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.
153 This research has provided an extensive analysis and syn- thesis of travel forecasting, best practices, as well as opera- tional research approaches to the modeling highway pricing projects. The conclusions and recommendations are summa- rized below in four major groups: ⢠Existing practices and identified gaps, ⢠Recommended short-term improvements, ⢠Major long-term improvements and strategic directions, and ⢠Suggestions for future research. 7.1 Existing Practices and Identified Gaps The review and analysis of the travel models and network simulation tools applied for T&R studies in practice has revealed a highly diverse picture, with a large proportion of simplified methods commonly applied, along with a growing number of applications of more advanced modeling tools. The following main conclusions can be made regarding the general tendencies observed and the identified gaps where improvements are needed: ⢠There is a great deal of variation in approaches. In most cases, the model applied for the highway pricing project was essentially a quite modest modification of the existing regional model available for the study. Thus, limitations and deficiencies of the existing regional model were inevi- tably adopted for the study. ⢠In most cases, only route itinerary (assignment) and binary route type choice (toll versus non-toll) models were employed for comparison and evaluation of pricing alternatives. This achieves reasonable results under the assumption that pricing would not affect mode choice, time-of-day choice, distribution of the origins and destinations of travel, or travel generation. While this simplification might be in some cases acceptable for intercity highways, it is difficult to defend for most analysis of pricing in metropolitan and urban settings. ⢠Pricing effects on trip distribution have been incorpo- rated by using mode choice Logsums as the measure of accessibility in destination choice or gravity-type dis- tribution models. The use of mode choice Logsums in gravity models needs to be tested for validity. Unlike in the logit destination choice framework, where appropri- ate elasticities with respect to cost are expected when reasonable Logsum parameters are used, it is not clear that doubly constrained gravity models behave appro- priately to changes in LOS variables such as the intro- duction of tolls. ⢠In some cases there is an inconsistency between the travel times and costs used for mode choice models, trip distri- bution, and assignment, in that the costs and travel times that reflect priced conditions are used in mode choice, and generated in assignment, while the toll costs do not enter the impedance function used for distribution. This is the case when travel times are fed back from a generalized cost assignment into a distribution model that is a function of travel times only. ⢠In a few cases utility functions in multinomial or nested logit mode choice models are miss-specified. Undesirable specifications include toll utilities that are a function of both the toll alternative travel time and travel time sav- ings with respect to the free alternative. This type of speci- fication may result in counter-intuitive results when the LOS attributes change on either the toll or the free routes. Another potentially problematic specification is the use of thresholds, such as making the toll alternative available only if it meets a pre-defined minimum time savings goal. The nesting coefficients on these models sometimes result in models with unreasonably high elasticity to toll, or time differences when the toll diversion is examined at the root level of the model (where they are comparable with the elasticity of route type binary choice models). C h a p t e r 7 Conclusions and Recommendations for Further Research
154 ⢠There is no consensus whether road pricing costs should be shared among vehicle occupants, and if so how. Most mod- els either assume that the full toll cost is either borne by all occupants or that it is equally shared among the occupants. Some models differentiate between cost sharing for HBW trips and cost sharing for other purposes. Sharing road pricing costs among vehicle occupants makes carpools less cost-sensitive, an assumption that may be acceptable for work trips, but is questionable for other purposes, where the majority of carpools are among members of the same household and oftentimes include minors. ⢠In some regional modeling systems that were specifically modified for congestion pricing projects, peak-spreading models were applied. Trip-based 4-step models are normally based on time-of-day (peak) factors that are not sensitive to relative congestion levels at different periods of the day. AMBs can offer a better framework where peak-spreading effects are captured by time-of-day choice sub-model. ⢠Peak-spreading or time-of-day models are sensitive to dif- ferences in travel times by time of day, but not to differ- ences in toll costs by time of day. This may be simply a result of the few localities where road pricing costs vary by time of day combined with the lack of observed data suf- ficient to estimate appropriate model parameters. ⢠Very few models to date have incorporated all trip and tour-level dimensions in a consistent way, and there have not yet been any practical examples of the incorporation of pricing impacts on the day-level, mid-term, and long- term choices, even with the activity-based models that have recently come into use. ⢠Almost all models, including ABMs, are characterized by a significant discrepancy between the user segmentation by VOT in the demand model compared to network simula- tion. While at the demand modeling stage, segmentation normally includes several trip purposes, income groups, car occupancy, and time-of-day periods; network simula- tions are characterized by a limited segmentation. Traffic assignments are implemented by periods of the day and for multiple vehicle classes that typically include vehicle type and occupancy. Trip purposes and income groups, however, are blended together before assignment, creating strong aggregation biases with respect to VOT. ⢠There are also discrepancies in the cost functions used to build best paths between the network simulations used to build travel time and cost matrices for the demand mod- els and the network simulations used to assign trips to the highway network. Best paths for the demand model may be built on the basis of travel time only, while the assignment is performed on the basis of generalized cost, or vice-versa. ⢠In almost all modeling efforts where route type choice (toll vs. non-toll) was involved, a problem of inconsistency between the generated trip tables for toll-users and their assignment onto the highway network was reported. This âleakageâ of toll users in the network simulation can be significant and constitutes a non-trivial analytical problem that requires special modeling efforts to resolve. ⢠Most models attempt to equilibrate supply and demand by feeding back travel times and cost from the assignment step to the trip distribution or mode choice steps. In most cases, feedback is executed for a fixed number of iterations, so convergence is not necessarily guaranteed. This may be particularly problematic when forecasting under conditions of high population growth, where congestion effects may be far more pronounced than in the base calibration year. ⢠Most models break down the network simulation into four broad time periods, typically AM Peak (2 to 4 hours long), Midday, PM Peak (2 to 4 hours long) and Night, and are therefore able to compute LOS differences by time of day only at this level of aggregation. Only one of the regional models reviewed performs the network simulation at a finer time of day disaggregation. 7.2 Possible Short-Term Improvements The short-term improvements summarized in this section are primarily applicable to trip-based 4-step models. A trip- based 4-step model, in combination with conventional static assignment, represents a modeling tool of a limited capabil- ity compared to more advanced ABMs and DTA. Although the major strategic directions for improvement of models are strongly associated with a new generation of advanced ABMs and network simulation tools like DTA, there are many practically useful steps that can be taken to improve 4-step models, as well as simple ABMs, in order to better prepare them for T&R forecasting and ensure reasonable model sen- sitivities to different pricing projects and policies in practical terms. The following main recommendations can be made: ⢠A travel model to be applied for highway pricing studies should comply with a minimal set of structural require- ments. Foremost among these is reasonable model sensi- tivities to tolls across all travel dimensions that could be affected by pricing actions to be studied, including: route choice, mode (and car occupancy) choice, trip distribu- tion, and time-of-day choice. Across all these choices, a reasonable level of segmentation and correct VOT esti- mates (with the necessary aggregations) should be applied. ⢠The demand model should be segmented by at least four to five travel purposes and three to four income groups, with VOT specific for each combined segment. An additional step that can be effective is to apply differential travel time coefficients by segments, and consequently make VOT values differentiated by network congestion levels. This in
155 effect represents a simple proxy for measures of travelersâ aversion to congestion (other than average travel times alone), including a lack of reliability associated with con- gested facilities. ⢠Network procedures that incorporate differential tolls and vehicle categories relevant to the pricing study are neces- sary. The traffic assignment should incorporate and distin- guish relevant vehicle classes (auto, commercial vehicles, trucks, taxis, etc.) with corresponding average VOT per class. The multi-class assignment technique is supported in all major transportation software packages (TransCAD, EMME, and Cube) and can be further applied to differen- tiate between VOT groups within the same vehicle class. If tolls or vehicle eligibility are differentiated by vehicle occu- pancy (HOV/HOT lanes) the auto vehicle class should be additionally segmented by the relevant occupancy catego- ries (SOV, HOV2, HOV3, etc.). ⢠It is highly recommended (although it is not an absolute requirement in the early stages of pricing studies) to incor- porate a binary route type choice model (toll versus non- toll facility), either as a lower-level, sub-nest in mode choice or as a pre-assignment procedure. This sub-model allows for capturing a toll bias associated with the percep- tion of a superior level of reliability and safety of the toll facility, as well as provides for better (non-linear) speci- fications of the tradeoffs between travel time savings and extra costs. ⢠It is essential for congestion pricing studies to include an improved time-of-day choice (peak-spreading) model sensitive to congestion levels and pricing. Although the trip-base 4-step model structure is not as flexible as ABM structure in addressing time-of-day choice factors, it can incorporate a time-of-day choice model with a fine level of temporal resolution (1 hour or less) that would roughly correspond to the outbound and inbound components of a tour-based time-of-day choice model applied separately for each trip segment. ⢠There are a growing number of applications where mode and/or occupancy choices are included. In several cases, mode, occupancy, and binary route type choices were combined in one multi-level nested logit choice model structure, where occupancy and route type choice served as lower-level sub-choices. These improvements can be implemented and are equally relevant for both 4-step models and ABMs. ⢠It is essential to equilibrate the demand model (at least mode choice and route type choice) and the highway assignment to ensure that the results correspond to (or at least approxi- mate) a stable equilibrium solution. It is more difficult to include the trip distribution (and other sub-models like time-of-day choice and/or trip generation) in the global equilibrium, which can require multiple iterations and special averaging algorithms. However, it is essential to eventually ensure a reasonable level of convergence of the entire model system. Recent experience with the New York ABM has shown that effective strategies of equilibration, based on a parallel averaging of trip tables and LOS skims, can achieve a reasonable level of convergence in three to four global iterations, even in one of the largest and most congested regional networks. ⢠Network simulations should be carefully validated and cali- brated to replicate period-specific traffic volumes, as well as period-specific LOS attributes. In this regard, the prevailing practice of model validation by daily traffic counts has to be replaced with more extensive and elaborate validation/ calibration by four to five time-of-day periods. ⢠There are many reserves for improvements that relate to a better understanding and incorporation of rules of finan- cial world. Many of them relate to the way in which a model is used, rather than to its structure per se. These include more thorough procedures for assessing non-modeled days (weekends and holidays) and time-of-day periods (if the model does not cover an entire weekday), as well as explicit consideration of possible ramp-up dynamics during the first several years of the project. The model structure and output should be made to produce the necessary inputs to the Financial Plan. Of special importance is the issue of quantification of risk factors. Risk analysis essentially rep- resents an important strategic direction with many aspects that have yet to be explored by travel forecasters. Some simplified procedures, however, are based on the possible scenarios for main input factors can be applied even with a simple travel model. 7.3 Major Long-Term Improvements and Strategic Directions The main avenues for improvement of modeling tools applied for pricing studies are seen to be associated with the advanced ABM framework on the demand side and DTA on the network simulation side. ABMs provide clear advantages over trip-based models in the analysis of pricing policies. In particular, such known limitations of trip-based models as a lack of policy sensitivity and insufficient market segmenta- tion can be overcome with these more advanced models. The main advantages of ABM structure for modeling highway pricing scenarios can be categorized according to the follow- ing model features: ⢠Tour-based structure that is essential for accounting for tolls applied by both directions by time-of-day periods, in a consistent and coherent way. This is, however, condi- tional upon obtaining a level of temporal resolution that matches the details of pricing schedules. Since variable
156 pricing schemes are frequently the focus of pricing studies, it is essential to have a large set of period-specific simula- tions, ideally, hourly assignments (or a full-day DTA as a better option as discussed below) in order to address dif- ferent pricing schedules. ⢠Microsimulation of individuals that allows for probabilis- tic variation of individual parameters including VOT, car rationing by license plate, toll discounts associated with dif- ferent payment types and/or population groups. In addition to that, a fully disaggregate structure of the model output is extremely convenient for reporting, analysis, and evalu- ation of the pricing scenarios, in particular for screening winners and losers, and for equity analysis across different population groups, etc. ⢠Entire-day individual activity pattern that allows for a consistent modeling of non-trip pricing options, such as a daily area pricing fee. There are, however, a number of issues that remain to be addressed by ABMs in practice. First, most ABMs continue to rely on static equilibrium highway assignment algorithms. It is common knowledge that such techniques fail to adequately address congestion due to their lack of ability to reflect queu- ing. One of the advantages of priced facilities (particularly dynamically priced facilities) is that they offer more reliable travel times than competing congested facilities where the variability of travel time can be quite onerous. From this per- spective, the integration of an ABM and DTA in one coherent modeling framework represents one of the most important strategic directions for the field. The advanced and flexible microsimulation modeling par- adigm embedded in ABM and DTA structures opens a con- structive way to include many recent theoretical advances in applied operational models. The following main aspects and directions were identified in this research: ⢠Heterogeneity of road users with respect to their VOT and willingness to pay. This requires a consistent segmenta- tion throughout all of the demand modeling and network simulation procedures to ensure compatibility of implied VOTs. In addition to an explicit segmentation, random coefficient choice models represent a promising tool for capturing heterogeneity. ⢠Proper incorporation of toll road choice in the general hierarchy of travel choices in the modeling system. Addi- tional travel dimensions (such as whether to pay a toll, car occupancy, and payment type/technology) and asso- ciated choice models should be properly integrated with the other sub-models in the model system. The impacts of pricing on long-term choices such as vehicle ownership, workplace location, residential location, and ultimately firm location need to be better understood. Most ABMs are based on cross-sectional data and are unable to fully capture long-term behavior associated with the introduc- tion of pricing policies. Hopefully, as more policies become implemented, more longitudinal data will be available to improve this critical aspect of travel demand models. ⢠Accounting for reliability of travel time associated with toll roads requires the incorporation of travel time reliability in applied models with quantitative measures that can be modeled on both demand and supply sides. ⢠More comprehensive modeling of time-of-day choice based on the analysis of all constraints associated with changing individual daily schedules. ⢠More comprehensive modeling of car occupancy related decisions, including differences in carpool types (planned intra-household, planned inter-household, and casual) and associated VOT impacts. ⢠More advanced traffic simulation procedures such as DTA and microsimulation, and better ways to integrate them with travel demand models. In this regard, future research needs to systematically incorporate features such as heterogeneous users in response to dynamic tolls, and develop efficient heterogeneous intermodal shortest path algorithms. Many of these research topics are being addressed in ongoing NCHRP and SHRP 2 projects. Incorporation of the results of these studies in models applied for highway pricing studies in practice represents an important challenge for the transportation modeling profession. 7.4 Suggestions for Future Research in Adjacent and Related Areas Highway pricing issues are closely intertwined with many general aspects of highway planning and modeling. The fol- lowing list of topics deserving of further investigation are either directly related to the modeling of pricing or are indi- rectly related to adjacent research areas that interact strongly with pricing: ⢠Effects of pricing on environmental quality and energy consumption. These measures are important in assessing the overall pricing benefits. ⢠Emerging automatic methods for the collection of infor- mation on highway volumes, speeds, and reliability. These new sources of information can be effectively used for gen- eral model improvement. ⢠Model development strategies for small MPOs. In general, it is almost impossible to outline a decent and defend- able analytical procedure for T&R forecasting without a regional travel model. Simplified sketch-planning tools can be applied at the initial phases of project development
157 to make a go/no-go decision, as well as to narrow the scope of possible alternatives. As the pricing project progresses to the phases of Environmental Impact and Investment Grade Studies, however, more substantial modeling work must be done. This represents a challenge for small MPOs that do not have sufficient modeling staff and resources to deploy an advanced regional ABM or DTA. ⢠Household and person travel time and cost budgets. It is known and well established in the micro-economic the- ory that the willingness to pay for any product (includ- ing travel time savings) is a strong function of the overall time and budget constraints. From this perspective, VOT cannot be explained for a particular trip, tour, or even travel day without taking into account the bigger picture of household and person behavior for a longer period of time. This aspect is still missing in almost all travel models, including the most advanced ABMs. ⢠Time scales for traveler responses to different pricing schemes. An important additional aspect of modeling traveler responses to congestion and pricing relates to dif- ferent time scales associated with different measures. The range of possible relevant time scales extends from a nearly instantaneous response (like changing a route as the result of real-time travel information or choice of a dynamically priced lane based on the current toll and congestion level on the general-purpose lanes) to the long-term effects (observed only in 20â30 years) like changes in population residential location or business activities. These particular topics have been identified in many pric- ing studies as deserving attention and came up frequently in the discussions of the project team with the panel of experts. This can be future research aimed at advancing the theory and practice of modeling road pricing.
