Developing a Method Selection Tool for Travel Forecasting (2017)

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.

Final Report Project No. 08-94 7 CHAPTER 2. RESEARCH APPROACH 2(A) STATE-OF-THE-PRACTICE REVIEW Travel forecasting tools and methods can be classified, organized, and presented in myriad ways. The research team researched a broad two-dimensional scheme of travel models and decision- support tools. The research team then developed the state-of-the-practice review to support two subcategories of methods or tools used in travel forecasting at transportation agencies in the United States: 1) travel demand models; and 2) traffic assignment and simulation models. The research team conducted additional reviews on the evaluation of effects and project prioritization methods that were not included in the final software tool, although these areas are proposed for future enhancements (see Chapter 4). This state-of-the-practice review identifies literature documenting and comparing travel forecasting methods. References from these sources are provided (where beneficial) for comparison purposes, but extensive descriptions of the methods are not included in this brief review to avoid redundancy with the original sources. The state-of-the-practice review is annotated with call-out boxes to connect the current practice with the methods described in the next section (see Figure 5). In some cases, the methods in TFGuide are described using different terminology, so these call- out boxes are designed to specifically connect the methods with the practice review. For example, the trip generation methods in the state-of-the-practice review are described in TFGuide as trip production (cross-classification), trip production (regression), or trip attraction (regression), which are more detailed descriptions for trip generation methods. Travel Model Improvement Program Agency Needs Assessment Survey In 2013, the Travel Model Improvement Program (TMIP)—supported by Federal Highway Administration (FHWA)—conducted a survey of transportation agency analytical needs to support transportation planning activities. The TMIP Agency Needs Assessment Survey comprised detailed questions regarding the types of analyses, tools, and methods used at transportation planning agencies across the country, including state DOTs, regional MPOs, and transit and toll agencies. The survey included several methods and tools used to support multiple planning activities. A total of 203 agencies responded. Figure 2 presents the planning priorities for the transportation agencies in the survey, with higher values representing the highest priorities. Overall, travel demand forecasting tools are used primarily to support long-range transportation plans and transportation improvement programs. Call-out boxes describe methods in TFGuide that are referenced by the practice review.

Final Re FIGURE 2 The TMI transport forecasts Twenty-e agencies highlight also the i demand m port : PLANNING P P survey als ation foreca , and the rem ight percen use other m not only the mportance o odels. RIORITIES FO o reported t sts. Eighty-f aining 15% t of agencies ethods—or p importance f other meth R TRANSPO he role of tra ive percent of agencies use a trave ostprocessi of travel de ods used ei 8 RTATION AG vel demand of agencies use differe l model outp ng—to prod mand mode ther separate ENCIES models in d use a travel nt methods t ut directly, uce forecas ls in suppor ly or in com Pro eveloping a model to de o produce f and the rem ts (Table 1). ting plannin bination w ject No. 0 n agency’s velop travel orecasts. aining 72% These data g activities, ith travel 8-94 of but

Final Report Project No. 08-94 9 TABLE 1: ROLE OF TRAVEL DEMAND MODELS IN DEVELOPING AGENCY'S TRAFFIC/TRANSIT/FREIGHT FORECASTS  TRAVEL DEMAND MODEL ROLE COUNT PERCENT We do not have a travel demand model. We rely entirely on other methods (e.g., sketch-planning models, growth factoring, diversion curves, etc.). 29 15.4% We have a travel demand model but use it in conjunction with other independent methods (sketch-planning models, growth factoring, diversion curves). 27 14.4% We have a travel demand model and generally rely exclusively on its direct outputs. 53 28.2% We have a travel demand model, but typically postprocess or otherwise adjust its outputs to produce forecasts. 79 42.0% Total 188 100.0% Recently, newer methods within travel demand models—those that link trips into tours or produce linked daily patterns of travel—have gained attention. Transportation agencies have not yet fully adopted these new methods of travel modeling methods, since most agencies (90%) still rely on traditional trip-based travel models. However, the landscape for these newer methods is changing; 28% of agencies are either planning for or considering the newer tour/activity methods, 25% of agencies are undecided, and 37% of agencies are not planning to adopt the new methods.  The TMIP survey explored details of additional methods and tools used by agencies in conjunction with travel and land-use modeling methods; these results are provided in Figure 3. Most agencies (60%) use data visualization tools. Many agencies also use vehicle emissions models (48%), traffic optimization or traffic microsimulation tools (45%), or dynamic traffic assignment (DTA) (33%). Thirty-one percent of agencies also use economic benefit-cost tools.

Final Re FIGURE 3 Importan methods project co sketch-pl emission scenario Transpor Figure 4. limited st more wit Transpor challenge destinatio sensitivit port : SHARE OF A tly, transpor and tools (e sting, air di anning and s (4%) forec planning, la tation agenc The limitat aff time and h less. tation agenc s linked to i n travel, fre y to the buil GENCY USE tation agenc .g., sketch-p spersion, an strategic mo asts. The re nd-use alloc ies face num ion of curren budgets we ies also repo naccurate o ight, and tra t environme OF METHODS ies indicate lanning and d decision-s dels, most f maining 7% ation, traffic erous chall t methods a re paramou rted on the r uncertain f vel times. A nt and poor 10 AND TOOLS d they had n strategic mo upport tools ocused on p of agencies optimizatio enges to con nd tools wa nt, pointing biggest tech orecasts of s dditional te spatial and t ot widely ad dels, crash ). Of the 25 roducing tra identified w n, and data duct planni s only of me to a need fo nical challe ocioeconom chnical chal emporal res Pro opted sever forecasting, % of agenci vel (14%) o ere using th visualizatio ng analyses, dium impor r agencies to nges, with f ic variables lenges inclu olution. ject No. 0 al categorie noise effect es who use r vehicle ese models n purposes. as shown in tance, while be able to ive of the to , origin- ded a lack o 8-94   s of , for do p six f

Final Re FIGURE 4 Special N The TMI the Unite provide a confirm t methods) Travel D The curre passenge literature Tradition Passenge These are and Simu predict th are typica (NCHRP demand f 2012). Th the 4-step choice, a assignme port : IMPORTANC ote: Importa P survey pro d States. Al useful fram he importan . emand M nt state-of-t r- and freigh comparing al Passeng r models pre the four ste lation Mode e time perio lly aggrega ) has produc orecasting g e most rece planning m nd highway nt) have bee E OF AGENC nce is on a sc vided a bro though man ework for f ce of the top odels he-practice t-demand fo methods on er Models dict the am ps in the 4- ls section.) d for reside te trip-based ed three ver uidance and nt version is odel. The 4 assignment. n added to t Y CHALLENG ale of 1-9, w ad overview y of the topi urther enhan ics that wer in travel dem recasting m metropolita ount, locatio step plannin Additional s nt and nonre models. Th sions of a re reflect trav a thorough -step proces Additional he 4-step pr 11 ES TO COND ith 1 being le of topics o cs were bey cements to e included i and model odels. The r n, statewide n, mode, an g process. (R teps have b sident trave e National C port (Repor el character description s refers to tr steps (i.e., t ocess over m UCT ANALYS ast important f interest to ond the scop this research n this resear s includes tr esearch team , and projec d route for r oute is disc een added o l patterns. T ooperative ts 187, 365, istics (TRB, of techniqu ip generatio rucks, extern any decade Pro ES FOR PLAN and 9 being transportatio e of this res (see Chapt ch (travel fo aditional an summariz t-level dema esident pass ussed in the ver the last raditional p Highway R and 716) to 1978; TRB es and param n, trip distri als, time-of s. For refer ject No. 0 NING most importa n agencies earch, they er 4) and recasting d advanced ed select nd modelin enger trave Assignmen three decade assenger mo esearch Pro provide tra , 1998; TRB eters used bution, mod -day, and tr ence, the 8-94 nt. in g. l. t s to dels gram vel , in e ansit

Final Report Project No. 08-94 12 Trip Production (Regression) following identifies the most common methods in traditional passenger demand models from this report. Trip generation is commonly considered the first step in the 4-step modeling process. It addresses the question of how many trips of each type begin or end in each location. It is standard practice to aggregate trips to a specific unit of geography (e.g., a traffic analysis zone)….Trip generation models require explanatory variables that are related to trip-making behavior and functions that estimate the number of trips based on these explanatory variables. While these functions can be nonlinear, they are usually assumed to be linear equations, and the coefficients associated with these variables are commonly called trip rates. Whether the function is linear or nonlinear, it should always estimate zero trips when the values of the explanatory variables are all zero. Mathematically, this is equivalent to saying that the trip generation equations should include no constant terms. The gravity model is the most common type of trip-distribution model used in 4-step models. In the equation, the denominator is a summation that is needed to normalize the gravity distribution to one destination relative to all possible destinations. This is called a “doubly constrained” model because it requires that the output trip table be balanced to attractions, while the numerator already ensures that it is balanced to productions. Most mode choice models use the logit formulation. In a logit mode choice model, the alternatives represent the modes. The utility is a function of the explanatory variables. These variables may include the following: • Modal level-of-service—Auto in-vehicle time, transit in-vehicle time, wait time, walk access/egress time, auto access time, transit fare, parking cost, number of transfers. • Traveler characteristics—Vehicle availability (sometimes relative to other potential drivers), household income, gender, age, worker/student status. • Area characteristics—Development density, pedestrian environment. FHWA, Federal Transit Administration (FTA), TRB, and the Office of the Secretary of Transportation sponsored a study to gather information needed to determine the national state-of- the-practice in metropolitan-area travel demand forecasting by MPOs and state DOTs (TRB, 2007). This special report evaluated the current state of practice and identified shortcomings along with recommendations for improvements; its audience was officials and policymakers who rely on the results of travel forecasting. The transportation analysis methods discussed in this report (3-step, 4-step, 5-step, population synthesis, household activity-based, and traffic microsimulation) are matched with land-use analysis methods (geographic information systems, accessibility models, real estate market models, input-output models, and business and residential location models) that can be tailored to the issues being addressed, based primarily on the level of detail required for analysis. The research concluded that no single approach to travel forecasting procedures is appropriate for all planning and policy needs. Trip Distribution (Gravity)

Final Report Project No. 08-94 13 In a study by Mishra et al. (2013), a comparison of destination choice and gravity models has been presented using a real case study applied within the Maryland Statewide Transportation Model. The authors noted several trip- distribution models in the paper (e.g., Fratar Growth model, Simple Growth Factor model, Furness model, Detroit model, Gravity model, and Destination Choice model), but focused their discussion on the last two models. The authors found that while the gravity model has some limitations as an aggregate model, disaggregate approaches have recently gained more attention with the development of logit and other discrete choice techniques. The application of destination choice models is still being studied, and while these models include the capability to accommodate population heterogeneity, there is not broad adoption. Destination choice models are estimated at the disaggregate level, and impedance variables interact with individuals’ demographic and socioeconomic characteristics. As a result, a destination choice model with continuous attributes interacting with impedance cannot be applied in practice since individuals’ continuous attributes are unavailable at the traffic analysis zone (TAZ) level unless an activity-based (AB) model is applied to simulate each individual’s travel behaviors. In addition, when a destination choice model is developed at the individual level, and then applied at the TAZ level, a spatial aggregation error will occur in both model estimation and application. Mishra et al. (2013) also discussed some of the advantages of destination choice models over gravity models, which included the presence of agglomeration effects and hierarchy in destination choices. Moreover, simultaneously modeling the effects of land use on travel behavior was found to be another advantage of destination choice models. The authors noted the inability to evaluate many policy issues for long-range planning as a disadvantage of gravity models. Mishra et al. (2013) concluded that the destination choice model better replicates the observed trip length and origin-destination matrix for home-based work trips and supports the hypothesis that the destination choice model provides more accurate results than gravity models in statewide travel demand model applications. In a study completed at Queensland University of Technology by Khan (2004), various nested logit mode choice models for different trip lengths and trip purposes were estimated using the data from a stated preference survey conducted in the study area. As part of this study, the author conducted a state-of-the-practice literature review on passenger mode choice modeling, focusing on modeling specifications and estimation techniques. Two types of trip generation models were evaluated: 1) multiple linear regression (zone- or household-based); and 2) cross-classification. The study observed the following: Both the trip generation models have been highly assessed by transport planners due to their distinct paradigms; however, linear regression continues to be most popular due to its simple formulation and analytical tractability. Shen (1994) developed a trip generation model for New Jersey and New York in the United States using both the regression and classification approaches. He demonstrated that regression analysis is a very practical tool for trip generation analysis under the circumstances where no household survey data is available and Destination Choice Model

Final Report Project No. 08-94 14 socioeconomic data is at the aggregated zonal level. Szplett and Kieck (1995) found that the individuals residing in rural areas generally have to take more long distance trips than the households in urban areas. Therefore, the trip generation rates are always unpredictable and unstable and cannot be accurately determined using cross-classification approach. Freedman et al. (1999) further supported the analysis of Szplett and Kieck by criticizing the cross- classification approach due to the inaccuracy of the results proposed by the model as compared to the actual data. (Khan, 2004) Three types of mode choice models were reviewed and considered in this research: 1) logit choice (multinomial or binary); 2) probit; and 3) general extreme value. The most commonly used choice models are logit models, with probit or general extreme value models working under less- strict and more-complex conditions. The report presents a comparison of these three methods based on the basic hypothesis, major constraints, model formulations, model estimation complexity, introduction of access modes, and practical applicability. Probit and general extreme value models are typically not used in practice because of the more-complex estimation process. Horowitz (2006) conducted a NCHRP synthesis that detailed current knowledge and practice on statewide travel forecasting models designed to address planning needs. The author discussed the types and purposes of models being used, integration of state and urban models, data requirements, computer needs, resources, limitations, and overall benefits. The core of this synthesis included results of surveys received from every state that has a statewide travel forecasting model. Information about modeling activities was provided by 49 states returning at least one questionnaire. The responses to the synthesis questionnaires, along with those from an earlier questionnaire distributed by the TRB Statewide Travel Demand Models Peer Exchange in 2004, permitted a general assessment of the state-of-the-practice. The questionnaires focused on individual components of the models and the modeling processes. This report presented a list of statewide travel demand model uses, with the top 10 including the following:  Corridor planning.  Statewide system planning or system environmental impact statement (EIS).  Bypass studies.  Regional planning, assisting an MPO model.  Project-level traffic forecasts or project EIS.  Regional planning, substituting for a local model.  Air quality conformity analysis, freight and intermodal planning, traffic impact studies, economic impact studies and long-term investment studies (tied for 7th place). States with experience in model usage had a greater confidence in model validity and tended to report more uses. The report included a wide variety of applications for statewide models. None of the states reported using their models either for truck-weight studies or for safety analyses. The FHWA Office of Project Development produced procedural guidance on travel and land-use forecasting in the context of the National Environmental Policy Act (NEPA) process (Resource Mode Choice (Disaggregate) Model

Final Report Project No. 08-94 15 Systems Group, Inc., 2010). The purpose of the interim guidance was to encourage improvement in the state-of-the-practice in project-level forecasting as it is applied in the context of the NEPA process. This guidance shares key considerations, lessons learned, and best practices regarding how to apply forecasting. DOTs can use the guidance to avoid common issues and improve the quality of forecasts, resulting in faster and more effective project delivery. Others had documented technical guidelines for producing forecasts for projects; however, guidance on procedural or process considerations in forecasting was not provided. Advanced Passenger Models TRB sponsored a synthesis as part of NCHRP to conduct a comprehensive review of current and past efforts in advanced modeling (Donnelly, 2010). The synthesis explored the use of travel modeling and forecasting tools that represent significant advances. The synthesis included five types of models: 1) AB; 2) dynamic network; 3) land-use; 4) freight; and 5) statewide models. The synthesis included a literature review; detailed interviews among federal, state, and metropolitan agencies, and consulting firms; and case studies. The study team interviewed more than 30 practitioners and researchers for the study and the study team discussed the highlights of each interview, with the interviewer(s) summarizing the key topics and discussion items. The major findings of those discussions were that the right model is the one that best meets the policy needs of the agency. This research summarizes the types of policy questions, whether they can be answered with traditional models, what type of advanced model would be most beneficial, and the benefit provided by the model. Policy issues for highway, transit, emissions and greenhouse gases, pricing, land use, validation, interaction with population, and other topics are included. The Association of Metropolitan Planning Organizations (AMPO) also sponsored a study to assess advanced travel model performance for use by MPOs (Vanasse Hangen Brustlin et al., 2011). The study provided technical guidance to the MPO community on the relative costs and benefits of committing resources to the development and implementation of AB models as a replacement for traditional models. The study provided qualitative and quantitative data to inform and support educated, fact-based decisions by MPOs on the allocation of resources to advanced travel forecasting techniques—specifically, the implementation of AB models. The Virginia Department of Transportation (VDOT) commissioned research to understand the feasibility of implementing AB models in Virginia (Virginia Department of Transportation, 2009). The report summarized the theory and practice of 4-step models and AB models, assessed the costs and benefits of moving to an AB model framework, and reviewed the practical concerns regarding the performance of such models. The research included interviews with agency staff in many of the metropolitan areas in which AB models were being developed or were in use; the research also reviewed published literature on the theory and practice of AB models. The report also described techniques that are available to improve traditional 4-step models, and noted many techniques can achieve benefits attributed to AB models. In a TRB paper by Lemp et al. (2007), two competing approaches to travel demand modeling were compared (traditional 4-step travel demand models and AB models) using an application in Austin, Texas. The paper revealed several differences in model performance and accuracy, in

Final Report Project No. 08-94 16 terms of replicating travel survey and traffic count data. The authors concluded that AB models—while requiring more data manipulation, model calibration, and application—are generally more sensitive to changes in model inputs, showing that aggregate models omit important behavioral distinctions across the population. FHWA, in cooperation with the Ohio Department of Transportation (ODOT), has also conducted a comparison of the forecasting results for three projects, at the regional and local level, for trip- based and tour-based models (Ferdous et al., 2011). The three projects represented growth in land-use and transportation investments. The results indicated that tour-based models performed slightly better compared to observed data at the regional scale, and both models performed equally well at the local level. TRB-sponsored research has described a decision-making framework and has cataloged analytical tools that detail likely effects of user-fees and tolling on revenue generation and system performance (Parsons Brinckerhoff, Inc. et al., 2012). Volume 2: Travel Demand Forecasting Tools of NCHRP Report 722 focuses on travel forecasting tools available for providing demand estimates for priced facilities. The report contains the following recommendations for models used to analyze demand for priced alternatives:  The travel model should be sensitive to pricing across several dimensions, including route choice, mode choice, destination choice, and time-of-day.  The travel model should have a minimum segmentation of 4–5 purposes and 3–4 income groups.  A binary toll\nontoll choice model should be considered either as a sublevel nest in mode choice or as a preassignment step to overcome toll bias and allow for nonlinear specification of the tradeoff between time savings and toll cost.  An improved time-of-day choice model is recommended, though primarily in the context of tour-based decisions. The report’s authors recommended development of an AB microsimulation model as a long-term enhancement, to leverage flexibility in the definition of tolling options and the availability of a synthetic population for sensitivity to tolls and equity analysis. Transit Passenger Models Transit ridership forecasts use transit passenger models for different short- and long-range planning purposes:  Service expansions or reductions (short-range).  Fiscal year budgeting (short-range).  Fare increases, special fair policies, and new fare instruments (short-range and long- range).  Station planning and design (short-range and long-range).  Projections for short- and long-range transit plans.  FTA New Starts and Small Starts fixed-guideway funding applications (current-year forecasts plus optional 10-year or 20-year forecasts).

Final Report Project No. 08-94 17 The methods chosen for these applications will often vary based on whether the planning purpose is to produce a short- or long-range forecast. Many transit agencies view regional travel models as primarily focused on automobiles, including transit model components aimed at removing a reasonable share of the region-wide person trip demand. MPOs in urban areas with significant transit share have invested in and developed transit model components that include more sophisticated path building and market segmentation routines with the goal of producing more accurate results. Several methods are utilized that often borrow certain elements from a regional travel model; this is because of the inherent complexity associated with accurately predicting transit ridership. These methods incorporate data in off-model methodologies that go beyond simple validation. In support of ridership forecasting for New and Small Starts applications, the FTA has given practitioners guidance at a series of presentations delivered at multiday workshops in Minneapolis, Minnesota (June 2006); St. Louis, Missouri (September 2007); and Tampa, Florida (March 2009). The workshops, designed for practitioners and project sponsors, were developed to communicate New Starts and Small Starts requirements, describe the current best practices regarding the use of data and the selection of methods, and illustrate what constitutes good model development and validation procedures. The reader should review the information on FTA’s Travel Forecasts web page for the latest from FTA about ridership forecasting for proposed New Starts and Small Starts projects. As of this writing, sponsors can choose among three methods to prepare ridership forecasts: 1. Regional travel models. 2. Incremental data-driven methods. 3. FTA’s Simplified Trips-on-Project Software (STOPS). The FTA developed STOPS as a simplified ridership-forecasting method for local agencies planning major transit projects. Local agencies can opt to use STOPS to meet all the forecast-related requirements for transit projects proposed for federal funding. STOPS is a region-wide travel model, similar to traditional trip-based models maintained by MPOs in larger metropolitan areas. The package is simplified in two ways. First, its development has already accomplished the specification and calibration of its component models—in this case, using data assembled nationally from transit systems with fixed guideways. Second, its application relies primarily on already-available data: 1) the Census Transportation Planning Package, for worker commuting patterns; 2) the General Transit Feed Specification data, for detailed representation of local transit services; and 3) information from the local MPO travel model to represent zone-level population, employment, and highway impedances. STOPS also has an “incremental” option that grounds transit forecasts in user-provided transit passenger trip tables derived from well-implemented transit rider surveys. Like MPO-maintained travel models, the STOPS mode choice and transit-loading models predict the number of zone-to-zone trips on transit, the distribution of those transit trips by access mode (including park-ride and kiss-ride), and the volume of trips by access mode at each boarding location (including designated park-ride facilities on fixed guideways or bus routes). Pivot-Point and Incremental Models Direct-Demand Model

Final Report Project No. 08-94 18 STOPS is intended for use by capable travel forecasting professionals. FTA provides detailed information on STOPS, as well as downloads of the software and a contact name for technical assistance, on the STOPS webpage that is accessible from FTA’s Travel Forecasts page. A TRB report by Coffel et al. (2012) includes guidelines for arranging and integrating various station design elements to promote access to public transportation. As part of the study, the authors assessed several evaluation tools that can support transit planning for access to transit stations. Travel demand modeling was one of the primary tools identified. Travel demand models are a familiar tool for estimating transit ridership, with 51% of transit agencies reporting that they use 4-step travel demand models for forecasting ridership. The report observed: Numerous travel demand models capable of assessing the impacts of at least some transit access alternatives have been developed. However, these models are not generally available to transit agencies for planning for access to transit stations, as the cost and data requirements of developing such sophisticated demand modeling tools are prohibitive for many MPOs and few transit agencies have resources to develop their own models. As a result, many transit agencies use other methods to estimate ridership. Table 5 summarizes the responses collected for TCRP Synthesis 66, which collected information on ridership estimation methods currently used by transit agencies. This report found that just over half of all transit agencies use their regional travel demand models for ridership estimating. Instead, the majority of transit agencies rely on more qualitative methods of forecasting ridership, such as judgment or rules of thumb. Relatively few transit agencies use econometric models or regression analyses, with only one out of every five identifying them as a forecasting technique. Coffel et al. (2012) Freight and Goods Movement Models Freight and goods movement models predict the amount, location, mode, and route of commodity flows. These models parallel passenger models with trip generation, trip distribution, mode choice, and trip assignment components. Commodity-based models segment goods movement into commodity groups and apply parameters specific to an industry. Supply chain models also predict business relationships between buyers and suppliers, along with the supply chains that result from these business relationships. Tour-based truck models predict the pick-up and delivery system of local goods movements. Microsimulation models predict individual shipments and then convert these annual shipments to daily or weekly deliveries in each vehicle (e.g., truck railcar). TRB sponsored research as part of the National Cooperative Freight Research Program (NCFRP) that presented an evaluation of possible improvements in freight-demand models and other analysis tools, and provided a guidebook to assist model developers in implementing these improvements (Cambridge Systematics, Inc. and GeoStats, LLP, 2010). The report focused on use of existing data to develop data inputs for the model, and found that existing and readily available data could be used to develop the inputs required by freight models. The study team also developed a proposed 10-step process to assist model developers in implementing freight transportation planning, including these improvements. The report developed the guidance based

Final Report Project No. 08-94 19 on the literature review and model framework, a survey of public decision-makers, and additional research in support of this project. The report noted the following: The approach to this study was driven by a desire to understand the needs of decision-makers and planners and to assess the degree to which existing technical tools meet these needs. This is a departure from the more traditional method of reviewing existing models and determining the most feasible improvements. As a result, this approach relied heavily on information collected from a series of interviews conducted concurrently with freight modelers, planners, and decision-makers from state DOTs and MPOs.…[M]ost of the tools are widely used in practice and can be used to answer a number of freight related planning and policy questions. (Cambridge Systematics, Inc. and GeoStats, LLP, 2010) This report includes descriptions for several freight models, including time series, behavioral, commodity-based and input- output, multimodal network, microsimulation and agent-based, supply chain and logistics, network design, routing and scheduling, and other emerging topics. These descriptions included an assessment of how widely used the methods were for model development, model implementation, and public-sector applications. Aggregate-level approaches to freight modeling focus on modeling the conglomeration of user demand. The typical example of such an approach is the 4-step planning method that includes trip generation, trip distribution, mode split, and traffic assignment. Different measurement units of freight demand used (e.g., commodities or vehicle trips) lead to commodity-based or vehicle- trip-based models. Regardless of units used for measurement, these approaches result in commercial vehicle trips. Since the TRB-sponsored research, FHWA has supported the development of supply chain and tour-based methods to support regional and statewide freight forecasting (Wies et al., 2013). The complete framework comprises two parts: 1) a national part that focuses on applying supply chain methods to model commodity shipments; and 2) a regional part that focuses on tour-based methods to model truck movements. Shipments developed in the national part of the framework are tracked through the regional part of the framework as they are moved by trucks. Wang and Holguín-Veras (2010) have also studied tour-based urban freight travel demand models, and have developed a summary of types of urban freight models that segments urban freight travel demand models into aggregate and disaggregate models; then by trip-based, tour-based, and AB; and then by commodity-based and vehicle trip-based methods. The broad category of commercial vehicles also includes a significant share of vehicles that do not carry freight, including but not limited to business services, landscaping, other household maintenance services, public safety, construction activity, vehicle rentals, and paratransit services. In 2002, FHWA sponsored research to account for all commercial vehicles in urban transportation models (Cambridge Systematics, Inc., 2004). The Commodity Flow Model Supply Chain Model Vehicle Types and Tour Patterns Freight and Services Tours and Stops

Final Report Project No. 08-94 20 research estimated the share of nonfreight truck traffic at approximately 5.6% of total vehicle miles traveled (VMT) on the interstate system, and approximately 5% of VMT on principal arterials. The report identifies three broad methods of estimating commercial vehicle demand: 1) aggregate methods that apply national parameters to estimate commercial vehicle trips, like the aggregate-level models; 2) network-based quick-response methods; and 3) locally estimated models. These methods are still in use as of 2017. The report recommended tour-based and supply chain models as avenues for future research, and much of this research has taken place in the intervening years. Stefan et al. (2005) describes just such an approach in a TRB paper. The paper describes an agent-based microsimulation framework that uses a tour-based approach and emphasizes important elements of urban commercial movement, including the role of service delivery, light- commercial vehicles, and trip chaining. The microsimulation uses Monte Carlo techniques to assign tour purpose, vehicle type, next-stop purpose, next-stop location, and next-stop duration. Tours are “grown” from the sequential processing of decisions. Assignment and Simulation Models The current state-of-the-practice in the use of assignment and simulation models in travel forecasting includes four types of models: 1) integrated and multiresolution models; 2) static equilibrium assignment; 3) DTA; and 4) traffic microsimulation. Integrated Multiresolution Models Each domain scale (i.e., macro-, meso-, and micro-) of simulation has level-of-resolution and spatial-scope limitations. Macroscopic models can analyze large networks and estimate mode shift, but these models cannot provide detailed information about individual vehicles or interactions between vehicles. Microscopic models can estimate individual vehicle movements, but only on relatively small networks. Mesoscopic simulations attempt to strike a balance between the two by modeling route choices of individual drivers with a limited level of detail pertaining to driver behavior. In order to incorporate the benefits of each single resolution simulation type, many practitioners have begun to utilize an integrated approach to transportation modeling. This integrated (or Multiresolution Modeling [MRM]) utilizes each modeling domain in the following manner (FHWA, 2013):  Macroscopic. Trip table manipulation for the discernment of overall trip patterns.  Mesoscopic. Analysis of the effect of driver behavior in reaction to mitigation strategies.  Microscopic. Analysis of the effect of traffic control strategies at roadway junctions. FHWA’s (2013) The Effective Integration of Analysis, Modeling and Simulation provides a thorough evaluation of each domain’s practicality and applicability and establishes means by which several domains are correlated. For example, many MPOs and transit agencies currently utilize macroscopic travel demand models for transit network improvements; incorporating DTA or microsimulation into an integrated modeling approach may better serve their objectives. The research presents potential relationships between different analysis, modeling, and simulation domains (land-use models, macroscopic travel demand models, AB models, mesoscopic DTA Integrated Multiresolution Models

Final Report Project No. 08-94 21 models, microscopic simulation models, deterministic operation models, emissions models, safety models, and sketch-planning tools). However, the tools to connect these areas may still be in development, or only available through proprietary software. The California Department of Transportation (Caltrans) (2012) sponsored a study that evaluated traffic simulation model use in the development of Corridor System Management Plans (CSMP). The project sought to evaluate the effectiveness and value of using traffic simulation in the development of CSMPs, and to determine whether the use of simulation models provided benefits that exceeded their actual and apparent costs. The study reviewed traffic simulation capabilities. Moreover, the study assessed experiences with traffic simulation for corridor evaluations through a survey of simulation modelers and decision-makers, and a detailed review of simulation modeling efforts. Static User Equilibrium Assignment Static User Equilibrium assignment methods are the prevailing state-of-the-practice at agencies across the United States. The user equilibrium (UE) concept in traffic assignment is an attractive modeling assumption because it reflects the volume dependence of roadway network performance, the natural tendencies of users to seek their best routes, and the need for consistency between the travel times used in modeling trip distribution and those generated by traffic assignment (FHWA, 2013). Newer static assignment methods have emerged in recent years that seek to improve the convergence of the UE problem. This report does not present literature on this research since it is not germane to project objectives. Traffic assignment models are used to assign the vehicular origin-destination table to the highway network. These models employ some measure of impedance, combining travel times, cost, and (sometimes) reliability to estimate which routes will be used by travelers. These assignment (or route-choice) models may use a relatively simple model of traffic congestion to predict travel times (e.g., the Bureau of Public Roads equation). Alternatively, the assignment models may use a more elaborate multiresolution process, combining DTA with mesoscopic or microscopic simulation models to estimate traveler route preferences and traffic congestion. Dynamic Traffic Assignment Use of DTA models has become more prevalent among MPOs and other transportation agencies in the United States that are tasked with assessing the effects of potential transportation projects. These time-variant traffic flow models provide valuable insights for practitioners toward the macroscopic or mesoscopic effects of a change to the system through analysis of queuing, route assignment, and travel time reliability on a link (or even network) level. Due to the popularity of DTA models, their appropriate application in travel demand modeling has been the subject of several detailed investigations. One recent exploration into the state-of-the-practice of DTA comes through the FHWA’s Traffic Analysis Toolbox Volume XIV: Guidebook on the Utilization of Dynamic Traffic Assignment in Modeling (Sloboden et al., 2012). The primary purpose of the guidebook is to provide engineers User Equilibrium Traffic Assignment Dynamic Traffic Assignment

Final Report Project No. 08-94 22 and planners in various MPOs and state DOTs with direction on the appropriate applications of DTA tools for transportation decision-making. Since DTA models can capture the interactions between travelers and the network, they allow practitioners to test various network structures and schemes, including alternative capacity, traffic control (e.g., signals or ramp meters), pricing, and evacuation planning. Some of the recommended applications include the following:  Bottleneck removal studies.  Active transportation and demand management strategies.  Integrated corridor management strategies.  Operational strategies.  Incident response management scenarios.  Special events.  Work-zone effects and construction diversion. The toolbox contains a complete list of applications; however, before applying DTA modeling to any of these potential projects, practitioners must be aware of the basic requirements for the proper utilization of DTA. Foremost, origin-destination (O-D) data, categorized into peak periods/hours, is a fundamental input to any DTA; without this information, developing an accurate DTA model is difficult. Additionally, extensive temporal data collection efforts are required to calibrate the model. Additional requirements include access to software with DTA capabilities, staff members with expertise in transportation modeling, and extensive temporal data collection efforts to conduct the required model calibration (Sloboden, 2012). TRB’s (2011) Dynamic Traffic Assignment: A Primer supplements these findings by recommending numerous additional potential applications; the document contains a full list of such potential applications. Despite the document’s observation that DTA requires extensive data collection, model calibration, and expertise, these models can be useful for MPOs and state DOTs conducting operational planning projects. DTA models are especially useful when these planning projects involve changes to roadway configuration, freeway expansions, development of a city bypass, the addition of high-occupancy toll (HOT)/high-occupancy vehicle (HOV) lanes, integrated corridor improvements, and travel demand management strategies (e.g., congestion pricing). DTA models are also recommended for engineers working on large-scale, real-time traffic management or information provision issues; this is due to their ability to address these issues in a “systematic manner because they provide capabilities to estimate future network conditions (flow patterns) that will result from a particular…strategy” (TRB, 2011). Traffic Microsimulation Models Although macroscopic and mesoscopic DTA models allow practitioners to simulate large networks, they do not provide a detailed analysis into individual motorists’ route choices. As a result, the use of traffic microsimulation models has become the prevailing methodology for modeling the response of individual drivers to changes within the network. A microscopic model allows practitioners to evaluate the behavior of an individual vehicle within the traffic stream, whether done with a car-following, lane-changing, or route- choice model. A microscopic simulation requires detailed input; the calibration and validation of Traffic Microsimulation Model

Final Report Project No. 08-94 23 model inputs can be arduous. When selecting a simulation type, practitioners must be cognizant of the level of detail that their projects require. FHWA’s Traffic Analysis Toolbox Volume III: Guidelines for Applying Traffic Microsimulation Modeling Software provides transportation professionals with a recommended process (seven steps) for properly applying microsimulation software to transportation analyses (Dowling et al., 2004). It recommends using microsimulation software to model the traffic performance of highways, streets, transit, and pedestrian facilities where second-by-second (or subsecond-by- subsecond) analysis is constructive. However, it is important to consider the scope of such projects, where careful consideration must be paid to securing the proper expertise, allotting sufficient time and financial resources, and fostering a detailed database for the development of a base model (Dowling et al., 2004). The American Association of State Highway and Transportation Officials’ (AASHTO) 2010 report, Best Practices in the Use of Micro Simulation Models (Sbayti and Roden, 2010), identifies several instances where microsimulations were found to be warranted and cost effective. The report defines worthwhile projects as endeavors where the following applies:  Detailed interactions of vehicle movements are the principal motivation for the study (these include interactions of vehicles with pedestrians/bicyclists, HOVs and buses, and congested traffic).  Alternative design considerations are needed related to lane changing.  The visual animation of traffic conditions will improve the credibility of potential solutions. Cervenka (1997) assessed the use of traffic microsimulation at the regional scale for use by MPOs. Using the Dallas-Fort Worth region as a study area, Cervenka (1997) identified factors that could hinder the successful application of the 4-step travel model toward simulating individual activities and travel. Many of the same drawbacks identified in previous studies (i.e., extensive data and computational requirements) challenged Cervenka’s (1997) study; however, traffic microsimulation showed potential for significantly improving existing travel forecasting procedures used by MPOs and state DOTs. Summary Travel demand forecasting models employ different, purpose-specific methods (i.e., specific methods exist for transit and freight models, although traditional and advanced travel forecasting models will also produce ridership forecasts for transit). Some freight models rely on methods like passenger models, but freight-specific methods are also currently in use. Travel forecasting methods also vary by the type of agency using them and the purpose (i.e., methods for statewide, regional, local, or project-specific planning purposes or for transit or toll agencies):  Metropolitan-area models tend to be more complex, representing multiple modes, more- complex policy needs, and congestion problems.  Statewide models range from sketch models to advanced models, depending on policy needs.

Final Report Project No. 08-94 24  Project-level forecasting ranges from site-specific approaches to full-scale travel models, depending on project size and complexity. Transit and toll agencies favor methods that are more detailed in their representation of transit or tolling, but include much less or no detail on other types of travel. Many of these methods are similar, but there are often specialized methods or tools identified in the literature. The brevity of this literature review necessitated highlighting literature in each area without detailing individual methods. Travel models are typically developed to meet specific policy needs—as these needs become more complex, so do the models needed to support them. The following is a summary of model types and uses:  4-step models are useful for many transit and highway forecasting situations.  AB models are useful for evaluating pricing, greenhouse gas emissions, peak spreading, and where non-home-based travel effects are significant.  Transit passenger models are an element of passenger models, or developed separately using methods focused on transit demand.  Freight models are useful where trucks are an important component of travel and can be scaled to match policy needs. From the brief literature review, several key conclusions can be made about the state-of-the- practice in analysis and simulation:  Static equilibrium assignment models are prevalent and used for macroscopic applications.  DTA models are used more extensively on projects where the traffic operational effects of alternative network and demand management schemes, such as pricing and work zones, are evaluated at a network level. DTA models are often built for a subarea of a regional travel demand model. The O-D table is a fundamental input to a DTA model. DTA models are also useful for assessing the queuing, route assignment, and travel time reliability of several alternatives. DTA models can be used for macroscopic or mesoscopic applications.  Microscopic modeling should only be utilized when sufficient time and effort can be applied toward model development and validation. Many MPOs and state DOTs have found that microscopic simulation provides useful and detailed results regarding vehicle interactions; however, the applications may become more limited due to time-consuming procedures, and MPOs and other transportation agencies are continuing to assess whether an increase in detailed output is worth the effort. To date, the use of integrated, multiresolution models has not been prevalent within MPOs, state DOTs, and other transportation agencies; however, these methods hold significant potential for improving current industry practices. 2(B) SOFTWARE DESIGN The software design focused on framing the questions a travel forecasting practitioner must answer when considering which methods to apply for a specific planning activity. That said, the

Final Report Project No. 08-94 25 software design accommodates schedule and budget constraints and other requirements for defining the most appropriate methods to answer these planning questions. An agency seeking guidance and recommendations for how best to analyze and evaluate a transportation program, plan, or policy would be the target group for these questions. A spirited back-and-forth discussion between a stakeholder and an “expert” is not possible since the flow of information will be mostly in a single direction from the user to the guidance. With these considerations in mind, the research team’s conceptualized guidance asks the following five questions:  What planning program or plan do you need travel forecasts for? − The user identifies the program, plan, or policy for which she or he is seeking guidance on technical methods and tools.  What are the requirements for the work? − The user identifies the level of detail, regulations, sensitivities, and markets needed for the planning exercise. This defines the scope of the technical methods that are needed.  What performance measures are important for evaluation? − The user identifies the performance measures that should be considered for the evaluation of the plan, program, or policy under consideration.  What are the constraints for the work? − The user identifies the budget and schedule constraints for developing, implementing, and validating the methods.  What current methods and resources are available? − The user identifies methods, expertise, hardware, software, and data that the agency already has that can be used to support the work. These questions frame the five types of inputs that the decision-support system uses to produce recommendations. The software design also adopted the following assumptions:  Each user is only selecting a single planning program or plan to evaluate at one time. (Multiple selection was considered as an enhancement, but this was found to add too many complexities for the first draft.)  The requirements and constraints are used to tailor methods presented to those that best fit the user’s situation.  The performance metrics of interest to the agency and necessary to evaluate the plan, program, or policy under consideration provide additional requirements, allowing the methods presented for consideration to be further refined.  The information about the agency’s current analytical methods and resources is optional. If the agency includes their current methods, then the recommended methods are considered as enhancements over the current methods (and do not include any cost or benefits for these current methods).

Final Report Project No. 08-94 26  The recommended methods are presented as individual methods or packages of methods, as appropriate and to meet the needs of the planning program or plan.  Data, hardware, software, and staff expertise required to support the recommended methods are identified. If any of these resources are identified as available within the current resources, then they are not included in this list.  Methods are prioritized according to how well they meet the requirements, constraints, and performance measures. The prioritization process includes an assessment of costs and benefits for each method. The prioritization process is presented with a list of possible options for the user to choose from. The user can identify the importance of various requirements, including budget and schedule, so the recommendations can reflect these. Information on each method is available as background material whenever a method is presented for consideration. Method Selection Components Step 1—Planning Programs and Plans In the first step, the user identifies the planning program or plan of interest to her or his agency or organization. These transportation-related policies, programs, plans, or investment projects address one or more planning issues or questions. Programs and plans include five major categories. Table 2 presents the major categories with planning programs and plans within each category. TABLE 2: STEP 1—PLANNING PROGRAMS AND PLANS TRANSPORTATION ENVIRONMENT  Bike/Pedestrian Capital Investments  Comprehensive Plans  Highway Detailed Design  Highway Preliminary Engineering  Intelligent Transportation Systems Plan  Travel Demand Management Program  Traffic Impact Study  Transit Operations Study  Transit-Oriented Development Study  Long-Range Transportation Plan  Transportation Improvement Program  Freight Plan  Major Highway Corridor Study  Major Transit Corridor Study  Bicycle and Pedestrian Plan  Freeway Operations and Management Study  Arterial Operations and Management Study  Congestion Management Plan  Pricing Study (Tolls, Fees, Fares, Gas, Parking)  Project Prioritization  Air Quality Emissions Inventory for Conformity Analysis  Energy Use Study  Greenhouse Gas Mitigation Study  Sustainable Community Strategies  Environmental Clearance and Preliminary Design for Transportation Projects ECONOMY  Economic Development Plan  Economic Impact Analysis SAFETY AND HEALTH  Health and Physical Activity Plan  Emergency Evacuation Plan  Safety Program EQUITY  Environmental Justice Plan Major categories list specific programs and plans and only one program or plan can be selected at a time. These categories do not limit the performance metrics or alternatives that can be tested, but identify the primary use of the results. For example, a Long-Range Transportation Plan is a planning program in the Transportation category, but this program will have environmental,

Final Report Project No. 08-94 27 land-use, economic, transportation, quality-of-life, and equity performance measures. The mappings, equivalencies, and relationships between programs and plans and performance measures are included in the guidance. If an agency seeks to simultaneously evaluate methods for multiple planning programs or plans, then it will need to evaluate each planning program or plan separately and then identify methods that can support multiple purposes (and therefore provide additional benefit for the same cost). The ability to evaluate multiple planning programs or plans together is considered as a future enhancement in Chapter 4. Step 2—Planning Context In the second step, the user will provide details regarding the overall planning context of the program or plan for which they are seeking analytical method recommendations. The user will also provide additional details about requirements, such as the level of detail, the agency’s planning area, regulations that may apply and the sensitivities and travel markets that the model should represent. To help the user fully understand the planning context, the software tool will evaluate the requirements, analytical rigor, and sensitivities necessary to address the planning issue and question. Table 3 presents the planning context categories and elements within each category, along with one sample question associated with each element. TABLE 3: STEP 2—CATEGORIES AND ELEMENTS OF THE PLANNING CONTEXT CATEGORY ELEMENT QUESTION GENERAL Regulation Are there regulatory or other requirements? Analytical Rigor Is this a feasibility study, a planning study, or an investment- grade analysis? LEVEL OF DETAIL Demographics What is the desired level of demographic detail? Geography What is the geographic area to be studied? Networks What is the desired level of network detail? Spatial Detail What is the desired level of spatial detail? Temporal Detail What is the desired level of temporal detail? SCOPE Pricing How does pricing affect user choices? Sensitivities What factors must the analysis be sensitive to? Traveler What travelers are affected? The research team programmed the software tool with the universe of possible responses to each question pertaining to these requirements. The user is required to input the population of the region or state of interest to scale responses where appropriate.

Final Report Project No. 08-94 28 Step 3—Performance Measures In the third step, the user chooses performance measures of interest to her or his agency. The performance measures should be those limited to and associated with the planning program and plan for which she or he is seeking analytical method recommendations. Performance measures have increased considerably in the past 10 to 20 years, and there are now hundreds of different performance measures in use across the United States. This discussion provides an overview of the distinct types of performance measures and their relation to travel demand modeling. Many performance measures are reported as part of a monitoring process (Florida Department of Transportation, 2014; Washington State Department of Transportation, 2016); not all are amenable to forecasting. Hence, the discussion is limited to those performance measures that would be or could be used in conjunction with forecasts from travel demand models. The guidance includes six categories of transportation planning performance measures. Table 4 presents the performance measure categories and measures, along with metrics for each measure. TABLE 4: STEP 3—PERFORMANCE MEASURES AND METRICS CATEGORY MEASURES METRICS PASSENGER TRANSPORT MEASURES  Mode Shares  Demand  Traffic Measures  Transit Measures  Mobility  Auto Congestion  Person miles traveled by mode  Number of person trips  Highway level-of-service  VMT  Auto speeds  Auto volumes  Transit Ridership  Passenger flows  Person delay hours  Congested miles ENVIRONMENTAL MEASURES  Air Quality  Energy Consumption  Noise  Criteria pollutant emissions  Greenhouse gas emissions  Fuel consumption  Time above certain levels ECONOMIC MEASURES  User Benefits  Revenues  Economic Development  Accessibility  Value of time saved  Fares  Tolls  Access to jobs in cluster employment  Access to jobs

Final Report Project No. 08-94 29 CATEGORY MEASURES METRICS FREIGHT TRANSPORT MEASURES  Mobility  Demand  Mode Shares  Traffic Measures  Congestion  Commodity flows  Number of commercial vehicle trips  Freight miles traveled by mode  Freight travel times  Truck volumes  Truck speeds  Truck delay hours COMMUNITY MEASURES  Land Use  Schools  Environmental Justice  Social Equity  Access to services  Access to schools  Travel costs for minorities and low- income populations  Travel costs by income group HEALTH AND SAFETY MEASURES  Active Transportation  Safety  Bike miles  Nonmotorized volumes  Number of vehicle collisions Step 4—Constraints Budget and schedule are two crucial factors that constrain the possibilities for developing any new method. Budget and schedule are constraints in TFGuide and present opportunities for information on the cost and timeline for implementing a method (or package of methods). The software tool compares the estimated cost and timeline with the user inputs on budget and schedule to include as part of the prioritization of the method recommendations. This comparison assigns a higher rank to methods that fall within (or close to) the budget and schedule constraints. Step 5—Current Agency Methods In the fourth step of the software tool, the user can choose to provide background information about the agency and—more importantly—details regarding their current analytical capabilities in the realm of transportation planning, modeling, and simulation. This portion of the software tool is optional; this is to lessen the burden on users and to allow users to evaluate the recommendations from a blank slate. The assessment of current agency methods will help identify an agency’s current analytical capabilities; this will allow the software tool to provide two valuable services to the end user:  Present recommendations unique and specific to each individual agency given the tools and methods currently at their disposal.  Present an improvement plan that would help the agency transition from older practices to newer and improved methods in a targeted and stepwise manner.

Final Report Project No. 08-94 30 Having an agency answer additional questions about their current methods allows the tool to provide more-customized responses, which contrasts with developing an easier-to-use tool that would provide more generic responses. Users may choose to run the software tool without any background on their current methods and then refine it to produce a set of recommended methods that are enhancements of current methods, without suggesting major model development efforts that might not be appropriate in the context of an agency’s existing methods. Step 6—Menu of Recommended Methods Figure 5 presents the list of methods that are included in the guidance. These methods include passenger travel demand model components, other travel demand model methods, commercial vehicle travel demand model components, and assignment and simulation methods. The research team considered a wider range of methods during the development of the software design. The research team could add these additional methods (presented in Chapter 4) to future versions of the guidance, but these were not the focus of this initial effort. The guidance will include information on which method is appropriate for each program (Table 2), planning context (Table 3), and performance measure (Table 4) element. For example, traffic assignment methods would be recommended for a traffic impact study, but would not be appropriate for a health and physical activity plan. In the guidance, there may be a wide range of methods that are possible for each program or plan; the software tool will narrow these down using the information provided by the user in the requirements and selected performance metrics. For example, in a transit corridor study, the user may choose to select highway network supply as a sensitivity factor, in which traffic assignment methods would be listed as options, but if the user chose transit network supply as a sensitivity factor, then transit assignment methods would be listed as an option. Some of the methods may require data or other resources to implement. The software tool will identify these resources and the cost of obtaining these resources.

F F inal Report IGURE 5: STEP 6 Passenger Tr Compo • Aggregate Mod • Trip Production • Trip Production • Trip Attraction • Trip Distribution • Mode Choice ( • Time of Day (F • Disaggregate M • Population Syn • Daily Activity P • Tour Frequency • Destination Ch • Mode Choice M • Time of Day Ch • Parking Locatio • Transit Pass O —LIST OF METHO avel Demand nents els (Cross-Class) (Regression) (Regression) (Gravity) Fixed Factors ) ixed Factors) odels thesis attern Model Model oice Model odel oice Model n Choice wnership DS BY CATEGO Other T M • Direct Dem • Origin-Dest Estimation • Pivot-Point Model RY ravel Demand ethods and ination Matrix and Incremental 31 Comm De • Aggre • Trip P • Trip A • Trip D • Mode • Time • Comm • Disagg • Firm S • Supp • Supp • Mode • Vehic Patte • Freigh Stops • Stop ercial Vehicle T mand Compone gate Models roduction (Cross-C ttraction (Regress istribution (Gravity Choice (Fixed Fac of Day (Fixed Fact odity Flow Models regate Models ynthesis lier Selection Mode ly Chain Model and Shipment Siz le Type and Tour rns t and Services To Sequence and Dur ravel nts lass) ion) ) tors) ors) l e urs and ation A • S A • M A • S A • V • R • In • F A • S A • D • D T • T • In M • P Project No ssignment and S ingle-Class Equilib ssignment ulti-Class Equilibriu ssignment tochastic User Equ ssignment olume-Delay Funct oute-Path Choice M tersection Delay requency-based Tr ssignment chedule-based Tra ssignment ynamic Traffic Ass ynamic Capacity-C ransit Assignment raffic Microsimulati tegrated, Multi-res odel edestrian/Bicycle S . 08-94 imulation rium Traffic m Traffic ilibrium ions odel ansit nsit ignment onstrained on olution imulation

Final Report Project No. 08-94 32 The outcome of the guidance is a menu of methods recommendations. The word “menu” was chosen as several individual enhancements to methods may be possible within an agency’s available budget or schedule. The menu presents users with methods’ advantages and disadvantages to consider. Multiple methods may be employed in sequence or in parallel. The guidance produces information on these costs and benefits for each method (or package of methods). The software tool allows the user to adjust the requirements and performance measure inputs to achieve a different menu of recommendations; specifically, this process redefines the requirements and constraints that the menu or recommendations are based on. The following items detail the potential benefits of a specific method:  Performance Measures. Specific performance metrics produced by an individual method are provided, since not all methods can produce all required performance metrics.  Sensitivities. If the user has selected factors that the methods should be sensitive to, then the software tool provides additional enhancement details for methods that are sensitive to these factors.  Spatial Detail. The spatial detail is an important element that goes into selecting a method, since more detailed methods often take longer to develop and cost more. Some methods are focused and provide sufficient spatial detail without additional time or cost, but with limited other benefits.  Temporal Detail. The temporal detail is important for certain programs or projects and will be explicitly defined for each method. The software tool suggests multiple enhancements; these can be customized by adding more requirements or weighting the requirements to consider the priorities of these requirements. The following items detail the costs of a specific method:  Data Needs. Many types of data can be obtained from available sources (secondary data collection) or collected directly (primary data collection) to support the development, calibration, validation, or forecasting aspects of analytical tools. This research focused on the analytical methods, so this assessment was limited to required data needs.  Staff Training and Expertise. The software tool provides information on staff training needs, which may include courses, seminars, webinars, or guidance for specific methods.  Industry Adoption. The guidance includes information on how well accepted and how well validated a method is. Newer methods and tools may have new capabilities or sensitivities, but older tools may have undergone more rigorous testing and use.  Cost and Time. Cost and time are approximated so that individual methods can be combined into a model improvement plan. Costs and times will have ranges since these depend on the details of the application of a specific method. The results should be useful for order-of-magnitude decisions regarding the methods, but the results should not be used without review as costs can vary. The method selection guidance does not evaluate individual data needs for each method. This could be a future enhancement to the method selection tool, but adding this function was not possible within the scope and budget confines of this project. Instead, the guidance identifies

Final Report Project No. 08-94 33 data needs in broad terms and includes a cost needed to acquire these data if the agency does not currently have access to these data sources. Collecting, obtaining, or processing data for use in travel forecasting is nuanced, so these cost and time requirements are solely for estimation purposes. The data sources are scaled to the size of the region or state if the cost for these data are dependent on the population. Available budget and schedule are two primary constraints. In the software design, these constraints are defined separately from the rest of the requirements so they can be adjusted in real time to produce different recommendations. Individual methods are reported with ranges of budget or schedule expected for each method and combined to produce a total range for budget and schedule that can be compared to the specified cost and schedule. The software tool applies a score to each element, with higher scores for higher benefits and lower costs, as shown in Table 5. Costs and benefits are most useful when they can be combined to produce a cumulative score. This requires quantifying the value of each cost or benefit via subjective weighting provided by the user, allowing the user to determine if each benefit or cost has the same value or if some benefits or costs are valued more or less than others. For example, one agency may determine that budget and time are more important than performance measures or sensitivities and weight these factors accordingly. TABLE 5: COST AND BENEFIT SCORES COSTS AND BENEFITS SCORES PERFORMANCE MEASURES One point for each performance metric that the method can produce. SENSITIVITIES One point for each sensitivity factor that the method can address. REQUIREMENTS One point for each requirement that the method can address. INDUSTRY ADOPTION Two points if the method is proven and validated; one point if the method is somewhat proven. COST Ratio of budget to cost required (rounded). Cost based on the highest part of the range. TIMEFRAME Ratio of schedule to timeframe required (rounded). Timeframe based on the highest part of the range. Software Development The software design included the primary elements and relationships among these elements for the application architecture. In addition, the database schema and graphical user interface (GUI) was established along with an identification of output results to complete the software tool design. Database Schema Figure 6 presents the decision-engine entity relationship (ER) diagram for the software tool.

F Method Sele IGURE 6: DECISI ction for Tra ON-ENGINE ER D vel Forecasti IAGRAM ng 34 Project No. 08-94

Final Report Project No. 08-94 35 This diagram represented in Figure 6 comprises four parts:  User Inputs. The User Inputs data tables store the dynamic data of individual travel forecasting practitioners using the software tool: their login credentials and the inputs that they provide to specify the specific planning needs for which the software tool will provide method selection options. These inputs are chosen by the user from selection sets of enumerated options stored in the Domain Taxonomy database.  Domain Taxonomy. The Domain Taxonomy data tables codify a preenumerated set of input parameters that allow the user to specify the problem domain for which the method selection tool will find solutions. Specifically, the set of Planning Programs and Plans the system recognizes, the characteristics of the Planning Context, and the various Performance Measures of interest to the user or planning agency. Although the domain knowledge codified in these tables is static with respect to a user analysis, the Domain Taxonomy is data driven and extensible over the life cycle of the system. The Domain Taxonomy is associated with Requirements and Constraints via several association tables that identify which specific detailed Requirements and Constraints are driven by the user’s Program, Metrics, and Contexts. The end user’s set of inputs creates a list of Requirements and Constraints derived from the domain knowledge codified in the Domain Taxonomy and its association “driver” tables.  Methods and Parameters. The Methods and Parameters data tables codify the preenumerated characteristics of the solution set of Methods and their parameters. These parameters are associated with Requirements and Constraints via association tables that identify the set of requirements that can be satisfied by those parameters, and the constraints that may be influenced by them.  Requirements and Constraints. Requirements and Constraints are the common currency that permits linking of the user-specified problem domain to the set of Methods chosen from the range of known methods to be proposed as possible solutions by the method selection tool. Although represented in the schema as simple atomic entities for simplicity of presentation, the Requirement and Constraint entities may be annotated to support weighting and costing of proposed solution sets. The beginning of this chapter describes the four user inputs in the decision-engine ER diagram and the attributes in each of these user inputs:  UserScenario_PlanningProgram, detailed in Table 2 on page 26.  UserScenario_PlanningContextResponse, detailed in Table 3 on page 27.  UserScenario_PerformanceMetric, detailed in Table 4 on page 28.  UserScenario_Method, detailed in Figure 5 on page 31. The method that is an output of the decision-engine ER diagram has the same list of methods (Figure 5) as the UserScenario_Method that is input; the difference between the user method and the output method is that users have the option to input the methods they are already using. The research team used these details from the software design to build the database schema.

Final Report Project No. 08-94 36 Application Architecture The research team implemented the software tool as a browser-based web application that runs on modern web browsers (e.g., Chrome, Firefox, Internet Explorer, Safari) with cookies and JavaScript enabled. The research team implemented the web application using a customizable web application framework with a relational database back-end store. The application can run on a standard Linux distribution server, with the target production platform being a full-service, cloud-based service provider with streamlined maintenance and system administration. The web user interface for tool configuration (specification of Program, Requirements, Performance Metrics, and Constraints) and result set display and evaluation (Recommendations) uses JavaScript to provide interactive and graphical features. The research team used JQuery, a client-side framework, to support JavaScript coding. The provisional web application framework is Django/Python with a PostgreSQL database. PostgreSQL is a robust, scalable relational database management system. The standard Django Object-Relational Mapping (ORM) tool was used to mediate access to the database for a substantial portion of the application interface, and raw PostgreSQL queries were used when the complexity of the logic exceeded Django’s ORM. Graphical User Interface The research team discussed the GUI as part of the software tool design to highlight some aspects of the tool that could be beneficial for users. One option includes separating mutually exclusive options in the menu of recommendations. The research team envisions a combination of other travel demand methods that are mutually exclusive from a combination of travel demand model component methods, or assignment methods that could be mutually exclusive or combined with other assignment methods. Another concept for the GUI involves setting a real-time adjustment for the constraints (i.e., budget and time/schedule) in the resource needs. These are buttons on the interface to demonstrate that users can increase or decrease the available budget and schedule, resulting in a different menu of recommendations that best fit these resource constraints. The user interface also includes a presentation of data needs and a breakdown of the cost for each element. The guidance in this research does not discuss data needs in detail, but does identify data needs for various methods and includes costs for data collection if the agency does not already have these data. The costs include data collection, model development, calibration and validation, and training. Reporting Results The software tool reports results for each method selection recommendation. This report will contain all the assumptions for a specific method selection and the menu of options, with scoring included. This is described earlier in this chapter for the method selection components. The reporting function allows users to compare different sets of assumptions and outputs. The report also includes the weights provided by the user and the scores developed for each cost or benefit included in the method selection process. The weights provide a transparent consideration of priorities for agencies to consider when reviewing the scores of each method.

Final Re Final Us Wirefram The best wizard— menu of into this functiona wirefram challenge the range software paid to th Page elem breadcrum color, fon ultimate “recomm FIGURE 7 Wirefram iterations phase. Co port er Functio es description an online to options or o early prototy lity. Follow ing. Wirefra s. As a resu of content a developmen e what (con ents and in b navigatio t, and butto behavior thr endations” p : EXAMPLE W ing was com . The team c nsideration nality of T for the type ol where use utputs at the pe web app ing this initi ming often lt, the resear nd back-en t projects w tent), the wh teraction—s n—and loc n design can ough annota age is captu IREFRAME A pleted for onsidered p s included th FGuide of web appl rs are seque end. The ba lication and al prototypi precedes pro ch team dec d data struct here a team ere (layout) uch as mou ation are typ be made la tion. A simp red in Figu ND NOTES FO each page of age function e following 37 ication need ntially guid sic function several itera ng, the proje totyping, bu ided to prot ures. Wirefr designs a si , and the ho se-hover beh ically speci ter, but the w lified exam re 7. R THE MENU the agency ality and th : ed was dete ed through t ality of this tions establ ct team ent t this tool p otype first t aming is a s te blueprint w (interactio avior, acco fied at this s ireframe sh ple of wiref OR RECOMM selection si e user exper Pro rmined to re he input ste wizard con ished the on ered a phase resented un o become co pecific step or schemati n/behavior rdion tables tage. Final c ould reflec raming the ENDATIONS te; this requ ience durin ject No. 0 semble a ps to reach a cept was bui -screen of site ique design mfortable w in many c, with atten ) for each pa , and hanges like t the page’s PAGE ired several g this design 8-94 lt ith tion ge.

Final Re 1. H 2. H pr 3. H ef 4. H Compreh A full, co wirefram designs i FIGURE 8 User Log Users can version o The adm administr for plann port ow weightin ow high-lev ocess they w ow a user co fects on the ow the tool ensive Lay mprehensiv ing process nto a functio : FINAL COMP in and Lan reach the T f TFGuide w inistrator co ative contro ing guidanc g should be el navigatio ere. uld return t menu of rec could provi outs e layout—w was comple nal web app REHENSIVE ding Page FGuide log ill need an ntrols user p l; the ability e. applied and n could allo o prior steps ommendati de feedback ith the corre ted (Figure lication. LAYOUT DES in page via w account; thi ermissions to create an 38 adjusted by w users to e and make i ons. if a user fai ct colors an 8). The softw IGN OF MENU eb browse s can be cre and can gran d edit techn users. asily see wh nput adjustm ls to enter a d logo—wa are engine OF RECOMM r (Figure 9). ated by a sit t several ca ical content Pro ere in the m ents while required fie s achieved o ering team t ENDATIONS All users o e administra pabilities, in ; the ability ject No. 0 ethod select seeing the ld. nce the hen turned t PAGE f the final tor on reque cluding: ful to use the to 8-94 ion hese st. l ol

Final Re FIGURE 9 After site either beg Previous FIGURE 10 The follo Profile A user sh profile id of resour presents port : USER LOGIN authenticat in a new m session wor : LANDING P wing section ould set a p entifies the ces that are the cost adju SCREEN ion, the agen ethod select k is automat AGE AFTER U s assume a rofile for the population o directly or in stment facto cy user arri ion session ically saved SER AUTHE new scenari region of in f the region directly rel rs for each 39 ves at the si (scenario) or and presen NTICATION o has been s terest befor or state; thi ated to the s population r te’s landing edit/delete ted in the tab tarted and d e beginning s informatio ize of the re ange. Pro page (Figur a preexistin le shown in escribe func work on a n is used to gion or state ject No. 0 e 10) and ca g scenario. Figure 10. tionality. scenario. Th adjust the c . Table 6 8-94 n e ost

Final Report Project No. 08-94 40 TABLE 6: REGION OR STATE POPULATION ADJUSTMENT FACTORS REGION OR STATE POPULATION LOWER LIMIT UPPER LIMIT COST ADJUSTMENT Over 25,800,000 Population 25,800,000 30,900,000 2.7 18,900,000 to 25,800,000 Population 18,900,000 25,800,000 2.6 15,800,000 to 18,900,000 Population 15,800,000 19,800,000 2.5 10,900,000 to 15,800,000 Population 10,900,000 15,800,000 2.4 9,800,000 to 10,900,000 Population 9,800,000 10,900,000 2.3 6,900,000 to 9,800,000 Population 6,900,000 9,800,000 2.2 5,800,000 to 6,900,000 Population 5,800,000 6,900,000 2.1 4,900,000 to 5,800,000 Population 4,900,000 5,800,000 2.0 4,000,000 to 4,900,000 Population 4,000,000 4,900,000 1.9 3,300,000 to 4,000,000 Population 3,300,000 4,000,000 1.8 2,600,000 to 3,300,000 Population 2,600,000 3,300,000 1.7 2,100,000 to 2,600,000 Population 2,100,000 2,600,000 1.6 1,600,000 to 2,100,000 Population 1,600,000 2,100,000 1.5 1,300,000 to 1,600,000 Population 1,300,000 1,600,000 1.4 1,000,000 to 1,300,000 Population 1,000,000 1,300,000 1.3 900,000 to 1,000,000 Population 900,000 1,000,000 1.2 800,000 to 900,000 Population 800,000 900,000 1.1 700,000 to 800,000 Population 700,000 800,000 1.0 600,000 to 700,000 Population 600,000 700,000 0.9 500,000 to 600,000 Population 500,000 600,000 0.8 400,000 to 500,000 Population 400,000 500,000 0.7 300,000 to 400,000 Population 300,000 400,000 0.6 200,000 to 300,000 Population 200,000 300,000 0.5 100,000 to 200,000 Population 100,000 200,000 0.4 50,000 to 100,000 Population 50,000 100,000 0.3 Less than 50,000 Population 10,000 50,000 0.2