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.

64 This report was designed to assess the benefits and challenges associated with moving beyond the traditional four-step travel demand modeling practice. A great deal of innovation in trans- portation and land use modeling has occurred over the past two decades, both in volume and its departure from their tradi- tional roots. However, a great number of the advances and lessons learned from them are readily accessible only to the relatively few who have been actively engaged in their pursuit. Therefore, the goal of this report was to condense this new knowledge into a form accessible to a much larger audience. Several major themes emerged during the review and study supporting this report. Perhaps the most fundamental issue sur- rounds the rationale for such models, where the motivating factors fall into two categories: those seeking more accuracy, and those seeking more sensitivity. This, in turn, appears to be influenced by the institutional settings of the modeler and their clients. Those who primarily use models to support the expan- sion of the existing transportation system appear focused on the need for more accurate models. To date there is no evidence that the advanced approaches described herein are inherently more accurate or more capa- ble of replicating observed traffic flows than their pre- decessors. Nor could examples of more accurate forecasts be identified. These modelers are likely to be disappointed by the gains in accuracy—or perceived lack thereof—obtained from advanced models. However, to the pioneers of these new approaches that largely misses the point. These models are compelling because of the questions that can be posed to them and the resolution and fidelity at which they can be approached. The enhanced framework of advanced models allows for the model system to be sensitive to policy changes in a consistent way across more dimensions. For example, when the time or cost of travel changes, an activity-based model can capture changes in route, mode, time-of-day, destination, party composition, frequency of travel, or auto ownership. A traditional model would only consider changes to route, mode, or destination, often with route and destination choice sensitive only to highway travel time and not to changes in cost or transit time. Such responses are important in certain applications, especially when issues such as induced demand, equity, or the response to changes in accessibility are considered. Many of these requirements and issues were never faced or anticipated by the creators of the four-step modeling paradigm. Such is hardly a criticism, as their work provided the tools that helped usher in an unprecedented era of construction of transportation infrastructure and systems. In places where construction of new major facilities continues today such tools unquestionably retain their utility. However, in many other places, the idea that we can “build our way out of congestion,” or that congestion is necessarily a failure that must be alle- viated, no longer describes the priorities of federal, state, or metropolitan transportation agencies. Many agencies we inter- viewed or interact with are grappling with issues such as equity, growth management, environmental quality, or the need to study scenarios such as fuel scarcity. To the extent that their focus has changed it hardly appears surprising that travel demand modeling must change in equally significant ways. Given the wide range of expectations and needs it is diffi- cult to define success for these models. Two studies are cur- rently underway that are closely examining the differences between trip- and tour-based models; however, neither are far enough along to report even preliminary results. Although the same criteria used to judge trip-based models are likely to apply to tour-based models the comparison is short-sighted, for the latter can provide measures and benefits simply un- attainable with the former. Although standards for calibra- tion and validation are still being defined, it can be argued that success could be measured in terms of client (decision- maker) satisfaction, ability of the model to appropriately respond to specific policy and investment scenarios, trans- parency, and tractability (in terms of run times, data require- ments, and other resources). It is not possible at this time to point to a large body of literature that can substantiate such benefits of advanced models. In the absence of such, an attempt has been made to report anecdotal evidence provided by the users of such models. With this broad context in mind we can return to the spe- cific issues addressed in this report. In chapter two, the report describes the current state of the practice in advanced land use and transportation models. The coverage is not comprehen- sive by any means, but rather focused on operational models used in practice in North America. Several successful exam- ples of advanced models are provided, suggesting that ample evidence of their efficacy exists and that tangible benefits CHAPTER SEVEN CONCLUSION

65 have emerged from their implementation. The benefits of these models, as reported by their developers and users, are summarized in chapter three. Many of the topics covered in chapters two and three have already been reported in the lit- erature and discussed at recent conferences, but are summa- rized here for completeness. What has long been sought is an examination of the institu- tional issues surrounding the adoption of such models and the key lessons learned along the way. Chapters four and five are devoted to these topics and considered the key contributions of this report. Summarizing the most important aspects of what has been learned is challenging, because the significance attributed to them varied by agency and many are interrelated. However, several issues appear to define the challenges faced by the pioneers, and there is ample reason to believe that others will face them as well. There was widespread agreement that human assets are the key factor limiting the adoption of advanced models. In almost all instances—and most certainly in the most suc- cessful ones—a visionary champion was clearly identified as the sustaining force behind the adoption of advanced models. It was widely believed that this champion, with the support of upper management in the agency, was the single most important ingredient for success. Although a consultant can play this role the results to date have been less satisfactory. Absent strong federal leadership in this area it is likely that the importance of the champion—from both management and technical standpoints—will remain high in the move toward advanced models. An equally significant human constraint is the lack of agency staff capable of creatively and competently building and using advanced models. Most agencies do not have devel- opers on staff, because they are difficult to find and afford. Moreover, unless the agency is continually developing new models they often seek out new challenges elsewhere once development gives way to application. As a consequence, there is a heavy reliance on a small number of consultants with the skills necessary to develop such models. The obvious potential for over-subscription aside, it was found from our interviews that many believed that the rest of the profession is falling progressively further behind the developers. That is, the knowledge and experience of the pioneers is perceived to be increasing more rapidly than their colleagues not involved in such pursuits. It is apparent that reading papers and attend- ing conferences and short seminars cannot impart the neces- sary skills, yet absent studying under the few academics active in this area there appear to be no available resources for train- ing new practitioners and, more importantly, retraining the current workforce. Moreover, there appeared to be agreement that widespread successes with innovative models cannot be achieved until this limitation is overcome. There appeared to be a widespread perception that advanced models are far more data-intensive than traditional models. This might be true for land use and freight models if they replace sketch planning methods, models dependent on obsolete data, or no models at all. An agency using a land use model in conjunction with a travel demand model will have larger data requirements than one that does not. Current land use modeling practice is characterized by complex models with large data requirements. However, where this complex- ity is not required the data requirements are much more mod- erate. Dynamic network models are also likely to have larger data requirements. More detailed network representations and explicit representation of traffic signals and control schemes, not required by traditional static traffic assignment models, are required. Adopters of all such models will unquestionably bear the burden of expanding both their data collection and modeling efforts. Ironically, the same is not necessarily true for tour- or activity-based travel demand models, against which such objections are more commonly lodged. Small changes to the structure of traditional household travel surveys are required, but otherwise they require substantially the same data as sequential trip-based models. To date, larger surveys have been specified for some of the models developed; however, their size has been driven by the desire for greater resolution and fidelity in the models. Equally larger surveys would have been required had the same increased level of detail been sought for traditional models. Otherwise, the data required are very similar for both modeling approaches. The microsimulated households—which do result in much richer detail within the model—are synthetically generated from easily tabulated marginal summaries rather than highly detailed exogenous data. Thus, the evidence gained by the experience with such models to date cannot support the per- ception that they have substantially more onerous data require- ments simply because they are activity-based models, but rather because of the improved sensitivity of the model and greater levels of detail. All of this is not to say that the costs of adopting advanced models are small. They are not. Although efforts to collect hard data on the costs of projects to date has proven diffi- cult (and even harder to draw conclusions from), it can be observed that the implementation of advanced models has taken several years, required outside assistance, and would not have succeeded without a highly committed and empow- ered internal champion. In many instances the development of these models had to be preceded by data collection efforts, and their continued evolution depends on their continuation. Whether such an investment can be justified or sustained depends on the issues the agency is facing or expects to tackle within the next decade. For the pioneers the opportunity cost (of not moving forward) was too large a price to pay. For them the decision is easy, as the investment represented the cost of remaining relevant to decision makers facing much different issues than those that traditional models are ideally suited to address. These pioneers have borne the higher ini- tial development costs. Subsequent adopters will benefit

greatly from the knowledge, software, and successes they have put in place. It might be argued that the cases for some of the advanced models described in this report are less compelling than for tour-based travel modeling, and even more dictated by local needs and capabilities. There are not as many success stories to learn from, although this situation will be remedied over the next five years. In the case of freight models, the consultant and internal costs appear to be less than for activity-based person- travel models, but that depends on the level of detail and sophis- tication sought. The evidence that such models perform better than more traditional urban truck models is mixed. Moreover, it is clear that the evolution of such models will entail signifi- cant data collection efforts. There are too few successes in land use modeling at the urban or statewide scale to generalize about what is required to deploy them and, as of this writing, there are only a few prototypes of dynamic network models at a city- wide or regional scale. The few that have been attempted have devoted a substantial amount of resources to testing, debug- ging, and developing the underlying software, making them unreliable indicators of the likely costs faced by future imple- menters. As with the advanced travel models, the pioneers have absorbed the initial research and development costs. The effort and resources required to implement these models will be sig- nificantly reduced for later adopters. Finally, it can be noted that any foray into advanced mod- eling is as much a change in mindset as methodology. The four-step sequential modeling paradigm, and to some extent the modeling of land use with DRAM-EMPAL, are mature practices whose evolution has virtually ceased. Equally mature software is available for them, and the challenges in maintaining existing models are small. The opposite is true for advanced models, which in all cases are still evolving in both theoretical and methodological terms, and their software implementation is far from standard. These issues will be overcome in time, and the evidence to date indicates that they are not insurmountable obstacles. However, advanced models and their implementations are likely to continue to evolve in significant ways over the next decade as they adapt to the need to respond to challenges such as carbon footprint- ing and pricing, continued uncertainty about fuel futures, tech- nological changes, and even greater needs to assess the micro- economic impacts and benefits of transportation projects. If that were not enough, some of the luminaries in the field of modeling are also questioning the very foundation of the practice. Wegener in particular advocates a return from mod- els defined by statistical data mining to more theoretical structures capable of supporting explicit risk and uncertainty analyses in forecasts. Given the changing political and eco- nomic landscape the need for expansive thinking on such fundamental levels is imperative. This, in turn, will continue to drive the evolution of advanced models used by acade- mics and practitioners. As such, advanced modeling is bet- ter thought of as a journey than a destination. The tools, methods, and practices described in this report mark the start- ing point of such a journey. 66