National Academies Press: OpenBook
Page 1
Suggested Citation:"May 10, 2004." National Academies of Sciences, Engineering, and Medicine. 2004. Review of Travel Demand Modeling by the Metropolitan Washington Council of Governments: Second Letter Report. Washington, DC: The National Academies Press. doi: 10.17226/22067.
×
Page1
Page 2
Suggested Citation:"May 10, 2004." National Academies of Sciences, Engineering, and Medicine. 2004. Review of Travel Demand Modeling by the Metropolitan Washington Council of Governments: Second Letter Report. Washington, DC: The National Academies Press. doi: 10.17226/22067.
×
Page2
Page 3
Suggested Citation:"May 10, 2004." National Academies of Sciences, Engineering, and Medicine. 2004. Review of Travel Demand Modeling by the Metropolitan Washington Council of Governments: Second Letter Report. Washington, DC: The National Academies Press. doi: 10.17226/22067.
×
Page3
Page 4
Suggested Citation:"May 10, 2004." National Academies of Sciences, Engineering, and Medicine. 2004. Review of Travel Demand Modeling by the Metropolitan Washington Council of Governments: Second Letter Report. Washington, DC: The National Academies Press. doi: 10.17226/22067.
×
Page4
Page 5
Suggested Citation:"May 10, 2004." National Academies of Sciences, Engineering, and Medicine. 2004. Review of Travel Demand Modeling by the Metropolitan Washington Council of Governments: Second Letter Report. Washington, DC: The National Academies Press. doi: 10.17226/22067.
×
Page5
Page 6
Suggested Citation:"May 10, 2004." National Academies of Sciences, Engineering, and Medicine. 2004. Review of Travel Demand Modeling by the Metropolitan Washington Council of Governments: Second Letter Report. Washington, DC: The National Academies Press. doi: 10.17226/22067.
×
Page6
Page 7
Suggested Citation:"May 10, 2004." National Academies of Sciences, Engineering, and Medicine. 2004. Review of Travel Demand Modeling by the Metropolitan Washington Council of Governments: Second Letter Report. Washington, DC: The National Academies Press. doi: 10.17226/22067.
×
Page7
Page 8
Suggested Citation:"May 10, 2004." National Academies of Sciences, Engineering, and Medicine. 2004. Review of Travel Demand Modeling by the Metropolitan Washington Council of Governments: Second Letter Report. Washington, DC: The National Academies Press. doi: 10.17226/22067.
×
Page8
Page 9
Suggested Citation:"May 10, 2004." National Academies of Sciences, Engineering, and Medicine. 2004. Review of Travel Demand Modeling by the Metropolitan Washington Council of Governments: Second Letter Report. Washington, DC: The National Academies Press. doi: 10.17226/22067.
×
Page9
Page 10
Suggested Citation:"May 10, 2004." National Academies of Sciences, Engineering, and Medicine. 2004. Review of Travel Demand Modeling by the Metropolitan Washington Council of Governments: Second Letter Report. Washington, DC: The National Academies Press. doi: 10.17226/22067.
×
Page10
Page 11
Suggested Citation:"May 10, 2004." National Academies of Sciences, Engineering, and Medicine. 2004. Review of Travel Demand Modeling by the Metropolitan Washington Council of Governments: Second Letter Report. Washington, DC: The National Academies Press. doi: 10.17226/22067.
×
Page11
Page 12
Suggested Citation:"May 10, 2004." National Academies of Sciences, Engineering, and Medicine. 2004. Review of Travel Demand Modeling by the Metropolitan Washington Council of Governments: Second Letter Report. Washington, DC: The National Academies Press. doi: 10.17226/22067.
×
Page12
Page 13
Suggested Citation:"May 10, 2004." National Academies of Sciences, Engineering, and Medicine. 2004. Review of Travel Demand Modeling by the Metropolitan Washington Council of Governments: Second Letter Report. Washington, DC: The National Academies Press. doi: 10.17226/22067.
×
Page13
Page 14
Suggested Citation:"May 10, 2004." National Academies of Sciences, Engineering, and Medicine. 2004. Review of Travel Demand Modeling by the Metropolitan Washington Council of Governments: Second Letter Report. Washington, DC: The National Academies Press. doi: 10.17226/22067.
×
Page14
Page 15
Suggested Citation:"May 10, 2004." National Academies of Sciences, Engineering, and Medicine. 2004. Review of Travel Demand Modeling by the Metropolitan Washington Council of Governments: Second Letter Report. Washington, DC: The National Academies Press. doi: 10.17226/22067.
×
Page15
Page 16
Suggested Citation:"May 10, 2004." National Academies of Sciences, Engineering, and Medicine. 2004. Review of Travel Demand Modeling by the Metropolitan Washington Council of Governments: Second Letter Report. Washington, DC: The National Academies Press. doi: 10.17226/22067.
×
Page16

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.

Page 1 of 23 May 10, 2004 The Honorable Christopher Zimmerman Chairman National Capital Region Transportation Planning Board Metropolitan Washington Council of Governments 777 North Capitol Street, NE, Suite 300 Washington, DC 20002 Dear Chairman Zimmerman: This letter is the second report of the Transportation Research Board’s (TRB’s) Committee for Review of Travel Demand Modeling by the Metropolitan Washington Council of Governments (MWCOG). The committee, which was appointed by the National Research Council to undertake this review, includes scholars and practitioners who collectively are familiar with metropolitan planning organization (MPO) modeling practices in many areas of the country. The committee’s membership is presented in Attachment 1.1 In a letter of May 8, 2002, Mr. Phil Mendelson, acting as Chairman of the National Capital Region Transportation Planning Board (TPB, a constituent unit of MWCOG), requested that TRB undertake this study. That request describes the present study as part of TPB’s ongoing program to upgrade its travel forecasting methods and respond to federal guidance on modeling in air quality nonattainment areas. The specific scope for that study is described in the Statement of Task approved by the Governing Board of the National Research Council on October 9, 2002. The Statement of Task specifies that the committee will “perform review of the state of the practice of travel demand modeling by the Transportation Planning Board (TPB) of the Metropolitan Washington Council of Governments.” The Statement of Task is presented in Attachment 2. The committee undertook to provide guidance on five specific elements listed in this statement of task: ƒ The performance of TPB’s latest travel model (Version 2) in forecasting regional travel, ƒ The proposed process for merging the latest travel model outputs to produce mobile source emissions, ƒ TPB’s proposed direction of future travel demand model upgrades, ƒ Travel survey and other data needed to accomplish future model upgrades, and ƒ The detail (grain) of travel analysis zones that should be developed for future upgrades. 1 Biographical information is available at http://www4.nas.edu/webcr.nsf/CommitteeDisplay/SAIS-P-02-07- A?OpenDocument.

Page 2 of 23 The committee’s initial letter report, dated September 8, 2003, addressed the first two items.2 This second letter addresses the last three items and other matters raised in a work program document that TPB presented to the committee. That document is included here as Attachment 4.3 In the course of preparing this report, committee members met as a group several times by teleconference and held a face-to-face meeting and teleconferences with TPB staff. The committee’s observations presented in this letter report have been reviewed independently under the procedures of the National Research Council by individuals selected for their diverse perspectives and technical expertise. We begin this letter as we did our first report, by stating our principal observations. We then proceed to explain the background and bases for these observations and other comments the committee offers to assist TPB. Principal Observations As we noted in our first letter, despite some four decades of experience with the use of travel demand models in transportation planning, there are few universally accepted guidelines or standards of practice for these models or their application. Any assessment of these models, their performance, and the current state of transportation demand modeling practice relies primarily on professional experience and judgment. Nevertheless, most transportation professionals will agree that improvements in demand modeling have been and continue to be made. These improvements have come from at least three sources. First, we have enhanced our understanding of factors that influence travel demand and have been able to apply that understanding to formulate models that represent the relationships among these factors in more realistic ways. Second, the power and flexibility of computers available for modeling have increased dramatically at the same time that their cost has fallen, allowing us to apply computationally more complex and data-intensive modeling methods. Finally, we have substantially enhanced our capabilities for data collection and database management. The state of transportation modeling practice today is marked by continuing evolutionary modification of the widely applied “four-step model” and development of revolutionary new models that are beginning to be applied in analyses of transportation policy matters. New “activity-based” or “tour-based” models and microsimulation of traffic flows using less aggregated data and providing more spatial and temporal detail promise substantial improvements in our ability to account for the influence of such factors as time of day, congestion, and travelers’ socioeconomic characteristics. These new models will improve our ability to represent the influence of investment decisions, traffic control strategies, and other public policy variables on regional travel patterns, which will in turn enhance our ability to 2 The committee’s first letter report is available at http://trb.org/publications/reports/mwcogsept03.pdf. The second report will be available at http://trb.org/publications/reports/mwcogapril04.pdf. 3 The document is a staff proposal and not an officially approved TPB work program.

Page 3 of 23 estimate air pollution emissions and other aspects of the region’s well-being. These new models, the committee agrees, are likely to be used widely within the next decade. Several MPOs have begun to adopt the new formulations. While the capabilities of the new models are being perfected and their application grows more widespread, many agencies may continue to rely on four-step modeling and seek to improve their existing modeling practices. An important implication of this situation is that MPOs must allocate their resources to strike an appropriate balance between maintaining their current models and preparing for the migration to new practices. That balance will change as time passes. Within this context, the committee makes the following points: 1. TPB’s proposal to develop a comparative analysis of modeling practices employed by other MPOs with similar characteristics may be useful but will be a challenging undertaking. (page 5) 2. TPB’s efforts to improve model calibration and validation statistics through improved representation of transit and highway network supply characteristics—such as refinements of volume-delay functions, free-flow speed and capacity values, linkages of transit speeds to highway speeds, and network coding—are steps in the right direction. (page 6) 3. TPB’s proposals to develop new model components to represent truck and commercial vehicle trips are a reasonable application of methods successfully adopted by other MPOs. The committee suggests that collection of new traffic classification counts begin as soon as possible and that model development work be initiated without waiting for completion of these counts. (page 7) 4. The committee is encouraged by TPB’s plan to work with the region’s transit agency and others to find a method to represent bus speeds in future years. (page 8) 5. The committee recognizes that the practice of using K-factors and other arithmetic adjustments to improve four-step models’ ability to represent base-year travel observations is not uncommon, but continues to find TPB’s use of such correction factors to be excessive. TPB should proceed aggressively with its plan to document the logical basis and need for these adjustments. (page 8) 6. Having previously questioned TPB’s feedback procedures to estimate transit and highway network travel times, the committee agrees that TPB’s proposed exploration of alternatives to its rule-based heuristic approach for approximating equilibrium conditions and current representation of highway-transit composite times in distribution and mode choice is helpful. The committee notes that there are accepted feedback algorithms for obtaining convergence of travel times. (page 10) 7. The committee believes that time-of-day link volumes estimated in the four-step model process should be more directly linked to TPB’s postprocessing procedures. TPB should develop postprocessing procedures that maintain consistency with the agency’s four-step travel demand modeling procedures. (page 11) With regard to other questions TPB posed, the committee offers the following points: 8. While large survey sample sizes generally yield statistically more precise estimates of

Page 4 of 23 important modeling parameters, large surveys can be expensive. For the purposes of model calibration, surveys that incorporate selective sampling of stratified populations can be more effective and efficient than those that entail larger random samples. (page 14) 9. The committee believes that TPB’s proposal to consider a nested logit model is appropriate. However, determining the specific nesting structure requires extensive empirical analysis. The committee recommends that TPB also consider other discrete- choice model formulations that allow more flexible representation of competition among different transportation modes. (page 12) 10. The committee agrees that TPB should actively monitor the progress of early adopters of new models and take appropriate action to ensure that the agency’s modeling meets current standards of good practice. At the same time, TPB should maintain balance in its work programs to achieve shorter-term objectives of producing forecasts required to meet agency responsibilities. (page 12) 11. Many factors influence the appropriate number and size of analysis zones to be used in modeling. The committee can offer only general comments, within the context of the scope of this study and the specific information available, concerning the grain size of TPB’s zone system. (page 13) 12. The committee commends TPB’s plan to conduct a new regional household survey and offers a number of comments on details of the proposed survey. (page 14) Background The committee’s first report stated eleven points concerning its assessment of the performance of TPB’s travel demand models. Six of these points, which questioned or suggested changes in various aspects of TPB’s models, are addressed directly in the work program document TPB prepared for the committee. Other points in the first report addressed matters on which the committee substantially agreed with TPB’s procedures. TPB’s work program document proposes work elements for the model development program organized in five parallel “tracks,” reflecting TPB’s historical approach to advancing its travel models: Track 1—Application: Improvement of the currently adopted model set to produce adequate forecasts while enhanced models are in development. Track 2—Methods development: The incorporation of advanced practice in travel demand modeling that can be made operational in the next few years. Track 3—Research: Keeping abreast of research developments in areas of travel modeling, surveying, data (GIS) maintenance practices and integration, and simulation. Track 4—Data collection: The implementation of data collection designed to meet the needs of Tracks 1, 2, and 3. Track 5—Maintenance: Documentation of the current modeling applications, including recent improvements in software and data requirements. This track includes an ongoing effort to train staff in the use of current and updated application procedures.

Page 5 of 23 The document also poses a number of questions soliciting the committee’s advice on several matters concerning modeling strategy and data collection. The following sections present the committee’s findings and conclusions, first on the points of concern raised in the committee’s initial report and addressed in the work program document and second on TPB’s questions. The latter questions include matters of travel surveys and model granularity cited in the scope of the study initially presented to the committee. Throughout this letter, we have not followed the structure of TPB’s proposed work program but do make reference to specific elements of the program in our comments. The committee remarked generally that the five tracks for model development are a reasonable organization for the agency’s work program but voiced concern that activities identified by such terms as “methods development” or “research” may be viewed by some people as not likely to contribute immediately to an MPO’s day-to-day operations. These activities then are more difficult to justify in budgeting discussions and are targeted for funding cutbacks when budgets become tight. Strong leadership and commitment are needed to maintain such an active model development program, to ensure that progress is maintained on all tracks, and to ensure that the results from the program are promptly incorporated into the production models used for travel demand forecasting. An awareness of what is being done at other MPOs can be valuable to technical staff and senior managers responsible for providing such leadership and commitment. While the models most MPOs use embody similar logic and assumptions, there are no widely accepted guidelines explicitly delineating best practices or even presenting a comprehensive comparison of various regions’ practices. TPB has undertaken to collect information from other MPOs with similar characteristics4 for comparative analysis of modeling practices and demand estimation results. However, TPB reports that progress has been hampered by difficulty in obtaining detailed and comparable current documentation on the various MPOs’ modeling practices. The committee anticipates that this effort will continue to be challenging. TRB, with sponsorship from the U.S. Department of Transportation, is undertaking a study to gather information and prepare a synthesis of practice on metropolitan area travel demand modeling. The study should be useful to TPB in determining modeling practices at other MPOs. Comments on TPB Proposals Responding to Previous Committee Assessments TPB’s discussion of proposed work elements includes responses to six areas of concern mentioned by the committee in its first letter report: model validation, travel estimation for trucks and commercial vehicles, bus network representation, uses of adjustment factors, applications of feedback through mode choice in reaching final travel estimates, and 4 TPB lists eleven peer MPOs and includes preliminary results of the analysis in the work program’s Appendix A.

Page 6 of 23 procedures in postprocessing for estimating hourly highway traffic volumes and speeds. The committee considers each of these areas in turn. Improving Model Validation The committee commented in its first report that statistical measures indicated that base-year modeled link volumes do not match observed traffic counts and transit ridership as closely as the committee would typically expect in model validation. TPB’s work program document addresses this concern primarily in the highway and transit validation work element (1.A.). This work element entails several steps intended to improve the match between modeled and actual link volumes, including network enhancements to better reflect actual conditions (1.A.1), short-term modeling improvements (1.A.2), longer-term modeling improvements (1.A.3), and testing of the SUMMIT model for use as a diagnostic tool (1.A.4).5 TPB refers to staff’s continuing efforts to achieve improvements in validation statistics. It cites as examples recent applications of its model set to study travel demand in specific regional corridors, where improved statistics have been achieved. TPB attributes these improvements to the use of refined free-flow speed and capacity values, a refinement of the zonal area–type assignments, adjusted link volume-delay functions, and improvements related to network coding. TPB believes that improvements such as these can be of benefit to the regional model. The committee finds these results encouraging and proposes that TPB staff continue to evaluate whether systematic adjustments of the sort used in these corridor studies may lead to improved highway assignments on a regional level. However, adjustments of selected link impedances merely to correct over- or underassigned volumes should be avoided. With regard to the match between estimated base-year and observed transit ridership, TPB has outlined several short-term enhancements to its current model package, such as incorporating existing sub–mode choice and rail access mode choice models, relating bus travel times to highway link times, and evaluating the weighting of in-vehicle and out-of-vehicle travel times in transit route choice. The committee was encouraged by the interest shown by the region’s principal transit agency (WMATA). It hopes that other agencies similarly will recognize the value of TPB’s model development activities and cooperate in supplying traffic and passenger counts and other operational data, and in collaborative data collection efforts. TPB proposes to investigate the new SUMMIT mode choice software being developed with Federal Transit Administration (FTA) support. The use of SUMMIT for projects seeking to qualify under FTA’s New Starts program has typically revealed problems in three areas: the physical and operating plans for the alternatives; errors in network coding; and fundamental limitations, errors, and other weaknesses in the mode choice model. The activity of modifying the mode choice model code to produce data files compatible with SUMMIT input requirements exposes these problems, making the SUMMIT software an excellent diagnostic tool as well as a means for estimating total user benefits of a transit improvement. User benefits are computed at the upper level of the mode choice model, so it can be equally applicable to multinomial or nested logit model formulations. 5 Titles and reference numbers for subtasks are presented in TPB’s schedule charting the timing of planned work elements (Figure 1 of the TPB document) and are included in discussion of the primary work element.

Page 7 of 23 The committee agrees that these enhancements, along with the longer-term recalibration and restructuring of the agency’s mode choice model, should yield benefits. Overall, these various efforts to improve model validation are commendable. In addition, the committee has the following comments, which may be useful to TPB in the validation portion of its work program. TPB’s reported error statistics (%RMSE) comparing highway counts and link volumes may be somewhat overstated because of small numbers of observations in some of the tabulated link count categories. The values in some cases also appear to include links in large buffer zones outside the MWCOG study area, which could add to the overstatement of error. The committee proposes that TPB staff revise the categories used in these tables and not include links in large buffer zones in the computations. The committee notes that TPB completes a relatively small number of iterations of the equilibrium highway-assignment algorithm and does not indicate a criterion for determining how many iterations may be appropriate. The committee believes that improvements in base- year highway link volume validation through additional iterations may be possible. Some testing of how the number of iterations affects fitting results could be included in TPB’s model maintenance work track. In such testing, the number of iterations may be limited by monitoring some standard measure of convergence.6 Truck and Commercial Vehicle Travel The committee commented in its first report that combining business and commercial trips in the non-home-based trip category is not advisable. The committee was concerned that commercial vehicle travel is influenced by factors fundamentally different from those influencing personal travel and was not persuaded by TPB’s explanations with regard to the use of light-duty trucks for commuter travel in the region. TPB has proposed, in the business and commercial trips work element (1.B.), to develop a set of truck and commercial vehicle models and thereby to separate the modeling of commercial and personal travel. The approach proposed is similar to that used by the Baltimore Metropolitan Council (BMC) and entails use of truck and commercial vehicle traffic classification counts to adjust base commercial vehicle trip tables. Other MPOs have used similar procedures to update their truck and commercial vehicle forecasts. The committee finds this proposal encouraging as a near-term solution, but it is concerned that model development work is scheduled to commence only after a series of classification counts are conducted in 2006 (work element 1.B.2). The committee recommends that new counts be collected sooner, if at all possible, to accelerate this work activity’s completion. Documentation provided by BMC indicates that link counts from 550 locations in the Baltimore region and an additional 50 locations in the Washington region for 2000 were used to adjust the base truck trip table and estimate a commercial travel trip table. BMC staff report that other counts are available from the Maryland State Highway Administration. The 6 One widely used measure is calculated as [Σ (link volume*link time) – Σ (O-D volume*minimum path time)].

Page 8 of 23 committee surmises that some truck count data currently available in the MWCOG study area7 would allow work on the truck and commercial vehicle trip tables to begin before additional classification counts are collected. The committee recommends that preliminary work be scheduled to examine the availability and coverage of truck counts, to support early updating of the TPB truck and commercial vehicle trip tables. Other techniques suggested in the literature might also be helpful in accelerating this work.8 With regard to our earlier discussion of the likely evolution of travel demand models, we note that the type of model being proposed is fairly crude. Over the longer term, the committee anticipates that TPB will find it appropriate to upgrade its truck and commercial travel modeling through a more behavioral approach. TPB should consider conducting a survey of commercial firms, stratified by types and volume of goods shipped, to provide a stronger basis for model development. Truck and commercial vehicle trips entering and departing the region, as well as through trips, may be estimated from a cordon intercept survey. Bus Network Characterization The committee commented in its first report that TPB’s use of fixed bus speeds and other coding details in its networks may misstate the influence of transit in estimates of future trip distribution and mode choice. TPB acknowledges the committee’s concern and plans to work with the region’s transit agency and others to find a method for representing bus speeds in future years (work element 1.C.). The committee finds TPB’s plan encouraging and notes that some agencies use estimated highway travel times from traffic assignment to modify bus travel times employed in mode choice and trip distribution modeling. Network coding for express and major line-haul bus lines on freeways and major arterials, in the committee’s view, certainly should reflect future congestion levels and travel times. However, care is needed in linking the underlying network of local and feeder bus schedules to less reliable assignment travel times on minor arterials and local streets. As we noted with regard to model validation, TPB proposes to investigate FTA’s SUMMIT software as a tool for assessing transit network quality. The committee agrees that this is a worthwhile activity. Use of Adjustment Factors The committee discussed TPB’s use of adjustment factors in trip generation, trip distribution, and mode choice in its first report. TPB’s work program document includes an element to minimize the use of adjustment factors (1.D.) that entails work of two types. TPB staff plan first to document more fully the bases for the various adjustment factors currently used and actually commence to do so in the work program document.9 The committee appreciates this 7 Truck volumes must be submitted for Highway Performance Monitoring System sections, for example. 8 For example, see Quick Response Freight Manual, prepared for the Federal Highway Administration, Office of Planning and Environment Technical Support Services for Planning Research, by Cambridge Systematics, Inc., with Comsis Corporation, and University of Wisconsin–Milwaukee, DTFH61-93-C-00075, September 1996. 9 The work program document’s Appendix D cites four principal reasons for using adjustment factors: substantial underreporting of nonwork travel in travel surveys, aggregation errors associated with trip production

Page 9 of 23 effort and agrees that the work should continue. The second type of work TPB proposes is a sensitivity analysis to investigate whether the use of factors can be reduced. TPB’s schedule indicates that trip generation, trip distribution, and mode choice will be tested and may be modified. To the extent that this work results in reduced use of factors, the committee finds this effort to be positive as well. As indicated in our first report, we recognize that the use of adjustment factors is not uncommon. However, justification for the use of such factors is based on current conditions. There is little theoretical basis for anticipating that such adjustments will remain constant. The effect of a physical barrier may change as development patterns shift over time, and jurisdictional barriers can be readily altered by changes in local tax policies, school characteristics, and real estate values. When primary model estimates (trips, interchanges, and mode share results) are heavily factored, future estimates of these quantities will often reflect the factors more than they do the underlying behavioral relationships embodied in the models. This result is particularly likely when large factors are applied to modeled trips and trip interchanges that initially are small values but are projected to become much larger in future forecasts. The committee notes that as a practical matter it is difficult to trace cause and effect when multiple model results are factored. Some factors in the later stages of the four- step process may simply be compensating for factors applied in earlier stages. The committee recognizes that professionals may reasonably differ on what represents “extensive” or “excessive” use of adjustment factors but continues to believe that TPB’s modeling relies unduly on adjustment factors. The committee agrees, therefore, that TPB should not only document clearly its reasons for applying each adjustment to its current models, but also periodically review those reasons and confirm that they remain valid. TPB should examine whether applied factors remain applicable and consistent with the assumptions underlying forecasts of future travel patterns. TPB should conduct sensitivity tests to assess the magnitude of the combined influence of factors applied in all stages of the four-step model. Otherwise, even though the adjustments are apparently justifiable, as they become more extensive they could weaken the fundamental behavioral logic of the modeling process and threaten its ability to provide defensible forecasts. The committee found the newest information presented in Appendix D of TPB’s work program to be helpful in understanding some of the adjustments being made, but questions remain. The Potomac River and jurisdictional boundaries in the Washington region, for example, may skew travel patterns. Trips originating in a zone near such a perceived barrier may be more likely to terminate in a zone on the same side of the barrier, as compared with otherwise equally attractive destinations on the other side of the barrier. Arguably, the classic and most clearly justifiable use of adjustment factors (in this case, K-factors) is to adjust interzonal impedances for zonal pairs that have the barrier between them. The committee was puzzled, however, that links between Montgomery and Fairfax Counties, for example, appear to require no K-factors, despite their separation by the Potomac River, while factors are models, inadequacies of explanatory variables to account for important factors influencing travel patterns, and limited geographic scope of travel survey data compared with the region within which travel is forecast.

Page 10 of 23 abundantly applied to other intercounty links. Speed Feedback Incorporating Mode Choice The committee commented in its first report that TPB’s feedback of highway and transit times to trip distribution bypasses mode choice and is not typical of good modeling practice in regions with significant transit services and ridership. TPB has proposed in its speed feedback work element (1.E.) to review speed feedback practices of other MPOs and to conduct sensitivity analyses to determine whether changes in procedures will improve modeling results. Sensitivity analyses conducted as part of TPB’s proposed work program explore the consequences of introducing additional feedback iterations as well as including mode choice in the feedback computations. The analyses indicate that including mode choice in the feedback does indeed influence transit travel forecasts, and TPB suggests adjustments that might be made in the model application. The committee found these analyses and suggested adjustments to be helpful. However, the committee suggests that clearer explanations of the bases for judging the adequacy of the feedback procedure might help to explain its influence on travel forecasts. The committee notes that there is a well-known algorithm for establishing equilibrium among trip distribution, mode choice, and assignment. The algorithm is applied by iterating through distribution, mode choice, and assignment, successively averaging link volumes over all completed iterations, computing new link travel times using the resulting average link volumes, building new paths and travel times between origin-destination zones, and then returning for another iteration. There is a theoretical basis for the algorithm, and it has been proved mathematically to converge to consistent link times.10 TPB’s work plan document, in contrast, describes a heuristic approach that approximates equilibrium conditions. The committee notes that the number of iterations used in TPB’s procedures appears small, and it did not find adequate documentation of how close the final assignment times are to convergence. Furthermore, the committee observes that average regional speed is not a good measure of convergence. It is possible for the regional average speed to remain nearly constant without achieving reasonable convergence in zone-to-zone travel times. The committee proposes that TPB use the equilibrium algorithm. Because network analyses can be unreliable under severe conditions of congestion, experimentation with alternative feedback approaches may be useful.11 The committee suggests that TPB test different methods for weighting highway and transit times to produce a composite travel time for distribution. It may be more effective to weight transit times by the mode share in the 10 This and alternative feedback algorithms are evaluated by Boyce, D. E., Y.-F. Zhang, and M. R. Lupa in Introducing “Feedback” into Four-Step Travel Forecasting Procedure Versus Equilibrium Solution of Combined Model, Transportation Research Record 1443, TRB, National Research Council, Washington, D.C., 1994, pp. 65–74. 11 See, for example, Boyce, D., B. Ralevic-Dekic, and H. Bar-Gera, Convergence of Traffic Assignments: How Much Is Enough? Journal of Transportation Engineering, Vol. 130, No. 1, Jan.–Feb. 2004, pp. 49–55; and Incorporating Feedback in Travel Forecasting: Methods, Pitfalls and Common Concerns, Report DOT-T-96-14, Comsis Corporation, Silver Spring, Md., and Federal Highway Administration, Washington, D.C., March 1996.

Page 11 of 23 interchange than to calculate an origin-destination composite time that weights the transit time by the regional transit mode share. The former approach should increase the model’s sensitivity to transit improvements by increasing the importance of transit in trip distribution for those interchanges with a high transit mode share while reducing its importance for those interchanges with negligible transit trips. Traffic Speed and Volume Estimation for Air Pollution Emissions Estimation The committee commented in its first report that TPB’s procedure for estimating hourly traffic volumes and speeds for air quality modeling was questionable. The postprocessing procedure entails two steps: first, aggregating peak- and off-peak-period traffic assignments to a 24-hour total that is redistributed to hourly periods as a percentage of daily volume; and second, adjusting the initially estimated hourly volumes as necessary to meet link hourly capacity constraints. The second step is referred to as “peak spreading.” The committee expressed two concerns in its first letter. The first is that TPB’s aggregation of peak and off-peak travel model estimates to a 24-hour volume and subsequent redistribution to hourly estimates based on a percentage of daily volume essentially dissociates the hourly volumes, and subsequently the final emissions estimates, from the peak and off-peak projections produced by the four-step model. The second is that TPB’s peak- spreading method may create unintended emissions impacts. TPB’s work plan addressed the second concern with additional sensitivity analysis but did not comment on the committee’s first concern. We have elaborated on this first concern with a set of computations discussed in the following paragraphs and Attachment 3. To compare the differences between time-of-day distributions of traffic produced by the four- step model with those produced by the postprocessing procedure, the committee conducted a simple analysis. We compared the peak-period traffic volumes from TPB’s four-step model with the peak-period volumes estimated by the hourly profiles used in TPB’s postprocessing. The analysis results are shown graphically in Attachment 3. The ratio of peak-period to daily traffic estimated in TPB’s postprocessing is necessarily a single number for each of the categories of links defined for the postprocessing procedure. The ratio estimated by the four-step model can vary. Ideally, the ratios of peak-period to daily traffic produced by the four-step model would be tightly clustered in a balanced distribution around the single-number estimate used in the postprocessing procedure. However, we found differences between the two sets that are in many cases strikingly large and skewed. The current postprocessing procedure undermines the relationship that ought to exist between the hourly volumes used for mobile source emissions estimates and the AM, PM, and off-peak volume estimates produced by TPB’s four-step model. The estimates of hourly volumes and speeds must be associated directly with the time-of-day (AM, PM, off-peak) travel model output. A simple method for accomplishing this would be to allocate volumes proportionally within each time period (i.e., the percentages of hourly

Page 12 of 23 volume within a time period sum to 100 percent). TPB staff have been aware of this issue and noted the possible need for hourly volume distributions as a percentage of the period volume instead of as a percentage of the daily volume.12 The committee asserts that such an effort is necessary to produce hourly volumes for the mobile source emissions process that are credibly linked to travel demand estimates and should be included in TPB’s work program. Response to TPB’s Questions to the Committee The committee’s responses to questions posed in TPB’s proposed work program document are given in the following sections. The questions address matters associated with items in the study’s initial scope or raised during the course of committee discussions with TPB staff. Nested Logit Models in Mode Choice TPB asked what level of survey sampling would be needed to support a nested logit model choice formulation of the size and structure TPB proposes. TPB asked further whether the committee could suggest a different structure that might be less difficult to estimate and calibrate. The nested logit formulation is often an effective tool for travel demand estimation, and software exists for its estimation. Determining the specific nesting structure requires extensive empirical analysis. While the estimation of nested logit models may require fewer observations of data, they often require higher-quality data in terms of information obtained per observation.13 The committee suggests that TPB consider other currently available discrete choice formulations that might be more flexible in representing the behavioral characteristics of competition among transportation modes. Explanations of such formulations are available in the literature.14 Matters of survey sampling are addressed below (page 14). Alternatives to the Four-Step Model In the work program TPB presented to the committee, TPB asked what direction the committee might suggest that TPB take with respect to the development of tour- and activity- based models. As we explained earlier in this letter, the committee anticipates that microsimulation, tour-, and activity-based models will be increasingly used within the coming decade for travel demand modeling at MPOs. These alternative formulations are being adopted now in some MPOs’ modeling practices. 12 See Metropolitan Washington Council of Governments memo file from Mike Freeman on Development and Recommendations of Hourly Distributions of Daily Traffic Volume, Aug. 27, 2002. 13 Meyer, M. D., and E. J. Miller, Urban Transportation Planning: A Decision-Oriented Approach (2nd ed.), McGraw-Hill, 2001, p. 303. In addition, some practitioners have encountered problems with overspecification; the problem is explored by Bierlaire, M., T. Lotan, and P. Toint in On the Overspecification of Multinomial and Nested Logit Models due to Alternative Specific Constraints, Transportation Science, Vol. 31, No. 4, 1997, pp. 363–371. 14 See, for example, Bhat, C. R., Random Utility-Based Discrete Choice Models for Travel Demand Analysis, in Transportation Systems Planning: Methods and Applications (K. G. Goulias, ed.), CRC Press, Boca Raton, Fla., 2003.

Page 13 of 23 The work of these early adopters will benefit the profession as a whole as well as the stakeholders in those specific agencies’ travel demand forecasting. The committee recommends that TPB actively monitor the early adopters’ progress and appreciates TPB’s efforts to learn more about such models. We note that these new models can be formulated and estimated on the basis of TPB’s existing survey data, although data collection with survey instruments designed to support such models can enhance the result. For example, collecting in-home activity information and using 2-day activity diaries may produce useful results. Agencies that have conducted such surveys—for example, the San Francisco Bay Area Metropolitan Transportation Commission—may be helpful sources of advice. The committee recognizes that responsible agency officials must ensure that the agency’s ability to produce forecasts suitable for decision making in the region is not compromised. However, it is important that TPB keep abreast of developments in new approaches to modeling and maintain the currency of the agency’s practices. Grain Size in Travel Modeling In the scope of work presented to the committee at the start of the study and in its recent work program, TPB asked the committee to comment on the grain of travel analysis zones. TPB staff have commented in meetings with the committee that zone size is an issue particularly with regard to estimation of pedestrian and other nonmotorized travel and transit usage in areas of higher-density development. Many people suggest that larger numbers of smaller zones—a finer-grained representation of the region—will produce better travel demand forecasts. However, as zone size is reduced, data requirements increase. Software and computing resources available to the agency often control the maximum number of zones that is practical for regional analysis. In general, zone sizes should take into account how grain size will influence variables important to the policy questions the model is to address. Grain size should be fine enough to provide a robust basis for making decisions, but not so fine as to result in excessive data and computing costs. That said, some practitioners suggest rules of thumb. For example, the number of zones actually used in designing a model is likely to be at least equal to the number of census tracts in the study area. The number should be sufficiently fewer than the practical maximum (e.g., considering data costs and computational resources) to allow for an expansion of the number of zones later for special studies and to subdivide zones to account for major land developments. Local factors will affect the preferred grain size of the analysis zone structure. For example, how is access to transit handled in the models? How many paths between zones are loaded in each model run? What is the geographic basis for the zone system? Answers to such

Page 14 of 23 questions are important in this context. TPB staff might gain insights by reviewing the literature on the topic.15 Travel Surveys and Other Data for Travel Modeling The scope of work presented to the committee at the start of the study called for commentary on travel survey and other data needed for future TPB model upgrades. In its work program document, TPB asked several questions concerning a proposed regional household travel survey. Overall, the committee commends TPB for its aggressive attention to maintaining the currency of its data. Trade-Offs of Increased Household Sample Size TPB asked about factors to be considered in increasing its survey sample size from 5,000 to 10,000 or 15,000 households. Given the need for data to calibrate a mode choice model with several alternatives, it is probably necessary to think beyond random sampling of households in order to improve the efficiency of the survey. Sample Size for Model Development TPB asked whether there is a “compelling need” to have a sample size greater than 10,000 households for model development. The committee recognizes that larger sample sizes will generally give statistically more precise estimates of important model parameters. Sample sizes generally may be determined by considering the statistical variation anticipated in the underlying population and the level of reliability needed to support robust decision making. The committee presumes TPB’s situation is similar to that of other regions and on this basis finds it difficult to fault TPB’s proposal to take a 15,000- household sample. Nevertheless, the committee encourages TPB first to identify the parameters with significant influence on policy decisions that are to be estimated from survey results and then to determine sample sizes necessary to yield reliable estimates of those parameters. As we have commented in preceding sections, response rates are crucial.16 The problem is to collect enough observations for households with (for example) transit trips to support reliable inference of causal relationships. The solution will likely depend less on the total number of households sampled than on how the survey is structured. The committee suggests that 15 For example, the following papers and reports may be helpful: Khatib, Z., K.-T. Chang, and Y. Ou, Impacts of Analysis Zone Structures on Modeled Statewide Traffic, Journal of Transportation Engineering, Vol. 127, No. 1, Jan.–Feb. 2001, pp. 31–38. You, J., Z. Nedovic-Budic, and T. J. Kim, A GIS-Based Traffic Analysis Zone Design: Technique, Transportation Planning and Technology, Vol. 21, 1997, pp. 45–68. You, J., Z. Nedovic-Budic, and T. J. Kim, A GIS-Based Traffic Analysis Zone Design: Implementation and Evaluation, Transportation Planning and Technology, Vol. 21, 1997, pp. 69–91. Wilbur Smith and Associates, The Effect of Zone Size on Traffic Assignments, Commonwealth Bureau of Roads, Melbourne, Australia, 1971. Hanscom, E. W., and K. C. Sinha, The Effects of Simplifying Traffic-Zone and Street-Network Systems on the Accuracy of Traffic Assignments in Small Urban Areas in Indiana, Interim Report, No. FHWA/IN/JHRP-79/15, Purdue University, West Lafayette, Ind., 1979. 16 Refer, for example, to Special Report 277: Measuring Personal Travel and Goods Movement: A Review of the Bureau of Transportation Statistics’ Surveys, TRB, National Research Council, Washington, D.C., 2003. Low response rates can introduce systematic selectivity bias in the estimation of key parameters and bring about a waste of resources associated with simply increasing sample size in an attempt to reduce estimation errors.

Page 15 of 23 stratified and choice-based sampling will be appropriate strategies for addressing this problem. Sampling Households Outside Planning Area TPB asked whether there a “compelling need” to collect a minimum number of household samples for jurisdictions that are modeled but are beyond the TPB planning area. The committee sees no reason to collect household data from outside the planning area unless those households are anticipated to have unique and underrepresented characteristics that could have critical influence on travel demand projections. Activity-Based Travel Diaries TPB asked whether there is a “compelling reason” to use activity-based travel diaries for future travel surveys and how the use of diaries would influence response rates and overall survey data quality. As explained previously, activity diaries are not needed as part of a survey to calibrate activity- and tour-based models. However, the activity-based travel diaries can provide useful information concerning in-home activities and joint activities pursued outside the home, which can enhance an activity-based model system. Multiday Survey TPB asked about the value of a multiday survey, particularly with regard to survey response rate and data quality. An attractive feature of a multiday survey is the moderation of normal day-to-day variation in household travel activity. Reports in the literature suggest that a 2-day period may be optimal, and the committee observes that the marginal costs of gathering a second day are probably low. However, arguments might be made that a new survey should collect travel data on every day of the week, including weekends. The committee cannot offer a compelling reason for making decisions about such matters without additional information on survey costs and the plausible relationship of survey duration and variance in parameter estimates, but it notes that shifts in the patterns of travel by time of day and day of week may occur over the multiyear periods between major household travel surveys. TPB’s FY 2003 Unified Planning Work Program indicates that a sixth wave of data collection in the continuing longitudinal household travel survey is scheduled for spring 2004; the survey is intended to track and assess changes in travel behavior over time. Random-Digit Dialing Sample Frame TPB asked about the value of continuing to use random-digit dialing (RDD) sample frames in light of declining response rates. As we have already commented, the committee agrees with TPB’s observation that declining response rates are a problem generally. We note also that a number of troublesome issues arise when household surveys rely on the initial recruitment call with RDD lists. For example, cell phones are an increasing proportion of telephone numbers, and the variety of ways cell phones are used pose problems for the RDD survey methodology. Increasing numbers of households are or will be on the “do not call” list or do not speak English as a primary language. Such trends are increasing the potential difficulties of recruiting travel survey respondents and avoiding sampling bias.17 17 The National Research Council–appointed Committee to Review the Bureau of Transportation Statistics’ Survey Programs noted that survey methodologists generally are concerned about declining response rates and resulting bias in telephone survey results. For a variety of reasons, dial-up surveys are not likely to remain a

Page 16 of 23 We nevertheless find the RDD sample frame to be a practical way of coupling household surveys with the computer-assisted telephone interview. The committee proposes that TPB explore recent applications of other survey design and recruitment methods. Some surveys, for example, have used mailing of introductory letters and other preparatory activities to improve response rates. Global Positioning System Household Vehicle Add-On TPB asked about the costs and benefits of a Global Positioning System (GPS) household vehicle add-on subsample to the regional household survey. The committee has reservations about this technology and cannot recommend its use. However, the technology may record travel that is otherwise unreported or underreported and so may be justified as a small add-on as described in TPB’s work program document. Reconciling GPS readings with travel diary and interview data may be very labor-intensive and yield relatively little information beyond what can be obtained by observing household vehicle odometer readings. Year-Round Travel Survey TPB asked about the value of moving from a “typical season” to a year-round data collection strategy for future household surveys. The committee notes that year-round survey procedures will capture seasonal employment and the effects of school- year travel variation. Additional factoring will be required to convert such survey results into the typical-day format used in the traditional four-step model. In addition, year-round surveys require extra care in their design and execution to ensure that each subsample is representative and that the survey yields robust statistics.18 Concluding Observations The committee has found TPB staff to be open in their discussion of the various matters we have addressed in this study. While many MPOs undertake to have a peer review of their modeling practices, these reviews are usually conducted within the confines of the agency by professionals selected by the agency’s leadership. We commend TPB for requesting that the National Research Council appoint an independent TRB committee to conduct this review and for its responsiveness to the committee’s several requests for additional information. We appreciate this opportunity to assist TPB in dealing with the complex and sometimes controversial issues of travel demand modeling. Yours very truly, David J. Forkenbrock Chair viable data-gathering technique in the long run, but replacement techniques are not yet adequately developed. See Special Report 277. 18 For example, the American Community Survey (a U.S. Census Bureau activity; see http://www.census.gov/acs/www/) relies on annual waves of samples. While TPB’s survey may be conducted over a shorter time period, the design issues (e.g., updating and data blending) could be similar.

Next: Attachments »
Review of Travel Demand Modeling by the Metropolitan Washington Council of Governments: Second Letter Report Get This Book
×
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

The TRB Committee for Review of Travel Demand Modeling by the Metropolitan Washington Council of Governments has issued the second of two letter reports to the National Capital Region Transportation Planning Board (TPB). This report reviews TPB’s proposed direction of future travel demand model upgrades. The first report reviewed performance of the TPB's travel forecasting model and processes for estimating mobile source emissions.

Appendix A

  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!