The Relationship Between Land Use and Trip Internalization Behaviors: Evidence and Implications

Size: px
Start display at page:

Download "The Relationship Between Land Use and Trip Internalization Behaviors: Evidence and Implications"

Transcription

1 The Relationship Between Land Use and Trip Internalization Behaviors: Evidence and Implications Michael J. Greenwald Dept. of Urban Planning School of Architecture and Urban Planning University of Wisconsin-Milwaukee Milwaukee, WI Tel: Fax: Word Count: 5,987 Words + 5 Tables * 250 words + 1 Figure *250 words 7,487 words Draft Date: November 14, 2005

2 Greenwald i Abstract: This paper addresses the relationship between land use and destination selection, and the question of destination selection on travel mode choice. Specifically, this work focuses on internalized trips, a sub-category of trip making where both trip origin and trip destination are contained in the same geographic unit of analysis. This investigation uses data from the 1994 Household Activity and Travel Diary Survey conducted by Portland Metro. Using multinomial logit and binary logistic models to measure travel mode choice and decision to internalize trips, the evidence here supports three conclusions: 1.) urban design elements do more to alter travel mode choice than alter trip destination; 2.) there is a threshold effect in the ability of mixed use to alter travel behavior; and 3.) greater emphasis to destinations within the area where the home is located needs to be given in trip distribution models.

3 Greenwald Page 1 of 18 INTRODUCTION Assumptions about travel behavior feed directly into the travel modeling process, explaining why better understanding of land use impacts on travel behavior, particularly on travel making decisions, has generated so much interest recently. The research up to now focused on questions of trip generation (e.g., mode choice, VMT, travel time, etc.) and the associated consequences (e.g., environmental quality, economic efficiency, public expenditure, etc.). That is the first step. The underlying land use theories on which these investigations are based emphasize both the ability of land use practices to eliminate or alter trips, and to alter the choices of trip ends. The current approach relating urban form to travel behavior is neotraditional design. These designs rest on the assumption smaller scale urban design, rectilinear street layout and variety of smaller scale economic activity closer to one s residence will inspire changes in travel patterns which simultaneously cause substitution in activity location (locations closer to home substituted for activity centers further away), and substitution of travel mode (travelers choose to walk to closer locations rather than drive, due to increased convenience) (1). This paper addresses the relationship between land use and destination selection, and the question of destination selection on travel mode choice. Internalized trips are the focus of this paper. Internalized trips are a sub-category of general trip making behavior, where both trip origin and trip destination are contained in the same area. This definition is both intuitively simple, and identifies issues requiring special attention. The next section describes current methods of travel modeling, including a brief discussion of difficulties with these approaches as they relate to understanding internalized trips. Then we consider what land use/transportation policy research related to internalized trips already exists. Next, summary information and discrete choice models related to trip internalization are presented, helping put internalized trip making into a behavioral context. Finally, policy and modeling implications are considered. CURRENT STATE OF KNOWLEDGE The need for better understanding of land use influences on trip internalization was discussed by Ewing, et.al. in their investigation of rates of internal trip capture for 20 neighborhoods in South Florida (2). The authors found that land use mix and regional accessibility accounted for a significant proportion of trips which started and ended within a specific community. They also called for greater understanding of internal capture in traffic impact research. The question is, how does the issue of internal capture fit conceptually in travel modeling? The most commonly used method of regional travel modeling is the Four Step Model (FSM). According to Bates, FSM consists of separate models which calculate and distribute travel behaviors by transportation analysis zone (TAZ). TAZs are not defined by consistent geographic scales, but zone borders often correspond to major streets in a particular region under analysis. The models applied to these TAZs are 1.) trip attraction and generation; 2.) distribution of trips between zones; 3.) travel mode split for distributed trips; and 4.) route assignment (2). The models are linked sequentially, with the assumption that the inputs to each model are independent. This independence assumption, particularly on the connection between trip generation/attraction and distribution, has been the subject of serious scrutiny. According to Bates the problem lies not with the logic underlying the relationships, rather the specification and execution of those relationships (3). Trip internalization is most relevant to the trip distribution model, since by definition internalized trips should not be distributed. Bates describes two methods of trip distribution in FSM travel modeling. The first is a variant of Newton s gravity model, where the number of

4 Greenwald Page 2 of 18 trips distributed between any two zones is a function of trip generating elements in the originating zone (e.g., housing), trip attracting elements in the destination zone (e.g. employment), and is inversely proportional to some measure of travel impedance between the zones (either travel time or distance). When both trip generation and attraction forces are well specified and trip origins and destinations fixed (e.g., distributing work commutes based on housing and employment figures by zone) distribution devolves into correctly specifying the impedance (3). These requirements also limit the usefulness of this method. Newer models integrating transportation and land use forecasting use discrete choice models to assign trips between zones. Examples include the TRANUS and MEPLAN packages described by Hunt et. al., comparing performance of network models for the Sacramento area (4). The advantage of discrete choice is only the origin must be predetermined; the choice of end point is allowed to vary based on characteristics of the traveler and the potential destinations. Borrowing Bates description, these discrete choice destination selection models are generalized as: k k k p j [ i : k] = f ( c... :, Z { } ) j c X i i{ j} j where p j [i:k] = proportion of travelers for a particular purpose k traveling from zone i to zone j c k ij = associated costs of traveling from zone i to a particular zone j {j} = the set of all possible destination zones X k = a set of characteristics affecting travel for purpose type k (e.g., Z {j} sociodemographics, household income, scale of activity, etc.) = a general set of zonal characteristics for all elements of {j} (e.g., parcel size, land use diversity, housing balance, etc.) This specification is more consistent with individual selection behavior for destination zones, though a special problem exists for internal trip making. Ben Akiva and Lerman describe one of the basic premises behind discrete choice modeling is that the process depends on differences observed in the explanatory factors governing mutually exclusive choices (5). In this case, discrete choice trip distribution models depend on differences between origin and destination zone on variables such as cost structures and zone attributes. For internalized trips, there is no difference on the zone elements which are part of the Z {j} set, including elements of local land use. The consequences are not mathematical abstractions. Hunt demonstrates by pointing out TRANUS does not model intrazonal trips. Hunt further explains the issue is circumvented for inter-zonal trip models by incorporating internalized trip making as part of an alternative specific catchall constant for interactions between households, institutions and firms. This catchall described by Hunt et. al. provides no insight on how land use affects someone s decision to make trips locally, and the authors specifically call for research on neighborhood attractiveness to reduce the impact these constants have on the trip modeling process (4). The current land use/travel behavior research addresses trip internalization only tangentially. Cervero s description and subsequent investigations of jobs-housing balances are some of the pioneering works in understanding trip internalization. Using data from the San Francisco Bay area, Cervero explained how combining residential and employment in the same area need not force mode splits away from the personal automobile, but could provide individuals with the option to reside closer to their place of employment, and thus alter their general trip making behavior to include travel and destination options which previously may not

5 Greenwald Page 3 of 18 have been viable (6). While this description connects directly to the question of trip distribution, subsequent investigations of the jobs housing balance paradigm have focused on the impact of employment and residential land uses on vehicle miles traveled, travel mode splits and trip generation by purpose (work vs. non-work) to validate the concept (6,7,8). These findings, emphasizing land use impacts on changes in trip generation, attraction, and mode choice, are explained as changes in trip distribution; they do not describe the impact of land use on trip distribution behaviors themselves. Handy and Clifton make important contributions in bridging this gap by coherently segmenting questions about trip distribution into related but distinct queries in their investigation of land use impacts on local shopping in Austin neighborhoods. Internalization is often a question of simultaneously determining mode and destination (e.g., what causes one to choose to walk to a local store vs. drive to a store in another zone). Handy and Clifton separate this into selection of location (i.e., local vs. non-local store), land use impacts on mode choice (i.e., travel mode throughout local neighborhood), and location dependent mode choice (i.e., choosing to walk to the local store vs. drive to a non-local store). They concluded local shopping cannot reduce overall vehicle dependence, finding that intangible factors such as loyalty to a store, and the high variability in ability of the local outlet to satisfy the immediate need (elements of X k modeling characteristics defined above), act against reducing vehicle travel (9). Subsequent to Handy and Clifton, Krizek demonstrated how neighborhood scale land use can cause changes in travel behavior, using longitudinal data from Puget Sound (10). Examining changes in travel for members of 430 households which changed locations in the Puget Sound, Krizek found significant reductions in vehicle miles traveled, overall travel regardless of mode, touring behavior and tour length for households with higher indices of neighborhood accessibility, calculated by combining residential density, neighborhood retail, and size of municipal block. The fact that Krizek examined changes in travel behavior as it related to change in environment strengthens the case for a causal relationship between land use and travel behavior. Similarly, this author has described how land use patterns more consistent with neotraditional design practices may induce travel mode substitution more conducive to walking instead of driving, relating urban design to mode specific travel time (11). That investigation was based on work by Crane explaining how travel time was an appropriate metric of personal trip cost for utility maximizing (i.e., efficient) trip makers (12). A recurring concern about land use impacts on travel behavior is the potential for land use being endogenously defined on a preferred travel mode choice. Boarnet and Greenwald have consistently tried to address this problem either through two stage least squares, or inclusion of supplementary probability models into the travel mode choice framework (11, 13, 14). While trip internalization may be a function of land use, the standard concern of travelers endogenously selecting environments which promote a preferred travel option is mitigated. One may choose to live in an environment which promotes a particular travel mode, based on preference, but decisions about where to complete subsequent activities in relation to current location are likely to be governed by time, sequencing and accessibility constraints of the immediate surroundings. Substituting the home neighborhood for other zones may be a less time consuming way of achieving a set of tasks, but that is a question of travel cost, not preference. Taken together, the studies by Cervero, Krizek and Greenwald suggest that land use may result in more localized trip distribution. Handy and Clifton suggest how the problem of investigating internalized trip making might be addressed, and the issues identified by Hunt explain how such an inquiry might be useful in a travel modeling context. Since trip distribution

6 Greenwald Page 4 of 18 is fundamentally a question of choice, the variables used in the discussion must reflect aspects of that choice. DATA DESCRIPTION AND SUMMARY STATISTICS This inquiry uses data from a variety of related sources. The main sources are the 1994 Household Activity and Travel Behavior Survey (1994 Travel Diaries), and the associated Regional Land Information System (RLIS) database, both produced by the Portland Metropolitan Services District (Portland Metro) (15, 16). The 1994 Travel Diaries contain information on 50,623 trips in the Portland Metro area, comprised of Multnomah, Washington and Clackamas counties in Oregon. The survey contains information on individual trip making behavior; traveler demographics, origin and destination location, mode choice, distance, parking costs and transit fares, travel times, and duration of activity related to trip purpose. After removing incomplete records, and selecting only those trips which had exact address matches on the location of trip origins and destinations, the final data set was reduced to 20,245 trips. The independent variables influencing trip making are grouped into four categories. The first are standard sociodemographic traits, traditionally included as part of travel choice behavior, mostly to check for any effect of a particular demographic suggesting a full set of choices is not available. Age (for persons 16 or older), possession of a license, gender, or ethnicity are indicators of individual ability to select a full range of travel mode options. The second set of traits address costs of trip making behavior. Household income is included to test wealth as factor in choosing destination; higher income would indicate access to a wider variety of travel modes. The measure of household income used is a projection based on the response of one of 13 income categories, ranging from zero to over sixty thousand dollars, in increments of five thousand dollars. Costs of driving are measured as the aggregate of vehicle operating cost (calculated using data from U.S. Dept. of Energy and the U.S. EPA,) as well as total parking charges the traveler either paid, or would have had to pay (17,18). Aggregation was deemed necessary because in several cases travelers could not estimate or reported no cost to parking. The monetary cost of transit is measured by the transit fare the traveler would have had to pay, based on their age. Mode specific, rush hour adjusted, travel times are also included as costs, using estimates from the EMME2 network model provided by Portland Metro. Walking time is calculated using network distance between origin and destination, applying a walking speed of three miles per hour, derived by Untermann (19). Third, travel mode opportunities and constraints are considered. Work and shopping trips are considered as special cases in determining mode choice; work trips are considered the furthest distance a person will travel for economic activity, and thus may not be representative of more general trip making. Handy s investigations of shopping behavior suggest that convenience goods (i.e., high frequency of consumption, immediate need and low impedance, such as groceries), were more inelastic with respect to travel time than comparison goods (i.e., selection based on traits such as style, price and quality) (9). Vehicle availability and transit viability are included as standard components of a choice framework; one cannot choose an alternative travel mode if it is not available. Transit availability follows the definition of Beimborn, et. al., that transit (either bus or light rail) must be within a quarter mile radius at both the origin and destination in order to be a viable choice (20). The number of persons joining the traveler are considered because additional travelers can imply additional constraints on travel logistics, limiting mode choice. Finally, land use measures are considered as promoters or inhibitors of particular mode choices. The focus here is on destination land uses, because internal trip making emphasizes the

7 Greenwald Page 5 of 18 attractiveness of localized land use to act as an alternative to external locations for a particular activity. Number of intersections and parcel size within a half mile of the center of the TAZ jointly indicate the degree of density and grid nature of the land use pattern in the zone; more intersections and smaller zones suggest a more rectilinear street pattern. The dummy variable for the Pedestrian Environmental Factor (PEF) reflects the degree of support for pedestrian travel in the TAZ as evaluated by 1000 Friends of Oregon. The original index measured the streetscape of each zone on four criteria (grid vs. cul-de-sac, topography, ease of street crossing and sidewalk continuity), using a three point scale (poor, fair or good), resulting in a constrained scale ranging from four to twelve points; the higher the score, the better the zone accommodates walking. This evaluation method raises questions of inter-rater reliability, so dummy coding is applied for zones which have low and high PEF scores. The implicit assumption is there will be more agreement on what constitutes an extreme environment than on a specific score. The housing balance and entropy indices indicate the degrees to which a TAZ is in balance in terms of housing stock and diverse in economic activity. In both cases, the values were calculated by aggregating point data on residential (N=358,795) and business location (N=41,890) to the TAZ in which the residence or business was located. Organizing the data by TAZ keeps subsequent discussion at only one level of geography. Size of destination TAZ is included as a control variable in models involving land use elements, since geographically larger zones are more likely to self contain trips. The mean size for TAZ in the RLIS is 2.99 square miles, with a standard deviation of square miles; the smallest zone was.02 square miles, the largest square miles. Smaller zones were associated with areas closer to the Portland downtown area. The housing balance variable is a ratio of number of taxlots zoned multifamily compared to single family, within a TAZ. The employment entropy indices are slightly more sophisticated, based on a formula derived from Cervero and Kockelman (8) and Ewing, et. al. (2): K Entropy = 1* prop( k) * ln( prop( k)) / ln( K) i= 1 Where K = Index of economic activity type Prop (k) = Proportion of a specific economic activity type, compared to all economic activity in the zone The business location data provided through RLIS included both the U.S. Economic Census Standard Industrial Classification (SIC) for the business and an ordinal scale for the number of employees engaged in the particular activity. Table 1 contains information on these scales. TABLE 1 ABOUT HERE Using the ordinal classifications in conjunction with the SIC classification entropy indices were generated for different employment scales. This allows considering not only the total economic diversity, but also the scale at which the diversity is taking place. In contrast to Cervero s work, all categories of economic activity, versus only retail and service employment, are considered. This is because all types of economic activity are generators and receivers of traffic. MODELS AND RESULTS TABLE 2 ABOUT HERE Table 2 summarizes difference of means tests, based on whether or not the trip was internalized, for all 50,623 trips. Two important points emerge. First, it appears that internally generated trips are significantly shorter in distance. This is not as obvious as it first appears; someone living on the edge of a TAZ might make a shorter trip traveling to a neighboring TAZ than staying within the TAZ. Second, the activities associated with an internalized trip are

8 Greenwald Page 6 of 18 shorter in duration than with a non-internalized trip. Using Crane s ideas about utility maximization, this means that for the same amount of time available during the day, more activities associated with internalized trips can be completed than for activities requiring external trips. Table 3 shows the results of a multinomial logit model relating travel mode choice to sociodemographic traits of the traveler and characteristics of the trip, including whether or not the trip was internally contained within the TAZ. The results are reported as relative risk ratios rather than standard regression coefficients. As described by Liao, relative risk ratios indicate the change in the ratio of probabilities for a particular outcome, relative to a fixed base choice (in this case, selecting either walking or non-school bus transit over automobile travel mode) attributable to changes in a specific independent variable (21). Reporting results in this manner simplifies the discussion in three ways; first it automatically accounts for the impact of the regression constant on the probability outcome for a particular category. Second, it allows the discussion to focus on the relative contribution of dummy variables by clearly stating the change in likelihood of a particular outcome depending on the state of a dummy variable. Third, it helps in describing the contribution of a particular variable to the travel decision; relative risk ratios greater than or less than one (i.e., neutral effect) exert influence in direct proportion to their distance from one. TABLE 3 ABOUT HERE The results in Table 3 appear consistent with previous research. The statistical significance and directions on the sociodemographic and monetary cost variables are as expected, but since the relative risk ratio is close to one, these elements do not stand out as major contributors to mode choice for an individual trip. The real contribution is made by the characteristics of the trips and availability of travel options. Vehicle availability is by far the strongest predictor of mode choice, with relative risk ratios close to zero for both walking and transit. Working trips are over 50% more likely to be made by transit compared to automobile, and shopping trips are only.51 times as likely to be made by walking, and.34 times as likely to be made by transit compared to automobile, all other things being equal. Higher personal vehicle travel times and shorter route transfer times increase the likelihood of (public) transit, while suppressing the likelihood of walking. More people in the traveling party reduces the likelihood of transit. Longer activity durations do little to inspire choosing walking over automobile, although each additional minute of activity increases the relative risk of selecting transit compared to automobile by a factor of The most significant finding is the impact of trip internalization on mode choice. The relative risk ratio indicates that an internalized trip is over 6.4 times more likely to be made by walking compared to driving, although it has no statistically significant impact on the relative risk of selecting transit. Combining these results with Table 2, a more complete picture of trip internalization emerges; internalized trips are shorter in distance, the activity quicker, and more likely to be walking trips. These are precisely the types of trips which should be most influenced by urban design, leading to the question what land use practices affect trip internalization? Table 4 models the decision to internalize a trip on the same sociodemographic, trip cost and trip opportunity characteristics, adding the land use information. The decision is modeled using a binary logit, reporting the odds ratios for explanatory factors. These ratios serve the same purpose as the relative risk ratios in the multinomial logit model (21). TABLE 4 ABOUT HERE

9 Greenwald Page 7 of 18 From the results in Table 4, again it appears trip characteristics and mode accessibility, rather than trip costs, play stronger roles in whether or not a trip is internalized in the originating TAZ. Shopping trips are less likely to be internalized, and higher transit and walking times between origin and destination reduce the likelihood even more. Land use elements, as a group, contribute to a more detailed understanding. The first two models in Table 4 explain the impact of urban form, housing mix, and diversity of economic activity within each zone. The results in the first two columns show greater density of street grid (as measured by the number of intersections and the average parcel size variables) and higher proportion of multifamily relative to single family tax lots all act to reduce the ratio for trip internalization. Statistical significance aside, the effect does not appear to be particularly strong, as the ratios on all of these land use measures are close to one. Contrary to expectation, a high value on the PEF index initially looks like it induces externalizing trips, although the effect disappears when the industry entropy index is disaggregated. The dummy variable for the trip destination being the individual s home TAZ substantially increased the likelihood that the trip was internalized (ratios ranging from 9.7 to 11.3). This is not surprising, since managing trips to minimize time away from home is simply making efficient use of one s travel time. Some interesting patterns emerge from the entropy indices. The ratio on the overall entropy index is over 15.5, representing an average value of the increase in likelihood of walking vs. driving since the entropy index is based on a continuous measurement. The model using entropy index by employer size reveals other important points. Economic diversity among very small (i.e., Size K) and very large (i.e., Size Q) employers may do more to expel trips than to retain them. For the larger employers, this may be a function of the change in ranges of the ordinal categories. For smaller employers, this might indicate a problem. Neo-traditional urban design theory suggests diversity among smaller scale employers work to contain trips locally; inconsistencies on this point weaken that premise. FIGURE 1 ABOUT HERE Figure 1 shows the distribution of businesses by employer size. The distribution is highly skewed towards the smaller employment categories; over half the businesses in this dataset have four employees or less. Because of their similarity in employment size, diversity among Size K and Size M employers were subsequently considered as one combined group. Similarly, the change in statistical significance for entropy among Size N and Size O employers lead to questions about range specification masking a significant relationship. To account for this, the Size N and Size O employment categories were also collapsed. Entropy indices were calculated for these new groupings. In combining the employment groups in this way, the disparity between employment ranges was minimized (though not eliminated), and thus it would still be possible to see if there was any consistent effect on trip internalization for diversity of economic activity conditional on employment size. Table 4, third column considers this new composite index model. Separately the Size K and Size M indices were significant but pulling in opposite directions. When the two employment scales are combined, diversity at this level does not appear to either promote or curtail trip internalizing. Combining the Size N and Size O categories does show a significant positive effect for increased diversity on trip internalization (in fact, there is a slight increase in the ratio for the Size N/Size O composite index compared to the index for Size N alone). Entropy indices for other employer size categories maintain the same relationships as seen in the second column. TABLE 5 ABOUT HERE

10 Greenwald Page 8 of 18 Table 5 examines the general and disaggregated models for any special impact on trip internalization for walking and automobile modes. On the sociodemographic variables, the pattern appears that age and being female increase the likelihood of making internal walking trips, while minorities have lower of driving for internal trips. Again, income and monetary costs tend to have little significant effect on trip internalization, though travel times associated with mode choice do affect trip internalization. Transit accessibility substantially reduces the likelihood of walking and car trips (ratios of.1342 to.1343 and.5666 to.5912, respectively). Controlling for the size of the zone, TAZ street grid density does not appear to affect internal generation for walking trips, though it does minimally act to reduce trip internalization for private automobiles (just barely; ratios ranging from.9442 to.9962 for intersections and parcel sizes). The most interesting findings are what isn t there; neither the general entropy index, nor the employer subtype indices, suggest a significant relationship between greater diversity of economic activity and internally generated walking trips. Higher proportions of multifamily compared to single family taxlots are statistically significant in reducing the likelihood of internalized walking trips, but again the magnitude of the effect is small, and there is no similar effect on internal vehicle trip making. High or low PEF scores in the destination zone similarly do not factor into decision to internalize, regardless of mode. For automobile trips, the entropy indices were important both in statistical significance and magnitude in explaining the decision to internalize a trip; the overall entropy index had an ratio over 30, and the entropy indices for smaller composite scale employment categories were generally significant; the Size N/Size O ratio was 5.47, and the Size P ratio was Similar to Table 5, the entropy score for the Size K/Size M composite index was not significant. Lastly, having the home TAZ as the destination increased the likelihood that the trip would be internalized by over a factor of 18 for a walking trip, and by a factor of 8.3 to 9.7 for an automobile trip. This is both consistent, and more pronounced, than the results in Table 4. ANALYSIS The results from Tables 3, 4 and 5 provide several validating and explanatory insights into questions of land use impacts on trip distribution. These findings support the arguments by Cervero that diversification of land use can lead people to alter location of activities (the impact of the entropy indices, both aggregate and disaggregate on trip internalization, both generally and for walking and car specific mode choices). This in turn might cause them to alter travel mode (the impact of internalization on travel mode choice). It is important to note, when disaggregated, the effect of smaller scale employers was initially inconsistent, then insignificant at the lowest employer level when recombined. This suggests a threshold effect. This point about sufficient scale also speaks to confirming the findings and conclusions of Handy and Clifton with respect to localized shopping not being a silver bullet strategy to reorient travel behavior. Shopping trips consistently were more likely to be externalized than other trip types. However, the findings are not all one sided. While the ratio for shopping trips on the decision to internalize was near.45 for trip internalization generally, and as low as.1108, or roughly 1/9, for internalized walking trips, the influence of being in one s home TAZ had ratios of to for trip internalization generally, and for walking trips (for car trips, the corresponding values from Table 5 are.65 and 9.79, respectively). Holding all other aspects of the travel decision constant, a shopping trip with the destination as the home TAZ is generally 11.02*.45 = 4.96 times more likely to be internalized than not; for walking trips, the factor is 18.15*.1108 = 2.01 times more likely to be internalized, though for car trips the factor is 9.79*.65 = 6.36 more likely to be internalized. This speaks to the ability of urban design to

11 Greenwald Page 9 of 18 facilitate individual travel decision making by allowing travelers to reorient the sequencing of trips; if retail opportunities in the home zone are sufficiently large and diverse, it may be more efficient for a person to begin a new sequence of trips after stopping at home, rather than add stops on the way. Last but not least, the impact of design and density variables needs to be addressed. PEF scores, parcel size and number of intersections (all indicators of neo-traditional design standards) and the housing balance index variable were not consistent with expectations. Lower PEF scores and lower concentration of multifamily compared to single family taxlots were more likely to internalize trips overall (Table 4). Looking more closely, the PEF effect vanished when trip internalization was considered by specific travel mode choice, and the housing balance index effect only appears to apply to decisions to internalize walking trips (Table 5). More street intersections and larger parcel sizes both acted to reduce trip internalization, pulling in opposite directions with respect to determining if tighter street grid density increases trip internalization. The ratios for parcel size, intersections and housing balance are all quite close to one, meaning that minor changes in these variables, on average, do little to affect the decision of whether or not to make a trip internally. Good design may induce people to walk more, but it is the variety of needs satisfied which determine whether or not a destination is viable. CONCLUSION Trip internalization is not well accounted in current transportation models. This paper suggests how different elements of urban form apply in the travel modeling process. Street design and housing concentration belong in the trip generation part of the process. Accessibility, variety and scale of economic activity are best considered as part of the trip distribution process. Further, the existence of the employer threshold effect for variety of economic activity affecting trip internalization suggests that there is such a thing as too small in terms of the ability of land use diversity affecting travel behaviors. These results also indicate that constraints and opportunities of the trip itself that most consistently guide the mode choice and trip distribution. The consistent significance of the destination being the home TAZ in promoting trip internalization suggests extra weight might be given to characteristics of locations closer to home in discrete choice trip distribution models.

12 Greenwald Page 10 of 18 References 1. Calthorpe, P. The Next American Metropolis: Ecology, Community and the American Dream, New York: Princeton Architectural Press, Ewing, R., Dumbaugh E. and Brown, M. Internalizing Travel by Mixing Land Uses: Study of Master-Planned Communities in South Florida. In Transportation Research Record: Journal of the Transportation Research Board, No. 1780, TRB, National Research Council, Washington, D.C., 2001, pp Bates, J. History of Travel Demand Modeling. In Hensher, D. and Button, K. Handbook of Transportation Modeling. Elsevier Science, Ltd., Oxford, UK, Hunt, J.D. et. al. Comparisons from Sacramento Model Test Bed. In Transportation Research Record: Journal of the Transportation Research Board, No. 1780, TRB, National Research Council, Washington, D.C., 2001, pp Ben-Akiva, M. and Lerman, S. Discrete Choice Modeling: Theory and Application to Travel Demand. MIT Press, Cambridge, Cervero, R. Jobs-Housing Balancing and Regional Mobility. Journal of the American Planning Association, Vol. 55 no. 2, 1989, pp Cervero, R. Jobs-Housing Balance Revisited: Trends and Impacts in the San Francisco Bay Area. Journal of the American Planning Association, Vol. 62 no. 4, 1996, pp Cervero, R., and Kockelman, K. Travel Demand and the 3Ds: Density, Diversity and Design. Transportation Research, Part D. Vol.2, no. 3, 1997, pp Handy, S. and Clifton, K. Local Shopping As a Strategy for Reducing Automobile Travel. Transportation, Vol. 28, 2001, pp Krizek, K. Residential Relocation and Changes in Urban Travel: Does Neighborhood-Scale Urban Form Matter? Journal of the American Planning Association, Vol. 69, No. 3, 2003, pp Greenwald, M. The Road Less Traveled: New Urbanist Inducements to Travel Mode Substitution for Non-Work Trips. Journal of Planning Education and Research, Vol. 23, No. 1, 2003, pp Crane, R. On form versus function: Will the New Urbanism reduce traffic, or increase it? Journal of Planning Education and Research, Vol. 15 No. 2, 1996, pp

13 Greenwald Page 11 of Boarnet, M. and Greenwald, M. Land Use, Urban Design and Non-Work Travel: Reproducing other Urban Areas Empirical Test Results in Portland, Oregon. In Transportation Research Record: Journal of the Transportation Research Board, No. 1722, TRB, National Research Council, Washington, D.C., 2000, pp Greenwald, M. and Boarnet, M. Built Environment as Determinant of Walking Behavior: Analyzing Non-work Pedestrian Travel in Portland, Oregon. In Transportation Research Record: Journal of the Transportation Research Board, No. 1780, TRB, National Research Council, Washington, D.C., 2001, pp Portland Metropolitan Services District Household Travel Diary and Activity Survey. Unpublished raw data, Portland Metropolitan Services District, Portland, OR. 16. Portland Metropolitan Services District Regional Land Information System (CD- ROM). Portland, OR: Portland Metropolitan Services District. 17. U.S. Heating Oil, Diesel Fuel, And Distillate Most Recent Weekly Data Weekly On- Highway Diesel Prices. U.S. Department of Energy Washington, D.C. Accessed January Download the Fuel Economy Database. U.S. Environmental Protection Agency. (2000). Washington, D.C. Accessed January Untermann. R. Accommodating the pedestrian: Adapting towns and neighborhoods for walking and bicycling. Von Nostrand Reinhold, New York, Beimborn, E., Greenwald, M. and Jin, X. Transit Accessibility and Connectivity Impacts on Transit Choice and Captivity. In Transportation Research Record: Journal of the Transportation Research Board, No TRB, National Research Council, Washington, D.C., 2003, pp Liao, T. Interpreting Probability Models: Logit, Probit, and Other Generalized Linear Models. Sage Publications, Thousand Oaks, CA 1994

14 Greenwald Page 12 of 18 List of Figures and Tables Figure 1: Table 1: Table 2: Table 3: Table 4: Table 5: Distribution of Businesses by Employer Size Economic Activity and Employment Categories for Businesses in the RLIS Database Differences of Means for Trip Length and Activity Duration for Internalized vs. Non-Internalized Trips Impact of Trip Characteristics on Selection of Non-Automobile Mode Choice Impact of Land Use Factors on Decision to Internalize Trips Impact of Land Use Factors on Decision to Internalize Walking or Automobile Trips

15 Greenwald Page 13 of 18 FIGURE 1- Distribution of Businesses by Employer Size Number of Businesses K-Size M-Size N-Size O-Size P-Size Q-Size R-Size S-Size T-Size Employer Size

16 Greenwald Page 14 of 18 TABLE 1: Economic Activity and Employment Categories for Businesses in the RLIS Database SIC 1-Digit Industry Types Agriculture Construction Electric Manufacturing FIRE Finance, Insurance, Real Estate Manufacturing Retail Services Transportation, Communications & Public Utilities Wholesale Employment Size Categories Category Number of Employees K 1-2 M 3-4 N 5-9 O P Q R S T 500+

17 Greenwald Page 15 of 18 TABLE 2: Differences of Means for Trip Length and Activity Duration for Internalized vs. Non-Internalized Trips (Unequal Variances Assumed) Difference in Trip Distance Externalized Trips Internalized Trips Difference Mean Standard Error N T-Test Degrees of Freedom Pr (Diff = 0) <.0001 Difference in Activity Duration (in Minutes) Externalized Trips Internalized Trips Difference Mean Standard Error N T-Test Degrees of Freedom Pr (Diff = 0)

18 Greenwald Page 16 of 18 TABLE 3: Impact of Trip Characteristics on Selection of Non-Automobile Mode Choice, Reported as Relative Risk Ratios Walking vs. Automobile Transit vs. Automobile Variable RRR Z RRR Z Age of Traveler Ethnicity (1 = White, 0 = Non White)) Gender (1 = Female) License (1 = Yes) Household Income Total Car Trip Cost Transit Fare Work Trip (1 = Yes) Shopping Trip (1 = Yes) Vehicle Available (1 = Yes) 8.65E E Transit Available (1 = Yes) Total Number of Travelers Vehicle M inutes* Transit M inutes* Transit Transfer M inutes* Walk Minutes Activity Duration (M inutes) Internal Trip? (1 = Yes) Number of Obs Log Likelihood Psuedo R^ Note: * All vehicle and transit times are adjusted for rush hour conditions Coefficients in bold are significant at the 5% level Coefficients in bold italic are significant at the 10% level

19 Greenwald Page 17 of 18 TABLE 4: Impact of Land Use Factors on Decision to Internalize Trips Entropy Index, General Entropy Index, Specific Entropy Index, Composite Variable Odd Ratio Z Odd Ratio Z Odd Ratio Z Age of Traveler Ethnicity (1 = White, 0 = Non White) Gender (1 = Female) License (1 = Yes) Household Income Total Car Trip Cost Transit Fare Work Trip (1 = Yes) Shopping Trip (1 = Yes) Vehicle Available (1 = Yes) Transit Available (1 = Yes) Total Number of Travelers Vehicle Minutes* Transit Minutes* Transit Transfer Minutes* Walk Minutes TAZ Size (in sq. miles) Number of Intersections Average Parce Size Low PEF Zone (1 = Yes) Hi PEF Zone (1 = Yes) Housing Balance Ratio Economic Entropy - General Economic Entropy - Size K Economic Entropy - Size M Economic Entropy - Size N Economic Entropy - Size O Economic Entropy - Size P Economic Entropy - Size Q Economic Entropy - Size R Economic Entropy - Size S Economic Entropy - Size T Economic Entropy - Size K & Size M Economic Entropy - Size N & Size O Activity Duration (Minutes) Destination Is Home TAZ? (1 = Yes) Number of Obs Log Likelihood Psuedo R^ Note: * All vehicle and transit times are adjusted for rush hour conditions Coefficients in bold are significant at the 5% level Coefficients in bold italic are significant at the 10% level Variable not used in this model

20 Greenwald Page 18 of 18 TABLE 5: Impact of Land Use Factors on Decision to Internalize Walking or Automobile Trips Walking Trips, General Entropy Index Walking Trips, Composite Entropy Indices Car Trips, General Entropy Index Car Trips, Composite Entropy Indices Odds Ratio Z Odds Ratio Z Odds Ratio Z Odds Ratio Z Age of Traveler Ethnicity (1 = White, 0 = Non White) Gender (1 = Female) License (1 = Yes) Household Income Total Car Trip Cost Transit Fare Work Trip (1 = Yes) Shopping Trip (1 = Yes) Vehicle Available (1 = Yes) Transit Available (1 = Yes) Total Number of Travelers Vehicle Minutes* Transit Minutes* Transit Transfer Minutes* Walk Minutes TAZ Size (in sq. miles) Number of Intersections Average Parce Size Low PEF Zone (1 = Yes) Hi PEF Zone (1 = Yes) Housing Balance Ratio Economic Entropy - General Economic Entropy - Size K & Size M Economic Entropy - Size N & Size O Economic Entropy - Size P Economic Entropy - Size Q Economic Entropy - Size R Economic Entropy - Size S Economic Entropy - Size T Activity Duration (Minutes) Destination Is Home TAZ? (1 = Yes) Number of Obs Log Likelihood Psuedo R^ Note: * All vehicle and transit times are adjusted for rush hour conditions Coefficients in bold are significant at the 5% level Coefficients in bold italic are significant at the 10% level Variable not used in this model Vehicle availability is assumed for all car trips