. To date, the MoPeD tool has been piloted with success in the Portland region using data unique to Metro, the metropolitan planning organization. However, there is increasing interest from planning agencies in adapting the pedestrian modeling tools and their inputs for use in their own jurisdictions. Unfortunately, other regions often do not have uniform access to the same kinds of pedestrian environment data as Metro, particularly at such a fine-grained scale. In this next phase of our pedestrian modeling work (see Clifton et al., 2013, 2015), this project focuses on making our measures, models, and methods more transferable to other locations. Specifically, we will re-evaluate, compare and test our pedestrian index of the environment (PIE) measure using data resources more commonly available to planning agencies across the country. Next, we test the results of PIE and its input data in models of pedestrian mode choice for stability of estimation results within a region (intraregional) and between regions (interregional). This research is the next logical step in the MoPeD’s enhancement and is critical to enabling its utility beyond the Portland region.In terms of index inputs, the results of this project show that population density and pedestrian connectivity had the most consistent and strong relationship to walk mode choice across all of our regions, which echoed the long literature on this topic. However, the other components of the built environment included in PIE had more variability in their ability to explain walk mode choice. Employment density and its subset urban living infrastructure (ULI), intended to capture retail and service access, had less explanatory power and stability in the cities tested. Based upon these findings, we provide several guidelines for the construct of walkability indices, including variables and spatial scales. Our findings raise questions about the relationship between walking and the built environment within a region and thus, the intraregional transferability of one walkability index is suspect. Estimation results suggest that there may be different responses to the built environment in lower-density vs. higher density regimes and that these relationships may be nonlinear. However, smaller sample sizes of travel survey data in high density areas in all of the US cities tested pose limitations to drawing more confident conclusions from these results. The interregional comparisons of PIE and walk mode share between Los Angeles and Portland showed promise for the use of the index in different regions. In these two regions, model results showed a similar walk mode share for the same values of PIE constructed at the block group level. This provides initial support that the PIEbg construct may be transferrable between metropolitan regions, in part, due to population density's prominent role in PIE. IntroductionThere is a growing demand for tools that can estimate pedestrian activity, forecast future walking levels, and be used for other transportation planning applications. Many metropolitan planning organizations (MPOs), in particular, seek tools to estimate pedestrian demand that can provide greater support for planning and policy-making activities, including: increased sensitivity to pedestrian facilities, land use changes, and the built environment; better accounting for mode shifts and resulting changes in greenhouse gas emissions; more accurate estimates of physical activity levels for health impact assessments and of pedestrian exposure for traffic safety analyses; etc. While not yet widespread, regional travel demand forecasting for pedestrians continues to develop ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"a1j6uhcs6n4","properties":{"formattedCitation":"(Kuzmyak et al. 2014; Singleton et al. 2018)","plainCitation":"(Kuzmyak et al. 2014; Singleton et al. 2018)"},"citationItems":[{"id":121,"uris":[""],"uri":[""],"itemData":{"id":121,"type":"article-journal","title":"Estimating Bicycling and Walking for Planning and Project Development: A Guidebook","container-title":"NCHRP Report","issue":"770","source":"trid.","URL":"","ISSN":"0077-5614","shortTitle":"Estimating Bicycling and Walking for Planning and Project Development","author":[{"family":"Kuzmyak","given":"J. Richard"},{"family":"Walters","given":"Jerry"},{"family":"Bradley","given":"Mark"},{"family":"Kockelman","given":"Kara M."}],"issued":{"date-parts":[["2014"]]},"accessed":{"date-parts":[["2018",4,14]]}}},{"id":194,"uris":[""],"uri":[""],"itemData":{"id":194,"type":"article-journal","title":"Making Strides: State-of-the-Practice of Pedestrian Forecasting in Regional Travel Models","container-title":"Transportation Research Record: Journal of the Transportation Research Board","abstract":"Much has changed in the 30 years since non-motorized modes were first included in regional travel demand models. As interest in understanding behavioral influences on walking and policies requiring estimates of walking activity increase, it is important to consider how pedestrian travel is modeled at a regional level. This paper evaluates the state-of-the-practice of modeling walk trips among the largest 48 metropolitan planning organizations (MPOs) and assesses changes made over the last five years. By reviewing model documentation and responses to a survey of MPO modelers, this paper summarizes current practices, describes six pedestrian modeling frameworks, and identifies trends. Three-quarters (75%) of large MPOs now model non-motorized travel, and over two-thirds (69%) of those MPOs distinguish walking from bicycling; these percentages are up from nearly two-thirds (63%) and one-half (47%), respectively, in 2012. This change corresponds with an increase in the deployment of activity-based models, which offer the opportunity to enhance pedestrian modeling techniques. The biggest barrier to more sophisticated models remains a lack of travel survey data on walking behavior, yet some MPOs are starting to overcome this challenge by oversampling potential active travelers. Decision-makers are becoming more interested in analyzing walking and using estimates of walking activity that are output from models for various planning applications. As the practice continues to mature, the near future will likely see smaller scale measures of the pedestrian environment, more detailed zonal and network structures, and possibly even an operational model of pedestrian route choice.","author":[{"family":"Singleton","given":"Patrick A."},{"family":"Totten","given":"Joseph C."},{"family":"Orrego-O?ate","given":"Jaime P."},{"family":"Schneider","given":"Robert J."},{"family":"Clifton","given":"Kelly J."}],"issued":{"date-parts":[["2018"]],"season":"forthcoming in"}}}],"schema":""} (Kuzmyak et al. 2014; Singleton et al. 2018); however, different MPOs have used various ad-hoc techniques that depend on their unique decision support needs, data sources, and modeling capabilities, with little consistency. Other agencies lack the technical capacity and funding to undertake model improvements on their own. Facing this reality, travel model transferability is of increasing interest ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"SDGGUevB","properties":{"formattedCitation":"(Rossi and Bhat 2014)","plainCitation":"(Rossi and Bhat 2014)"},"citationItems":[{"id":137,"uris":[""],"uri":[""],"itemData":{"id":137,"type":"article-journal","title":"Guide for Travel Model Transfer","source":"trid.","URL":"","author":[{"family":"Rossi","given":"Thomas F."},{"family":"Bhat","given":"Chandra R."}],"issued":{"date-parts":[["2014",10]]},"accessed":{"date-parts":[["2018",4,14]]}}}],"schema":""} (Rossi and Bhat 2014). A framework for modeling pedestrian demand with consistent data inputs that can be more easily transferred between regions would be of great value to practitioners, especially those at agencies with limited resources.Despite a long history of research documenting relationships between walking and environmental conditions—including effects of the built environment—models used in practice lag in their representation of the pedestrian environment. Pedestrian travel has been positively related to higher residential and employment densities, greater land use mix or diversity, more connected networks or higher intersection densities, greater accessibility to transit, and sidewalk and crossing conditions (e.g., ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"OmYpJVsM","properties":{"formattedCitation":"(Saelens and Handy 2008)","plainCitation":"(Saelens and Handy 2008)"},"citationItems":[{"id":125,"uris":[""],"uri":[""],"itemData":{"id":125,"type":"article-journal","title":"Built Environment Correlates of Walking: A Review","container-title":"Medicine and science in sports and exercise","page":"S550-S566","volume":"40","issue":"7 Suppl","source":"PubMed Central","abstract":"Introduction\nThe past decade has seen a dramatic increase in the empirical investigation into the relations between built environmental and physical activity. To create places that facilitate and encourage walking, practitioners need an understanding of the specific characteristics of the built environment that correlate most strongly with walking. This paper reviews evidence on the built environment correlates with walking.\n\nMethod\nIncluded in this review were 13 reviews published between 2002 and 2006 and 29 original studies published in 2005 and up through May 2006. Results were summarized based on specific characteristics of the built environment and transportation walking versus recreational walking.\n\nResults\nPrevious reviews and newer studies document consistent positive relations between walking for transportation and density, distance to non-residential destinations, and land use mix; findings for route/network connectivity, parks and open space, and personal safety are more equivocal. Results regarding recreational walking were less clear.\n\nConclusions\nMore recent evidence supports the conclusions of prior reviews, and new studies address some of the limitations of earlier studies. Although prospective studies are needed, evidence on correlates appears sufficient to support policy changes.","DOI":"10.1249/MSS.0b013e31817c67a4","ISSN":"0195-9131","note":"PMID: 18562973\nPMCID: PMC2921187","shortTitle":"Built Environment Correlates of Walking","journalAbbreviation":"Med Sci Sports Exerc","author":[{"family":"Saelens","given":"Brian E."},{"family":"Handy","given":"Susan L."}],"issued":{"date-parts":[["2008",7]]}}}],"schema":""} Saelens and Handy 2008). Yet, most MPO regional models use only rough density measures to predict pedestrian demand ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"fh7kUorg","properties":{"formattedCitation":"(Singleton et al. 2018)","plainCitation":"(Singleton et al. 2018)"},"citationItems":[{"id":194,"uris":[""],"uri":[""],"itemData":{"id":194,"type":"article-journal","title":"Making Strides: State-of-the-Practice of Pedestrian Forecasting in Regional Travel Models","container-title":"Transportation Research Record: Journal of the Transportation Research Board","abstract":"Much has changed in the 30 years since non-motorized modes were first included in regional travel demand models. As interest in understanding behavioral influences on walking and policies requiring estimates of walking activity increase, it is important to consider how pedestrian travel is modeled at a regional level. This paper evaluates the state-of-the-practice of modeling walk trips among the largest 48 metropolitan planning organizations (MPOs) and assesses changes made over the last five years. By reviewing model documentation and responses to a survey of MPO modelers, this paper summarizes current practices, describes six pedestrian modeling frameworks, and identifies trends. Three-quarters (75%) of large MPOs now model non-motorized travel, and over two-thirds (69%) of those MPOs distinguish walking from bicycling; these percentages are up from nearly two-thirds (63%) and one-half (47%), respectively, in 2012. This change corresponds with an increase in the deployment of activity-based models, which offer the opportunity to enhance pedestrian modeling techniques. The biggest barrier to more sophisticated models remains a lack of travel survey data on walking behavior, yet some MPOs are starting to overcome this challenge by oversampling potential active travelers. Decision-makers are becoming more interested in analyzing walking and using estimates of walking activity that are output from models for various planning applications. As the practice continues to mature, the near future will likely see smaller scale measures of the pedestrian environment, more detailed zonal and network structures, and possibly even an operational model of pedestrian route choice.","author":[{"family":"Singleton","given":"Patrick A."},{"family":"Totten","given":"Joseph C."},{"family":"Orrego-O?ate","given":"Jaime P."},{"family":"Schneider","given":"Robert J."},{"family":"Clifton","given":"Kelly J."}],"issued":{"date-parts":[["2018"]],"season":"forthcoming in"}}}],"schema":""} (Singleton et al. 2018). While the lack of pedestrian-sensitive built environment data was a barrier in the past, today fine-grained archived spatial datasets are becoming more widely available, including point-, parcel-, or block-level measures of the pedestrian environment. This has led to the development of a number of measures of the pedestrian environment, including the Pedestrian Index of the Environment, ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"bU4HSXLR","properties":{"formattedCitation":"(Singleton et al. 2014)","plainCitation":"(Singleton et al. 2014)"},"citationItems":[{"id":139,"uris":[""],"uri":[""],"itemData":{"id":139,"type":"paper-conference","title":"The Pedestrian Index of the Environment: Representing the Walking Environment in Planning Applications","source":"trid.","event":"Transportation Research Board 93rd Annual MeetingTransportation Research Board","URL":"","shortTitle":"The Pedestrian Index of the Environment","author":[{"family":"Singleton","given":"Patrick A."},{"family":"Schneider","given":"Robert J."},{"family":"Muhs","given":"Christopher"},{"family":"Clifton","given":"Kelly J."}],"issued":{"date-parts":[["2014"]]},"accessed":{"date-parts":[["2018",4,14]]}}}],"schema":""} (Singleton et al. 2014), the Pedestrian Environment Factor ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"lBVCMvEJ","properties":{"formattedCitation":"(Greenwald and Boarnet 2001)","plainCitation":"(Greenwald and Boarnet 2001)"},"citationItems":[{"id":210,"uris":[""],"uri":[""],"itemData":{"id":210,"type":"article-journal","title":"Built Environment as Determinant of Walking Behavior: Analyzing Nonwork Pedestrian Travel in Portland, Oregon","container-title":"Transportation Research Record: Journal of the Transportation Research Board","page":"33-41","volume":"1780","source":"trrjournalonline. (Atypon)","abstract":"Much has been written about the connection between land use/urban form and transportation from the perspective of affecting automobile trip generation. This addresses only half the issue. The theoretical advances in land use-transportation relationships embodied in paradigms such as the jobs-housing balance, neotraditional design standards, and transitoriented development rely very heavily on the generation of pedestrian traffic to realize their proposed benefits. The present analysis uses models and data sets similar to those used in previous work for the Portland, Oregon, area but applies them toward analysis of nonwork walking travel. The results suggest that regardless of the effects that land use has on individual nonwork walking trip generation, the impacts take place at the neighborhood level.","DOI":"10.3141/1780-05","ISSN":"0361-1981","shortTitle":"Built Environment as Determinant of Walking Behavior","journalAbbreviation":"Transportation Research Record: Journal of the Transportation Research Board","author":[{"family":"Greenwald","given":"Michael"},{"family":"Boarnet","given":"Marlon"}],"issued":{"date-parts":[["2001",1,1]]}}}],"schema":""} (Greenwald and Boarnet 2001), the Pedestrian Environment Index ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"L3vnWUwD","properties":{"formattedCitation":"(Peiravian et al. 2014)","plainCitation":"(Peiravian et al. 2014)"},"citationItems":[{"id":128,"uris":[""],"uri":[""],"itemData":{"id":128,"type":"article-journal","title":"Development and application of the Pedestrian Environment Index (PEI)","container-title":"Journal of Transport Geography","page":"73-84","volume":"39","source":"ScienceDirect","abstract":"The objective of this work is to develop a new and easily computable measure of pedestrian friendliness for urban neighborhoods that makes the best use of the available data and also addresses the issues concerning other models in use. The Pedestrian Environment Index (PEI) is defined as the product of four components representing land-use diversity (based on the concept of entropy), population density, commercial density, and intersection density. The final PEI is bound between 0 and 1, and uses data that typically are readily available to planners and metropolitan planning organizations (MPO). The results of this method are region-specific; they are comparable only between the zones within the given study area. As a case study, the city of Chicago is analyzed at the sub-traffic analysis zone (sub-TAZ) level. The results agree closely with the expectation of pedestrian friendliness across different parts of the city. Possible extensions are also listed, including a further study to determine statistical relationships between the PEI and common socio-economic characteristics. The method could also be further improved should more types of data become available.","DOI":"10.1016/j.jtrangeo.2014.06.020","ISSN":"0966-6923","journalAbbreviation":"Journal of Transport Geography","author":[{"family":"Peiravian","given":"Farideddin"},{"family":"Derrible","given":"Sybil"},{"family":"Ijaz","given":"Farukh"}],"issued":{"date-parts":[["2014",7,1]]}}}],"schema":""} (Peiravian et al. 2014), and others. The present-day issue is that the availability and formatting of these built environment data vary widely across and within regions, yielding challenges to the transferability of composite measures of the pedestrian environment (and models of pedestrian demand). A related issue involves finding the appropriate scale that balances tradeoffs between the use of small-scale and behaviorally-sensitive pedestrian environment data, the computational processing challenges that such data impose, and the general availability of those data now and in the future.Pedestrian modeling has also not kept pace with nor taken full advantage of developments happening in pedestrian data collection. Historically, one reason why few MPO models represented pedestrian travel was that the regional household travel surveys used to estimate such models captured few walking trips, often by design ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"j4VVCuOh","properties":{"formattedCitation":"(K. Clifton and Muhs 2012)","plainCitation":"(K. Clifton and Muhs 2012)","dontUpdate":true},"citationItems":[{"id":131,"uris":[""],"uri":[""],"itemData":{"id":131,"type":"article-journal","title":"Capturing and Representing Multimodal Trips in Travel Surveys","container-title":"Transportation Research Record: Journal of the Transportation Research Board","page":"74-83","volume":"2285","source":"trrjournalonline. (Atypon)","abstract":"Multimodal trips, or trips that use more than one means of transportation, have historically been underrepresented in travel surveying efforts. This lack of consideration has implications for widely accepted statistics for nonmotorized travel behavior (walking, bicycling, etc.) and affects researchers and professionals in travel modeling, urban planning, public health, and urban design. However, interest in the \"last mile\" connections to transit, aggregate health impacts of short walking trips, and emphasis on local connectivity require more detailed information on these typically short but important stages of travel. This study reviews approaches to multimodal travel behavior in travel surveys, analyzes their implications, and makes recommendations to improve data collection for the purpose of improved representation of multimodal travel. Particular attention is given to transit access and egress trip segments, nonmotorized travel, use of technology in travel surveys, reporting data, and dissemination of the travel survey beyond the travel forecasting community.","DOI":"10.3141/2285-09","ISSN":"0361-1981","journalAbbreviation":"Transportation Research Record: Journal of the Transportation Research Board","author":[{"family":"Clifton","given":"Kelly J."},{"family":"Muhs","given":"Christopher"}],"issued":{"date-parts":[["2012",10,18]]}}}],"schema":""} (Clifton and Muhs 2012). Today, surveys have improved their capture of pedestrian activity, and data collection efforts are increasingly being supplemented by short- and long-term pedestrian counts ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"GPLIyQQo","properties":{"formattedCitation":"(Schneider et al. 2005)","plainCitation":"(Schneider et al. 2005)"},"citationItems":[{"id":134,"uris":[""],"uri":[""],"itemData":{"id":134,"type":"article-journal","title":"Case Study Analysis of Pedestrian and Bicycle Data Collection in U.S. Communities","container-title":"Transportation Research Record: Journal of the Transportation Research Board","page":"77-90","volume":"1939","source":"trrjournalonline. (Atypon)","abstract":"Federal funding for pedestrian and bicycle transportation has increased over the past 15 years, with a resulting increase in shared-use pathways, paved shoulders, bicycle lanes, and sidewalks in many parts of the United States. This has caused communities to ask questions: Where is pedestrian and bicycle activity taking place? What effect does facility construction have on levels of bicycling and walking? What are the characteristics of nonmotorized transportation users? How many miles of pedestrian and bicycle facilities are available? Where are existing facilities located? This paper provides a summary of recent research that was sponsored by FHWA and the Pedestrian and Bicycle Information Center to review and evaluate bicycle and pedestrian data collection methods throughout the United States. It uses a case study approach to evaluate pedestrian and bicycle data collection in 29 different agencies throughout the country in communities ranging in size from 6,000 residents (Sandpoint, Idaho) to 8 million residents (New York City). These case studies are analyzed in the following data collection categories: manual counts, automated counts, surveys targeting nonmotorized transportation users, surveys sampling a general population, inventories, and spatial analyses. The results provide information about the methods and the optimum timing for pedestrian and bicycle data collection; emerging technologies that can be used to gather and analyze data; the benefits, limitations, and costs of different data collection techniques; and implications for a national data collection strategy.","DOI":"10.3141/1939-10","ISSN":"0361-1981","journalAbbreviation":"Transportation Research Record: Journal of the Transportation Research Board","author":[{"family":"Schneider","given":"Robert"},{"family":"Patten","given":"Robert"},{"family":"Toole","given":"Jennifer"}],"issued":{"date-parts":[["2005",1,1]]}}}],"schema":""} (Schneider et al. 2005). Surprisingly, these count data are rarely used in the development of regional pedestrian demand forecasting models and analysis tools, despite being a potential source of data for external validation, if not calibration.This project begins to address some of these research gaps and practical needs. It builds upon previous work developing a framework for Model of Pedestrian Demand (MoPeD) and a Pedestrian Index of the Environment (PIE) variable ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"2dTRIecY","properties":{"formattedCitation":"(K. Clifton et al. 2013, 2015)","plainCitation":"(K. Clifton et al. 2013, 2015)"},"citationItems":[{"id":187,"uris":[""],"uri":[""],"itemData":{"id":187,"type":"article-journal","title":"Improving the Representation of the Pedestrian Environment in Travel Demand Models, Phase I","container-title":"Civil and Environmental Engineering Faculty Publications and Presentations","URL":"","DOI":"10.15760/trec.120","author":[{"family":"Clifton","given":"Kelly"},{"family":"Singleton","given":"Patrick"},{"family":"Muhs","given":"Christopher"},{"family":"Schneider","given":"Robert"},{"family":"Lagerwey","given":"Peter"}],"issued":{"date-parts":[["2013",9,1]]}}},{"id":185,"uris":[""],"uri":[""],"itemData":{"id":185,"type":"article-journal","title":"Development of a Pedestrian Demand Estimation Tool","container-title":"Civil and Environmental Engineering Faculty Publications and Presentations","URL":"","DOI":"10.15760/trec.124","author":[{"family":"Clifton","given":"Kelly"},{"family":"Singleton","given":"Patrick"},{"family":"Muhs","given":"Christopher"},{"family":"Schneider","given":"Robert"}],"issued":{"date-parts":[["2015",9,1]]}}}],"schema":""} (Clifton et al. 2013, 2015) to make measures and models more transferrable and useful for forecasting. PIE is a composite index that is constructed from a set of models that estimate the probability of walking trips as a function of a built environment attribute. The value of PIE is calculated from a set of built environment variables representing activity density, block density, sidewalk density, transit access, neighborhood-oriented businesses and other factors ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"zhMt7u0Y","properties":{"formattedCitation":"(Singleton et al. 2014)","plainCitation":"(Singleton et al. 2014)"},"citationItems":[{"id":139,"uris":[""],"uri":[""],"itemData":{"id":139,"type":"paper-conference","title":"The Pedestrian Index of the Environment: Representing the Walking Environment in Planning Applications","source":"trid.","event":"Transportation Research Board 93rd Annual MeetingTransportation Research Board","URL":"","shortTitle":"The Pedestrian Index of the Environment","author":[{"family":"Singleton","given":"Patrick A."},{"family":"Schneider","given":"Robert J."},{"family":"Muhs","given":"Christopher"},{"family":"Clifton","given":"Kelly J."}],"issued":{"date-parts":[["2014"]]},"accessed":{"date-parts":[["2018",4,14]]}}}],"schema":""} (Singleton et al. 2014). PIE was originally developed for Portland, OR using the 2011 Oregon Household Travel Survey (OHAS) and built environment data at a small grid scale of 80 by 80 meters. MoPeD is a tool for predicting pedestrian demand using trip generation, mode choice, and destination choice models. These models predict walking trip probabilities for a variety of travel purposes using PIE and other socioeconomic and demographic information. MoPeD has been calibrated with OHAS data and applied to 80 by 80-meter grid cells in Portland, OR.These tools and others are increasingly requested by those interested in increasing walking activity and creating more walkable places. While MoPeD and PIE have been piloted with success in the Portland region using data unique to Metro, the metropolitan planning organization, there is increasing interest from planning agencies within and outside of Portland and Oregon (e.g.: City of Tigard, OR; Metropolitan Council of the Twin Cities, MN; San Francisco Public Health Department, CA) in adapting these pedestrian modeling tools for use in their own jurisdictions. Local governments desire to apply these tools for a variety of planning and forecasting purposes, including but not limited to regional demand modeling. Unfortunately, other regions often do not have uniform access to the same kinds of pedestrian environment data as Metro, particularly at such a fine-grained scale. Important challenges remain in model development that must be overcome if these tools are to achieve widespread application. Among the most critical needs are the standardization and forecasting of model inputs, particularly measures of the built environment.In this next phase of our pedestrian modeling work, we focus on testing the ability of our measures, models, and methods to be more transferable to other locations. Specifically, we capitalize on the increased availability of pedestrian environment data to create a new PIE (called PIEbg) that utilizes more widespread data sources available at the census block group level. We re-evaluate, compare (with the original PIE, called PIE0 in this report) and test our PIEbg measure using data resources more commonly available to planning agencies across the country. Next, we examine this PIEbg variable and its association with walking in Portland amid both urban and suburban contexts (intra-regionally). Additionally, we examine differences in the PIEbg measure and its performance in regions outside of Portland (interregional transferability). These tasks balance data availability, scale, computational capacity, and behavioral realism. This updated suite of pedestrian modeling products has the potential to be more widely transferable and applicable to MPOs and other planning agencies beyond Portland. Finally, we test this new measure across a large range of urban environments by estimating models of probability of walking using PIEbg as a single explanatory variable. This analysis tests the issue of linearity in the parameters and questions the validity of using a single parameter to represent the behavioral response to walking environment for all locations. Our overall research objective is to increase the availability of pedestrian demand tools for use by various planning agencies around the country. To do this, we are guided by the following research questions.How can PIE be constructed using data that are widely available to planning agencies across the country, but remain sensitive to pedestrian scales and variations in environmental conditions? PIE is central to representing the built environment in the trip generation and destination choice models within MoPeD. In its current form, PIE relies on detailed, disaggregate, spatial data that are available in the Portland region; but, in other areas of the country, these data are not uniformly available. In order for the MoPeD model to be transferable to other locations, we redesign PIE to make use of data resources that are more commonly available around the US and elsewhere and compare results with our original PIE construct.Should the measures of the built environment comprising PIE be weighted differently in different urban/suburban contexts? People walking in suburban locations may actually care about and respond to various environmental characteristics differently than people walking in urban locations. Suburban areas have been given relatively less attention in studies of pedestrian behavior, despite evidence of significant levels of walking observed there ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"tQ6LvBVz","properties":{"formattedCitation":"(Larco et al. 2013)","plainCitation":"(Larco et al. 2013)"},"citationItems":[{"id":195,"uris":[""],"uri":[""],"itemData":{"id":195,"type":"article-journal","title":"Trips to Strips: Walking and Site Design in Suburban Multifamily Housing","container-title":"Journal of Urban Design","page":"281-303","volume":"18","issue":"2","source":"Taylor and Francis+NEJM","abstract":"With over nine million units in the country, suburban multifamily housing is a widespread and overlooked example of density located within walking distance of commercial development in suburbia. This paper reports on resident demographics, attitudes, and perceptions as they relate to mode choice in 14 suburban multifamily sites in Eugene, Oregon. Through site analysis and resident surveys, this study shows that residents of well-connected suburban multifamily housing developments walk or bike for nearly half of their trips to the local commercial area (LCA). In addition, residents of well-connected multifamily developments reported walking to their LCA 60% more (one more trip per week) than residents of less-connected developments who took a similar number of total trips. Quantifying the degree to which site design, and specifically connectivity, makes a difference in residents' mode choice is a first step towards increasing the amount of active transportation in these areas. The results of this research provide planners and designers a basis for re-evaluating suburban multifamily site design and zoning codes.","DOI":"10.1080/13574809.2013.772886","ISSN":"1357-4809","shortTitle":"Trips to Strips","author":[{"family":"Larco","given":"Nico"},{"family":"Stockard","given":"Jean"},{"family":"Steiner","given":"Bethany"},{"family":"West","given":"Amanda"}],"issued":{"date-parts":[["2013",5,1]]}}}],"schema":""} (Larco et al. 2013). To better understand walking behavior across a spectrum of contexts, we test for differences in the combinations of environmental features and thresholds used in PIE between urban and suburban locations for Portland. Additionally, we extend these tests to different cities to assess how PIE behaves and if we can find evidence that, the measurements are transferrable. The remaining report is organized into five additional sections. The Section 2.0 provides an introduction to the pedestrian model. Then, a brief summary of the background of the project, a review of the main issues about walking and travel behavior, and a synthesis of the transferability measures are provided in Section 3.0. The next section, Section 4.0 is a description of the data inputs and outputs, and discusses the main assumptions in the analysis, and reviews the construction of the data sets including the built environment and travel data. Section 5.0 covers the research methods and the implications of our approach and presents results. It is separated in four subsections: the reconstruction of our new index (PIEbg), the comparison with the older index (PIE0), and two sections analyzing the intra- and intercity transferability of our new index. Finally, Section 6.0 is a discussion and conclusions of this project with suggestions for policy and future research.BackgroundThe relationship between walking and the built environment has had a particular emphasis on the study of travel behavior. In 1997, ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"yeu2F3uk","properties":{"formattedCitation":"(Cervero and Kockelman 1997)","plainCitation":"(Cervero and Kockelman 1997)"},"citationItems":[{"id":141,"uris":[""],"uri":[""],"itemData":{"id":141,"type":"article-journal","title":"Travel demand and the 3Ds: Density, diversity, and design","container-title":"Transportation Research Part D: Transport and Environment","page":"199-219","volume":"2","issue":"3","source":"ScienceDirect","abstract":"The built environment is thought to influence travel demand along three principal dimensions —density, diversity, and design. This paper tests this proposition by examining how the ‘3Ds’ affect trip rates and mode choice of residents in the San Francisco Bay Area. Using 1990 travel diary data and land-use records obtained from the U.S. census, regional inventories, and field surveys, models are estimated that relate features of the built environment to variations in vehicle miles traveled per household and mode choice, mainly for non-work trips. Factor analysis is used to linearly combine variables into the density and design dimensions of the built environment. The research finds that density, land-use diversity, and pedestrian-oriented designs generally reduce trip rates and encourage non-auto travel in statistically significant ways, though their influences appear to be fairly marginal. Elasticities between variables and factors that capture the 3Ds and various measures of travel demand are generally in the 0.06 to 0.18 range, expressed in absolute terms. Compact development was found to exert the strongest influence on personal business trips. Within-neighborhood retail shops, on the other hand, were most strongly associated with mode choice for work trips. And while a factor capturing ‘walking quality’ was only moderately related to mode choice for non-work trips, those living in neighborhoods with grid-iron street designs and restricted commercial parking were nonetheless found to average significantly less vehicle miles of travel and rely less on single-occupant vehicles for non-work trips. Overall, this research shows that the elasticities between each dimension of the built environment and travel demand are modest to moderate, though certainly not inconsequential. Thus it supports the contention of new urbanists and others that creating more compact, diverse, and pedestrian-orientated neighborhoods, in combination, can meaningfully influence how Americans travel.","DOI":"10.1016/S1361-9209(97)00009-6","ISSN":"1361-9209","shortTitle":"Travel demand and the 3Ds","journalAbbreviation":"Transportation Research Part D: Transport and Environment","author":[{"family":"Cervero","given":"Robert"},{"family":"Kockelman","given":"Kara"}],"issued":{"date-parts":[["1997",9,1]]}}}],"schema":""} Cervero and Kockelman proposed three attributes to characterize the built environment: diversity, design, and density (the D’s). Analyzing different metro areas from the US, they found that relationship between these variables and travel behavior are modest to moderate. Although they did not find strong relationships for each variable, they did suggest that the synergy among the three could cause impacts that are more appreciable. Other studies, focusing explicitly on walking, have found a stronger association with the built environment. Two reviews ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"hO89NUp1","properties":{"formattedCitation":"(S. L. Handy et al. 2002; Frank 2000)","plainCitation":"(S. L. Handy et al. 2002; Frank 2000)"},"citationItems":[{"id":146,"uris":[""],"uri":[""],"itemData":{"id":146,"type":"article-journal","title":"How the built environment affects physical activity: Views from urban planning","container-title":"American Journal of Preventive Medicine","page":"64-73","volume":"23","issue":"2","source":"","DOI":"10.1016/S0749-3797(02)00475-0","ISSN":"0749-3797, 1873-2607","shortTitle":"How the built environment affects physical activity","journalAbbreviation":"American Journal of Preventive Medicine","language":"English","author":[{"family":"Handy","given":"Susan L."},{"family":"Boarnet","given":"Marlon G."},{"family":"Ewing","given":"Reid"},{"family":"Killingsworth","given":"Richard E."}],"issued":{"date-parts":[["2002",8,1]]}}},{"id":144,"uris":[""],"uri":[""],"itemData":{"id":144,"type":"article-journal","title":"Land Use and Transportation Interaction: Implications on Public Health and Quality of Life","container-title":"Journal of Planning Education and Research","page":"6-22","volume":"20","issue":"1","source":"SAGE Journals","abstract":"Increases in per capita vehicle usage and associated emissions have spawned an increased examination of the ways in which our communities and regions are developing. Associated with increased vehicle usage are decreased levels of walking and biking, two valid forms of physical activity. The Surgeon General’s 1996 report, Physical Activity and Health, highlights the increasing level of physical inactivity as a growing cause of mortality. The costs and benefits of contrasting land development and transportation investment practices have been the subject of considerable debate in the literature. Findings have been refuted based on methodological grounds and inaccurate interpretation of data. Several of these studies, their methodological approaches, and their critiques are analyzed. While most agree that the built environment influences travel, considerable disagreement exists over the likely impacts of increased density, mix, and street connectivity on air quality, and on transportation system performance and household activity patterns.","DOI":"10.1177/073945600128992564","ISSN":"0739-456X","shortTitle":"Land Use and Transportation Interaction","journalAbbreviation":"Journal of Planning Education and Research","language":"en","author":[{"family":"Frank","given":"Lawrence D."}],"issued":{"date-parts":[["2000",9,1]]}}}],"schema":""} (Handy et al. 2002; Frank 2000) found positive and significant relationships between density and mixed land use with walking. In a 2006 review, ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"cgKxPS9Q","properties":{"formattedCitation":"(Sallis et al. 2006)","plainCitation":"(Sallis et al. 2006)"},"citationItems":[{"id":149,"uris":[""],"uri":[""],"itemData":{"id":149,"type":"article-journal","title":"An Ecological Approach to Creating Active Living Communities","container-title":"Annual Review of Public Health","page":"297-322","volume":"27","issue":"1","source":"Annual Reviews","abstract":"The thesis of this article is that multilevel interventions based on ecological models and targeting individuals, social environments, physical environments, and policies must be implemented to achieve population change in physical activity. A model is proposed that identifies potential environmental and policy influences on four domains of active living: recreation, transport, occupation, and household. Multilevel research and interventions require multiple disciplines to combine concepts and methods to create new transdisciplinary approaches. The contributions being made by a broad range of disciplines are summarized. Research to date supports a conclusion that there are multiple levels of influence on physical activity, and the active living domains are associated with different environmental variables. Continued research is needed to provide detailed findings that can inform improved designs of communities, transportation systems, and recreation facilities. Collaborations with policy researchers may improve the likelihood of translating research findings into changes in environments, policies, and practices.","DOI":"10.1146/annurev.publhealth.27.021405.102100","note":"PMID: 16533119","author":[{"family":"Sallis","given":"James F."},{"family":"Cervero","given":"Robert B."},{"family":"Ascher","given":"William"},{"family":"Henderson","given":"Karla A."},{"family":"Kraft","given":"M. Katherine"},{"family":"Kerr","given":"Jacqueline"}],"issued":{"date-parts":[["2006"]]}}}],"schema":""} Sallis et al. found that studies have yielded surprisingly consistent results but that utilitarian walking (and cycling) is generally higher in the areas with mixed uses, good street connectivity, and higher population densities. In another 2008 review, ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"nYo5fH6H","properties":{"formattedCitation":"(Saelens and Handy 2008)","plainCitation":"(Saelens and Handy 2008)"},"citationItems":[{"id":"un1Ot1kw/oiS6XK39","uris":[""],"uri":[""],"itemData":{"id":"un1Ot1kw/oiS6XK39","type":"article-journal","title":"Built Environment Correlates of Walking: A Review","container-title":"Medicine and science in sports and exercise","page":"S550-S566","volume":"40","issue":"7 Suppl","source":"PubMed Central","abstract":"Introduction\nThe past decade has seen a dramatic increase in the empirical investigation into the relations between built environmental and physical activity. To create places that facilitate and encourage walking, practitioners need an understanding of the specific characteristics of the built environment that correlate most strongly with walking. This paper reviews evidence on the built environment correlates with walking.\n\nMethod\nIncluded in this review were 13 reviews published between 2002 and 2006 and 29 original studies published in 2005 and up through May 2006. Results were summarized based on specific characteristics of the built environment and transportation walking versus recreational walking.\n\nResults\nPrevious reviews and newer studies document consistent positive relations between walking for transportation and density, distance to non-residential destinations, and land use mix; findings for route/network connectivity, parks and open space, and personal safety are more equivocal. Results regarding recreational walking were less clear.\n\nConclusions\nMore recent evidence supports the conclusions of prior reviews, and new studies address some of the limitations of earlier studies. Although prospective studies are needed, evidence on correlates appears sufficient to support policy changes.","DOI":"10.1249/MSS.0b013e31817c67a4","ISSN":"0195-9131","note":"PMID: 18562973\nPMCID: PMC2921187","shortTitle":"Built Environment Correlates of Walking","journalAbbreviation":"Med Sci Sports Exerc","author":[{"family":"Saelens","given":"Brian E."},{"family":"Handy","given":"Susan L."}],"issued":{"date-parts":[["2008",7]]}}}],"schema":""} Saelens and Handy (2008) found more consistency among these same relationships even after controlling for self-selection. However, they found evidence in some studies indicating that the random samples of travel observations taken from the general population do not ensure sufficient variation and that multicollinearity among built environment characteristics makes it hard to identify the unique contribution of a specific attribute. None of the studies acknowledges any location bias in the selection of study participants. Specifically, there is a lack of variation in the overall urban structure and range of built environments tested in the literature. As noted by Saelens and Handy, the majority of studies are based on a small number of countries including the US. Overall residential densities in the US are the lowest in the world ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"SAaVjJWQ","properties":{"formattedCitation":"(Newman 2014; Huang et al. 2007)","plainCitation":"(Newman 2014; Huang et al. 2007)"},"citationItems":[{"id":154,"uris":[""],"uri":[""],"itemData":{"id":154,"type":"article-journal","title":"Density, the Sustainability Multiplier: Some Myths and Truths with Application to Perth, Australia","container-title":"Sustainability","page":"6467-6487","volume":"6","issue":"9","source":"","abstract":"The paper suggests that the divisive urban issue of density has critical importance for sustainability. It is particularly important to resolve for the low density car dependent cities of the world as they are highly resource consumptive. Ten myths about density and 10 truths about density are proposed to help resolve the planning issues so commonly found to divide urban communities. They are applied with data to Perth to illustrate the issues and how they can be resolved.","DOI":"10.3390/su6096467","shortTitle":"Density, the Sustainability Multiplier","language":"en","author":[{"family":"Newman","given":"Peter"}],"issued":{"date-parts":[["2014",9,25]]}}},{"id":155,"uris":[""],"uri":[""],"itemData":{"id":155,"type":"article-journal","title":"A global comparative analysis of urban form: Applying spatial metrics and remote sensing","container-title":"Landscape and Urban Planning","page":"184-197","volume":"82","issue":"4","source":"ScienceDirect","abstract":"Currently, debates over urban form have generally focused on the contrast between the “sprawl” often seen as typical of the United States and “compact” urban forms found in parts of Europe. Although these debates are presumed to have implications for developing worlds as well, systematic comparison of urban forms between developed and developing countries has been lacking. This paper utilized satellite images of 77 metropolitan areas in Asia, US, Europe, Latin America and Australia to calculate seven spatial metrics that capture five distinct dimensions of urban form. Comparison of the spatial metrics was firstly made between developed and developing countries, and then among world regions. A cluster analysis classifies the cities into groups in terms of these spatial metrics. The paper also explored the origins of differences in urban form through comparison with socio-economic developmental indicators and historical trajectories in urban development. The result clearly demonstrates that urban agglomerations of developing world are more compact and dense than their counterparts in either Europe or North America. Moreover, there are also striking differences in urban form across regions.","DOI":"10.1016/j.landurbplan.2007.02.010","ISSN":"0169-2046","shortTitle":"A global comparative analysis of urban form","journalAbbreviation":"Landscape and Urban Planning","author":[{"family":"Huang","given":"Jingnan"},{"family":"Lu","given":"X. X."},{"family":"Sellers","given":"Jefferey M."}],"issued":{"date-parts":[["2007",10,17]]}}}],"schema":""} (Newman 2014; Huang et al. 2007). Additionally, the low level of compactness ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"qplK3p0c","properties":{"formattedCitation":"(Huang et al. 2007)","plainCitation":"(Huang et al. 2007)"},"citationItems":[{"id":155,"uris":[""],"uri":[""],"itemData":{"id":155,"type":"article-journal","title":"A global comparative analysis of urban form: Applying spatial metrics and remote sensing","container-title":"Landscape and Urban Planning","page":"184-197","volume":"82","issue":"4","source":"ScienceDirect","abstract":"Currently, debates over urban form have generally focused on the contrast between the “sprawl” often seen as typical of the United States and “compact” urban forms found in parts of Europe. Although these debates are presumed to have implications for developing worlds as well, systematic comparison of urban forms between developed and developing countries has been lacking. This paper utilized satellite images of 77 metropolitan areas in Asia, US, Europe, Latin America and Australia to calculate seven spatial metrics that capture five distinct dimensions of urban form. Comparison of the spatial metrics was firstly made between developed and developing countries, and then among world regions. A cluster analysis classifies the cities into groups in terms of these spatial metrics. The paper also explored the origins of differences in urban form through comparison with socio-economic developmental indicators and historical trajectories in urban development. The result clearly demonstrates that urban agglomerations of developing world are more compact and dense than their counterparts in either Europe or North America. Moreover, there are also striking differences in urban form across regions.","DOI":"10.1016/j.landurbplan.2007.02.010","ISSN":"0169-2046","shortTitle":"A global comparative analysis of urban form","journalAbbreviation":"Landscape and Urban Planning","author":[{"family":"Huang","given":"Jingnan"},{"family":"Lu","given":"X. X."},{"family":"Sellers","given":"Jefferey M."}],"issued":{"date-parts":[["2007",10,17]]}}}],"schema":""} (Huang et al. 2007) adds the question of whether the density gradients have a role. Furthermore, having low residential densities on average means higher density areas may be scarce, thus, making them “hidden populations” or “hard-to-reach populations” for gathering travel behavior data from surveys. This means that the sampling process has to be very large to have a good accuracy level, or that places with a desirable density for walking may be nonexistent in many cases, especially in smaller size cities or towns. Thus, empirically-derived relationships between walking and the built environment using only lower-density observations may not apply to higher-density places or regions. A study of differences between major cities and rural small towns found that smaller towns did not have the nearby destinations to support walking, compared with city centers of major metropolitan areas that did have close destinations ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"OsKqwbly","properties":{"formattedCitation":"(Stewart et al. 2016)","plainCitation":"(Stewart et al. 2016)"},"citationItems":[{"id":157,"uris":[""],"uri":[""],"itemData":{"id":157,"type":"article-journal","title":"Comparing Associations Between the Built Environment and Walking in Rural Small Towns and a Large Metropolitan Area","container-title":"Environment and Behavior","page":"13-36","volume":"48","issue":"1","source":"SAGE Journals","abstract":"The association between the built environment (BE) and walking has been studied extensively in urban areas, yet little is known whether the same associations hold for smaller, rural towns. This analysis examined objective measures of the BE around participants’ residence and their utilitarian and recreational walking from two studies, one in the urban Seattle area (n = 464) and the other in nine small U.S. towns (n = 299). After adjusting for sociodemographics, small town residents walked less for utilitarian purposes but more for recreational purposes. These differences were largely explained by differential associations of the BE on walking in the two settings. In Seattle, the number of neighborhood restaurants was positively associated with utilitarian walking, but in small towns, the association was negative. In small towns, perception of slow traffic on nearby streets was positively associated with recreational walking, but not in Seattle. These observations suggest that urban–rural context matters when planning BE interventions to support walking.","DOI":"10.1177/0013916515612253","ISSN":"0013-9165","journalAbbreviation":"Environment and Behavior","language":"en","author":[{"family":"Stewart","given":"Orion T."},{"family":"Vernez Moudon","given":"Anne"},{"family":"Saelens","given":"Brian E."},{"family":"Lee","given":"Chanam"},{"family":"Kang","given":"Bumjoon"},{"family":"Doescher","given":"Mark P."}],"issued":{"date-parts":[["2016",1,1]]}}}],"schema":""} (Stewart et al. 2016).Another significant approach to determine the degree to which built environment characteristics affects travel behavior has been the calculation of elasticities from a meta-analysis or meta-regressions ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"Ch3HRMz0","properties":{"formattedCitation":"(Ewing and Cervero 2010; Stevens 2017)","plainCitation":"(Ewing and Cervero 2010; Stevens 2017)"},"citationItems":[{"id":160,"uris":[""],"uri":[""],"itemData":{"id":160,"type":"article-journal","title":"Travel and the Built Environment","container-title":"Journal of the American Planning Association","page":"265-294","volume":"76","issue":"3","source":"Taylor and Francis+NEJM","abstract":"Problem: Localities and states are turning to land planning and urban design for help in reducing automobile use and related social and environmental costs. The effects of such strategies on travel demand have not been generalized in recent years from the multitude of available studies. Purpose: We conducted a meta-analysis of the built environment-travel literature existing at the end of 2009 in order to draw generalizable conclusions for practice. We aimed to quantify effect sizes, update earlier work, include additional outcome measures, and address the methodological issue of self-selection. Methods: We computed elasticities for individual studies and pooled them to produce weighted averages. Results and conclusions: Travel variables are generally inelastic with respect to change in measures of the built environment. Of the environmental variables considered here, none has a weighted average travel elasticity of absolute magnitude greater than 0.39, and most are much less. Still, the combined effect of several such variables on travel could be quite large. Consistent with prior work, we find that vehicle miles traveled (VMT) is most strongly related to measures of accessibility to destinations and secondarily to street network design variables. Walking is most strongly related to measures of land use diversity, intersection density, and the number of destinations within walking distance. Bus and train use are equally related to proximity to transit and street network design variables, with land use diversity a secondary factor. Surprisingly, we find population and job densities to be only weakly associated with travel behavior once these other variables are controlled. Takeaway for practice: The elasticities we derived in this meta-analysis may be used to adjust outputs of travel or activity models that are otherwise insensitive to variation in the built environment, or be used in sketch planning applications ranging from climate action plans to health impact assessments. However, because sample sizes are small, and very few studies control for residential preferences and attitudes, we cannot say that planners should generalize broadly from our results. While these elasticities are as accurate as currently possible, they should be understood to contain unknown error and have unknown confidence intervals. They provide a base, and as more built-environment/travel studies appear in the planning literature, these elasticities should be updated and refined. Research support: U.S. Environmental Protection Agency.","DOI":"10.1080/01944361003766766","ISSN":"0194-4363","author":[{"family":"Ewing","given":"Reid"},{"family":"Cervero","given":"Robert"}],"issued":{"date-parts":[["2010",6,21]]}}},{"id":118,"uris":[""],"uri":[""],"itemData":{"id":118,"type":"article-journal","title":"Does Compact Development Make People Drive Less?","container-title":"Journal of the American Planning Association","page":"7-18","volume":"83","issue":"1","source":"Taylor and Francis+NEJM","abstract":"Problem, research strategy, and findings: Planners commonly recommend compact development in part as a way of getting people to drive less, with the idea that less driving will lead to more sustainable communities. Planners base their recommendations on a substantial body of research that examines the impact of compact development on driving. Different studies, however, have found different outcomes: Some studies find that compact development causes people to drive less, while other studies do not. I use meta-regression analysis to a) explain why different studies on driving and compact development yield different results, and b) combine different findings from many studies into reliable statistics that can better inform planning practice. I address the following questions: Does compact development make people drive less, and if so, how much less? I find that compact development does make people drive less, because most of the compact development features I study have a statistically significant negative influence on driving. The impact, however, is fairly small: Compact development features do not appear to have much influence on driving. My findings are limited to some extent because they are derived from small sample sizes.Takeaway for practice: Planners should not rely on compact development as their only strategy for reducing driving unless their goals for reduced driving are very modest and can be achieved at a low cost.","DOI":"10.1080/01944363.2016.1240044","ISSN":"0194-4363","author":[{"family":"Stevens","given":"Mark R."}],"issued":{"date-parts":[["2017",1,2]]}}}],"schema":""} (Ewing and Cervero 2010; Stevens 2017). This has been done for different cities and samples, with estimated elasticities of built environment variables ranging 0.0–0.3, suggesting an inelastic relationship with walking. Further discussions have focused on establishing the relative vs. absolute magnitude of this number (is it large or not?) but have neglected the potential complementarity among those variables characterizing the built environment. Besides, this approach avoids the discussion of multicollinearity in the characteristics of the built environment. The latest debates have centered on the usefulness of these methods ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"D8Y6Y1Hk","properties":{"formattedCitation":"(K. J. Clifton 2017; S. Handy 2017; Ewing and Cervero 2017)","plainCitation":"(K. J. Clifton 2017; S. Handy 2017; Ewing and Cervero 2017)"},"citationItems":[{"id":163,"uris":[""],"uri":[""],"itemData":{"id":163,"type":"article-journal","title":"Getting From Here to There: Comment on “Does Compact Development Make People Drive Less?”","container-title":"Journal of the American Planning Association","page":"148-151","volume":"83","issue":"2","source":"Taylor and Francis+NEJM","DOI":"10.1080/01944363.2017.1290494","ISSN":"0194-4363","shortTitle":"Getting From Here to There","author":[{"family":"Clifton","given":"Kelly J."}],"issued":{"date-parts":[["2017",4,3]]}}},{"id":166,"uris":[""],"uri":[""],"itemData":{"id":166,"type":"article-journal","title":"Thoughts on the Meaning of Mark Stevens’s Meta-Analysis","container-title":"Journal of the American Planning Association","page":"26-28","volume":"83","issue":"1","source":"Taylor and Francis+NEJM","DOI":"10.1080/01944363.2016.1246379","ISSN":"0194-4363","author":[{"family":"Handy","given":"Susan"}],"issued":{"date-parts":[["2017",1,2]]}}},{"id":169,"uris":[""],"uri":[""],"itemData":{"id":169,"type":"article-journal","title":"“Does Compact Development Make People Drive Less?” The Answer Is Yes","container-title":"Journal of the American Planning Association","page":"19-25","volume":"83","issue":"1","source":"Taylor and Francis+NEJM","DOI":"10.1080/01944363.2016.1245112","ISSN":"0194-4363","shortTitle":"“Does Compact Development Make People Drive Less?","author":[{"family":"Ewing","given":"Reid"},{"family":"Cervero","given":"Robert"}],"issued":{"date-parts":[["2017",1,2]]}}}],"schema":""} (Clifton 2017; Handy 2017; Ewing and Cervero 2017), as they have become common among different fields. The main concerns regarding these methods are the absence of longitudinal studies as well as the assumptions of linearity behind them, as there may be evidence that is not the case ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"vsT5EfPK","properties":{"formattedCitation":"(Talen and Koschinsky 2014)","plainCitation":"(Talen and Koschinsky 2014)"},"citationItems":[{"id":172,"uris":[""],"uri":[""],"itemData":{"id":172,"type":"article-journal","title":"Compact, Walkable, Diverse Neighborhoods:Assessing Effects on Residents","container-title":"Housing Policy Debate","page":"717-750","volume":"24","issue":"4","source":"Taylor and Francis+NEJM","abstract":"What research supports the view that compact, walkable, diverse (CWD) neighborhoods are beneficial for urban residents? To make this assessment, we searched the literature to try to understand the current status of evidence regarding claims about the CWD neighborhood. We find that research linking CWD neighborhoods to effects on residents coalesces around three main topics: social relations, health, and safety. We conclude that on the basis of the literature reviewed, most of the intended benefits of the CWD neighborhood have been researched and found to have significant, positive effects for urban dwellers. While physical factors are but one element affecting behavior and outcomes, and the issues of self-selection and causality remain, overall, key dimensions of the CWD neighborhood have been found to positively affect social interaction, health, and safety.","DOI":"10.1080/10511482.2014.900102","ISSN":"1051-1482","shortTitle":"Compact, Walkable, Diverse Neighborhoods","author":[{"family":"Talen","given":"Emily"},{"family":"Koschinsky","given":"Julia"}],"issued":{"date-parts":[["2014",10,2]]}}}],"schema":""} (Talen and Koschinsky 2014). Although these issues may be responsible for mixed results in studies of the relationships between walking and the built environment, there is extensive evidence that these relationships exist. Nevertheless, there is no consensus around the level or magnitude of effects and suggests that findings may have limited applicability in other locations. This project tries to fill these gaps by building and testing new indicators of the built environment from data sources that are more available and from there test how transferable are the measures among cities. ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"ikOgcMCS","properties":{"formattedCitation":"(Fox et al. 2014)","plainCitation":"(Fox et al. 2014)"},"citationItems":[{"id":66,"uris":[""],"uri":[""],"itemData":{"id":66,"type":"article-journal","title":"Temporal transferability of models of mode-destination choice for the Greater Toronto and Hamilton Area","container-title":"Journal of Transport and Land Use","page":"41","volume":"7","issue":"2","source":"CrossRef","DOI":"10.5198/jtlu.v7i2.701","ISSN":"1938-7849","author":[{"family":"Fox","given":"James"},{"family":"Daly","given":"Andrew"},{"family":"Hess","given":"Stephane"},{"family":"Miller","given":"Eric"}],"issued":{"date-parts":[["2014",7,28]]}}}],"schema":""} Fox et al. (2014) identify a clear distinction between two types of model transferability: temporal transferability that refers to the validity of the models for long-term predictions, and spatial transferability that refers to the validity of models across different regions. For both types of transferability, there are methods to test model transferability ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"4xcBaIyk","properties":{"formattedCitation":"(Ben-Akiva 1981; Lerman 1981)","plainCitation":"(Ben-Akiva 1981; Lerman 1981)"},"citationItems":[{"id":198,"uris":[""],"uri":[""],"itemData":{"id":198,"type":"chapter","title":"Issues in transferring and updating travel behavior models","container-title":"New Horizons in Travel-Behaviour Research","publisher":"Lexington Books","publisher-place":"Lexington","page":"665-686","source":"trid.","event-place":"Lexington","URL":"","author":[{"family":"Ben-Akiva","given":"Moshe"}],"editor":[{"family":"Stopher","given":"Peter R."},{"family":"Meyburg","given":"A.H."},{"family":"Br?g","given":"W."}],"issued":{"date-parts":[["1981"]]},"accessed":{"date-parts":[["2018",4,15]]}}},{"id":200,"uris":[""],"uri":[""],"itemData":{"id":200,"type":"chapter","title":"A comment on interspatial, intraspatial, and temporal transferability","container-title":"New Horizons in Travel-Behaviour Research","publisher":"Lexington Books","publisher-place":"Lexington","page":"628-632","event-place":"Lexington","author":[{"family":"Lerman","given":"R"}],"editor":[{"family":"Stopher","given":"Peter R."},{"family":"Meyburg","given":"A.H."},{"family":"Br?g","given":"W."}],"issued":{"date-parts":[["1981"]]}}}],"schema":""} (Ben-Akiva 1981; Lerman 1981) and approaches to updating the model parameters ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"5R3QI7d7","properties":{"formattedCitation":"(Ben-Akiva and Bolduc 1987; Badoe and Miller 1995)","plainCitation":"(Ben-Akiva and Bolduc 1987; Badoe and Miller 1995)"},"citationItems":[{"id":201,"uris":[""],"uri":[""],"itemData":{"id":201,"type":"article-journal","title":"APPROACHES TO MODEL TRANSFERABILITY AND UPDATING: THE COMBINED TRANSFER ESTIMATOR","container-title":"Transportation Research Record","issue":"1139","source":"trid.","URL":"","ISSN":"0361-1981","shortTitle":"APPROACHES TO MODEL TRANSFERABILITY AND UPDATING","author":[{"family":"Ben-Akiva","given":"Moshe"},{"family":"Bolduc","given":"Denis"}],"issued":{"date-parts":[["1987"]]},"accessed":{"date-parts":[["2018",4,16]]}}},{"id":206,"uris":[""],"uri":[""],"itemData":{"id":206,"type":"article-journal","title":"ANALYSIS OF TEMPORAL TRANSFERABILITY OF DISAGGREGATE WORK TRIP MODE CHOICE MODELS","container-title":"Transportation Research Record","issue":"1493","source":"trid.","URL":"","ISSN":"0361-1981","author":[{"family":"Badoe","given":"Daniel A."},{"family":"Miller","given":"Eric J."}],"issued":{"date-parts":[["1995"]]},"accessed":{"date-parts":[["2018",4,16]]}}}],"schema":""} (Ben-Akiva and Bolduc 1987; Badoe and Miller 1995). However, the focus in the literature has been on identifying the statistical methods to transfer models and testing the accuracy of the prediction, rather than understanding the behavioral responses under different contexts. For example, ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"2gz9cgUx","properties":{"formattedCitation":"(Nowrouzian and Srinivasan 2012)","plainCitation":"(Nowrouzian and Srinivasan 2012)"},"citationItems":[{"id":67,"uris":[""],"uri":[""],"itemData":{"id":67,"type":"article-journal","title":"Empirical Analysis of Spatial Transferability of Tour-Generation Models","container-title":"Transportation Research Record: Journal of the Transportation Research Board","page":"14-22","volume":"2302","source":"CrossRef","DOI":"10.3141/2302-02","ISSN":"0361-1981","language":"en","author":[{"family":"Nowrouzian","given":"Roosbeh"},{"family":"Srinivasan","given":"Sivaramakrishnan"}],"issued":{"date-parts":[["2012",12]]}}}],"schema":""} Nowrouzian and Srinivasan (2012) test spatial transferability of activity-based models in Florida and identify a gap in the study of transferability in this types of models. They found that transferability is limited and that further research must be done to identify the range of contexts in which the model can be applied. ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"1AIpoth6","properties":{"formattedCitation":"(Karasmaa 2007)","plainCitation":"(Karasmaa 2007)"},"citationItems":[{"id":51,"uris":[""],"uri":[""],"itemData":{"id":51,"type":"article-journal","title":"Evaluation of transfer methods for spatial travel demand models","container-title":"Transportation Research Part A: Policy and Practice","page":"411-427","volume":"41","issue":"5","source":"CrossRef","DOI":"10.1016/j.tra.2006.09.009","ISSN":"09658564","language":"en","author":[{"family":"Karasmaa","given":"Nina"}],"issued":{"date-parts":[["2007",6]]}}}],"schema":""} Karasmaa (2007) tests the spatial transferability of the simple four-step regional model in two different urban areas in Finland. The research shows that the models are largely transferable between these locations; however, doubts arise as to whether this evidence supports the idea of more universal model transferability or is limited to similar urban structures and cultural contexts. Another limitation in the transferability literature is that regional models focus more on motorized vehicles than walking and biking (see ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"Sfsh2cD7","properties":{"formattedCitation":"(Singleton et al. 2018)","plainCitation":"(Singleton et al. 2018)"},"citationItems":[{"id":194,"uris":[""],"uri":[""],"itemData":{"id":194,"type":"article-journal","title":"Making Strides: State-of-the-Practice of Pedestrian Forecasting in Regional Travel Models","container-title":"Transportation Research Record: Journal of the Transportation Research Board","abstract":"Much has changed in the 30 years since non-motorized modes were first included in regional travel demand models. As interest in understanding behavioral influences on walking and policies requiring estimates of walking activity increase, it is important to consider how pedestrian travel is modeled at a regional level. This paper evaluates the state-of-the-practice of modeling walk trips among the largest 48 metropolitan planning organizations (MPOs) and assesses changes made over the last five years. By reviewing model documentation and responses to a survey of MPO modelers, this paper summarizes current practices, describes six pedestrian modeling frameworks, and identifies trends. Three-quarters (75%) of large MPOs now model non-motorized travel, and over two-thirds (69%) of those MPOs distinguish walking from bicycling; these percentages are up from nearly two-thirds (63%) and one-half (47%), respectively, in 2012. This change corresponds with an increase in the deployment of activity-based models, which offer the opportunity to enhance pedestrian modeling techniques. The biggest barrier to more sophisticated models remains a lack of travel survey data on walking behavior, yet some MPOs are starting to overcome this challenge by oversampling potential active travelers. Decision-makers are becoming more interested in analyzing walking and using estimates of walking activity that are output from models for various planning applications. As the practice continues to mature, the near future will likely see smaller scale measures of the pedestrian environment, more detailed zonal and network structures, and possibly even an operational model of pedestrian route choice.","author":[{"family":"Singleton","given":"Patrick A."},{"family":"Totten","given":"Joseph C."},{"family":"Orrego-O?ate","given":"Jaime P."},{"family":"Schneider","given":"Robert J."},{"family":"Clifton","given":"Kelly J."}],"issued":{"date-parts":[["2018"]],"season":"forthcoming in"}}}],"schema":""} Singleton et al. 2018) and thus a specific examination of the transferability of relationships between the built environment and walking behavior is lacking. Thus, this project will try to fill this gap to test if there are systematic behavioral responses across and within different cities.PEdestrian Index of the Environment (PIE)This study builds on our previous work by examining and improving upon the Pedestrian Index of the Environment (PIE), developed and described in the project reports ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"6kZLRrHF","properties":{"formattedCitation":"(K. Clifton et al. 2013, 2015)","plainCitation":"(K. Clifton et al. 2013, 2015)"},"citationItems":[{"id":187,"uris":[""],"uri":[""],"itemData":{"id":187,"type":"article-journal","title":"Improving the Representation of the Pedestrian Environment in Travel Demand Models, Phase I","container-title":"Civil and Environmental Engineering Faculty Publications and Presentations","URL":"","DOI":"10.15760/trec.120","author":[{"family":"Clifton","given":"Kelly"},{"family":"Singleton","given":"Patrick"},{"family":"Muhs","given":"Christopher"},{"family":"Schneider","given":"Robert"},{"family":"Lagerwey","given":"Peter"}],"issued":{"date-parts":[["2013",9,1]]}}},{"id":185,"uris":[""],"uri":[""],"itemData":{"id":185,"type":"article-journal","title":"Development of a Pedestrian Demand Estimation Tool","container-title":"Civil and Environmental Engineering Faculty Publications and Presentations","URL":"","DOI":"10.15760/trec.124","author":[{"family":"Clifton","given":"Kelly"},{"family":"Singleton","given":"Patrick"},{"family":"Muhs","given":"Christopher"},{"family":"Schneider","given":"Robert"}],"issued":{"date-parts":[["2015",9,1]]}}}],"schema":""} (Clifton et al. 2013, 2015). In order to understand this study, it is important to give a brief overview of our original concept of PIE, referred to as PIE0 in this report. PIE0 is an index comprising six built environment measures, listed below, and was computed for each PAZ in the system (N= 1,465,252 in the four-county metro area of Portland, OR). comfortable cycling facilities; a proxy for low volume streets with traffic calming (local density of weighted bicycle network links in a one-mile buffer of each PAZ (1.6 km)); block size (average block size in a quarter-mile (0.4 km) buffer of each PAZ); people per acre (population plus employment density in a quarter-mile (0.4 km) buffer of each PAZ); sidewalk density (total length of metro sidewalk inventory within a quarter-mile buffer of each PAZ);transit access (a density measure of transit stops weighted by the service frequency in a quarter-mile buffer); and urban living infrastructure (ULI) (the number of services, entertainment, and retail services in a quarter-mile buffer of the origin PAZ). In the calculation of the index, each of the built environment measures is weighted based upon their relative importance to walking behavior (see Clifton et al., 2013 for more information). To calculate these weights, we used the standardized coefficients from a univariate model that predicts the likelihood of walking as a function of each of the six built environment measures for the PAZ of the trip origin. For each built environment attribute j, a univariate (walk/don't walk), binary-logit mode choice model Pj is estimated to predict walking/other mode using as an explanatory variable the attribute score zj . Each model calibration delivers a utility function:Uj=β0j+β1jzjAfter this, PIE0 was constructed by weighting each built environment variable by the corresponding estimated coefficient β1j and summing the values over a specific spatial unit. PIE0=jβ1jzjThe dataset used to provide the built environment characteristics used in PIE0 is the Context Tool, developed by Portland Metro. This dataset calculates the built environment characteristics within 80-meter by 80-meter grid cells (called Pedestrian Analysis Zones, or PAZs) and associates each raw measure with discrete values between 1 and 5. These discrete values were assigned using natural breaks in the distribution of each measure. Additionally, the travel behavior data used in the estimation of the binary logit models for the weights used in PIE0 were the Oregon Household Travel Survey (OHAS) from 2011. The sample consisted of 56,634 trips for a regular day, 36,463 trips of which correspond to the area where Portland Metro Context Tool had measured. The coefficients for PIE0 are shown in REF _Ref521597375 \h Table 1 and the values for PIE0 for the Portland region are shown in REF _Ref513458620 \h Figure 1. The advantages of PIE0 were that it captured fine-grained variations in the walking environment at the PAZ scale (80-meters by 80-meters). This improves upon the course scale used in most regional travel models, where Traffic Analysis Zones (TAZs) are the unit of analysis. Additionally, this index weighted the different built environment attributes by their association with walking behavior and improves our characterization of “walkability”. Thus, PIE0 was statistically significant in our later analysis of travel behaviors (mode choice and trip generation). In addition to the contributions made by PIE0, MoPeD can be used as a stand-alone demand tool or in concert with Metro's regional travel model. Despite these improvements, the methodology still has some limitations. First, the data sources used for the model estimation used discrete values between 1 and 5 using natural breaks to transform the data rather than the direct values of the built environment variables. This presents problems in transferring the measures and the models to other places since each location will have its own distributions of these variables (and different corresponding natural breaks). Table SEQ Table \* ARABIC 1 REF _Ref513459399 \h Binary Logit Models for PIE0Context Variable (zn)Coefficient (β)p-valueModel pseudo-R2Model 1Bicycle Access 0.4940.000.057Constant-4.0470.00Model 2Block Size0.5430.000.096Constant-3.7290.00Model 3People per Acre0.8120.000.095Constant-4.3040.00Model 4Sidewalk Density0.5000.000.083Constant-3.9000.00Model 5Transit Access0.6210.000.083Constant-3.3860.00Model 6ULI Density0.5490.000.073Constant-3.2040.00Data used for all modelsTrips (n)36,463Walk3,560Not Walk32,903Figure 1 Regional Map of the Built Environment Index PIE0 Data and MEthods for NEW PIE (PIEbg) Our objective at large was to re-create pedestrian-specific measures of the built environment at a coarser scale (for the purpose of using more readily available data nation-wide), compare them with PIE0, and test whether their predictive capacity in models of walking mode choice in Portland, OR. The extension to this methodology was to estimate a new index that we called PIEbg, named for the spatial scale of its construct: The Census block group geographic unit. There is a total of 1,324 block groups with an average area of 11.4 sq. meters (SD = 111.1 sq. meters). REF _Ref511669394 \h Figure 2 shows the difference between both units of analysis. The block group scale is similar to the traffic analysis zones (TAZ) commonly used in transportation planning methods. PIEbg was calibrated using an analogous methodology as PIE0. One of our objectives was to test the loss of power of our models in changing the scale and the data source, so we framed the PIEbg score to be similarly constructed as PIE0. In PIE0 the built environment data were scaled using natural breaks and assigned discrete values between 1 and 5. However, in PIEbg, we decided that this method lose too much information and the induce error could even be higher at the larger scale of the block group, particularly when using a block group scale and thus, continuous values were assigned. For that reason, we scaled the data to be between 1 and 5 using continuous breaks. This is done by using the following formula:znew=5-1max?(Z)-min?(Z)*zold-minZ+1,where z is the value that wants to be transform and the max and min values correspond to the maximum and minimum value of the built environment data set for each city.After this, we used the same methodology of PIE0 where the weights of each attribute of the built environment are the coefficients of univariate logit models of the choice to walk estimated using household travel survey data for each region. The new indicator PIEbg is again transformed by multiplying with a constant to have a minimum value of 20 and a maximum value of 100 as with PIE0.constant= 20β1j*minzjOne of the objectives of this project was to use widely-available data in order to make our measures of PIE transferrable. The US Environmental Protection Agency (EPA) created a unique dataset called the Smart Location Database (SLD) ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"V008b6ih","properties":{"formattedCitation":"(Ramsey and Bell 2014; US EPA 2014)","plainCitation":"(Ramsey and Bell 2014; US EPA 2014)"},"citationItems":[{"id":192,"uris":[""],"uri":[""],"itemData":{"id":192,"type":"article-journal","title":"The Smart Location Database: A nationwide data resource characterizing the built Environment and destination accessibility at the neighborhood scale","container-title":"Cityscape","page":"145","volume":"16","issue":"2","source":"Google Scholar","shortTitle":"The Smart Location Database","author":[{"family":"Ramsey","given":"Kevin"},{"family":"Bell","given":"Alexander"}],"issued":{"date-parts":[["2014"]]}}},{"id":208,"uris":[""],"uri":[""],"itemData":{"id":208,"type":"webpage","title":"Smart Location Database Technical Documentation and User Guide","container-title":"US EPA","genre":"Data and Tools","abstract":"This report has a detailed description of the data sources and methodologies used to calculate the variables in the Smart Location Database.","URL":"","language":"en","author":[{"family":"US EPA","given":"OA"}],"issued":{"date-parts":[["2014",2,27]]},"accessed":{"date-parts":[["2018",4,16]]}}}],"schema":""} (Ramsey and Bell 2014; US EPA 2014) that includes information from the Census, the street design from Navteq’s NAVSTREETS street data, and transit data form each local agency, in General Transit Feed Specification (GTFS), that include all the time schedule and transit routes. This data source was selected because it is a resource that offers a large number of built environment variables, measured using equivalent methods, and it covers the entire country at a relatively small spatial scale. The Census block groups are clusters of blocks with a population between 600 and 3,000 people. Although pedestrian behavior is associated with a broad array of built environment characteristics, the following were selected based upon their availability in the SLD database (and thus their availability in all of our US study locations), their strong association with walking, and similarity to that used in PIE0.People per acre: It is the addition of population and employment density. Population density is the gross population density of people per acre using the population of the decennial 2010 Census and the unprotected area defined in the Census Block Group. Employment density is calculated using the Census Longitudinal Employer-Household Dynamics (LEHD) data and dividing it by the unprotected area defined in the Census Block Group in acres.Urban living infrastructure (ULI): Using the same data from employment density we filtered to include only commercial and entertainment employment. However, unlike the original ULI measure, which is a count of these establishments within a quarter-mile buffer, this ULI measure is a density measure. While they are both effectively describing density, the coefficients from these two different ULI measures cannot be directly compared because of their different construction. Transit accessibility: Aggregate frequency of all transit in a buffer of 0.25 miles of the block group in the peak hour using GTFS data from the local transit agencies. Total road network density: The total miles of roadway per square mile. Highways that are divided are counted as two separate roadways. These variables in this new index differ from those used in the original PIE0 in two ways. First, our original measures of connectivity, block size, was substituted with total road network density. The reason for this is partially convenience: it is not available in the SLD dataset. The other alternative, intersection densities, were poorly defined in the SLD database making it difficult to interpret. Additionally, ADDIN ZOTERO_ITEM CSL_CITATION {"citationID":"T4dPLLEJ","properties":{"formattedCitation":"(Berrigan, Pickle, and Dill 2010)","plainCitation":"(Berrigan, Pickle, and Dill 2010)","noteIndex":0},"citationItems":[{"id":199,"uris":[""],"uri":[""],"itemData":{"id":199,"type":"article-journal","title":"Associations between street connectivity and active transportation","container-title":"International Journal of Health Geographics","page":"20","volume":"9","issue":"1","source":"Crossref","abstract":"Background: Past studies of associations between measures of the built environment, particularly street connectivity, and active transportation (AT) or leisure walking/bicycling have largely failed to account for spatial autocorrelation of connectivity variables and have seldom examined both the propensity for AT and its duration in a coherent fashion. Such efforts could improve our understanding of the spatial and behavioral aspects of AT. We analyzed spatially identified data from Los Angeles and San Diego Counties collected as part of the 2001 California Health Interview Survey.\nResults: Principal components analysis indicated that ~85% of the variance in nine measures of street connectivity are accounted for by two components representing buffers with short blocks and dense nodes (PRIN1) or buffers with longer blocks that still maintain a grid like structure (PRIN2). PRIN1 and PRIN2 were positively associated with active transportation (AT) after adjustment for diverse demographic and health related variables. Propensity and duration of AT were correlated in both Los Angeles (r = 0.14) and San Diego (r = 0.49) at the zip code level. Multivariate analysis could account for the correlation between the two outcomes. After controlling for demography, measures of the built environment and other factors, no spatial autocorrelation remained for propensity to report AT (i.e., report of AT appeared to be independent among neighborhood residents). However, very localized correlation was evident in duration of AT, particularly in San Diego, where the variance of duration, after accounting for spatial autocorrelation, was 5% smaller within small neighborhoods (~0.01 square latitude/longitude degrees = 0.6 mile diameter) compared to within larger zip code areas. Thus a finer spatial scale of analysis seems to be more appropriate for explaining variation in connectivity and AT.\nConclusions: Joint analysis of the propensity and duration of AT behavior and an explicitly geographic approach can strengthen studies of the built environment and physical activity (PA), specifically AT. More rigorous analytical work on cross-sectional data, such as in the present study, continues to support the need for experimental and longitudinal study designs including the analysis of natural experiments to evaluate the utility of environmental interventions aimed at increasing PA.","DOI":"10.1186/1476-072X-9-20","ISSN":"1476-072X","language":"en","author":[{"family":"Berrigan","given":"David"},{"family":"Pickle","given":"Linda W"},{"family":"Dill","given":"Jennifer"}],"issued":{"date-parts":[["2010"]]}}}],"schema":""} Berrigan et al. (2010) showed that intersection density and network density have almost a 90% of correlation, making them very similar. Therefore, we used total road network density to assess connectivity, which also gives the best fit in the model. Second, we did not include measures of sidewalk density and comfortable cycling facilities in our new construct. This omission was justified as both measures are not widely available. Further, the comfortable cycling facilities variable itself has a weak theoretical association with pedestrian travel. It was originally incorporated in PIE0 a proxy measure for the degree to which streets accommodated multimodal travel. Furthermore, the primary motivation for this exercise is to test how transferable are the relationships in PIE are across different locations and not necessarily to estimate a model to use for forecasting. Figure 2 Comparison between block groups (bold line) and PAZ (grid) in downtown Portland, ORThe next step in the process was to construct the built environment measures from the SLD data. The dataset, as explained previously, includes the residential density and employment density, so those measures were directly incorporated. To create ULI, we used commercial and retail employment density measures that also was a direct calculation. The total road network and the transit aggregated frequency were defined directly in the database so no calculation was needed.In synthesis, PIEbg is estimated with the same method of aggregating the univariate coefficients of the four variables described above. This process is as follows: Rescale the built environment variables from the SLD values to be a continuous measure between 1 and 5, with 1 representing the group of lowest values and 5 the highest.Do a spatial join of the Oregon Household Survey trips origins data with the SLD. In this way, we manage to get a data set of each trip with the associated data of the block group at the trip origin.Estimate individual univariate binary logit models of walking mode choice as a function of each of the built environment variables. Take the coefficients of each model and generate a PIEbg value for the block group level with the weights. The coefficients are multiplied by a constant so that PIEbg has a minimum value of 20 and a maximum value of a 100. For this task, we replicate the same method used in PIE0. The final PIEbg score is calculated using the following formula.PIEbg=4.6*people per acre+4.8*ULI+4.7*transit accessibility+6.1*street network densityTesting the transferability of PIEbg This is section we present the results of our testing, comparison, and analysis of both walkability measures PIE0 and PIEbg constructed at different scales and with different data inputs. First, we compare our both walkability measures. Then, we test the relationship between PIEbg and walking within the Portland region to investigate whether one walkability measure is suitable for all urban contexts within one region. Or said differently, we are testing whether there is a different behavior response to PIE in urban vs. suburban areas. Finally, we introduce a more extensive analysis that includes different cities in the US. Our aim here is to test the inter-regional transferability by examining how these measures perform in different metropolitan areas of the paring PIEbg and PIE0To compare the new measures of the walkability PIEbg constructed at the block group scale with nationally available data to the original measure PIE0 at a finer scale with region-specific data, we needed to compute PIEbg for the Portland region. The distributions of various data used to construct PIEbg in Portland are shown in REF _Ref511734391 \h Figure 3. ULI, shown in orange, has a highly positive skew, meaning that there are very few block groups with scores above 2 and most of the block groups have low ULI score with a score of 1. The people density measure, shown in green, is less skewed, but also shows a large concentration of low values. This means that high values of both of these measures are scarce in the total composition of the region. Transit access, shown in pink, is also skewed to be high in very few places. The total road network density shown in purple is the only measure that it distributed more evenly across all of the values. Overall, this distribution makes sense for a region such as Portland, where the total land area is dominated by suburban landscapes and large park reserves with only a few highly urbanized nodes. This supports the notion that the landscape of North American cities may not have enough locations with sufficient densities to support non-automobile travel. Figure 3 Distribution of the scaled attributes of the built environment for PIEbg in Portland, ORUsing these data inputs, PIEbg was computed using the process outlined in the previous section. The resulting coefficients from the univariate binary logit models of walk mode choice for each built environment measure are shown in REF _Ref513715450 \h Table 2. The estimated coefficients range between 0.5-0.7 with the total road network density showing the highest coefficient value. These coefficients are used to weight each individual measure in the construction of PIE, after scaling to give PIE a range of values between 20 and 100. Table SEQ Table \* ARABIC 2 PIEbg estimation results: binary logit models of walk mode share Built Environment variable (zn)Coefficient (β)p-valueModel pseudo-R2Model 10.03People (employment and residential) per acre [1-5]0.520.00Constant-3.010.00Model 20.02ULI [1-5]0.540.00Constant-2.940.00Model 30.05Road network density [1-5]0.690.00Constant-4.120.00Model 40.03Transit accessibility [1-5]0.500.00Constant-3.100.00Data used for all modelsTrips (n)41,316Walk3,946Not Walk37,370The resulting weights for PIEbg are shown in REF _Ref513710909 \h Table 3 below and compared to the commensurate measures from PIE0. Here, we can see that the person density has a similar weight of 4.6; however, the other variables were apparently more sensitive to the change in scale and resulted in higher weights in PIEbg than in PIE0. Table SEQ Table \* ARABIC 3 Estimated weights for each built environment measure for PIE0 and PIEbgComponentRange of valuesWeights for PIE0Weights for PIEbgPeople per acre1 to 54.64.6ULI1 to 53.14.8Road network density1 to 5-6.1Block size1 to 53.1-Transit access1 to 53.54.7Sidewalk completeness1 to 52.8-Bicycle access 1 to 52.8- REF _Ref511734950 \h Figure 4 shows the histogram of the resulting values of PIEbg for each block group computed using these data and weights across the Portland region. The most frequent scores occur in a range between 26 and 38. We see a limited number of locations that have high scores of PIEbg, which is characteristic of many North American cities. Figure 4 Frequency distribution of PIEbg scores by block groupFigure 5 Land percentage by PIEbg score REF _Ref511735095 \h Figure 5 shows this distribution spatially. The highest scores of PIEbg (shown in red) are all concentrated in the Central City of Portland. In contrast, the vast majority of the region has scores that are under 30 (dark green). When analyzing the large patterns of the metro area, around the 90% of the land area has slightly less than half of the total population, a large share of this area is mainly rural/exurban. Furthermore, the areas with the highest walkability have scores that ranges between 40 and 100 represent less than the 1% of the area and only 5% of the population (~6.5% of the households sampled) and generates almost 14% of the total trips of the region. Visually from REF _Ref513458620 \h Figure 1 and REF _Ref511735095 \h Figure 5, we can see that despite the differences in the data sources, scale, and weighting scheme, the two walkability measures track similarly across the region. We will investigate these differences statistically and in models of walking below. Figure 6 PIEbg distribution in PortlandThe two versions of PIE, PIE0, and PIEbg are compared for each block group in the study area. PIE0 is calculated at the pedestrian analysis zone, was averaged over the block group and plotted against the value of PIEbg. Both indices were calculated using dependent variables between 1 and 5 and were weighted to have values between 20 and 100. REF _Ref513461870 \h Figure 6 shows the comparison of the two constructs in Portland, OR. As mentioned earlier, PIEbg scores are generally lower in value than PIE0 scores. This is due to differences in our approaches to scaling the originally built environment data (discrete values with natural breaks for PIE0; continuous rescaled data for PIEbg). However, both measures are coherent as the differences only reflect a change in the scale of the score, but both rise together.An interesting observation of this graph the breakpoint in the PIEbg score of 40 or PIE0 of 70. This is what REF _Ref511735095 \h Figure 34 would be the central city of Portland plus some outer neighborhoods or cities commercial areas. The notion is that in more walking areas the difference between both PIE become smaller, almost reaching the same value in the highest values. This would mean, that in higher values the continuous rescaling of PIEbg is more similar to the natural breaks approach of PIE0. Consequently, this could be interpreted as that the higher values of the characterization of the built environment are more clustered, meaning that the measures used by both versions of PIE become similar. In sum, PIEbg is coincident with an aggregated PIE0 with some caveats. PIEbg is less sensitive to changes in the built environment for suburban contexts and more sensitive in more urban areas than PIE0. These differences are likely due to the reliance on natural breaks in PIE0 and continuous breaks in PIEbg than the use of different data sources or the change in scale. The next section will explore how PIEbg as a measure of walkability is associated with walk mode share at the block group level across the region. Figure 7 PIE0 versus PIEbgIntraregional measure transferability We now turn our attention to examining the question of how walk mode share varies within a region and the relationship with our measure of walkability, PIEbg. Here we question if the relationship between walking and the built environment changes over the spectrum of urban spaces and if one walkability measure (or one model of walking behavior estimated from a pooled sample from the region) is adequate. Or contrarily, do we need to segment models (or walkability measures) by the urban regime in order to better reflect this varying relationship over the region. To do this, we used the weighted expansion of Oregon Household Travel Survey, which inflates the sample of trips to represent the total population of trips, to compute the total amount of trips and walk trips in each block group (walk mode share). REF _Ref511839778 \h Figure 7 shows the resulting plot of the walk modal share by deciles of PIEbg. The graph shows two trends (or three if we include the first rise from decile 1 to 2). For the bottom, 60% of PIEbg areas have a walking share that varies between 6-8%. The second tendency in the graph indicates an increase of walking share with the PIEbg scores at the seventh decile or greater. These two main differences in the slopes suggest that they might be two different behavioral responses to the built environment. Figure 8 PIEbg aggregated by deciles vs walking shareThese two regimes could be identified as the separation between a suburban environment with low people density and an urban environment with a higher people density. This suggests that the relationships between walking and the built environment should not be represented by just one formal functional and that this function may not be linear. More detailed information about the walking behavior and more data points across the spectrum of PIEbg are needed before we can discern and quantify relationship. In the next section, we are going to test the different regime hypothesis by contrasting two separated models versus a single one. These results suggest that it may not be appropriate to model the relationship between walk mode share and the built environment as a simple linear relationship or estimating a single model parameter for this relationship across all locations. Our findings suggest that there may be a threshold that separates two different regimes of walking behavior relative to the built environment. To examine this more closely, we test the results of a) a singular model with one parameter estimate for the built environment across all urban environments, with b) a model estimated at two intervals, with one parameter for areas that are more urban, another representing suburban environments. The model is a univariate binary logit model (the same of section REF _Ref511842358 \r \h 3.1) that predicts the probability of walking as a function of a single independent variable, an index of the built environment, which is represented as PIEbg. The model estimation results are shown in REF _Ref516029136 \h Table 4 REF _Ref511212849 \h . Model 1 and 2 are separate models for urban and suburban areas, respectively, where urban areas are defined as those areas with a PIEbg score of 40 or greater and suburban areas are those with a score below 40. Model 3 is a pooled model across all values of PIEbg. We used a PIEbg score of 40 because this was the approximate breakpoint in the relation between PIE0 and PIEbg in REF _Ref513461870 \h Figure 6 and the natural break in the distribution of PIE shown in REF _Ref513458620 \h Figure 1. However, one potential topic for future research is determining the exact breakpoint(s) of PIE in the defining these different regimes. For the purposes of this research, we will use 40 as the breakpoint and test the relationship between two regimes only due to sample size limitations and the maximum values of PIE in Portland, we could not test more with statistical accuracy. The three models had very low explanatory power, which is expected from behavioral models in general and from a univariate model such as this. However, we are only estimating these models for comparison. From simple inspection of the results, it is noticeable that the coefficients are different between the three models. We tested the validity of separating the models using the likelihood ratios test. The test rejects the pooled model as valid (p<0.000), meaning that the correct specification should be two separated models. Therefore, this evidence supports our hypothesis that the built environment does not have the same effect on the choice to walk in a city. However, this difference may be due to a spatial sorting of people according to the differences in the attitudes and inherent preferences of the people in those respective areas. However, we are lacking information about those aspects and cannot test them in this study. A visualization of these model results is shown in REF _Ref511212929 \h Figure 9. This figure shows the predicted walking mode share plotted against the PIEbg score. The graph shows a different prediction for the probability of walking for the different set of models. From the two separated models (urban and suburban), we can see that the effect of the built environment is larger in the suburban context and less steep in the urban environment, but still considerable. Is important to notice that almost all Portland has a low PIE score, as only 0.6% of the block groups have over 60 in the score, and 0.04% of the total land area (the very central city). Remember the area in the analysis is the Metropolitan Statistical Area that is based in counties and includes plenty of rural lands. Therefore, the larger portion of Portland as shown in REF _Ref521939572 \h Figure 4 is around a PIEbg score of 30 that is consistent with the 8% overall walking share in the region. The pooled model in the vicinity of the score 30 shows a similar walking probability, however, there is an overestimation of probability trips in the rural/suburban area and underestimation in the transition between urban and suburban, and in the more urban areas.Despite the limitations, the exercise done here lends empirical evidence to support that concern that the relationship between walking and the built environment may be different (non-linear or linear in different regimes with different coefficients) across different environments within the same region. Thus, one model estimate may not be sufficient to capture this relationship and thus not transferable to all areas within a region. Table SEQ Table \* ARABIC 4 Pooled vs. Urban/Suburban Walk Mode Share ModelsContext variable (xn)Coefficient (β)p-valueModel pseudo-R2Model 1 Urban0.03PIEbg (40-100)0.020.00Constant-2.830.00Log-Likelihood-2993Chi-squared166Model 2 Suburban0.03PIEbg (20-40)0.100.00Constant-5.730.00Log-Likelihood-9364Chi-squared521Model 3 All urban types0.02PIEbg (20-100)0.0360.00Constant-3.560.00Log-Likelihood-12489Chi-squared1060Log-Likelihood ratio test segmentation of the model0.00Data used for all modelsTrips (n) – all urban types41,316Walk3,946Not Walk37,370Trips (n) – urban6,350Walk1,199Not Walk5,151Trips (n) – suburban34,966Walk2,747Not Walk32,219Figure 9 Predicted walk probabilitiesInterregional measure transferabilityIn this section, we test the differences in various estimations of PIEbg for different cities. As the objective was to assess the transferability of our walkability measure, we needed to associate it with travel behavior data from each city in our study. To test relationships within and across regions, we assembled four US travel surveys: those covering the state of California (Kunzman and Daigler 2013) as well as the metropolitan areas of Minneapolis–St. Paul (Cambridge Systematics 2014), Portland (Oregon Modeling Steering Committee 2012), and Seattle (Cambridge Systematics 2006). The California travel survey allowed for the inclusion of the Metropolitan Statistical Areas of Los Angeles, San Diego, and San Francisco-Oakland (without San Jose), bringing the total of US regions to six.Following the general procedures outlined in Section 4.0, we estimate PIEbg for five additional cities (Los Angeles, Minneapolis, San Diego, San Francisco, and Seattle) using household travel surveys and built environment data from each region. Then we compare the coefficient estimates across these regions to test if they are comparable and thus, indicate that PIE estimated from one region could indeed be applicable in another. REF _Ref522024223 \h Table 5 shows details from each metropolitan area travel survey and built environment attributes statistics. Table SEQ Table \* ARABIC 5 Trip information and built environment measures in each metropolitan areaLOS ANGELESMINNEAPOLIS – ST. PAULPORTLANDSAN DIEGOSAN FRANCISCOSEATTLEYear of survey2010-12201020112010-122010-122006Total trips32.8M11.1M6.5M7.8M13.4M12.6MPopulation12.8M3.3M2.2M3.1M4.3M3.4MWalking mode share (%)11.0%6.2%8.6%8.5%15.9%7.9%No. of block groups824823141555179529032483Population density range [ppl/acre][0,300][0,180][0,100][0,93][0,319][0,173]Population density mean [ppl/acre] (std. deviation)20 (16)8 (9)8 (7)14 (11)21 (24)9 (10)Employment density range [ppl/acre][0,579][0,354][0,251][0,330][0,900][0,1120]Employment density mean [ppl/acre] (std. deviation)14 (5)16 (4)14 (4)12 (4)36 (8)29 (5)ULI range [ppl/acre][0,66][0,127][0,56][0,73][0,795][0,115]ULI mean [ppl/acre] (std. deviation)1 (3)1 (4)1 (3)1 (3)3 (24)1 (4)Road network density range [miles/sq. miles][0,69][0,55][0,54][0,62][0,65][0,86]Road network density mean [miles/sq. miles] (std. deviation)21 (7)17 (8)18 (10)19 (8)22 (9)17 (9)Transit accessibility range [aggregated frequency][0,4401][0,5038][0,2351][0,526][0,2028][0,1839]Transit accessibility mean [aggregated frequency] (std. deviation)71 (140)128 (230)151 (211)35 (45)86 (144)49 (97)Some basic analysis shows that the city that has the most walking is San Francisco (15.9%). The larger city in population is Los Angeles with 12.8 million people. The population density ranges are higher in Los Angeles and San Francisco while the employment density is higher in Seattle and San Francisco. Following the assembly and standardization of the datasets, we made a few exceptions to the methodology. We did not scale each of the built environment data to have values between 1 and 5. Instead, we scaled the data to have an average of 0 and a standard deviation of 1. This allowed us to have standardized coefficients that let us directly compare the magnitude of the coefficients between cities. In doing this, the estimated standardized coefficients are only meant to be used for comparison across regions and not for application. Finally, we did not sum the employment and population density variables as we wanted to test the independent effect of each measure, however, we re-combine them as people density later in this section in an application to Los Angeles. Figure 10 Standardized coefficients for different built environment measuresThe results of these estimations for the different cities are shown in REF _Ref511988862 \h \* MERGEFORMAT Figure 10. All the estimates were statistically significant (p < 0.001). Of the various built environment measures tested, the standardized coefficients for population density appear to have the most consistent values across the cities in the study, with values ranging from 0.37 in Portland to 0.51 in San Francisco. For the Los Angeles and San Diego regions, population density was the most important built environment characteristic in explaining walk mode choice. Given its stability and its relatively high standardized coefficient, population density may be the most important built environment measure associated with the odds of walking across all of the cities, which is consistent with the literature.In contrast, road network density has much more variability based upon the statistics in REF _Ref522024223 \h Table 5 that is reflected in the values of the standardized coefficients in REF _Ref511988862 \h \* MERGEFORMAT Figure 10, which range from a low of 0.20 in Los Angeles to a high of 0.72 in Minneapolis. This was the most important measure in explaining walking for several cities: Minneapolis, Portland, San Francisco, and Seattle. This range of values is perplexing but may be a function of the computation of the attribute itself. Road network density includes all facility types, including freeways, and for that reason, it may have confounding results. For future work, we suggest using intersection density as the measure of pedestrian connectivity. The standardized coefficients for employment density and retail and entertainment employment density are also variable across the cities. The values for employment density range from a low of 0.02 in Los Angeles to a high of 0.14 in San Francisco. Overall, the coefficient values are low compared to the other built environment attributes, and probably the spatial distribution of employment, something that we have not covered here, might explain its influence on walking behavior. It may also be that employment density is an imperfect proxy for access to destinations and a local accessibility measure or a mixed use measure may be a more stable measure across locations. Narrowing the scope to retail, service, and entertainment establishments, as measured by ULI did not result in a decrease in the variation across places. The standardized coefficients varied from a low of 0.02 for San Diego to a high of 0.23 for Minneapolis. The suggestion to improve the measure of access to destinations holds. These results raise a lot of questions about the cause of these results, including how these variables are distributed across the different regions, where are the locations where walking trips are observed in the sample, and perhaps variations in the trip purposes. But the results suggest that a walkability measure estimated in one location may only be cautiously applied in other regions, if at all. To explore this issue further, we want to test the predictive ability of PIEbg estimated in one region and applied in a model of walk mode share in another. In order to make the broadest comparison, we chose to utilize PIEbg estimated from Portland in a univariate model of walk mode share in Los Angeles - the largest city in our study. We make additional modifications to the process of estimating PIEbg using Portland data. This time, the data for each built environment measure (xn) used in the calibration are not scaled at all because our aim is not to compare the coefficients but rather apply PIE in a model of walking. Because Portland and Los Angeles have different ranges and distributions of these variables, as shown in REF _Ref522024223 \h \* MERGEFORMAT Table 5, we needed to ensure that the data were unscaled in the application across different locations. Second, population and employment density are re-combined into people per acre. The results are shown in REF _Ref516488802 \h Table 6.Table SEQ Table \* ARABIC 6 Unscaled PIEbg coefficients estimated from Portland dataBuilt Environment variable (xn)Coefficient (β)p-valueModel pseudo-R2Model 10.03People per acre 0.0070.00Constant-2.4860.00Model 20.02ULI0.0390.00Constant-2.4020.00Model 30.05Road network density0.0520.00Constant-3.4630.00Model 40.04Transit access0.0010.00Constant-2.5980.00Data used for all modelsTrips (n)41,316Walk trips3,946Not walk trips37,370Using the coefficients for each of these built environments as weights, we computed PIEbg for each block group in the Los Angeles region by using the following formula:PIEbg=0.007*people per acre+0.039*ULI+0.052*road network density+0.001*transit accessibility REF _Ref512001404 \h Figure 11 shows both the distribution in the city of unscaled PIEbg. Both areas have a clear central area that it is more walkable. In both cases, this area is small compared to the region. Even though the urban structure of Los Angeles seems very different from that of Portland, it seems from a visual inspection of the map that the PIEbg specification of Portland applied to Los Angeles make some intuitive sense. However, as seen in REF _Ref522024220 \h REF _Ref522024223 \h Table 5 many of the built environment attributes are much larger in Los Angeles than Portland. For example, the mean population density in each block group in Los Angeles is 20 people per acre, while in Portland is only 8. This means that the model estimation in Portland can omit and effect of what larger values of the built environment attributes may cause, causing that the PIEbg scores in Los Angeles only reach higher values in very few places. Additionally, as shown also in REF _Ref522024223 \h Table 5 is that Los Angeles have a lower average transit service than Portland and also as REF _Ref511988862 \h Figure 10 shows the coefficient for transit is smallest in Los Angeles than in the other cities, for that reason, the effect of transit might also be creating some noise in the scores. Figure 11 Unscaled PIEbg distribution in Los Angeles and PortlandDiscussion and conclusionsThe analysis of aggregate and disaggregate pedestrian trips presented here examined the potential transferability of relationships between walking and the built environment within and across different regions. To summarize some of these findings: The probability of walking has a clear relationship with PIEbgAn examination of PIEbg and walk mode share suggests that there may be different relationships between walking behaviors and the built environment across the spectrum of environments within a region. Our intraregional tests for transferability of models and measures revealed two regimes where these relationships may differ: urban versus suburban contexts. However, more research is needed to better understand these variations within a region and the conditions defining these regimes. Population density has the strongest and most consistent relationship with walk mode share across the six US metropolitan regions tested in our study.Interregional tests of transferability of PIEbg revealed similar walk mode share results in Los Angeles and Portland. This provides initial support that the PIEbg construct may be transferrable between metropolitan regions. More detailed discussion of these conclusions is below. Transferability of models and measuresThe findings were mixed in terms of the degree to which PIE could be applicable to areas beyond its estimation region. In our examination of intraregional transferability, there appeared to be different relationships between walking and the built environment in different parts of the region, divided simply into lower density (suburban) and higher density (urban) areas. This suggests that perhaps different constructs may be needed to represent different walking regimes within a region. Yet, the characteristics (thresholds) defining each walking regime are not the same in Portland and Los Angeles. Further, our findings also suggest that the relationship between PIEbg and walk mode shares may not be linear. This notion suggests the existence of different regimes of response from people to the built environment that should be better represented by our predictions tools. One reason that defining these regimes and the functional form of the relationship between walking and the built environment was difficult is because of the relatively smaller sample sizes in the urban areas resulted in noise in the estimates. Thus, larger sample sizes from higher density areas are needed to better understand these relationships. Further, there are potential correlations with the socio-economic characteristics of travelers and the characteristics of the locations of travel, which point to the need to control for this in walkability measures (discussed in the next subsection). Based upon our analysis, travel behavior surveys from the US cities we studied tend to have greater proportion of sample from lower density areas associated with American suburban form and smaller sample sizes in areas with higher walkability and overall walking activity, which represent a smaller proportion of both land mass and population of those cities. This may not be the case in international contexts, where more of the urbanized regions are of a higher density than the US. Creating pooled samples with comparable and consistent data from cities around the world is becoming a more realistic endeavor with standardization of travel surveys and the availability of built environment data. This would be an asset to understanding the relationships between travel behavior and the built environment, broadly, and identifying the various regimes of walking, specifically. From our analysis, the distribution of walkable places and their relative locations across the region appear to be important. For example, Los Angeles region has more variation in its urban structure and is more polycentric than Portland. This raises the question of the role of the larger urban spatial structure in supporting walking activity. By urban structure we mean the number and distribution of centers or sub-centers across a region, the density gradients, the clustering or contiguity of walkable places, and the spatial extent of the urbanized area (total land area). These are characteristics that are not captured well in local and regional accessibility measures and perhaps represent a meso-scale description of the built environment. Figure 12 The role of walkability gradients REF _Ref522299462 \h Figure 12 shows an example of walkability across a fictitious cross section of an urban region, from the city center outward. While all areas across the red dashed line have the same objective measure of walkability, there is significant variation in the walkability gradients as well the total space that has that level of walkability or more. One question to investigate is would those locations proximate to the city center, where walkability is relatively high and sustained across the space, have the same walk mode share or absolute levels of walking as those farther from the center with the same objective measure of walkability but different gradients and total area of walkable space. Or put another way, would we expect pockets of walkability to perform similarly as places where walkability is sustained over a larger area? At the moment, our methods of analysis do not consider this and this may be an area of fruitful future work that may help to explain the differences we see between regions. Representing the built environmentThe ways that we have been construction PIE have limited its transferability and have problematic issues that limit its usefulness. This new construct, PIEbg has some important differences to our original PIE0, in part because of our need to have comparable measure across cities. When PIE0 was developed in 2011, the data available in Portland were unique and at a scale not widely available (PAZs - 80m x 80m grid cells). This presented the opportunity to examine walking at a scale more consistent with the behavior of interest and with zones of uniform size. However, these data were originally continuous and were then reduced to categorical on a scale of 1 to 5 based upon natural breaks in their distribution over the Portland region in the base year. This resulted in the loss of variation as well as reducing its applicability to other regions or even future Portland conditions. In this research, we attempted to address this limitation in the creation of PIEbg. However, this new built environment construct also has limitations. In using the SLD data at the block group level, we then had comparable data across regions but the fine-grained and consistently-sized spatial resolution of the PAZ grid cells was lost. Further, we realize that in estimating the various coefficients (or weights) for PIE with built environment variables while not controlling for socio-economic characteristics may bias those estimates. Also, there may be econometric issues with our constructs in that we are using the coefficients from a walk mode choice model as weights in our final construction of PIE and then in turn, using that very construct in other models of mode choice and trip distribution in MoPeD. These limitations are problematic.Now at the end of this research process, we have some important recommendations about future attempts to create a walkability measure using various built environment characteristics. The literature has revealed a myriad of associations between the built environment and walking behavior. Many of these characteristics are highly correlated with one another and then, are problematic when including them as separate independent variables in a model. Thus, this is one of the rationales for creating an aggregate index or measure that reflects the walking environment. Given the variations in the availability of data, the difficulties in interpretation, and the correlations between these phenomena, we recommend parsimony in the selection of built environment characteristics to represent in such a measure. We recommend using continuous data to represent the built environment rather than reducing the information to a categorical measure. The reasons for this are maintaining the variation in the measure as well as the difficulty in representing environments that exceed the conditions present in the estimation year. Of those measures tested in our study, population density consistently explained more of the variation in walk mode share, followed by measures representing pedestrian connectivity. These two built environment characteristics are important to include and not as highly correlated with one another as we originally believed. Several measures of connectivity were tested over the course of this project: average block size, road network density, and average streets per node. There are various concerns and limitations with each. Employment density was highly variable, including the ULI measure, that attempts to capture access to local destinations. Contemporaneous with this study, new guidebook on measuring non-motorized connectivity was completed (FHWA, 2018) with the input and participation of this study's author, and offers suggestions that may resolve these issues. Based upon this guidebook, we recommend including a pedestrian connectivity measure that captures access to destinations into future walkability indices. This compensates for the low performance of employment density and incorporates network connectivity in one measure. Transit accessibility and service is a difficult measure to incorporate consistently. From a theoretical perspective, the links between transit and access and egress mode of travel is logical. However, in this work, we are only considering trips where walking is the single mode of transport. Thus, the rationale for including transit accessibility is less clear. One could argue that in areas with good transit, more trips would be made on foot by persons traveling to the area via transit. However, this may be a proxy for (or confounded with) the built environment conditions in these locations. In development of a walkability measure, it may be best to omit transit accessibility from the direct representation of walkability and instead include as an additional measure for better interpretation of results. When data permit, using uniform grid cells as the spatial unit of analysis is preferable to block groups, which suffer from a size differential that is correlated with population density. Others have examined the effects of the size of these grid cells on model performance and have found tradeoffs between size, model run times, and accuracy (Zhang et al. 2018). However, the loss of accuracy from doubling the grid size from 80m grids to 160m grids is nominal compared to the advantages for model runs and data availability. In terms of the econometric issues with PIE, future representations of walkability using PIE index will be computed as the sum of the portion of utility function of a walking mode choice model that deals with the built environment. By minimizing the number of built environment variables included in this estimation, correlation between built environment variables does not exceed the threshold for concern. Thus, they can be included directly without the use of an index. Here socio-economic and other important characteristics will be controlled for to include inflation of the coefficients for the built environment variables. This development and presentation of these results are outside the scope of this report but are based upon lessons learned during the process of this research. Future workSeveral limitations inhibit stronger conclusions from this analysis. First, there are a limited number of cities included in this exploratory analysis. A more comprehensive study needs a larger number of cities with different activity density distributions (particularly on the high end) and varying urban spatial structure, including those from other countries. This expansion of the work could help to define regimes, and how these measures interact at different thresholds. Additional analysis may also include more advanced spatial analysis techniques. In addition to considering the recommendations for the construct of PIE and other walkability measures, future work may want to include more course measures of the built environment to better capture urban structure, such as amount and distribution of density, total land area, density gradients and continuity, and mixed-use, as they may play a role in understanding traveler response to the built environment, including pedestrian behaviors. The consideration of non-linearities in the built environment-walking relationship is one contribution of this work that deserves more exploration. Although more analysis is needed to understand the specific functional form that these non-linearities take, our analysis provides evidence that continuing to assume linear relationship across the urban spectrum is questionable.References ADDIN ZOTERO_BIBL {"custom":[]} CSL_BIBLIOGRAPHY Badoe, Daniel A., and Eric J. Miller. 1995. “Analysis of Temporal Transferability of Disaggregate Work Trip Mode Choice Models.” Transportation Research Record, no. 1493. , Moshe. 1981. “Issues in Transferring and Updating Travel Behavior Models.” In New Horizons in Travel-Behaviour Research, edited by Peter R. Stopher, A.H. Meyburg, and W. Br?g, 665–86. Lexington: Lexington Books. , Moshe, and Denis Bolduc. 1987. “Approaches to Model Transferability and Updating: The Combined Transfer Estimator.” Transportation Research Record, no. 1139. , David, Linda W Pickle, and Jennifer Dill. 2010. “Associations between Street Connectivity and Active Transportation.” International Journal of Health Geographics 9 (1): 20. , Robert, and Kara Kockelman. 1997. “Travel Demand and the 3Ds: Density, Diversity, and Design.” Transportation Research Part D: Transport and Environment 2 (3): 199–219. (97)00009-6.Clifton, Kelly J. 2017. “Getting From Here to There: Comment on ‘Does Compact Development Make People Drive Less?’” Journal of the American Planning Association 83 (2): 148–51. , Kelly J., and Christopher Muhs. 2012. “Capturing and Representing Multimodal Trips in Travel Surveys.” Transportation Research Record: Journal of the Transportation Research Board 2285 (October): 74–83. , Kelly, Patrick Singleton, Christopher Muhs, and Robert Schneider. 2015. “Development of a Pedestrian Demand Estimation Tool.” Civil and Environmental Engineering Faculty Publications and Presentations, September. , Kelly, Patrick Singleton, Christopher Muhs, Robert Schneider, and Peter Lagerwey. 2013. “Improving the Representation of the Pedestrian Environment in Travel Demand Models, Phase I.” Civil and Environmental Engineering Faculty Publications and Presentations, September. , Reid, and Robert Cervero. 2010. “Travel and the Built Environment.” Journal of the American Planning Association 76 (3): 265–94. .———. 2017. “‘Does Compact Development Make People Drive Less?’ The Answer Is Yes.” Journal of the American Planning Association 83 (1): 19–25. , James, Andrew Daly, Stephane Hess, and Eric Miller. 2014. “Temporal Transferability of Models of Mode-Destination Choice for the Greater Toronto and Hamilton Area.” Journal of Transport and Land Use 7 (2): 41. , Lawrence D. 2000. “Land Use and Transportation Interaction: Implications on Public Health and Quality of Life.” Journal of Planning Education and Research 20 (1): 6–22. Highway Administration (FHWA). 2018. "FHWA Guidebook for Estimating Multimodal Network Connectivity." US Department of Transportation, Washington, DC. Available online: , Michael, and Marlon Boarnet. 2001. “Built Environment as Determinant of Walking Behavior: Analyzing Nonwork Pedestrian Travel in Portland, Oregon.” Transportation Research Record: Journal of the Transportation Research Board 1780 (January): 33–41. , Susan. 2017. “Thoughts on the Meaning of Mark Stevens’s Meta-Analysis.” Journal of the American Planning Association 83 (1): 26–28. , Susan L., Marlon G. Boarnet, Reid Ewing, and Richard E. Killingsworth. 2002. “How the Built Environment Affects Physical Activity: Views from Urban Planning.” American Journal of Preventive Medicine 23 (2): 64–73. (02)00475-0.Huang, Jingnan, X. X. Lu, and Jefferey M. Sellers. 2007. “A Global Comparative Analysis of Urban Form: Applying Spatial Metrics and Remote Sensing.” Landscape and Urban Planning 82 (4): 184–97. , Nina. 2007. “Evaluation of Transfer Methods for Spatial Travel Demand Models.” Transportation Research Part A: Policy and Practice 41 (5): 411–27. , J. Richard, Jerry Walters, Mark Bradley, and Kara M. Kockelman. 2014. “Estimating Bicycling and Walking for Planning and Project Development: A Guidebook.” NCHRP Report, no. 770. , Nico, Jean Stockard, Bethany Steiner, and Amanda West. 2013. “Trips to Strips: Walking and Site Design in Suburban Multifamily Housing.” Journal of Urban Design 18 (2): 281–303. , R. 1981. “A Comment on Interspatial, Intraspatial, and Temporal Transferability.” In New Horizons in Travel-Behaviour Research, edited by Peter R. Stopher, A.H. Meyburg, and W. Br?g, 628–32. Lexington: Lexington Books.Newman, Peter. 2014. “Density, the Sustainability Multiplier: Some Myths and Truths with Application to Perth, Australia.” Sustainability 6 (9): 6467–87. , Roosbeh, and Sivaramakrishnan Srinivasan. 2012. “Empirical Analysis of Spatial Transferability of Tour-Generation Models.” Transportation Research Record: Journal of the Transportation Research Board 2302 (December): 14–22. , Farideddin, Sybil Derrible, and Farukh Ijaz. 2014. “Development and Application of the Pedestrian Environment Index (PEI).” Journal of Transport Geography 39 (July): 73–84. , Kevin, and Alexander Bell. 2014. “The Smart Location Database: A Nationwide Data Resource Characterizing the Built Environment and Destination Accessibility at the Neighborhood Scale.” Cityscape 16 (2): 145.Rossi, Thomas F., and Chandra R. Bhat. 2014. “Guide for Travel Model Transfer,” October. , Brian E., and Susan L. Handy. 2008. “Built Environment Correlates of Walking: A Review.” Medicine and Science in Sports and Exercise 40 (7 Suppl): S550–66. , James F., Robert B. Cervero, William Ascher, Karla A. Henderson, M. Katherine Kraft, and Jacqueline Kerr. 2006. “An Ecological Approach to Creating Active Living Communities.” Annual Review of Public Health 27 (1): 297–322. , Robert, Robert Patten, and Jennifer Toole. 2005. “Case Study Analysis of Pedestrian and Bicycle Data Collection in U.S. Communities.” Transportation Research Record: Journal of the Transportation Research Board 1939 (January): 77–90. , Patrick A., Robert J. Schneider, Christopher Muhs, and Kelly J. Clifton. 2014. “The Pedestrian Index of the Environment: Representing the Walking Environment in Planning Applications.” In . , Patrick A., Joseph C. Totten, Jaime P. Orrego-O?ate, Robert J. Schneider, and Kelly J. Clifton. 2018. “Making Strides: State-of-the-Practice of Pedestrian Forecasting in Regional Travel Models.” Transportation Research Record: Journal of the Transportation Research Board.Stevens, Mark R. 2017. “Does Compact Development Make People Drive Less?” Journal of the American Planning Association 83 (1): 7–18. , Orion T., Anne Vernez Moudon, Brian E. Saelens, Chanam Lee, Bumjoon Kang, and Mark P. Doescher. 2016. “Comparing Associations Between the Built Environment and Walking in Rural Small Towns and a Large Metropolitan Area.” Environment and Behavior 48 (1): 13–36. , Emily, and Julia Koschinsky. 2014. “Compact, Walkable, Diverse Neighborhoods:Assessing Effects on Residents.” Housing Policy Debate 24 (4): 717–50. EPA, OA. 2014. “Smart Location Database Technical Documentation and User Guide.” Data and Tools. US EPA. February 27, 2014. , Qin; Clifton, Kelly J.; and Moeckel, Rolf, 2018. "Investigation of an Appropriate Spatial Resolution for Large-scaled Pedestrian Travel Demand Models." presented at Mobil.TUM 2018 Urban Mobility: Shaping the Future Together, Munich, Germany, June 13-14, 2018. ................
................

In order to avoid copyright disputes, this page is only a partial summary.

Google Online Preview   Download