Abstract
As agencies develop more robust planning objectives for creating sustainable and livable communities, the research community has continued developing tools for estimating transportation impacts for site-level development review. This article provides a review of the state-of-the-art trip generation methods. We then offer a discussion about the consistency of approaches compared with theories of travel behavior and urban economics. The main objective is to identify the largest and potentially problematic gaps in methods available for practice in order to allow researchers, agencies, and practitioners to both be aware of these limitations and forge forward new innovations to solve these ongoing problems.
Introduction
The Institute of Transportation Engineers’ (ITE) Trip Generation Handbook (2014) and the corresponding Manual (2012) are the predominant resources for estimating transportation impacts generated by new development in the United States. Over the past fifteen years, a substantial amount of research has been published, evaluating the ability for current state-of-the-practice methods, namely, ITE’s Trip Generation Handbook, to more accurately predict multimodal traffic impacts in urban areas, such as (Currans 2013; Clifton, Currans, and Muhs 2012; Bochner et al. 2011; Daisa and Parker 2009). While the Handbook has remained a resource of “guidance” (ITE 2014)—recommending analysts collect their own data where the context of the site does not reflect typical ITE locations: suburban, single-use, vehicle-oriented development with unconstrained parking—constrained budgets for both agencies and practitioners have caused a call for more urban data and methods (Bochner et al. 2016). In response, the most recent edition of the Handbook (ITE 2014) has begun to incorporate the growing number of studies aimed at improving the collection of multimodal data and the estimation of multimodal impacts at new developments in urban areas, such as (Clifton, Currans, and Muhs 2015; Schneider, Shafizadeh, et al. 2013; Ewing, Greenwald, et al. 2011; Daisa et al. 2013). However, the existing state-of-the-art methods do not control for a number of important aspects, which are outlined in this article.
While others have evaluated the error in prediction of these methods (San Diego Association of Governments 2010; Shafizadeh et al. 2012; Millard-Ball 2015; Shoup 2003), discussed evolving data collection for urban, multimodal contexts (Clifton, Currans, and Muhs 2013; Schneider, Shafizadeh, et al. 2013; Dock et al. 2015), and explored how these methods are implemented in practice (Bochner et al. 2011; Clifton, Currans, and Muhs 2012), the focus of this article is aimed at reviewing these methods and others for consistency with theories of travel behavior and urban economics, of which the literature is both far-reaching and plentiful, but rarely framed around transportation impact analyses (TIAs).
To identify methods, multiple Google Scholar and library searches were used to identify a list of studies and methods aimed at improving trip generation estimation for TIAs or transportation impact studies (TISs). Phrases including “trip generation” and “transportation/traffic impacts analysis/studies” were used to identify a first cut list of methods. Methods that were not (a) developed using data collected in the United States, (b) within the past fifteen years, and (c) published within a peer review process (journal article or published institutional reports at the time this review was completed) were excluded.
In an effort to remain concise, the focus of this article remains directed toward the state-of-the-art methods for trip generation estimation, particularly for urban contexts. Description of state of the practice can be found in the ITE’s (2014) Trip Generation Handbook, which is now in its third edition. The following subsection provides context for the applications of trip generation data. Following, an overview of the general state-of-the-art methods for estimation of urban-centric transportation impacts is provided, including a table describing the methods evaluated in this study (Table 1). This article ends with a longer discussion section focusing on comparing these methods for consistencies with travel behavior and economic theories and conclusions for moving forward.
Urban Trip Generation Estimation Methods.
Note: HTS = household travel survey data; ITE = Institute of Transportation Engineers; MXD = mixed-use development; TOD = transit-oriented development.
aSite indicates travel behavior observed at individual sites. Elasticities are derived from external and prior studies. Other allows external data, assumptions, and information to be incorporated into model estimates.
bInfill+ may be applicable to larger areas that are not single use. Flexible method is flexible to development type.
cGeneral definitions include AM: peak hour, 7 AM–9 AM; PM: peak hour, 4 PM–6 PM; Daily: twenty-four hour counts.
dOn-street parking is an indicator for on-street parking present. Supply/price is the elasticities for constrained supply or parking pricing pulled from prior studies.
eArea wide indicates variables describing the surrounding area of the development, such as the block group area of study location; trip maker indicates characteristics of observed trip makers were incorporated into the method development; land use indicates characteristics of the nature of the land use itself (e.g., “luxury condominiums” or “discount grocery store”).
fInternal capture is only relevant for MXDs; methods developed for infill, but not listed as mixed use, do not provide a means for estimated internal capture.
Applications of Trip Generation Data and Methods
The question of how to properly estimate the multimodal transportation impacts of urban development is more pressing as urban areas struggle with the challenge of creating sustainable futures, supporting multimodal development, and reducing greenhouse gas emissions from the transportation sector given ever-constrained public resources. And as performance measures evolve, so must the data (California: The Governor’s Office of Planning and Research 2014). Because the current methods for estimating transportation impacts rely on these existing methods that have been shown to have varying applicability and accuracy in urban areas (Shafizadeh et al. 2012; Millard-Ball 2015; Weinberger et al. 2015), the implications trickle down into many different components of engineering and planning for new development including site design, scaling or scoping development, traffic impacts, system development charges, impact fees, emissions estimates, and sometimes regional travel demand modeling.
The use of trip generation data has a broad set of applications in transportation engineering and planning. The first, and most well-known application, is the use of trip generation data in TIA. Trip generation data refer to the counts of people entering and exiting a site. As with many studies of trip generation data, this article will refer to trip “ends” as trip counts interchangeably. Trip generation data are used to estimate the relevant demand of new development or redevelopment, estimate “new trips” derived from new development, and provide an estimate of total impact that allows for an assessment of future travel at the site for the year of build-out relative to area-wide rates of growth (which are also sometimes estimated using trip generation rates; McRae, Bloomberg, and Muldoon 2006). Many agencies in the United States rely on ITE’s approach as a defensible method for assessing the impacts of new development (Clifton, Currans, and Muhs 2015; Bochner et al. 2011).
Trip generation rates are also used as a proxy to estimate whether or not the developer needs to conduct a full TIA (Clifton, Currans, and Muhs 2012). If a new development is estimated to produce more than the threshold number of vehicle trips, as outlined by a given agency, the process triggers a TIA to review the relevant impacts of the new development. The thresholds that trigger TIA are often arbitrarily chosen, occasionally specified differently for districts throughout the city. For most agencies, only vehicle-oriented triggers are used; some have suggested the use of nonmotorized- or transit-based triggers may encourage more thorough multimodal development review, particularly in evaluating the safety of nonmotorized modes of travel surrounding the development (Ridgway and Tabibnia 1999).
Although the original intent of creating a compilation of trip generation was for use in traffic impact analyses, it is far from the only application of these data. As such, the implications of imprecision, inaccuracy, and inappropriate applications extend far beyond site-level traffic mitigations. Vehicle trip generation rates have also been used to estimate system development or impact fees—to accommodate improvements in network capacity or service—and transportation utility fees—to accommodate costs of operation and maintenance. While practices in applying impact and utility fees vary, these transportation charges are often applied on a “per trip basis,” which are estimated based on ITE’s data and methods, for example (Junge and Levinson 2012).
And while these data are occasionally used within four-step travel demand models to produce attraction trip generation rates—specifically for special generators or where there exist limited household travel survey data—they have been more recently incorporated in models estimating emissions of development in California. The California Emissions Estimator Model (CalEEMod) estimates greenhouse gas emissions for personal vehicle travel at new developments using a combination of ITE trip generation rates and locally derived trip length distributions (ENVIRON International Corporation, California Air Districts 2013), allowing users to evaluate greenhouse gases through vehicle miles traveled to satisfy Senate Bill 743 on Environmental Quality.
Urban Estimation: A Paradigm Shift
As researchers and agencies become more interested in improving traffic impact analyse practices for their regions, there has been a shift in the type of trip generation data collected. Alternative trip generation sources include person trip rates, mode shares (and mode-specific count estimates), contextual information (e.g., density, diversity, design, multimodal facilities, parking, and sociodemographics), and site information in addition to the “size” of the development (e.g., the cost of dwelling units, bike parking, year built, and transportation demand management [TDM] strategies). Moreover, as more agencies are the drivers of funding data collection and method revision research, some are also requiring, as much of the site-level information be both free and readily available for researchers, developers, and practitioners.
Multimodal trip counts cannot always be easily collected using observation or passive data collection—such as cordon counts. The state of the practice for collecting multimodal trip counts relies on both person counts entering and exiting the site to establish an overall person trip rate and a visitor intercept survey to calculate a multimodal mode share and an automobile occupancy rate. The combination of the person trip rate, mode share, and automobile occupancy rate provides an estimate of multimodal person trip counts and rates (e.g., person trips by car, bike, walk, transit, and vehicle trips). Additional information is sometimes collected but not always provided along with the cleaned data; these data may include on-street or off-street automobile/bike parking of visitors, trip length, trip purpose, demographics (age, gender, or income), group size, frequency of site visit, and frequency of mode used (Clifton, Currans, and Muhs 2013).
Although the procedures from person trip generation data collection are far from widely adopted by agencies across the United States, ITE’s Handbook has adopted some of the suggestions for guidance based on a few papers derived from the recent research with common features. ITE’s most recent addition of the Handbook included methods to collect multimodal person-based trip generation for infill development—single land uses developed on unused or vacant land within urban areas that are already mostly developed (ITE 2014) and are planning increased updates later this year (Bochner et al. 2016). The data collection methods adopted reflected the input of authors of several recent papers and data collections. These revised guidelines do not recommend a unified method to account for differences in urban behavior but rather introduce multimodal assessment using a wide range of approaches, each with its own limitations and constraints.
There are thirteen methods available (per my review standards stated in the Introduction section) and tested to predict urban vehicle trip generation impacts. To simplify the discussion, the following cited methods have been labeled in no particular order by letters. The characteristics of each method are described in Table 1. The methods discussed in this article, with their corresponding reference letter, include: Urban context adjustment (Clifton, Currans, and Muhs 2015); Smart growth trip generation adjustment (Schneider, Shafizadeh, and Handy 2015); Household travel survey urban context adjustment (Currans and Clifton 2015) based on Clifton, Currans, Cutter, et al. (2012); Report 758, National Cooperative Highway Research Program (NCHRP; Daisa et al. 2013); Report 684, NCHRP (Bochner et al. 2011), an updated version of the ITE’s Multiuse Method (ITE 2004) not discussed here; Environmental Protection Agency (EPA) MXD (Ewing, Greenwald, et al. 2011) based on Ewing, Dumbaugh, and Brown (2001); MXD + (Walters, Bochner, and Ewing 2013); Report 128, Transit Cooperative Research Program (Arrington and Cervero 2008); Urban Emissions Model (URBEMIS) (Nelson/Nygaard 2005; Jones and Stokes Associates 2007); CalEEMod (ENVIRON International Corporation, California Air Districts 2013); San Francisco Traffic Impact Guidelines (The Planning Department of the City and County of San Francisco 2002); New York City (NYC) Transportation Guidelines—Section 311 (2014); and Washington DC Department of Transportation (District Department of Transportation [DDOT] 2015).
To the author’s knowledge, there are also currently nine large research projects in progress across the United States with the intent to improve our understanding of how transportation impacts vary in transit-oriented development, 1 smart growth areas, 2 areas that allow no new parking to be included in new development, 3 developing more locally sensitive rates, 4 affordable housing with TDM strategies, 5 and one focusing on how to identify which method is best suited for different environments. 6
While most of these studies have collected or will collect urban trip generation, there remain only three methods that directly estimates person trips. While a few methods utilize household travel survey data, organized in a format that allows for parity with more traditional methods, for most methods, there exist too few person counts for any one land use to estimate multimodal impacts directly from establishment-level studies and control for the various aspects of new development believed to influence transportation impacts (such as the built environment, sociodemographic, etc.). As it stands, most existing methods that account for any of these issues are adjustments—modifying ITE’s suburban, vehicle-oriented data and methods, most often to correct for relative measures of the built environment.
The most common way to estimate person trip rates is to indirectly adjust ITE’s Trip Generation Handbook vehicle trip generation rates based on assumed mode share and vehicle occupancy rates for ITE’s study sites. This adjustment method considers ITE’s Handbook study sites and assumes an automobile mode share and vehicle occupancy rates for these ITE-type locations. The range of assumed automobile mode share rates varies by land uses but is generally considered between 95 percent and 100 percent automobile uses based on the vehicle-oriented, suburban, single-use establishment descriptions within the Handbook. Transit use was not collected very often in ITE’s baseline sites, but occasionally, it was recorded and can be used to adjust these assumptions. Vehicle occupancy rates were reported for only a select few land uses but can be used to refine the general assumption of vehicle occupancy rates between 1.0 and 1.2 persons per vehicle. The ITE vehicle trip estimation is then adjusted using these assumed rates to derive a person trip estimate (see
Equation (1) adjusting ITE vehicle trip rates into ITE person trip rates
This estimate, generally derived using loosely assumed mode share and occupancy rates, represents the person trip rate of ITE-type locations—suburban, vehicle-oriented, single-use locations with no shared parking, little to no transit access, and no bicycle or pedestrian activity. These person trip rates are assumed constant across urban contexts, or rather the assumption is that an estimate of person trips derived from ITE’s suburban sites is relevant for more urban locations as well. The implications of this assumption are discussed in the Estimating People subsection. However, for study sites in urban contexts, this ITE-based estimate of person trip rates is used to estimate overall activity and then context-based estimates for the urban area are used to determine the mode share for the study area, allocating the person trip count estimates into relative mode counts (see equation [2] for the mathematical form of this description).
Equation (2) Reallocating ITE Person Trip Rates into Context-based Modal Trip Estimates
While the industry shifts toward collecting multimodal trip generation data, practitioners continue to struggle estimating the impacts of urban development. In lieu of waiting for person-based trip generation estimation methods to become available, the industry remains reliant on methods that adjust vehicle trip generation—most often estimated using ITE’s Trip Generation Handbook. These adjustments are developed from urban trip generation data, or they are developed using secondary household travel survey data. Eleven of the thirteen major adjustment methods discussed here rely on some “base estimate” adjustment—always ITE’s Trip Generation Handbook vehicle trip generation rates, but sometimes allows for some locally collected data. Table 1 provides a summary of all thirteen methods for urban trip generation estimation. The following section provides a discussion of the similarities and differences between these methods, aligning the research with themes and theories of travel behavior and urban economics.
Discussion: State-of-the-art Methods
The methods summarized in Table 1 are explored below with respect to several elements of urban trip generation estimation, travel behavior theory, and urban transportation economics. The methods—identified using the IDs provided in Table 1—and their relative contributions and performances are discussed this section. Among these methods, there is no clear indication that any of these methods do a better job of estimating urban influences in trip generation—for example, Shafizadeh et al. (2012) and Weinberger et al. (2015)—likely due to limited data, so instead, we focus on the relationship between these approaches and travel behavior, land use development, and economic theories.
Estimating People
The three of the thirteen urban methods that estimate person trip rates directly (K, L, M) are agencies that have compiled their own data repository. The other nine methods adjust ITE’s vehicle trip rates either (a) directly adjust vehicle trip rates for urban context through reductions in vehicle trips (A, B) or (b) adjusting from a baseline estimated person trip rate derived from ITE’s suburban rates. In these nine methods, there are no adjustments for changes in person trip rates across urban areas—adopting the assumption that person trip rates are constant across all areas.
To explore this assumption, we first examine on the travel behavior literature. In reference to trip generation, most of the literature focuses on estimating the relationship between the built environment, demographics, and mode-specific travel (e.g., vehicle trips and walk trips; Ewing and Cevero 2010), mainly focusing on travel described from a home-based orientation, which limits the ability to transfer findings to a wide range of development types. While much can be explained from independent analyses of mode-specific travel, few studies have focused on understanding the overall demand for travel (e.g., total trips or activity)—or rather the joint effects of land use upon mode choice and trip frequency—potentially leading to over- and underestimation of overall activity (Guo, Bhat, and Copperman 2007).
In lieu of substantial support from the travel behavior literature, we turn to urban economics. The theory of bid rent has indicated that as regional accessibility decreases, so does the value of land, for example (Alonso 1964; Mills 1969; Giuliano and Small 1991). It follows that businesses pay a premium to locate in areas with higher levels of accessibility—defined as access to destinations or economic potential. Studies indicated that even residents pay more to locate in areas with greater accessibility in terms of retail and total employment destinations (Kockelman 1998; Srour, Kockelman, and Dunn 2002), lower accessibility to workplace competition (Srour, Kockelman, and Dunn 2002), greater access to transportation facilities, such as highways (Iacono and Levinson 2012) or metro lines (Anas 1995), although some suggest there is no (significant) added value in locating near multimodal facilities (including bicycle and pedestrian; Iacono and Levinson 2011). Many studies have also found significant relationships between accessibility and employment (Srour, Kockelman, and Dunn 2002), retail (Srour, Kockelman, and Dunn 2002), business districts (Cervero and Duncan 2002), population (Srour, Kockelman, and Dunn 2002), transit (Anas 1995; Cervero and Duncan 2002), and toward facilities (Targa, Clifton, and Mahmassani 2005). More directly, the success (sales) of the businesses—which must then off-set any premiums paid by increased accessibility of the location choice—is determined by the regional accessibility (population accessible to the site) and the economic potential of the location (income of the population that may access the site discussed later in this section; Des Rosiers, Theriault, and Menetrier 2005).
Further research is necessary to determine whether the outcomes suggested by this theory hold for TISs—in other words, do person trip rates vary by accessibility to destinations and consumers, land value, or the economic potential of sites? Instead, this theory calls into question the assumption that person trip rates do not vary across contexts, which is prominent in nearly every state-of-the-art method. If regional accessibility is the metric that reflects how reachable the location is relative to other areas in the region—for which, no existing study to the best of the authors knowledge has tested—it is included in only three methods to capture variations in mode share (C) or vehicle trip rates (B, H). But none of these adjustments account (or test) for variation in person trip rates across any definition of accessibility—even New York’s approach provides a single person trip rate for all five boroughs. The approach used in San Francisco (K) indirectly accounts for regional accessibility (as well as demographics and densities) in the estimation of mode shares within predefined districts. The New York approach (L) for estimating mode share accounts for regional accessibility indirectly through qualitative assessment and selection of previously collected data for location-by-location application. Furthermore, understanding the overall flows of activity to and from any one development requires a better understanding of who is traveling in the first place, which leads us to examine the ways in which demographics are incorporated into site-level transportation impact estimation methods.
Who the People Are?
Few methods account for sociodemographic or economic-demographic indicators. There are two ways to incorporate demographics in trip generation analysis: studying the trip makers or studying the market in the study area. The former approach is not utilized in any of the methods (except for an early version of the method C in which mode share varied significantly with income; Clifton, Currans, Cutter, et al. 2012), mainly citing issues in the practical application—it is difficult to apply when we do not know who may be coming to the sites. The alternative method to account for demographic is to use some average or median values representing the site’s surrounding areas. While developing the smart growth trip generation adjustment (B) and EPA MXD (F), contextual information about the types of average households located within the mixed-use development (MXD) study area—including children, household size, and vehicle ownership—was included. For adjustment B, there was not enough evidence to suggest the variables were significant. For adjustment F, there was evidence to suggest the variables were significant; analysts applying the model rely on area-wide descriptions of demographics to apply adjustments. Methods D and K use district-based analysis to estimate mode shares—the benefit being that relative difference in travel behavior due to trip-maker demographics is incorporated indirectly through aggregation of trips that occur in those areas. For example, trips from districts with high land values reflect trip-maker decisions that would normally travel to high-land-value districts.
To fully understand the travel demand, we must also understand who is traveling and why. Within the theory of derived demand, it is recognized that activity patterns of individuals and households are constrained in both time and monetary budgets, requiring certain types of activities to satisfy both individual and household needs but constraining travel to activities—as well as the activity itself—within time and cost budgets (Goulias, Pendyala, and Kitamura 1994; Bhat and Koppelman 1993). Activities are correlated among members of a household (Goulias, Pendyala, and Kitamura 1994), particularly among households with children whose dependence is so great, shifts of activities patterns for every member of the household can be seen (Pas 1985). In practice, trip generation studies rarely consider the sociodemographics of the establishment’s market. Without information about who is traveling to these establishments, we have a limited ability to control for variations in demographics using existing data.
An alternative to accounting for demographics using explicit variables in analysis is to incorporate the measure in how the land use is defined (e.g., luxury condominiums, discount superstores or grocery stores, toy/children’s store, baby store; ITE 2012), although this segmentation may be a statistically inefficient use of the data—the documentation of the decisions to segment these categories is not publicly documented. San Francisco (K) segments residential land use types by the number of bedrooms, attempting to capture variations in household size of the residents. ITE considers luxury condominiums as a separate category from condominiums and includes a “discount grocery store” category (2012), although the definition for luxury and discount in monetary terms is not provided. We are left to assume that any data in the land use category not specified by some measure of price are actually market rate, but this information is neither solicited nor regularly collected for uses prior to creating these categories, making demographic-based adjustments to ITE impractical.
There is a substantial amount of interest in investigating the relationship between trip-maker behavior and sociodemographics or economic-demographics for development-level evaluation of transportation impacts—particularly related to multifamily housing, income, and vehicle ownership (see notes 2, 4, and 6).
Land Use Categorization and Aggregation
Next, the detailed categorization of ITE’s land use categories needs consideration. ITE’s Handbook divides their data into over 150 different definitions of land use, with little published discussion about the process in which new categories are added or aggregated and whether that level of cataloging is necessary for practice. It is not clear whether the process of aggregation is based on the definition of land use alone or on some form of statistical testing of behavior. Regardless, methods that directly adjust ITE’s vehicle trip rates rely heavily on ITE’s detailed categorization of land use types (A, B, C, I, K, and L), while methods that use household travel survey data (C, D, and F) or are constrained by too few data (E) use aggregate land use categorization into broader categories that reflect zoning definitions (e.g., retail, service, and residential). Since the user of these data is often the developer—and the stage of development review often only provides a rough estimate (Mehra and Keller 1985) not always corresponding with the final product (McRae, Bloomberg, and Muldoon 2006)—the (dis)aggregation of categories plays a big role in how efficiently these data are used.
Recalling our discussion of “derived demand” in the previous subsection, the incentive for studying the trip maker’s motivations for activities results in a more complete understanding of why we observe variations in travel behavior and how we might better estimate or predict it. Based on this view of travel, more attention should be paid to the reasons for the demand for activities themselves and the corresponding derived nature of travel (Pas 1985). The result could be a land use taxonomy for trip generation estimation at the establishment level that is based in the theory of derived demand, supported by activity-based research and theory, and considerate of applied practice and the multidimensional information available to predicting land use types at new developments (Guttenberg 2002; American Planning Association 2001). By considering the patterns of travel behavior and motivations related to specific types of activities, we can begin to identify potential similarities and differences between land uses to identify patterns of behavior that allow us to more accurately and precisely predict transportation impacts of development.
Built Environment and Multimodal Travel
Some previous research suggests that travel behavior varies across different measures of the built environment (Ewing and Cevero 2001, 2010). The built environment may include any of the six Ds: density, (land use) diversity, design, destinations, distance to transit, and demand management (Walters, Bochner, and Ewing 2013; Cevero and Kockelman 1997). (The seventh D, development scale, is accounted for in both the categorizing mixed-use and infill development and development size. The eighth D, demographics, is discussed in the previous subsection.) Although there is a lack of consensus for whether behavior, such as mode choice and trip rates, do vary by the built environment (Ewing and Cevero 2010), ten methods evaluated here include the built environment in their estimation process, placing a great importance of urban context in estimating variations in trip generation—specifically, as it pertains to changes in mode shares or mode-specific trips.
Household travel surveys are commonly used as a method to estimate multimodal mode share and vehicle occupancy rates (C, D, F, G, I, J, K, and L). Alternatively, methods can utilize intercept surveys performed during the site-level data collection (E, G, and L), but these data are both expensive to collect and difficult to synthesis for future use. In New York (L), mode shares are not attached to person trip generation rates but provided location-by-location based on the land use and urban context of the development. In Washington, DC (M), due to small variation in densities, multimodal trips are estimated as a function of development size, without controlling for the built environment.
Only two methods account for TDM strategies (beyond transit access): metered parking within 0.1 miles of the development (B), proportion of surface parking (B), and various TDM programs (I). Agencies, like San Francisco (K) and NYC (L), often negotiate credits for adopting strategies allocated through a separate process. NCHRP 684 (E) includes a small sample size of study sites (six) but uses proportions of land uses to interpolate potential mixing of land uses. The combination of NCHRP 684 with EPA MXD (F) into MXD+ (G) allows the user to control for variations in mode share based on a wider sample of sites and built environments provided by household travel surveys used in EPA MXD while maintaining the robust analysis of how trips within developments are captured by other land uses. These methods are discussed in the following subsection. Seven methods account for the built environment using either continuous measures describing the built environment (A, C, F, G, H, and I) or a distilled measure using factor analysis (B). Two methods (D, K) account for the built environment by using districts or zones to estimate variations in mode shares.
Mixed-use or Multiuse Methods
Adopted in the second edition of the Handbook (ITE 2004), ITE incorporated a method to estimate impact adjustments for mixed-used development. An MXD is defined as “an integrated development (usually master planned) consisting of at least two complementary and interactive land uses designed to foster synergy among activities generated by the land uses” (ITE 2014, 138). Literature discussing MXDs, trip generation, and internal capture tends to reflect the data analysis of large planned communities. By ITE’s definition, however, the scale of these developments tends to include mostly single developments (planned simultaneously but built out in stages), ranging from 7 to 300 acres in scale (Bochner et al. 2011), but other comparable studies have even focused on developments anywhere between 5 to over 2,000 acres (San Diego Association of Governments 2010; Ewing, Greenwald, et al. 2011). In MXD analysis, “internal capture” is defined as “a person trip made between two distinct on-site land-uses at a mixed-use site without using an off-site road system” (ITE 2014, 129). This type of trip can be made by any type of transportation mode.
By removing trips that are internally captured from the overall estimate of transportation demand, the estimate reflects trips that are added to the existing network after the development occurs. For new development (or rezoned development), this means that only the change in transportation demand, before and after development, is used to assess the impacts—through impact fees or charges or when evaluating necessary mitigations to the adjacent transportation network (e.g., roadway widening, turning bays, and intersection upgrades). For MXDs, ignoring internal capture would result in overdeveloping for the automobile—which inhibits precisely the goal that MXDs are trying to achieve walkable, connected, and planned neighborhoods. Similarly, for infill development, analysts also assume that a proportion of travel to new development is “pass by” traffic or does not necessarily add traffic to the network. The methods of collecting and applying pass by data were not subject for review in this study.
Three methods (E, F, and G) were developed for MXDs, ranging from single-building developments (F, G) to 800-acre planned development, and these methods account for whether person trips (by automobile, foot, bicycle, or transit) generated to the study area are external or internal. Because NCHRP 684 (E) was developed using site-level data from only six locations, the authors combined their estimates with the results provided from EPA MXD (F) to derive reconciled estimates in MXD+ (G). The other methods mentioned in Table 1 are primarily for infill, although each of these methods on their own can be used to estimate establishment rates (although not internal capture) located within MXD to refine rate estimates.
Hooper (1990) and Bochner et al. (2011) have set the standard for mixed-use and multiuse development data collection at large (3–800 acre) mixed-used development. They approach the complexity of capturing the internal trips between land uses within the development with a system of cordon counts (automobiles), manual person counts, intercept surveys at establishments and transit access points, and intercept surveys along internal sidewalks. These data collections are often the most expensive to perform—costing upward of a US$50,000 per sites (Bochner et al. 2011)—and therefore are much harder to come by than single-use or single-building sites (as much as US$10,000 per site; see note 2).
The term “mixed-use development,” however, includes a broader definition in practice than considered in ITE-related studies. Mixed-use includes any area where the mix of land uses results in trip chaining between the land uses. While there is a growing literature on the overall transportation impacts of mixed-use planned developments toward an analysis that examines the influences of mixed uses on infill development within existing communities—we understand less about how these infill developments function within an existing mixed-use community—like historic downtowns, urban commercial corridors, or the central business district. In a 2013 study, the authors surveyed visitors to shopping districts in suburban and urban areas and found that 65 percent of trips between land uses in all shopping districts were walking trips, but in urban centers, approximately 96 percent of trips between land uses were walking trips (Schneider 2013). Furthermore, we have only begin to understand the ways in which trips are captured within mixed-use buildings (Walters, Bochner, and Ewing 2013), but the available methods have not yet been adequately tested within a dense urban range of contexts—particularly because of the extensive costs of data collection. As such, new approaches to capturing and understanding the interactions between infill development and the surrounding existing area are needed.
Conclusions
Trip generation estimation methods for TIAs were developed with an eye for simplicity—a quick rule of thumb reference—but estimating transportation demand accurately can be more complex and nuanced than methods in practice would lead one to believe. Research developed in response to this review may increase the flexibility of the data available for practice, extending the life of information by being more efficient with how it is understood and being used. This will allow agencies, developers, and practitioners to recognize which elements of a new development and its environment might influence the expected transportation impacts, permitting more appropriate mitigations to be considered to achieve planned results. This review identifies eligible methods for TIAs, providing a critique in the successes of existing methods as well as the gaps supported by the literature.
The findings of this review indicate strong support for understanding the influences of the built environment on vehicular trips, but not necessarily on multimodal trips. The assumptions used for most of the existing methods—adjustments to ITE’s method—have not been tested for conflicts with theories of urban economics, such as (Alonso 1964; Des Rosiers, Theriault, and Menetrier 2005), to the best of the author’s knowledge. There appears to be little-to-no sensitivity toward the relationship between the sociodemographic of the trip maker and behavior for TIA, which may cause over- or underestimation of vehicle travel in areas on either end of the income distribution. Moreover, the current detailed segmentation of land use categories may not provide any additional benefit for evaluating new development, particularly as the TIA process is held very early in the development stage and detailed accounts of the businesses occupying the development may not be known. Overall, the gaps identified from this review of state-of-the-art methods suggests consistencies with travel behavior theory related to identifying likely trip makers—sociodemographic and economic constraints that define a time and monetary budget for travel to land use development. While the extent of the corresponding biases for these issues is yet unknown, multiple existing and ongoing projects aim to target several of these problematic areas.
Although we have identified several gaps and issues in this process, primed for future research, the responsiveness to these themes in state-of-the-art methods in urban TIA has improved substantially within the past two decades. However, there has been little effort in the literature to identify the widespread use, substitution or replacement of existing methods in practice, suggesting only a limited and anecdotal view of the state of the practice of TIS approaches.
While this article has focused upon existing, peer-reviewed and published methods for estimating urban trip generation for TIA, many new forms of data have become more readily available to agencies and analysts. While ITE has only recently accepted and incorporated adjustment methods developed from more traditional and pervasive household travel surveys into use, there remains an ever-growing list of new technologies that may be applicable to such circumstances; examples of which include, but are not limited to, smartphone tracking data and “push” surveys, transaction count data, Google data like “popular times” activity distributions, and passive data collection technology including Bluetooth tracking and various forms of sensors. Likewise, the need for responsiveness in urban trip generation methods to planning policy goals and indicators requires the merging of multiple forms of data to describe urban form, transportation facility pricing, and, not the least of which, parking. Few methods consider and explore the endogeneity of parking availability and pricing in vehicle and multimodal trip generation estimation. The practicality and effectiveness of these types of data in site-level impact analyses will be directly related to the capacity, support, and willingness of agencies to test and adopt new technologies that may improve accuracy and precision as well as the theoretical understanding of transportation impacts at urban land use development.
Additionally, few have discussed the uncertainty and limited information available to developers and analysts during the time many TIA are completed. Many building permits are tied up in the process of site-level evaluation, leaving impact fee estimates and TIA studies tied up in rough predictions of what the development may become. McRae et al. (2006) reviewed twelve TIA studies after development and found that a third were not developed as planned in the TIA. This, combined with the inherent uncertainty existing in all transportation demand modeling predictions, leads to the question: is it reasonable to evaluate new development so early in the development process and per estimates from models not necessarily sensitive to planned outcomes seen as influencing behavior? Or rather, should the evaluation of impacts along a singular metric of trip generation—that so often leads to incremental overdevelopment of automobile facilities (Manville 2017), regularly in direct conflict with regional plans—be the primary means of determining whether mitigations to the network be made? This is certainly a necessary area of future research and thought.
The methods available today, albeit adjustments to existing data of limited contexts, provide a means for planners and engineers, agencies and practitioners, to respond more flexibly toward planning outcomes, specifically the built environment. That said, there exists only limited evaluation of the performance and improvement of these methods for widespread applications in practice, which may orient the user toward methods that perform better for their specific contexts or land uses. Furthermore, more could be done to assess how these approaches are being adopted, substituted, tailored for local context, or even prohibited by agencies and practitioners around the United States. As such, one of the main objectives of this article is to provide a landscape from which researchers, agencies, and practitioners can pull from to continue to move the state-of-the-practice forward.
Footnotes
Declaration of Conflicting Interests
The author declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The author(s) received financial support from the Dwight David Eisenhower Graduate Transportation Fellowship and the National Institute of Transportation and Communities Fellowship.