265 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 nd 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 Travel and the Built Environment A Meta-AnalysisReid Ewing and Robert CerveroSome of today's most vexing problems, including sprawl, congestion, oil dependence, and climate change, are prompting states and localities to turn to land planning and urban design to rein in automobile use. Many have concluded that roads cannot be built fast enough to keep up with rising travel demand induced by the road building itself and the sprawl it spawns. The purpose of this meta-analysis is to summarize empirical results on associations between the built environment and travel, especially nonwork 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 nd 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 condence intervals. They provide a base, and as more builtenvironment/travel studies appear in the planning literature, these elasticities should be updated and rened. Keywords :vehicle miles traveled (VMT), walking, transit, built environment, effect sizes Research support :U.S. Environmental Protection Agency. About the authors : Reid Ewing (firstname.lastname@example.org) is professor of city and metropolitan planning at the University of Utah, associate editor of the Journal of the American Planning Association, columnistfor Planning magazine, and fellow of the Urban Land Institute. Robert Cervero (email@example.com) is professor of city and regional planning at the University of California, Berkeley, director of the University of California Transportation Center, and interim director of the Institute of Urban and Regional Development. Journal of the American Planning Association, Vol. 76, No. 3, Summer 2010 DOI 10.1080/01944361003766766 American Planning Association, Chicago, IL. RJPA_A_477198.qxd 6/11/10 3:25 PM Page 265
travel. A number of studies, including Boarnet and Crane (2001), Cao, Mokhtarian, and Handy (2009b), Cervero (2002a), Cervero and Kockelman (1997), Crane (1996), Kockelman (1997), and Zhang (2004), provide economic and behavioral explanations of why built environments might be expected to inuence travel choices. We do not repeat them here, focusing instead on measuring the magnitude of such relationships. We aim to quantify effect sizes while also updating earlier work, including walking and transit use as outcome measures, and addressing the methodological issue of self-selection. Little work on this topic to date has generalized across studies or helped make sense of differing results. Without this, readers have glimpses of many trees rather than a panoramic view of this complex and rich forest of research. We authored one previous attempt, a literature review (Ewing & Cervero, 2001), in which we derived composite elasticities by informal inspection, an inherently imprecise process. The current meta-analysis, by contrast, is a more systematic way to combine information from many studies, arriving at weighted averages as bottom lines. There are now more than 200 built-environment/ travel studies, of which most were completed since our 2001 review.1Compared to earlier studies, the newer ones have estimated effects of more environmental variables simultaneously (expanding beyond density, diversity, design, and destinations, to include distance to transit), controlled for more confounding inuences (including traveler attitudes and residential self-selection), and used more sophisticated statistical methods. In response to the U.S. obesity epidemic, the public health literature has begun to link walking to dimensions of the built environment. The rst international studies have appeared using research designs similar to those of U.S. studies. This collective and enlarged body of research provides a substantial database for a meta-analysis. The transportation outcomes we studied in 2001, vehicle miles traveled (VMT) and vehicle trips (VT), are critically linked to trafc safety, air quality, energy consumption, climate change, and other social costs of automobile use. However, they are not the only outcomes of interest. Walking and transit use have implications for mobility, livability, social justice, and public health. The health benets of walking, in particular, are widely recognized (Badland & Schoeld, 2005; Cunningham & Michael, 2004; Frank, 2000; Frank & Engelke, 2001; Humpel, Owen, & Leslie, 2002; Kahn, Ramsey, Brownson, Heath, & Howze, 2002; Krahnstoever-Davison & Lawson, 2006; Lee & Moudon, 2004; McCormack, Giles-Corti, Lange, Smith, Martin, & Pikora, 2004; Owen, Humpel, Leslie, Bauman, & Sallis, 2004; Saelens & Handy, 2008; Transportation Research Board & Institute of Medicine Committee on Physical Activity, Health, Transportation, and Land Use, 2005; Trost, Owen, Bauman, Sallis, & Brown, 2002). Transit use is less obviously related to public health, but is still classied as active travel since it almost always requires a walk at one or both ends of the trip (Besser & Dannenberg, 2005; Edwards, 2008; Zheng, 2008). So, to VMT we add walking and transit use as outcomes of interest.2More than anything else, the possibility of selfselection bias has engendered doubt about the magnitude of travel benets associated with compact urban development patterns. According to a National Research Council report (Transportation Research Board & Institute of Medicine Committee on Physical Activity, Health, Transportation, and Land Use, 2005), "If researchers do not properly account for the choice of neighborhood, their empirical results will be biased in the sense that features of the built environment may appear to inuence activity more than they in fact do. (Indeed, this single potential source of statistical bias casts doubt on the majority of studies on the topic to date)" (pp. 134135). At least 38 studies using nine different research approaches have attempted to control for residential selfselection (Cao, Mokhtarian, & Handy, 2009a; Mokhtarian & Cao, 2008). Nearly all of them found "resounding" evidence of statistically signicant associations between the built environment and travel behavior, independent of self-selection inuences (Cao, Mokhtarian, et al. 2009a, p. 389). However, nearly all of them also found that residential self-selection attenuates the effects of the built environment on travel. Using travel diary data from the New York/New Jersey/Connecticut regional travel survey, Salon (2006) concluded that the built environment accounted for one half to two thirds of the difference in walking levels associated with changes in population density in most areas of New York City. Using travel diary data from the Austin travel survey, Zhou and Kockelman (2008) found that the built environment accounted for 58% to 90% of the total inuence of residential location on VMT, depending on model specications. Using travel diary data from northern California, Cao (2010) reported that, on average, neighborhood type accounted for 61% of the observed effect of the built environment on utilitarian walking frequency and 86% of the total effect on recreational walking frequency. Using data from a regional travel diary survey in Raleigh, NC, Cao, Xu, and Fan (2009) estimated that anywhere from 48% to 98%3of the difference in vehicle miles driven was due to direct environmental inuences, the balance being due to self-selection. Using data from the 2000 San 266Journal of the American Planning Association, Summer 2010, Vol. 76, No. 3 RJPA_A_477198.qxd 6/11/10 3:25 PM Page 266
Francisco Bay Area travel survey, Bhat and Eluru (2009) found that 87% of the VMT difference between households residing in conventional suburban and traditional urban neighborhoods is due to "true" built environment effects, while the remainder is due to residential selfselection. So, while the environment seems to play a more important role in travel behavior than do attitudes and residential preferences, both effects are present.The D Variables as Measures of the Built EnvironmentThe potential to moderate travel demand by changing the built environment is the most heavily researched subject in urban planning. In travel research, such inuences have often been named with words beginning with D. The original "three Ds," coined by Cervero and Kockelman (1997), are density, diversity, and design, followed later by destination accessibility and distance to transit (Ewing & Cervero, 2001; Ewing et al., 2009). Demand management, including parking supply and cost, is a sixth D, included in a few studies. While not part of the environment, demographics are the seventh D, controlled as confounding inuences in travel studies. Density is always measured as the variable of interest per unit of area. The area can be gross or net, and the variable of interest can be population, dwelling units, employment, building oor area, or something else. Population and employment are sometimes summed to compute an overall activity density per areal unit. Diversity measures pertain to the number of different land uses in a given area and the degree to which they are represented in land area, floor area, or employment. Entropy measures of diversity, wherein low values indicate single-use environments and higher values more varied land uses, are widely used in travel studies. Jobsto-housing or jobs-to-population ratios are less frequently used. Design includes street network characteristics within an area. Street networks vary from dense urban grids of highly interconnected, straight streets to sparse suburban networks of curving streets forming loops and lollipops. Measures include average block size, proportion of fourway intersections, and number of intersections per square mile. Design is also occasionally measured as sidewalk coverage (share of block faces with sidewalks); average building setbacks; average street widths; or numbers of pedestrian crossings, street trees, or other physical variables that differentiate pedestrian-oriented environments from auto-oriented ones. Destination accessibility measures ease of access to trip attractions. It may be regional or local (Handy, 1993). In some studies, regional accessibility is simply distance to the central business district. In others, it is the number of jobs or other attractions reachable within a given travel time, which tends to be highest at central locations and lowest at peripheral ones. The gravity model of trip attraction measures destination accessibility. Local accessibility is different, dened by Handy (1993) as distance from home to the closest store. Distance to transit is usually measured as an average of the shortest street routes from the residences or workplaces in an area to the nearest rail station or bus stop. Alternatively, it may be measured as transit route density,4distance between transit stops, or the number of stations per unit area. Note that these are rough categories, divided by ambiguous and unsettled boundaries that may change in the future. Some dimensions overlap (e.g., diversity and destination accessibility). We still nd it useful to use the D variables to organize the empirical literature and provide order-of-magnitude insights. LiteratureQualitative ReviewsThere are at least 12 surveys of the literature on the built environment and travel (Badoe & Miller, 2000; Cao, Mokhtarian, et al., 2009a; Cervero, 2003; Crane, 2000; Ewing & Cervero, 2001; Handy, 2004; Heath, Brownson, Kruger, Miles, Powell, Ramsey, & the Task Force on Community Preventive Services, 2006; McMillan, 2005, 2007; Pont, Ziviani, Wadley, Bennet, & Bennet, 2009; Saelens, Sallis, & Frank, 2003; Stead & Marshall, 2001). There are 13 other surveys of the literature on the built environment and physical activity, including walking and bicycling (Badland & Schoeld, 2005; Cunningham & Michael, 2004; Frank, 2000; Frank & Engelke, 2001; Humpel et al., 2002; Kahn et al., 2002; KrahnstoeverDavison & Lawson, 2006; Lee & Moudon, 2004; McCormack et al., 2004; Owen et al., 2004; Saelens & Handy, 2008; Transportation Research Board & Institute of Medicine Committee on Physical Activity, Health, Transportation, and Land Use, 2005; Trost et al., 2002). There is considerable overlap among these reviews, particularly where they share authorship. The literature is now so vast it has produced two reviews of the many reviews (Bauman & Bull, 2007; Gebel, Bauman, & Petticrew, 2007). From our earlier review (Ewing & Cervero, 2001), the most common travel outcomes modeled are trip frequency, Ewing and Cervero: Travel and the Built Environment: A Meta-Analysis267 RJPA_A_477198.qxd 6/11/10 3:25 PM Page 267
trip length, mode choice, and VMT (as a composite measure of travel demand). Hence, we can describe measured associations between D variables and these outcomes with more condence than we could for outcomes studied less often, like trip chaining in multipurpose tours or internal capture of trips within mixed use developments. Our earlier review (Ewing & Cervero, 2001) held that trip frequency is primarily a function of socioeconomic characteristics of travelers and secondarily a function of the built environment; trip length is primarily a function of the built environment and secondarily of socioeconomic characteristics; and mode choice depends on both, although probably more on socioeconomics. VMT and vehicle hours of travel (VHT) also depend on both. Trip lengths are generally shorter at locations that are more accessible, have higher densities, or feature mixed uses. This holds true both when comparing home-based trips from different residential neighborhoods and trips to nonhome destinations in different activity centers. Destination accessibility is the dominant environmental inuence on trip length. Transit use varies primarily with local densities and secondarily with the degree of land use mixing. Some of the density effect is, no doubt, due to better walking conditions, shorter distances to transit service, and less free parking. Walking varies as much with the degree of land use mixing as with local densities. The third D, design, has a more ambiguous relationship to travel behavior than the rst two. Any effect is likely to be a collective one involving multiple design features. It also may be an interactive effect with other D variables. This is the idea behind composite measures such as Portland, Oregon's urban design factor, which is a function of intersection density, residential density, and employment density.Our Earlier Quantitative SynthesisUsing 14 travel studies that included sociodemographic controls, we previously synthesized the literature on the elasticities of VMT and VT with respect to density, diversity, design, and destination accessibility (Ewing & Cervero, 2001). The U.S. Environmental Protection Agency (EPA) incorporated these summary measures into its Smart Growth Index (SGI) model, a widely used sketch-planning tool for travel and air quality analysis. The SGI model measures density as residents plus jobs per square mile; diversity as the ratio of jobs to residents divided by the regional average of that ratio; and design as street network density, sidewalk coverage, and route directness (road distance divided by direct distance). Two of these three measures relate to street network design. Our 2001 study (Ewing & Cervero, 2001) suggested, for example, that a doubling of neighborhood density would reduce both per capita VT and VMT by approximately 5%, all else being equal. We also concluded that VMT was more elastic with respect to destination accessibility than the other three built environmental measures, meaning that highly accessible areas such as center cities produce substantially lower VMT than dense mixed-use developments in the exurbs. However, as noted earlier, our 2001 study relied on only 14 studies, and the elasticities were imprecise, some obtained by aggregating results for dissimilar environmental variables (e.g., local diversity measured as both entropy and jobs-housing balance). In this update, we compute weighted averages of results from a larger number of studies, and use more uniformly dened built-environmental variables.Meta-Analyses in PlanningUnlike traditional research methods, meta-analysis uses summary statistics from individual primary studies as the data points in a new analysis. This approach has both advantages and disadvantages for validity and reliability, as every standard textbook on meta-analysis explains (Borenstein, Hedges, Higgins, & Rothstein, 2009; Hunter & Schmidt, 2004; Lipsey & Wilson, 2001; Littell, Corcoran, & Pillai, 2008; Lyons, 2003 Schulze, 2004). The main advantage of meta-analysis is that it aggregates all available research on a topic, allowing common threads to emerge. The pooling of samples in a carefully constructed meta-analysis also makes its results more generalizable than those of the smaller primary studies on which it is based. But meta-analysis has drawbacks too. Combining stronger studies with weaker ones may contaminate the results of the former. Further, meta-analysis inevitably mixes apples and oranges due to the variation among studies in modeling techniques, independent and dependent variables, and sampling units. If we compare only very similar studies, sample sizes can become small, threatening statistical reliability, a problem that we admit characterizes some of the subcategories for which we present results in this article. Last, the studies for a metaanalysis are usually chosen from the published literature. This can result in publication bias since studies that show statistical signicance are more likely to be published (Rothstein, Sutton, & Borenstein, 2005). Publication bias may inate the absolute size of the effects estimated with a meta-analysis. Addressing these potential weaknesses involves tradeoffs. We sought to minimize publication bias in this metaanalysis by searching for unpublished reports, preprints, and white papers, as well as published articles. Online searches using Google Scholar and the Transportation 268Journal of the American Planning Association, Summer 2010, Vol. 76, No. 3 RJPA_A_477198.qxd 6/11/10 3:25 PM Page 268
Research Information Service (TRIS) were particularly helpful in this regard. We sought to minimize the apples-and-oranges problem by focusing on a subset of studies that employed disaggregate data and comparably dened variables. Yet, our efforts to avoid publication bias may have exacerbated the strong-weak study problem, and our efforts to achieve greater construct validity by segmenting the analysis into subgroups sharing comparably dened dependent and independent variables produced small sample sizes. More than a dozen studies have applied meta-analytical methods to the urban planning eld (Babisch, 2008; Bartholomew & Ewing, 2008; Bunn, Collier, Frost, Ker, Roberts, & Wentz, 2003; Button & Kerr, 1996; Button & Nijkamp, 1997; Cervero, 2002b; Debrezion, Pels, & Rietveld, 2003; Duncan, Spence, & Mummery, 2005; Graham & Glaister, 2002; Hamer & Chida, 2008; Lauria & Wagner, 2006; Leck, 2006; Nijkamp & Pepping, 1998; Smith & Huang, 1995; Stamps, 1990, 1999; Zhang, 2009). Bartholomew and Ewing (2008) combined results from 23 recent scenario planning studies to calculate the impacts of land-use changes on transportation greenhouse gas emissions. Button and Kerr (1996) explored the implications of urban trafc restraint schemes on congestion levels. Cervero (2002b) synthesized the results of induced travel demand studies. Debrezion et al. (2003) measured the impact of railway stations on residential and commercial property values. Nijkamp and Pepping (1998) analyzed factors critical to the success of sustainable city initiatives. Smith and Huang (1995) calculated the public's willingness to pay for cleaner air. Stamps (1990, 1999) applied meta-analysis to the visual preference literature. Most relevant to the present study, Leck (2006) identied 40 published studies of the built environment and travel, and selected 17 that met minimum methodological and statistical criteria. While Leck's meta-analysis stopped short of estimating average effect sizes, it did evaluate the statistical signicance of relationships between the built environment and travel, nding residential density, employment density, and land use mix to be inversely related to VMT at the p < .001 signicance level.ApproachSample of StudiesWe identied studies linking the built environment to travel using the Academic Search Premier, Google, Google Scholar, MEDLINE, PAIS International, PUBMED, Scopus, TRIS Online, TRANweb, Web of Science, and ISI Web of Knowledge databases using the keywords "built environment," "urban form," and "development," coupled with keywords "travel," "transit," and "walking." We also reviewed the compact discs of the Transportation Research Board's annual programs for relevant papers, contacted all leading researchers in this subject area for copies of their latest research, and put out a call for built-environment/ travel studies on the academic planners' listserv, PLANET. Finally, we examined the bibliographies of the previous literature reviews in this topic area to identify other pertinent studies. We inspected more than 200 studies that relate, quantitatively, characteristics of the built environment to measures of travel. From the universe of built-environment/ travel studies, we computed effect sizes for the more than 50 studies shown in Table 1.These studies have several things in common. As they analyze effects of the built environment on travel choices, all these studies control statistically for confounding inuences on travel behavior, sociodemographic inuences in particular. They use different statistical methods because the outcome variables differ from study to study.5All apply statistical tests to determine the signicance of the various effects. Almost all are based on sizeable samples, as shown in the appendix tables. Most capture the effects of more than one D variable simultaneously. Most importantly, we selected only studies for which data were available for computing effect sizes. We left out many quantitative studies for various reasons. Many studies did not publish average values of dependent and independent variables from which point elasticities could be calculated. Although we followed up with authors to try to obtain these descriptive statistics, in many cases the research was several years old and the authors had moved on to other subjects. In a few cases, we could not track the authors down or get them to respond to repeated data requests. Many studies used highly aggregated city, county, or metropolitan level data (e.g., Newman & Kenworthy, 2006; van de Coevering & Schwanen, 2006). Such studies have limited variance in both dependent and independent variables with which to explain relationships. More importantly, it is inappropriate to make causal and associative inferences about individuals based on results obtained from aggregate data, an error called the ecological fallacy As we would like our elasticities to be suitable for use in models predicting individual behavior, we did not use studies relying on aggregate data. Several studies used statistical methods from which simple summary effect size measures could not be calculated, including some using structural equation models (SEM) to capture complex interactions among built environment and travel variables (e.g., Bagley & Mokhtarian, Ewing and Cervero: Travel and the Built Environment: A Meta-Analysis269 RJPA_A_477198.qxd 6/11/10 3:25 PM Page 269
270Journal of the American Planning Association, Summer 2010, Vol. 76, No. 3 Table 1. Studies included in the sample. Selfselection Study sitesDataMethodsControlscontrolled fora Bento et al., 2003Nationwide Personal Transportation DLNR/LGRSE/LS/OTno Survey (114 metropolitan statistical areas) Bhat & Eluru, 2009San Francisco Bay Area, CADCOPSE/OTyes Bhat, Sen, et al., 2009San Francisco Bay Area, CADMDC/LGRSE/OTno Bhatia, 200420 communities in Washington, DCALNRSEno Boarnet et al., 2004Portland, ORDLNR/PRRSE/OTno Boarnet et al., 2008Portland, OR DTORSEyes Boarnet et al., in press8 neighborhoods in southern CADNBRSEno Boer et al., 200710 U.S. metropolitan areasDPSMSE/WEno Cao et al., 20066 neighborhoods in Austin, TXDNBRSE/ATyes Cao, Mokhtarian, et al., 2009b8 neighborhoods in northern CADSURSE/ATyes Cao, Xu, et al., 2009Raleigh, NCDPSMSE/ATyes Cervero, 2002aMontgomery County, MDDLGRSE/LSno Cervero, 2006225 light rail transit stations in 11 metropolitan areasALNRST/LSno Cervero, 200726 TODs in ve CA regionsDLGRSE/ LS/WP/ATyes Cervero & Duncan, 2003San Francisco Bay Area, CADLGRSE/OTno Cervero & Duncan, 2006San Francisco Bay Area, CADLNRSE/WPno Cervero & Kockelman, 199750 neighborhoods in the San Francisco Bay Area, CADLNR/LGRSE/LSno Chapman & Frank, 2004Atlanta, GADLNRSEno Chatman, 2003Nationwide Personal Transportation SurveyDTORSE/WPno Chatman, 2008San Francisco, CA/San Diego, CADLNR/NBRSE/LS/OTno Chatman, 2009San Francisco, CA/San Diego, CADNBRSE/LS/OT/ATyes Ewing et al., 1996Palm Beach County/Dade County, FLDLNRSEno Ewing et al., 200952 mixed use developments in PortlandDHLMSEno Fan, 2007Raleigh-Durham, NCDLNRSE/LS/OT/ATyes Frank & Engelke, 2005Seattle, WADLNRSE/LSno Frank et al., 2008Seattle, WADLGRSE/LSno Frank et al., 2009Seattle, WADLNRSEno Greenwald, 2009Sacramento, CADLNR/TOR/SEno NBR Greenwald & Boarnet, 2001Portland, ORDOPRSE/LSno Handy & Clifton, 20016 neighborhoods in Austin, TXDLNRSEno Handy et al., 20068 neighborhoods in northern CADNBRSE/ATyes Hedel & Vance, 2007German Mobility Panel SurveyDLNR/PRRSE/OTno Hess et al., 199912 neighborhood commercial centers in Seattle, WAALNRSEno Holtzclaw et al., 2002Chicago, IL/Los Angeles, CA/San Francisco, CAANLRSEno Joh et al., 20098 neighborhoods in southern CADLNRSE/CR/ATyes Khattak & Rodriguez, 20052 neighborhoods in Chapel Hill, NCDNBRSE/ATyes Kitamura et al., 19975 communities in San Francisco, CA regionDLNRSE/ATyes Kockelman, 1997San Francisco Bay Area, CADLNR/LGRSEno Kuby et al., 2004268 light rail transit stations in nine metropolitan areasALNRST/OTno Kuzmyak et al., 2006Baltimore, MDDLNRSEno Kuzmyak, 2009aLos Angeles, CADLNRSEno Kuzmyak, 2009bPhoenix, AZDLNRSEno Lee & Moudon, 2006aSeattle, WADLGRSE/LSyes Lund, 20038 neighborhoods in Portland, ORDLNRSE/ATyes Lund et al., 200440 TODs in four CA regionsDLGRSE/LS/WP/ATyes Naess, 200529 neighborhoods in Copenhagen, DenmarkDLNRSE/WP/ATyes RJPA_A_477198.qxd 6/11/10 3:25 PM Page 270
Ewing and Cervero: Travel and the Built Environment: A Meta-Analysis271 Table 1. (continued). Selfselection Study sitesDataMethodsControlscontrolled fora Pickrell & Schimek, 1999Nationwide Personal Transportation SurveyDLNRSEno Plaut, 2005American Housing SurveyDLGRSE/OTno Pushkar et al., 2000795 zones in Toronto, Ontario, CanadaASLESE/LSno Rajamani et al., 2003Portland, OR DLGRSE/LSno Reilly, 2002San Francisco, CADLGRSE/OTno Rodriguez & Joo, 2004Chapel Hill, NCDLGRSE/LS/OTno Rose, 20043 neighborhoods in PortlandDLNR/PORSEno Schimek, 1996Nationwide Personal Transportation SurveyDSLESEno Shay et al., 20061 neighborhood in Chapel Hill, NCDNBRSE/ATyes Shay & Khattak, 20052 neighborhoods in Chapel Hill, NCDLNR/NBRSEno Shen, 2000Boston, MAALNRSEno Sun et al., 1998Portland, ORDLNRSEno Targa & Clifton, 2005Baltimore, MDDPORSE/ATyes Zegras, 2007Santiago, ChileDLNR/LGRSEno Zhang, 2004Boston, MA/Hong KongDLGRSE/LS/OTno Zhou & Kockelman, 2008Austin, TXDLNR/PRRSEyes Notes: We use the following abbreviations: Data: A = aggregate D = disaggregate Methods:COP = Copula-based switching model GEE = generalized estimating equations HLM = hierarchical linear modeling LGR = logistic regression LNR = linear regression MDC = multiple discrete continuous extreme value model NBR = negative binomial regression NLR = nonlinear regression OPR = ordered probit regression POR = Poisson regression PRR = probit regression PSM = propensity score matching PSS = propensity score stratication SLR = simultaneous linear equations SUR = seemingly unrelated regression TOR = Tobit regression Controls:AT = attitudinal variables CR = crime variables LS = level of service variables OT = other variables SE = socioeconomic variables ST = station variables WE = weather variables WP = workplace variables a. Cao, Mokhtarian, et al. (2009a) notes nine different approaches used to control for residential self-selection. The least ri gorous incorporates attitudinal measures in multivariate regression models, while the most rigorous jointly estimates models of residential choice and travel behavior, treating residential choice as an endogenous variable. RJPA_A_477198.qxd 6/11/10 3:25 PM Page 271
2002; Cao, Mokhtarian, & Handy, 2007; Cervero & Murakami, 2010). In SEM, different equations represent different effects of variables on one another, both direct and indirect through intermediate variables. These cannot be aggregated into a single elasticity.6We excluded many studies because they dealt with limited populations or trip purposes (e.g., Chen & McKnight, 2007; Li, Fisher, Brownson, & Bosworth, 2005; Waygood, Sun, Kitamura, 2009). Notably, several recent studies of student travel to school cannot be generalized to other populations and trip purposes. The literature suggests that the choice of mode for the journey to school is based on very different considerations than those for other trip making (Ewing, Schroeer, & Greene, 2004; Yarlagadda & Srinivasan, 2008). We excluded some studies because they characterized the built environment subjectively rather than objectively, that is, in terms of qualities perceived and reported by travelers rather than variables measured in a standardized way by researchers (e.g., Craig, Brownson, Cragg, & Dunn, 2002; Handy, Cao, & Mokhtarian, 2005). Subjective measures are common in public health studies. While perceptions are important, they differ from objective measures of the built environment and are arguably more difcult for planners and public policymakers to inuence (e.g., Livi-Smith, 2009; McCormack et al., 2004; McGinn, Evenson, Herring, Huston, & Rodriguez, 2007). For studies that include both types of measures, we analyzed relationships only for the objective measures. Finally, we excluded several studies because they created and then applied built environmental indices without true zero values (e.g., indices derived through factor analysis). There is no defensible way to compute elasticities, the common currency of this article, for such studies (e.g., Estupinan & Rodriguez, 2008; Frank, Saelens, Powell, & Chapman, 2007; Livi-Smith, 2009). For the same reason, we excluded several excellent studies whose independent variables, although initially continuous, had been converted to categorical variables to simplify the interpretation of results (e.g., Lee & Moudon, 2006b; McGinn et al., 2007; Oakes, Forsyth, & Schmitz, 2007). We analyzed studies using nominal variables to characterize the built environment separately from those using continuous variables. Examples of the former include studies distinguishing between traditional urban and conventional suburban development or between transitoriented and auto-oriented development. We only included such studies if they analyzed disaggregate data and controlled for individual socioeconomic differences across their samples, thereby capturing the marginal effects of neighborhood type.7Common MetricsTo combine results from different studies, a metaanalysis requires a common measure of effect size. Our common metric is the elasticity of some travel outcome with respect to one of the D variables. An elasticity is the ratio of the percentage change in one variable associated with the percentage change in another variable (a point elasticity is the ratio when these changes are innitely small). Elasticities are dimensionless (unit-free) measures of the associations between pairs of variables and are the most widely used measures of effect size in economic and planning research. For outcomes measured as continuous variables, such as numbers of walk trips, an elasticity can be interpreted as the percent change in the outcome variable when a specied independent variable increases by 1%. For outcomes measured as categorical variables, such as the choice of walking over other modes, an elasticity can be interpreted as the percent change in the probability of choosing that alternative (or the percent change in that alternative's market share) when the specied independent variable increases by 1%.Elasticities in Individual Studies in the SampleWe obtained elasticities from the individual studies in our sample in one of four ways, just as in Ewing and Cervero (2001). We either: (1) copied them from published studies where they were reported explicitly; (2) calculated them ourselves from regression coefcients and the mean values of dependent and independent variables; (3) derived them from data sets already available to us or made available by other researchers; or (4) obtained them directly from the original researchers. Most commonly, we used one of the formulas shown in Table 2to compute elasticities, depending on which statistical method was used to estimate coefcient values. When regression coefcients were not signicant, we could have chosen to drop the observations or substitute zero values for the elasticities, since the coefcients were not statistically different from zero, but we chose instead to use the reported coefcients to compute elasticities, again using the formulas in Table 2. Dropping the observations would have biased the average elasticities away from the null hypothesis of zero elasticity, and thus we rejected this option. Substituting zero values for computed elasticities would have had the opposite effect, biasing average values toward the null hypothesis, thus we rejected it as well. Instead, we used the best available estimates of central tendency in all cases, the regression coefcients themselves, to compute elasticities. This is the standard approach in meta-analysis (see, e.g., Melo, Graham, & Noland, 2009). 272Journal of the American Planning Association, Summer 2010, Vol. 76, No. 3 RJPA_A_477198.qxd 6/11/10 3:25 PM Page 272
Borenstein et al. (2009) argue against another possibility, using signicance levels as proxies for effect size, since they depend not only on effect size but also on sample size: "Because we work with the effect sizes directly we avoid the problem of interpreting nonsignicant p -values to indicate the absence of an effect (or of interpreting signicant p -values to indicate a large effect)" (p. 300). Ideally, the original studies would have computed elasticities for each observation (trip, traveler, or household) and then averaged them over the sample. Indeed, a few of the researchers who reported elasticities did exactly that (e.g., Bento, Cropper, Mobarak, & Vinha, 2003; Bhat, Sen, & Eluru, 2009; Rodriguez & Joo, 2004). However, since we could not ask all these busy people to go back and compute elasticities, we have instead estimated elasticities at the overall sample means of the dependent and/or independent variables, as indicated in Table 3. While commonplace, this procedure could introduce a fair amount of error in the elasticity estimates. Elasticities calculated at mean values of dependent and independent variables may differ signicantly from the average values of individual elasticities due to the nonlinear nature of many of the functions involved (e.g., logistic functions). "In general, the probability evaluated at the average utility underestimates the average probability when the individuals' choice probabilities are low and overestimates when they are high" (Train, 1986, p. 42). Train (1986) cites work by Talvitie (1976), who found in a mode choice analysis that elasticities at the average representative utility can be as much as two to three times greater or less than the average of individual elasticities. This is a greater concern with discrete choice models than with the linear regression models that Table 1 shows are most commonly used to study the built environment and travel. Due to the large number studies we summarize here, we show the effect sizes for individual studies in appendix tables for each travel outcome of interest (VMT, walking, and transit use) with respect to each built environment variable of interest (density, diversity, design, destination accessibility, distance to transit, and neighborhood type). All effect sizes are measured as elasticities, except those for neighborhood type, which is a categorical variable. The effect size for neighborhood type is the proportional difference in a travel outcome between conventional suburban neighborhoods and more compact, walkable neighborhoods. Ewing and Cervero: Travel and the Built Environment: A Meta-Analysis273 Table 3. Weighted average elasticities of VMT with respect to built-environment variables. Total number Number of studies with Weighted average of studiescontrols for self-selectionelasticity of VMT( e ) DensityHousehold/population density91 ÂŠ 0.04 Job density610.00 DiversityLand use mix (entropy index)100 ÂŠ 0.09 Jobs-housing balance40 ÂŠ 0.02 DesignIntersection/street density60 ÂŠ 0.12 % 4-way intersections31 ÂŠ 0.12 Destination Job accessibility by auto50 ÂŠ 0.20 accessibilityJob accessibility by transit30 ÂŠ 0.05 Distance to downtown31 ÂŠ 0.22 Distance to transitDistance to nearest transit stop61 ÂŠ 0.05 Table 2. Elasticity estimation formulas. Notes: is the regression coefcient on the built-environment variable of interest, the mean value of the travel variable of interest, and the mean value of the built-environment variable of interest. a.is is the mean estimated probability of occurrence. b. Applied only to positive values of the Tobit distribution (i.e., where > 0). Regression specicationElasticity Linear Log-log Log-linear Linear-log Logistica Poisson Negative binomial Tobitb RJPA_A_477198.qxd 6/11/10 3:25 PM Page 273
We consistently report the elasticity values with a positive sign indicating the effects of greater accessibility, which required reversing signs in many cases, as noted in the tables. Thus, for example, a negative elasticity of VMT with respect to measures of destination accessibility in our appendix tables always indicates that VMT drops as destination accessibility improves. Where destination accessibility was measured originally in terms of jobs reachable within a given travel time, our sign is the same as that obtained by the original study. However, where destination accessibility was measured in terms of distance to downtown, for example, we reversed the sign of the elasticity in the original source so that higher values of the independent variable correspond to better, not worse, accessibility. Where studies reported results for general travel and, in addition, for different trip purposes or different types of travelers, we report effect sizes only for the most general class of travel. Thus, for example, if a study estimated VMT models for all trips and for work trips alone, we present only the former. A few studies analyzed only subcategories of travel, and in these cases, we sometimes present more than one set of results for a given study. Weighted Average ElasticitiesWe used individual elasticities from primary studies to compute weighted average elasticities for many dependent/independent variable pairs representing travel outcomes and attributes of the built environment. We show the resulting weighted average elasticities in Tables 3, 4, and 5.We calculated averages where three conditions were met: (1) a sample of at least three studies was available; (2) for these particular studies, dependent and independent variables were comparably dened; and (3) for these particular studies, disaggregate travel data were used to estimate models. The numbers of studies in each sample are as indicated in Tables 3, 4, and 5. These results should be used only as ballpark estimates, both because of the minimum sample size we chose and because of how we computed weighted average elasticities. We settled on a minimum sample size of three studies8due to data limitations (as in Tompa, de Oliveira, Dolinschi, & Irvin, 2008). While the relationship between the built environment and travel is the most heavily researched subject in urban planning, when studies are segmented by variable type, samples never reach what some would consider a reasonable minimum sample size (Lau, Ioannidis, Terrin, Schmid, & Olkin, 2006). Also, to maximize our 274Journal of the American Planning Association, Summer 2010, Vol. 76, No. 3 Table 4. Weighted average elasticities of walking with respect to built environment variables. Total number Number of studies with Weighted average of studiescontrols for self-selectionelasticity of walking ( e ) DensityHousehold/population density1000.07 Job density600.04 Commercial oor area ratio300.07 DiversityLand use mix (entropy index)810.15 Jobs-housing balance400.19 Distance to a store 530.25 DesignIntersection/street density720.39 % 4-way intersections51 ÂŠ 0.06 Destination accessibilityJob within one mile300.15 Distance to transitDistance to nearest transit stop320.15 Table 5. Weighted average elasticities of transit use with respect to built environment variables. Total number Number of studies with Weighted average of studiescontrols for self-selectionelasticity of transit use DensityHousehold/population density1000.07 Job density600.01 DiversityLand use mix (entropy index)600.12 DesignIntersection/street density400.23 % 4-way intersections520.29 Distance to transitDistance to nearest transit stop310.29 RJPA_A_477198.qxd 6/11/10 3:25 PM Page 274
sample sizes, we mixed the relatively few studies that control for self-selection with the many that do not. We advise readers to exercise caution when using the elasticities based on small samples of primary studies (see Tables 3, 4, and 5), but rather than omit the categories for which only small samples were available, we aimed in this analysis to seed the meta-study of built environments and travel, expecting that others would augment and expand our database over time. We computed weighted average elasticities using sample size as a weighting factor because we lacked consistent standard error estimates from individual studies. Weighting by sample size is by far the most common approach in meta-analyses, since sample sizes are nearly always known (Shadish & Haddock, 1994, p. 264). However, it is not the optimal weighting scheme. Hedges and Olkin (1985) demonstrated that optimal weights are related to the standard errors of the effect size estimates, and this has become the gold standard in meta-analysis. Specically, because larger standard errors correspond to less precise estimates of effect sizes, the preferred method is to calculate a meta-analysis weight as an inverse variance weight, orthe inverse of the squared standard error (Borenstein et al., 2009; Hunter & Schmidt, 2004; Lipsey & Wilson, 2001; Littell et al., 2008; Schulze, 2004). From a statistical standpoint, such weights are optimal since they minimize the variance of the average effect size estimates. They also make intuitive sense, as they give the greatest weight to the most precise estimates from individual studies. No weighting factor except standard error allows judging whether the resulting weighted averages are statistically different from zero. Since we combine signicant and insignicant individual effect sizes, and do not have the data necessary to test for signicance, we do not report statistical condence for any of the results. It is thus possible that any given meta-elasticity is not signicantly different from zero. We particularly advise readers to exercise caution in using weighted average elasticities when the elasticities on which they are based are statistically insignificant, as shown in the appendix tables.DiscussionFor all of the variable pairs we discuss here, the relationships between travel variables and built environmental variables are inelastic. The weighted average elasticity with the greatest absolute magnitude is 0.39, and most elasticities are much smaller. Still, the combined effect of several built environmental variables on travel could be quite large. As in our 2001 meta-study (Ewing & Cervero, 2001), the D variable most strongly associated with VMT is destination accessibility. Our elasticity of VMT with respect to "job accessibility by auto" in this meta-analysis, 0.20, is identical to the elasticity in the earlier study. In fact, the 0.20 VMT elasticity is nearly as large as the elasticities of the rst three D variables (density, diversity, and design) combined; this too is consistent with our earlier meta-study. Equally strongly, though negatively, related to VMT is the distance to downtown. This variable is a proxy for many Ds, as living in the city core typically means higher densities in mixed-use settings with good regional accessibility. Next most strongly associated with VMT are the design metrics intersection density and street connectivity. This is surprising, given the emphasis in the qualitative literature on density and diversity, and the relatively limited attention paid to design. The weighted average elasticities of these two street network variables are identical. Both short blocks and many interconnections apparently shorten travel distances to about the same extent. Also surprising are the small elasticities of VMT with respect to population and job densities. Conventional wisdom holds that population density is a primary determinant of vehicular travel, and that density at the work end of trips is as important as density at the home end in moderating VMT. This does not appear to be the case once other variables are controlled. Our previous study (Ewing & Cervero, 2001) did not address walking and transit use, thus we have no benchmarks against which to compare the results in Tables 4 and 5. The meta-analysis shows that mode share and likelihood of walk trips are most strongly associated with the design and diversity dimensions of built environments. Intersection density, jobs-housing balance, and distance to stores have the greatest elasticities. Interestingly, intersection density is a more signicant variable than street connectivity. Intuitively this seems right, as walkability may be limited even if connectivity is excellent when blocks are long. Also of interest is the fact that jobs-housing balance has a stronger relationship to walking than the more commonly used land use mix (entropy) variable. Several variables that often go hand-in-hand with population density have elasticities that are well above that of population density. Also, as with VMT, job density is less strongly related to walking than is population density. Finally, Table 5 suggests that having transit stops nearby may stimulate walking (Cervero, 2001; Ryan & Frank, 2009). The mode share and likelihood of transit trips are strongly associated with transit access. Living near a bus stop appears to be an inducement to ride transit, supporting the transit industry's standard of running buses within a quarter mile of most residents. Next in importance are road network variables and, then, measures of land use Ewing and Cervero: Travel and the Built Environment: A Meta-Analysis275 RJPA_A_477198.qxd 6/11/10 3:25 PM Page 275
mix. High intersection density and great street connectivity shorten access distances and provide more routing options for transit users and transit service providers. Land use mix makes it possible to efciently link transit trips with errands on the way to and from transit stops. It is sometimes said that "mass transit needs mass'"; however, this is not supported by the low elasticities of transit use with respect to population and job densities in Table 5. No clear pattern emerges from scanning across the Tables 3, 4, and 5. Perhaps what can be said with the highest degree of condence is that destination accessibility is most strongly related to both motorized (i.e., VMT) and nonmotorized (i.e., walking) travel and that among the remaining Ds, density has the weakest association with travel choices. The primacy of destination accessibility may be due to lower levels of auto ownership and auto dependence at central locations. Almost any development in a central location is likely to generate less automobile travel than the best-designed, compact, mixed-use development in a remote location. The relatively weak relationships between density and travel likely indicate that density is an intermediate variable that is often expressed by the other Ds (i.e., dense settings commonly have mixed uses, short blocks, and central locations, all of which shorten trips and encourage walking). Among design variables, intersection density more strongly sways the decision to walk than does street connectivity. And, among diversity variables, jobs-housing balance is a stronger predictor of walk mode choice than land use mix measures. Linking where people live and work allows more to commute by foot, and this appears to shape mode choice more than sprinkling multiple land uses around a neighborhood. Controls for residential self-selection appear to increase the absolute magnitude of elasticities if they have any effect at all. This conclusion follows from a simple review of elasticities in the appendix. There may be good explanations for this unexpected result. In a region with few pedestrianand transit-friendly neighborhoods, residential selfselection likely matches individual preferences with place characteristics, increasing the effect of the D variables, a possibility posited by Lund, Willson, and Cervero (2006). if people are simply moving from one transit-accessible location to another (and they use transit regularly at both locations), then there is theoretically no overall increase in ridership levels. If, however, the resident was unable to take advantage of transit service at their prior residence, then moves to a TOD (transit-oriented development) and begins to use the transit service, the TOD is fullling a latent demand for transit accessibility and the net effect on ridership is positive. (p. 256) Similarly, Chatman (2009) hypothesizes that "[r]esidential self-selection may actually cause underestimates of built environment inuences, because households prioritizing travel accessparticularly, transit accessibilitymay be more set in their ways, and because households may not nd accessible neighborhoods even if they prioritize accessibility" (p. 1087). He carries out regressions that explicitly test for this, and nds that self-selection is more likely to enhance than diminish built environmental inuences. Still, we are left with a question. Most of the literature reviewed by Cao, Mokhtarian, et al. (2009a) shows that the effect of the built environment on travel is attenuated by controlling for self-selection, whereas we nd no effect (or enhanced effects) after controlling for self-selection. The difference may lie in the different samples included in our study and that of Cao, Mokhtarian, et al. (2009a), or in the crude way we operationalized self-selection, lumping all studies that control for self-selection together regardless of methodology.ApplicationsThis article provides elasticities in two forms that may be useful to planners: elasticity estimates from primary studies (in the appendix tables) and average elasticities from our pooled samples (in Tables 3, 4, and 5). If a planner happens to have an application in a location near one of those listed in the appendix tables, if not too many years have intervened since that study was completed, and if the study included the right D variables, he or she can simply borrow an elasticity estimate from the appendix, provided that the appendix table indicates it meets conventional statistical signicance criteria. Thus, for applications in Boston in the near future, Zhang's (2004) estimate of the elasticity of walk/bike mode choice with respect to population density (0.11) may be used without modication. More commonly, geographic and functional gaps in the literature may make the elasticities in Tables 3, 4, and 5 useful to planners. These elasticities may be applied in sketch planning to compute estimates of VMT, walking, and transit use relative to a base case, or in post-processing travel and activity forecasts from four-step travel demand models to reect the inuence of the ve Ds. The literature covers post-processing applications well (Cervero, 2006; DKS Associates, 2007; Johnston, 2004; Walters, Ewing, & Allen, 2000). These new elasticity values can be used in exactly the same way as earlier elasticity estimates. Sketch planning applications are limited only by the creativity of planning analysts. To illustrate, climate action planning of the type currently underway in California and 18 other states will require VMT estimates in order to 276Journal of the American Planning Association, Summer 2010, Vol. 76, No. 3 RJPA_A_477198.qxd 6/11/10 3:25 PM Page 276
extrapolate current trends and project an alternative lowercarbon future. These states have set greenhouse gas emission reduction targets and, with their metropolitan planning organizations, will need to pull together veriable plans that include smart growth elements. If planners are willing to make assumptions about the increases in density and other D variables that can be achieved with policy changes, they can use elasticity values from this article to estimate VMT reductions in urbanized areas and to translate these in turn into effects on CO2. Another potential sketch planning application could be to assess health impacts. Rates of physical activity, including walking, are inputs to health assessment models. Again, once planners make assumptions about changes in the D variables under future scenarios, increases in walking can easily be computed using elasticities. Until now there has been no empirically grounded methodology for making such projections. Elasticities could also be applied to trafc impact analysis. There has been no way to adjust the Institute of Transportation Engineers' (ITE) trip generation rates for walking and transit use, which has left developers of dense developments at urban sites paying impact fees and other exactions at the same rate as their suburban counterparts. The only adjustment previously allowed was for internal capture of trips within mixed-use developments, which did nothing for the typical inll project. Elasticity values can be used to adjust ITE trip rates for suburban developments to reect how greater densities and other environmental attributes would affect trip making. The elasticities in this meta-analysis are based on the most complete data available as of late 2009. However, as we acknowledge, sample sizes are small and the number of studies controlling for residential preferences and attitudes is still miniscule. We also do not know the condence intervals around our meta-analysis results. Users should weigh these shortcomings when applying results to any particular context or local setting. However, they provide a base on which to build. As more built environment-travel studies appear in the planning literature, it will be important to update and rene our results.AcknowledgmentsThe authors wish to acknowledge funding for this study from the Development, Community, and Environment Division of the U.S. Environmental Protection Agency. We also wish to acknowledge data and other assistance from the following individuals, hoping we didn't miss anyone: Chandra Bhat, Marlon Boarnet, Rob Boer, Mark Bradley, Jason Cao, Dan Chatman, Cynthia Chen, Mike Duncan, Yingling Fan, Ann Forsyth, Larry Frank, Jessica Greene, Mike Greenwald, Daniel Hess, Ken Joh, Kara Kockelman, Rich Kuzmyak, Chanam Lee, Tracy McMillan, Petter Naess, Mike Reilly, Daniel Rodriguez, Elizabeth Shay, C. Scott Smith, Qing Shen, Xiaoduan Sun, Chris Zegras, Ming Zhang, and Brenda Zhou.Notes1. A full list of studies is available from the corresponding author. 2. Vehicle trips (VT) is not studied as widely as these other outcome measures and is not related to as many important outcomes. However, it is a critical determinant of regulated vehicle emissions, which was the focus of our 2001 literature review. 3. The percentage varied depending on which locations were paired and compared, whether urban and suburban locations, urban and exurban, etc. 4. Transit route density is measured by miles of transit routes per square mile of land area. 5. Linear regression is used where the travel variable in continuous, Poisson regression where the travel variable is a count, logistic regression where the dependent variable is a probability, and so forth. 6. Several studies applied ordered probit regression to data on counts of walk and transit trips. We excluded all but one of these studies from the meta-analysis because the breakpoint parameters ( ) for the ordered categories were unavailable, which meant we could not calculate marginal effects. These parameters were available for one ordered probit study (Greenwald & Boarnet, 2001), and Jason Cao computed elasticities for us. We used elasticities for the median ordered category. 7. Due to a dearth of solid research, we could not study certain important travel outcomes with meta-analysis. Most notably, this article is silent regarding the effects of the built environment on trip chaining in multipurpose tours, internal capture of trips within mixed-use developments, and the choice of bicycling as a travel mode. 8. The following quotation from Rodenburg, Benjamin, de Roos, Meijer, and Stams (2009) explains that a meta-analysis in another eld settled on seven studies as a minimum sample size: Some limitations of this meta-analytic study should be mentioned. Although the minimum number of studies to permit a metaanalysis is only three studies (Treadwell, Tregear, Reston, & Turkelson, 2006) and many published meta-analyses contain nine or fewer studies (Lau, Ioannidis, Terrin, Schmid, & Olkin, 2006), the small number of seven studies included in this meta-analytic review limits the generalizability of our ndings and the possibilities of examining and adjusting for publication bias by means of more complex analytic methods (Macaskill, Walter, & Irwig, 2001). (p. 605)ReferencesBabisch, W. (2008). Road trafc noise and cardiovascular risk. Noise & Health 10 (38), 2733. Badland, H., & Schoeld, G. (2005). Transport, urban design, and physical activity: An evidence-based update. Transportation Research D, 10 (3), 177196. Badoe, D. A., & Miller, E. J. (2000). Transportation-land-use interaction: Empirical ndings in North America, and the implications for modeling. Transportation Research D, 5 (4), 235263. Bagley, M., & Mokhtarian, P. (2002). The impact of residential neighborhood type on travel behavior: A structural equations modeling approach. Annals of Regional Science 36 (2), 279297. Bartholomew, K., & Ewing, R. (2008). Land use-transportation scenarios and future vehicle travel and land consumption: A metaanalysis. Journal of the American Planning Association 75 (1), 115. Bauman, A. E., & Bull, F. C. (2007). Environmental correlates of physical activity and walking in adults and children: A review of reviews London, UK: National Institute of Health and Clinical Excellence. Ewing and Cervero: Travel and the Built Environment: A Meta-Analysis277 RJPA_A_477198.qxd 6/11/10 3:25 PM Page 277
Bento, A. M., Cropper, M. L., Mobarak, A. M., & Vinha, K. (2003). The impact of urban spatial structure on travel demand in the United States. (World Bank policy research working paper #3007). Besser, L., & Dannenberg, A. (2005). Walking to public transit: Steps to help meet physical activity recommendations. American Journal of Preventive Medicine 29 (4), 273280. Bhat, C. R., & Eluru, N. (2009). A Copula-based approach to accommodate residential self-selection effects in travel behavior modeling. Transportation Research B 43 (7), 749765. Bhat, C. R., Sen, S., & Eluru, N. (2009). The impact of demographics, built environment attributes, vehicle characteristics, and gasoline prices on household vehicle holdings and use. Transportation Research B 43 (1), 118. Bhatia, R. (2004, June). Land use: A key to livable transportation Paper presented at the 40th International Making Cities Livable conference, London, UK. Boarnet, M. G., & Crane, R. (2001). The inuence of land use on travel behavior: Specication and estimation strategies. Transportation Research A 35 (9), 823845. Boarnet, M. G., Greenwald, M., & McMillan, T. (2008). Walking, urban design, and health: Toward a cost-benet analysis framework. Journal of Planning Education and Research, 27 (3), 341358. Boarnet, M. G., Joh, K., Siembab, W., Fulton, W., & Nguyen, M. T. (in press). Retrotting the suburbs to increase walking: Evidence from a land use-travel study. Urban Studies Boarnet, M. G., Nesamani, K. S., & Smith, C. S. (2004, January). Comparing the inuence of land use on nonwork trip generation and vehicle distance traveled: An analysis using travel diary data. Paper presented at the 83rd annual meeting of the Transportation Research Board, Washington, DC. Boer, R., Zheng, Y., Overton, A., Ridgeway, G. K., & Cohen, D. A. (2007). Neighborhood design and walking trips in ten U.S. metropolitan areas. American Journal of Preventive Medicine 32 (4), 298304. Borenstein, M., Hedges, L.V., Higgins, J. P. T., & Rothstein, H. R. (2009). Introduction to meta-analysis Chichester, UK: Wiley. Bunn, F., Collier, T., Frost, C., Ker, K., Roberts, I. & Wentz, R. (2003). Trafc calming for the prevention of road trafc injuries: Systematic review and meta-analysis. Injury Prevention 9 (3), 200204. Button, K., & Kerr, J. (1996). Effectiveness of trafc restraint policies: A simple meta-regression analysis. International Journal of Transport Economics 23 (2), 213225. Button, K., & Nijkamp, P. (1997). Environmental policy analysis and the usefulness of meta-analysis. Socio-Economic Planning Sciences 31 (3), 231240. Cao, X. (2010). Exploring causal effects of neighborhood type on walking behavior using stratication on the propensity score. Environment and Planning A 42 (2), 487504. Cao, X., Handy, S. L., & Mokhtarian, P. L. (2006). The inuences of the built environment and residential self-selection on pedestrian behavior: Evidence from Austin, TX. Transportation, 33 (1), 120. Cao, X., Mokhtarian, P. L., & Handy, S. L. (2007). Do changes in neighborhood characteristics lead to changes in travel behavior? A structural equations modeling approach. Transportation, 34(5), 535556. Cao, X., Mokhtarian, P. L., & Handy, S. L. (2009a). Examining the impacts of residential self-selection on travel behaviour: A focus on empirical ndings. Transport Reviews 29 (3), 359395. Cao, X., Mokhtarian, P. L. & Handy, S. L. (2009b). The relationship between the built environment and nonwork travel: A case study of northern California. Transportation Research Part A 43 (5), 548559. Cao, X., Xu, Z., & Fan, Y. (2009, January). Exploring the connections among residential location, self-selection, and driving behavior: A case study of Raleigh, NC Paper presented at the 89th annual meeting of the Transportation Research Board. Cervero, R. (2001). Walk-and-ride: Factors inuencing pedestrian access to transit. Journal of Public Transportation 3 (4), 123. Cervero, R. (2002a). Built environments and mode choice: Toward a normative framework. Transportation Research D 7 (4), 265284. Cervero, R. (2002b). Induced travel demand: Research design, empirical evidence, and normative policies. Journal of Planning Literature 17 (1), 320. Cervero, R. (2003). The built environment and travel: Evidence from the United States. European Journal of Transport and Infrastructure Research 3 (2), 119137. Cervero, R. (2006). Alternative approaches to modeling the traveldemand impacts of smart growth. Journal of the American Planning Association 72 (3), 285295. Cervero, R. (2007). Transit oriented development's ridership bonus: A product of self-selection and public policies. Environment and Planning Part A 39 (9), 20682085. Cervero, R., & Duncan, M. (2003). Walking, bicycling, and urban landscapes: Evidence from the San Francisco Bay Area. American Journal of Public Health 93 (9), 14781483. Cervero, R., & Duncan, M. (2006). Which reduces vehicle travel more: Jobs-housing balance or retail-housing mixing? Journal of the American Planning Association 72 (4), 475490. Cervero, R., & Kockelman, K. (1997). Travel demand and the 3Ds: Density, diversity, and design. Transportation Research D 2 (3), 199219. Cervero, R., & Murakami, J. (2010). Effects of built environments on vehicle miles traveled: Evidence from 370 U.S. metropolitan areas. Environment and Planning A 42 (2), 400418 Chapman, J., & Frank, L. (2004). Integrating travel behavior and urban form data to address transportation and air quality problems in Atlanta, Georgia (Research Project No. 9819, Task Order 9713). Washington, DC: U.S. Department of Transportation.Chatman, D. G. (2003). How density and mixed uses at the workplace affect personal commercial travel and commute mode choice. Transportation Research Record 1831 193201. Chatman, D. G. (2008) Deconstructing development density: Quality, quantity and price effects on household non-work travel. Transportation Research Part A 42 (7), 10091031. Chatman, D. G. (2009). Residential self-selection, the built environment, and nonwork travel: Evidence using new data and methods. Environment and Planning A,41 (5), 10721089. Chen, C., & McKnight, C. E. (2007). Does the built environment make a difference? Additional evidence for the daily activity and travel behavior of homemakers living in New York City and suburb. Journal of Transport Geography,15 (5), 380395. Craig, C. L., Brownson, R. C., Cragg, S. E., & Dunn, A. L. (2002). Exploring the effect of the environment on physical activity: A study examining walking to work. American Journal of Preventive Medicine, 23 (2), 3643. Crane, R. (1996). On form versus function: Will the new urbanism reduce trafc, or increase it? Journal of Planning Education and Research 15 (2), 117126. Crane, R. (2000). The inuence of urban form on travel: An interpretive review. Journal of Planning Literature 15 (1), 323. Cunningham, G. O., & Michael, Y. L. (2004). Concepts guiding the study of the impact of the built environment on physical activity for older adults: A review of the literature. American Journal of Health Promotion 18 (6), 435443. 278Journal of the American Planning Association, Summer 2010, Vol. 76, No. 3 RJPA_A_477198.qxd 6/11/10 3:25 PM Page 278
Debrezion, G., Pels, E., & Rietveld, P. (2003). The impact of railway stations on residential and commercial property value: A meta analysis (Tinbergen Institute Discussion Paper No. TI-2004-023/3). Amsterdam, The Netherlands: Tinbergen Institute. DKS Associates. (2007). Assessment of local models and tools for analyzing smart-growth strategies: Final report Sacramento, CA: California Department of Transportation. Duncan, M. J., Spence, J. C., & Mummery, W. K. (2005). Perceived environment and physical activity: A meta-analysis of selected environmental characteristics. International Journal of Behavioral Nutrition and Physical Activity,2 (11). doi:10.1186/1479-5868-2-11. Edwards, R. D. (2008). Public transit, obesity, and medical costs: Assessing the magnitudes. Preventive Medicine,46 (1), 1421. Estupinan, N., & Rodriguez, D. A. (2008). The relationship between urban form and station boardings for Bogota's BRT. Transportation Research A,42 (2), 296306. Ewing, R., & Cervero, R. (2001). Travel and the built environment. Transportation Research Record, 1780 87114. Ewing, R., DeAnna, M. & Li, S. (1996). Land use impacts on trip generation rates. Transportation Research Record,1518 17. Ewing, R., Greenwald, M. J., Zhang, M., Walters, J., Feldman, M., Cervero, R., Thomas, J. (2009). Measuring the impact of urban form and transit access on mixed use site trip generation ratesPortland pilot study. Washington, DC: U.S. Environmental Protection Agency. Ewing, R., Schroeer, W., Greene, W. (2004). School location and student travel: Analysis of factors affecting mode choice. Transportation Research Record 1895 5563. Fan, Y. (2007). The built environment, activity space, and time allocation: An activity-based framework for modeling the land use and travel connection (Unpublished doctoral dissertation.) University of North Carolina, Chapel Hill, NC. Frank, L. D. (2000). Land use and transportation interaction: Implications on public health and quality of life. Journal of Planning Education and Research 20 (1), 622. Frank, L. D., Bradley, M., Kavage, S., Chapman, J., & Lawton, K. (2008). Urban form, travel time, and cost relationships with tour complexity and mode choice. Transportation 35 (1), 3754. Frank, L. D., & Engelke, P. (2001). The built environment and human activity patterns: Exploring the impacts of urban form on public health. Journal of Planning Literature 16 (2), 202218. Frank, L. D., & Engelke, P. (2005). Multiple impacts of the built environment on public health: Walkable places and the exposure to air pollution. International Regional Science Review 28 (2), 193216. Frank, L. D., Kavage, S., Greenwald, M., Chapman, J., & Bradley, M. (2009). I-PLACE3S health & climate enhancements and their application in King County Seattle, WA: King County HealthScape. Frank, L. D., Kerr, J., Chapman, J., & Sallis, J. (2007). Urban form relationships with walk trip frequency and distance among youth. American Journal of Health Promotion, 21 (4), 305311. Frank, L. D., Saelens, B. E., Powell, K. E., & Chapman, J. E. (2007). Stepping towards causation: Do built environments or neighborhood and travel preferences explain physical activity, driving, and obesity? Social Science & Medicine 65 (9), 18981914. Gebel, K., Bauman, A. E., &Petticrew, M. (2007). The physical environment and physical activity. A critical appraisal of review articles.American Journal of Preventive Medicine, 32 (5), 361369. Graham, D., & Glaister, S. (2002). Review of income and price elasticities of demand for road trafc. London, UK: Centre for Transport Studies Imperial College of Science, Technology and Medicine. Greenwald, M. J. (2009). SACSIM modeling-elasticity results: Draft. Unpublished manuscript, Fehr & Peers Associates, Walnut Creek, CA. Greenwald, M. J., & Boarnet, M. G. (2001). The built environment as a determinant of walking behavior: Analyzing non-work pedestrian travel in Portland, Oregon. Transportation Research Record 1780 3343. Hamer, M., & Chida, Y. (2008). Active commuting and cardiovascular risk: A meta-analytic review. Preventive Medicine,46 (1), 913. Handy, S. L. (1993). Regional versus local accessibility: Implications for non-work travel. Transportation Research Record, 1400 5866. Handy, S. L. (2004). Critical assessment of the literature on the relationships among transportation, land use, and physical activity. Davis, CA: Department of Environmental Science and Policy, University of California, Davis. Prepared for the Transportation Research Board and Institute of Medicine Committee on Physical Activity, Health, Transportation, and Land Use. Handy, S. L., Cao, X., & Mokhtarian, P. L. (2005). Correlation or causality between the built environment and travel behavior? Evidence from Northern California. Transportation Research D 10 (6), 427444. Handy, S. L., Cao, X., & Mokhtarian, P. L. (2006). Self-selection in the relationship between the built environment and walkingEmpirical evidence from Northern California. Journal of the American Planning Association 72 (1), 5574. Handy, S. L., & Clifton, K. J. (2001). Local shopping as a strategy for reducing automobile travel. Transportation 28 (4), 317346. Heath, G. W., Brownson, R. C., Kruger, J., Miles, R., Powell, K. E., Ramsey, L. T., & the Task Force on Community Preventive Services. (2006). The effectiveness of urban design and land use and transport policies and practices to increase physical activity: A systematic review. Journal of Physical Activity and Health 3 (1), 5576. Hedel, R., & Vance, C. (2007, January). Impact of urban form on automobile travel: Disentangling causation from correlation Paper presented at the 86th annual meeting of the Transportation Research Board, Washington, DC. Hedges, L. V., & Olkin, I. (1985). Statistical methods for meta-analysis Orlando, FL: Academic Press. Hess, P. M., Moudon, A. V., Snyder, M. C., & Stanilov, K. (1999). Site design and pedestrian travel. Transportation Research Record, 1674 919. Holtzclaw, J., Clear, R., Dittmar, H., Goldstein, D., & Haas, P. (2002). Location efciency: Neighborhood and socioeconomic characteristics determine auto ownership and use-studies in Chicago, Los Angeles, and San Francisco. Transportation Planning and Technology 25 (1), 127. Humpel, N., Owen, N., & Leslie, E. (2002). Environmental factors associated with adults' participation in physical activity: A review. American Journal of Preventative Medicine 22 (3), 188199. Hunter, J. E., & Schmidt, F.L. (2004). Methods of meta-analysis: Correcting error and bias in research ndings Newbury Park, CA: Sage. Joh, K., Boarnet, M. G, & Nguyen, M. T. (2009, January). Interactions between race/ethnicity, attitude, and crime: Analyzing walking trips in the South Bay Area Paper presented at the 88th annual meeting of the Transportation Research Board, Washington, DC. Johnston, R. (2004). The urban transportation planning process. In S. Hanson and G. Guiliano (Eds.), The geography of urban transportation (pp. 115140). New York, NY: Guilford. Kahn, E. B., Ramsey, L. T., Brownson, R. C., Heath, G. W., Howze, E. H., Powell, K., & Stone, E. (2002). The effectiveness of interventions to increase physical activity: A systematic review. American Journal of Preventive Medicine 22 (4), 73107. Ewing and Cervero: Travel and the Built Environment: A Meta-Analysis279 RJPA_A_477198.qxd 6/11/10 3:25 PM Page 279
Khattak, A. J., & Rodriquez, D. (2005). Travel behavior in neotraditional neighborhood developments: A case study in USA. Transportation Research A 39 (6), 481500. Kitamura, R., Mokhtarian, P. L., & Laidet, L. (1997). A micro-analysis of land use and travel in ve neighborhoods in San Francisco Bay Area. Transportation 24 (2), 125158. Kockelman, K. M. (1997). Travel behavior as a function of accessibility, land use mixing, and land use balance: Evidence from the San Francisco Bay Area. Transportation Research Record, 1607 116125. Krahnstoever-Davison, K., & Lawson, C. T. (2006). Do attributes in the physical environment inuence children's physical activity? A review of the literature. International Journal of Behavioral Nutrition and Physical Activity 3 (19). doi: 10.1186/1479-5868-3-19. Kuby, M., Barranda, A., & Upchurch, C. (2004). Factors inuencing light-rail station boardings in the United States. Transportation Research A 38 (3),223258. Kuzmyak, R. (2009a). Estimating the travel benets of blueprint land use concepts Unpublished manuscript, Southern California Association of Governments, Los Angeles, CA. Kuzmyak, R. (2009b) .Estimates of point elasticities. Phoenix, AZ: Maricopa Association of Governments. Kuzmyak, R., Baber, C., & Savory, D. (2006). Use of a walk opportunities index to quantify local accessibility. Transportation Research Record,1977 145153. Lau, J., Ioannidis, J. P. A., Terrin, N., Schmid, C. H., & Olkin, I. (2006) The case of the misleading funnel plot. British Medical Journal, 333 (7568) 597600. Lauria, M., & Wagner, J. A. (2006). What can we learn from empirical studies of planning theory? A comparative case analysis of extant literature. Journal of Planning Education and Research 25 (4), 364381. Leck, E. (2006). The impact of urban form on travel behavior: A metaanalysis. Berkeley Planning Journal 19 3758. Lee, C., & Moudon, A. V. (2004). Physical activity and environment research in the health eld: Implications for urban and transportation planning practice and research. Journal of Planning Literature 19 (2), 147181. Lee, C., & Moudon, A. V. (2006a). Correlates of walking for transportation or recreation purposes. Journal of Physical Activity and Health 3 (1), 7798. Lee, C., & Moudon, A. V. (2006b). The 3Ds + R: Quantifying land use and urban form correlates of walking. Transportation Research Part D 11 (3), 204215. Li, F., Fisher, K. J., Brownson, R. C., & Bosworth, M. (2005). Multilevel modelling of built environment characteristics related to neighbourhood walking activity in older adults. Journal of Epidemiology and Community Health, 59 (7), 558564. Lipsey, M. W., & Wilson, D. B. (2001). Practical meta-analysis Thousand Oaks, CA: Sage. Littell, J. H., Corcoran, J., & Pillai, V. (2008). Systematic reviews and meta-analysis New York, NY: Oxford University Press. Livi Smith, A. D. (2009, January) Contribution of perception in analysis of walking behavior Paper presented at the 88th annual meeting of the Transportation Research Board, Washington, DC. Lund, H. M. (2003). Testing the claims of new urbanism: Local access, pedestrian travel, and neighboring behaviors. Journal of the American Planning Association 69 (4), 414429. Lund, H. M., Cervero, R., & Wilson, R. W. (2004). Travel characteristics of transit-oriented development in California Sacramento, CA: California Department of Transportation. Lund, H. M., Willson, R., & Cervero, R. (2006). A re-evaluation of travel behavior in California TODs. Journal of Architectural and Planning Research 23 (3), 247263. Lyons, L. (2003). Meta-analysis: Methods of accumulating results across research domains Retrieved June 26, 2009, from http://www.lyonsmorris. com/MetaA Macaskill, P., Walter, S. D., & Irwig, L. (2001). A comparison of methods to detect publication bias in meta-analysis. Statistics in Medicine 20 (4), 641654. McCormack, G., Giles-Corti, B., Lange, A., Smith, T., Martin, K., & Pikora, T. J. (2004). An update of recent evidence of the relationship between objective and self-report measures of the physical environment and physical activity behaviours. Journal of Science and Medicine in Sport 7 (1), 8192. McGinn, A. P., Evenson, K. E., Herring, A. H., Huston, S. L., & Rodriguez, D. A. (2007). Exploring associations between physical activity and perceived and objective measures of the built environment. Journal of Urban Health,84 (2), 162184. McMillan, T. E. (2005). Urban form and a child's trip to school: The current literature and a framework for future research. Journal of Planning Literature, 19 (4), 440456. McMillan T. E. (2007). The relative influence of urban form on a child's travel mode to school. Transportation Research A 41 (1), 6979. Melo, P. C., Graham, D. J., & Noland, R. B. (2009). A meta-analysis of estimates of urban agglomeration economies. Regional Science and Urban Economics 39 (3), 332342. Mokhtarian, P. L., & Cao, X. (2008). Examining the impacts of residential self-selection on travel behavior: A focus on methodologies. Transportation Research B,43 (3), 204228. Naess, P. (2005). Residential location affects travel behaviorBut how and why? The case of Copenhagen metropolitan area. Progress in Planning,63 (1), 167257. Newman, P. W. G., & Kenworthy, J. R. (2006). Urban design to reduce automobile dependence. Opolis: An International Journal of Suburban and Metropolitan Studies 2 (1), 3552. Nijkamp, P., & Pepping, G. (1998). A meta-analytical evaluation of sustainable city initiatives. Urban Studies 35 (9), 14811500. Oakes, J. M., Forsyth, A., & Schmitz, K. H. (2007). The effects of neighborhood density and street connectivity on walking behavior: The Twin Cities walking study. Epidemiologic Perspectives & Innovations 4 (16). doi:10.1186/1742-5573-4-16. Owen, N., Humpel, N., Leslie, E., Bauman, A., & Sallis, J. F. (2004). Understanding environmental influences on walking: Review and research agenda. American Journal of Preventive Medicine 27 (1), 6776. Pickrell, D., & Schimek, P. (1999). Growth in motor vehicle ownership and use: Evidence from the Nationwide Personal Transportation Survey. Journal of Transportation and Statistics, 2 (1),117. Plaut, P. O. (2005). Non-motorized commuting in the U.S. Transportation Research D, 10 (5), 347356. Pont, K., Ziviani, J., Wadley, D., Bennett, S., & Abbott, R. (2009). Environmental correlates of children's active transportation: A systematic literature review. Health & Place 15 (3), 827840. Pushkar, A. O., Hollingworth, B. J., & Miller, E. J. (2000, January). A multivariate regression model for estimating greenhouse gas emissions from alternative neighborhood designs Paper presented at the 79th annual meeting of the Transportation Research Board, Washington, DC. Rajamani, J., Bhat, C. R., Handy, S., Knaap, G., & Song, Y. (2003). Assessing the impact of urban form measures in nonwork trip mode 280Journal of the American Planning Association, Summer 2010, Vol. 76, No. 3 RJPA_A_477198.qxd 6/11/10 3:25 PM Page 280
choice after controlling for demographic and level-of-service effects. Transportation Research Record, 1831 158165. Reilly, M. K. (2002, January). Inuence of urban form and land use on mode choice: Evidence from the 1996 bay area travel survey Paper presented at the 81st annual meeting of the Transportation Research Board, Washington, DC. Rodenburg, R., Benjamin, A., de Roos, C., Meijer, A. M. & Stams, G. J. (2009). Efcacy of EMDR in children: A meta-analysis. Clinical Psychology Review, 29 (7), 599606. Rodriguez, D. A., & Joo, J. (2004). The relationship between nonmotorized mode choice and the local physical environment. Transportation Research D 9 (2), 151173. Rose, M. (2004). Neighborhood design and mode choice. (Unpublished doctoral dissertation.) Portland State University, Portland, OR. Rothstein, H. R., Sutton, A. J., & Borenstein, M. (2005). Publication bias in meta-analysis. In Rothstein, H. R., Sutton, A. J. & Borenstein, M. (Eds.). Publication bias in meta-analysis: Prevention, assessment and adjustment (pp.17). Chichester, UK: Wiley. Ryan, S., & Frank, L. D. (2009). Pedestrian environments and transit ridership. Journal of Public Transportation,12 (1), 3957. Saelens, B. E., & Handy, S. (2008). Built environment correlates of walking: A review. Medicine & Science in Sports & Exercise, 40(S), S550S567. Saelens, B. E., Sallis, J. F., & Frank, L. D. (2003). Environmental correlates of walking and cycling: Findings from the transportation, urban design, and planning literatures. Annals of Behavioral Medicine 25 (2), 8091. Salon, D. (2006). Cars and the city: An investigation of transportation and residential location choices in New York City Davis, CA: Agricultural and Resource Economics, University of California, Davis. Schimek, P. (1996). Household motor vehicle ownership and use: How much does residential density matter? Transportation Research Record 1552 120125. Schulze, R. (2004). Meta-analysis: A comparison of approaches Cambridge, MA: Hogrefe & Huber. Shadish, W. R., & Haddock, C. K. (1994). Combining estimates of effect size. In H. Cooper & L. V. Hedges (Eds.), The handbook of research synthesis (pp. 261280). New York, NY: Russell Sage Foundation. Shay, E., Fan, Y., Rodriguez, D. A., & Khattak, A. J. (2006). Drive or walk? Utilitarian trips within a neo-traditional neighborhood. Transportation Research Record 1985 154161. Shay, E., & Khattak, A. J. (2005). Auto ownership and use in neotraditional and conventional neighborhoods. Transportation Research Record 1902 1825. Shen, Q. (2000). Spatial and social dimensions of commuting. Journal of the American Planning Association,66 (1), 6882. Smith, V. K., & Huang, J. (1995). Can markets value air quality? A meta-analysis of hedonic property value models. Journal of Political Economy, 103 (1), 209227. Soltani, A., & Allan, A. (2006). Analyzing the impacts of microscale urban attributes on travel: Evidence from suburban Adelaide, Australia. Journal of Urban Planning and Development 132(3), 132137. Stamps, A. E. III. (1990). Use of photographs to simulate environments: A meta-analysis. Perceptual and Motor Skills,71 (3), 907913. Stamps, A. E. III. (1999). Demographic effects in environmental aesthetics: A meta-analysis. Journal of Planning Literature 14 (2), 155175. Stead, D., & Marshall, S. (2001). The relationships between urban form and travel patterns. An international review and evaluation. European Journal of Transport and Infrastructure Research,1 (2), 113141. Sun, X., Emany, M., Pendyala, R., & Wilmot, C. G. (1998). Household travel, household characteristics, and land-use: An empirical study from the 1994 Portland travel survey. Transportation Research Record, 1617 1017. Talvitie, A. (1976). Disaggregate travel demand models with disaggregate data, not with aggregate data, and for what (Working Paper 7615). Berkeley, CA: Institute of Transportation Studies. Targa, F., & Clifton, K. (2005). The built environment and trip generation for non-motorized travel. Journal of Transportation and Statistics, 8 (3), 5570. Tompa, E., de Oliveira, C., Dolinschi, R. & Irvin, E. (2008). A systematic review of disability management interventions with economic evaluations. Journal of Occupational Rehabilitation 18 (1), 1626. Train, K. (1986). Qualitative choice analysis: Theory, econometrics, and an application to automobile demand Cambridge, MA: The MIT Press. Transportation Research Board & Institute of Medicine Committee on Physical Activity, Health, Transportation, and Land Use. (2005). Does the built environment inuence physical activity? Examining the evidence (Special Report 282). Washington, DC: National Academy of Sciences. Treadwell, J. R., Tregear, S. J., Reston, J. T., & Turkelson, C. H. (2006). A system for rating the stability and strength of medical evidence. Medical Research Technology 6 (52). doi:10.1186/1471-2288-6-52. Trost, S. G., Owen, N., Bauman, A. E., Sallis, J. F., & Brown, W. (2002). Correlates of adults' participation in physical activity: Review and update. Medicine & Science in Sports & Exercise 34 (12), 19962001. van de Coevering, P., & Schwanen, T. (2006). Re-evaluating the impact of urban form on travel patterns in Europe and North-America. Transport Policy,13 (3), 229239. Walters, J., Ewing, R., & Allen, E. (2000). Adjusting computer modeling tools to capture effects of smart growth. Transportation Research Record 1722 1726. Waygood, E. O. D., Sun, Y., Kitamura, R. (2009, January). Children in transit oriented development: Travel patterns, independence, and exercise. Paper presented at the 88th annual meeting of the Transportation Research Board, Washington, DC. Yarlagadda, A. K., & Srinivasan, S. (2008). Modeling children's school travel mode and parental escort decisions. Transportation 35 (2), 201218. Zegras, P. C. (2007, January). The built environment and motor vehicle ownership and use: Evidence from Santiago de Chile Paper presented at the 86th annual meeting of the Transportation Research Board, Washington, DC. Zhang, M. (2004). The role of land use in travel mode choice: Evidence from Boston and Hong Kong. Journal of the American Planning Association 70 (3), 344361. Zhang, M. (2009, January). Bus vs. rail: A meta-analysis of cost characteristics, carrying capacities, and land use impacts Paper presented at the 88th annual meeting of the Transportation Research Board, Washington, DC. Zheng, Y. (2008). The benet of public transportation: Physical activity to reduce obesity and ecological footprint. Preventive Medicine 46 (1), 45. Zhou, B., & Kockelman, K. (2008). Self-selection in home choice: Use of treatment effects in evaluating relationship between built environment and travel behavior. Transportation Research Record,2077 5461. Ewing and Cervero: Travel and the Built Environment: A Meta-Analysis281 RJPA_A_477198.qxd 6/11/10 3:25 PM Page 281
282Journal of the American Planning Association, Summer 2010, Vol. 76, No. 3 Appendix:Individual Study Results Table A-1. Elasticity of VMT with respect to density. In metaStudy Nyxe analysis? Bhatia, 200420VMT per householdHousehold ÂŠ 0.34 Boarnet et al., 20046,153Nonwork VMT per personPopulation ÂŠ 0.04 Boarnet et al., 20046,153Nonwork VMT per personJob 0.03 Boarnet et al., 20046,153Nonwork VMT per personRetail job ÂŠ 0.02 Chatman, 200314,478VMT for commercial trips per personHousehold ÂŠ 0.58 Chatman, 200314,478VMT for commercial trips per personJob ÂŠ 0.34 Chatman, 2008527Nonwork VMT per personPopulation per road mile ÂŠ 1.05 Chatman, 2008527Nonwork VMT per personRetail job ÂŠ 0.19 ** Ewing et al., 1996 (Dade 1,311VHT per householdPopulation and employment ÂŠ 0.05 County) Ewing et al., 1996 (Palm 764VHT per householdPopulation and employment0.00 Beach County) Ewing et al., 20091,466VMT per householdPopulation 0.00y Ewing et al., 20091,466VMT per householdJob ÂŠ 0.06y Fan, 20077,422Miles traveled per personParcel ÂŠ 0.07 ** Frank & Engelke, 20054,552VMT per householdNet residential 0.00y Greenwald, 20093,938VMT per householdNet residential ÂŠ 0.07y Greenwald, 20093,938VMT per householdNet job 0.01y Hedel & Vance, 200728,901VKT per personCommercial ÂŠ 0.01 Holtzclaw et al., 2002314VMT per householdHousehold ÂŠ 0.14 (Chicago) Holtzclaw et al., 2002 (Los 1,459VMT per householdHousehold ÂŠ 0.11 Angeles) Holtzclaw et al., 2002 (San 1,047VMT per householdHousehold ÂŠ 0.14 Francisco) Kockelman, 19978,050VMT per householdPopulation 0.00y Kockelman, 19978,050VMT per householdJob0.00y Kuzmyak, 2009a5,926VMT per householdHousehold ÂŠ 0.04 **y Kuzmyak, 2009b3,615VMT per householdHousehold 0.00y Naess, 20051,414Weekday travel distance by car per personPopulation and employment 0.00 Pickrell & Schimek, 199940,000Miles driven per vehiclePopulation ÂŠ 0.06 ** Schimek, 199615,916VMT per householdPopulation ÂŠ 0.07y Sun et al., 19984,000VMT per householdJob 0.00y Zegras, 20074,279Daily automobile use per householdDwelling unit ÂŠ 0.04 **y Zhou & Kockelman, 20081,903VMT per householdPopulation ÂŠ 0.12 **y Zhou & Kockelman, 20081,903VMT per householdJob 0.02 y p < .10* p < .05** p < .01 RJPA_A_477198.qxd 6/11/10 3:25 PM Page 282
Ewing and Cervero: Travel and the Built Environment: A Meta-Analysis283 Table A-2. Elasticity of VMT with respect to diversity. In metaStudy Nyxe analysis? Bento et al., 20036,808VMT per householdJob-housing imbalance ÂŠ 0.06 ay Cervero & Kockelman, 1997896VMT per householdLand use dissimilarity0.00 Cervero & Kockelman, 1997896VMT per householdProportion vertical mix0.00 Cervero & Kockelman, 1997896VMT per householdProportion of population within 0.00 1/4 mile of store Chapman & Frank, 20048,592VMT per personLand use mix (entropy index) ÂŠ 0.04 **y Ewing et al., 1996 (Palm Beach 764VHT per householdJob-population balance ÂŠ 0.09 County) Ewing et al., 20091,466VMT per householdJob-population balance0.00y Fan, 20077,422Miles traveled per personRetail store count0.00 Frank & Engelke, 20054,552VMT per householdLand use mix (entropy index) ÂŠ 0.02 **y Frank et al., 20092,697VMT per householdLand use mix (entropy index) ÂŠ 0.04y Greenwald, 20093,938VMT per householdNon-retail job-housing balance0.03y Greenwald, 20093,938VMT per householdRetail job-housing balance ÂŠ 0.01y Greenwald, 20093,938VMT per householdJob mix (entropy index)0.01 Hedel & Vance, 200728,901VKT per personLand use mix (entropy index) ÂŠ 0.06y Kockelman, 19978,050VKT per householdLand use dissimilarity ÂŠ 0.10 ** Kockelman, 19978,050VKT per householdLand use mix (entropy index) ÂŠ 0.10 *y Kuzmyak et al., 20062,707VMT per householdLand use mix (entropy index) ÂŠ 0.09y Kuzmyak et al., 20062,707VMT per householdWalk opportunities within 1/2 ÂŠ 0.10 *y mile of home Kuzmyak, 2009a5,926VMT per householdLand use mix (entropy index) ÂŠ 0.27 **y Kuzmyak, 2009b3,615VMT per householdLand use mix (entropy index) ÂŠ 0.09 **y Pushkar et al., 2000795VKT per householdLand use mix (entropy index) ÂŠ 0.11 ** Sun et al., 19984,000VMT per householdLand use mix (entropy index) ÂŠ 0.10y Zegras, 20074,279Automobile use per householdLand use diversity ÂŠ 0.01 **y p < .10* p < .05** p < .01 Note: VKT is vehicle kilometers of travel. a. Sign reversed. RJPA_A_477198.qxd 6/11/10 3:25 PM Page 283
284Journal of the American Planning Association, Summer 2010, Vol. 76, No. 3 Table A-3. Elasticity of VMT with respect to design. In metaStudy Nyxe analysis? Bhat & Eluru, 20093,696VMT per householdBicycle lane density ÂŠ 0.08 ** Bhat, Sen, et al., 20098,107VMT per householdBicycle lane density ÂŠ 0.05 Bhat, Sen, et al., 20098,107VMT per householdStreet block density0.01 Boarnet et al., 20046,153Nonwork VMT per personIntersection density ÂŠ 0.19 ** Boarnet et al., 20046,153Nonwork VMT per personProportion 4-way intersections ÂŠ 0.06 Boarnet et al., 20046,153Nonwork VMT per personPedestrian environment factor0.05 Cervero & Kockelman, 1997896VMT per householdProportion 4-way intersections0.00y Cervero & Kockelman, 1997896VMT per householdProportion quadrilateral blocks0.19 ** Cervero & Kockelman, 1997896VMT per householdSidewalk width0.00 Cervero & Kockelman, 1997896VMT per householdProportion front and side parking0.00 Chapman & Frank, 20048,592VMT per personIntersection density ÂŠ 0.08 **y Chatman, 2008527Nonwork VMT per person4-way intersection density ÂŠ 0.06 Ewing et al., 20091,466VMT per householdIntersection density ÂŠ 0.31 *y Fan, 20077,422Miles traveled per personProportion connected intersections ÂŠ 0.11y Fan, 20077,422Miles traveled per personSidewalk length ÂŠ 0.02 Frank & Engelke, 20054,552VMT per householdIntersection density ÂŠ 0.10 **y Frank et al., 20092,697VMT per householdIntersection density ÂŠ 0.11 **y Greenwald, 20093,938VMT per householdIntersection density ÂŠ 0.29 **y Hedel & Vance, 200728,901VKT per personStreet density ÂŠ 0.04 *y Pushkar et al., 2000795VKT per householdIntersections per road km ÂŠ 0.04 Zegras, 20074,279Automobile use per householdProportion 3-way intersections ÂŠ 0.15 *ay Zegras, 20074,279Daily automobile use per householdPlaza density ÂŠ 0.03 p < .10* p < .05** p < .01 Note: a. Sign reversed. RJPA_A_477198.qxd 6/11/10 3:25 PM Page 284
Ewing and Cervero: Travel and the Built Environment: A Meta-Analysis285 Table A-4. Elasticity of VMT with respect to destination accessibility. In metaStudy Nyxe analysis? Bento et al., 20036,808VMT per householdPopulation centrality ÂŠ 0.15 ** Bhat & Eluru, 20093,696VMT per householdAccessibility to shopping ÂŠ 0.01 ** Bhatia, 200420VMT per householdJob/household accessibility by ÂŠ 0.19 transit Boarnet et al., 20046,153Nonwork VMT per personDistance to CBD ÂŠ 0.18 ** Cervero & Duncan, 200616,503Work VMT per personJob accessibility by auto ÂŠ 0.31 ** Cervero & Duncan, 200616,503Shopping VMT per personRetail job accessibility by auto ÂŠ 0.17 ** Cervero & Kockelman, 1997896VMT per householdJob accessibility by auto ÂŠ 0.27 **y Ewing et al., 1996 (Palm 764VHT per householdJob accessibility by auto ÂŠ 0.04 ** Beach County) Ewing et al., 1996 1,311VHT per householdJob accessibility by auto ÂŠ 0.15 ** (Dade County) Ewing et al., 20091,466VMT per householdJob accessibility by auto ÂŠ 0.03y Frank et al., 20092,697VMT per householdJob accessibility by transit ÂŠ 0.10 **y Greenwald, 20093,938VMT per householdJob accessibility by auto ÂŠ 0.06 **y Kockelman, 19978,050VMT per householdJob accessibility by auto ÂŠ 0.31 **y Kuzmyak et al., 20062,707VMT per householdJob accessibility by auto and transit ÂŠ 0.13 Kuzmyak, 2009a5,926VMT per householdJob accessibility by transit ÂŠ 0.04 **y Kuzmyak, 2009b3,615VMT per householdJob accessibility by transit ÂŠ 0.03 **y Naess, 20051,414Weekday travel distance by car Distance to downtown ÂŠ 0.27 **ay per person Pushkar et al., 2000795VKT per householdDistance to CBD ÂŠ 0.20 **aShen, 20003,565Average commute timeJob accessibility by auto and transit ÂŠ 0.18 Sun et al., 19984,000VMT per householdJob accessibility by auto ÂŠ 0.17 **y Sun et al., 19984,000VMT per householdHousehold accessibility by auto ÂŠ 0.34 ** Zegras, 20074,279Daily automobile use per householdDistance to CBD ÂŠ 0.20 **ay p < .10* p < .05** p < .01 Note: a. Sign reversed. RJPA_A_477198.qxd 6/11/10 3:25 PM Page 285
286Journal of the American Planning Association, Summer 2010, Vol. 76, No. 3 Table A-5. Elasticity of VMT with respect to transit access. In metaStudy Nyxe analysis? Bento et al., 20036,808VMT per householdDistance to transit stop ÂŠ 0.08 **aFrank & Engelke, 20054,546VMT per householdDistance to bus stop ÂŠ 0.01 ay Frank et al., 20092,697VMT per householdDistance to bus stop squared ÂŠ 0.04 **a,by Hedel & Vance, 200728,901VKT per individualWalk minutes to transit ÂŠ 0.02 ay Naess, 20051,414Weekday travel distance by car per personDistance to rail station ÂŠ 0.14 *ay Pushkar et al., 2000795VKT per householdDistance to transit station ÂŠ 0.03 **aZegras, 20074,279Daily automobile use per HouseholdDistance to Metro ÂŠ 0.19 **ay p < .10* p < .05** p < .01 Notes: a. Sign reversed. b. Sign reversed and multiplied by 2 to make x variable equivalent to others. Table A-6. Effect on VMTaof neighborhood type. In metaStudy Nyxe analysis? Bhat & Eluru, 20093,696VMT per householdUrban neighborhood ÂŠ 0.34 ** Cao, Xu, et al., 20093,376Vehicle miles driven per personUrban neighborhood ÂŠ 0.28 ** Cervero, 2007226Commute VMT per personTransit-oriented development ÂŠ 0.29 ** Khattak & Rodriguez, 2005302Daily miles traveled per householdNew urbanist neighborhood ÂŠ 0.20 Shay & Khattak, 2005399Auto VMT per householdNew urbanist neighborhood ÂŠ 0.22 p < .10* p < .05** p < .01 Note: a. Proportional reduction relative to conventional suburban neighborhood. RJPA_A_477198.qxd 6/11/10 3:25 PM Page 286
Ewing and Cervero: Travel and the Built Environment: A Meta-Analysis287 Table A-7. Elasticity of walk trips with respect to density. In metaStudy Nyxe analysis? Bhatia, 200420Walk trips per householdHousehold density0.83 ** Boarnet et al., 20086,362Miles walked per personPopulation density0.13 Boarnet et al., 20086,362Miles walked per personRetail job density0.07** Boarnet et al., 20086,362Miles walked per personJob density0.00 Boarnet et al., 20091,370Walk trips per personResidential density ÂŠ 0.50y Boarnet et al., 20091,370Walk trips per personBusiness density0.14 *y Boer et al., 200729,724Miles walked per personHousing density0.21 bChatman, 2009999Walk/bike trips per personPopulation per road mile0.16 Chatman, 2009999Walk/bike trips per personRetail job density0.00 Ewing et al., 2009 3,823Walk mode choicePopulation density0.01y Ewing et al., 2009 3,823Walk mode choiceJob density0.10y Fan, 2007988Daily walking time per personParcel density0.08 Frank et al., 20088,707Walk mode choice for work tripsRetail oor area ratio0.07 Frank et al., 200810,475Walk mode choice for other tripsRetail oor area ratio0.04 Frank et al., 20092,697Walk trips per householdRetail oor area ratio0.20 ** Frank et al., 20092,697Walk trips per householdNumber of retail parcels0.08 ** Greenwald & Boarnet, 20011,084Walk trips per person for nonwork Population density0.34 **ay purposes Greenwald & Boarnet, 20011,084Walk trips per person for nonwork Retail job density0.11 *apurposes Greenwald, 20093,938Walk/bike trips per householdResidential density0.28 **y Greenwald, 20093,938Walk/bike trips per householdJob density0.03y Hess et al., 199912Pedestrians per hourPopulation density1.39 Joh et al., 20092,125Walk trips per personNeighborhood business density0.19 ** Kockelman, 19978,050Walk/bike mode choicePopulation density0.00y Kockelman, 19978,050Walk/bike mode choiceJob density0.00y Naess, 20051,406Weekday travel distance by walk/bike Population + employment density0.00 per person Rajamani et al., 20032,500Walk mode choice for nonwork tripsPopulation density0.01y Reilly, 20027,604Walk mode choice for nonwork tripsPopulation density0.16 **y Targa & Clifton, 20052,934Walk trips per personHousehold density0.03y Zhang, 2004 (Boston)1,619Walk/bike mode choice for work tripsPopulation density0.11 *y Zhang, 2004 (Boston)1,619Walk/bike mode choice for work tripsJob density0.03 *y Zhang, 2004 (Boston)1,036Walk/bike mode choice for nonwork tripsPopulation density0.06 *y Zhang, 2004 (Boston)1,036Walk/bike mode choice for nonwork tripsJob density0.00y p < .10* p < .05** p < .01 Notes: a. Computed at median cutpoint by Jason Cao. b. Signicance level indeterminate. RJPA_A_477198.qxd 6/11/10 3:25 PM Page 287
288Journal of the American Planning Association, Summer 2010, Vol. 76, No. 3 Table A-8. Elasticity of walk trips with respect to diversity. In metaStudy Nyxe analysis? Bento et al., 20034,456Walk/bike mode choiceJob-housing imbalance0.30 *ay Boer et al., 200729,724Miles walked per personBusiness types in neighborhood0.20 bCao, Mokhtarian, et al., 2009b1,277Nonwork walk trips per personBusiness types within 400 meters0.07 ** Cao et al., 2006837Walk trips to store per personDistance to store0.56 **ay Cervero & Kockelman, 19972,850Non-person vehicle choice for Land use dissimilarity0.00 nonwork trips Cervero & Kockelman, 19972,850Non-person vehicle choice for Proportion vertical mix0.00 nonwork trips Cervero & Kockelman, 19972,850Non-person vehicle choice for Proportion of population within 0.00 nonwork trips1/4 mile of store Ewing et al., 2009 (Portland)3,823Walk mode choiceJob-population balance0.18y Frank et al., 20088,707Walk mode choice for work tripsLand use mix (entropy index)0.22 **y Frank et al., 200810,475Walk mode choice for other tripsLand use mix (entropy index)0.03 *y Frank et al., 20092,697Walk trips per householdLand use mix (entropy index)0.08 yGreenwald, 20093,938Walk/bike trips per householdNon-retail job-housing balance0.25 y Greenwald, 20093,938Walk/bike trips per householdRetail job-housing balance0.02y Greenwald, 20093,938Walk/bike trips per householdJob mix (entropy index)0.09 Handy & Clifton, 20011,368Walk trips to store per personDistance to nearest store0.48 **ay Handy et al., 20061,480Walk trips to store per person# Business types within 800m0.29 ** Handy et al., 20061,480Walk trips to store per personDistance to nearest grocery0.17 **ay Kitamura et al., 199714,639Fraction walk/bike tripsDistance to nearest park0.11 *aKockelman, 19978,050Walk/bike mode choiceLand use mix (entropy index)0.23 *y Rajamani et al., 20032,500Walk mode choice for nonwork tripsLand use mix (diversity index)0.36 *y Reilly, 20027,604Walk mode choice for nonwork tripsDistance to closest commercial use0.16 **ay Shay et al., 2006348Walk trips per householdDistance to commercial center0.98 **ay Targa & Clifton, 20052,934Walk trips per personLand use mix (entropy index)0.08 **y Zhang, 2004 (Boston)1,619Walk/bike mode choice for work tripsLand use mix (entropy index)0.00y Zhang, 2004 (Boston)1,036Walk/bike mode choice for nonwork tripsLand use mix (entropy index)0.12y p < .10* p < .05** p < .01 Notes: a. Sign reversed. b. Signicance level indeterminate. RJPA_A_477198.qxd 6/11/10 3:25 PM Page 288
Ewing and Cervero: Travel and the Built Environment: A Meta-Analysis289 Table A-9. Elasticity of walk trips with respect to design. In metaStudy Nyxe analysis? Boarnet et al., 20086,362Miles walked per personIntersection density0.45 ** Boarnet et al., 20086,362Miles walked per personPedestrian environment factor0.04 Boarnet et al., 20091,370Walk trips per personBlock size0.35 ay Boarnet et al., 20091,370Walk trips per person% 4-way intersections ÂŠ 0.09y Boer et al., 200729,724Miles walked per personProportion 4-way intersections0.39 dBoer et al., 200729,724Miles walked per personBlock length (long side) ÂŠ 0.31 a,dCervero & Kockelman, 19972,850Non-private vehicle choice for nonwork tripsProportion 4-way intersections0.00 Cervero & Kockelman, 19972,850Non-private vehicle choice for nonwork tripsProportion quadrilateral blocks0.00 Cervero & Kockelman, 19972,850Non-private vehicle choice for nonwork tripsSidewalk width0.09 Cervero & Kockelman, 19972,850Non-private vehicle choice for nonwork tripsProportion front and side parking0.12 **aChatman, 2009999Walk/bike trips per person4-way intersection density0.30 Ewing et al., 2009 3,823Walk mode choiceIntersection density0.43 **y Ewing et al., 2009 3,823Walk mode choiceSidewalk coverage0.27 **y Fan, 2007988Daily walking time per person% connected intersections0.40 ** Fan, 2007988Daily walking time per personSidewalk length0.12 ** Frank et al., 20088,707Walk mode choice for work tripsIntersection density0.21 **y Frank et al., 200810,475Walk mode choice for other tripsIntersection density0.28 **y Frank et al., 20092,697Walk trips per householdIntersection density0.55 **y Greenwald, 20093,938Walk/bike trips per householdIntersection density1.11 **y Greenwald & Boarnet, 20011,084Walk trips per person for nonwork purposesPedestrian environment factor0.25 bHess et al., 199912Pedestrians per hourBlock size0.35 **aJoh et al., 20092,125Walk trips per personBlock size0.01 ay Joh et al., 20092,125Walk trips per person% 4-way intersections ÂŠ 0.27y Rajamani et al., 20032,500Walk mode choice for nonwork trips% Culs-de-sac0.00 **cy Rodriguez & Joo, 2004448Walk mode choice for commute tripsSidewalk coverage1.23 ** Rodriguez & Joo, 2004448Walk mode choice for commute tripsPath directness0.03 Soltani & Allan, 20061,842Walk/bike mode choicePath directness0.11 Targa & Clifton, 20052,934Walk trips per personBlock size0.32 **ay Zhang, 2004 (Boston)1,619Walk/bike mode choice for work tripsStreet connectivity0.07 y Zhang, 2004 (Boston)1,036Walk/bike mode choice for nonwork tripsStreet connectivity0.05 p < 0.10* p < 0.05** p < 0.01 Notes: a. Sign reversed. b. Computed at the median cutpoint by Jason Cao. c. Because either the elasticity or signicance level must be misreported in the published article we dropped this observation from the meta-analysis. d. Signicance level indeterminate. RJPA_A_477198.qxd 6/11/10 3:25 PM Page 289
290Journal of the American Planning Association, Summer 2010, Vol. 76, No. 3 Table A-10. Elasticity of walk trips with respect to destination accessibility. In metaStudy Nyxe analysis? Bento et al., 20034,456Walk/bike mode choicePopulation centrality1.00 Boarnet et al., 20086,362Miles walked per personDistance to cbd0.49 **aCervero & Duncan, 20037,836Walk mode choiceJobs within one mile0.04y Cervero & Kockelman, 19972,850Non-person vehicle choice for nonwork tripsJob accessibility by auto0.00 Chatman, 2009999Walk/bike trips per personDistance to downtown0.29 aEwing et al., 2009 3,823Walk mode choiceJobs within one mile0.23 *y Greenwald, 20093,938Walk/bike trips per householdJob accessibility by auto ÂŠ 0.32 ** Kockelman, 19978,050Walk/bike mode choiceJob accessibility by walking0.22 **y Naess, 20051,406Weekday travel distance by walk/bike per personDistance to downtown0.29 **a p < .10* p < .05** p < .01 Note: a. Sign reversed. Table A-11. Elasticity of walk trips with respect to transit access. In metaStudy Nyxe analysis? Bento et al., 20034,456Walk/bike mode choiceDistance to nearest transit stop0.30 ay Boarnet et al., 20086,362Miles walked per personDistance to light rail ÂŠ 0.17 *aKitamura et al., 199714,639Fraction walk/bike tripsDistance to nearest bus stop0.10 *ay Naess, 20051,406Weekday travel distance by walk/bike per personDistance to closest rail station0.00 aRajamani et al., 20032,500Walk mode choice for nonwork trips% within walking distance of bus0.02 aTarga & Clifton, 20052,934Walk trips per personDistance to nearest bus stop0.08 **ay p < .10* p < .05** p < .01 Note: a. Sign reversed. Table A-12. Effect on walk tripsaof neighborhood type. In metaStudy Nyxe analysis? Cao, Mokhtarian, et al., 2009b1,277Nonwork walk trips per personTraditional neighborhood0.44 ** Handy & Clifton, 20011,368Walk trips to store per personTraditional neighborhood1.20 ** Khattak & Rodriguez, 2005302Walk trips per householdNew urbanist neighborhood3.06 ** Lund, 2003427Destination walk trips per personNeighborhood with retail0.38 ** Lund, 2003427Destination walk trips per personNeighborhood with retail and park0.85 ** Plaut, 200526,950Walk mode choice for commute tripsNeighborhood with retail0.79 ** Rose, 2004244Walk trips per personNew urbanist neighborhood0.35 p < .10* p < .05** p < .01 Note: a. Proportional increase relative to conventional neighborhood. RJPA_A_477198.qxd 6/11/10 3:25 PM Page 290
Ewing and Cervero: Travel and the Built Environment: A Meta-Analysis291 Table A-13. Elasticity of transit trips with respect to density. In metaStudy Nyxe analysis? Bhatia, 200420Transit trips per householdHousehold density0.37 Cervero, 2002a427Transit mode choiceGross population density0.39 *y Cervero, 2006225Weekday boardings per stationPopulation density0.19 ** Ewing et al., 2009 3,823Transit mode choicePopulation density ÂŠ 0.01y Ewing et al., 2009 3,823Transit mode choiceJob density0.08y Fan, 2007154Daily transit travel time per personParcel density0.00 Frank et al., 20088,707Transit mode choice for work tripsRetail oor area ratio0.21 **y Frank et al., 200810,475Transit mode choice for nonwork tripsRetail oor area ratio0.17 **y Greenwald, 20093,938Transit trips per householdNet residential density0.41 **y Greenwald, 20093,938Transit trips per householdNet job density ÂŠ 0.05 *y Kuby et al., 2004268Weekday boardings per stationPopulation within walking distance0.11 Kuby et al., 2004268Weekday boardings per stationEmployment within walking distance0.07 Rajamani et al., 20032,500Transit mode choice for nonwork tripsPopulation density0.08y Reilly, 20027,604Transit mode choice for nonwork tripsPopulation density0.20 *y Rodriguez & Joo, 2004454Transit mode choice for commute tripsPopulation density ÂŠ 0.20y Zhang, 2004 (Boston)1,619Transit mode choice for work tripsPopulation density0.12 *y Zhang, 2004 (Boston)1,036Transit mode choice for nonwork tripsPopulation density0.13 *y Zhang, 2004 (Boston)1,619Transit mode choice for work tripsJob density0.09 *y Zhang, 2004 (Boston)1,036Transit mode choice for nonwork tripsJob density0.00y Zhang, 2004 (Hong Kong)20,246Transit mode choice for work tripsPopulation density0.01y Zhang, 2004 (Hong Kong)15,281Transit mode choice for nonwork tripsPopulation density0.01 *y Zhang, 2004 (Hong Kong)20,246Transit mode choice for work tripsJob density0.01 **y Zhang, 2004 (Hong Kong)15,281Transit mode choice for nonwork tripsJob density0.01y p < .10* p < .05** p < .01 RJPA_A_477198.qxd 6/11/10 3:25 PM Page 291
292Journal of the American Planning Association, Summer 2010, Vol. 76, No. 3 Table A-14. Elasticity of transit trips with respect to diversity. In metaStudy Nyxe analysis? Bento et al., 20034,456Transit mode choiceJob-housing imbalance0.60 ay Cervero, 2002a427Transit mode choiceLand use mix (entropy index)0.53 *y Cervero & Kockelman, 19971,544Non-personal vehicle choice for work tripsLand use dissimilarity0.00 Cervero & Kockelman, 19971,544Non-personal vehicle choice for work tripsProportion vertical mix0.00 Cervero & Kockelman, 19971,544Non-personal vehicle choice for work tripsProportion of population 0.00 within 1/4 of store Fan, 2007154Daily transit travel time per personRetail store count ÂŠ 0.04 Frank et al., 20088,707Transit mode choice for work tripsLand use mix (entropy index)0.09 *y Frank et al., 200810,475Transit mode choice for nonwork tripsLand use mix (entropy index)0.19y Greenwald, 20093,938Transit trips per householdJob-housing balance0.23 *y Greenwald, 20093,938Transit trips per householdJob mix (entropy index)0.04 Kitamura et al., 199714,639Fraction transit tripsDistance to nearest park0.11 Rajamani et al., 20032,500Transit mode choice for nonwork tripsLand use mix (diversity index) ÂŠ 0.04y Reilly, 20027,604Transit mode choice for nonwork tripsDistance to closest commercial use ÂŠ 0.19 ** Zhang, 2004 (Boston)1,619Transit mode choice for work tripsLand use mix (entropy index)0.00y Zhang, 2004 (Boston)1,036Transit mode choice for nonwork tripsLand use mix (entropy index)0.12y p < .10* p < .05** p < .01 Note: a. Sign reversed. RJPA_A_477198.qxd 6/11/10 3:25 PM Page 292
Ewing and Cervero: Travel and the Built Environment: A Meta-Analysis293 Table A-15. Elasticity of transit trips with respect to design. In metaStudy Nyxe analysis? Cervero, 2002a427Transit mode choiceSidewalk ratio0.16 Cervero, 2007726Transit mode choice for work trips% 4-way intersections1.08y Cervero & Kockelman, 19971,544Non-personal vehicle choice for work tripsProportion front and side parking0.00 Cervero & Kockelman, 19971,544Non-personal vehicle choice for work tripsProportion 4-way intersections0.00 Cervero & Kockelman, 19971,544Non-personal vehicle choice for work tripsSidewalk width0.00 Cervero & Kockelman, 19971,544Non-personal vehicle choice for work tripsProportion quadrilateral blocks0.19 Fan, 2007154Daily transit travel time per person% connected intersections0.27 Fan, 2007154Daily transit travel time per personSidewalk length0.00 Frank et al., 20088,707Transit mode choice for work tripsIntersection density0.20 *y Frank et al., 200810,475Transit mode choice for nonwork tripsIntersection density0.24 y Frank et al., 20092,697Transit trips per householdIntersection density0.12y Greenwald, 20093,938Transit trips per householdIntersection density0.37 *y Lund et al., 2004967Transit mode choice % 4-way intersections at destination1.08 **y Rajamani et al., 20032,500Transit mode choice for nonwork trips% Culs-de-sac0.00 ay Rodriguez & Joo, 2004454Transit mode choice for commute tripsSidewalk coverage0.28 Rodriguez & Joo, 2004454Transit mode choice for commute tripsPath directness0.01 Zhang, 2004 (Boston)1,619Transit mode choice for work tripsStreet connectivity0.08 y Zhang, 2004 (Boston)1,036Transit mode choice for nonwork tripsStreet connectivity0.04y p < .10* p < .05** p < .01 Note: a. Sign reversed. Table A-16. Elasticity of transit trips with respect to destination accessibility. In metaStudy Nyxe analysis? Bento et al., 20034,456Transit mode choicePopulation centrality0.00 Cervero, 2006225Weekday boardings per stationDistance to CBD0.21 **aEwing et al., 2009 3,823Transit mode choiceJob accessibility by transit0.29 ** Frank et al., 20092,697Transit trips per householdJob accessibility by transit0.16 Greenwald, 20093,938Transit trips per householdJob accessibility by auto0.05 Kuby et al., 2004268Weekday boardings per stationAverage time to other stations0.95 **aLund et al., 2004967Transit mode choice Job accessibility by auto ÂŠ 0.70 ** p < .10* p < .05** p < .01 Note: a. Sign reversed. RJPA_A_477198.qxd 6/11/10 3:25 PM Page 293
294Journal of the American Planning Association, Summer 2010, Vol. 76, No. 3 Table A-17. Elasticity of transit trips with respect to transit access. In metaStudy Nyxe analysis? Bento et al., 20034,456Transit mode choiceDistance to transit stop1.00 ay Ewing et al., 2009 3,823Transit mode choiceBus stop density0.08 Frank et al., 20092,697Walk trips per householdDistance to bus stop squared0.02 by Kitamura et al., 199714,639Fraction transit tripsDistance to rail station0.13 **ay Rajamani et al., 20032,500Transit mode choice for nonwork trips% within walking distance of bus0.42 p < .10* p < .05** p < .01 Notes: a. Sign reversed. b. Sign reversed and multiplied by 2 to make x variable equivalent to others. Table A-18. Effect on transit tripsaof neighborhood type. In metaStudy Nyxe analysis? Rose, 2004244Transit trips per personNew urbanist neighborhood0.66 Note: a. Proportional increase relative to conventional neighborhood. RJPA_A_477198.qxd 6/11/10 3:25 PM Page 294
Copyright of Journal of the American Planning Association is the property of Routledge and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use.
xml version 1.0 encoding UTF-8
REPORT xmlns http:www.fcla.edudlsmddaitss xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.fcla.edudlsmddaitssdaitssReport.xsd
INGEST IEID ETWU88DA1_FX6Y8F INGEST_TIME 2014-10-03T22:10:11Z PACKAGE UFE0046816_00001
AGREEMENT_INFO ACCOUNT UF PROJECT UFDC
HOW THE BUILT ENVIRONMENT AFFECTS ELDERLY TRAVEL BEHAVIOR: AN ACTIVITY BASED APPROACH FOR SOUTHEAST FLORIDA By RUOYING XU A THESIS PRESENTED TO THE GRADUATE SCHOOL OF THE UNIVERSITY OF FLORIDA IN PARTIAL FULFILLMENT OF THE REQUIREMENTS FOR THE DEGREE OF MASTER OF ARTS IN URBAN AND REGIONAL PLANNING UNIVERSITY OF FLORIDA 2014
2014 Ruoying Xu
To m y f amily
4 ACKNOWLEDGMENTS I am deeply grateful to my committee chair, Dr. Ruth Steiner, for her guidance during my research. She provided me with her insights in transportation and land use interaction, and guide me through confusion s and doubts. I would also like to appreciate my co chair, Dr. Siva Srinivasan, for his guidance and help in my modelling techniques. He provided me with a lots of valuable informations and experiences in activity based model. Add itionally I want to thank Dr. Zhong Ren Peng for his advices during my research. I give my special appreciation to Roosbeh Nowrousian, a PhD student in civil engineering, who provides tremendous help to me through my research. I would also thank my friends and colleagues from both the Department of Urban and Regional Planning and Department of Civil and Coastal Engineering. It i s great to live, learn and cooperate with other Gators. This research is sponsored by the Eisenhower Transportation Fellowship prov ided by the Federal Highway Administration (FHWA) from the United States Department of Transportation. Here I thank them for their financial support in the last two years. Last but not least, I would like to thank my parents, who h a v e always been my most a r d e n t supporters. They give me strength and willingness to continue studying in transportation planning.
5 TABLE OF CONTENTS page ACKNOWLEDGMENTS ................................ ................................ ................................ .. 4 LIST OF TABLES ................................ ................................ ................................ ............ 7 LIST OF FIGURES ................................ ................................ ................................ .......... 9 ABSTRACT ................................ ................................ ................................ ................... 10 CHAPTER 1 INTRODUCTION ................................ ................................ ................................ .... 12 2 LITERATURE REVIEW ................................ ................................ .......................... 16 3 MODEL FRAMEWORK ................................ ................................ .......................... 19 4 DATA ................................ ................................ ................................ ...................... 21 The Tour Setting ................................ ................................ ................................ ..... 22 Control Variables ................................ ................................ ................................ .... 25 Measures of the Built Environment ................................ ................................ ......... 26 Street Connectivity ................................ ................................ ........................... 26 Accessibility to Transit ................................ ................................ ...................... 27 Regional Accessibility ................................ ................................ ....................... 27 Density ................................ ................................ ................................ ............. 28 Diversity ................................ ................................ ................................ ............ 28 Descriptive Analysis of Land Use Variables ................................ ........................... 29 Regional Context ................................ ................................ .............................. 29 Connectivity and Transit Accessibility ................................ .............................. 32 Mixed use ................................ ................................ ................................ ......... 34 5 MODEL RESULTS ................................ ................................ ................................ 36 Activity Generation ................................ ................................ ................................ .. 36 Tour Generation ................................ ................................ ................................ ...... 36 Tour Mode Choice ................................ ................................ ................................ .. 38 Street Connectivity ................................ ................................ ........................... 38 Transit Acce ssibility ................................ ................................ .......................... 39 Regional Accessibility ................................ ................................ ....................... 39 Density ................................ ................................ ................................ ............. 40 Diversity ................................ ................................ ................................ ............ 41 6 CONCLUSIONS ................................ ................................ ................................ ..... 42
6 7 SUGGESTIONS FOR FUTURE RESEARCH ................................ ......................... 45 REFERENCES ................................ ................................ ................................ .............. 51 BIOGRAPHICAL SKETCH ................................ ................................ ............................ 54
7 LIST OF TABLES Table page 4 1 Comparison between All Sample and Elderly Who Made Tours ........................ 23 4 2 Tour Purpose Distribution ................................ ................................ ................... 24 4 3 Tour Type Distribution ................................ ................................ ........................ 24 4 4 Distribution of Tour Mode ................................ ................................ ................... 25 4 5 Descriptive Analysis of Control Variables ................................ ........................... 26 4 6 Descriptive Analysis of the Built Environment Variables ................................ ..... 30 4 7 Distance to Regional Center for Each Tour Based on Different Tour Mode ....... 32 4 8 Block Level Density ar ound Household for Each Tour Based on Different Tour Mode ................................ ................................ ................................ .......... 32 4 9 Tract Level Density around Household for Each Tour Based on Differ ent Tour Mode ................................ ................................ ................................ .......... 32 4 10 Transit Accessibility around Household for Each Tour Based on Different Tour Mode ................................ ................................ ................................ .......... 33 4 11 No. of Intersections around Household for Each Tour Based on Different Tour Mode ................................ ................................ ................................ .......... 33 4 12 No. of Cul de sacs around Household for Each Tour Based on Different Tour Mode ................................ ................................ ................................ .................. 33 4 13 Connected Node Ratio around Household for Each Tour Based on Different Tour Mode ................................ ................................ ................................ .......... 34 4 14 Mixed Development Index around Household for Each Tour Based on Different Tour Mode ................................ ................................ ............................ 34 4 15 Entropy Index around Household for Each Tour Based on Different Tour Mode Motorized Mode and Transit. ................................ ................................ .... 35 5 1 Activity Generation Estimation Results (Binary Logit Model) .............................. 37 5 2 Tour Generation Estimation Results (Multinomial Logit Model). ......................... 47 5 3 Tour Mode Choice for Medical Tours Estimation Results (Multinomial Logit Model) ................................ ................................ ................................ ................ 48
8 5 4 Tour Mode Choice for Maintenance Tours E stimation Results (Multinomial Logit Model) ................................ ................................ ................................ ........ 49 5 5 Tour Mode Choice for Discretionary Tours Estimation Results (Multinomial Logit M odel) ................................ ................................ ................................ ........ 50
9 LIST OF FIGURES Figure page 3 1 Activity based Model Framework ................................ ................................ ........ 20 4 1 Location of Regional Activity Center in Southeast Florida ................................ .. 27
10 Abstract of Thesis Presented to the Graduate School of the University of Florida in Partial Fulfillment of the Requirements for the Degree of Master of Arts i n Urban and Regional Planning HOW THE BUILT ENVIRONMENT AFFECTS ELDERLY TRAVEL BEHAVIOR: AN ACTIVITY BASED APPROACH FOR SOUTHEAST FLORIDA By Ruoying Xu May 2014 Chair: Ruth Loraine Steiner Cochair: Sivaramakrishnan Srinivasan Major: Urban and Regional Planning Elderly travel behavior can potentia lly be shaped by changes in the built environment. However the debate over the connection of transportation and land use has not yet reached a consensus. In order to better understand the relationship between the built environment and elderly travel behavi or, this analysis adopt s an integrated activity based type approach to study the impact of the built environment represented by disaggregate land use characteristics on different levels of elderly travel decisions (activity generation, tour generation and tour based mode choice). Using data from National Household Travel Survey (NHTS) 2009 Florida Add on and GIS data from the Florida Geographic Data Library (FGDL), a case study i s presented on the Southeast Florida Region. The results show that different le vels of travel decisions are affected by different built environment factors. Employment density can encourage elderly travel. Better street connectivity increases the likelihood of travelers engaging in a simple tour, while living in a neighborhood with a n office area may result in less time constrained tour. Street connectivity, regional accessibility and transit accessibility are found to be correlated with elderly mode choice. This study provided a more
11 comprehensive interpretation of the travel patterns, and subsequent travel behavior and needs of the elderly.
12 CHAPTER 1 INTRODUCTION The rapid growth of population over 65 years old, usually referred to as elderly, is increasingly evident throughout the world. According to U.S. Department of Health and Human Services, the American elderly population is expected to reach 72.1 million in 2030, almost twice as large as in 2000 (Wan et al. 2005). Such growth can potentially bring about greater challenge s to t he transportation system, since elderly in the near future are likely to expect the same level of mobility as younger generation, which is higher than the expectation of current elderly (Buehler and Nobis 2010; Van den Berg, Arentze, and Timmermans 2011; K arimi et al. 2012). Previous research suggested that elderly w ill try to keep their auto ownership in order to retain their mobility (Rosenbloom 2001). Compounding this issue are consistent safety concerns that elderly are more likely to be involved in cra shes as drivers, despite their self regulation of the amount of driving (Giuliano, Hu, and Lee 2003; Hilderbrand 2003; Burkhardt 1999). As a result, a shift from an auto dependen t travel pattern is essential for the elderly drivers to maintain their mobili ty level. However research also found that elderly are less likely to rely on public transit, and are more likely to suffer serious or fatal injury in pedestrian crashes due to higher exposure rates (Giuliano, Hu, and Lee 2003). Therefore as individuals be came older, physical conditions and the lack of alternative s to automobile travel may hinder them from sustaining their expected level of mobility. Without proper mobility,
13 elderly may suffer from isolation and depression, thus compromise their general qua lity of life. Land use planning is increasingly used as a strategy to improve the viability of the alternatives to automobile (Handy 2005). Such policies are to utilize the interaction between transportation and land use which assumes that travel demand c an be shaped by urban development pattern. Since the elderly are more sensitive to local accessibility, it is expected that promoting more transit friendly, mixed use communities will be effective in improving elderly mobility (Giuliano, Hu, and Lee 2003). However previous findings raised more questions than answers about this issue. Some research suggested that higher density development can significantly increase elderly mobility level, and reduce the use of automobile by increasing the probability of wal king and cycling (Kim and Ulfarsson 2004; Mercado and Paez 2009; Van den Berg, Arentze, and Timmermans 2011; Sikder and Pinjari 2012). On the contrary, Oaks et al. (2007) concluded that the effects of density and block size on total walking and physical ac tivity are modest to non existent, if not contra positive to hypotheses. In general, previous literature does not reach a consensus on how and why the built environment affects elderly travel behavior. Traditionally two approaches have been taken to study the traffic impact of land use: to analyze trip generation, usually in terms of number of trips or the distance of travel; and to use discrete choice model to study mode choice or time of day choice at
14 trip level. However the accuracy of these methods has been questioned for the fact that travel pattern has become much more complicated since the introduction of these models. The availability of activity based model provides us with new opportunities to perform detailed analysis on how land use interacts wi th elderly travel behavior, particularly at a level involving activity engagement. Travel related choices, in activity based models, become part of the activity pattern and scheduling process. Such model will capture the demand for activity rather than dem and for trips. This provides us with a sound and viable approach to forecast travel demand since activity based model tours (Bhat and Koppelman 1999). The aim of this researc h is to analyze the effects of the built environment characteristics on different levels of travel decision making process using a simplified activity based type model system. Although this model framework is descriptive in nature, it has two advantages co mpared to traditional methodologies. First, the decision of travel in this analysis is layered based on a certain hierarchy: willingness to travel (activity generation), tour purpose split (tour generation), and mode choice at tour level. This model consid ers travel as a whole, starting from the activity generation, to tour generation and then to the travel decision at tour level. The impact of land use is tested in each level of the travel decision, thus making it clear how the built environment interacts with elderly travel behavior.
15 Second, a tour level mode choice set was created based on the assumption of the trip level. This setting implies the availability of different modes when making travel decision. For example, the tour mode is au to means that a car is available in all trips in the same tour. This method takes into account the interaction between trips in the same tour and avoids duplication of the same subjects traveling on a sequence of trips in the same tour using the same mode, thus providing us with an accurate interpretation of travel behavior. The State of Florida ha s the highest percentage (17.3%) of elderly people among all states in the United States (Himes 2002). The analysis will use southeast Florida as a case study, an d combine various data source from National Household Travel Survey (NHTS) 2009 Florida Add on for household and trip related data, and GIS data by county from Florida Geographic Data Library (FGDL). By using this data sources we intended to construct a st atistically efficient model to find out how much the built environment affect elderly travel behavior.
16 CHAPTER 2 LITERATURE REVIEW Much of the research on urban form and elderly travel behavior has been focused on whether mixed used, higher density co mmunity can increase physical activity and the use of transit by elderly (Giuliano, Hu, and Lee 2003; Kim and Ulfarsson 2004; Mercado and Paez 2009; Van den Berg, Arentze, and Timmermans 2011; Sikder and Pinjari 2012). The majority of the research has adop ted a traditional method of evaluating trip generation and trip level travel decision. These methodologies are generally in consistent with studies that analyze land use and general travel behavior. Similar to the findings of elderly travel behavior resear ch, the results of these studies that address the coordination of transportation and land use bring about more questions. Many researchers suggested that higher density development can significantly affect travel behavior (Cervero and Seskin 1995; Cevero a nd Kockelman 1997; Kitamuram Mokhtarian and Laidet 1997, Steiner et al., 2008). Other studies, which adopted similar methodologies as those mentioned above, showed that the built environment does not have a significant impact on travel demand and travel be havior (Giuliano 1995; Crane and Crepeau 1998; Boarnet and Sarmineto 1998). In general, we can conclude that previous literature does not reach a consensus on how and why the built environment affects travel demand and travel behavior.
17 Recent improvements in the research of elderly travel behavior have included studying mobility level and preference of elderly (Siren and Blomqvist 2009; Sikder and Pinjari 2012), and using multiple correspondence analysis to study nonlinear and non monotonic relationships b etween socioeconomic characteristics and elderly travel behavior at a trip chain level (Golob and Hensher 2007). There has been little research on how the built environment impact s elderly travel behavior, particularly the travel behavior involving activity generation. It is clear that further research on this topic is required to better understand the connection between elderly travel behavior and the built environment. Travel dem and modeling has made significant advances in the past 35 years. Discrete choice modeling techniques were first developed in order to study the choice of travelers on a trip based scenario (McFadden and Talvitie 1977; Ben Akiva and Lerman 1985). This metho dology was further developed into tour based models that capture the interrelated decision making in a trip chain (Daly, van Zwam, and van der Valk 1983; Gunn 1994). Later on, activity based modeling concept s were developed in order to report the constrain ts of activity schedule and important activity based demand responses (Ben Akiva, Bowman, and Gopinath 1996; Bowman and Ben Akiva 2000). Activity based model can capture the subtle impact of explanatory variables to the travel decisions on different layers Such characteristics give us opportunities to implement this methodology into the analysis of land use policies (Shiftan 2008).
18 The analysis presented in this paper incorporates an activity based approach into the study of land use and elderly travel beh avior interaction. It is different from previous study in the following ways. First, it takes into account that travel decision is not made solely based on a specific trip, instead people consider travel as a whole including willingness, purpose and mode c hoice together. Second, tour generation is represented by a binary model studying what aspects of land use can encourage time unconstrained tours, given that the increase in the proportion of unconstrained tours implies better elderly mobility level. Third this approach considers mode choice at a tour level instead of trip level, considering that mode choice for a chain of trips with in the same tour usually interrelates with each other. As such, the incorporation of activity based type model can potentiall y lead us into further understanding of how the built environment affects elderly travel behavior, thus bring new insights into the pool of current literature.
19 CHAPTER 3 MODEL FRAMEWORK The model framework in this study is an activity based type model system consists of a series of disaggregate logit models. Travel decisions are classified into three levels based on a hierarchy of decision making process: activity generation (willingness to travel), tour generation (purpose and complexity of tour), and tour level mode choice. Lower level choices are conditional on the decision of higher level. shows the diagram of the model framework. The underlying basis for such a model framework is that travel decisions on activity generation level are driven by the need of the travelers. Therefore activity generation is considered to be the highest level of the analysis hierarchy. For tour level decisi on such as mode choice, they tend to be driven by convenience, travel conditions, and short term temporal constraints. Therefore they are positioned at a lower level of the hierarchy. For each level of travel decision, a discrete choice analysis is conduc ted to estimate the effects of land use on travel decision. At activity generation level, a binary logit model is fit in order to find what affects elderly willingness to travel. At tour generation level, we choose a binary logit model that reports what f actors would encourage the elderly to make more time unconstrained travel over a regression model for studying how many trips or tours elderly person produced. The underlying logic is that higher proportion of unconstrained tours in the travel schedule of the elderly
20 person indicates mobility level, since tours with time constraints, such as work, school or medical, are always necessary regardless of the built environment. Higher proportion of complex tours means elderly is likely to plan their travel befor e hand as a compromise to their constraints. After controlling socioeconomic factors, the models are intended to explore the degree of association between multiple dimensions of land use and elderly travel behavior. Figure 3 1 Activity based M odel F ramework For tour based mode choice, a multinomial logit model is fit for medical, maintenance and discretionary tour purpose. Mandatory purpose is not estimated here since travel decisions on mandatory tours are generally inelasti c, and elderly are less likely to conduct mandatory travel compared to younger generation.
21 CHAPTER 4 D ATA This analysis will be conducted by using primary data from National Household Travel Survey (NHTS) 2009 Florida Add on. The case study area includes three counties from Southeast Florida which are Palm Beach, Broward and Miami Dade. The other component of the dataset is the parcel data from FGDL. In total 2557 households 2747 persons are in the data sample. The NHTS 2009 dataset is collected on daily trips taken in a 24 hour period. The purpose of NHTS 2009 is for researcher to have a better understanding of travel behavior. The design of NHTS 2009 data is a random digital dialing telephone interview survey conducted over an entire year (FHWA 2011). The dataset includes socio economic and trip related information at a household and person level, and information perception of the transportation system. In the add on samples such as Florida Add on data, O D information for all trips are included and it allows us to geocode the location of the respondents and study the relations between travel behavior and the built environment. The Florida Geography Data Library (FGDL ) provides a rich pool of GIS based data of the land use and built environment information in Florida. We utilize the parcel data from 2010, roadway data, and transit data for the analysis. The primary reason for
22 picking south Florida as the case study is that south Florida is a very urbanized area and it has a rich variety of land use patterns and different activity patterns. This can provide us with a significant diversity in the sample for the analysis. The Tour Setting Tours constitute a fundamental un it of analysis in activity based and tour based travel demand modeling systems (Nowrouzian and Srinivasan 2012). In this analysis, a tour is a sequence of trips that begin from home and return to home after one or more intermediate stops with none of them being home, therefore all tours studied here are home based tours. In the NHTS 2009 Florida Add on, trips are categorized into 36 different purposes, and then we combine each trips into tours based on their specific travel information. The following trips were excluded from the process of tour creation: Traveler did not start and/or end the survey day at home Traveler made one tour with no intermediate stops (i.e., a trip that starts and ends at home) At least one person made at least one tour with more th an six intermediate trips Based on the above criteria, we have 2099 out of 2747 elderly persons who generate at least one tour. Table 4 1 data. These 2099 persons produced 3130 observations of tours. Note t hat for activity generation, we still conduct our analysis with the complete sample.
23 Table 4 1 Comparison between A ll S ample and E lderly W ho M ade T ours Age group All Sample (Including Stay at Home) Elderly Who Made One or More Tours Frequency Percentage Frequency Percentage 65 69 669 24.4 567 27.0 70 74 591 21.5 476 22.7 75 79 566 20.6 451 21.5 80 84 520 18.9 374 17.8 85+ 401 14.6 231 11.0 Total 2747 100.0 2099 100.0 The 2099 elderly generated 3130 tours in total. Tour generation and tour level mode choice are conducted based on these 3130 tours. Tour mode is defined based on the most important activity of the tour, i.e., the purpose of a tour is the highest priority activity taking place in the tour. The most important activity is determined based on pre defined hierarchy considering flexibility in frequency, location, and scheduling of the activities. The lesser the flexibility of an activity in frequency, locatio n, scheduling, the higher is it s priority in hierarchy. The activity purposes are ranked in order as: mandatory (work, school or school related, pick up or drop off), medical, maintenance (shopping, eating out, etc.) and discretionary (social/recreational, exercise, etc.). Mandatory and medical tours are consid ered to be time constrained, while maintenance and discretionary tours are considered to be unconstrained. Table 4 2 shows the frequency and percentage of each trip purpose. The complexity of tours are define d by the number of stops in each tour. If a tour contains two stops, the tour is defined as simple tour. If the tour contains at least three stops, it is defined as complex tours. Table 4 3 shows the distribution of tour type
24 Table 4 2 Tour P urpose D istribution Tour generation Tour purpose Definition Frequency Percent Time constrained tour mandatory Work, School, escort, etc 215 6.9 medical go to doctor, dentist, etc 354 11.3 Time unconstrained tour maintenance Shopping, eating out, etc. 1331 42.5 discretionary social/recreational, exercise, etc. 1230 39.3 Total 3130 100.0 Table 4 3 Tour T ype D istribution Tour Type Frequency Percent Simple tour with time constraints 285 9.1 Simple tour without time constraints 1717 54.9 Complex tour with time constraints 284 9.1 Complex tour without time constraints 844 27.0 Total 3130 100.0 The travel mode of the entire tour is determined to be one of the following based on the modes of the individual trips in the tour and the vehicle occupancy levels: Drive Alone, Shared Ride 2, Shared Ride 3+, Non motorized, and Walk Transit. If all trips w ithin the tour are made by Auto, the tour mode is first broadly classified as auto. The tour level auto occupancy is then determined based on the maximum number of participants on the trip that occur within the tour. Based on the tour level auto occupancy, auto tours are further classified into Drive Alone, Shared Ride2, and Shared Ride 3+. If the mode for all trips in a tour is Walk or Bike, the tour mode is respectively defined as Non motorized. To complete a tour, if both Transit and Auto are used, tour mode is classified as Drive Transit. If Transit is the only mode to make a tour, tour mode
25 is defines a Walk Transit. We do not have any drive transit in our sample. shows distribution of tour mode Table 4 4 Distribution of T our M ode Tour mode All Mandatory Medical Maintenance Discretionary Freq. % Freq. % Freq. % Freq. % Freq. % Drive alone 1374 43.9 177 82.3 151 42.7 686 51.5 360 29.3 Share ride 2 999 31.9 24 11.2 144 40.7 401 30.1 430 35.0 Share ride 3+ 317 10.1 14 6.5 22 6.2 96 7.2 189 15.4 Non motorized 370 11.8 121 9.1 249 20.4 Walk transit 70 2.2 37 10.5 27 2.0 Total 3130 100.0 215 100 354 100 1331 100 1230 100 Literature expected the trip chaining behavior to increase in general as a population ages, due to the increasingly ageing society which a large portion of the population being 65 years or older, who are less constrained when undertaking single purpose commuting activity (Golob and Hensher 2007). The literature also suggested the growth in more active lifestyles of seniors and their ability through trip chaining to meet multiple objectives in one tour (Banister and Bowling 2004 ). Control Variables Socioeconomic characteristics and some travel information are used as control variables in this analysis. presents the de scriptive analysis of all control variables. Since medical conditions are an important aspect of elderly travel, we include two physical condition variables here: driving impaired, and mobility impaired. Driving impaired means a person has a medical condit ion that makes driving hard, and
26 mobility impaired means a person has medical condition that makes travel in general difficult. Table 4 5 Descriptive A nalysis of C ontrol V ariables Variables Definition Activity Generation Tour Generation and Mode Choice Mean Std. Dev. Mean Std. Dev. weekend Travel take place on weekend (dummy) .29 .454 .27 .443 age Age of subject 75.99 7.492 74.82 6.99 male Subject is male (dummy) .43 .495 .47 .499 employed Subject is employed (dummy) .15 .357 .18 .384 medium income Medium income household (dummy) .25 .436 .27 .445 high income High income household (dummy) .25 .434 .30 .457 no. of driver Number of drivers in household 1.58 .768 1.64 .70 no. of vehicle Number of vehicles in household 1.47 .871 1.55 .851 no. of adult Number of adults in household 1.91 .788 1.87 .753 driving impaired Medical condition which makes driving difficult (dummy) .11 .312 .06 .230 mobility impaired Medical condition which makes travel difficult (dummy) .26 .436 .18 .383 driver license Subject has a driver license (dummy) .81 .393 .91 .292 live alone Subject live alone (dummy) .26 .441 .27 .445 Measures of the Built Environment A variety of measures of the built environment are tested in this analysis. These variables are calculated using GIS technology which allows us to measure built environment variables with different scales. Densit ies are calculated based on census block and census tract, while other variables, such as connectivity and land use mix, are calculated by 0.25 miles buffer and 0.5 miles buffer around household locations. shows the descriptive analysis of all built environment v ariables.
27 Street Connectivity In this analysis, street connectivity is represented in terms of number of intersections and number of cul de sacs in a certain buffer area. Additionally, connected node ratio (CNR) is used to represent the overall connectivit y of local network. CNR is the number of street intersections divided by the number of intersections plus cul de sacs. The maximum value is 1.0. Accessibility to Transit Accessibility to transit is an important indicator of the viability of transit as an a lternative to automobile. In this analysis, three measures of accessibility to transit are tested: network distance to nearest bus stop, number of bus stops in a certain buffer area and total length of bus route in a certain buffer area. Regional Accessib ility In this analysis, the measure of regional accessibility determines the network distance of each neighborhood to each of four regional activity centers in southeast Florida ( ). The activity centers were defined as neighborhoods with the highest commercial square footage (Steiner et al. 2008). The distances were determined between the household locations to regiona l center along the roadway network.
28 Figure 4 1 Location of R egional A ctivity C enter in Southeast Florida Density Density is a common measure of the built environment in the literature which explores the interaction between urban form and travel behavior. Higher density usually implies better accessibility, higher proportion of mixed use area, and better transit servi ces. In this analysis, net jobs density, net residential density and net population density are tested. Diversity Land use types are divided into the following six categories: residential, commercial, office, institutional, industrial, and other. The firs t set of the land use variables captures the fraction of a certain area by each land use type. These variables represent the diversity of land use pattern around residential location. The next set of variables is the fraction of area that is developed, ca lculated as the ratio of the sum of the areas in the six land use categories (residential, commercial,
29 office, institutional, industrial and other) to the total buffer area around the neighborhood. For neighborhood located near the coast line, the total bu ffer area is smaller. Another set of variables is the entropy index around neighborhood. Entropy index is used to define the land use balance based on local or zonal characteristics (Kockelman, 1997). The equation for entropy is as follows. ( 4 1) Where Pj is the proportion of developed land in the j th land use type; in this analysis, J =6. The last set of variables is the mixed development index (MDI). It is a variable that characterizes the job housing balance. The definition of MDI is as follows. MDI=[ (ED)*(RD)]/[ED+RD ( 4 2) Where RD is Residential Density, and ED is Employment Density. Descriptive A nalysis of L and U se V ariables Regional Context The tables in this section only show the descriptive analysis on travel made by elderly. Younger travelers are excluded from the analysis. As we can see from T able 4 7 the distance to activity center for non motorized, and walk transit are shorter than other modes. For distance to nearest residential center, walk transit has the lowest value.
30 From table 4 8 we can see a potential correlation between density and walk transit. It is clear from this table that the higher the density, the more likely for elderly people to use transit.
31 Table 4 6 Descriptive A nalysis of the B uilt E nvironment V ariables Category Variables Activity Generation Tour Generation and Mode Choice Mean Std. Dev. Mean Std. Dev. Street Connectivity No. of intersections in .25 mile buffer 29.71 15.38 28.83 15.21 No. of cul de sacs in .25 mile buffer 6.21 6.49 6.22 6.33 No. of intersections in .5 mile buffer 115.3 48.71 112.44 47.40 No. of cul de sacs in .5 mile buffer 24.32 18.92 24.57 18.72 No. of intersections in 1 mile buffer 433.3 160.47 423.15 154.08 No. of cul de sacs in 1 mile buffer 88.29 51.55 88.99 50.73 CNR .25 mile buffer 0.82 0.15 0.81 0.15 CNR .5 mile buffer 0.82 0.11 0.81 0.11 CNR 1 mile buffer 0.82 0.09 0.82 0.08 Transit Accessibility number of bus station in 1 mile buffer 38.91 44.97 36.08 42.28 number of bus station in 0.5 mile buffer 10.35 13.32 9.40 12.53 Distance to nearest bus stop (1000 meters) 1.24 2.03 1.25 1.91 Total length of bus route in 0.5 mile buffer (1000 meter) 11.46 15.74 10.41 14.62 Regional Accessibility Distance to Regional Activity center (miles) 10.74 4.86 10.73 4.60 Density net job density blk (1000/sq mile) 2.69 3.19 2.53 3.04 net house density block level (1000/sq mile) 3.23 4.56 3.20 4.63 net population density block level (1000/sq mile) 6.45 7.74 6.16 7.53 net population density tract level (1000/sq mile) 6.23 4.87 5.96 4.76 net job density tract level (1000/sq mile) 2.28 1.75 2.15 1.69 net house density tract level (1000/sq mile) 3.27 3.57 3.24 3.70 net job density block group level (1000/sq mile) 2.42 2.19 2.24 2.06
3 2 Table 4 6. Continued Category Variables Activity Generation Tour Generation and Mode Choice Mean Std. Dev. Mean Std. Dev. Diversity Mixed Development Index block level (1000) 1.21 1.43 1.17 1.41 Mixed Development Index tract level (1000) 1.25 1.04 1.20 1.03 Fraction of .25 buffer area that is developed 0.46 0.22 0.46 0.21 Fraction of .25 buffer area that is residential 0.59 0.30 0.59 0.30 Fraction of .25 buffer area that is commercial 0.06 0.12 0.06 0.13 Fraction of .25 buffer area that is office 0.03 0.07 0.03 0.07 Fraction of .25 buffer area that is institutional 0.15 0.20 0.14 0.20 Fraction of .25 buffer area that is industrial 0.02 0.09 0.03 0.10 Entropy Index in .25 mile buffer 0.39 0.24 0.38 0.24 Fraction of .5 buffer area that is developed 0.41 0.20 0.41 0.21 Fraction of .5 buffer area that is residential 0.39 0.22 0.39 0.22 Fraction of .5 buffer area that is commercial 0.10 0.11 0.10 0.11 Fraction of .5 buffer area that is office 0.04 0.06 0.04 0.06 Fraction of .5 buffer area that is institutional 0.23 0.19 0.22 0.19 Fraction of .5 buffer area that is industrial 0.03 0.07 0.03 0.08 Entropy Index in .5 mile buffer 0.41 0.24 0.40 0.25
33 Table 4 7 Distance to R egional C enter for E ach T our B ased on D ifferent T our M ode distance to activity center miles distance to residential center miles Statistic Std. Error Statistic Std. Error Drive Alone 10.81 0.12 9.13 0.14 Shared Ride2 10.81 0.15 8.79 0.17 Shared Ride3 11.22 0.27 9.10 0.30 Non motorized 10.28 0.24 9.19 0.26 walk transit 8.96 0.55 6.69 0.52 Table 4 8 Block L evel D ensity around H ousehold for E ach T our B ased on D ifferent T our M ode net job density at blk level per square mile net house density at blk level per square mile net pop density at blk level per square mile Statistic Std. Error Statistic Std. Error Statistic Std. Error Drive Alone 2521.56 90.42 3454.65 137.57 6660.94 229.85 Shared Ride2 2406.57 91.64 3309.21 150.08 6277.71 237.97 Shared Ride3 2677.43 199.97 3566.39 343.21 6793.63 468.26 Non motorized 2558.05 185.57 3269.37 234.86 6880.10 475.61 walk transit 3999.11 540.99 5899.66 1123.57 9317.43 1280.47 Table 4 9 Tract L evel D ensity around H ousehold for E ach T our B ased on D ifferent T our M ode net job density at trct level per square mile net pop density at trct level per square mile net house density at trct level per square mile Statistic Std. Error Statistic Std. Error Statistic Std. Error Drive Alone 2163.23 44.36 5969.09 125.60 3212.33 95.71 Shared Ride2 2001.10 49.56 5527.84 137.34 2968.93 102.85 Shared Ride3 2350.95 113.47 6525.16 311.41 3575.70 236.76 Non motorized 2181.75 89.02 6089.20 267.57 3250.85 208.30 walk transit 3099.59 266.24 8839.78 654.87 6198.09 757.70 Similarly to density at the block level, walk transit has the potential correlation with density at tract level. Connectivity and T ransit A ccessibility The second set of land use variables is the street connectivity and transit accessibility.
34 Table 4 10 Transit A ccessibility around H ousehold for E ach T our B ased on D ifferent T our M ode number of bus station in 1 mile buffer dist to nearest bus stop meter NUMBER OF BUS STOP IN HALF A MILE total bus route length in half mile buffer Statistic Std. Error Statistic Std. Error Statistic Std. Error Statistic Std. Error Drive Alone 33.08 1.10 1328.70 53.92 8.59 0.33 9672.32 386.80 Shared Ride2 34.76 1.28 1208.04 58.99 9.20 0.39 9558.33 404.54 Shared Ride3 38.59 2.42 1353.53 118.37 9.57 0.69 10825.74 832.58 Non motorized 42.14 2.39 1049.26 81.80 10.85 0.68 12044.87 797.34 walk transit 70.21 6.34 742.25 150.40 19.74 1.82 26676.95 3095.40 We can see a smooth variation from auto (drive alone, carpool) to non motorized and finally to transit. This is a strong indication that transit accessibility is highly correlated with mode choice. Table 4 11 No. of I ntersections around H ousehold for E ach T our B ased on D ifferent T our M ode no. of intersections Buffer 0.25 miles Buffer 0.5 miles Buffer 1 mile Statistic Std. Error Statistic Std. Error Statistic Std. Error Drive Alone 28.84 .414 112.37 1.263 422.61 4.033 Shared Ride2 28.08 .466 110.20 1.458 416.25 4.811 Shared Ride3 29.19 .823 114.66 2.755 434.25 9.069 Non motorized 30.22 .804 115.31 2.447 431.08 8.243 walk transit 36.61 1.855 141.54 6.039 500.86 20.182 From T able 4 11 we can see the correlation between model choice and number of intersections around household location. Table 4 12 No. of Cul de sacs around H ousehold for E ach T our B ased on D ifferent T our M ode no. of culdesac 0.25 miles 0.5 miles 1 mile Statistic Std. Error Statistic Std. Error Statistic Std. Error Drive Alone 6.35 .165 24.87 .481 88.72 1.305 Shared Ride2 6.20 .197 24.56 .575 91.54 1.604 Shared Ride3 6.34 .362 25.68 1.086 92.53 3.109 Non motorized 5.92 .352 23.11 .983 84.54 2.596 walk transit 6.24 1.124 26.00 4.127 80.03 8.715
35 The number of cul de sacs around residential location, on the other hand, has a totally opposite effect compared to the number of intersections. Travelers tend to choose automobile more as the number of Cul de sacs increases. Table 4 13 Connected Node R atio around H ousehold for E ach T our B ased on D ifferent T our M ode connected Node ratio 0.25 miles 0.5 miles 1 mile Statistic Std. Error Statistic Std. Error Statistic Std. Error Drive Alone .8038 .00412 .8117 .00281 .8203 .00219 Shared Ride2 .8063 .00482 .8107 .00349 .8130 .00271 Shared Ride3 .8161 .00678 .8095 .00570 .8162 .00496 Non motorized .8237 .00825 .8281 .00544 .8276 .00455 walk transit .8706 .01389 .8592 .01305 .8662 .01091 Not surprisingly, the trend found in this table coincides with the finding in the number of intersections. People leaning towards non motorized mode and transit as the increase of connected road ratio. Mixed use The next set of land use variables are the mixed use variables. Table 4 14 Mixed D evelopment I ndex around H ousehold for E ach T our B ased on D ifferent T our M ode mixed development index tract level block level Statistic Std. Error Statistic Std. Error Drive Alone 1199.30 26.65 1149.46 37.24 Shared Ride2 1118.22 29.54 1147.13 47.63 Shared Ride3 1312.85 71.09 1207.92 87.21 Non motorized 1210.92 55.39 1137.84 73.82 walk transit 1890.84 179.47 1735.34 288.64 Mixed use always considered to be correlated with choice over non motorized mode and transit. Form T able 4 14 and Table 4 15 we can clearly see the trend.
36 Table 4 15 Entropy I ndex around H ousehold for E ach T our B ased on D iffere nt T our M ode M otorized M ode and T ransit. Similarly to mixed development index, entropy increases with the choice of non entropy index 0.25 Statistic Std. Error Drive Alone .380 .006 Shared Ride2 .369 .007 Shared Ride3 .376 .013 Non motorized .407 .013 walk transit .470 .028
37 CHAPTER 5 MODEL RESULTS Activity Generation Table 5 1 reports the binary logit model that estimates activity generation. The dependant variable is whether elderly travel (1) or stay at home (0) at travel day. Controlling for other correlates, the results show that higher net job density and higher MDI is associated with a higher likelihood of travel, which indicates that higher density and mixed use is likely to encourage elderly to conduct activity outdoors. Other variables tested in this model have the expected signs. Proximity to regional activity center will increase the probability of travel, potentially due to the increased number of opportunity near the activity center. Age, travel day (weekend or not), driver license, auto ownership all have impacts on the decision to trav el. It is worth noticing that mobility impaired elderly are much less likely to travel according to the significance and magnitude of the estimation results. We can conclude that medical conditions of elderly are one of the most important determinants of t ravel. Tour Generation Table 5 2 presents the second stage of the model that focused on tour generation. This model uses a binary logit model to estimate elderly travel decision over time constrained tour or unconstrained tour. Those respondents in the sur vey who did not report making a tour are eliminated from this analysis. Constrained tours are usually for things that have to be done, including mandatory tours (work, school, and escort)
38 and medical tours; while unconstrained tours include maintenance tou r (shopping, eating out) and discretionary tours (socio recreational, exercise). Table 5 1 Activity G eneration E stimation R esults (Binary Logit M odel) Base case: stay at home Base Model Final Model Variable Coef. Std. Err. Coef. Std. Err. Constant 3.202 ** .596 3.517** .633 employed .939 ** .214 .933** .214 age .029 ** .007 .032** .007 weekend .383 ** .105 .376** .106 Mobility Impaired .503 ** .111 .511** .112 Driver License 1.044 ** .128 .961** .130 no. of car .190 ** .077 .233** .081 no. of adult .318 ** .078 .323** .079 Distance to Activity Center .028** .013 net Job Density tract level (1000/mile 2 ) .455** .103 MDI tract level (1000unit) .746** .182 Pseudo R 2 0.127 0.133 Log likelihood 1585.150 2526.969 Sample size 2747 *:significant at 90% confidence level; **: significant at 95% confidence level Better transit accessibility around residence increase s the likelihood of engaging in a time unconstrained tour, while living in a neighborhood with much diversity in land use area may also result in more time unconstrained tour. This finding may partly be attribute d to the fact that elderly people who lived in a business area are likely to find more opportuni ties in the neighborhood T hus the number of unconstrained tours produced by these elderly people is higher. The results show here are good indicators that mixed use s with better accessibility can potentially encourage elderly people to travel, since the more simple tour and unconstrained tours are related to a more relaxed life style.
39 Tour Mode Choice For each tour purpose, namely medical, maintenance and discretionary, a multinomial logit model is estimated for tour mode choice ( Table 5 3 toTable 5 5 ). A ll four models have a significant increase in Pseudo R 2 compared to their base model, indicating that the inclusion of built environment factors strengthens the models, substantially reduced the unexplained variation in different dependent variables. Drive alone is used for base case in all four models in order to test the viability of alternatives to drive alone. The availability of modes varies according to different tour purposes. In maintenance tours, both non motorized mode and transit are properly rep resented in the sample, however elderly only choose transit as an alternatives to driving for medical tours, and non motorized travel is the only options other than automobile in discretionary tours. Street Connectivity Street connectivity is found to be highly correlated with the choice for non motorized and transit. Connected node ratio (CNR), in particular, affects tour level mode choice in both significance and magnitude. Better street connectivity would increase the likelihood of choosing non motorize d and transit for medical, maintenance, and discretionary tours. This finding is consisten t with previous literature that walking is found to be most strongly related to measures of intersection density (Ewing et al. 2010).
40 One interesting finding is that street connectivity seems to affect walking and biking more than transit. In Table 5 3 Table 5 4 and Table 5 5 we can see that the parameters of connected node ratio ( CNR ) for non motorized in both 0.5 miles buffer and 1 mile buffer area appears to be sig nificant, and much greater than the parameters for walk transit, implying that the choice of non motorized mode is much susceptible to street connectivity. Transit Accessibility Not surprisingly, we found that better transit accessibility is likely to inc rease transit use, however among all the transit accessibility variables we estimate for this analysis, the number of bus stops within a certain buffer area of residence turns out to be the most significant indicator, in some cases it is the only significa nt transit related variable. Although the estimation results show the expected sign of transit parameter, the magnitude is small compared to other sets of variables such as street connectivity. Other variables, such as total bus route length, are not signi ficant, implying that the convenience to reach a station is more important than the availability of different bus routes around neighborhood. Regional Accessibility The d istance to a regional activity center is correlated with elderly mode choice when con ducting maintenance tours. Longer distance s to activity center s would decrease the utility of travel by non motorized mode and transit. Residential areas that
41 are farther away from an activity center are usually suburban communities with single land use. T herefore distance to activity center implies the land use type of the residential neighborhood, thus affects the decision of mode choice. Longer distance to activity center would also increase the probability of carpool ing for the same reason. Density The modeling results show that density has a relatively small effect on elderly mode choice, as previous literature suggest s (Ewing et al. 2010). Both significance level and the magnitude of the parameters of density are moderate to marginal compared to other variables. The weak relationship between density and mode choice indicates that density has an intermediate connection between travel behavior and other built environment variables such as accessibility and connectivity. Although the impact of density on travel behavior is small, we cannot ignore the existence of this influence. The model results show that for tours without time constraints (maintenance and discretionary), density variables seems to have a greater impact o n mode choice compared to medical tours. For maintenance tour, five density variables turned out to be significant, including job density, house density and population density etc. This facts may have a profound implication suggesting that elderly are more likely to consider the effect of the built environment when conducting unconstrained travel.
42 Diversity Diversity can shape elderly mode choice in a larger scale, in terms of significance level and magnitude, compared to density. For medical, maintenance a nd discretionary tours, higher entropy or higher percentage of commercial and office area tends to promote non motorized mode and transit. The e ntropy index calculated for smaller buffer areas (0.25 miles) seems to be more relevant than the entropy of larg er buffer areas (0.5 miles), therefore we can conclude that the diversity of land use matters to mode choice only in a smaller spatial area. Job housing balance ( represented by MDI), however, does not increase the viability of non motorized mode and transi t, even given the relatively larger sample size in maintenance tours and discretionary tours.
43 CHAPTER 6 CONCLUSIONS Based on the results of the models, we can conclude that the built environment has greater impact on the tour level travel decision tha n on activity generation and tour generation. The demand for activity is still largely driven by the socioeconomic characteristics, while the travel behavior at lower level of activity engagement, such as tour level mode choice, is heavily affected by the built environment. However, the effect of built environment on activity generation and tour generation cannot be underestimated. Although density is found to be moderately correlates with mode choice as previous research suggested (Ewing et al. 2010), it i mpacts activity generation in a greater significance level. According to the parameter estimated for job density in activity generation impli cates that the increase of 1000 employees per square mile will increases the utility of elderly travel in a magnitude that larger than most of the socioeconomic variables. This finding suggests that density is significantly related to travel not by affecti ng trip length or mode choice, but by shaping the decision of elderly on whether to travel or not. The diversity of land use around residential neighborhood area can increase the proportion of tours without time constraints. Larger business area can poten tially bring new opportunities and conveniences for elderly, thus increase their unconstrained travel. This conclusion is based on the assumption that constrained travels are for mandatory
44 or medical purpose, therefore the increase in the proportion of unc onstrained travel means the increase in the total number of travel in general. Job housing balance ( represented by MDI) is found to be significant in activity generation, implying that diversity of land use is also strongly related with elderly willingness housing balance can significantly increase walk and transit use (Cervero et al. 2006) Entropy index, on the other hand, is found to be correlated with choosing non motorized mod es and transit, although this analysis also finds that only calculated in a smaller area can entropy become significant. Street connectivity and regional accessibility are highly correlated with elderly tour level mode choice as previous research predicted (Ewing et al. 2010). Street connectivity has the largest effect on elderly tour level mode choice, and it has larger impact on non motorized mode than on transit. The findings on transit accessibility, suggest that access to transit stations, rather than the availability of multiple transit routes, is the primary reason which elderly choose transit. In another word, if elderly can get to a transit station easily, they are more likely to use transit regardless how many transit routes are available for them Another interesting finding is that we do not find any consistency in how the built environment affects carpool. Most indicators turned out to be insignificant, while those significant sometimes reports contradictory results. This may imply that the bui lt
45 of transit. The choice of carpool is largely determined by the need of travel, the purpose of travel, and availability of companions.
46 CHAPTER 7 SUGGESTIONS F OR FUTURE RESEARCH This analysis adopted an activity based model system which allows us to examine the effects of land use on different layers of travel decision. One improvement of this model system is to incorporate a logsum term which makes the model mo re of a layers of travel decision together, and choices at lower level are conditioned on choices at higher level, while choices at higher level also reflect the choices at lower level. We can also improve the viability of the methodology by introducing a more specific classification on tour purpose. In this analysis, tour purposes are only divided into tours with time constraints and without time constraints. Using a more complete activity based model system can increase the accuracy of the results, which are likely to report the effects of land use on conducting different purposes of travel. This research shows that elderly travel behavior is correlated with several built environment factors. However, for elderly population, the lack of mobility is usually caused by their physical conditions. Potential longitudinal research can be conducted to study how the elderly travel behavior changes through time when they become incre asing mobility impaired and how these may vary across different neighborhood type.
47 Another limitation of this study is that it does not control for self selection. Tour based analysis requires each tour to be started and ended at home, therefore making ana lysis based on destination location difficult. One possible solution is to use interaction terms which connect demographic variables with land use variables in order to examine who locates in mixed use development area and why. This method can potentially give some insights for the self selection issue.
48 Table 5 2 Tour G eneration E stimation R esults (Multinomial L ogit M odel) Base case: tour with time constraint Base Model Final Model Variable Coef. Std. Err. Coef. Std. Err. Constant 1.984 ** .139 1.422 ** .301 Weekend 1.874 ** .164 1.814 ** .166 Male .258 ** .105 .253 ** .106 Employed 2.065 ** .121 2.078 ** .123 High income .350 ** .116 .252 ** .119 No. of adult .203 ** .060 .235 ** .063 Mobility Impaired .665 ** .129 .661* .131 No. of Bus stop in .5 mile buffer .013 ** .004 Fraction of .25 buffer area that is office 1.336* .786 Fraction of .25 buffer area that is commercial .696 ** .252 Fraction of .5 buffer area that is institutional 1.059* .612 Pseudo R 2 0.154 0.167 Log likelihood 305.914 2049.807 Sample size 2099 *:significant at 90% confidence level; **: significant at 95% confidence level
49 Table 5 3 Tour M ode C hoice for M edical T ours E stimation R esults (Multinomial L ogit M odel) For medical tour Shareride 2 Shareride 3+ Walk transit Base case is Drive alone Base Model Final Model Base Model Final Model Base Model Final Model Variable Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Constant 1.626 .664 2.693 3.485 1.274 .913 .116 6.911 1.505 1.646 34.720 13.251 No. of driver 1.299** .413 1.777** .489 1.803** .625 4.349** 1.001 1.710** .552 1.392* .723 No. of vehicle .669** .252 .742** .291 .560 .466 .230 .572 3.008** .594 3.955** .833 No. of adult .949** .359 1.731** .495 1.729** .453 3.119** .726 1.460** .717 2.777** .934 Mobility Impaired 1.139** .313 1.262** .353 .444 .564 .064 .790 1.005* .526 1.365** .683 Live alone 2.070** .475 1.560** .569 .971 .752 .800 1.049 1.427 1.048 .376 1.234 Street Connectivity No. of intersection in .25 mile buffer .008 .013 .064** .031 .041 .026 No. of cul de sacs in .5 mile buffer .012 .017 .088* .046 .110** .055 CNR 1 mile buffer .849 3.805 3.536 7.313 39.096** 13.769 Density Net population density tract level .443** .143 .805** .364 .268 .260 Net house density tract level .268** .131 .585 .365 .323 .207 Net job density block group level .092 .142 .584 .385 .466** .220 Diversity MDI tract level (1000unit) 2.552** .805 5.611** 2.150 1.201 1.603 Fraction of .25 buffer area that is developed .463 .851 2.304 1.834 5.907** 2.203 Fraction of .25 buffer area that is office 5.748* 2.997 5.414 4.946 16.772** 7.160 Entropy Index .25 buffer .081 .859 1.182 1.951 6.053** 2.144 Fraction of .5 buffer area that is residential .727 .878 5.957** 2.061 2.360 1.773 Fraction of .5 buffer area that is commercial 2.350** 1.571 9.344** 3.093 6.224 3.949 Pseudo R 2 0.208 0.398 Log likelihood 190.356 452.196 Sample size 354 *:significant at 90% confidence level; **: significant at 95% confidence level
50 Table 5 4 Tour M ode C hoice for M aintenance T ours E stimation R esults (Multinomial L ogit M odel) For maintenance tour Shareride 2 Shareride 3+ Non motorized Walk transit Base case is Drive alone Base Model Final Model Base Model Final Model Base Model Final Model Base Model Final Model Variable Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Constant 4.009 .821 1.125 1.991 1.868 .864 .637 3.177 1.255 .894 1.107 3.192 1.971 1.566 8.027 13.228 Male .091 .142 .167 .163 .540** .251 .721** .305 .125 .210 .050 .250 .884 .548 2.590** 1.153 h igh income .376** .166 .408** .198 .145 .285 .789** .379 .349 .241 .421 .313 20.169 0.001 21.616 0.001 N o. of driver .656** .209 .651** .253 .372 .329 .553 .430 .776** .331 .074 .417 .647 .648 .850 1.249 N o. of vehicle .636** .123 .635** .139 .581** .208 .607** .255 .235 .166 .056 .187 3.215** .539 3.178** .946 N o. of adult .135 .198 .129 .210 .383* .224 .634** .292 .016 .271 .244 .330 .491 .638 1.147* .698 D rive Impaired 1.717** .587 1.392** .623 2.052** .647 1.575** .731 .228 .828 .091 .887 1.721** .746 2.412* 1.285 D river License 4.039** .762 4.118** .783 4.064** .819 4.169** .878 4.480** .827 3.811** .886 3.271** 1.081 3.104* 1.587 l ive alone 1.768** .262 1.964** .301 .969** .373 .210 .450 .663** .323 .425 .417 .704 .839 1.007 1.157 S treet Connectivity N o. of cul de sacs in .25 mile buffer .031 .023 .074 .046 .037 .032 .298** .122 N o. of cul de sacs in .5 mile buffer .008 .014 .006 .026 .061** .022 .097 .074 N o. of cul de sacs in 1 mile buffer .005 .004 .009 .008 .016** .007 .010 .033 C NR .5 mile buffer 3.289 2.039 3.819 3.557 11.347** 3.751 1.163 15.262 CNR 1 mile buffer 1.422 2.809 4.224 5.159 12.420** 4.542 2.504 23.241 T ransit Accessibility N o. of bus stops in 1 mile buffer .004 .003 .004 .004 .003 .004 .043** .017 R egional Accessibility D istance to Activity Center .021 .024 .057 .044 .031 .034 .313** .154 D ensity Net house density block level .031 .070 .025 .106 .250** .077 .323 .235 N et population density block level .017 .042 .016 .071 .169** .050 .270 .219 D iversity MDI tract level (1000unit) 1.474** .656 .532 1.103 1.962* .837 1.664 2.330 F raction of .25 buffer area that is residential 1.833** .614 1.826* 1.067 1.858* 1.093 6.359* 3.538 F raction of .25 buffer area that is office 2.269 1.702 6.599 4.314 5.783** 1.982 8.833 7.784 Fraction of .5 buffer area that is residential .643 .652 4.648** 1.204 .864 .937 7.037* 3.907 Pseudo R 2 0.139 0.238 Log likelihood 769.401 1845.398 Sample size 1331 *:significant at 90% confidence level; **: significant at 95% confidence level
51 Table 5 5 Tour M ode C hoice for D iscretionary T ours E stimation R esults (Multinomial L ogit M odel) For discretionary tour Shareride 2 Shareride 3+ Non motorized Base case is Drive alone Base Model Final Model Base Model Final Model Base Model Final Model Variable Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Coef. Std. Err. Constant 2.664 .759 6.003 1.492 1.349 .797 4.729 1.757 2.170 .809 4.927 1.692 weekend .256* .157 .298* .165 .083 .195 .005 .208 .507** .196 .589** .210 Male .155 .155 .210 .163 .595** .197 .677** .206 .229 .182 .195 .194 No. of driver .185 .300 .374 .315 .459 .352 .471 .378 .663** .328 .793** .348 No. of vehicle .169 .118 .164 .126 .143 .149 .085 .155 .432** .150 .403** .157 No. of adult .069 .305 .245 .324 .669** .336 .679* .360 .722** .328 .864** .352 Mobility Impaired .363 .229 .306 .249 .399 .264 .476* .285 .122 .268 .173 .293 Driver License 1.729** .603 1.826** .675 1.932 .633 2.102** .706 2.057** .616 2.272** .690 live alone 1.825** .322 1.747** .342 .738** .357 .718* .382 .518 .353 .353 .379 Street Connectivity No. of cul de sacs in .25 mile buffer .046** .021 .023 .026 .047* .026 No. of cul de sacs in .5 mile buffer .023** .008 .010 .010 .025** .010 CNR 1 mile buffer 4.068** 1.533 4.409** 1.900 3.283* 1.780 Transit Accessibility No. of bus stops in 1 mile buffer .009** .003 .006 .004 .008** .003 Density Net house density block level .093 .061 .086 .069 .238** .088 Net population density block level .062* .036 .057 .041 .080* .045 Net population density tract level .120 .095 .235** .115 .142 .110 Net job density tract level .149 .180 .437** .219 .077 .220 Net house density tract level .125 .079 .151* .089 .199** .089 Diversity Fraction of .25 buffer area that is commercial .804 .720 1.986** 1.013 .613 .841 Fraction of .5 buffer area that is commercial .473 .904 2.495** 1.023 1.057 1.004 Fraction of .5 buffer area that is office .362 1.741 2.295 1.986 4.683** 1.743 Pseudo R 2 0.065 0.095 Log likelihood 798.977 2550.096 Sample size 1230 :significant at 90% confidence level; **: significant at 95% confidence level
52 REFERENCES Banister, D., & Bowling A. ( 2004 ) Quality o f Life f or t he Elderly: The Transport Dimension. Transport Policy 11 105 115. Ben Akiva, M. & Lerman S. R. ( 1985 ) Discrete Choice Analysis: Theory a nd Application t o Travel Demand Cambridge, MA: MIT Press. Ben Akiva, M. & Bowman J. ( 1998 ) Activity Based Travel Demand Model System. Equilibrium Advanced Transportation Modeling 27 46. Ben Akiva, M., Bowman J., & Gopinath D. ( 1996 ) Travel D emand M odel S ystem f or t he I nformation E ra. Transportation 23 241 266. Bhat, C.R. & Koppelman F. S. ( 1999 ) A Retrospective a nd Prospective Survey o f Time Use Research. Transportation 26(2) 119 139. Boarnet, M.G., & Sarmineto S. ( 1998 ) Can Land Use Policy Really Affect Travel Behavior? Urban Studies 35(7) 1155 1169. Bowman, J.L., & Ben Akiva M. ( 2000 ) Activity Based Disaggregate Travel Demand System w ith Activity Schedules. Transportation Research Part A 35 1 2 8. Buehler, R. & Nobis C. ( 2010 ) Travel Behavior i n Aging Societies: C omparison o f Germany And The United States. Transportation Research Record: Journal o f The Transportation Research Board 2182 62 70. Burkhardt, J. E. ( 1999 ) Mobility Changes: Their Nature, Effects, a nd Meaning f or Elders Who Reduce o r Cease Driving. Transportation Research Record 1671 11 19. Cervero, R., & Kockelman K. ( 1997 ) Travel Demand a nd The 3Ds: Density, Diversity, a nd Design. Transportation Ressear ch Part D 2(3) 199 219. Cervero, R., & Seskin S. ( 1995 ) An Evaluation Of The Relationship Between Transit a nd Urban Form. TCRP Research Results Digest No. 7. Crane, R., & Crepeau R. ( 1998 ) Does Neighborhood Design Influence Travel? A Behavioral Analys is o f Travel Diary a nd GIS Data. Transportation Research Part D: Transportation Environment 3 225 238. Daly, A.J., H. H. P. Van Zwam, & Van Der Valk J. ( 1983 ) Application o f D isaggregate M odels f or A R egional T ransport S tudy i n The Netherlands. World Conference o n Transport Research Ewing, R., & Cervero R. ( 2010 ) Travel a nd The B uilt E nvironment. Journal o f The American Planning Association 76(3) 265 294. Federal Highway Administration. ( 2011 ) Guide.
53 Giuliano, G. ( 1995 ) The Weakening Transportation Land Use Connection. Access 6 3 11. G iuliano, G., H. Hu, & Lee K. ( 2003 ) Travel Patterns o f The Elderly: The Role o f Land Use (No. FHWA/CA/OR 2003/06). METRANS Transportation Center. Golob, T., & Hensher D. A. ( 2007 ) The Trip Chaining Activity o f Sydney Residents: A Cross Section Assessment b y Age Group w ith A Focus o n Seniors. Journal o f Transport Geography 15 298 312. Handy, S. ( 2005 ) Smart G rowth a nd t he T ransportation L and U se C onnection: W hat D oes t he R esearch T ell U s? International Regional Science Review 28(2) 146 167. Hilderbrand E.D. ( 2003 ) Dimensions i n Elderly Travel Behavior: A Simplified Activity Based Model Using Lifestyle Clusters. Transportation 30: 285 306. Himes, C.L. ( 2002 ) Elderly Americans. Population Bulletin 56(4 ). Karimi, B., T. H. Rashidi, A. Mohammadian, & St urm K. ( 2012 ) Young Old Elderly a nd Baby Boomers: E xplanatory Analysis o f Activity Duration, Time o f Day Choice, a nd Planning Time Horizons. Transportation Research Record: Journal o f The Transportation Research Board 2322 51 59. Kim, S. & Ulfarsson G. F. ( 2004 ) Travel Mode Choice o f The Elderly: E ffects o f Personal, Household, Neighborhood, a nd Trip Characteristics. Transportation Research Record: Journal o f The Transportation Research Board 1894 117 126. Kitamura, R., P. Mokhtarian, & Laidet L. ( 1 997 ) A Micro Analysis o f Land Use a nd Travel i n Five Neighborhoods i n t he San Francisco Bay Area. Transportation 24(2) 125 158. Kockelman, K.M. ( 1997 ) Travel Behavior a s Function of Accessibility, Land Use Mixing, a nd Land Use Balance: Evidence f rom San Francisco Bay Area Transportation Research Record: Journal o f t he Transportation Research Board 1607 116 125. Mcfadden, D., & Talvitie A. ( 1977 ) Demand Model Estimatio n a nd Validation Vol. 5 Institute f or Transportation Studies, University o f California, Berkeley. Mercado, R. & Paez A. ( 2009 ) Determinants o f Distance Traveled w ith a Focus o n t he Elderly: A Multilevel Analysis i n t he Hamilton CMA, Canada. Journal o f T ransport Geography 17 65 76. Nowrouzian, R & Srinivasan S. ( 2012 ) Empirical Analysis o f Spatial Transferability o f Tour Generation Models. Transportation Research Record: Journal o f t he Transportation Research Board 2302 14 22.
54 Oaks, J. M., A. Forsyth, & Schmitz K. H. ( 2007 ) The E ffects o f N eighborhood D ensity a nd S treet C onnectivity o n W alking B ehavior: The Twin Cities W alking S tudy. Epidemiologic Perspectives & Innovations 2007 4 :16. Rosenbloom, S. ( 2001 ) Sustainability a nd Automobility a mong t he El derly: A n International Assessment. Transportation 28 (4) 375 408. Shiftan, Y. ( 2008 ) The Use o f Activity Based Modeling t o Analyze t he Effect o f Land Use Policies o n Travel Behavior. The Annals o f Regional Science 42 79 97. Sikder, S., & Pinjari P. R. ( 2012 ) Immobility Levels a nd Mobility Preferences o f t he Elderly i n t he United States: E vidence f rom The 2009 National Household Travel Survey. Transportation Research Record: Journal o f t he Transportation Research Board 2318 137 1 47 Siren, A & Blomqvist L. H. ( 2009 ) Mobility a nd Well Being i n Old Age. Topics i n Geriatric Rehabilitation 25 (1) 3 11. Steiner, R., S. Srinivasan, R. Provost, J. Mackey, A. Arafat, N. Anderson, & Delarco L (2008). VMT Based Traffic Impact Assessment: Development o f a Trip L ength Model (CMS Project No. 2008 007). Van Den Berg, P., T. Arentze, & Timmermans H. ( 2011 ) Estimating Social Travel Demand o f Senior Citizens i n t he Netherlands. Journal o f Transport Geography 19 323 331. Wan, H., M. Sengupta, V. A. Velkoff, & Debarros K. A. ( 2005 ) 65+ i n t he United States: 2005. U.S. Census Bureau Current Population Report 23 209.
55 BIOGRAPHICAL SKETCH The author graduated fro m Tongji University in Shanghai with a bachelor degree in transportation engineering, and then attended the University of Florida as a graduate research assistant working with Dr. Ruth Steiner. He graduated with a Master of Science in c ivil e ngineering and a Master of Arts in u rban and r egional p lanning in the University of Florida in 2014 elderly travel behavior: an activity based approach worked with other research topics such as impact of land use on VMT, travel behavior with High occupancy/ t olled lanes, pedestrian access range to transit station, etc.