Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
TCRP H-l ~ Final Report APPENDIX A THE TRANSPORTATIONS - LAND-USE INTERACTION EMPIRICAL FINDINGS . This Appendix was prepared with He assistance of Dr. Dame! A. Badoe. .: .
TCRP H-17 Final Report A.1 EMPIRICAL STUDIES ON THE IMPACT OF URBAN FORM ON TRAVEL BEHAVIOR A.. Introduction The review of the literature dealing with urban fonn impacts on travel behavior presented in this section is divided into four sub-sections. The first two deal with "real-worId" observations concerning urban form impacts on usage of transit and other vehicular modes (Section A. ~ .2), and - alk mode usage (Section A.I.31. Section A.!.4 then reviews simulation studies of urban form impacts on travel behavior. Section A.. concludes this portion of the review with a brief discussion of some common problems and identified directions for further research which emerge from this review-. A.~.2 Urban Form Impacts on the Use of Transit and Other Vehicular Modes PBQD tI99~] provide a very comprehensive review of the important studies that investigated the relationships between transit and urban form conducted over the last three decades. The key domestic U.S. study mentioned is that by Pushkarev and Zupan tI977] who developed a set of "land-use thresholds" that were necessary to justify financially different types of transit investments, based on intermodal comparisons of transit unit costs and intercity comparisons of transit trip generation rates. The land-use characteristics identified to be the determinants of transit demand in this study were the size of the downtown, measured by the non-residential floor space, the distance of a site to the downtown, and residential densities. In a subsequent study Pushkarev and Zupan tI980] developed six demand-based threshold criteria to determine the financial feasibility of fixed guide-way transit. The Pushkarev stuffy was based on two assumptions that may not be presently applicable "Steiner, ~ 9944. First it assumed that all work trips are to the central business district' in other words, a monocentric cites. This contrasts with the multi-centered character of most metropolitan regions in the U.S today. Second. non- residential lancI-uses were assumed to be segregated from residential uses. This contrasts sharply with the mixed land-uses found in many neighborhoods tPBQD. 19954. Not~thstar~ding the difference in Dreary form implied in the PusDkare~ study from today's urban structure. the PusDkarev's study is still widely quoted and employed in determining feasibility of proposed rail projects. In another study~ Smith tI984] compared transit usage in six U.S. metropolises. and found that the renumber of trips made by transit increased sharply when residential densities rose from approximately 7 dwelling units to ~ 6 dwelling units per acre. In flew York city, for example. such a density increase resulted in an increase in the renumber of transit mps made per person Tom a figure of 0.2 to 0.6. A study by Peat Warwick & Mitchell tI97 5] using data collected In the 1973 Nationwide Personal Transportation Survey (NPTS) found that for both bus and raid systems, density among other variables did not explain much of the variation observed in transit usage. Rather socioeconomic characteristics of the residents appeared to better explain the observed variations. A-1
TCRP H-17 Final Report Levinson and Kumar tI994], who investigated for relationships between density and several indicators of travel behavior using the 1990 NETS data found that the density-threshold for a relationship between density and mode choice was ~ 0.000 persons per square mile. They, however were critical of the thesis that density alone can explain travel behavior. Burby, et at. t~ 974] compared travel behavior in fifteen new communities that had mixed land- uses, with fifteen "semi-planned' conuoT suburbs. Notwithstanding the differences in urban designs their study results shoaled no significant reductions in vehicle miles traveled or transit usage for all trip purposes save recreational trips. Fr~edmar~, et at. tI994] compared household trip-rates of residents of communities designed as standard post-war suburbs with that for residents of older, more traditionally designed communities. Data for this study were extracted from household travel surveys conducted in the San Francisco Bay Area. The authors excluded data from households located within the city of San Francisco. stating that its high level of transit service and utilization, and high jobs/housing ratio could not be replicated in any new town or community to the extent that it exists in San Francisco. They also excluded households with very Tow or very high incomes from the analysis, stating that based on the price range of houses in some of the neo-traditionally designed neighborhoods under construction at the time, Hey were urdikel, residential candidates. Trips were stratified by purpose and then by mode, and household modal trip rates computed. The major findings of this study include the 25 percent higher daily trip-rate and 32 percent higher auto-dnver trip-rate that households in the "standard suburbs', made compared to households in traditional neighborhoods. Auto-based modes had a mode-share of 86 percent in the suburbs compared to 76 percent in Me traditional communities. Transit use in the traditional areas stood at 7 percent compared to 3 percent for the suburban areas, and walk mode share for all trips was 12 percent in the ~aditiona1 communities compared to ~ percent for the suburban communities. Clearly all this study reveals is the existence of a difference in household tr~p-rate between residents of two types of neighborhoods. No attempt Bras made to isolate the effects of other important factors that could have contributed to this difference in observed tr~p-rates. Hence no conclusions can be drawn Dom this study on the efficacy of neighborhood design schemes to reduce vehicle miles traveled or increase the attractiveness of non-automobile modes. Dur~phy and Fisher t! 996] report their findings of art investigation into the relationships between urban densities, socioeconomic characteristics of residents and their travel characteristics. The analysis was focussed first on regional comparisons of travel characteristics, using as the data-source, the 1991 FHWA Highway Statistics. Specifically. the authors investigated the relationship between average density of a number of metropolitan regions in the U.S. and annual vehicle miles traveled per capita, as well as the annual transit trips per capita. The authors found that there was a general tendency for less driving in higher density regions, with the notable exceptions of Sari Francisco and Sari Jose where residents droxte far more than residents of other regions of similar average densities. The analysis of trips by transit indicated a clear pattern of higher levels of transit trips per capita in regions of higher density. leading the authors to conclude the existence of a positive relationship between density and transit usage. In a bid to better understand the observed behavior of travelers living in the different communities, as opposed to entire regions, the second part of the analysis investigated the relationship between density. calculated at the zip-code level, with demographic and A-2
TCRP H-17 Final Report . socioeconomic characteristics of residents, as well as their travel characteristics. The findings were as follows: I. household travel. measured in miles per household, increased with income; per capita VMT was substantially lower at higher levels of density, up to a factor of 9 when compared to driving by residents of the lowest density areas similar to Handy's t! 995] finding that the average number of per capita trips by all modes declined very little Mom the Cow- density neighborhoods to the high density neighborhoods; 4. the average number of trips per capita by the auto-driver mode declined by one-half as density increased from 4.500 to 30.000 persons per square-mile, automobile person trips similarly declined from 3.6 person trips to 0.6 person trips per day; S. transit trips increased sharply above densities of ~ 0,000 residents per square mile, from less than 0.2 to 0.5 hips daily; and 6. stalking and bicycling also become significant at higher densities, growing from about 0.3 trips by walk per day at densities of 4~00 persons per square-mile to I.S trips by the average resident living at densities greater than 40.000 persons per square mile. This latter study. although insightful and useful, suffers Tom a fen shortcomings. First. the use of a single average density for a large urbanized region masks a lot of information, and does not help to explain Ably why residents of communities of different densities exhibit differences in travel behavior. As an example. the influence of the different neighborhood designs on mode usage is completely ignored. Further, Me stud`; results demonstrated that notwithstanding Los Angeles' high urban-density, it had much lower transit usage compared to New York. which had a lower population density, suggesting that factors other than density also had an important role to play. Second, the results are primarily drawn Tom one-~-ay cross-tabulations and simple plots. Thus the effect of each variable is assessed without consideration of the impact of the remaining variables, or for any possible interactions among them. Clearly, the explicit role each vanable plays in explaining travel behavior can only be fully determined when all the variables and their interactions are simultaneously considered in the analytical framework. Third. in comparing the different metropolitan regions, the historical development of the cities is completely ignored. Fourth, no explicit consideration is given in the analysis to the transit service provided in the various communities or regions. Fifth. in the regional arlalysis, hardly any figures are quoted in the paper to enable one to assess the extent of the relationship between density and travel behavior. In a departure from the simple linear correlation approach of ins estigating the role, if any, of urban form on travel behavior, Schimek tI996b] developed a multiple linear regression model of vehicle travel which includes vehicle ownership as an intermediate factor, and which treats a household's pick of neighborhood density and the amount of travel as a simultaneous relationship. To better isolate the impact of density on trace] behavior, income and demographics were controlled for in the mode} specification. The issue as Schimek, points out, "is not only whether density affects automobile use, but whether the effect is strong enough so that attainable changes in density could AN
TCRP H-1 ~ Final Report make a substarthal contribution to efforts to improve air quality and provide other social benefits of reduced automobile use". Using data drawn Tom the ~ 990 Nationwide Personal Transportation Survey, Schimek obtained the following results hom his estimated models. Members of households in higher-densit,~- areas make fewer automobile trips and travel fewer automobile kilometers than those in low--density areas. 2. For trips, most of the effect of density is direct, but for distance driven, two-thirds of the effect of density on automobile use comes through the mechanism of Tower rates of car ownership in high-density areas. This lower rate in car ownership is not the result of the association between income and density. because the model controls for the influence of household income on residential density. Lower rates of vehicle ownership In higher-density areas is likely the result of greater attractiveness of alternatives (walking and public transit) and the greater difficulties and higher costs of motor vehicle storage in high-density areas. 3. The hypothesis of shorter car trips in higher density areas because of land-use clustering appears to have only a small effect. In other words there may be more destinations within a shorter distance, but in practice people appear not to take advantage of this density to reduce vehicle travel. In shorts density matters. but not much. ~J 4. With respect to public transit use, much of Me difference in household travel associated with the presence of transit comes from lower rates of vehicle ownership. This suggests that at least some households locate near transit routes to reduce their vehicle ownership needs or having done so. find that they can reduce automobile ownership without an unacceptable loss of mobility. Schimek concluded that his study results provided evidence that households in higher-densi0 areas travel less in private cars, all else being equal. However, the effect of density was found to be so small Mat even a relatively large-scale shift to urban densities would have a negligible impact on total vehicle travel. In quantitative teens, a 10 percent increase in der~sity leads to only a 0.7 percent reduction in household automobile travel. By comparison, a 10 percent increase in household income leads to a 3 percent increase in automobile travel. Schimek's study pushes the state of the art forward In that it attempts to address the endogenous nature of some of the explanatory variables specified in the models, particularly, vehicle ownership. However, the vehicle ownership mode] is simple' and not derived from any explicit theory of decision making. The study also uses spatial-units defined by zip-code to compute densities, units that can hardly be described as homogeneous in urban character. Thus, spatial organization within the zip-code areas is not controlled for in this study. Kockelman tI997] employs multiple regression analysis and the binary logit model to explore the association between several dimensions of urban form and travel behavior, after controlling for socioeconomic factors. Several variables describing land patterns, such as accessibility, land-use balance, mix and density are defined and tested in the mode] specifications. Travel data for the study came Dom the ~ 990 San Francisco Bay Area Travel Survey which involved over 9,000 households while the land-use data was largely constructed from the 1990 Association of Bay Area A-4
TCRP H-12 Final Report Gove~ment's land-use file. Several models were developed for: vehicle miles traveled (VMT) per household. non-work home-based VMT per household. auto ownership models, and No mode choice models. The study results showed that after controlling for demographic and socioeconomic factors. measures of accessibility. land-use mixing and land-use balance were influential in their impact on all measures of travel behavior. With the exception of the vehicle-ownership model. the impact of density on travel behavior was found: to be negligible after controlling for accessibility. Further. the author found that arise balance. mix and accessibility were more relevant to travel behavior prediction than several household and Raveler characteristics commonly used for this purpose. The author concluded that the study lent empirical support to the neo-traditionalist claim that land-use integration and compact development reduce automobile reliance. This paper does well In defining several objective variables that describe the built environment for testing in venous mode] specifications. However. a look at the estimation results show- some of the models to be of quite low explanatory power. As an example the non-work home-based VMT per household base mode! has a coefficient of determination of 0.0366, and after inclusion of the best set of urban-form explanatory vanables, rises to 0.0~67. Thus, although signs of coefficients In these models may reveal the impact of individual explanatory variables on the dependent variable. the estimation results also indicate that as a whole, the models are able to account for only a small proportion of the variation in the observed values of the dependent variables, or, in other words, that other unidentified factors are primarily responsible for the variation in the data. Secondiv. auto- ownership arid mode choice are not modeled within a consistent travel-behavioral framework. Auto Ownership is modeled using linear regression anally sis, while mode choice is modeled with a binary logit model, without any theoretically derived link between these models. Cervero arid Kockelman [19973. in a very similar study to Kockelman tI997], investigated the influence of the built environment on travel Remarry along the thee principal dimensions of density, diversity, and design. The Ravel and socioeconomic data used in the analysis were drawn from the 1990-1991 Bay Area Travel Survey. The land-use and design-features data were collected through field surveys of the 50 sampled neighborhoods Tom the San Francisco area the Association of Bay Area Governments (ABAG) land-use inventory, and the Census Transportation Planning Package (CTPP). Multiple linear regression and binary logit models were estimated to test for any association between trip-rates, mode choice and vehicle miles traveled respectively, and several factors describing the built environment. The model estimation results led to the conclusions that density. land-use diversity, and pedestrian-oriented designs generally reduce trip rates and encourage non- auto travel in statistically significant ways. though their influences appear to be fairly marginal. Elasticities between each dimension of the built environment and travel demand were found to be modest to moderate, though certainly not inconsequential. Residents of neighborhoods ~ ith grid- iron street designs and resmcted commercial parking were found to average significantly less vehicle miles of travel and rely less on single-occupant vehicles for non-~-ork trips. Within neighborhood retail shoes were most strongly associated with mode choice for work trios. .. ~ . ~ ~ ~ ~ ~ ,, Schimek tI996a] compared public transit use in Toronto to that in Boston, with the objective of identifying the factors that would account for the differences in patronage levels. Choice of these cities lay in their similarity in terms of size of population and geographic area. On the basis of the A-:
TCRP H-12 Final Report l estimated hme series models of transit use for the two regions, which had several policy and control variables specified, the author concluded that the higher residential densities, the greater concentration of jobs in the CBD and the inner suburbs, the greater transit service provided, =d relatively lower incomes and income growth in Toronto contributed to the higher observed use of transit, compared to Boston. Frank and Pivo ~1994] studied the impacts of land-use mix and density on use of the single- occupant vehicle (SOV), transit and walk modes respectively for shopping and work trips. The analysis was done at the census tract level. Sources of data for their study were the Puget Sound Transportation Panel, the U.S. Census Bureau and a three local agencies in Washington State. Similar to Schimek t] 996b]- a multivariate statistical approach was used in the analysis to control for non-urbar~ form factors hypothesized to influence travel behavior, while assessing the possible role density and land-use mix might have on travel behavior. The percentage of SOV, transit and walking trips Mat onginated or terminated in a census tract were calculated for each census tract, and each in turn served as the dependent variable in a multiple regression model. Four specific hypothesis were tested: I. population density, employment density, and land-use mix are related to mode choice; population density, employment density, and land-use mix are related to mode choice when non-urban form factors are controlled for; a stronger relationship exists between mode choice and urban-form characteristics at both trip ends than at one trip end; and 4. the relationship between population density. employment density, land-use mix, and mode choice is non-linear. From the results of a simple correlation analysis, the authors concluded that urban form and mode choice were significantly related. The strongest linear relationships for work and shopping trips respectively were between employment density and transit and walking. Land-use mix was not found to be significantly correlated with any of the three modes for shopping trips. Results Tom the regression analysis fed the authors to conclude that urban form is significantly related to mode choice when non-urban form factors are controlled for. The authors state that the percentage of transit arid walk trips for both work and shopping respectively had the highest relationships with the urban-form variables. Urban form factors were found to be consistently ne~ativeiv associated with the percent of SOV use and were nositivelv a~.int~1 with n~rn.ent · . ~ tot · ~ ~ ~ ~ . ~ . ~ ~ ~ ~ ~ . , · ~ · ~ ~ _ . transit use anct walking. ~ anally, the authors concluded that the relationship between mode choice and employment density is nonlinear. Two employment-densit~r ranges at which significant modal shifts from SOV use to transit and walking were identified. The first is at 75 employees per acre. beyond which the percentage of trips by transit or walk increased dramatically. The second is between 20 and 50 employees per acre at which there are moderate increases in the use of transit and walk modes. The study findings suggest that population densities need to exceed ~ 3 residents per acre for changes in mode choice to be detected. Further, the reduction in SOV travel was not as significantly associated with increases in population density as it was with employment density. The A-6
TCRP H-17 Final Report relationships between land-use mix and mode choice were found to be relatively weak. Handy tI993] addressed the question of how- alternative forms of development could affect shopping travel patterns using the concept of accessibility. Two measures of accessibility were defined In this process to reflect the Inking that Me amount oftravel made by a person is influenced by bow the character ofthe local community person resides in, and the spatial structure ofthe region of which the local community is a part. Both accessibility measures were calculated using art exponential form of the gravity model. The first. local accessibility. was a measure of proximity to locally oriented centers of activity. The measure of attractiveness was defined as the sum of retail, service and other employment within the zone. The off-peak intra-zonal travel time by auto was used in the impedance function. The second regional accessibility, measured how good transportation links to large regionally oriented concentrations of activity were. The measure of attractiveness was retail employment, with the impedance to travel being a Traction ofthe inter-zonal travel time by auto. Handy tested four hypotheses: I. that accessibility levels will be negatively related to travel distances 2. that there will be a positive relationship between trip frequency and accessibility; 3. that accessibility will have little impact on total travel as measured in average person kilometers traveled; and 4. that the balance between regional and local accessibility of a community significantly influences the travel patterns of its residents. The study results showed that high levels of either local or regional accessibility were associated with shorter average shopping distances, but not with trip frequency as hypothesized. This suggested that there may be an average or standard number of trips that residents make, regardless of the distance dew must travel. The amount of non-~-ork travel measured in person kilometers traveled, was found to be significantly Tower in areas that had higher levels of accessibility at both the local and regional level. Har~dy's study, though useful. has a number of drawbacks. First, the measure of accessibility treats all households/persons residing within a particular commuru~- as having ache same accessibility regardless of whether they have a vehicle available to them or not. Second the measure of accessibility employed is sensitive, in teens of system attnbutes. to only travel-time by automobile. Thus, the level of transit service provided or network of ~alk-paths that determine Cracking distances to retail activity do not reflect in the re:,ional or local accessibility of a location. Third, local accessibility was aggregated to obtain an average value at a super-district level. From the discussion ~ the paper, the heterogeneity within these super-districts far exceeded that between districts -- this raises the serious question of how representative the average values were in reflecting accessibility for the entire super-district. Further, averaging reduces the amount of variability to be explained, and this can lead to spurious results and conclusions. Aggregation to a super district level also meant that an accessibility value did not consistently clearly map a neighborhood into one of the alternate neighborhood types considered, namely. traditional. or neo-traditional, or suburban-type neighborhood. Thus implications for policy may not be clearly defined. FiDch, the possible role A-7
TCRP H-l ~ Final Report played by socioeconomic and demographic characteristics of residents of the various local communities in accounting for the vocations in observed shunning travel patterns is cc~mr,ieteiv ignored in the analysis process. -rr---= ~red -----a---- Handy (199~) describes an alternative approach to researching the link between urban form and travel behavior. This approach is based on a theory analogous to that underlying the development of discrete choice models. Succinctly, it suggests that urban fond must be evaluated in terms of the sets of choices that it provides, in terms of the kinds of destinations that are found in the various areas, in terms of the transport modes serving these areas, and the characteristics of these choices, including the cost and comfort of travel, the amount and quality of activity at the destination, etc. By this differences in travel characteristics in different neighborhoods are not simply attributed to some being old while others are new. Rather the differences in travel characteristics are ascribed to the different sets of choices inherent in the form of the neighborhood Hat influences travel behavior. Thus urban form is evaluated in teens of the range arid nature of the choices inherent within it. Handy's studs- focussed on non-work trips, specifically shopping Imps, which she argues are more likely to be greatly- influenced by urban form due to the greater flexibility in terms of time and space associated with this trip. Handy selected four neighborhoods in the San Francisco Bay Area for her study. Selection -as based on three factors, namely, location writhing the region and accessibility to regional centers of retail activity, the type of neighborhood, that is, whether "traditional" or "typical", and finally, the socioeconomic characteristics of the residents in the communities. By limiting the number of neighborhoods studied, Handy- was able to devote more attention to obtaining good measurements of the attributes residents of the different neighborhoods were faced with in making decisions concerning shopping in addition to their socioeconomic attributes. By studying the variation in the travel patterns of residents of the different neighborhoods using the analysis of variance procedure and controlling for socioeconomic effects, Handy found the following: 1. residents appeared to compensate for longer distances by making fewer trips, 2. residents in the traditional neighborhoods, which had more people living within walking distance of a supermarket. had a higher percentage of the walk trips; 3. having a single supermarket close to a resident did not necessarily minimize the total amount of travel to supermarkets since residents choose to shop at a variety of supermarkets; 4. residents of traditional neighborhoods, compared to residents of typical neighborhoods, had a greater frequency of trips to convenience stores because of the greater accessibility to convenience stores in these areas. These trips did not appear to substitute for trips to supermarkets, as the frequency of trips to supermarkets by residents of the traditional neighborhoods was not significantly different from the other study areas; 5. average frequency of trips using the walk mode for the purposes of pleasure or exercise did not differ significantly between neighborhoods, suggesting that the differences in pedestrian quality of these areas did not seem to foster walk-trips; and A-8
TChP H-19 Final Report 6. point (5) notwithstar~ng. the percent of respondents who walked to a commercial area at least once in the previous month and the average frequency of ways to commercial areas did vary significantly between the traditional neighborhoods and the typical neighborhoods. This finding was partly explained by residents of traditional neighborhoods being more closely located to commercial activity. Interestingly. in areas where links between residential areas and commercial areas could be described as "pedestrian unfriendly" residents still used the walk mode In significant numbers. Thus, Handy, concluded that having commercial activity within walking distance is enough to encourage at least some walking but nedestnan- oriented design encouraged it even more. 4 , ~ Handy concluded that urban form does in fact make a difference in determining whether residents perceive walking as an option available to them. The relationship between residential and commercial areas is important; distances must be short, and barriers such as major arterials should be eliminated or avoided when possible. Commercial areas need to be designed for pedestnan access as well as automobile access and for pedestnar~ circulation within the commercial area. This study found no evidence that residential design (things like Dont porches, varied of materials and designs, etc.) is particularly important in the decision to walk, contrary to what neo-traditionalists assert. There was also no evidence that the option to walk would significantly reduce automobile travel. A greater range of destination choices. at least up to a point, is valued by residents. although this results in more travel and longer average tr~p-lengths. Cervero tI996] examined how mixed land-uses and features of the built environment, like residential densities, influences the travel choices of residents from large metropolitan areas. The Gavel choices considered are the mode used for commuting, the commuting distance arid household vehicle ownership level. Data for the stuffy were drawn from the 1985 American Housing Survey and embraced ~ ~ Metropolitan Statistical Areas (MSAs). The mode choice models developed are based on random utility theory and are modeled as Gnaw decisions. Thus. even though three modes were modeled, namely, automobile (drive-alone arid shared ride), public transit and walk/bicycle, each was modeled within a beam framework. The explanatory variables specified in the modal utility functions were either classified as land-use variables or control variables. No attributes explicitly describing the transport system In terms of times and costs were considered. The analysis of commuting distance and household vehicle ownership levels employed multiple regression models to identify the relevant related urban form factors. Five hypotheses were tested in the study. These are (p.3631: - 1. mixed-use neighborhoods induce higher shares of non-auto commuting among residents 2. mixed-use neighborhoods exert their strongest influence on non-motor~zed commuting, specifically walk and bike travel to work; 3. m~xed-uses only have a positive influence on transit-riding, walking, and bicycling to work if they are close by (i.e., within several blocks of a residence); non-residential uses. such as grocery and drug stores that lie between several blocks and a mile or so of a residence induce auto-commuting and trip-chaining; and A-9
TCRP H-12 Final Report 5. mixed-use neighborhoods are associated with shorter distance commutes and lower vehicle ownership rates. The following conclusions were reached based on Me estimated coefficients (p.373~. Neighborhood densities have a stronger influence than mixed land-uses on all commuting mode choices, except for walking arid bicycling. For commuting by bicycle or walking having or not having neighborhood shops was more important than residential densities in predicting the mode chosen. 3. Non-auto commuting increases with increased neighborhood densities. 4. 5. Relative proximity to mixed-use development influenced mode choice. The likelihood of non-auto commuting, for example, increased significantly with increased proximity to mixed used development. If retail shops were within 300 feet, or several city blocks from a dwelling unit, workers were more likely to commute by transit, foot or bicycle. Beyond this distance. however. mixed use activities appeared to induce auto-commuting. Neighborhood density and mixed land-uses affected vehicle ownership rates and commute distances. Neighborhoods where residential densities were high and non-residential activities were relatively close by, after controlling for other factors such as household income. had comparatively lower automobile -ownership rates and shorter commutes. This latter study represents a good attempt at finding out what it is about people and households that results in the differences In travel behavior of residents of high- and Tow-densitv neighborhoods. However, there are concerns with some aspects of the study. First is the data employed in the modeling effort. Because many ofthe households did not supply all the requested information. more than two-~irds of the 42 200 surveyed households had to be dropped from the analysis tCervero, p.3644. The study does not investigate whether the remaining data have similar attribute distributions to the parent sample, arid hence makes no statement on the biases that could exist in the modeling data. The study results should therefore be treated as questionable. Second even though some of the defined dummy variables included in the mode choice mode] specifications may be associated with the transport system, they do not adequately reflect the role ofthe trar~sport system in influencing the travel choices of commuters. Such exclusion of relevant variables in the binary Togit model specification biases the estimates of the remaining slope coefficients tCramer ~ 99 ~ ]. Third the impact of mixed land-uses on travel may be somewhat overstated in the conclusions. in that, its impact on probability of choice ofthe automobile or transit for a trip, at a given density level, appears quite minimal. Fourth, very little attention is given to the goodness-of-fit statistics in the assessment of the models. The likelihood ratio index for the transit-commute mode] is O.O87. It would be interesting to know how this model compares to the market-share model, since such a low value of the index implies very little improvement in the log-likelihood value by the specified mode! relative to the equal modal-probability model. This low index value means that the specified variables in the modal utility functions do not serve as good predictors of the travel choices of the individuals in the modeling data. In similar vein to the poor mode-choice mode] fit to the data. the coefficient of determination of the regression mode] for predicting commute distance is only 0.05 A-10
TCRP H-19 Final Report indicating that the included explanatory variables are able to explain only A. ~ percent of the ~ Cation in the tnp-distance data. Further, if the omitted variables are correlated with the included variables then this would lead to bias ~ the estimated mode! parameters. PBQD [1 996a] investigated the nature of the impact of the residential built environment on transit usage. More specifically, they examined the ways mixed land-uses and urban design in residential neighborhoods affect travel choices alter controlling for densities, household income. and transit service characteristics. Data for their study were from three sources. First is a sample of ~ ~ large metropolitan areas Dom the 198: American Housing Survey -- these data were used to model mode choice for the work tup. Second is the Chicago arouse and transportation databases -- these data were used to model the transit trips produced per person, and rapid raid and commuter rail boardings. Third, data were collected from 12 neighborhoods in the San Francisco Bay Area. and were used to mode! mode choice for non-work and work trips. The study fond that density explains more of the vanabilit T in transit use than land-use mix or urban design, stating that "density explained by ~ 0-20 times more, transit use for commute trips than land-use mix". In particulars residential density was found to erratic influence commuter mode choices, transit trips made per person arid rapid rail station boardings~ while station area employ ment density was fourth to influence We number of boardings at commuter rail stations. Land-use mid; was found to be statistically significant in explaining transit use a few times. In particular, residing in a mixed-use block or close to non-residential uses was found to result in a comparatively- higher probability of commuting by transit or other non-auto mode. Based on He San Francisco data. it was found that few of the individual neighborhood design variables tamed out to be significant in determining transit usage or for that matter, mode choice for non-work tnps. However. when the design variables were "aggregated" and treated as a dummy variable. which reflected either a "traditional neighborhood" or "non-traditional neighborhood", the variable became statistically significant In some instances, with residents of traditional neighborhoods being more likely- to use non-automobile modes for non-work trips than residents of suburban (non-traditional neighborhoods) neighborhoods [PBQD, ~ 996b]. A particularly thorny issue in this research was the difficulty He authors had In isolating the roles of land-use mix arid urban design in influencing travel behavior because of their strong correlation with density. Cervero tI 989] analyzed the travel choices of workers and local traffic conditions of see eral suburban employment centers (SEC) to find out if they were influenced in any significant -at by site characteristics, namely. (~) employment densities; (2) site designs; (3) land-use composition, particularly the level of mixed use activities, (4) suburban levels of iohs housing balance (~ ~ land lotting arid ownership patterns, and (6) parking provisions. Cervero employed the step-wise linear regression procedure to develop aggregate models using data on the above mentioned variables. The model results led Cervero to draw the following conclusions. - J ~--D ~\- J T. Employment densities and a variable which measured the proportion of square footage of floor space to square footage of larch in SEC. were the most important in influencing local traffic conditions. Furler, that high employment densities worked in favor of commute alternatives to the Unve-alone automobile. High density SECs were also found to generally have relatively low- levels of parking supply. A-1 1
TCRP H-17 Final Report 2. Of the variables examined, the degree of land-use mixing appeared to have had the greatest influence on the modal choices of SEC workers. Projects made up of predominantly of rices were associated with solo-commutin~. Cervero's results showed that all else being equal, every 20 percent increase in the share of total floorspace devoted to office use resulted in a 2.4 percent increase in the single occupant driver mode-share for the work trip. On the other harts, the availability of nearby retail services appeared to induce ridesharing and attract more work trips by the walk and bicycle modes. 3. The size and scale of SEC activities appeared to influence mode choice. Mode shares of car- and van-pools appeared to increase with increasing size of the project. Cervero reports that every :000 workers added to office developments increased ride sharing by 3.: percent. all else being equal. ~us. Cervero states that suburban work environments with a critical mass of employees appear to be an important pre-requisite for mounting and sustaining successful ridesharing programs. 4. Suburban work settings more evenly balanced in terms of jobs and housing had higher shares of employees walking and cycling to work, although they at the same time had lower percentages of ridesharing. Cervero, by his own admission. obtained relationships between frequency of trips by mode and the above mentioned explanatory vanables that were not strong. The lacklustre model-showing was attributed to the use of aggregate data and models. It remains to be demonstrated, however, that Farther data disaggregation, which would be closer to reality, and increase the variability to be explained, would have yielded better model-estimation results. PBQD tI996c] investigated how the mode of access to and from rail stations, and the size of the station-catchment area are affected by the built environment. Three rail transit systems were selected for this study: one in San Francisco (BART) and two in Chicago, METRA and Chicago Transit Authonty (CTA). Travel data on the users of BART were obtained in an on-board survey conducted in ~ 992, while Mat for transit users in Chicago were collected in a household Gavel survey conducted in 1990 by the Chicago Area Transportation Study (CATS). Land-use data for B ART were drawn from a 1990 digital inventory of dominant land-uses compiled by the Association of Bay Area Governments. Supplementary information on the areas around the stations was taken Tom the ~ 990 Census Summary Tape File and the Census Transportation Planning Package (CTPP) for San Francisco Bay Area. Land-use data for Chicago were provided by the Northeastern TIlinois Planning Co~runission (NIPC). Supplementary information on the areas around stations was obtained from the ~ 990 Census Summary Tape File and the Census Transportation Planning Package (CTPP) for Chicago. Cluster analysis Vitas used to classify stations with similar -use characteristics, measured in terms of density and land-use composition. The objective was to study how access and egress trips varied by land-use environment. To test the he pothesis of how the built environment influenced access characteristics of BART, regression equations to predict the share of each access mode as a function of urban densities, land-use compositions, and various control variables, were estimated. The major findings of the study included the following. A-12
TCRP H-12 Final Report 1. The built environment influences choice between the rail-access modes of wralking and driving. Thus, residents in dense, mixed-use residential neighborhoods were found to be more likely to use walk as the access mode to transit, while residents of Tow-densitv suburban areas use cars. 2. Use of bus as an access mode to rail-transit is influenced strongly by transit service levels and parking supply and much less so by densities and land-use mix near stations. ~- The distances people walk to stations is not determined by the densit: and mix of land-use near the stations. 4. Availability of parking at stations encouraged driving and discouraged walking. Size of catchment area was influenced by the land-use environment. Stations in the iow-- density suburbs, arid where parking was ample, had much larger catchment areas than stations located in the downtown or urban areas. This study suffered from a paucity of data for transit users in Chicago, thereby making it impossible to reach any strong conclusions based on this data set. Further it would have been interesting to has e known whether demographics played any role in the choice of access mode particularly at stations located in the dense urban areas. One factor of the built environment that has more recently been suggested by some researchers (e.g., Cervero tI989] and Nowlan and Stewart tI991~) to affect the commuting patterns of workers sign~ficar~tly, is the jobs-housing balance. Nowlan and Stewart tI991], for example. report that notwithstanding the substantial new office construction in the Toronto central area between 197: and 1988. the impact on peak-hour work trips entering the area was offset by the substantial residential construction that also occulted In Me area. The inference was that people who occupied the new- residential construction also worked In the central area, thus resulting in no net increase in - trips to the area. Giuliano and Small [1993] examined the jobs-housing imbalance factor more closely investigating whether it has an important influence on commuting-distances or times on a metropolitan-w~de basis. ~ sing data collected Tom the Los Angeles region in ~ 980, the authors find the commuting pattern that would minimize average commuting time or distance, given the actual spatial distributions of job and housing locations. The authors found that the amount of commuting required by the spatial distributions of jobs and housing locations is far less than the actual commuting. In other words that people live much farther from their place of work than the standard mode} would predict. Further they found that variations in required commuting across job locations only weakly explained the variations in actual commuting. The authors concluded that "other factors must be more important to location decisions than commuting cost, and that policies aimed at changing the jobs-housing balar~ce would have only a minor effect on commuting" tp. ~ 48S]. McNally arid Kulkarni tI997] also examined relationships between the land-use - transportation system and travel behavior to determine if policies advocating land-use modifications were likely to promote travel behavior changes. Unlike several studies where residents are classified into two neighborhood types. namely. traditional neighborhoods or planned urban neighborhoods. a third neighborhood type is introduced, referred to as the mixed neighborhood types. This neighborhood A-1~
TCRP H-12 Final Report - t~tpe has a mixture of the elements of the former m-o hence its name. Land-use and trave! data for the study were collected in Orange Countv in 1991. ~, . . ~· ~ The county was partitioned into several ne~gnuorhoo~s, and a cluster analysts procedure was used to assign each county to a neighborhood- t,vpe based on the values of several network, arouse and design variables. The statistical procedure of analysis of vary ce (ANOVA) was then employed to study the vanation in household tup-rates arid mode split. Socioeconomic factors were included In the analysis in order to assess the role land- use factors have on travel behavior in an unambiguous fashion. McNally and Kulkarni found from the ANOVA results that neighborhood type was not a statistically significant factor in explaining the variation In household trip generation and mode choice. Rather, income was found to be the single most important factor influencing travel behavior. The authors Bus concluded that the relationships between travel behavior and land-use - transportation system was rather weak, therefore casting doubt on the efficacy of desi~n-oriented solutions to address problems of congestion and air pollution. In an interesting departure from the majority of earlier works, Messenger and Ewing tI996] employ a multivar~ate framework to investigate the factors influencing the share of travel by bus for the work-tr~p. In particular, the study attempts to identify how density influences travel choices and behavior. The study area was the Metropolitan Dade County, win the analysis unit for the study being the traffic zone. The data-source was the U.S. census of population and housing for 1990. Several vanables. that can be categonzed as socioeconomic, land-use, road network, transit service and auto- related~ were investigated. The transit service and route data came from the transit route map in effect at the time of the ~ 990 AS. Census of Population and Housing, while walk access times =d bus travel times to the downtown were from travel time skims generated by the regional travel model. This study is the only one encountered so far that explicitly considers system-at~bute values for transit and a few for auto In the mode] specification. Bus mode-share by home-zone and work- zone respectively, are each modeled with a system of three equations. This happens because two of the explanatory variables in the bus mode share model-specification, namely the proportion of households with 0 or ~ automobiles and the loganthm of the number of local bus routes per square mile, are endogenously determined. Messenger and Ewing, from their model-estimation results, concluded that socio-demo~raphic, land-use and transit service attributes all affect bus use through a web of relationships; bus mode share was found to be primanly related to automobile ownership and secondarily to jobs-housing balance and bus service frequency. Automobile ownership, in turn, was related to household income. overall density, arid transit access to the downtown. Density was found to be a significant determinant of bus mode share, although indirectly, through automobile ownership and parking charges. They also estimated that between 8.4 and ~ 9.4 dwellings per acre was required to support transit service. A variable measuring the effect of land-use mix in each analysis unit had either no effect or an adverse effect on bus use leading the authors to suggest that the effect of mixed uses on transit ridership could be erroneous or the variable may have been mix-specified in their model. Road network design had no apparent effect on bus use. Bus mode share by Place of work was found to ~,& ,& 1 1 ~ 1 . , 1 , ~- 1 · , ', . , 1 1 . 1 ~ ~ ~ · . ~ . ~ 1 . be related to the cost or parings transit access to the downtown and overall density, although these variables were found to explain only a small proportion of the variation in the bus-mode share data by place of work. Recognizing this, the authors suggest that an alternative approach to modeling the A-14
TCRP H-12 Final Report data should be sought. Miller and Tbrahim tI998] provide an empirical investigation into the relationship between the physical dimensions of urban form namely, density. degree of decentralization. structure. and automobile travel (Vehicle Kilometers Traveled - VKT). which serves as a surrogate for energy use. Data for the investigation were drawn Dom a cross-sectional grave} survey conducted in the Greater ~ , Toronto Area in the Fall of 1 986. More specifically the analysis made use of 24-hour Home-Based- Work trips aggregated by home zone and work zone respectively. Equilibrium Origin-Destination (O-D) travel distar~ces were generated using the EMMEi2 road network assignment procedure and the product of these distances with the corresponding O-D flows gave estimates of VKT. Densities calculated were based on gross land area, while distances were measured as Euclidean distances between centroids. The study led to the following important findings: 1. VKT per worker increased With increasing distance from the CBD, author other major employment zones within the urban area; 2. a region-wide multi-center ss stem of high density employment nodes had lower VKT per worker compared to the case of these centers being absent; 3. increased jobs-housing balar~ce did not appear to result in lower VKT per worker. and 4. population density was not a strong causal variable in the explanation of variations in VKT per worker across the urban area. This study pros ides good insight into some effects of land patterns, and avoids some of the pitfalls that result Tom using average densities, computed over fairly large heterogenous land areas. However, some aggreaai~on still exists here since VKT is calculated at He zone] level, and this could result in the "burying" of substantial variance, unless He distribution of VKT by worker within each zone is relatively uniform. The regression analysis also has as dependent variable VKT per worker. This suggests that the variance of the dependent vanable is no longer constant, but dependent on the number of workers in each zone. Weighted regression analysis, rather than ordinary least squares, would have under such circumstances provided better estimates of the mode} parameters and standard deviation. In addition potential auto-correlation effects among adjacent zones have not been accour~ted for. Finally it would also have been interesting to have known -~That sort of impact variations in household s~uctureisocioeconomic at~nbutes might have had on VKT. Ewing ti99 S] investigates the effects of Arouse patterns on household travel behavior. This study works with the number of tups in a tour in order to capture the effects of trip-cha~ning. which earlier descriptive analysis had shown to be quite prevalent for suburban dwellers. Tours were classified as either `~-ork-related, if they include a stop at work, or non-work related. Data for the study were collected in a diar-~-based travel survey conductecl in Palm Beach County, Florida. The households in the database number 1,000, while the trip records number 1 6,000. The behavioral-unit of ar~alysis employed was the household. By using the household unit, it was possible to control for the effects of socioeconomic attributes (household income, autos per person household size. number of workers and housing Opel on travel behavior in the models developed. Several Arouse related variables were also explored. These included' at the origin zone, residential density, employment A-13
TCRP H-l ~ Final Report density jobs-housing balance dummy. and accessibility; and at the destination zone. employment density and accessibility for non-home based trips. Four attributes of travel were modeled using linear regression. These were trip rates. trip times, mode shares and vehicle hours of travel. The findings of the study include: I. development patterns do have a significant impact on household Aver beyond that explained by socio-demographic attributes -- "placing households in more accessible residential locations will cut down significantly on their vehicular travel" good regional accessibility cuts down on household vehicular travel to a far greater extent than does localized density or mixed use -- the jobs-housing balance. after accounting for socioeconomic attributes. had a pronouncedly weak relationship with travel behavior; 3. accessibility of residences to a mix of land-uses as opposed to one tone of use (e at. work shopping) had a significant impact on ~ ehicular travel; and , ~. , good accessibility of work place to other activities resulted in additional trips being made from work. although the average length of these trips was shorter, and reduced the number of trips made independent of work. This study does well in attempting to isolate the effects of urban form factors on travel behavior. However, some of the models on which the latter findings are based, have extremely low coefficient of determinations - four of Me eight models has-e R' values that range from 0.02 to 0. ~ 0, suggesting that as a whole, the vanables specified provide very little explanation of the variation in the respective dependent variable. A.~.3 Urban Form Impacts on Walk Mode Handy tl996b] points out that oniv few of the studies investigating the relationship between urban form and navel behavior have attempted to understand the more complex causal links between urban fond and travel behavior; most have typically relied on simple correlations between variables to establish association. This has led to little being known about exactly what attributes of the built environment Influence travel behavior. and how important these built-environment attributes are relative to the other factors influencing travel. Handy tI996b] therefore explores hove urban form fits into a more comprehensive mode] of choices about pedestnan trips. The focus is on walking hips. which is posited to be influenced be the surrounding environment and urban form more so than trips by any other mode. The hypothesis is that walk-trips, classified as "strolling trips (objective is the walk itself)" or "walks to a destination (objective is to reach an activity of interest)" have as primary causal factors individual motivations and individual limitations. Urban form is hypothesized to be an external factor that could encourage or discourage chalk trips, given motivations and ~ . · . mltatlons. Using data collected on six neighborhoods in the Austin area. selected on the basis of the era of development, location within the city, aIld to control as much as possible for average socioeconomic characteristics, He author explored the relationship between urban form and pedestrian choices. The neighborhoods were selected such that pairs of them were similar. Three pairs were thus obtained A-16
TCRP H-1 ~ Final Report and these were descnbed as "traditional", "early-modern", and "late-modern". Having similar pairs allowed for testing of the differences between types of neighborhoods as well as an analysis of variation between neighborhoods of the same type. Several variables were then used to describe the neighborhood transportation and -use characteristics. The v~abies fall under the categories: street system, transit system. commercial area and commercial establishments per ~ 0~000 population. Data on pedesman choices were collected through a mail-out. mail-back survey. The survey included sections on supermarket trips, walking trips trips to local commercial areas. and socio- demo=~raphic characteristics. as well as questions on feelings about and perceptions of a varietal- of up form charactenstics. A thousand questionnaires were mailed out, with the overall response rate being 25 percent Handy's study had the follow-in" findings: 1. the results supported the proposed model that is, Mat individual motivations and limitations are central to the decision to walk ': urban form was foment to be a secondary- factor that encouraged or discouraged walking, given the motivation to walk and the absence of limitations. A. 4. urban form was also found to play a greater role than other factors in the choice to walk to a destination, with the distance from home to destination being the most important factor. the latter factor mentioned in point (3), together with the quality of the pedestrian environment at the destination. outweighed the quality of the pedestrian environment around home in the choice to walk to a store; and S. a crude assessment of the possibility that walk trips to the store might replace drives to the store, under He most optimistic of assumptions' showed that only 0.4 percent of the average distance driven per month would be substituted for by walking. .. Handoffs study, although focussed on pedestriarls. represents one of the few- attempts at identifying, if anti what it is about ache urban environment that may hex e some relationship with the travel behavior of residents. Loutze~eiser [1997] reports on a study; to identify the factors influencing the choice of the access-mode of stalk for passengers using the BART system. Data for the -study was collected through an outboard survey conducted over a two-day penod. in the Fall of ~ 992. These data were supplemented with data from the ~ 990 US Census. The approach adopted was first to attempt to identify the factors influencing the choice of the walk access-mode for BART users. second, to examine the role of urban fonn/desi~n and station area characteristics in the decision to access BART by the walk-mode; and finally. to identify the impact of urban design factors on access-mode choice after controlling for the individual characteristics of B ART users. Three binary logit models of access mode choice there developed to identify the factors influencing the choice of the walk mode. These models were (~) walk versus all non-walk trips, (2) walk versus transit and (3 ) stalk versus the automobile. The author hypothesized that the choice of BART access mode is a function of the characteristics of the traveler" the characteristics of the access trip, and the assailability of alternative access modes. The variables found to be the primary determinants of access mode choice, from the model-estimation results. were distance-to-station, gender, ethnicity, age and car A-17
TCRP H-19 Final Report availability, Furler analysis revealed that the importance of these factors varied widely by B ART stations, and that specifying dummy variables to capture these station area effects was not adequate in explaining the relative roles of these vanables. Hence, the second part of the study investigated the impact of urban fonn and design on access mode choice. Multiple linear regression models. which related walk mode share at each station to various urban fo~m/design characteristics and average socioeconomic characteristics of residents of the neighborhoods around the station, were developed. The regression-results showed that population density was by far the most important variable to explaining the variation in walk mode share. Further, it was found that the proportion of BART users in the neighborhood of a station. who walked increased with the level of education of the residents. and with income. Parking capacity was found to have a negative impact on walk mode share, while proximity to an activity center had a positive effect on walk-share. To properly assess the effects of urban form factors on access mode choice, the aggregate station area characteristics and indix idual characteristics of respondents were combined in a single specification for mode] estimation. This so-called "combined model's show-cd that population and dwelling unit densit arid transit availability were no longer significant factors in the decision to walk. Median household income which previously had a counter intuitive sign, had a negative coefficient, hence impact on choice of the walk mode. The impact of urban design attributes was mixed. Loutzenheiser's results demonstrate the dangers of conducting a partial analysis; that is either developing a mode choice mode] that omits some variables, or developing an aggregate station-area mode] of walk mode share alone that omits variables descnbing individuals. As the study showed, this can lead to erroneous results and conclusions. Clearly, for the impact of any variables ore choice to be properly assessed, it must be included in a specification that includes the effects of all other variables. The study also illustrates some of the potential dangers in using average values for development of regression models -- it can lead to coefficient estimates with counter intuitive signs te.g. De Donnea. 1971], which the author found for the case of income. Other concerns with this study are with the vanables omitted in the transit utility, which may perhaps be responsible for the relative!! low percent-correct reported for transit. goodness of fit measures for the binary- logit models are not reported, and this prevents any comments being made about He overall utility of the models. Further, only a third of the available data were used for modeling, and no figures are given on the mode split of this modeling-data. Data from five stations, which had relatively low walk access-mode-shares for BART users, were deleted from the analysis in order to improve the fit of the model to the data. The objective of the study was to identify factors that influenced use of the walk mode for accessing BART; but, clearly. such data trimming measures have the unfortunate impact of not helping to identify why certain areas have low use of the walk mode, and perhaps of over-estimating the importance of other attributes. A.~.4 Simulation Studies In an interesting presentation on the impacts sprawl has on transportation demand, Eager ~ ~ 995] uses data from the 1990 National Personal Transportation Survey to argue that the expectation of increased travel by transit, walking and bicycle and decreased travel by automobile through increased density, may occur only at very- high densities, far beyond densities of development that would be realistic for 95 percent of urban America. More importantly his test scenario of doubling urban A-18
TCRP H-17 Final Report densities shows that even though household automobile trip-rates and automobile share of trips would fed at very high densities. the total number of trips bit automobile would also increase several times or en Easter argues Mat t~sit-proponents only take a partial look at the picture of travel: the increased use of transit, without looking at the grown In the overall number of trips by automobile. Eager's study concludes Mat the spot light should not only be on mode share and trip rates' but also on the total trips by mode. Bland tI983] examined the relationship between land-use patterns, particularly town shape, size and population density, aIld travel for all purposes in a set of nine hypothetical towns. based in part on modeling using the LUTE-model of walking, car arid bus travel Bland. 19824. The LUTE mode! was calibrated using data from the UK National Travel Survey (NTS), and a land-use pattern synthesized Mom UK Defacto Urban Areas population density data to fix the modal split parameter and the behavioral values of three and money for each mode and tup-pu~pose, prior to examining travel in the nine hypothetical areas. These latter areas vaned in tenth Tom 3 to ~ Km, and in width from ~ to 3 Km, with rural population density values of 2 persons/hectare to urban values of 10 to ALSO persons/ha up to 400 persons/ha. From the mode} results obtained, Bland concluded that the model-predictions were in good agreement with observed travel patterns. Specific conclusions include: I. use per car varies little with land-use pattern or bus service levels, although car ownership `~-as lower where Incomes are lower, or where congestion and parking difficulties or Rood access by public ~ar~sport or on foot make car ownership less worthwhile; 2. higher densities are not effective in reducing car ownership or use; further. Me UK NTS data did not show higher densities to reduce the amount of travel; A. moving homes arid jobs closer together did not appear to reduce travel based on the modeling results and from comparisons of more arid less compact cities; and 4 higher densities favor public transport use, although this did not imply any net savings in the resources devoted to mechanized transport, since the majority of trips carried by public transport would o~erv`Tise be made on foot rather than by car. ... . .. . .. . . . . McNally and Ryan t1993] examined the claim Mat transportation benefits. namely. reduced travel distances and times, cart be derived from ne - traditionally designed neighborhoods. Two hypothetical networks were developed to replicate a neo-traditional and a conventional subdivision. These two networks were developed with the guidance of several sources to ensure consistency with realistic networks arid land-uses. The road-attr~bute of prime concern in this studs was the shape of the networks, with factors such as street-width, s~eet-environment etc. being neglected. The hypothetical subdivisions had approximately the same level of activity. However, certain aspects of site-design were not modeled. e.g. mixed land-uses in traditional neighborhoods. Travel parameters, such as land-use trip-rates. were adopted from those developed for the City of Irvine, California Conventional transportation planning models were used as tools to evaluate the performance ofthe hypothetical networks in the two community types. Results of the study showed that the neo-traditional network resulted in approximately ~ O percent fewer vehicle-kilometers of travel during the morning peak compared to the conventional network. Further total hours spent A-19
TCRP H-17 Final Report traveling during the morning peak period in the traditional network was approximately 27 percent less than on the conventional network. Finally. the mean trip-length in the neo-traditional network was approximately T: percent shorter than in the conventional network. Ryan and McNally ticks] also cite the results of the work by Stone arid Johnson tI 992], who used site impact assessment techniques. to compare two hypothetical subdivisions and found that the neo-traditional neighborhood had 2: percent less vehicle delay, 20 percent fewer Hips generated, and 30 percent more entry points (used to define accessibility) than a conventional suburban neighborhood. to support their latter results. Handy [1 996a] also reports a study by Rabeiga and Howe [1 994]. who examined the impact of neighborhood design on selected Gavel charactenstics, using four neighborhood types that differed with respect to street layout and location of retail activity. "A sample of 120 origin points was selected in each neighborhood. hip tenths were calculated to a random choice of retail destination, and total travel was then calculated for these 120 tnps" tHandy, 1996a, p. 1543. Both measures showed a reduction in total travel coupled with an increase in average speeds in favor of neo- traditional networks relative to conventional networks. Although these simulation studies may yield useful insights of how different neighborhood- and street-designs may impact on travel patterns, they are based on simplifying assumptions of urban form and travel behavior, and hence their conclusions should be treated as unproven and speculative. A.~.S Problems and Some Directions for Future Work The empirical studies reviewed have all used cross-sectional data and models for their analysis. Yet, the dimension of interest is a temporal one; that is, the goal has been to learn how residents in a given nei~hborhood-type would respond to changes in their built environment. Thus, the use of cross-sectional data in these analyses is with the assumption that travel-behavioral response of an analysis unit over time can be inferred from the differences in the behavior of the analysis-units surveyed at a specific point in time. In the travel demand literature, generally, this has not been found to be so ~Kitamura. ~ 9903. Use of cross-sectional data therefore essentially allows only for the associations between variables describing urban form and travel-behavior, at a given point in time, to be determined, they do not provide any direct evidence of how residents of a typical suburban-type neighborhood would respond. in the travel context, to changes in the buiTt- environment of their neighborhood. To appropriately address the question of travel behavioral response to neighborhood design and land-use changes, time-series data would be necessary. As a first step, before and after studies should be conducted to give insights into how residents are likely to respond to changes in the built environment. Very often, the data available for analysis have been at a gross spatial level. This has led to the majority of studies making use of averages (e.g.. averages densities) for their analysis. First, these gross spatial units can hardly be described as homogenous in terms of neighborhood design, density and land-use mix, and hence do not always lead to clear-cut conclusions. Second, these average values can not be described as representative of the analysis-units, given that "within-unit heterogeneity" is in some cases comparable to "between ts heterogeneity". Future research should ensure that spatial definitions do respect the differences in density, Arouse mix and neighborhood A-20
TC8P Final Report . . design to allow for clear interpretable results. The literature review revels Me complexly of He phenomenon under study -- how people make choices with respect to where they will reside in space (residential location), the household auto- holdings, the mode for the journey to work. and the travel-modes and frequencies for non-work travel. This is a multidimensional choice problem. which involves a set of related choices. As an example, the number of vehicles a household ma\; decide to own is related to where they decide to live and the level of service provided by alternate modes serving the area. Thus any attempt at modeling arty of the dimensions of this multidimensional choice problem should first, respect the inter-related nature of the choices on hand arid therefore provide an internally consistent method of modeling the dimension, and second, provide for a theoretically sound approach for linking these sub-models. The majority of the studies reviewed have been pumarilv correlative in nature, developed partial models in that they have botany ignored impacts on the other choices on hand, and not been based on any explicit theory of human decision making. ~ ~ ~ · ~ ~ ~ ~ ~ ~ · ~ . . Clearly. if progress is to be made In determining the likely responses of residents to changes in urban for =d transport system attributes. then a comprehensive behavioral approach to modeling the several choices on hand has to be adopted. As Anas and Moses fI978, p. 163] point out, "elasticity series do not throw- much light on the locational adjustments households will make in response to changes in the cost of travel. This problem is better addressed by urban simulation models in which the location of activity and the demand for transportation are determined simultaneously". Consumer theory in economics provides such an approach, the random utility approach, for modeling these different dimensior~s of choices in a theoretically sound fashion. This involves the establishment of a choice hierarchy, which could presumably have residential location at the top of Me hierarchy, and household-vehicle Moldiness work-mode choice respectively at Tower levels tBen-Akiva and Lerman, 1985~. Sub-models, based on this theory have feedback between them, hence choices made at a higher level are done with Input from Tower-leve] choices. McFadden [1978] develops He theoretical framework for such models and their estimation, arid this could form the basis of fixture mode! development ~ the quest to better understand how- urban form impacts on travel behavior. A.2 EMPIRICAL STUDIES OF TRANSIT IMPACTS ON URBANE FORM Transit undoubtedly had a great impact on urban grown In the ~ 9th century.- -The dominant mode of travel prior to transit was the walk mode, hence the geographic spread of cities was limited to walking distances tSchaeffer and Scalar. 19753. The arrival oftransit resulted in decentralization and relatively Tower density development [Muller, 1995; Black. 19954. PBQD tI995] summarize the classic work by plainer tI962], who traces how the extension of electric streetcar lines to suburbia arour~d He turn-of-the-century led to massive decentralization in Boston, the San Francisco Bay Area, and Southern California Smerk tI967] estimated that as much as one quarter ofthe U.S. population resided at that time in urban and suburban areas whose spatial organization was shaped by the street car. Black tiffs] also cites a study by Hoyt tI933] who confine that transit with its most important routes being the radio lines that met at the city center, strengthened the CBD and made cities monocentnc. The main alternative mode to transit during the latter ~ gth century and eddy part of the 20th centur was the walk mode hence the significant impact transit made to urban development and structure. However, in the latter part of the 20th century, transit's major modal A-21
TCRP H-12 Final Report competition has been the auto, a far more convenient mode, Thus as Knight tI980] points out, transit improvements today do not result in the kind of drastic improvement in overall accessibility associated w~ earlier transit improvements. Hence new- transit lines are less likely to affect urban form they were they did several decades ago tBlack, 19953. Evidence from recent studies has been mixed, with some concluding transit to have a major impact on urban form under "naht" conditions. ___1_ ·1 _ _ `1_ _ _ _ 1~ _ _ _ 1 ~ _ ~ ~1 ~ 1 _ _ (_ 1 _ _ _ · , ~ ~ ,1 , t wrn~e owners nave concluded anal ~ nas tar less impact. come of these squares are presented below. Heilburn tI981] reports the findings of Webber's tI976] assessment of BART after a year of operations. This was an extremely short period for assessing its possible impact on urban form, hence the evaluation was pnmarily on the other touted benefit of diverting suburban car users to rail. From a survey conducted In 1976, Webber found that only about 44,000 daily B ART users were recorded as previously making the trip by car. This represented about 35 percent of BART trips. UnexpectedIx7, however, "about 50,000 BART riders, and as much as half ofBART's transbay traffic, were diverted hom the over form of mass transportation -- He motor bus" [Heilburn, ~ 98 I, p. 2553. Overall, Webber concluded that BART had not effected a significant charge In auto use habits. Giuliano tI995] also reports on a survey of the impacts of B ART, five years after it came into service. Although still relatively not too long a period for assessing its impact on urban for, the findings are nevertheless important. The survey showed that transportation access was seldom a factor in job location choice, and access to B ART was a minor consideration in household location choice. Important factors influencing household location decisions were housing tYne. general ~r _ . ~· ~t ~. · . · ~At ~. _% ~ ~ 1, . ~ access to the workplace and nelg~oor~ooa characteristics. further access to bAK1 was not an important factor in the location decision of employers -- important location factors included site availability-. price, and proximity to other fimns. The conclusion of the B ART-study team was that BART's impact on land-use after its five years of operation was insign~ficar~t tGinTiano, ~ 9954. Cervero arid Landis tI997] conducted an in-depth study on the Impacts of BART on land-use and development twenty years abler service started. The authors found that land-use changes associated with BART have been largely localized, limited to downtown Sari Francisco and Oakland and a handfi~} of suburban stations. Elsewhere, they report Hat few land-use changes occurred either due to neighborhood opposition or a lack-lustre local real estate market. Although crediting B ART with bringing about a multi-centered regional settlement pattern, they also note that it has done little to stem the tide of freeway-oriented suburban employment growth over the last two decades. Office development seen near some of the suburban stations is found to pale in comparison to the amount of floor space built in non-BART freeway corridors. The multi-famils~ housing built around several suburban stations is in a large part credited to lock redevelopment authorities who helped leverage _ ~ . , . _ .. . . . . ,. . ~ . . . .. . . . ... . . .. these projects by providing various financial Incentives and assistance with land assemblage. Cervero and Landis concluded that BART, in and of itself, has not been able to induce large-scale land-use changes although under the right circumstances, it appeared to have been an important contributor. Stronger public policy Initiatives was stated by the authors as a necessary condition for BART to be able to achieve the compact, multi-centered urban form initially envisaged. Workman and Brod t! 997] investigated the neighborhood benefits of rail transit accessibility. A hedonic price function specified in the form of a multiple regression model, was used to estimate property values and the impact of proximity to rail stations. Two sets of data from different cities were employed in the study. The first source of data was collected for homes around Pleasant Hill, A-22
TCRP H-12 Final Report one of the B ART stations In the San Francisco Bay Area. The second data set was collected for homes around Free stations on Po~and's MAX light rail system. The researchers found proximity to Pleasant Hill station to be a key determinant of property values in the Pleasant Hill neighborhood. Their results showed that single family homeowners were willing to pay. on average, nearly $ ~ 6 more in home-pr~ce for each foot closer to BART within the study area. Further they found that homeowners were willing to pay nearly $8 in home-price for every foot further from the freeway interchange nearest the study area. The Portland data, however. gave much weaker resets. Indeed. based on the entire three-station Portiar~d data He key viable of "distar~ce of home from station" had a coefficient sign contrary to the expectation ofthe researchers. One interpretation given to the contrary- result was that possibly the effects of light rail could be different from that of heavy rail (e.~.. BART). Further. the light rail line runs close to a major arterial, whose effect the researchers thought confounded the effect of distance to the station. Massaging of the data Cats done to include homes within 2~00 and :280 feet Of the line. This time amours, He results Bough very weak, agreed with intuitive expectations. The authors concluded from their mode] that property values increased by about $0.76 for every foot closer to light raid within the 2500-5280 feet bared defined. A similar study on property values, which focussed on the same stations in Portland by Al- Mosaind, et al [19933 ford a statistically insignificant property value premium for station proximity. Black tI995] also reports a study by Arrington tI9893. who tabulated $693 million of new development adjacent to stations on the Portland light rail line after the decision to build it in ~ 979. Arrington also suggested that light raid might have a greater impact that heard rail on urban form because light rail operates on land-surface arid makes adjacent businesses visible to thousands of passengers. which is contrary to the suggestions of Workman and Brod tI9973. Nelson and Sanchez tI997] came up with mixed results when they studied the influence of the Metropolitar1 Atlanta Rapid Transit Authority (MARTA) on population and employment location. While the population of the Atlanta metropolitan area grew by 29 percent or nearly half-a-million people, He population within one-half mile of MARTA stations fell by more than ~ ~ percent, with the regional share of population living within one-half mile falling bit one-third. - The regional share of employment within walking distance of MARTA stations also fell by one-quarter. However, actual employment within half-a-mile of the stations rose 13 percent, indicating that job growth outside the station area of Influence was much greater. Hunt, et al. [1994] in Heir investigation ofthe relative impacts of various transportation and non- transportation factors on the perceived attractiveness of residential locations using stated preference techniques. concluded that the transport system has ~ strong influence on the attractiveness of residential locations. The authors found from their empirical study that being within walking distance of a light-rai] transit station was worth about 217 Canadian dollars per month. Gentlemen. ef al. tI983] undertook an extensive study to identify the effects of two major rail investments in Glasgow: the British Rail (BR) Argyle Line and the modernized Underground on travel, activities and land-use in Glasgow. The study conducted several before-and-aher surveys: a survey of all travel by households within the raid combor, survey s of BR. bus and underground passengers, arid surveys of users of shops, workplaces, hospitals. health centers. and a number of A-23
TCRP H-1 ~ Final Report leisure facilities. The latter surveys were supplemented by data over a longer period on planning applications and house paces, rail ticket sales and turnover and catchments from shops and other services. Since this study was conducted one year aver the opening ofthese services, major changes in land-use and property development were not expected. since these are more long term. Further, neither rail services was entirely new: the Argyle line replaced a line closed ~ 5 years previously, and the Underground was withdrawn for 3 years. Thus, the previous services helped shape the urban structure existing then, and therefore may have limited the impact of the "new" services. Nevertheless, the authors reported an increase, relative to the rest of Glasgow in planning applications for retail, office, storage and manufacturing development in the inner areas served by the Argyle Line and Underground. Based on sales data, the authors also reported clear evidence of a reversal of a downward trend in house prices, relative to the rest of Glasgow, in the areas associated with the new services. The authors also reported no systematic take-up of v acant land near raid lines, which could be identified as being due to the new- services. Small increases in population, stemming population decline, were noted around some of the Underground stations though further evidence was needed to confirm this as a trend. The raid lines also had some impact on travel. After the opening of the BR Argyle line, 24 percent of the passengers were new to the BR network. 75 percent of them had previously made the same journey by another mode and the rest were making new trips. Ofthose who had changed mode. 72 percent were from bus, 19 percent from car and 9 percent previously walked. All Underground travelers were effectively new users because the service had been closed for several years. The authors report that the proportion who transferred from bus or walk was similar to BR, but here, only 9 percent previously used car. Other studies on the impact of transit and development include that by Baker tI983] who fourth that the Washington D.C. Metro attracted considerable non-residential development to station areas. Baker reports that between 1979 and 1982. $2 billion out of $3.7 billion of public and private nonresidential construction in the metropolitan area occurred within 7/10's of a mile of stations. Green and James tI993] in a study of the Washington system as well, also concluded that the rail system had significant development impacts, noting that rail corridors had developed more than other places, and within raid corridors, that station areas had grown much more than other areas. Knight and Trygg tI977] examined the land-use impacts of rapid transit in several North American cities. Toronto, one of the few North American cities in which transit has done relatively well, is orate of those reviewed. Significant urban development is reported by the authors to have taken place in the lands adjacent to the rapid transit corridor dunng the period of the transit systems staged development. Heenan t1966, 1968] in his papers on the impact rapid transit had on land values in Toronto, claimed that the subway was responsible for igniting $10 billion in development along its route, and also for two-thirds of the appreciation in the physical value of all lands and facilities in Metropolitan Toronto during a 10 year period. Heenan's findings were however contradicted by Meyer and Gomez-Ibanez tI981], who studied the Toronto case, and concluded that the subway had a much less impact. Knight and Trygg t! 977] also found Heenan's findings of the Toronto rapid-transit impact on lar~d-use to be considerably overstated. Knight and Trygg also report the cor~clusions of two statistical studies conducted with the objective of assessing the subway systems impact on land value. The first of these. bar Abouchar tI973], concluded that the subway had no impact on the value of properties studied. Data for A-~)4
TCRP H-17 Final Report Abouchar's study however, came from the period eleven years after the first subway- line was opened, and also after most of the rest of the system was either approved or well under construction. Thus, this study was criticized on the grounds that it was likely most of the impact on values had occulted prior to the period from which the analysis data was drawn. The second by Dewees tI975], however, concluded that the subway line did have a positive impact on residential values, but did not attempt to quantify this. Davies tI 974] examined the impact of the Yonge street subway line on population density, charges. Census data for the years of 1951 1956 and 1961 were used in the study. Davies concluded that there was no effect in 1956, which was two years after the line was opened. However, density near the line was fourth to have increased significantly; faster between 1956 and 1961 compared to areas much farther array from the fine. Knight arid Trying after examining the Toronto case more closely concluded that it was quite unique. in that zoning was closely coordinated win the rail lines. a coordination that has not been common for new transit facilities in the United States. In short that transit. in arid of itself. is unlikely' to have chalked up the relative success seen in Toronto, without the complement of urban policies to ensure this. Deakin tI990] notes that the impact of raid transit on urban development has been mixed; in some cases they have helped achieve more compact and dense growth, but at the same time, at a regional level, these same rail systems have made previously poorly accessible suburban locations able to support suburban development. Deakin. similar to Knight and Trying, echoes the important impact that transit availability and quality have on location and land-use, but goes on to stress that they represent just two of many other factors. Thus, urdess the remaining factors are also supportive of transit, then transit investments would be unlikely to make a difference on development. Knight and Trvgg tI977] also surveyed rapid transit systems and lines in a number of other cities including Montreal, Boston, Chicago, Cleveland and Philadelphia. In Montreal after reviewing the evidence and local studies. they concluded that the nature and intensity of retail shopping in the COD had been influenced by Metro, although other factors, such as the developable Iand, had played a strong role in the downtown revitalization. Outside the CBD, with the exception of two stations, little development had occulted. In Phiiadelphia. the focus Bras on the Lindewold high speed line. Here the authors concluded that the single most important impact of the line in the areas it served was the substantial increase in residential properly values. Notwithstanding He evidence that office development occurred who equal or greater intensity in some other areas of Philadelphia and South Jersey not served by the line, the line was cited for its contributory effect on the location of new suburban offices and apartment developments nearby, and its influence in local zoning decisions as well as in actual investments. Evidence from the extensions and improvement to the Boston rail system on land-use appeared to be mixed. In North Quincy, the line service was cited as a strong influencing factor in the selection of that location for several major developments, though other important complemental,- factors. such as land a~'a~labilin,;. reasonable land prices. etc. played a role. North Quincy aside, verily little impact was observe ed at the stations on the line. Improvements to the Chicago system generated no land-use impacts. Several reasons are given for this, notably, the already well developed nature of the downtown district and the high land costs. Similar to Chicago, the impact of the Cleveland rapid transit system was found to be minimal. Clearly, Dom the foregoing rail transit is one of several factors that contribute to the shaping of art urban area. The rail lines and stations in arid of themselves cannot shape urban form without the other factors also in sync. A-~:
TCRP H-12 Final Report A-26
