trb14-bridging the gap between the new urbanist ideas and transportation planning practice

M.Zhang, H. Pang, and A. Kone, TRB 2014 1

BRIDGING THE GAP BETWEEN THE NEW URBANIST IDEAS AND 1

TRANSPORTATION PLANNING PRACTICE 2

3 Submitted to the 93rd TRB Annual Meeting, January 12-16, 2014, Washington, D.C. 4

5 6 7

8 Ming Zhang*, Ph.D. 9

Associate Professor, University of Texas at Austin 10 1 University Station, B7500 11

Austin, TX 78712, USA 12 Tel. 512-471-0139 13

FAX: 512-471-0716 14 Email: [email protected] 15

16 Hao Pang 17

MSCRP Candidate, University of Texas at Austin 18 1 University Station, B7500 19

Austin, TX 78712, USA 20 FAX: 512-471-0716 21

Email: [email protected] 22 23

And 24 25

Alex Kone 26 Planner, Capital Area Metropolitan Planning Organization 27

505 Barton Springs Road, Suite 700 28 Austin, TX 78704 29

Email: [email protected] 30 31 32

Word Count: body & ref.: 5171 + table: 7 + figure: 2 = 7421 33 (* Corresponding author) 34

35

mailto:[email protected]


ABSTRACT 1 Metropolitan planning organizations (MPOs) have become increasingly interested in 2

urban design solutions to transportation problems. Many design ideas under the umbrella of the 3 New Urbanism appear promising; yet in practice they hardly get fully implemented in the 4 standard transportation planning procedures. Partly it is because of remaining skepticisms on the 5 role that the built environment could play to influence travel. Partly it is due to technical, 6 institutional, and political issues. This paper sets a two-fold objective. First, it intends to 7 contribute to the continuing debate on urban form-travel connection by adding further empirical 8 evidence from the Austin, TX region. Differing from most existing empirical work, the paper 9 studies and reports evidence in metrics commonly used by transportation planners and engineers 10 for travel demand analysis. Second, it demonstrates ways to integrate urban design variables in 11 transportation demand analysis. The study identifies 42 mixed use districts (MXD) in the Austin 12 region. It then analyzes the following aspects of travel behavior in MXDs and non-MXDs: trip 13 production rates, trip duration and length distribution, internal rate of capture, departure time, 14 and travel mode choice; which correspond to the first three of the Four-Step planning models. 15 The study contributes to transportation planning and policy making in Central Texas by 16 providing local empirical evidence on urban form-travel connection. The study’s method and 17 process can be of interest to a broad audience in academia and practice. 18 19 20 Keywords: 21 Travel behavior, urban form, four-step modeling, the New Urbanism, Austin TX 22 23


1. INTRODUCTION 1 Transportation professionals have become increasingly interested in new urban design ideas 2 when seeking solutions to such enduring transportation problems as roadway congestion, vehicle 3 emissions, and traffic accidents. Examples of the design ideas, under the umbrella of the New 4 Urbanism, include transit oriented development (TOD), traditional neighborhood design, 5 compact development, mixed use development, and pedestrian/cyclist friendly environmental 6 design (popularly referred as the D-factors). Many MPOs (metropolitan planning organization) 7 have created programs around the New Urbanism ideas. In Texas, for example, the Capital Area 8 Metropolitan Planning Organization (CAMPO) in the Austin area is adapting a regional growth 9 concept of “Activity Centers” featured with high-density, mixed use districts (MXD) for its 10 long-range transportation plan (1). In the Dallas area, the North Central Texas Council of 11 Governments (CTCOG) is expanding its program of Transit-Oriented Development (TOD) along 12 with the expansion plan of the Dallas Area Rapid Transit (DART) network (2). In the greater 13 Houston area, the Houston-Galveston Area Council (H-GAC) has been implementing a “Livable 14 Centers” program that promotes clustering development of jobs, shopping, entertainment, and 15 housing (3). 16

However, when it comes to the making of long-range regional transportation plans, the 17 urban design ideas hardly get fully implemented in the standard procedures, for a variety of 18 reasons. First of all, there remain skepticisms on the role that the built environment could play to 19 influence travel (e.g., 4). The place-making initiatives such as those mentioned above may help 20 improve livability in selected neighborhoods; whether they will generate significant 21 transportation benefits at the regional scale are still unconvincing to many transportation 22 engineers and to some urban planners as well. Furthermore, there are technical and institutional 23 issues (5). As of today the majority of MPOs in the US apply the Four-Step modeling procedures 24 (i.e., trip generation, trip distribution, modal split and traffic assignment) for demand analysis 25 and forecasting. These modeling tools were first developed in the 1950s mainly for 26 highway-based transportation planning. They are rather insensitive to changes in urban form at 27 the site scale as the New Urbanists propose. Take the first step, trip generation modeling, as an 28 example. Typically trip productions and attractions are estimated based on the trip rate tables 29 recommended by national agencies such as NCHRP and ITE or developed by local regions (6). 30 The tables provide trip rates varying along income, household size, vehicle ownership, and 31 metropolitan populations. Urban design variables rarely enter into trip generation equations. 32

There have been efforts to better integrate urban design ideas into transportation planning 33 practice. A variety of tools has been or is being developed (7). Generally speaking, the efforts fall 34 into two categories. One is called the ‘post-processing’ approach (8). The approach takes the 35 output of the conventional four-step models as input and post-processes travel outcome by 36 making empirical adjustments. For example, empirical studies have reported travel behavior 37 elasticities of urban form attributes such as density, land use mixture, and intersection 38 configuration (9). The post-processing approach applies the elasticities to adjust up or down the 39 modeled trip volumes, modal splits and other aspects of trip making. While the approach offers a 40 improved solution technically, it may not work due to policy or political constraints. In Austin, 41


TX, for instance, the governing board of CAMPO does not approve plans that are made using 1 borrowed data. 2

The second approach is what we call ‘pre-processing’, referring to the effort of 3 developing large scale, integrated land use-transport models, for example, UrbanSim, PECAS, 4 and region-specific models. The effort attempts to develop new modeling tools that eventually 5 replace the conventional, highway focused four-step modeling procedures. Nevertheless, despite 6 major progress achieved in the field, the integrated land use-transport models remain operational 7 largely in academia. It may take years or even longer for them to become a common practice 8 among MPOs due to known technical and institutional reasons. 9

This paper sets a two-fold objective. First, it intends to contribute to the continuing 10 debate on urban form-travel connection by adding further empirical evidence from the Austin, 11 TX region. Differing from most existing empirical work in the area, the paper studies and reports 12 evidence in metrics commonly used by transportation planners and engineers for travel demand 13 analysis. Second, it presents what we call a ‘present-processing’ approach to integrating urban 14 design ideas with transportation practice. The proposed approach lies between the pre-processing 15 and the post-processing approach mentioned above. It incorporates urban design variables 16 directly into the conventional demand analysis procedures. A number of MPOs throughout the 17 country has made such efforts, for example, in Southern California (10). The case example of 18 Austin, TX presented in the paper adds to the efforts that help develop a practical solution to 19 bridge the gap currently existing between the New Urbanist ideas and transportation planning 20 practice. . 21

The remainder of the paper includes three parts. Part II describes the rationale and 22 method of the study. It explains why and how MXDs are identified in the Austin area. The part 23 also explains the process of geocoding residents’ activity and travel locations recorded in the 24 Austin 2005 Activity and Travel Survey. Linking the locations in GIS allows us to derive trip 25 lengths along the transportation network. Part III reports study results and shows the operational 26 ways of integrating urban form features into three of the Four-Step models. Specific travel 27 aspects studied include trip production, trip duration and length distribution, internal rate of 28 capture, and departure time. Multinomial logit models of travel mode choice are also estimated 29 for different trip purposes, with urban form variables included. 30

2. RATIONALE AND METHOD OF THE STUDY 31 In essence the New Urbanist design ideas are not entirely new. Rather they are the 32 re-representation of the built environmental amenities observed in the traditional neighborhoods 33 built around the beginning of the 20th century. Austin, like other communities in the nation, has 34 traditional neighborhoods throughout the region. There are also some contemporary places with 35 features promoted by the New Urbanists. Studying the travel behavior of those living in the 36 traditional neighborhoods and alike would extract empirical evidence (or lack of it) from Austin 37 on urban form-travel connection, which can be used for travel forecasting for design-based future 38 scenarios. Since the evidence comes locally, plans made with the empirical support would be 39 more receptive to local communities and political forces. For this reason, the first step is to 40


identify the MXDs in the region. 1 2

2.1 Identification of MXDs 3

Focusing on MXDs was originally part of a national study sponsored by the United States EPA 4

and the Institute of Transportation Engineers (ITE) to improve traffic impact analysis pertaining 5

to MXDs. The Austin region was one of the six cases for the national study (for some 6

publications out of the national study, see (11), (12)). Identifying MXD’s in the Austin region 7

followed aa ‘bottom up’ approach taken by the national study. Specifically, it was based upon 8

local knowledge of city officials, professional planners, CAMPO staff and academic experts. The 9

process involved three working steps. First, a list of 49 communities in the region was created 10

and the contact information of representative planners or public officials collected. The research 11

team then interviewed by phone the planners or officials, asking them to identify MXD’s based 12

on their professional and personal knowledge of their own communities. Instructed by the 13

national study, the interviewee was first given a definition of MXD: “A mixed-use development 14

or district consists of two or more land uses between which trips can be made using local streets, 15

without having to use major streets. The uses may include residential, retail, office, and/or 16

entertainment. There may be walk trips between the uses.” 17

The MXD definition given in this study was relatively expansive and inclusive in order to 18 garner a significant number and variety of samples for statistical analysis. The study did not 19 establish criteria for minimum size, density, or number of land uses for a MXD. A general 20 reference is the area reachable by walking. For example, a circle of ¼~ ½ -mile in radius has an 21 area of approximately 125~502 acres. Downtown districts (excluding downtown Austin) and 22 traditional neighborhoods were the primary areas cited by local planners. 23

The second step includes two work sessions with experts from CAMPO and from the 24 University of Texas at Austin (UTA). The experts were presented with maps of land use and 25 street network for the study area and asked to draw on the maps the MXD-like developments. 26 CAMPO staff reviewed the preliminary set of MXDs and offered their own identification of 27 MXD samples. UTA planning faculty members who have decades of working knowledge on land 28 use and community development in Central Texas were invited to provide their expert 29 knowledge of Central Texas geography and urban planning. 30

Finally, the research team using land use GIS and Google aerial photos refined the MXDs 31 identified from previous steps and finalized the boundaries of the MXD’s to complete the sample 32 set. The final sample set contains 42 MXD’s in the region. The expert-GIS combined approach is 33 superior to the GIS-only approach as the latter cannot distinguish among different functions that 34 are simply spatially adjacent but actually no synergetic relationships due to physical (e.g., a fence 35 not recorded in the GIS database) or non-physical barriers. 36

37


1

2.2 Geocoding of Trip Records in GIS 2

Travel data used for this study comes from the 2005 Austin Activity Travel Survey. The survey 3 records geographic coordinates of activity locations and trip ends (origins and destinations) of 4 the surveyed travelers. These trip ends are geocoded in TransCAD GIS. Network distance is 5 estimated based on the assumption that the traveler took the shortest path in network distance 6 between trip origin and destination. 7

8

FIGURE 1 MXD Distribution in the Austin, TX Region


1 2

Table 1 reports descriptive statistics of the households located inside and outside MXDs. 3 Notably, households outside MXDs having an average number of 2.82 persons per household are 4 larger than those inside MXDs (2.29 persons per household). The statistical test of difference in 5 sample means suggests that the difference in average household size is significant. This 6 difference exists mainly due to a larger number of non-working dependents in non-MXD 7 households than MXD households (The MXD and non-MXD households appear to have the 8 same average number of workers). On average, MXD households exhibit similar characteristics 9 to the average non-MXD households in terms of income, vehicle ownership, and tenure. The 10 descriptive statistics shown in Table 1 suggest to a certain extent the representativeness of the 11 sampled MXD households for the 2005 surveyed households. 12

13 14 15 16 17 18 19 20

FIGURE 2 Geocoding of Trip Ends and Trip Distance Estimate


TABLE 1 Sample Household Characteristics 1

HH Inside MXDs (n=69) HH Outside MXDs (n=1,354)

Variable Mean Std. Dev Min Max Mean Std. Dev Min Max t-test

# Persons in HH 2.29 1.20 1.00 5.00 2.82 1.54 1.00 13.00 -2.75

# Workers in HH 1.08 0.83 0.00 2.00 1.12 0.80 0.00 2.00 -0.44 HH Income (2005 $1000’s) 45.35 36.30 5.00 150.00 54.38 38.33 5.00 150.00 -1.86

Income/Person (2005 $1000s) 22.21 17.19 2.50 87.50 22.92 18.47 0.83 150.00 -0.30

Vehicles in HH 1.80 0.96 0.00 4.00 1.91 0.91 0.00 7.00 -0.93

Vehicles/Person 0.87 0.46 0.00 3.00 0.79 0.41 0.00 5.00 1.59

Vehicles/Worker 1.24 0.46 0.00 2.00 1.41 0.71 0.00 5.00 -1.54

Bikes in HH 0.85 1.39 0.00 7.00 1.16 1.59 0.00 10.00 -1.56

Years in Residence 3.80 1.73 0.00 5.00 3.98 1.58 0.00 5.00 -0.89

2

3. EMPIRICAL ANALYSIS AND RESULTS 3

This section reports individual and household travel characteristics from analyzing the 2005 4 Austin Activity Travel Survey. The main interests are the differences in travel behavior 5 between those who are associated (living in, traveling from or to) with the MXDs and those who 6 are not. Seven aspects of travel behavior analyzed include trip production rates, trip duration and 7 length distribution, internal rate of capture, departure time, and travel mode choice; they 8 correspond to the first three of the Four-Step models. Person miles of travel (PMT) and vehicle 9 ownership were also part of the analysis. They are not reported here due to space constraints. 10

3.1 Trip Generation by MXD Travelers 11

Modeling trip generation (production and attraction) is the first and a critical step of travel 12 demand analysis. Transportation professionals use three primary methods to model trip 13 generation: cross-classification, regression, and discrete choice. Here we use cross-classification 14 methods to analyze trip production in MXDs vs. non-MXDs. 15

Cross-classification methods separate the population in the study area into relatively 16 homogenous groups based on selected socioeconomic characteristics and then derive empirically 17 trip rates for each group. CAMPO in their 2005 base models used three variables for grouping: 18 household size, income, and number of workers in households. In this case study, we consider 19 household size only due to limitation in sample size of MXD households. The sample descriptive 20 statistics (Table 1) show an equal per capita income between MXD and non-MXDs. Hence it is 21 reasonable to say that the income effects on trip rate variation between MXD- and non-MXD 22 travelers are partly controlled. Aside from the socioeconomic variables, a spatial variable is used 23 for grouping, that is, home location in MXDs and non-MXDs. 24


Table 2 reports trip production rates by purpose, household size, and MXD indicator. 1 Note that the 69 households living in Austin’s MXDs made 139 home-based work (HBW) trips, 2 giving an average rate of 2.014 person trips per household. For those living in non-MXDs, the 3 rate is 1.636. The sample size is relatively small yet the result is consistent with that from 4 multi-region national study. Nevertheless, in practice, should the CAMPO’s three-way tabulation 5 for trip rates be used, additional sampling of trip-making would be needed from the MXDs. 6 Validation is also necessary as done by SANDAG (10). 7

CAMPO splits Home Based Work (HBW) Trips into Direct, Strategic and Complex, 8 aiming at obtaining additional insight into the mode choice decision. A Direct HBW (HBW-D) 9 trip is part of a trip tour (or trip chain) that consists of both home-to-work and work-to-home 10 trips as being direct. Some travelers may make intermediate stops between home- and work-ends. 11 If the stops are for personal convenience, for example, buying a cup of coffee and (by a rule of 12 thumb) lasts less than 5 minutes, the trips are still considered as being linked and direct. 13

If the trip links an intermediate stop for the traveler to fulfill child-caring obligations, for 14 example, dropping off or picking up a child at day-care, nursery school, baby sitter, pre-school, 15 elementary or secondary school, the trip is classified as HBW Strategic (HBW-S). A Complex 16 HBW (HBW-C) trip is part of a trip “tour” that consists of one trip between home and work and 17 another trip between home and work which involves an intermediate stop at any destination for 18 personal or family business except for school drop-off/pick-up (HBNW). Complex work trips are 19 part of a trip “tour” where workers’ choice of mode is conditioned to some extent on the tasks 20 that the worker must accomplish on either one or both legs of the journey between home and 21 work. 22

This study follows CAMPO’s method to split HBW trip chains into Direct, Strategic, and 23 Complex. Table xx reports derived trip chain rates by household size and home locations in 24 MXDs vs. non-MXDs. Despite limitation in sample size, the cross-tabulation shows interesting 25 trip chaining patterns: MXD households make slightly more HBW Direct trips (1.507 person 26 trips/household) than non-MXD households (1.415 person trips/household), but much less 27 HBW-Strategic trips (0.043 and 0.080 person trips/household in MXDs and non-MXDs, 28 respectively). For HBW-Complex, the average MXD trip chain rate (0.464) is much higher than 29 that for non-MXD (0.14). The variations may be attributed to the siting of schools and locations 30 of community services. In MXDs, schools are relatively close to homes. School-age children are 31 more likely to go to schools by themselves than those in non-MXDs (13). Similarly, banks, 32 stores, hospitals and other services tend to be more conveniently located in MXD neighborhoods 33 than in non-MXDs. MXD residents thus are more likely to chain these activities with their 34 commuting than non-MXD residents. 35

For HBNW, MXD residents make more retail trips than non-MXD residents (1.869 vs. 36 1.537), likely due to more convenient access to retail shops that induce more trip making. In 37 contrast, non-MXD residents make more NHB-other trips (1.66) than MXD residents (1.016). To 38 understanding this difference, we may speculate that the non-MXD residents live relatively 39 farther away from ‘other’ service destinations and are thus more likely to perform NHB activities 40 once they are away from homes. 41


1 2

TABLE 2 Trip Production by Purpose and MXD (Person trips per household) 3

# of HHs

# of Trips

Trip Rates

Trip Rates by Household Size (# of Persons in HH)

One Two Three Four Five+

Regional 1389 2298 1.654 0.613 1.389 2.279 2.112 2.254

MXD 69 139 2.014 0.647 2.765 2.875 6.000 0.733

Non-MXD 1320 2159 1.636 0.611 1.337 2.240 2.044 2.392

HBW_Direct MXD 69 104 1.507 0.588 2.000 2.500 3.000 0.533

Non-MXD 1320 1868 1.415 0.518 1.203 1.894 1.645 2.193

HBW_Strategic MXD 69 3 0.043 0.000 0.000 0.000 0.750 0.000

Non-MXD 1320 106 0.080 0.000 0.000 0.203 0.202 0.060

HBW_Complex MXD 69 32 0.464 0.059 0.765 0.375 2.250 0.200

Non-MXD 1320 185 0.140 0.093 0.134 0.142 0.197 0.139

HBNW Retail MXD 61 114 1.869 1.000 1.647 2.313 2.250 3.667

Non-MXD 1328 2041 1.537 1.053 1.641 1.589 1.675 1.634

HBNW Other MXD 61 134 2.197 1.222 1.647 2.063 3.000 6.500

Non-MXD 1328 3568 2.687 0.862 1.610 2.276 4.114 6.543

NHB_Work MXD 61 51 0.836 0.333 0.765 1.250 1.500 1.000

Non-MXD 1328 1056 0.795 0.320 0.612 0.931 1.132 1.251

NHB Other MXD 61 62 1.016 0.444 0.647 0.625 1.750 4.333

Non-MXD 1328 2205 1.660 0.880 1.187 1.451 2.132 3.571

Notes 4

5 6

The above analysis of trip production can be performed for trip attraction as well. The 7 results demonstrate that urban form does have an influence on trip generation; adding a spatial 8


dimension to the conventional tool would enable the capture of urban form effects on trip 1 making. 2

3

3.2 MXD Modifying Trip Distribution 4

3.2.1 Trip Duration and Length Distribution 5

The analysis of trip length (time and distance) provides essential information needed for trip 6 distribution modeling. 7

Table 3 shown below compare average trip length (times and distances) between MXD 8 and Non-MXD trips for four trip purposes. MXD trips refer to those with trip ends, either origins 9 or destinations, falling within MXDs. Trip times are derived from travel logs of departures and 10 arrivals reported by the surveyed individuals in the 2005 Austin Activity-Travel Survey. A 11 number of records show exceptionally long trip times. The analyses for this study exclude the 12 records with one-way trip time longer than 180 minutes. The last column of each table shows 13 statistical test of the difference in average trip length between MXD and non-MXD trips. 14

On average MXD trips are 0.9 minute shorter than non-MXD HBW trips. However, test 15 of the difference in sample means suggests that the difference in average trip times between 16 MXD and non-MXD trips is attributable to sampling errors. For HBW travel purpose, average 17 MXD trip distance (8.33 miles) is statistically significantly shorter than non-MXD trips (9.16 18 miles), indicating a higher average travel speed for non-MXD trips than for MXD trips. 19

For NHBW trips, MXD travelers travel shorter in both time and distance than non-MXD 20 travelers. The differences in average trip length are statistically significant. For non-home based 21 other trips, there are no statistically significant differences in average trip time and distance 22 between MXD and non-MXD trips. 23

Given the difference in trip length between MXD and non-MXD trips, we estimated 24 parameters of trip length distribution (TLD) for MXD and non-MXD trips where TLD is 25 assumed to take a Gamma function (Table 4). Figure 3 show images of the TLD. The information 26 can be used to re-calibrate friction functions that would become sensitive to urban design 27 variables. 28

3.2.2 Internal Rate of Capture and Departure Time Analysis 29

The study also looked at two other aspects of travel that would affect trip distribution 30 modeling. Internal rates of capture are calculated for each of the 42 MXDs in the study area. On 31 average, 7.4% of MXD trips are internal, with both trip origins and destinations falling within the 32 MXD boundaries. The highest rate of internal capture is nearly 35%. Table 5a compares internal 33 trip rates of MXDs with TAZs in comparable size. Table 5b reports departure times of MXD vs. 34 non-MXD travelers. Travelers living in MXDs leave home in the morning approximately 9 35 minutes later those living in non-MXDs. Departure time analysis provides information for 36 calibrating time-of-day distribution and for better understanding peaking effects of traffic. 37

38


TABLE 3 Average Trip Time and Distance by Trip Purposes 1

Home-Based Work Trips

MXD Trips Non-MXD Trips t-test

Variable Mean Std. Dev Min Max Mean Std. Dev

Min Max

Time (minutes) 22.92 17.30 2.00 165.00 23.82 16.27 1.00 175.00 -0.93

Distance (miles) 8.87 8.33 0.12 41.96 10.75 9.16 0.01 49.30 -3.90

n=393 n=1530 Home-Based Non-Work Trips

MXD Non-MXD t-test Variable Mean Std. Dev Min Max Mean Std. Dev Min Max Time (minutes) 15.47 10.94 1.00 90.00 14.56 11.90 1.00 180.00 2.50 Distance (miles) 6.07 6.44 0.01 45.82 5.41 6.40 0.01 65.87 3.13

n=1104 n=6245 Non-Home-Based Work Trips

Variable Mean Std. Dev Min Max Mean Std. Dev Min Max Time (minutes) 13.12 10.08 1.00 75.00 16.09 13.98 1.00 120.00 -3.80 Distance (miles) 5.39 6.25 0.10 43.66 6.59 7.22 0.03 43.03 -2.69

n=310 n=694 Non-Home-Based Other Trips

Variable Mean Std. Dev Min Max Mean Std. Dev Min Max Time (minutes) 13.11 11.79 1.00 120.00 13.25 11.40 1.00 75.00 -0.25 Distance (miles) 4.68 5.68 0.01 35.50 4.86 6.26 0.01 40.01 -0.66

n=571 n=1641 2

TABLE 4 Calibration of Gamma Functions for Trip Length Distribution 3

MXD Non-MXD A B C A B C HBW 0.0040 1.3560 -0.1044 0.0046 1.2302 -0.0940 HBNW 0.0116 1.2471 -0.1453 0.0182 1.0361 -0.1410 NHBW 0.0224 1.0277 -0.1545 0.0205 0.8105 -0.1126 NHBW 0.0332 0.7224 -0.1314 0.0359 0.6452 -0.1242 4

5


TABLE 5a Internal Trip Rates in MXDs vs. in TAZs

MXD (n=42) TAZ* (n=450) Variable Mean Std. Dev Min Max Mean Std. Dev Min Max

% Internal 7.64 9.28 0.00 34.78 4.57 9.37 0.00 50.00

Internal Trips 3.14 4.98 0.00 23.00 1.06 2.36 0.00 15.00

Total Trips 31.67 21.90 3.00 90.00 14.54 14.25 0.00 115.00

Area (acre) 205.89 113.09 25.10 549.50 253.27 143.04 25.44 547.97

*Note: Only include those TAZs comparable in size to MXDs.

TABLE 5b Departure Time for Morning Trips

Trip Ends in MXDs Trip Ends outside MXDs

Variable N Mean Std. Dev Min Max N Mean

Std. Dev Min Max

All Trips 204 544.9 123.8 0 710 1,028 533.6 110.9 0 715

HBW 33 568.3 167.4 0 708 121 557.5 177.5 0 715

HBNW 170 539.9 113.8 0 710 904 530.2 98.4 45 715

Note: Times measured as minutes from midnight.

1

3.3 Mode Choice Analysis with MXD Variables 2

The relationship between the built environment and travel behavior has been studied extensively. 3 Mode choice analysis for this study utilizes two sources of data: the 2005 Austin Activity Travel 4 Survey and the 2005 Transit On-Board Survey conducted by CapMetro. 5

The choice modeling began with specifications of nested logit (NL) structures. An NL 6 model structure groups similar modes into general categories, for example, bus, metro, and taxi 7 being “transit” and driving-alone and car-pool being “car”. It then specifies the models in a tree 8 structure with transit vs. car at the upper level and specific modes in the lower level. In contrast, 9 the conventional joint multinomial logit (MNL) structure treats all modes equally at the same 10 level. While NL and MNL assume different decision behavior, the choice between the two model 11 structures is determined empirically. Selection of the final models for reporting purposes in this 12 paper considers three aspects of model performance: 1) the sign of coefficients for system and 13 socio-demographic variables; 2) the theta coefficient of Inclusive Value (IV) or the logsum; and 14 3) the estimate of value of time (VOT). The coefficients for system and socio-demographic 15 variables are assessed based on travel behavior theories and/or common knowledge. For instance, 16 the coefficients for time and monetary cost variables are expected to have negative signs. If the 17 theta coefficient estimate is rejected statistically, the NL model collapses to the MNL structure 18 (14). Estimating VOT provides a quantitative assessment of model performance. It is expected 19 that a reasonable VOT for commuting trips ranges from 30% to 50% of wage rate (15). 20


Searching for global optimum solution to estimate theta turns out to be a tedious process 1 as the estimate is sensitive to the starting value of theta. This analysis carried out approximately 2 80 model runs. Final results suggest that joint MNL specifications outperformed NL 3 specifications. In the MNL and NL models, the coefficient estimates for system and 4 socio-demographic variables have expected signs. The theta estimates in NL models are between 5 0 and 1 as expected and statistically significant. However, VOT estimates with the NL models 6 appear unreasonably large. It shows a VOT at $43.75 per hour for a commuter with an annual 7 income of $45,000. The VOT estimates with MNL models are acceptable at $9.05/hour for the 8 region and $13.85/hour for City of Austin. Accordingly, for the final models the analysis 9 specifies no theta coefficient, essentially estimating MNL joint models. 10

Table 6 shows a base model and an expanded model of travel mode choice for HBW trips 11 in the Austin region. The base model specifies variables representing travel costs (time and 12 monetary) and traveler socio-demographic characteristics, whereas the expanded model adds in 13 variables of MXD features. Mode initials in parentheses indicate the modes to which variables 14 are specified. 15

Results from the expanded model for HBW trips suggest that higher population densities 16 at trip origins are associated with higher probabilities of choosing non-driving modes for work 17 commute. Increasing population densities at destinations encourage car-pooling and riding buses. 18 Concentration of jobs at higher densities supports more bus uses. Street connectivity matters: 19 cul-de-sac intersections (% 1-way) discourage walking. The results re-confirm the findings 20 reported from other studies on the role of job and population density at home and work locations 21 on commuting mode choice after the effects of time and monetary costs of travel are controlled 22 (16, 17). 23

Table 7 shows models of mode choice for HBNW trips. Coefficients for all cost variables 24 have expected negative signs. Except for taxi cost, all coefficients are significant at 95% or 25 above level. For HBNW trips, population and job densities at designations matter to mode choice 26 decisions. The coefficient for % 1-way (i.e., cul-de-sac) intersections has a positive sign, 27 seemingly counter intuitive. Future studies should explore the issue by estimating separate 28 models for various non-work purposes, for example, shopping, leisure, school, and personal 29 business. 30

Data on sidewalk provision is available for areas within City of Austin. To utilize the data, 31 MNL models HBW and HBNW trips are re-estimated with observations falling within City of 32 Austin. The results (not reported here) confirm that, aside from population and job densities, 33 sidewalk provision at trip origins significantly influence mode choice for walking to work. No 34 statistically significant effects are observed for non-work travel. 35

36 37 38 39 40 41


TABLE 6 Logit Model of Travel Mode Choice for HBW Trips 1

Base Model Expanded Model

Variable Coef. Std. Err. t-Stat. Coef. Std. Err. t-Stat.

Time (DA) -0.095 0.009 -10.09 ** -0.093 0.010 -9.31 ** Cost (DA) -27.196 2.066 -13.17 ** -27.717 2.110 -13.14 ** Time (CP) -0.056 0.011 -4.95 ** -0.048 0.012 -3.84 ** Cost (CP) -45.149 3.925 -11.50 ** -45.782 4.000 -11.44 ** Time (Taxi) -0.069 0.066 -1.04 -0.061 0.064 -0.95 Cost (Taxi) -3.288 0.862 -3.81 ** -3.333 0.871 -3.82 ** Time (Bus) -0.135 0.041 -3.33 ** -0.146 0.042 -3.44 ** Cost (Bus) -50.143 4.187 -11.98 ** -50.614 4.254 -11.90 ** Vehpc (DA, CP) 3.184 0.388 8.22 ** 3.336 0.412 8.10 ** HHSize (CP) 0.200 0.077 2.59 ** 0.186 0.075 2.49 ** Female (CP) 0.474 0.151 3.15 ** 0.439 0.156 2.82 ** Age2035 (WK) -2.637 0.868 -3.04 ** -2.808 0.931 -3.02 ** Age3550 (WK) -0.408 0.656 -0.62 -0.427 0.708 -0.60 Age5065 (WK) -0.251 0.739 -0.34 -0.554 0.818 -0.68

PopDen at Origin (Non-DR) 0.027 0.014 1.92 * PopDen at Destination (CP, BU)

0.040 0.014 2.93 **

JobDen at Destination (BU) 0.009 0.003 3.10 **

PCT1Way at Origin (WK) -10.732 2.875 -3.73 **

Block Size at Origin (WK) -0.001 0.044 -0.01 Constant (DRIVE) 120.887 0.668 181.05 ** 120.326 0.689 174.67 ** Constant (DA) 2.647 0.255 10.36 ** 3.060 0.280 10.93 ** Constant (TRANSIT) 124.311 0.660 188.44 ** 123.716 0.688 179.84 ** Constant (Taxi) -1.736 1.071 -1.62 -1.301 1.057 -1.23 Constant (WK) 125.138 0.737 169.75 ** 126.296 0.814 155.09 ** Log-Likelihood at Zero -3483.20 -3483.20 Log-Likelihood at Start -2687.25 -4404.23 Log-Likelihood at End -845.51 -826.80 -2 (LL(Zero) - LL(End)) 5275.38 5312.81 -2 (LL(Start) - LL(End)) 3683.47 7154.86 Asymptotic rho squared 0.7573 0.7626 Adjusted rho squared 0.7518 0.7557 Notes: Significance Level: ** < 0.01; * <0.10

2


TABLE 7 Logit Model of Travel Mode Choice for HBNW Trips 1

Base Model Expanded Model

Variable Coef. Std. Err. t-Stat. Coef. Std. Err. t-Stat.

Time (DA, CP) -0.042 0.002 -19.09 ** -0.043 0.002 -18.44 ** Cost (DA) -9.262 0.437 -21.20 ** -9.336 0.441 -21.18 ** Cost (CP) -15.256 0.661 -23.09 ** -15.209 0.667 -22.80 ** Time (Taxi) -0.193 0.039 -4.98 ** -0.197 0.040 -4.96 ** Cost (Taxi) -0.077 0.048 -1.61 -0.076 0.047 -1.62 Time (Bus) -0.180 0.031 -5.86 ** -0.175 0.032 -5.43 ** Cost (Bus) -6.659 0.852 -7.82 ** -6.708 0.914 -7.34 ** Vehpc (DA, CP) 2.254 0.163 13.80 ** 2.299 0.166 13.82 ** HHSize (CP) 0.063 0.043 1.47 0.084 0.044 1.92 * Female (CP) 0.339 0.050 6.78 ** 0.341 0.050 6.77 ** Ageto20 (WK, BK) -0.482 0.178 -2.71 ** -0.432 0.174 -2.49 ** Age3550 (WK, BK) -0.734 0.146 -5.04 ** -0.748 0.148 -5.04 ** Age65up (WK, BK) -0.757 0.262 -2.89 ** -0.747 0.253 -2.95 **

PopDen at Origin (Non-DR) 0.007 0.006 1.21

PopDen at Destination (CP, BU) 0.020 0.006 3.44 **

JobDen at Destination (BU) 0.031 0.003 9.01 **

PCT1Way at Origin (WK) 0.932 0.522 1.79 *

Block Size at Origin (WK) 0.001 0.002 0.51 Constant (DRIVE) 1.377 0.157 8.78 ** 1.142 0.174 6.57 ** Constant (DA) 0.123 0.123 1.01 0.322 0.132 2.45 * Constant (TRANSIT) 1.272 0.234 5.44 ** 0.695 0.254 2.74 ** Constant (Taxi) -0.872 0.604 -1.44 -0.258 0.620 -0.42 Constant (WK) 2.389 0.102 23.49 ** 2.213 0.155 14.24 ** Log-Likelihood at Zero -12001.90 -12001.90 Log-Likelihood at Start -9608.34 -9964.74 Log-Likelihood at End -6025.05 -5980.74 -2 (LL(Zero) - LL(End)) 11953.70 12042.32 -2 (LL(Start) - LL(End)) 7166.58 7968.01 Asymptotic rho squared 0.4980 0.5017 Adjusted rho squared 0.4965 0.4998 Notes: Significance Level: ** < 0.01; * <0.10 2

3

Mode choice analyses for the Austin area presented above report findings that are 4 consistent with the literature on the role of urban form in influencing travel. After the effects of 5


system performance and traveler socio-demographic characteristics are controlled, MXD features 1 such as high population and job densities, network connectivity, and sidewalk provision exhibit 2 additional influence on mode choice decisions. These features matter at both trip origins and 3 destinations. CAMPO can adapt this “normative framework” (8) for mode choice modeling by 4 including the urban form variables in model specifications. 5

4. SUMMARY AND CONCLUSIONS 6

The year of 2013 marks the 20th anniversary of the creation of the Congress for the New 7 Urbanism. Design and transportation professions have had converging interest in the potential of 8 altering urban form to alter travel outcome. Yet, when it comes to the implementation stage, 9 practitioners in the two fields remain largely on their own silos. Barriers come from both 10 technical and non-technical aspects. This paper focuses on the technical side. Supplementing to 11 the “pre-processing” and “post-processing” approaches to integrating engineering and design 12 practice, the paper proposes a “present-processing” approach to incorporating design variables 13 directly in the first three steps of the Four-Step demand modeling procedures. The approach is 14 illustrated through the Austin, TX example. 15

The study first identified MXD sites in the Austin, TX area and then analyzed travel 16 characteristics associated with the MXDs vs. non-MXDs. Main results are summarized below: 17 Trip generation related: a) People living in MXDs make 0.2 more daily trips /person for 18

HBW and 0.3 fewer daily trips/person for NHBO purposes. Per CAMPO HBW 19 classification, MXD households make slightly more HBW Direct trips than non-MXD 20 households, but much less HBW-Strategic trips. For HBW-Complex, the average MXD 21 trip chain rate is much higher than that for non-MXD. b) MXDs show 40% higher 22 internal rate of capture than non-MXD (TAZs). c) Travelers from MXD households leave 23 homes ~10 minutes later than others in the morning; 24

Trip distribution related: MXD trips are 1.9 miles shorter for HBW trips, 0.65 mile / 0.9 25 minute longer for HBNW trips, and 1.2 miles/3 minutes shorter for NHBW trips. Daily 26 PMT is 6.2 miles less for MXD households than for non-MXD households; 27

Mode choice related: About 41.5% of MXD households own one or zero vehicles. For 28 non-MXD households, the figure is 30%. On the role of urban form attributes, population 29 and job densities at origins and destinations influence travel mode choice independent 30 from the effects of system performance and socio-demographic factors. Network 31 connectivity and sidewalk provision also matter. 32 The results suggest areas in which CAMPO models can be modified or refined to capture 33

the potential effects of the Activity Centers growth strategy on regional travel, for instance, 34 revising trip rates for trip production and attraction modeling; re-calibrating friction functions for 35 trip distribution analysis; improving estimation of internal trip making; fine-tuning time-of-day 36 distribution; and re-fining travel mode choice models by including urban form indicators. 37 Differences between MXD and non-MXD in travel as reported above could have significant 38 implications region wide. For example, CAMPO is considering a regional development scenario 39


with 37 Activity Centers featured with MXD attributes. The 37 Activity Centers may house 1 approximately 20,000 homes, which would translate into 120,000 daily PMT reduction. 2

Yet it should also be pointed out that fully incorporating the results in CAMPO planning 3 process still requires additional efforts. For example, supplemental surveys of travel in the 4 MXDs will be needed in order to apply this spatial grouping method. It is non-trivial task to 5 accomplish what are suggested so far. 6

To conclude, the study contributes to transportation planning and policy making in 7 Central Texas by providing local empirical evidence on urban form-travel connection. The 8 study’s method and process can be of interest to a broad audience in academia and practice. 9

10 ACKNOWLEDGEMENT 11 The research was partially supported by the Capital Area Metropolitan Planning Organization 12 (CAMPO); City of Austin; Texas DOT; University of Texas at Austin Center for Sustainable 13 Development. 14 15 REFERENCE 16

1 CAMPO. CAMPO 2035 Regional Growth Concept Initiative. CAMPO, Austin. http://www.campotexas.org/programs_growth_concept.php. Accessed July 20, 2013. 2 DART. Transit-Oriented Development and Planning. NCTCOG, Dallas. http://www.dart.org/about/tod.asp. Accessed on July 20, 2013. 3 H-GAC. Livable Centers. H-GAC, Houston. http://www.h-gac.com/community/livablecenters/default.aspx. Accessed on July 20, 2013.

4 Echenique, M. H., Hargreaves, A. J., Mitchell, G., and Namdeo, A. Growing cities sustainably: Does urban form really matter? Journal of the American Planning Association, 78(2), 2012, pp. 121-137. 5 Eash, R. Incorporating Urban Design Variables in Metropolitan Planning Organizations’ Travel Demand Models,1997. http://tmiponline.org/Clearinghouse/Items/Incorporating_Urban_Design_Variables_in_Metropolitan_Planning_Organizations_Travel_Demand Models.aspx. Accessed July 18, 2013. 6 TRB/NRC, NCHRP Report 365: Travel Estimation Techniques for Urban Planning. Washington, D.C., 1998. 7 Moudon, A. V and Stewart, O. Tools for Estimating VMT Reductions from Built Environment Changes. Washington State Transportation Center, Research Report WA-RD 806.3, Seattle, Washington, 2013. 8 Cervero, R. Built Environments and Mode Choice: Toward a Normative Framework. Transportation Research Board of the National Academies, Washington, D.C., 2002, D 7, pp. 265-84. 9 Ewing, R. and Cervero, R. Travel and the Built Environment: A meta-analysis. Journal of the American Planning Association, 76 (3), 2010, pp. 265-294. 10 SANDAG, Trip Generation for Smart Growth: Planning Tools for the San Diego Region.

http://www.campotexas.org/programs_growth_concept.php

http://www.dart.org/about/tod.asp

http://www.h-gac.com/community/livablecenters/default.aspx

http://tmiponline.org/Clearinghouse/Items/Incorporating_Urban_Design_Variables_in_Metropolitan_Planning_Organizations_Travel_Demand%20Models.aspx

http://tmiponline.org/Clearinghouse/Items/Incorporating_Urban_Design_Variables_in_Metropolitan_Planning_Organizations_Travel_Demand%20Models.aspx


2010. 11 Ewing, R., Greenwald, M., Zhang, M., Walters, J., Feldman, M., Cervero, R., Frank, L., Kassa, S., and Thomas, J. Traffic generated by mixed-use developments – A six region study using consistent built environmental measures. Journal of Urban Planning and Development, American Society of Civil Engineers, 3, 2011, pp. 248-261. 12 Ewing, R., Greenwald, M., Zhang, M., Bogaerts, Meghan, and Greene, W. Predicting transportation outcomes for LEED projects. Journal of Planning Education and Research, 33(3), 2013, pp. 265-279. 13 Schlossberg, M., Greene, J., Phillips, P.P, Johnson, B. and Parker, B. School trips: Effects of urban form and distance on travel mode, Journal of the American Planning Association, 72(3), 2006, pp. 337-346. 14 Ben-Akiva, M., and Lerman, S. Discrete choice analysis. Cambridge, MA, 1985. 15 Small, K., Winston, C., and Yan, J. Uncovering the distribution of motorists' preferences for travel time and reliability, Econometrica, vol. 73(4), 2005, pp. 1367-1382. 16 Zhang, M. The role of land use in travel mode choice: Evidence from Boston and Hong Kong. Journal of American Planning Association, 70(3), 2004, pp. 344-361. 17 Chen, C., Gong, H., and Paaswell, R. Role of the built environment on mode choice decisions: additional evidence on the impact of density. Transportation, 35(3), 2008, pp. 285-299.

trb14-bridging the gap between the new urbanist ideas and transportation planning practice

Documents