National Academies Press: OpenBook

Statewide Travel Forecasting Models (2006)

Chapter: Chapter Two - Survey of Statewide Travel Forecasting Practice

« Previous: Chapter One - Introduction
Page 12
Suggested Citation:"Chapter Two - Survey of Statewide Travel Forecasting Practice." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 12
Page 13
Suggested Citation:"Chapter Two - Survey of Statewide Travel Forecasting Practice." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 13
Page 14
Suggested Citation:"Chapter Two - Survey of Statewide Travel Forecasting Practice." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 14
Page 15
Suggested Citation:"Chapter Two - Survey of Statewide Travel Forecasting Practice." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 15
Page 16
Suggested Citation:"Chapter Two - Survey of Statewide Travel Forecasting Practice." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 16
Page 17
Suggested Citation:"Chapter Two - Survey of Statewide Travel Forecasting Practice." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 17
Page 18
Suggested Citation:"Chapter Two - Survey of Statewide Travel Forecasting Practice." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 18
Page 19
Suggested Citation:"Chapter Two - Survey of Statewide Travel Forecasting Practice." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 19
Page 20
Suggested Citation:"Chapter Two - Survey of Statewide Travel Forecasting Practice." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 20
Page 21
Suggested Citation:"Chapter Two - Survey of Statewide Travel Forecasting Practice." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 21
Page 22
Suggested Citation:"Chapter Two - Survey of Statewide Travel Forecasting Practice." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 22
Page 23
Suggested Citation:"Chapter Two - Survey of Statewide Travel Forecasting Practice." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 23
Page 24
Suggested Citation:"Chapter Two - Survey of Statewide Travel Forecasting Practice." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 24
Page 25
Suggested Citation:"Chapter Two - Survey of Statewide Travel Forecasting Practice." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 25
Page 26
Suggested Citation:"Chapter Two - Survey of Statewide Travel Forecasting Practice." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 26
Page 27
Suggested Citation:"Chapter Two - Survey of Statewide Travel Forecasting Practice." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 27
Page 28
Suggested Citation:"Chapter Two - Survey of Statewide Travel Forecasting Practice." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 28
Page 29
Suggested Citation:"Chapter Two - Survey of Statewide Travel Forecasting Practice." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 29
Page 30
Suggested Citation:"Chapter Two - Survey of Statewide Travel Forecasting Practice." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 30
Page 31
Suggested Citation:"Chapter Two - Survey of Statewide Travel Forecasting Practice." National Academies of Sciences, Engineering, and Medicine. 2006. Statewide Travel Forecasting Models. Washington, DC: The National Academies Press. doi: 10.17226/13958.
×
Page 31

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.

13 SURVEY METHODOLOGY The survey of states involved four stages; taking advantage of a Statewide Travel Demand Models Peer Exchange held in September 2004. First, the 14 states planning to attend the Peer Exchange were asked to answer several open-ended questions about their models, model creation, and model ap- plication. Second, an analysis of these responses was used to create the multiple-choice questions that would be answered by some or all states. Third, a very short screening question- naire was prepared (see Appendix A) and e-mailed to all states to ascertain their general level of modeling capability and alternatives to modeling. Fourth, those states found to be reasonably far along in their model development process were mailed one of two follow-up questionnaires. The longer of the two follow-up questionnaires was sent to states not participating in the Peer Exchange and the shorter form was sent to those states represented at the Peer Exchange. The longer form is found in Appendix B. The shorter form omit- ted questions that appeared to be adequately covered by the Peer Exchange questionnaire. All states except Hawaii responded to the screening ques- tionnaire. Of all the states with models, only Louisiana and Oregon did not return the follow-up questionnaire; however, both states gave extensive responses to the Peer Exchange questionnaire and provided model documentation. Montana’s response to the survey indicated that it did not have a statewide model; however, its HEAT (Highway Eco- nomic Analysis Tool) (Cambridge Systematics, Inc.; Eco- nomic Development Research Group; ICF Consulting; and Short Elliott Hendrickson, Inc. 2004) has an embedded freight component that is similar in structure to those in other statewide models. Responses from Montana included here were based on a report about HEAT. RURAL TRAFFIC FORECASTING NOT INVOLVING STATEWIDE MODELS All states without models and some states with models han- dle project-level traffic forecasts through simpler techniques, such as growth factors and trend lines. Many states do not have a fixed methodology that applies to projects in general, whereas other states have implemented a standard technique for use everywhere. Of the 32 states reporting that they used simpler techniques, a majority reported using linear trend lines applied to historical count data. A few states use growth factors and one state (Wisconsin) uses Box–Cox re- gression, which produces a nonlinear trend line, heavily weighted toward higher traffic volumes. Box–Cox regres- sion is described in the Guidebook (Horowitz 1999). South Dakota establishes growth rates by regression analysis of business data and historical vehicle-miles traveled (VMT) treads by county (Johnson 2000). A number of states find it necessary to modify their historical trend lines with local knowledge and other forecasts of business and population growth. It is not uncommon for a state to develop growth fac- tors by highway functional class and by region within the state. Commercial vehicles are sometimes forecasted sepa- rately from passenger cars. Wyoming reported using a mov- ing-average linear regression technique. None of the states reported using Box–Jenkins (sometimes called ARIMA or autoregressive integrated moving average) methods, which have become essential tools of business forecasting. (For a more complete discussion of Box–Jenkins methods see the Guidebook.) Other approaches incorporate modeling concepts but stop short of a full-blown statewide model. Kansas reported using OD table estimation from ground counts as a stopgap before developing their own statewide model. New York encourages its MPOs to extend their models into rural areas to achieve a wider coverage and authorizes special models to be built when needed. Other states reported reliance on MPO models where possible. South Dakota has pursued an interesting variation on trend–trend line forecasting that seems to embody principles of behavioral travel forecasting. STATES WITH STATEWIDE MODELING CAPABILITY Unlike MPO models, which are often permanent components of the UTP process and get incremental upgrades, statewide models go through a life cycle. Many of the statewide models are in transition; they are either being developed or redevel- oped from scratch or are being extensively revised. Other models are dormant and one state is considering the possibil- ity of building a model. Table 1 gives an overview of the sta- tus, at the time of data collection for this synthesis, of all states’ CHAPTER TWO SURVEY OF STATEWIDE TRAVEL FORECASTING PRACTICE

modeling capabilities. The District of Columbia and Rhode Is- land do not have statewide models, because they are covered entirely by a single MPO model. This table does not reflect the common practice of states using MPO models for rural travel forecasting, when feasible. Those models shown as being re- vised are already functional, but are either being updated or be- ing given greater capabilities. Figure 1 provides an additional overview of the status as of spring 2005, as well as a rough es- timate of the cost of model development and, in a few cases, the amount of time allowed for model development. 14 RATIONALE FOR STATEWIDE MODELS Dynamics of Modeling Process Responses to the synthesis and Peer Exchange question- naires revealed that the statewide modeling process is dynamic. A generalized process that has been followed by several states is illustrated in Figure 2, which shows that the design of the statewide model is influenced by past experi- ences with the use of the model and by the levels of knowl- edge by both staff and decision makers. A statewide model State Model Condition Cost Development Time (years) Comments Alabama None Alaska None Arizona None Arkansas None California Operational $200,000 2.4 Colorado None $400,000 1 Connecticut Operational Delaware Operational District of Columbia MPO model Florida Operational $1,500,000 4 Georgia Operational $65,000 1 Hawaii None Individual island models Idaho Dormant Illinois Dormant Indiana Operational $1,500,000 3 7 more years for various upgrades Iowa Developing $300,000 2 Kansas Developing Has a dormant freight component Kentucky Operational $370,000 2 New model under development Louisiana Operational $500,000 Cost includes some applications Maine Operational $500,000 5 Being revised Maryland None Massachusetts Revising $800,000 Michigan Operational $1,000,000 2 Minnesota Partial Mississippi Developing Missouri Operational $500,000 Revision completion soon Montana Operational Freight only Nebraska Dormant Base year model Nevada None New Hampshire Revising $2,000,000 New Jersey Operational $500,000 Freight only New Mexico None New York None County-level OD assignment North Carolina None North Dakota None Ohio Operational $6,000,000 8 Being revised; $3,500,000 for data Oklahoma None Oregon Operational Being revised Pennsylvania Developing Rhode Island MPO model South Carolina Operational $25,000 0.5 South Dakota None Feasibility study being conducted Tennessee Developing Based on OD table estimation Texas Operational $1,700,000 4 Utah None Vermont Operational $730,000 2.5 Virginia Operational $1,500,000 3 Washington None West Virginia None Wisconsin Revising $850,000 2.5 Wyoming None Notes: MPO = metropolitan planning organization; OD origin–destination. TABLE 1 STATUS OF STATEWIDE MODELING CAPABILITY, SPRING 2005

15 can flourish or become dormant depending on the amount of positive reinforcement that the process provides. The modeling process is driven by needs in the form of general environmental and planning factors or by the require- ments of a specific project, such as a major new highway cor- ridor study. The process is also influenced by the needed level of spatial detail. These needs lead to the development of goals and objectives for the statewide model. The goals may be ex- plicit or implicit, but are most often created in collaboration with decision makers and other stakeholders. The actual design of the statewide model is dictated by the established goals; the level of funding available for model development; the state of the practice in statewide modeling; the state of the art in travel forecasting, in general; and the availability and quality of secondary data sources. The de- sign of the model is also influenced by the level of expertise of the DOT staff and their consultants. Primary data sources can supplement secondary data sources, but at much greater cost. As staff expertise increases, the model can be upgraded for better accuracy and applications to a greater variety of policies and projects. The most important feedback loop in the modeling process involves five stages, shown counterclockwise in Figure 2: • Goals and objectives, • Model development funding, • Statewide travel forecasting model, • Applications to plans and projects, and • Outreach to decision makers. Successful applications of the model lead to increased awareness and confidence among decision makers, who in turn find additional uses for the model and provide the nec- essary financial support. Models that fail to continuously prove their utility will eventually be discarded. Uses of Models The Guidebook mentioned the need to address certain Inter- modal Surface Transportation Efficiency Act of 1991 (ISTEA) planning factors as a major motivation for the development of statewide travel forecasting models. States with models reported an even broader rationale for model creation consid- ering how the model is applied. Corridor planning (19) ■■■■■■■■■■■■■■■■■■■ Statewide system planning or system environmental impact statement (EIS) (14) ■■■■■■■■■■■■■■ Bypass studies (13) ■■■■■■■■■■■■■ Regional planning, assisting an MPO model (12) ■■■■■■■■■■■■ Project-level traffic forecasts or project EIS (11) ■■■■■■■■■■■ Regional planning, substituting for a local model (9) ■■■■■■■■■ Air quality conformity analysis (6) ■■■■■■ Freight and intermodal planning (6) ■■■■■■ Traffic impact studies (6) ■■■■■■ Economic development studies (6) ■■■■■■ Long-term investment studies (6) ■■■■■■ Detour analysis (5) ■■■■■ Project prioritization (5) ■■■■■ Toll, pricing, or tax studies (5) ■■■■■ Border crossing or port-of-entry studies (4) ■■■■ Inputs to economic modeling (3) ■■■ Intercity bus planning (3) ■■■ Operational Dormant Developing Revising Partial FIGURE 1 Status of statewide travel forecasting models, Spring 2005. Environmental and Planning Factors Desired Spatial Detail Anticipated Projects Goals and Objectives Outreach to Decision Makers Model Development Funding Applications to Plans and Projects State-of-the- Art/Practice Statewide Travel Forecasting Model Secondary Data Sources Primary Data Sources Staff and Consultant Training FIGURE 2 Typical statewide model development process.

16 Land use planning (3) ■■■ Passenger rail planning (3) ■■■ Freight rail planning (2) ■■ Homeland security (2) ■■ Incident management planning (2) ■■ Operational level studies (2) ■■ Work zone planning (2) ■■ Airport planning (1) ■ Weigh station location (1) ■ Revenue forecasting (1) ■ Pavement life studies, equivalent single-axle loads (ESALs) (1) ■ Highway alternatives analysis (1) ■ Transit alternatives analysis (1) ■ Park-n-ride location analysis (1) ■ The list of uses is in order of prevalence. The number fol- lowing an item indicates the number of states reporting that item. States having a long history with models and great con- fidence in model validity tended to report a greater number of uses. The list illustrates the wide variety of applications for a statewide model. None of the states reported using their models for either truck weight studies or for safety analyses. Although most states have found broad uses for their mod- els once created, many models were initiated because of a very specific issue or project. For example Texas needed to analyze North America Free Trade Agreement (NAFTA) trade im- pacts, Louisiana and Wisconsin required input to their 2030 plans, Maine proposed to analyze a toll road, Rhode Island needed data for its air quality conformity analysis, and Indiana and Missouri were both writing an Interstate highway corridor plan. However, most models were started because of a real- ization by staff that there were general forecasting needs to be addressed. California and Ohio identified these needs through a formal set of workshops. In Ohio’s case, the process resulted in a model specification (for both an interim model and a final model) and a program of data collection. Many states found it useful to stage the development of the model, adding capabilities as the budget permitted. Cali- fornia, New Jersey, Ohio, Oregon, and Virginia are examples of states with a deliberate staging process—building a lim- ited model to address immediate needs and expanding upon this model to address a greater range of issues. Examples of Successful Applications of Statewide Models In some locations the use of statewide models is multifac- eted. Information about historical uses was solicited from states having recent, solid experience with applications. Sim- ilar information was also derived from the Peer Exchange questionnaire, where states had an opportunity to expound on the rationale for model investment. Although this section provides example success stories from Ohio, Indiana, Kentucky, Oregon, Florida, and Delaware, similar stories exist in other states. Ohio Since coming on line in 2002, the Ohio Interim Statewide Travel Demand Model has been used for many statewide and project- level analyses. The three most important are described here. The model was used to analyze, verify, and update Ohio’s Macro cor- ridor listing as part of the statewide long-range plan update. Macro corridors are those that receive priority for capacity expansion. The model analysis was able to verify the existing corridors, but also added several important corridors, many with out-of-state connectivity that were missed in the original selec- tion process. The model was also used to estimate truck diver- sion to the Ohio Turnpike, based on a truck speed limit increase and a decrease in truck tolls. This study was particularly sensi- tive because the bond rating of the Turnpike Commission was at stake if the toll decrease resulted in decreased revenues. The model predicted an approximately 20% increase in truck traffic on the Turnpike (enough to offset the toll reduction) and this amount was realized within a year of the changes. Finally, the model was used to estimate the road user benefits (in terms of travel time and vehicle operating cost but excluding crash costs) of the Governor’s Jobs and Progress Plan. This plan envisions $5 billion in major new construction over 10 years. The analysis focused on those projects involving capacity expansion (about half of the program in dollars) and demonstrated an annual user benefit of $390 million per year over the 20-year life of the proj- ects, which was enough to validate the proposal (G. Giaimo, per- sonal communication, 2005). A description of Ohio’s next generation model is found later in this synthesis. Indiana The Indiana DOT (INDOT) model was actually developed for the purpose of corridor-level economic development studies. The model has served as the basis for four corridor studies (I-69, SR-101, SR-37, and US-231). The model is used to produce VMT and vehicle-hours traveled (VHT) output of existing plus committed network and build networks for level-of-service (LOS) deficiency analysis, corridor-level and systemwide eco- nomic analysis (in conjunction with a benefit–cost add-on and an economic simulation model). The model was also used to pro- duce future year growth factors to forecast future traffic volumes in a statewide interchange assessment study. The model output regarding link-level LOS for build and no- build conditions is a factor in project prioritization. INDOT uses the FHWA Highway Economics Requirements System with a 100% database to provide project-level benefit–cost analysis and implementation phasing input to supplement the travel demand model. The statewide travel model provides future year traffic for input into the FHWA Highway Economics Requirements System. The model is used at a systemwide level for safety analysis. The NETBC [Network Benefit Cost] cost–benefit calculator computes accident reduction costs from model output VMT by functional classification and facility type. The statewide travel demand model has been a valuable tool for INDOT in developing our 2030 INDOT Long-Range Plan. The model is used to display existing and future year congestion problems for discussion at a series of INDOT consultation meet- ings with our Metropolitan Planning Organizations and Regional

17 Planning Organization. The model provides information on pro- posed INDOT improvements and has been very valuable in the evaluation of improvements providing bypass alignments in smaller communities not covered by an MPO planning process (S. Smith, personal communication, 2005, and excerpts from the Indiana response to the Statewide Travel Demand Models Peer Exchange 2004). The Indiana passenger component is described later in this synthesis. Kentucky The Kentucky model has been used primarily for corridor studies. • I-66 Corridor Study. The study limits were between the Kentucky–Virginia state line and the Kentucky–Illinois/ Missouri state line. The cost of the study was $1 million and recommendations from the study were to implement (build) portions of a new I-66 corridor. The Kentucky Statewide Model was instrumental in determining traffic volumes and the economic impact of the new corridor. A portion of I-66 is currently under design in Pulaski County. Other sections are being staged for later letting dates. • I-69 Corridor Study. This study stretched between Texas and Michigan. The Kentucky Statewide Model was used to deter- mine the traffic volumes in the state of Kentucky. I-69 will be built as a new facility in many locations (Indiana for instance), and in Kentucky existing four-lane highways will be improved and resigned. • Other Corridors. The Kentucky Statewide Model has been used for many other important corridor studies such as I-64 (widening), I-875 (proposed new Interstate between Berea, Kentucky, and Chattanooga, Tennessee), and I-74 (extension of Indiana Interstate through Kentucky to Maysville). The model is able to give more accurate future growth rates and traffic diversions than other available tools. • The Kentucky Statewide Model was used to optimize the lo- cation of potential commercial vehicle stations (weigh sta- tions) by identifying the number of trucks that would use each location (N.R. Bostrom, personal communication, 2005). The Kentucky passenger component is described later in this synthesis. Oregon Here is list of applications of the Oregon model for which case studies have been prepared. • Willamette Valley Livability Forum. The Forum initiated a comprehensive regional visioning process for the future of land use and transportation in Oregon’s populous Willamette Valley. The first generation of the Transportation and Land Use Model Integration Project (TLUMIP) model was used to model eight scenarios that varied by land use, road and public transit networks, and mileage tax. Results of modeling various combinations of land use, economic, and transportation policy options allowed decision makers to see the effects of each pol- icy and how it will shape the future of the Willamette Valley. • House Bill 3090: Eastern/Central Oregon Freeway. The 1999 Oregon Legislature directed the Oregon DOT (ODOT) to ana- lyze whether a freeway in eastern and central Oregon would off- load increasing traffic in the Willamette Valley. The TLUMIP model was used to evaluate the effectiveness of three alternative alignments to meet this objective. • Newberg–Dundee Bypass Induced Demand. An EIS is being prepared for a proposed highway bypass of two small com- munities between Portland and McMinnville. The TLUMIP model was used to evaluate induced demand potential in rural Yamhill County as a result of the new bypass highway. • Economic and Bridge Options Report. The TLUMIP model was used to examine the impacts of weight limits for vehicles using deteriorating bridges throughout Oregon. This analysis was the basis for a discussion with the 2003 legislature and resulted in a $2.5 billion investment in Oregon’s transportation infrastructure. • Oregon Transportation Plan Update. The ODOT strategic policy document for transportation is undergoing its first update since it was adopted in 1992. The statewide model is being used to help define a reference case and different transportation service and investment scenarios (W. Upton, derived from the Oregon re- sponse to the Statewide Travel Demand Models Peer Exchange Questionnaire, 2004, and personal communication, 2005). Florida The Florida model has a single focus. The Florida Legislature established the Strategic Intermodal System (SIS) in 2003 to enhance Florida’s economic competi- tiveness. The system encompasses transportation facilities of statewide and interregional significance and is focused on the efficient movement of passengers and freight. These facilities include Florida’s major highways, rail facilities, airports, sea- ports, and waterways, as well as the intermodal connectors join- ing the SIS ports and terminals to its corridors. The Florida Statewide Passenger and Freight Model has a highway network that includes all MPO model network links and major rural roadways. SIS highway links are identified in the model. Intermodal terminals, major seaports, and rail yards are included as special generators. The model will be used to ana- lyze and evaluate conditions and performance of passenger and freight transportation under different scenarios, which will lead to the prioritization of proposed projects for SIS planning analy- sis (H. Shen, personal communication, 2005). Delaware Delaware’s model was recently updated and the Delaware DOT (DelDOT) has not yet had extensive experience with it. The updated model was immediately put to use as an integral tool within major studies of two long-standing, critical trans- portation issues. The first effort was the US-113 North–South Study initiated in fall 2004. This involved analysis of projected traffic conditions and evaluation of more than 20 alternatives within a 50-mi-long corridor. The DelDOT Statewide Model was used to examine “average annual conditions.” Because the corridor is significantly impacted by beach resort-oriented travel patterns at least 6 months of the year, the model was expanded to include equations and models focusing on “average summer conditions” and “peak weekend conditions.” Development of the peak season models included postcard mailback surveys and other analyses to refine peak trip rates, OD patterns, develop a “day tripper” table, and re- view summer assignment patterns on “beach routes.” The second effort was the US-301 Environmental Impact Study initiated in spring 2005. This is a location and preliminary

18 design study for a 15-mi-long study area projected to double its population and triple its employment by 2030. The DelDOT model was used to examine approximately 10 alignment options with various access scenarios and was used to assess travel im- pacts for a number of toll rate possibilities (M. DuRoss, personal communication, 2005). GOALS AND OBJECTIVES States have found it important to define goals and objectives for their models before embarking on initial model develop- ment or performing major updates. Two examples are pro- vided here. Oregon Oregon set forth three goals and seven objectives for its sec- ond generation model development. • Goal #1: Develop a set of integrated land use and transporta- tion models that will enable ODOT and the MPOs to do [the] analysis needed to support land use and transportation deci- sion making. • Goal #2: Develop and maintain databases needed to make pe- riodic long-term economic, demographic, passenger, and com- modity flow forecasts for statewide and substate regions. • Goal #3: Develop the expertise, guidelines, and institutional support necessary to sustain the models and databases needed for integrated land use and transportation facility analysis. • Objective #1: Provide training on the integrated transportation and land use models. • Objective #2: Connect the statewide and substate models with the metropolitan area models. • Objective #3: Transfer the statewide and substate model to a platform that is extensible and can be modified by ODOT in the future. • Objective #4: Integrate rail transportation into the statewide and substate model. • Objective #5: Develop a working metropolitan model that in- tegrates transportation and land use components. • Objective #6: Establish data linkages between the statewide, substate, and metropolitan models and analytical software for assessing highway system performance. • Objective #7: Establish university research linkages. Wisconsin Wisconsin listed six practical objectives for its model. • Having the capability to analyze modal diversion impacts along major backbone and connecting corridors. • Having the capability to analyze route diversion impacts once corridor-level improvements are made, such as adding lanes and changing design from expressway to freeway, thus in- creasing the operating speed and lowering the travel time. • Analyzing the capacity (LOS) and safety impacts associated with increased truck travel on key Wisconsin interstates owing to the introduction of major new intermodal facilities such as Rochelle in north–central Illinois and with the ever-expanding regional commercial distribution centers like Wal-Mart, Lowe’s, etc. • Developing a planning and modeling process that integrates the on-going development of fourteen (14) MPO models and two (2) urban area models with our statewide model. • The statewide model has two components: a passenger model and a freight model. • To conduct AQ [air quality] regional emissions and conformity analysis for rural, isolated counties that do not have a MPO LRTP [Long-Range Transportation Plan] and TIP [Transporta- tion Improvement Program]. INSTITUTIONAL ARRANGEMENTS The development and maintenance of a statewide model is a ma- jor effort, involving costs for data, consultants, and in-house staff. Overall Costs There are no commonly accepted standards of statewide model design. Ideally, statewide models would be as detailed and accurate as the best of our urban models. This goal can be achieved in some smaller states (e.g., New Jersey or Rhode Island) either by expanding an urban model to encompass the whole state or by stitching together all of their urban models. For most states, however, compromises on quality must be made to stay within cost and time constraints. Some states find it difficult to estimate the full cost of their models be- cause the development occurs over a long period of time or because the costs of certain related activities cannot be wholly attributable to the model development process. Given this caveat, there is an extremely wide range of costs. At the low end of the scale, South Carolina paid just $25,000; whereas at the upper end of the scale, Ohio paid $8 million, of which $5.5 million covered the cost of data collection. Approximately half of Ohio’s data collection costs were for data that could be shared with MPOs. Both the Ohio and South Carolina mod- els would be classified as being unconventional. (See chapter three for a more complete discussion of the Ohio model.) For more conventional modeling approaches, costs range be- tween approximately $300,000 in less populated states (e.g., Delaware and Iowa) to approximately $1.5 million in popu- lous states (e.g., Florida and Texas). Most states were able to pay for their models exclusively with State Planning and Research funds, although a few states needed supplementary funds from either general pur- pose revenues or transportation-dedicated revenues. Other revenue sources were rare. Maine received funds from a toll road authority and New Hampshire used Congestion Mitiga- tion and Air Quality funds. Data Costs Data collection can be a large component of the development of a statewide model. For example, Ohio’s devoted almost 70% of its budget to data acquisition. Big ticket data items in Oregon included: • Continuous Survey for Modeling in Oregon Pilot Project ($250,000). • Freight commodity flow data collection ($390,000). • Freight shipper and carrier survey ($300,000).

19 • Truck intercept survey ($175,000). • Oregon Travel Behavior Survey ($125,000). • Recreation/Tourism Activity Survey ($150,000). • Household Activity and Travel Survey ($1,000,000). Wisconsin paid $2.5 million for an NHTS add-on; how- ever, the data are also usable by MPOs within the state. Ken- tucky paid just $176,000 for their NHTS add-on. Louisiana spent $100,000 on commodity flow data, which is typical. In- diana spent $60,000 on D&B employment data. Some other states reported negligible data acquisition costs. Staffing and Maintenance All states reported having the help of consultants when build- ing their models. In some cases teams of consulting firms contributed to model development. The dependence on con- sultants for maintenance varied considerably across states; however, most states reported that routine maintenance was done in-house. As with costs, staffing levels varied widely across states. Staffing levels ranged from a one-half full-time equivalent (FTE) in Florida, Indiana, and Kentucky to approximately three FTEs in Connecticut, Oregon, and Wisconsin. A little more than half of the states reported roughly one FTE. A few states noted that modeling responsibilities were spread across multiple staff members, each spending only a fraction amount of their time on the project. Some of the states with lower staffing levels reported having a larger amount of con- sultant help. Several states needed to add personnel as they increased their modeling activities. Model maintenance is required to keep it up to date in terms of network structure, demographic data, link data, and calibration data. Models not maintained become obsolete and useless. However, maintenance should not be so burdensome that staff does not have sufficient time for applications. States with new models find that there is little need for maintenance, but states with mature models experience a more constant ef- fort. A 50/50 split between maintenance and applications is typical among those states that were able to make an estimate. Time Frame Because situations vary significantly across states, there is no consensus as to how long it takes to build a model. Models in most states have evolved over many years; therefore, no time estimate is possible. A reasonable range for states that recently built their models from scratch is 1 year (Delaware) to 4 years (Florida and Texas). Ohio, with an unusually am- bitious model, is taking 8 years (see chapter three). Maintenance is largely a continuous process or on a very frequent cycle (1 to 2 years). Update cycles tend to coincide with statewide plan updates, with most states using a 5-year update cycle. Two states (Connecticut and Ohio) indicated using a 10-year cycle for major model revisions. Massachu- setts noted that its next update would likely be driven by air quality conformity needs, whereas Indiana performs updates as needed for specific projects. User Support Training is considered an essential element of model deploy- ment. A training session is often provided by the consultant on model delivery; thereafter, training happens sporadically. Only a few states have regular training cycles, with details differing from state to state. For example, Oregon has an arrangement with a local university to supply training, Connecticut sends employees to FHWA urban modeling courses, and Kentucky has an in-house annual training program lasting 2 days. Users of the model tended to be confined to the state DOT and its consultants. States with organized urban model user groups (e.g., Florida and Iowa) can call on them for assis- tance with the statewide model, even though their members are not primary users of the model. Web pages tended to be located in those states with user groups. A little more than half of the states with models have made provisions for distributing them to consultants and MPOs. Se- lected states will deliver their models to outside agencies or universities on request, although with conditions. For exam- ple, Texas asks borrowers to sign a confidentiality agreement, Kentucky requests that borrowers sign an agreement as to ac- ceptable use of the model, Wisconsin has procedures by which it will allow the use of its model and the modeling software by outside parties, and Michigan only distributes trip tables and networks. The remaining states have had no experience with model distribution, and it is not clear whether these states have a policy against or simply no need for distribution. All states with models will make results of model runs available on request. Requests are handled on a case-by-case basis. Some states will do custom model runs on request; however, those requests are fulfilled only for internal needs. Often the format of the requested data must be negotiated. For example, Vermont asks outside recipients of model re- sults to sign a binding nondisclosure agreement to protect sensitive employment information. INSTITUTIONAL BARRIERS TO MODELS Given that only about half of the states have active models, there would appear to be reluctance on the part of some states to proceed with model development. Agency Roadblocks Only a few states reported having institutional barriers that needed to be overcome, and these barriers were not critical.

Massachusetts and Ohio each had trouble obtaining employ- ment data from another state agency. A few states reported funding shortfalls until the need for the model could be con- vincingly demonstrated. Wisconsin found trouble getting good cooperation from the state’s two largest MPOs and needed to deal with a change in governor, who required time to under- stand issues related to the statewide transportation plan. Overcoming Resistance Literature on innovation often makes reference to the need for a “champion” to effect change. I’ve learned that in every state where models are maintained and actively used in the planning process that there is an evangelist and visionary that drives the program. This person, by force of person- ality or position, is the key driver behind the success of the model. If this person retires or moves on to other things the modeling pro- gram often dies. Thus, maintenance of the model is often more a reflection of the priorities and capabilities of the evangelist more than a systematic or carefully considered process (R. Donnelly, Statewide Travel Demand Models Peer Exchange, 2004). Cooperation is critical to an effective statewide travel modeling process. It is important to build and maintain relationships between tech- nical staff and management. Well-established relationships be- tween modeling staff and senior management make management more willing to take a chance on a process that does not support their initial preconceived ideas. Interest and support of model- ing from ‘outsiders’ is helpful. Those who used the modeling tools in the past support and advocate for its use on new projects. It is helpful to have advocacy from others external to the process that are perceived as nonbiased and those that may better under- stand non-traditional model outputs (W. Upton, Statewide Travel Demand Models Peer Exchange, 2004). Model Failure A number of statewide models have gone dormant. One model developer writes: Models fail for one of several reasons: • Vague or poorly defined goals and objectives. • Developed with single purpose in mind. • Higher than expected maintenance and application costs. This in- cludes the need for more highly skilled staff, the magnitude of data required (both in scale and scope), and inter-agency friction. • Lack of management support (read: the models do not provide information useful to decision makers in the metrics and time frames they need). • The models are cumbersome and inaccurate. The models are no better than the quality and quantity of the data used to de- velop them. Poor models are the only possible outcome from building them with poor or scarce data. • Failure to build linkages to economic models. Most state leg- islatures tend to look at transportation problems as economic problems. Models that simply address traffic flows do not pro- vide the information on key linkages (and benefits) between the economy and transportation. In some instances I’ve seen state legislators discount modeled outcomes because they are at odds with, insensitive to, or seem uninformed by economic and market trends (R. Donnelly, Statewide Travel Demand Models Peer Exchange, 2004). 20 COMPUTER HARDWARE AND SOFTWARE Questionnaire Results All states reported using a high-speed personal computer to run their existing models; typically running a version of the Windows operating system. No other hardware requirements were noted. A large majority of statewide models are built on software platforms originally designed for UTP. Oregon has constructed its own software specifically for statewide modeling. Most states use a GIS with their models; either a stand-alone GIS package or one built into their UTP software. Computation times vary considerably, ranging from only 30 s in South Car- olina to 12 h in Maine. The median computation time is some- where between 1 and 2 h; therefore, it is possible to conclude that the computational burden is not large. Example of the Use of Geographic Information Systems In Louisiana, the statewide model network was developed based on several existing DOT legacy databases including: • Louisiana Road GIS file in Geomedia format; • Surface Type log file, a Microsoft Access database containing mile post and key roadway attributes; and • Highway Needs Inventory Summary log file, another Mi- crosoft Access database, containing mile post and additional roadway attributes, roadway conditions, and future needs information. Substantial resources were devoted by the model development consultant Wilbur Smith Associates (WSA) to make sure these files were rendered suitable for modeling purposes and were lin- ear referenced, facilitating future network update activities. • WSA first converted the original Geomedia Road GIS file to ArcInfo, created a Route System, and used the dynamic seg- mentation method to link the Surface Type log and Needs In- ventory file to the GIS file. This process allowed WSA to access all the necessary network attributes from the two Microsoft Ac- cess databases. WSA decided to retain all of the links, including some local roads in the original GIS file for the Micro Model network. • The Road GIS file was designed for the Louisiana Department of Transportation and Development (LADOTD) for nonmod- eling functions and therefore was not suitable for modeling. The original GIS file did not represent a modeling network of links and intersections. Network editing to split links was nec- essary to represent intersections properly. Because some of these intersections could be overpasses or underpasses, each required review so that network connections would replicate ground conditions. • Many stub links were found in the original GIS file. Network connectivity checks and editing were performed to make sure the network was suitable for modeling. • Additional roads were added, particularly within MPO areas. The sources for the additions were from the MPOs’ modeling network file or Census Topologically Integrated Geographic En- coding and Referencing (TIGER) line files. Toll roads, bridges, and automobile ferry links were identified and added to the Mi- cro network, because they were not present in the state database. • With substantial manual editing and link additions, the existing GIS file mile post information was either missing or distorted.

21 WSA developed automatic procedures (TransCAD) and Ac- cess database macros, allowing for the update of attribute in- formation from LADOTD Summary log file or other files as long as these files contain beginning and ending mile posts for the updated attributes. With these macros, the updating net- work attributes become a simple task. For example, near the end of the model development work, LADOTD systematically reclassified their functional classification system, and WSA was able to incorporate this latest information easily for the fi- nal model revalidations. In summary, given the scale and extent of the statewide model network coverage, the ability to link these DOT existing, well-established attribute databases to the modeling GIS net- work becomes increasingly important once the statewide model is developed. It eliminates duplicate efforts, reduces network coding errors, and increases job satisfaction by eliminating te- dious manual work and increasing fast turnaround time in con- ducting alternative analysis, corridor studies, scenario planning, and other statewide planning activities (S. Yoder, personal com- munication, 2005, and Wilbur Smith Associates 2004). OVERALL MODEL CONSIDERATIONS All states with operational models have used them for long- range forecasting purposes. With the exception of Vermont, forecasts of 20 or more years have been done. Measures of Effectiveness A state selects measures of effectiveness (MOEs) that relate closely to the rationale of the model. MOEs are usually ag- gregations of results that would pertain to individual links (e.g., road segments) or nodes (e.g., intersections) and are aids to deciding between alternatives. MOEs are relied on during the decision-making process because people are able to readily grasp only a few indicators of system performance, and aggregate measures have a lower percentage of error than raw travel forecast outputs. The following is a complete list of MOEs used by states in order of prevalence. VMT (22) ■■■■■■■■■■■■■■■■■■■■■■ VHT (20) ■■■■■■■■■■■■■■■■■■■■ Volume and capacity ratios (18) ■■■■■■■■■■■■■■■■■■ Levels of congestion (15) ■■■■■■■■■■■■■■■ Traffic growth rates (14) ■■■■■■■■■■■■■■ System delay (11) ■■■■■■■■■■■ Passenger volumes by mode (9) ■■■■■■■■■ Corridor delay (9) ■■■■■■■■■ Employment by area (8) ■■■■■■■■ Time savings (8) ■■■■■■■■ Freight tonnages by mode (6) ■■■■■■ Air pollution emissions (3) ■■■ Crash reduction (2) ■■ Greenhouse gas emissions (2) ■■ Benefit–cost ratio (2) ■■ Goods production by area (2) ■■ Interregional travel (1) ■ Land prices (1) ■ Shipping costs (1) ■ Total trips by area (1) ■ MOEs are similar to those found in urban models. Among the seemingly obvious MOEs not mentioned were energy consumption and any user benefits other than time savings. States will often disaggregate MOEs by time of day or by lo- cation to better identify problems. For example, Massachu- setts looks at congestion measures by time of day. Ohio breaks down its MOEs by Ohio DOT district and by county. Oregon computes various MOEs depending on the issue, such as VMT by travel market segment, VHT by travel mar- ket segment, shipping costs by area, total production by area, employment by area, land prices by market segment and area, and trips by travel market segment. Montana’s HEAT included measures of accessibility, business activity within cities or markets, production costs, and personal income. Three states indicated that they had no MOEs. Employment Data Two particularly difficult aspects of travel forecasting are obtaining good TAZ level employment data and good long- range economic forecasts. Employment data from govern- mental sources are often restricted by confidentiality issues and incorrect street addresses. Many states have opted to ob- tain their employment data from commercial sources. Here are the primary sources of employment data reported by states, in order of prevalence. A state may have used more than one source. CTPP (10) ■■■■■■■■■■ MPO databases (10) ■■■■■■■■■■ Commercial data vendor (10) ■■■■■■■■■■ Department of Workforce/Employment/Labor Development (6) ■■■■■■ Workman’s compensation tax records (5) ■■■■■ Unemployment records (4) ■■■■ Employer or establishment survey (2) ■■ Regional economic model (2) ■■ Employer directory (1) ■ Other unspecified (1) ■ Many states have taken advantage of MPO models for employment data. Although the same data problems also ex- ist at an MPO, usually an individual(s) with good local knowledge has already confronted them. Ten states use the CTPP, which derives employee location from a large sample of households during the decennial census. The CTPP is es- pecially attractive because of its low marginal cost. Although unemployment records (ES-202) seem to be an attractive source of data, some states have reported considerable prob- lems in obtaining and using this database. Economic forecasts are done regularly by the BEA; how- ever, the BEA regions are usually too large for direct inclusion

into a statewide model. Therefore, many states have opted for other sources of economic forecasts. The following is a list of the sources, in order of prevalence. State agency forecast (8) ■■■■■■■■ A regional economic model (5) ■■■■■ An IO model (4) ■■■■ MPO databases (4) ■■■■ BEA (4) ■■■■ Commercial forecast vendor (3) ■■■ State DOT (2) ■■ University (1) ■ None or not mentioned (3) ■■■ The largest number of states obtain their economic fore- casts from another state agency. Five states use a regional economic model, either a commercial model or one devel- oped particularly for the statewide model, and three states use an IO model. As with employment data, a few states ef- ficiently rely on their MPOs for economic forecasts. Generic Model Structures There is an intrinsic relationship between model structure and the policies and projects that can be addressed, as illustrated in Figure 3. This figure can be interpreted either backwards or for- wards, depending on whether it is illustrating part of the model design process or model operation. The structure of the model dictates what it can reasonably produce as outputs, and the out- puts limit what can be accomplished by the model when adding information to the decision process. However, the design of the model is largely derived from what the model is intended to ac- complish. Conceptually this is a two-stage process, where the issues needing to be addressed dictate the required model out- puts, which further dictate the model structure. Figure 4 expands the relationship between the first two items in Figure 3, the generic structure and the range of model outputs. There are essentially six generic structures, ranging from statistical trend analysis to an integrated model of freight, passenger travel, and economic activity. The behavioral real- ism generally increases from left to right, except for the dis- tinction between freight-only and passenger-only models. Al- though freight-only models may be as sophisticated as passenger-only models, their use for traffic operational analy- sis is limited. The arrow in Figure 4, deliberately drawn to be 22 vague, shows how the behavioral realism of the model affects the uses to which the model can be put. PASSENGER COMPONENTS Statewide travel forecasting models are often thought of hav- ing two equally complex components: passenger and freight. In some models, vehicles in commercial service that do not carry freight are treated separately. With a few exceptions, passenger components look much like urban travel forecast- ing models in structure; containing the four major steps of trip generation, trip distribution, model spit, and trip assign- ment. Oregon’s model and Ohio’s new model (see chapter three) have more complex structures, but the four traditional steps are still present, conceptually. This section deals pri- marily with details of how the four steps are implemented. Passenger Component Data States use a wide variety of data sources to calibrate their statewide models, although only a few sources are used in each state. The following are the data sources identified by states in order of prevalence. CTTP (13) ■■■■■■■■■■■■■ Census journey-to-work data (11) ■■■■■■■■■■■ NCHRP Report 365 (10) ■■■■■■■■■■ NHTS normal sample (10) ■■■■■■■■■■ MPO household survey(s) or panel(s) (9) ■■■■■■■■■ ATS (9) ■■■■■■■■■ Own household survey (8) ■■■■■■■■ Institute of Transportation Engineers (ITE) Trip Generation (7) ■■■■■■■ PUMS (7) ■■■■■■■ Roadside survey(s) (6) ■■■■■■ NHTS add-on (5) ■■■■■ NCHRP Report 187 (5) ■■■■■ GPS-based survey (3) ■■■ Amtrak (2) ■■ Intercity bus service (2) ■■ FAA sample ticket data (2) ■■ Ferry service (1) ■ Tourism survey (1) ■ Own on-board rail survey(s) (1) ■ Bus on–off counts (1) ■ Other agency survey (1) ■ The extensiveness of this list of calibration data sources indicates that modelers are being quite resourceful; using what is readily available and augmenting as necessary. With home interview surveys costing approximately $165 per sample (based on the cost of an NHTS add-on), there is a strong advantage to exploiting whatever data have already been collected. The most often cited data source was the CTPP. A total of 11 states either did their own household Design Operation Generic Structure Range of Model Outputs Issues to Be Addressed FIGURE 3 Relationship between model structure and policies and project decision making.

23 survey or funded an NHTS add-on. Connecticut reported us- ing both a household survey and an NHTS add-on, although the survey dated back to the 1970s. California did the most extensive home interview survey of their own with 17,000 samples. Seven states tapped into MPO household surveys. Ten states used either NCHRP Report 187 (Sosslau et al. 1978) or NCHRP Report 365 (Martin and McGuckin 1998) for transferable parameters. Maine transferred parameters from the Michigan model. Although the ATS is now 10 years old, many states believe that it is still essential. The ATS is the only comprehensive data source on long distance travel. The NHTS add-ons varied greatly in size. Wisconsin bought the largest number of samples (17,610) and Massa- chusetts bought the fewest (500). Interestingly, Oregon did not use the NHTS, but performed four different surveys in support of its statewide models, as well as MPO models: Household Activity and Travel Survey, Oregon Travel Behavior Survey, Recreation/Tourism Activity Survey, and Continuous Oregon Survey for Oregon Models. Household socioeconomic data came from a few obvious sources as listed here. The U.S. Census dominated as a data source, followed by MPO databases, which most likely were derived largely from the U.S. Census. Five states obtained employment data from another state agency, although there were often considerable problems using such data. Other U.S. Census than CTPP (15) ■■■■■■■■■■■■■■■ CTTP (12) ■■■■■■■■■■■■ MPO databases (10) ■■■■■■■■■■ Another state agency (5) ■■■■■ A regional economic model (4) ■■■■ Commercial data vendor (4) ■■■■ School enrollment data (3) ■■■ A state natural resources department (2) ■■ Local property tax records (1) ■ GIS maintained by another agency within your state (1) ■ There were only a few sources of highway traffic data, with most states relying on their own counts or their own Highway Performance Monitoring System (HPMS) data- base. Only six states used either their own speeds or travel times. Massachusetts was the only state reporting that it ob- tained counts from other states. Own agency counts (18) ■■■■■■■■■■■■■■■■■■ HPMS (11) ■■■■■■■■■■■ Own agency speeds (6) ■■■■■■ Own agency travel times (5) ■■■■■ Toll or bridge authority counts (5) ■■■■■ Counts, speeds, or travel times from another agency (2) ■■ Other states (1) ■ Building passenger networks is an expensive and time- consuming task. Data that would allow the construction of statewide passenger networks (links and nodes) came mostly through MPO networks or through DOT road inventory sys- tems. The National Highway Planning Network (NHPN) was used principally for out-of-state portions of the network. Delaware and Rhode Island asked neighboring states for net- work data. For out-of-state highway networks, Florida and Trend Analysis OD Table Estimation & Assignment Freight Only PassengerOnly Combined Passenger & Freight Integrated Passenger, Freight & Economic Activity Individual Link ADT Freight or Passenger Volumes Across State Inputs to Traffic Operational Analysis Details of Freight and Passenger Volumes Transport Effect on Economic Development Generic Structures G en er al iz ed O ut pu ts FIGURE 4 Generic model structures and their potential outputs.

Texas and used a proprietary data source that came from their UTP model software vendor. Own agency road inventory/management system (17) ■■■■■■■■■■■■■■■■■ NHPN (7) ■■■■■■■ MPO networks (6) ■■■■■■ TIGER (6) ■■■■■■ Bus or rail published information (3) ■■■ Neighboring state agency road inventory(ies) or management system(s) (2) ■■ More than half of the states reported the need to obtain lo- cally collected data for their modeling efforts. Several states performed travel surveys, as noted previously. Texas per- formed a border survey. California, Delaware, and Ohio per- formed roadside surveys. The costs of the surveys varied considerably, from $2,000 in Virginia to more than $2 mil- lion in Michigan, Ohio, and Wisconsin. Only Delaware, Indiana, and Oregon reported ongoing data collection efforts to support or update their models. Ore- gon is conducting its Continuous Survey for Modeling in Oregon (COSMO), which collects additional time–series in- formation on household activities and travel. The update cycles for passenger networks tend to be long, with most states reporting that they wait more than a year be- tween updates. Networks are usually updated with DOT road inventory or MPO data. There was no consensus about data deficiencies. Table 2 lists data items modelers wanted but could not get or data items needing improvements. Long distance and tourism data appeared to be a need in states that would prefer to avoid the dated ATS (from 1995). Given the size of the databases necessary for statewide travel forecasting, a large majority of states with models are using GIS for storing passenger data or networks. Those states with a GIS integrated into their UTP software tended to use it instead of a stand-alone GIS product. Furthermore, a large majority of states obtained at least some of their net- work data from a GIS. 24 Passenger Level of Detail Statewide models are distinguished from urban models pri- marily in their spatial extent and their level of detail. Histor- ically, most statewide models were designed at a “sketch planning” level of detail so as to cover more area with fewer network elements. However, recent advances in computer hardware and GIS software have permitted much more detail in statewide models. The amount of detail relates to the num- bers of zones, network elements (nodes and links), trip pur- poses, special generators, time of day, and modes. MODAL CHOICE All states with passenger components have at least the pas- senger automobile as a mode. A majority of the statewide models are multimodal. Listed here are the modes cited by states. The only passenger mode sometimes seen in urban models that has been universally omitted from statewide models is the taxi, although statewide models are likely to contain some treatment of intercity modes, particularly pas- senger rail and passenger aviation. With very few exceptions, each mode in a statewide model has its own network. Passenger automobile (21) ■■■■■■■■■■■■■■■■■■■■■ Intercity passenger rail (conventional) (7) ■■■■■■■ Intercity bus (6) ■■■■■■ Local bus (6) ■■■■■■ Commuter rail (5) ■■■■■ Intercity passenger rail (high speed) (2) ■■ Passenger aviation (2) ■■ Metro rail or light rail (2) ■■ Ferry (1) ■ Intercity rail/bus (1) ■ Commuter express bus (1) ■ Time of Day Time of day is much coarser in statewide models than is typi- cal of urban models. Only five states reported the ability to run peak-hour analyses. The other states run their models for a full 24 h, either during a weekday, a summer day, or an average State Deficiency Ohio There was an underreporting of short and discretionary trips in survey data Massachusetts Lack of household trip data Michigan OD data are impossible to collect on major highways Indiana No external automobile trips from national sources Maine Nonwork origins and destinations; long distance travel patterns California Multimodal long distance or multiday trips Kentucky Up-to-date trip information; not enough samples from NHTS Virginia Long distance travel Florida Rural travel behavior characteristics and tourism trip OD data Notes: OD = origin–destination; NHTS = National Household Travel Survey. TABLE 2 REPORTED DATA DEFICIENCIES

25 annual day. The coarseness of time-of-day representations has implications for being able to calculate accurate delays and for identifying congestion hot spots. The critical issue in time of day, as identified in the Guidebook, is that many trips statewide are longer in duration than a peak hour or a short peak period, but more than likely shorter than a day. The stan- dard methods of overcoming this trip duration problem are dy- namic traffic assignment or traffic microsimulation. Traffic microsimulation was being explored as a possibility in Ohio and Oregon, but consideration of dynamic traffic assignment has not been reported for any statewide model. Zone Systems All statewide models have zone systems for organizing spa- tial information on a network. Zone size varies greatly among states, with the largest zones being counties and the smallest zones corresponding to MPO TAZs. Many states that are re- vising their models do not have a good idea as to the number of TAZs, so data on this issue are incomplete. The relation- ship between the number of TAZs and the size of the state is shown in Figure 5 for those states reporting. A quick inspec- tion of the graph indicates that the relationship between land area and zone size is, at best, not significant. However, if one were to separate out the five states with more than 3,000 zones but fewer than 70,000 square miles (Florida, Indiana, Ken- tucky, Massachusetts, and Ohio, in the upper left corner of the graph) two linear relationships emerge. The differences between the two sets of states appear to be related to model- ing philosophy, rather than to any intrinsic characteristic of the state itself. Oregon, which did not provide an exact esti- mate of the number of zones, would fall into this upper group. Virginia, although appearing in the lower group, uses sub- zones for various model steps and is able to achieve consid- erable spatial detail in this manner. The lowest point on the graph is Georgia, which uses counties for zones. Some states extend their zone systems beyond their bor- ders, but others do not. Kentucky has the most aggressive statewide model in this regard, having 1,109 zones, almost one-third of its total in neighboring states. Other states with a large number of zones outside their borders include Virginia (522), Louisiana (465), Maine (463), Massachusetts (431), and Rhode Island (400). Other states have signifi- cantly fewer out-of-state zones and are dependent on exter- nal stations to account for trips with at least one end outside their borders. The zone structures within the urbanized areas of the states were constructed using a variety of data sets. Most of the states borrowed MPO zones or aggregated MPO zones. The following are the methods used. Aggregations of MPO zones (11) ■■■■■■■■■■■ Adopted MPO zone structures (6) ■■■■■■ Census tracts or aggregations of census tracts (6) ■■■■■■ Census block groups or aggregations of block groups (3) ■■■ Census TAZ-UP (Update Program) (1) ■ Counties (1) ■ Six states (Iowa, Kentucky, Louisiana, Michigan, Ohio, and Virginia) reported that their internal zones systems cov- ered most or all of the United States. However, only two states (Maine and Michigan) mentioned having all or part of Canada or Mexico. Subzones or grid cells are a means of greatly increasing spatial detail in certain model steps, particularly traffic assign- ment, and are used by Kentucky, Ohio, Oregon, and Virginia. External stations are used in urban models to represent origins and destinations of trips that, at some point, leave the study area. Most statewide models do the same. States whose model zone systems fully or mostly encompass the United State do not have a need for external stations. In practice, external stations are placed just outside the study area along Interstate and major U.S. highways; therefore, their number is closely tied to the number of major roads entering or leav- ing the state. For example, Maine had just 20 external sta- tions and Texas had 142. Special Generators A special generator is a network element, often similar to a zonal centroid that represents a single site. A special genera- tor may be shown as its own node on the network or it may share a centroid with other land uses from the TAZs. Poten- tially, each special generator can have its own trip generation rates. Most states use special generators sparingly or not at all. However, two states, Michigan and Texas, use them ex- tensively, with each having nearly 4,000 special generators. Only Virginia has a specified minimum size threshold for special generators. The following is a list of types of special 6000 5000 4000 3000 2000 1000 0 0 50,000 100,000 150,000 200,000 250,000 300,000 N um be r o f T ra ffi c An al ys is Z on es Land Area, Square Miles OH IN FL TX KY MA GA FIGURE 5 Relationship between land area and number of zones in statewide models.

generators cited by states (not the number of such special generators). Tourist attractions (8) ■■■■■■■■ Major recreation sites (6) ■■■■■■ Universities (5) ■■■■■ Military bases (5) ■■■■■ Airports (5) ■■■■■ Shopping centers (4) ■■■■ Hotels (2) ■■ Hospitals (2) ■■ Public offices (1) ■ Bus terminals (1) ■ Michigan and Texas, as would be expected, cited the most types of special generators. The methods of determining trip generation rates for each special generator were split between dependence on ITE’s Trip Generation (1997) and locally determined rates. Here are the methods used. Counts, growth factors, or trends from actual trip making at sites (6) ■■■■■■ Trip rates from ITE’s Trip Generation (6) ■■■■■■ Trip rates from local trip generation studies (3) ■■■ Rates from MPO models (1) ■ California had access to a park attendance database. New Hampshire differed from all other states by using a multino- mial logit expression for tour formations that involved spe- cial generators. Trip Purposes Statewide models tend to have a long list of trip purposes to capture both urban trips and long distance trips. To keep models reasonably simple, the urban trip purposes are often limited to those of NCHRP Report 187: home-based work, home-based nonwork, and nonhome-based. These urban trip purposes are then supplemented with a few purposes that de- scribe long distance trips. Here are the trip purposes in statewide models, in order of prevalence. Home-based work (19) ■■■■■■■■■■■■■■■■■■■ Home-based nonwork (home-based other) (16) ■■■■■■■■■■■■■■■■ Non-home based (16) ■■■■■■■■■■■■■■■■ Long distance recreation/vacation (10) ■■■■■■■■■■ Long distance commute (7) ■■■■■■■ Long distance business (7) ■■■■■■■ Long distance other (7) ■■■■■■■ Home—shop (5) ■■■■■ Long distance personal business (3) ■■■ Home—recreation (3) ■■■ Home—other (3) ■■■ Home—social/recreation (3) ■■■ 26 Home—school (3) ■■■ Other—work (1) ■ Other—recreation (1) ■ Other—other (California) (1) ■ General (Georgia) (1) ■ Long distance, general (1) ■ Other (1) ■ Maine has separate trip purposes for both short and long distance trips for home-based social/recreation. Oregon seg- ments its trip purposes by income. Some of the newer statewide models contain very detailed networks, which are a consequence of incorporating most or all of the urban networks. Florida and Texas have approxi- mately 100,000 links and Wisconsin and Virginia each have approximately 200,000 links. Some states have found it pos- sible to work with smaller networks. For example, Delaware, New Hampshire, and Vermont have fewer than 7,000 links. Passenger Component Methods For the most part, statewide models have passenger compo- nents that are similar to those found in urban models. Mod- els for large urban areas are traditionally four-step, encom- passing trip generation, trip distribution, mode split, and traffic assignment. Many smaller urban models are three- step, replacing the mode split step with small downward ad- justments to trip generation rates. Beyond these four steps, specific procedures must be introduced to handle the distrib- ution of traffic across times of day and to calculate the aver- age numbers of persons in a vehicle (termed automobile oc- cupancy). The new models in Ohio and Oregon (see chapter three) deviate substantially from the norm, so it is difficult to classify their attributes in conjunction with traditional four- step models. A solid majority of the statewide models are traditional four-step. The models in Kentucky, Maine, and Massachusetts are better classified as three-step, because they omit a formal treatment of mode split. Massachusetts handles the large tran- sit ridership in Boston by removing riders at the trip genera- tion step, based on information obtained from the Boston MPO model. Ohio and Oregon have integrated land use and eco- nomic activity components, which encompass the functional- ity of trip generation, trip distribution, and mode split. Ohio implemented OD table estimation from traffic counts within its interim model. Montana uses OD table es- timation from traffic counts to provide background traffic for its economic model, HEAT. The calculation of trip productions during the trip genera- tion step is for the most part performed by a cross-classification procedure. Exceptions include the new Ohio and Oregon models (as discussed earlier), New Hampshire, and Virginia.

27 New Hampshire relied on its tour-based multinomial logit expression for trip productions, and Virginia factored data obtained from the 1995 ATS, the U.S. Census, and the NHTS. Although Connecticut, Indiana, and Vermont used cross- classification for some trip purposes, they also used trip rates or linear equations. Table 3 shows the variables within cross-classification models for trip productions for those states that provided the information. Most models combine household size (persons per household) with some measure of wealth (income, num- ber of workers, or automobile availability). Trip attraction calculations are dominated by the use of linear equations of demographic variables or trip rates. New Hampshire, Ohio, and Oregon are exceptions because they use destination choice models. California and Kentucky both reported referencing NCHRP Report 365 for trip rates. Automobile occupancy calculations convert passengers to automobiles and usually follow the standard urban practice of dividing numbers of passengers by an automobile occu- pancy rate that varies by trip purpose. Here are the methods adopted by states: Automobile occupancy values for each trip purpose (10) ■■■■■■■■■■ Rates that vary with trip distances (2) ■■ Multinomial logit mode split model that includes drive alone, high-occupancy vehicle 2, and high- occupancy vehicle 3 (1) ■ Rates that vary by metropolitan statistical area (MSA) size and Claritas Code (1) ■ None, generation is in vehicles already (1) ■ Rates that vary by vehicle ownership by TAZ (1) ■ Microscopic activity patterns; occupancy is based on the individual travel decision (1) ■ No state reported using a single automobile occupancy rate for all purposes or using automobile occupancy rates that vary by trip duration. The gravity expression remains popular as a method for trip distribution. Three states (California, Florida, and Texas) create composite impedances for multimodal trip making as an input to their gravity expressions. Virginia’s and Louisiana’s models and Ohio’s interim model rely heavily on Fratar factoring of existing OD tables. New Hampshire, Ohio, and Oregon use destination choice models. The fol- lowing list cites the numbers of states reporting each tech- nique. Gravity expression, without composite impedances across modes (12) ■■■■■■■■■■■■ Fratar factoring (3) ■■■ Gravity expression, with composite impedances across modes (3) ■■■ Logit expression, joint between distribution and mode split (2) ■■ Tour-based multinomial destination choice model (1) ■ Those statewide models that are considered multimodal require a mode split step. A variety of methods is used. Logit expression, mode split only (5) ■■■■■ Fixed shares (3) ■■■ Nested logit (3) ■■■ Logit expression, joint between distribution and mode split (3) ■■■ Diversion curves (1) ■ The preferred method of traffic assignment depends on the network detail in congested areas, typically in dense urban centers. Models with highly detailed networks can estimate volume-to-capacity ratios with some degree of certainty, so that equilibrium conditions can be estimated. Models with ab- breviated urban network representations are better off with a traffic assignment method that does not require delay infor- mation. The method of traffic assignment selected by most states is static equilibrium. Virginia uses stochastic multipath traffic assignment, whereas Maine, Michigan, and Montana use all-or-nothing traffic assignment. Dynamic traffic assign- ment (either equilibrium or all-or-nothing) is not used, even State Variables Kentucky MSA Size and Claritas Code (urban, second city, suburban, town, and rural) Louisiana Claritas Code (urban, second city, suburban, town, and rural) Wisconsin Household size by automobiles or workers by automobiles Delaware Income, employees per household, and persons per household Texas Household size by income Massachusetts Household size by automobile ownership; also household income, number of household workers, workers per vehicle, and numbers of school age children Connecticut Automobile availability by income category Maine Household size with either income or automobile ownership Michigan Household size and income and area type Indiana Household size by automobile availability by area type California Household size by income Vermont Household size by automobiles per household Note: MSA = metropolitan statistical area TABLE 3 VARIABLES USED IN CROSS-CLASSIFICATION MODELS FOR TRIP PRODUCTIONS

though trip lengths in statewide models exceed the duration of a peak period. Although some states intend to investigate traffic microsimulation for their statewide models, actual ap- plications of microsimulation have not yet been reported. Given the limitations of the available traffic assignment algorithms, most states have chosen to ignore the peak period or do simple factoring of 24-h traffic into a peak. Here are the adopted methods. Factored by percent of traffic in peak from traffic counts (7) ■■■■■■■ Peak period assigned directly (6) ■■■■■■ No factoring into peak (5) ■■■■■ Post-processed in another manner (2) ■■ No state overtly includes peak spreading. Massachusetts and Ohio reported having time-of-day models. Ohio’s model includes travel time as a variable in its utility expression; therefore, there is some sensitivity to traffic congestion. Every model uses speed and volume curves, such as the BPR curve, for delay calculations. Special Treatment of Long Distance Trips Beyond computational and data problems associated with the detail of networks in statewide models, the greatest obstacle for forecasting is good representation of long distance trip making. Obtaining good information on long distance trips has become more difficult as the 1995 ATS has aged; there- fore, states have discovered numerous ways to work around this limitation. All states with models are cognizant of the need to include long trips, and a little less than two-thirds of the states re- ported taking special actions to model long distance travel. A solid majority of these states create special trip purposes for long distance travel. There are three approaches: (1) seg- menting existing trip purposes into short and long distance categories; (2) creating separate trip purposes, such as recre- ation/tourism, to capture long trips; and (3) Fratar factoring an OD table of long distance trip purposes. Some other fixes were necessary, depending on the state. Delaware, Maine, and New Hampshire, in particular, reported the need to account for tourism during the summer months. California found it necessary to introduce k-factors during trip distribution; to use composite impedances for input to the gravity expression and to modify friction factors to account for long distance travel. A variety of data sources, cited here, were used specifi- cally to model long distance travel. There is no consensus as to the preferred data sources. Notably, Ohio performed a 2- week household survey of trips in excess of 50 mi. Four states are still using the 1995 ATS. 28 NHTS or NPTS (5) ■■■■■ ATS (4) ■■■■ Special long distance travel survey(s) (3) ■■■ Employment data from private vendor(s) (2) ■■ Employment data from public source(s) (2) ■■ Roadside surveys (2) ■■ Tourism economic or attendance data (2) ■■ U.S. Census (1) ■ Economic data from a private vendor(s) (1) ■ FAA data (1) ■ National and state park attendance database (1) ■ MPO survey data (1) ■ Borrowed data from another model (1) ■ Borrowed parameters from another model (1) ■ Own long distance survey (1) ■ Own household survey (1) ■ Seasonal traffic counts (1) ■ FREIGHT COMPONENTS Freight components of statewide models do more than simply complement passenger components, as is typical for urban models. Indeed, the driving forces behind statewide model development in many states are economic develop- ment issues that cannot be fully analyzed without a good freight component. In addition, freight is more easily ana- lyzed statewide than for urban areas because the scale of the geography is more compatible with available freight data sources. Thus, freight components for statewide mod- eling have evolved to a level of sophistication well beyond what is seen within MPO models. Freight components sometimes include commercial vehicles that are not carry- ing a commodity. The following discussion includes Montana’s freight component that is part of HEAT. HEAT is primarily a tool for economic forecasting, but contains a commodity-based truck model. Nature of Freight Components There are two fundamentally different styles of freight fore- casting: (1) direct forecast of vehicle flows without refer- ence to commodities or (2) forecasting of commodities, then using the commodity flow forecast to estimate vehicle flows. Of the 16 states reporting freight components as a part of their statewide travel forecasting model, 12 base their forecasts on commodities. Although they are much more complex, commodity-based models have a greater sensitiv- ity to economic conditions and to state policies toward in- dustrial development. Commodity flows are derived from data sources in either tons or dollars. Finding the effects of freight on the trans- portation system requires that commodity flows be converted to trucks, rail cars, shiploads, aircraft, barges, or containers.

29 The correct conversion requires knowledge of how much of a commodity is carried by a particular vehicle. These payload factors (tons per vehicle) can be obtained from several sources, as listed here. VIUS (Florida, Georgia, Michigan, Montana, Ohio, Wisconsin (6) ■■■■■■ Commercial freight data vendor (Kentucky, Louisiana, Tennessee, Texas) (4) ■■■■ Rail Carload Waybill Sample (Georgia, Indiana, Ohio) (3) ■■■ Data from another state or from an MPO (Kentucky, Virginia) (2) ■■ Truck intercept studies (Georgia) (1) ■ VIUS pertains only to trucks, and the Rail Carload Way- bill Sample only to railroads. However, the Rail Carload Waybill Sample can provide estimates of the density of com- modities, which can then be applied to other modes. Some freight components are closely tied to models of economic activity (e.g., Ohio’s new model and Oregon’s model) that account for commodity flows in units of dollars. To forecast vehicular flows there is an additional need for a conversion between dollars and tons. Two states, Georgia and Indiana, reported that their principal source of data on dollars per ton is the CFS. Many sources of freight data give commodity flows as yearly totals. For single-day forecasts (or peak periods with a single day) it is necessary to determine the fraction of yearly commodities transported in a day. This fraction can be obtained implicitly through OD table estimation techniques or explicitly by calculating the number of truck days in a year. The number of truck days ranges from 261 in Kentucky to 365 in Texas and Virginia. The distribution of commodities from zone to zone is han- dled by three methods, alone or in combination: (1) Fratar factoring a vehicle or commodity OD table that was created from data, (2) a gravity expression, or (3) a logit expression. Most states use a gravity expression. The new models in Ohio and Oregon, because of their economic activity and land use underpinnings, use logit expressions. Five states (Indiana, Louisiana, Ohio, Tennessee, and Vir- ginia) reported using techniques of OD table estimation from ground counts for improving their truck forecasts; however, Indiana used OD table estimation only for non-freight truck traffic and Ohio will be abandoning these techniques when its new model is completed. Four states (Florida, Indiana, Michigan, and Vermont) re- ported using “quick response” methods, such as the ones from the Quick Response Freight Manual, to supplement their freight forecasts. For example, Florida used these tech- niques only for non-freight truck trips. If a model is commodity-based, it is likely that states would need data on commodity flows for calibrating their trip generation and distribution steps. Slightly more then half of the states with freight components purchased the TRANSEARCH database from Reebie to understand com- modity flows. Three states were able to use the CFS instead. Oregon performed its own shipper and carrier survey. None of the major sources of commodity flows are com- plete. Some states have adopted different methods of dealing with missing commodities or industrial sectors and empty trucks, but the main objective is to adjust for the error by com- paring assignment results to truck counts. Indiana, Louisiana, and Virginia use OD table estimation from traffic counts to bridge the missing commodities and account for empties. Florida ignores the missing categories and Kentucky lumps all the missing categories into one catch-all commodity group. Ohio did its own establishment survey, which included all commercial vehicle movements, not just freight shipments. Freight components that are commodity-based usually re- quire that commodity production totals be estimated for each commodity category for each zone. Almost all states with this requirement derived commodity productions from em- ployment estimates and commodity output per employee. Kentucky obtained its production totals directly from Reebie. Similarly, freight components that are commodity-based usually require that commodity consumption totals be esti- mated for each commodity category for each zone. Estimat- ing consumption is more difficult that estimating production, because (1) the commodities consumed by an industry are not obvious by looking at the nature of the industry and (2) house- holds consume a large fraction of the commodities. One method of understanding commodity consumption is IO analysis (as suggested in NCHRP Report 260); however, only Michigan, Montana, Ohio, Oregon, and Vermont use IO. All states (except Kentucky) with a need to estimate commodity consumption by zone do it through employment estimates along with consumption per employee or through household estimates and commodity consumption per household. Commodity flow databases are often reported for fairly large spatial units such as counties or states. There is a need in some states to expand the flow matrices to cover much smaller spatial units. Half of the states with freight components created procedures for disaggregating their commodity flows. The method most often cited by states was to factor county-to- county flows into zone-to-zone flows using employment cat- egories and population totals. Commodity flows must be divided among modes. Only Florida, Ohio, and Oregon reported using mode split expres- sions (such as logit) to allocate commodities to modes. The remaining states use fixed shares from data. Indiana varies these shares by the distance of the shipment, which is facili- tated by the way data are reported from the CFS. A model

with fixed shares does not necessarily mean that the propor- tion of tonnage carried by each mode remains constant. Total mode shares can shift as commodity production and con- sumption patterns change in the future. However, fixed-share models are insensitive to changes in shipping costs that may give an advantage to one mode over another. A few commodity-based components further calculate the fraction of commodities carried by each truck type. All states reported using fixed shares, derived from the CFS, Reebie’s TRANSEARCH database, expert judgment, or the VIUS. Traffic assignments usually involve the mixing of passen- ger and freight traffic. States have adopted two methods of as- signing a mix of traffic: (1) preloading trucks onto highway links, and then performing a passenger car assignment or (2) loading trucks and passenger cars together. Preloading is often done with an all-or-nothing traffic assignment. When trucks and passenger cars are assigned together, a static user-optimal equilibrium traffic assignment is preferred. The decision be- tween the two methods (preloading or together) revolves around the question of whether truck routing is heavily influ- enced by traffic congestion, which is essentially ignored in an all-or-nothing assignment. When trucks are assigned together with passenger cars, a multiclass traffic assignment algorithm is required to account for the mix of vehicles on each link and to ensure that trucks are assigned only to legal routes. Because trucks have a greater impact on congestion than passenger cars, it is further necessary to weight truck volumes by a passenger- car-equivalent factor when calculating delays from a multi- class traffic assignment. For example, Louisiana derived a statewide average factor of 1.83 from the Highway Capacity Manual, a value that is appropriate for the terrain in that state. Only two states (Ohio and Wisconsin) explicitly handled transshipment of commodities. Four states used FHWA’s FAF for network development. A majority of states obtained network information from a GIS and used GIS for freight network storage. Freight Component Level of Detail Freight components can be either multimodal or concentrate on a single mode. No state reported concentrating on a single mode other than trucks, even though numerous railroad-only models have been described in the literature. Cited here are the modes reported as forecasted by statewide models. None of the models dealt directly with truck–rail intermodal, or indeed, any other intermodal pairings. Models also ignored categories of trucks that would be related to the economic structure of the trucking industry, such as a for-hire truck or a private truck. Truck, general (15) ■■■■■■■■■■■■■■■ Rail freight (5) ■■■■■ 30 Air freight (5) ■■■■■ Deep water shipping (4) ■■■■ Inland water shipping (3) ■■■ Less than truckload and truckload (1) ■ Florida, Ohio, and Wisconsin reported having all five of these major modes. All states with freight components have at least a truck network or a passenger network that has been modified for trucks. Ohio and Texas have networks for other modes besides trucks. Almost all states worked with all trucks together or just worked with heavy trucks. Michigan and Ohio divided trucks into heavy, medium, and light cate- gories, similar to the categories in the Quick Response Freight Manual. Montana’s HEAT divided trucks between truckload and less-than-truckload. All statewide models with a freight component do a 24-h truck forecast. Five states also reported the ability to do a peak-period truck forecast. It is desirable, but not necessary, that the in-state zone sys- tem for a freight forecast correspond to the zone system for a passenger forecast. All states reported consistent zone sys- tems except Texas, which has a coarser zone system for freight. Kentucky, Ohio, and Virginia use subzones or grid cells to increase the spatial detail where necessary. Because of the ready availability of freight OD data for the whole United States, a majority of statewide freight components cover most or all of the continental United States rather than relying on external stations at the state borders. Half of the statewide freight components cover parts of either Canada or Mexico. The following is a list of the ways in which zones are de- fined for out-of-state portions of the freight component. None of the states chose to use national transportation analy- sis regions. Some models used multiple sources of zones, de- pending on how far the area is from the state border. Counties or aggregations of counties (6) ■■■■■■ BEA regions or aggregations of BEA regions (6) ■■■■■■ States or aggregations of states (6) ■■■■■■ TAZs (2) ■■ External stations (1) ■ Multistate regions (1) ■ Freight components use special generators sparingly, and most models do not have any. New Jersey has the most with 200. Special generators include rail yards, airports, seaports, truck terminals, warehouses or distribution centers, pipeline terminals, and regional shopping malls. All truck networks have links that are coded to the same highway functional classes as passenger car networks. A state’s truck network has about the same number of links as its passenger car networks.

31 Those freight components that use commodities have many commodity categories. Vermont has the fewest cat- egories at 6 and Ohio has the largest number of categories at 32. There is a cluster of four states (Kentucky, Michi- gan, Virginia and Wisconsin) using between 25 and 28 categories. LAND USE AND ECONOMIC ACTIVITY A number of states formally consider economic activity as ei- ther an input to their forecasts or as a post-processor of model outputs. Ohio’s new model and Oregon’s model have land use and economic activity calculations that are tightly inter- woven with the rest of their components. Chapter three in- cludes a discussion of Ohio’s model. Indiana and Maine specifically mentioned using a commercial regional eco- nomic forecasting model. Montana’s HEAT is an economic model with a freight component. A few other states indicated that they are considering using a regional economic fore- casting model to post-process the results of their statewide travel forecasts. STATEWIDE AND URBAN MODEL INTEGRATION Good linkages between statewide and urban models are de- sirable, but not necessary. Rhode Island is a special case, be- cause its statewide model is an MPO model; therefore, there is no need to integrate. Here are some integration activities and the number of states participating in each. Statewide model provides independent estimates of traffic in areas covered by urban models (13) ■■■■■■■■■■■■■ Statewide model is used to develop external station forecasts for the urban models (13) ■■■■■■■■■■■■■ Statewide and urban models share geographic systems such as zones or networks (10) ■■■■■■■■■■ Statewide and MPO models use similar computational steps, trip purposes, base-year, or modes to promote compatibility (7) ■■■■■■■ Statewide model shares GIS databases with MPO models (6) ■■■■■■ Urban models incorporated as part of the statewide model (6) ■■■■■■ Institutional issues regarding the statewide model provide forecasts that might conflict with MPO models (3) ■■■ Statewide model provides impedances for use in the MPO models (1) ■ Most statewide models are coarser than MPO models within urban areas; therefore, the relative validity of the statewide versus urban models is obvious. Seven states commented that although their statewide model can pro- duce forecasts for urban areas, they defer to MPO model results if a conflict arises. As a condition for Wisconsin gaining the MPOs’ cooperation in building its model, the state needed to ensure their two largest MPOs that the statewide model would not be used for urban forecasts. Except for Rhode Island, where an MPO model is available, statewide models are not used directly for urban forecasts. Integration efforts thus far have been heavily influenced by the need to share data and to provide external station fore- casts for MPO models. VALIDATION All statewide models have been validated or are undergoing validation. The following is a list of the types of data used during validation. Passenger vehicle counts (24) ■■■■■■■■■■■■■■■■■■■■■■■■ Truck counts (15) ■■■■■■■■■■■■■■■ Comparisons to national default trip generation values (11) ■■■■■■■■■■■ Commuting OD flows from CTPP (11) ■■■■■■■■■■■ Comparisons to average values (or other statistics) from own travel surveys (8) ■■■■■■■■ Known trip length frequency distribution(s) (8) ■■■■■■■■ Comparisons to average values from similar states or cities (7) ■■■■■■■ MPO models (5) ■■■■■ Counts of passengers on buses (3) ■■■ Counts of passengers on trains (3) ■■■ MPO OD studies (2) ■■ Goods production by sector or zone (1) ■ Data from cordon surveys (1) ■ HPMS VMT estimates (1) ■ States tended to use a variety of data sources for valida- tion. All states already involved in validating their models used passenger car volumes. Most states also used truck counts. Criteria for validation of statewide models closely follow those found in urban models. Each state chose to use a variety of measures. VMT by functional class absolute deviation (18) ■■■■■■■■■■■■■■■■■■ Link root mean-square error (RMSE) by volume strata (17) ■■■■■■■■■■■■■■■■■ Screenline count absolute deviation (17) ■■■■■■■■■■■■■■■■■ Link absolute deviation (12) ■■■■■■■■■■■■ Cordon count absolute deviation (10) ■■■■■■■■■■ Correlation coefficient between link volume forecasts and counts (8) ■■■■■■■■

Link-by-link comparisons (1) ■ Other (1) ■ Only nine states reported using the Model Validation and Reasonableness Checking Manual (Barton–Aschman Asso- ciates, Inc. and Cambridge Systematics, Inc. 1997). Most states did not provide a qualitative assessment of how well their models validated. A few states gave vague responses, such as “well,” “acceptable,” and “fair.” Texas reported good comparisons between its freight component and flows from Reebie’s database. California stated that 44% of the links meet the “maximum desirable deviation” standard and an R-square of 0.83 between link counts and base case link volumes. Michi- gan reported that 80% of links in major corridors were within the “standard.” Louisiana provided a “maximum desirable de- viation” chart showing 95% of links meeting the standard. Only two states used OD table estimation from traffic counts, which would tend to arbitrarily improve the match between observed and forecasted volumes before validation. Because of the larger scales of statewide models, there is an expectation that the accuracy of these models would be less than urban models. Approximately half of the states ap- plied the same validation standards to statewide models as urban models. The other half used less stringent standards for their statewide models. Louisiana explained that because most of their links in the statewide model were low volume, it was possible to meet the looser criteria for urban roads of similar volumes. Oregon’s model, having an unusual structure, also had un- usual validation criteria. Research of current practices surprisingly found no existing clearly defined model calibration or validation criteria for inte- grated land use–transportation modeling. The modeling team and Peer Review Panel together developed several criteria for assessing model performance for the Gen1 Model: • Match production by sector and zone. • Match number of trips and average trip distances by trip purpose. • Minimize zone-specific constants by sector. 32 • Network flows to match counts by mode of transportation, with emphasis on interurban routes. • Match increments of land to changes in land price. • Match CTPP distribution for commuting flows. Each criterion has a specific numeric target. The network flows, for example, must fall within specified ranges based on total ob- served volume. Some targets are more liberal than for traditional urban travel models, owing to the complexity of the integrated models and their coarser geographic detail. Several subjective performance tests were also developed. Each required the model to produce sensible and reasonable results. Additional criteria for which specific numeric targets could not be defined include: • Destination and route choice response behavior. • Trip generation sensitivities. • Path and transportation cost testing. POST-PROCESSING Post-processing of model results is sometimes needed to ob- tain information that is compatible with decision processes on alternatives or policies. The need for post-processors de- pends on the already built-in capabilities of the state’s travel forecasting software package. States reported some post-pro- cessing for air pollution emissions, benefits evaluation, level of service determination, and economic impacts. Air pollution emissions (9) ■■■■■■■■■ Level of service determination (7) ■■■■■■■ Benefit–cost analysis (3) ■■■ Economic impact (2) ■■ Factoring volume-to-capacity ratios (1) ■ Validation and model performance statistics (1) ■ Indiana and Michigan use the same post-processor for economic impact, which is a commercial regional economic analysis package. Indiana assesses project benefits with Highway Economic Requirements System and in-house soft- ware, whereas Michigan uses a benefits module developed by a local university. Virginia reported the need to adjust volume-to-capacity ratios downward to account for the un- usually sparse networks with urban areas.

Next: Chapter Three - Case Studies »
Statewide Travel Forecasting Models Get This Book
×
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

TRB’s National Cooperative Highway Research Program (NCHRP) Synthesis 358: Statewide Travel Forecasting Models examines statewide travel forecasting models designed to address planning needs and provide forecasts for statewide transportation, including passenger vehicle and freight movements. The report explores the types and purposes of models being used, integration of state and urban models, data requirements, computer needs, resources (including time, funding, training, and staff), limitations, and overall benefits. The report includes five case studies, two that focus on passenger components, two on freight components, and one on both passenger and freight.

  1. ×

    Welcome to OpenBook!

    You're looking at OpenBook, NAP.edu's online reading room since 1999. Based on feedback from you, our users, we've made some improvements that make it easier than ever to read thousands of publications on our website.

    Do you want to take a quick tour of the OpenBook's features?

    No Thanks Take a Tour »
  2. ×

    Show this book's table of contents, where you can jump to any chapter by name.

    « Back Next »
  3. ×

    ...or use these buttons to go back to the previous chapter or skip to the next one.

    « Back Next »
  4. ×

    Jump up to the previous page or down to the next one. Also, you can type in a page number and press Enter to go directly to that page in the book.

    « Back Next »
  5. ×

    To search the entire text of this book, type in your search term here and press Enter.

    « Back Next »
  6. ×

    Share a link to this book page on your preferred social network or via email.

    « Back Next »
  7. ×

    View our suggested citation for this chapter.

    « Back Next »
  8. ×

    Ready to take your reading offline? Click here to buy this book in print or download it as a free PDF, if available.

    « Back Next »
Stay Connected!