Advanced Practices in Travel Forecasting (2010)

Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.

32 During the interview process, each agency was asked about the motivation for moving to advanced models and the benefits of doing so. The responses were specific to the context and stage of development of each agency: those agencies at the beginning of model development projects could only speak to the expected benefits that motivated them to implement advanced models, whereas those agencies that have used the models extensively in application could speak directly to those benefits that they have actually achieved. More weight is given to the latter in the discussion of those benefits that follows. The notion of benefits is best considered in the context of what an ideal model must do. A good model must do three things: 1. Replicate base year conditions, 2. Be sensitive to the policies being tested, and 3. Respond logically to changes in input. Model calibration and validation efforts in practice tend to focus on the first objective, which is a legitimate objective as a measure of the model’s ability to replicate observed behav- ior. Thus, it is fair to argue that an advanced model could val- idate at least as well as a traditional model against base year conditions, and one area of benefits identified focuses on an advanced model’s ability to do so. The most significant advantage of advanced models noted in validation is simply the ability to validate against a much broader range of criteria. That is, if an activity-based model has 12 core models (population synthesis, usual workplace location, auto ownership, tour generation, joint travel, tour destination, tour time-of-day, tour mode, stop location, trip time-of-day, trip mode choice, and assignment), each of those models can be individually calibrated. Furthermore, microsimulation allows for model results to be tabulated in any number of ways, just as with a household interview survey, allowing the analyst to compare the model and the survey in any number of ways. Ultimately, advanced models allow the analyst to dig much deeper into the model system to understand what is transpiring, and what aspects of the real world the model does or does not capture. In contrast, aggre- gate, trip-based models can be very good at hiding mistakes. It is easy for such model systems to contain compensating errors, and difficult to detect—just ask any modeler who has been involved in a New Starts forecast, and realized that they need to significantly adjust the mode choice model to over- come problems in trip distribution. Activity-based models allow for a much more detailed validation of the system, but also allow for much more targeted calibration. However, a model that achieves the validation objective while ignoring the other two objectives will be limited in application. There is far more to building a quality model than achieving the lowest possible root-mean-squared error compared with traffic counts, especially if achieving that goodness-of-fit involves sacrificing logical responsiveness in the model system. The ultimate goal of most model develop- ments is to test alternative policies and to forecast future con- ditions, and not to recreate reality. Although agencies did acknowledge the quality of validation as important, the moti- vations for moving to advanced models, and the benefits achieved by doing so, overwhelmingly involved the ability to evaluate more sophisticated policies than existing models could be used for. The remainder of the chapter discusses specific benefits cited by the agencies. It is not a comprehensive list, but seeks to identify those common themes mentioned multiple times. Clearly the benefits of different types of models differ; there- fore, the benefits are segmented by the class of model. A section is then presented that discusses the types of policies an agency might consider and the benefits of advanced mod- els for evaluating those policies. ACTIVITY-BASED MODELS The enhanced framework of an activity-based model system is to be sensitive to policy changes in a consistent way across more dimensions. In other words, when the time or cost of travel changes, an activity-based model would consider trav- elers responding by changing route, mode, time-of-day, des- tination, frequency of travel, or auto ownership. A traditional model would only consider changes to route, mode, or desti- nation, often with route and destination sensitive only to highway travel time and not to changes in cost. In certain applications, the added sensitivity is of particu- lar importance. For example, in the San Francisco Mobility, Access and Pricing Study (MAPS), peak period tolls were considered for a cordon surrounding downtown San Fran- cisco, with the goal of providing an incentive for travelers to shift out of the peak periods or to switch to transit. In this CHAPTER THREE BENEFITS OF ADVANCED MODELS

33 application it was important that the time-of-day models be sensitive to cost, and the choice of destinations consider cost so that the study team could understand how many people would shop or recreate elsewhere, rather than switching to transit or traveling in the off peak. Also, because the applica- tion linked trips into tours, and simulated individual travel- ers, it was able to model area pricing scenarios in which travelers would pay a single daily fee to bring their vehicle downtown and then could travel as much as they wanted for the rest of the day. In either case, the models are calibrated to match base-year conditions, so it is not clear that activity-based models are any better at replicating base-year traffic counts or transit volumes than traditional models. The true advantage is that they are sensitive to a broader range of policies and can answer more complicated questions. If planners and decision makers ask the same questions of the new models as the old, their value will be limited. However, if the planning and modeling processes evolve together, their value can be much greater. Beyond the additional dimensions of sensitivity, five pri- mary advantages of activity-based models were identified: (1) eliminating NHB trips, (2) allowing for a more detailed analysis of outputs, (3) improved ability to model pricing, (4) more detailed representation of time, and (5) a greater ease of extensibility. These are discussed in further detail here. Eliminating Non-Home-Based Trips As discussed in chapter two, traditional trip-based models commonly include a category of trips called non-home-based (NHB), which neither begin nor end at home. This could be a trip from work to a doctor’s appointment or from the gym to a grocery store. NHB trips are difficult to model because the information available to simulate them is limited in scope to the two endpoints of the trip. This limitation has two impli- cations. First, because neither end is at home, the model can- not include any demographic or socioeconomic characteris- tics of the traveler when modeling NHB trips. For example, although we know that auto ownership is a highly important predictor of transit use, NHB trips cannot be segmented by auto ownership, and the explanatory power of that variable is lost. Second, it is impossible to know the context of the trips—whether it is a stop for coffee as part of a longer work commute, a trip to get lunch, or a trip on a multi-part shop- ping tour. It would be logical to assume that a stop on a work commute would occur somewhere between home and work; however, a trip-based model does not have that context and cannot account for that. Therefore, in the words of Gordon Schultz, an early pioneer in travel modeling, “you just sort of smear them around.” Activity-based models overcome this limitation by chain- ing trips into tours that both start and end at home. In this way, NHB trips are connected to the home location and detailed attributes of the traveler, such as auto ownership and income, can be considered when modeling those trips. Fur- thermore, the locations and modes of those trips are con- strained to be consistent with the surrounding trips. Therefore, in the example of stopping for coffee on the way to work, the NHB trip will have one end at work and one end between home and work. Further, with the knowledge that the trip is really part of a larger work commute, specific preferences (propensity toward transit, time-of-day characteristics) asso- ciated with work commutes can be considered. Thus, the chaining of NHB trips is about making the mod- els more accurate, and making them respond more logically to changes. The SFCTA specifically reported that they have been pleasantly surprised with the way their model responds logically to the wide range of policy alternatives they tested. In addition, agencies mentioned several ancillary benefits to eliminate NHB as a category of trips. For one thing, agen- cies noted that the model is actually easier to explain when the model more closely replicates reality. Anyone who has tried to explain what a HBW trip or NHB trip is at a public meeting can attest to this. In an activity-based model, it is much easier to show the results for work tours and have it align with everyone’s intuitive understanding of what a work commute is. A related benefit noted is the ability to trace all of a house- hold’s travel back to the household itself. This issue is of par- ticular importance to MPOs in California who are subject to the planning requirements outlined in Senate Bill 375, which seeks to reduce greenhouse gasses. Both the Metropolitan Transportation Commission (MTC) and the Sacramento Area Council of Governments (SACOG) cited the ability to trace vehicle miles traveled (VMT) back to individual households as important when evaluating the effects of the location of new households in established urban areas versus in sub- urban locations. With this detailed information, the analyst is much better able to do a thorough accounting of transporta- tion affordability, an issue of growing importance to many agencies. By eliminating NHB trips, all costs can be traced back to the household, and with the simulated population it is possible to know which households are incurring which costs. Therefore, it is easy to know how many low-income versus high-income households are paying for a toll increase. Detailed Analysis of Outputs Another benefit to activity-based models that was cited sev- eral times is the ability to do a much more detailed analy- sis of the outputs. In a trip-based model, the results of the demand models are sets of trip tables segmented by purpose and mode. In an activity-based model, the decisions of indi- vidual travelers are simulated, so the results of the demand models are a list of individual households, persons, tours, and trips that look just like what you get from a household travel survey. Therefore, the results can be reported across any number of dimensions. For example, it is possible to tabulate

commuter rail riders by income, by age, by gender, by the number of trips on a tour, or by any other category included in the synthetic population or in the results. The ability to tabulate the results in this way truly enhances the analyst’s ability to do environmental justice analysis and understand how projects affect different categories of people. Moreover, the structure of these models appears to be more amenable to the modeling of nonmotorized transportation. This is partially because of the increased level of spatial reso- lution typically built into these models, which enables analy- ses at the same scale at which such travel often takes place. They can also incorporate more variables that directly influ- ence the choice of nonmotorized travel, as well as focusing on the subset of the population most likely to engage in it. How- ever, it should be noted that explicit representation of non- motorized transportation is not inherent in activity-based mod- els. As with traditional models their inclusion must be explicitly accounted for in data and model design and development. Pricing Another commonly mentioned benefit to activity-based mod- els is an enhanced ability to model pricing. Traditional models either assume that all travelers have the same value of time, or segment the value of time by three (or so) income groups. Research has shown that even at the same income level, trav- elers can have widely different values of time. Ignoring this reality leads to aggregation error, particularly at higher price points, where a relatively small, but still significant, number of travelers may be willing to pay. At the Innovations in Travel Modeling 2008 conference in Portland, Oregon, one of the core topics was pricing, and many speakers emphasized the impor- tance of accounting for this distribution in values of time. Because activity-based models simulate individual travel- ers instead of aggregate matrices of trips, they provide the ability to assign each traveler their individual value of time, thus simulating this continuous distribution. In addition to value of time, their disaggregate nature allows for a more coherent accounting of cost itself. In many downtown areas, for example, individual travelers pay vastly different parking costs depending on any subsidies or free parking provided by employers. Usually the high-income workers pay the least to park, because they are more likely to receive parking as an employee benefit. Typically, models use an average parking cost in modeling mode choice, but activity-based models allow for an explicit modeling of who pays what, enhancing both the model’s accuracy and the accounting of transportation affordability. Furthermore, a new type of pricing policy currently being implemented in London, Singapore, and Stockholm; previ- ously considered in New York; and currently under consider- ation in San Francisco, is area pricing. In such a scenario, trav- elers would pay to enter a downtown area, and once they have paid for that day, they can drive around as much as they want. A trip-based model would struggle with such a scenario, because there is no way to know whether a traveler has already paid or not; however, with the chaining of trips and traceabil- ity back to the traveler it is explicit in an activity-based model. The types of pricing policies that can be tested are numerous—it is possible to exempt low-income households from paying tolls, give seniors and youth transit pass dis- counts, test the effect of eliminated employer-paid parking, or vary prices by time-of-day. With their enhanced ability to model variation in the population and the wide range of poli- cies that can be tested, activity-based models offer a superior platform for evaluating pricing alternatives. Time-of-Day A situation where specific enhancements are warranted is not unusual in an advanced model. The ability to deal with complex policies, such as congestion pricing, in a meaning- ful way is not something obtained “out of the box” in an activity-based model. However, starting from the framework of an activity-based model opens the door to a credible and robust treatment of time, whereas the options are much more restricted with a trip-based model. For example, it is possible to build a time-of-day model in an activity-based model that does not consider travel time; however, the model will then be completely insensitive to congestion. Adequate attention must be paid to that detail to obtain reasonable sensitivity in the model, although the same would be required in a trip- based model as well. The difference is that there is more that can be done in an activity-based modeling framework. In such models the context of the trip is known in terms of what other activities happen before or after it. These other activities serve as constraints, making travelers less sensitive to congestion than would be apparent in an unconstrained (trip-based) approach. Also, the disaggregate framework allows for the considera- tion of individual preferences. This is useful in the San Francisco peak spreading models, where for each trip a “pre- ferred” departure time is selected. This then determines how much the traveler is willing to shift away from their preferred time to avoid congestion. Similarly, for an agency interested in peak period pricing, the disaggregate approach allows each simulated traveler to be assigned his or her own value of time. This heterogeneity reflects that some people are much more sensitive to pricing than others, which a trip-based model simply cannot account for. Extensibility One additional and unanticipated benefit has been realized by those agencies that have the longest history in applying and maintaining activity-based models—that of extensibil- ity. Because of the disaggregate nature of these models, it is 34

35 actually quite easy to add a new descriptive variable to the model system. In an activity-based model it is as simple as adding a column to a table, whereas in a trip-based model it involves further segmentation of trip matrices, which can quickly become unwieldy. Furthermore, the ability to simu- late individual travelers greatly enhances the types of policies that can be tested. One example is the New York Metropolitan Transportation Council model, which was successfully applied to test license plate rationing in Manhattan. The premise is that on any given day, only autos with license plate numbers ending in certain digits could enter lower Manhattan. License plate numbers were randomly assigned to each vehicle in the simulated house- holds and, depending on which vehicles were available for use, modal alternatives were made available or unavailable. Such a policy could not have been tested using a traditional model. Further, model enhancements undertaken as part of the San Francisco MAPS showed that extending the existing model system to account for distributed values of time, track area pricing, and enhanced peak spreading models were a relatively moderate effort compared with the contortions that would have to be made to a trip-based model system to achieve a similar result. Simply stated, disaggregate activity-based models offer a platform that is readily adaptable to evaluating a broad range of policy alternatives. The ability to model complex policies does not come without a cost—the modelers must still pay attention to the details to make sure that each indi- vidual model component is sensitive to the right variables; however, the platform provides the ability to do this. This attribute is extremely appealing in agencies where the next big policy issue may remain unknown. DYNAMIC NETWORK MODELS Traditional user equilibrium highway assignment models pre- dict the effects of congestion and the routing changes of traf- fic as a result of that congestion. They neglect, however, many of the details of real-world traffic operations, such as queuing, shock waves, and signalization. Such operational details are important both to reflect reality and to evaluate policies asso- ciated with improving traffic operations. Examples of such policies might include ramp metering, signal coordination, or targeted improvements at choke points. Currently, it is common practice to feed the results of user equilibrium traffic assignments into dynamic network models as a mechanism for evaluating these policies. The simula- tion models themselves, however, do not predict the routing of traffic, and therefore are unable to account for re-routing owing to changes in congestion levels or policy, and can be inconsistent with the routes determined by the assignment. Dynamic network models overcome this dichotomy by com- bining a time-dependent shortest path algorithm with some type of simulation (often meso- or macroscopic) of link travel times and delay. In doing so it allows added reality and con- sistency in the assignment step, as well as the ability to eval- uate policies designed to improve traffic operations. The nonlinear and chaotic nature of congestion at the micro scale and its effect on vehicular flow characteristics is well documented (May 1990; Newell 1995; Boyles et al. 2008). The benefits of modeling individual vehicle interactions and how they collectively give rise to level of service and conges- tion have long been captured in traffic simulation models. Such models have traditionally been tractable for small study areas, owing to their data requirements and heavy computa- tional demands. Other than a few isolated attempts to simulate large networks, such models were not considered practical at the urban or metropolitan level irrespective of what benefits they might have offered. TRANSIMS arguably changed that, demonstrating proof of concept from its inaugural case study in Dallas–Fort Worth to early deployments today. During the same timeframe that TRANSIMS has evolved, separate progress has been made in the use of DTA models in planning studies, an application not widely anticipated a decade ago. Given their successes and the growing interest in fusing activity-based travel demand models with dynamic network models it is likely that such models will become more widely used in practice over the coming decade. As with activity-based models in general, practitioners are eager to learn what practical advantages such models have over tradi- tional static traffic assignment models. There is arguably insufficient evidence at this writing to conclusively show their superiority. Regardless of the level of network resolution, such models have already provided indispensible benefits that cannot be obtained using traditional static network models: • Dynamic network models are capable of capturing the time-dependent effects of congestion, something that static models are not capable of doing. The incidence, location, and duration of congestion, often evidenced as bottlenecks in the roadway system, lead to highly unstable flows, high variabilities in travel times, and unreliable system states. These effects can only be represented in an aggregate sense—in both time and space—in macroscopic models, which assume invariant flows and travel times over the entire analysis period. Even when modeling peak periods the variability in travel times and their effect on departure time, mode, destination, and route choice varies considerably within that time. Static models are simply unresponsive to such variation, and to the extent that they can approximate the macroscopic outcomes, do so in a manner inconsistent with current traffic flow theory. • Macroscopic models represent traffic control in an abstract manner, such that improvements in that realm go unnoticed. Traffic signals are the most abundant control strategy in place at the present time, although areawide control schemes, ITS, and traveler informa-

tion systems are rapidly increasing in importance. With management and operation of the transportation system playing an ever-increasingly important role, the need for tools appropriately sensitive to such actions is criti- cal. Dynamic network models fill this gap. • Dynamic tolling and congestion pricing schemes rely heavily on accurate and detailed information about tem- poral patterns of demand and their response to changes in levels of service. To the extent that such schemes become more commonplace in the future investors in and operators of such systems will require more robust estimates of network performance and response than can be obtained from static models. Dynamic network models are ideally suited for such analyses. • Global climate change is renewing the focus on mobile source emissions. California has adopted statewide poli- cies to dramatically reduce mobile source emissions (California Air Resources Board 2009). The recent pro- posed federal “cap and trade” legislation includes regu- latory powers for the EPA over emissions estimation methods and standards. These changes will require more accurate estimation of link and network travel times, which in turn will require a more robust representation of the time-dependent effects of congestion and traffic control systems. By contrast, most static models resort to using post-processing of macroscopic assignment results to derive credible estimates of link travel times. It is expected that the various types of dynamic network models will converge at some point. Some commercial pack- ages offer the capability to seamlessly move between varying levels of detail and, depending on the flow levels, between dif- ferent models (macroscopic, mesoscopic, and microscopic). If DTA models fulfill expectations, it is possible they will be capable of providing most of the performance measures required, and at an acceptable level of resolution and accuracy, without resort to traffic simulation models. LAND USE MODELS Adding a land use model to a travel demand model adds a large set of land use-related policy scenarios that can be tested and improves traffic forecasts through better travel demand input data. If emissions are of interest, a land use model allows adding emissions from dwellings and firms to traffic emis- sions. Such models can be used to analyze land use policies as well, ranging from the implementation of growth bound- aries to tax incentives for transit-oriented development. Imple- menting a land use model may also improve the functioning of travel demand models, be they traditional or advanced for- mulations. For example, if an additional highway reduces congestion at the beginning, subsequent relocation of house- holds may dampen the congestion relief. If changes in land use are simulated explicitly, rather than using fixed exoge- nous forecasts, the quality of a base forecast as well as the responsiveness of the travel model to alternative policies may be significant. Extended Scenario Analysis For many agencies the driving motivation for using a land use model has been the capability to analyze a wider variety of sce- narios. Agencies seeking to reduce urban sprawl by zoning or an urban growth boundary have used land use models to better understand the impact of development restrictions. Concerns about rising land prices owing to zoning can be analyzed with a land use model before the urban growth boundary is estab- lished. Impacts on land use patterns and the demographic distribution of households by, for instance income, help to better understand if zoning policies may have unexpected side effects. After Cervero and Kockelman (1997) published a paper on how the three Ds (density, diversity, and design) affect travel demand, several agencies across the country have used land use models to simulate the impact of an alternative urban design. Land use models also allow for simulating the effect of subsidies. If an agency attempts to vitalize a depressed region by subsidizing the development of an industrial busi- ness park, land use models may be used to analyze the likely demand for such kind of development. Transportation–Land Use Feedback Cycle In chapter two, the transportation/land use feedback cycle was described, explaining how transportation influences land use and how, in turn, land use creates transportation demand. Integrating a transportation model with a land use model implements the entire feedback cycle, which helps account for induced travel demand. New transportation infrastructure affects land use pat- terns. Today, large-scale infrastructure developments are fairly uncommon. Kreibich (1978) proved that the exten- sion of the commuter rail system in Munich for the Olympic summer games significantly fostered urban sprawl within the catchment areas of the train stations. Likewise, high-speed rail or highway developments may encourage people to increase the distance between home and workplace. Conversely, a changing demography may alter travel demand substantially. If over time a neighborhood develops to become a retirement area the daily travel demand may be reduced significantly. A change in average income within a neighborhood has an effect on auto ownership and, hence, mode choice of people living in this area. Land use models allow accounting for these land use–transportation interactions. Emissions from Non-Transportation Sources Several states are beginning to analyze carbon dioxide (CO2) emissions as part of climate change and greenhouse gas emission reduction strategies. As this point, most efforts con- centrate on estimating the emissions from mobile sources. Given that less than half of the CO2 emissions in most areas originate from the transportation sector, a reasonable next step will be to estimate emissions from land uses. Oregon’s 36

37 GreenSTEP model (Gregor 2009) is an example of an emerg- ing methodology for estimating emissions from fixed-point sources. FREIGHT AND COMMERCIAL MOVEMENT MODELS Freight and commercial movement models offer the ability to analyze the effects of transportation improvements on freight. Furthermore, where they contribute a significant portion of traffic volumes, they enhance a model’s ability to properly forecast traffic congestion. Hunt and Stefan (2007) estimated that commercial traffic—of which freight is but one aspect— comprised roughly 20% of all VMT in Calgary in 2002. The percentage of VMT attributed to commercial vehicles appears to be increasing in most urban areas; growing faster than either the economy or auto flows in recent years (Trans- portation Research Board 2003b; Downs 2004). Moreover, the impact of trucks on air quality has become a major issue in many metropolitan areas. Although nonfreight commercial vehicle flows, such as ser- vice and sales travel, might arguably be better captured in tour- or activity-based travel models because of their ability to model multi-stop itineraries, it is likely that these unique travel pat- terns are poorly represented by current models. Attempting to account for them in trip-based models is typically accom- plished by expanding NHB trips to account for the missing trips. Such misses affect the dynamics of such flows as well as their spatial characteristics. The approach adopted in Calgary is one approach to explicitly modeling such behavior. However, further advances in tour- and activity-based models are likely to give rise to progressively more sophisticated models of firms and their contribution to travel demand. Although employment (as a surrogate for firms) is still modeled in aggregate in most advanced models, opportunities exist to better account for work-related person travel. This is a fertile area of research that is expected to expand in the next few years. The separate modeling of urban freight has been carried out for several decades. As noted in chapter two, most such mod- els are analogues of traditional trip-based person-travel mod- els. Although they perform acceptably in most cases, it is well known that some important dynamics of freight are missed. Tours are even more prevalent and more important than in- person travel. Moreover, the use of urban distribution centers has greatly increased over the past two decades. As a conse- quence, a large volume of goods that formerly was delivered directly to firms and households by long-distance trucks now unload at a single distribution center. Deliveries to local cus- tomers are consolidated and made as needed from these dis- tribution centers. Donnelly et al. (2002), analyzing data from Canada, found that one-half of all intercity truck flows were destined to or from a distribution center, a number that has likely increased since then. Thus, the dynamics of distribu- tion centers and their effects are so important they can no longer be ignored in freight models. Models capable of cap- turing their dynamics will provide a much more credible basis for representing urban freight. An advanced freight model will be capable of capturing such dynamics. It will also be capable of explicitly representing other important factors influencing the demand for freight and its impact on the transportation system. Freight movements are influenced by several different actors (e.g., shippers, con- sumers, carriers, and intermediaries) that often do not share the same goals or information. Almost all person trips begin and end within an urban area; however, freight movements often have one or both trip ends outside of the place under study. Thus, linkages to statewide and regional truck models, a likely trend in the near future, will continue to grow in importance. The freight industry has undergone tremendous change in recent years. These changes include the deregulation of the trucking industry, the birth and expansion of supply chain logistics and just-in-time deliveries, the advent of third-party logistics companies, the widespread adoption and tremendous growth in container traffic, the early adoption of ITS in freight, the birth and rapid growth of shippers such as United Parcel Service and Federal Express and retailers such as Wal-Mart, globalization of trade, e-commerce, and the rise of regional and urban distribution centers—to name but a few. More changes are on the horizon, such as the increasing sophistication of con- tainers that enable logging, remote tracking, and integrity monitoring. Radio frequency identification tagging and other digital short-range communications technology will heavily influence the freight industry in the coming years as well. It is unlikely that any urban freight model will capture all of these factors in the near future. However, the gulf between practice and need can be usefully bridged in two ways. The first is the adoption of a framework that allows for a flexi- ble but holistic representation of all actors, resources, and con- straints in the system. It is important that each actor or market segment be represented in the physical and behavioral struc- tures most appropriate to it, rather than forcing it to fit within a certain modeling framework. Moreover, response to policies or exogenous constraints could be exerted in a realistic manner recognizable to end users of the model. The effect of differen- tial pricing, for example, could be realized in the mode choice decisions made by those actors affected by the costs. It is also apparent that a system of models can be con- structed that will capitalize on the strengths of different modeling approaches at each level of the problem. Xu et al. (2003) describes a three-level modeling framework that uses price and commodity signals between them. The design is compelling and clearly demonstrates the utility and validity of multi-scale approaches to freight modeling. Taken together, a hybrid of both approaches offers advantages that cannot be obtained through either alone. The emerging crop of microsimulation-based freight mod- els be they based on rule systems, behavioral constructs, or

a combination thereof, are capable of incorporating these many unique facets of freight and nonfreight commercial travel patterns. To the extent that policymakers and interest groups are interested in better understanding and forecasting them, the various advanced options for modeling this market segment will prove invaluable. STATEWIDE MODELS Statewide travel models have been used over the past several decades to examine long-distance and intercity travel, review linkages between urban areas, focus on freight, and standard- ize modeling throughout a state. They are not a separate class of models per se, but rather are characterized by the scale at which they operate. Tour- and activity-based statewide travel models enjoy the same benefits described previously. How- ever, an increasing number of statewide models also have explicit economic and freight components, and several recent ones (including Maryland, Ohio, and Oregon) include formal land use models, with their attendant benefits, as well. A com- pilation of the benefits of statewide models is not reported in the literature, but was gleaned from interviews, review of peer reviews (Transportation Research Board 2005, 2006), and case studies cited by Horowitz (2006). Some of the major ben- efits cited follow. • Statewide models are capable of estimating intercity travel, which can be a significant share of statewide VMT in eastern states and in states with several metropolitan areas. Most urban models represent such trips as infor- mation-poor external trips. Statewide models can make explicit external tripmaker characteristics, origins, desti- nation, and other attributes. • Most statewide models do not end at the state border, but provide successively more aggregate spatial repre- sentation as distance outside the state increases. Such models can be used to study specific intercity markets, such as high-speed rail, corridors of national signifi- cance, the effect of major highway closures, and simi- lar scenarios where the influential factors are far larger than or far from an given urban area. • In several states statewide modeling is not only a single model at the statewide level, but a set of standard models and methods applied consistently across the state. Michi- gan and Oregon have used such an approach for several years. In both cases the major metropolitan area uses its own approach, but interfaces loosely with the statewide model. In the case of Oregon, a standard four-step model is used by all urban areas (including Portland). As a result, all agencies are able to share in the development costs, data collection, and knowledge transfer. • Statewide models are also used to ensure consistency in demand estimation and impact measurement between urban areas within a state. Statewide transportation plans and programming is seen as more consistent in such cases. Florida and Iowa have cited this benefit as a signif- icant motivating factor for investing in statewide models. • Owing to their inclusion of or linkage to macroeconomic models, some statewide models are capable of generating fairly detailed estimates of direct and indirect user bene- fits. Indiana and Ohio have both developed formal eco- nomic impact components of their statewide models. Oregon uses its statewide model to estimate the impact of transportation decisions on job retention and local economic impacts. In most states these analyses can be reported at the local, corridor, and statewide levels. • Freight flows are much more likely to cross the urban cor- don than person travel. In some instances the emphasis at the statewide level is as much on freight as it is for person travel. Even more so than for auto traffic, the delineation of origin–destination patterns and truck and cargo char- acteristics are important data for policy studies. The benefits and utility of statewide models can be extended even further through the adoption of multi-level modeling, as described in the previous chapter. In such cases infor- mation can flow between models at varying scales, such as between regional, statewide, and urban models. Data can be transformed as it passes between levels or simply used at the spatial and temporal scale at which it is provided. An excellent discussion of the benefits of multi-scale modeling can be found in Nagurney et al. (2002), with an innovative application to regional freight modeling described by Xu et al. (2003). SELECTING A MODEL APPROPRIATE TO POLICY QUESTIONS OF INTEREST The modelers at agencies that have moved or are moving to advanced models noted that they are motivated by the more complex policy questions that their boards and planners are asking. As this report has come together, it has become increasingly clear that the right model is the one that best meets the policy needs of the agency. Depending on the agency’s specific needs, the selection of a model system and the appro- priate allocation of resources will vary. As a tool to assist in selecting which advanced models may be most appropriate to evaluating specific policies and the benefits that such models may offer, Table 4 shows types of policy questions, whether they can be answered with traditional models, what type of advanced model would be most beneficial, and what that benefit would be. 38

39 TABLE 4 ADVANCED MODEL ADVANTAGES FOR SPECIFIC POLICY QUESTIONS Theme My Policy Issues Include Can a Traditional Model Answer These Questions? Advanced Modeling Should Focus on Advanced Models Would Offer These Benefits Highway capacity projects Yes Activity-based models Eliminate NHB trips HOV lanes and carpooling Yes, if it includes a mode choice model that includes a choice of drive alone, shared ride (2 persons), or shared ride (3+ persons) Activity-based models with joint intra- household travel The bulk of shared ride trips are composed of members of the same household. A model that does not account for this behavior risks overstating travelersí willingness to form inter- household carpools. Time-of-day and peak spreading Yes, if it includes a peak spreading model Activity-based models with time-of-day choice sensitive to level-of- service Trip-based models cannot account for the constraints of adjacent activities or travel, and therefore risk overstating travelers’ willingness to shift times of day in response to congestion or pricing. Highway Traffic operations analysis (queuing, choke points, etc.) No Dynamic network models Standard user-equilibrium traffic assignments do not account for the dynamics of traffic progression. Dynamic network models can overcome this limitation. Major transit investments Yes Activity-based models Trip-chaining allows mode choice to consider the context of the trips. For example, transit must be available in both the departure and return period for it to be available, so there is an advantage to having a tour-based model that considers the level-of-service in both directions. Transit New Starts analysis Yes, with careful attention to detail Activity-based models, with careful attention to detail Same as above, but also note that the microsimulation framework would allow a more detailed analysis of the forecast markets using transit because it allows the results to be sliced and diced in more ways. Note in both cases that the fundamentals, such as transit path building, logical mode choice coefficients, and an understanding of the markets, are crucial. This part is the same in either case. Air quality conformity Yes End output is the same, but may offer better sensitivity to specific policies. Tracing greenhouse gas emissions to households and tours No, can calculate the total emissions, but not which households are responsible. Activity-based model Activity-based models eliminate the problem of NHB trips, allowing all travel to be traced back to the household. This allows for a better analysis of VMT per household when households are located in different zones. Tracing greenhouse gas emissions to vehicles No Dynamic network models Most dynamic network models can trace emissions of individual vehicles, as well as their disaggregate acceleration and deceleration profiles. This meshes well with the MOVES approach. Emissions and Greenhouse Gases Reducing greenhouse gas in region by a given percent Yes, with an emissions model Comprehensive emissions models With motor vehicles accounting for about half of greenhouse gas emissions transportation models can provide important insight into the effectiveness of strategies for reducing greenhouse gas emissions, when connected with an emissions model such as MOBILE6 or MOVES. Specifically, any strategies that reduce motor vehicle travel, such as transit investment or land use changes can be evaluated. However, there is a whole range of strategies for which a transport model is irrelevant, such as point source and area source emissions. For these, a more comprehensive set of tools is necessary. Tolling and pricing Yes Activity-based model with distributed value of time The microsimulation structure allows each traveler to be assigned an individual value of time, drawn from a continuous distribution. In a trip-based model, the variation in value of time is constrained to the number of market segments in the model, leading to aggregation error. Pricing (continued on the following page)

40 TABLE 4 (continued) Further, an activity-based model allows for a wide range of pricing policies that can be tested. It is possible to exempt people from households earning under XX dollars per year, model senior transit fare discounts, or understand how eliminating employer parking subsidies might affect transit use. HOT lanes Yes Activity-based model with distributed value of time and intra-household interactions See comments on Tolling and Pricing and on HOV lanes. Transportation affordability Partially Activity-based model Allows all costs to be traced back to households, including the cost of owning and operating vehicles, gas, tolls, transit fares, etc. Also allows the results to be sliced and diced in any way imaginable, to understand the implications for different subsets of the population. Effects of land use on travel Yes Activity-based model with sensitivity to land use characteristics Either model could be built with sensitivity to small-scale land use characteristics. Both are already sensitive to the households and employment in each zone. Effects of transportation investments on land use No Land use model Assuming the land use model is sensitive to accessibility measures from the transportation model it will be able to evaluate the effects of improvements on land use. Evaluating “reality” of land use plans No Land use model Land use model can serve as a reality check on plans to understand if they are supported by the market. Land Use Urban growth boundaries or infill development incentives No Land use model A land use model with the appropriate policy sensitivity will allow for the evaluation of land use specific policies, such as urban growth boundaries, or infill development incentives. Matching base-year traffic counts Yes Can potentially do a better job of matching traffic counts, but the real measure of a model is how it responds to change. Validation Detailed validation and understanding of markets Partially Activity-based models Because the advanced models can be summarized in so many different ways, it opens up a whole range of additional validation checks, and evaluation of travel markets that can be done, in addition to those traditional checks done for a trip-based model. This has the potential to raise the bar, because for many analyses in a trip-based model we would never have the option of know that we’re wrong. Effect of changing demographics Partially Activity-based models Microsimulation structure allows for the inclusion of a much greater range of demographic variables, and thus sensitivity to changing demographic conditions, such as an aging population. Interaction with Population Environmental justice Partially Activity-based models Model results can be sliced and diced in many different ways, allowing for better analysis of effects on different sub-populations. Truck lanes No Commercial and freight model Commercial and freight models are designed to forecast truck volumes, in addition to other modes. Economic impacts No Integrated economic model An economic model sensitive to accessibility measures from the transportation model is needed to evaluate economic impacts of transportation investments. Other Intercity transportation infrastructure No Statewide model A model with a larger geographic scope than single urban area is needed to evaluate intercity transportation infrastructure investments as well as for understanding intercity commodity flows. HOV = high-occupancy vehicle; HOT = high-occupancy toll.