Skip to content U.S. Department of Transportation/Federal Highway AdministrationU.S. Department of Transportation/Federal Highway Administration
Office of Planning, Environment, & Realty (HEP)

Thurston Regional Planning Council (TRPC) Travel Model Peer Review Report

6.0 Peer Review Panel Response to Technical Questions

In its application to TMIP for this peer review and its presentation to the peers during the review TCRP proposed a series of specific questions. Those questions formed the basis of the peers' discussion and are listed below with the panel's responses in italics. The major headings in this section match those in Section 5 above for easy cross-reference.

6.1 Overall Modeling Framework

6.1.1 How can we make our model sensitive to travel demand management strategies on I-5, such as a potential addition of HOV lane, conversion of an existing general purpose lane to HOV, congestion pricing, etc.? In response to this question the panel proposed a multi-part approach: Build LOS sensitivity into trip generation to handle latent demand or flex work response to congestion Feed LOS and accessibility back to auto ownership (e.g. location-efficient mortgage scenario). See also section 6.1.4 below, "…peak hour to peak period…" The panel observed that there are two general options for structural changes to the model: a "basic" approach and a "state of the practice" approach; TRPC can choose based on its assessment of its needs and resources. Both approaches include borrowing PSRC's 11 assignment classes (HBW by 4 income categories, HOV2, HOV3+, etc.) or doing similar income stratification.

  1. Basic approach:
    1. Carry income classification through the entire model chain consistently (however, note that recent literature shows that low income travelers can have very high value of time(VoT) so this approach entails inaccuracies)
  2. State of practice approach:
    1. Replace the current trip generation submodel with a population synthesizer
    2. Use distributed VoT from the population synthesis in mode choice and assignment; even if the distribution is asserted it will be more useful than the static alternative.
    3. See Joan Walker's paper[8] on microsimulating a 4-step model

6.1.2 To better account for inter-regional commute pattern between Thurston County and Pierce plus King Counties, especially from Thurston County to the Joint Base Lewis McChord(JBLM), we are planning to expand our model boundary to include JBLM at an appropriate zone structure and network detail. Is this the right approach, and what are some related data and modeling issues we need to consider? Consistent internal representation (in other words, subject to complete treatment by all model components including trip generation, trip distribution, and so on) in both TRPC and PSRC models of zones and networks, will create useful sensitivity to factors affecting LOS in areas of "overlap." The preferable goal is to endogenously treat all factors driving LOS and trip generation. The extent of geographic expansion should be data-driven (using the 2010 External OD Survey) to ensure internalizing all significant origins and destinations WITHIN the enhanced model. For example, it may be advantageous to extend the TRPC model boundary farther north than SR 512. Borrowed survey data can inform this development task; see the question regarding travel survey data for related comments. A usual workplace location choice model (even at a coarse geography) would build in useful sensitivity; this could potentially be estimated from 2010 Census data, the 2010 OD Survey, and appropriate American Community Survey (ACS) Census Transportation Planning Package (CTPP) data. Ensure appropriate and consistent incorporation of special generators with their supporting network details (e.g. JBLM, campuses), informed by count and gate data already available, and consistent treatment of these elements in both the PSRC and TRPC models.

6.1.3 TRPC and PSRC travel demand models share the border at the north Thurston and Pierce Counties' boundary. We have found discrepancies between the two models at the boundary and want to work on resolving the differences. What are the core features and assumptions that must be in agreement between the models? The technical steps enumerated in section 6.1.2 "Model Boundaries" should ameliorate drastic discrepancies. Institutional steps will be necessary to enable the technical steps. The panel recommended establishing close collaboration with key partner agencies - especially PSRC and Pierce County - to address the technical needs.

6.1.4 We are transitioning from peak hour modeling to peak period modeling as part of this model update. Are there any factors to be aware of or lessons learned based on prior experience elsewhere? The panel advised that regardless of final assignment time periods it is most important to ensure accurate representation of peak-hour LOS (several methods are available; PSRC uses peaking factors derived from base year traffic counts to convert to hourly volumes within its multi-hour modeling periods). Add a time-of-day choice model: this makes the model sensitive to demand management policies and pricing strategies in ways that using static proportions does not. A time-of-day submodel should be responsive to the peak-hour LOS (some agencies like PSRC use smaller time periods for their time-of-day submodel) and is a useful way of addressing peak spreading issues. The panel endorsed the use of DTA for understanding fine-grained time-dependent response to congestion. Demand models can only be taken so far in capturing temporal responses.

6.1.5 What are some guidelines in modeling freeway auxiliary lanes and hard shoulder running? Within the limitations of the demand model (see DTA note just above) there are common and feasible treatments to ensure reasonable representations of the capacity two way left turn and other arterial treatments create. One approach to arterial geometries is to develop from observed traffic data customized volume-delay-functions by a somewhat detailed facility typology (e.g. 5-lane urban arterial, 4-lane urban arterial, 3-lane urban arterial, etc.). Another approach that may be applicable to hard shoulder running is to use fractional lane attributes (e.g. add 0.75 of a lane in time periods when hard shoulder running is activated). For auxiliary lanes, networks can be coded so that links end whenever the real-world cross section changes to avoid needing fractional lane attributes, or fractional lanes can be used. Fractional lane approaches should be supported by observed count and speed data to ensure accurate representation of capacity.

6.2 Modeling Framework Specific to Transit

6.2.1 TRPC posed several specific transit-related questions (6.2.2 through 6.2.4 below), but the discussion they engendered led the panel to make additional structural recommendations first: Build more robust treatment of zone-to-zone transit LOS in destination choice (i.e. incorporate transit LOS in demand portions of the model through transit-, bike-, and walk-inclusive logsums) AND in mode choice. Bus stop density was mentioned as a possibility but it is an ineffective proxy for zone-to-zone transit LOS. Consider more detailed wait-time treatment. This could be informed by a low-budget data collection effort (i.e. have staff spend short periods of time at selected key stops and time the actual waits).

6.2.2 What Park and Ride treatment can be done to model vanpool/carpool formation? Full treatment (for fixed-route transit also) of PNR is a high priority (see INRO's packaged PNR macros); in addition to internal trips, be sure to properly represent PNR access to northbound and southbound trips now treated as "internal-external" or "external-internal" in collaboration with PSRC to provide accurate LOS skims. Consider treating "park and pool" (vanpool/carpool) as PNR fixed-route transit supply; CTR data can help with this.

6.2.3 What transit data sources should TRPC maintain in-house? Establish a relationship with Intercity Transit to pull data as needed. Special attention should be given to getting datasets for validation and survey time periods and base years for projects. Route-level ridership by time of day from APC data has been useful for other agencies as has stop-level ridership for stops of interest (i.e. near important areas of interest and high transfer areas). Both APC and AVL data can also be useful for validating transit travel times in the travel model. Intercity Transit reports[9] that all its fixed-route fare services are instrumented with Automatic Passenger Count (APC) equipment plus Automatic Vehicle Location (AVL) systems, and that the AVL data is integrated with its GIS database in which all fixed-route stops are geocoded. See "New and Existing Data Sources" below for a suggested on-board survey.

6.2.4 Do we need to differentiate walk and bike access? If so, what additional data do we need to collect to account for the same? In the new, denser TAZs ensure sufficient network representation to accurately represent walk access. Treat walk-, bike-, and auto-access separately with appropriate "travel sheds" and speeds for each access mode. These can be calibrated based on on-board survey data and cross-checked with the HH survey. An oversample in the HH survey of this segment could help validate walk and bike skims and also allow for the estimation of separate models for each access mode.

6.3 Modeling Framework Specific to Freight

6.3.1 Can we account for major freight distribution centers in the region, using our fairly basic truck model? Is there a suggested high level freight model enhancement that we should consider incorporating in our scope of work? The panel offered a few overall freight-related observations:

  1. First, the panel understood that by "high level" TRPC desires a relatively low-cost means of improving the freight component of its model. Due to the intricacies of freight travel behavior and the challenges with getting relevant observed data, advanced freight modeling tends to be a significant financial investment.
  2. Since this level of effort is not within the TRPC budget at this time, the panel recommends not making any significant investments in the freight component of the travel model and instead recommends keeping it simple by relying on national parameters and existing data. Focus improvements on your major planning needs by ensure that freight flows on I-5 and US 101 are accurate. To this end consider engaging WSDOT for data acquisition support (and perhaps even model development support) given the statewide importance of these facilities. Employ Quick Response Freight Manual, second edition (QRFM2) techniques and parameters to make quick but well-founded improvements. Use more detailed employment categories in truck trip generation than simple total employment. A corollary suggestion is to ensure that the employment data used for truck trip generation is accurately categorized. For example, the Production Distribution and Repair (PDR) category should capture the freight distribution centers, but care should be taken to make sure that it is appropriately categorized in the TRPC land use data.

6.4 New and Existing Data Sources

6.4.1 Given that we are expanding the model boundary outside Thurston County; can we 'borrow' survey data from the adjacent regions? What do we need to keep in mind while dealing with surveys from different sources and different points of time? The PSRC 2006 HH Survey will be useful given the overlap in markets and the fact that PSRC survey records have specific locations for what PSRC considers to be "external" trip ends (some of which are likely to be in the TRPC region). A useful way to ensure compatibility between the PSRC 2006 HH survey and the upcoming TRPC survey would be to conduct an independent sample in the TRPC region using the PSRC survey design. The panel strongly recommended coordinating the "boundary expansion" of the model with the use of borrowed survey data and the targeted samples recommended below (military and outbound commuters) to ensure consistent representation in both model geography and collated survey data.

6.4.2 What aspects of the National HHTS are appropriate to use to augment our HHTS? The panel observed that the NHTS is not likely to be useful given a probable difference in trip length frequency distribution and the fact that the NHTS had a small sample size within the region.

6.4.3 We are experiencing pronounced growth in outbound commuting. How does that influence the design and analysis of our HHTS? Perform a targeted sample of outbound commuters.

6.4.4 Some parts of our County have experienced significant growth since the prior HHTS was conducted. Do we emphasize this area and collect more samples in these areas, while scoping our new HHTS? How to we take care of statistical validity across the region? First, to ensure representative demographics in the completed survey TRPC should consider a multi-frame sample, combining an address-based sample with a cell-phone sample. Second, geographic targeting is likely necessary but a proper overall survey design should sample from both high-growth and low-growth areas in order to provide statistically valid samples of both population groups.

6.4.5 The military population commuting from Thurston County to JBLM has unique travel patterns and constraints. Should we collect extra samples of this population? Yes, JBLM travelers should be the subject of a targeted sub-sample similar to the outbound commuting population cited earlier. These two groups are the most important targeted samples to obtain. The best approach would be a geographic over-sample of the areas where most off-base housing is known to exist.

6.4.6 Questions not to forget while designing the survey, given our vision for the model enhancements and how we plan to use our models? Capture all detail for all respondents (e.g. trip end locations for what TRPC now considers to be "external" trips); the entire state is the TRPC survey area in a very real sense given the interregional flows. Coordinate the model geographic expansion with the collation of borrowed surveys (PSRC) and targeted samples (e.g. JBLM) to ensure consistency.

6.4.7 What are the pros and cons of various technologies out there that we should consider in scoping our HHTS? Which of those would be best suited to our agency needs and why? Newer technologies are a must given the available budget. This means that the survey approach should be multi-modal - offering mail, phone and web options. With regards to some type of GPS collection, the most cost-effective approach would be a Smartphone application, similar to the SFCTA Cycle Tracks. A volunteer-based, smartphone-deployed instrument would capture the younger demographic group without costly special recruitment.

6.4.8 Additional panel discussion led to these observations/recommendations regarding data: A transit on-board survey is highly desirable, in same time frame as HH survey. It will be useful to mine the CTR data to supplement the new HHTS, particularly giving insights into the behavior of the in-bound commuters. The 2010 External Origin-Destination Survey is a critical source to mine, especially to inform the question of what geography to internalize within the full coverage of the model. In the longer term, TRPC should consider partnering with PSRC, WSDOT, and other western Washington agencies to conduct a future "Cascadia" survey covering both PSRC and TRPC regions together; one strategy for accomplishing this could be to do an "add-on" to the next PSRC HH survey. We also encourage PSRC to do an "add-on" to the next TRPC survey

6.5 Integration with simulation and Dynamic Traffic Assignment (DTA) tools

6.5.1 TRPC's vision is to develop and maintain a region wide Dynameq DTA model. We have a Dynameq model that we developed as part of our Smart Corridors project. Our strategy is to build on the existing model, and expand the model to some more priority corridors as part of this model update. Is this a valid approach? Does addition of more corridors to the model mean recalibration of the whole model? The panel observed that the network coding effort is low relative to the calibration effort; it can therefore be advantageous to proceed by coding all the base-year network detail across all priority corridors at one time. This would avoid the potential need for significant recalibration efforts if each corridor study were to add significant network coding detail. However, note that:

  1. One must have high-quality data with which to code network characteristics and understand the observed, base-year travel demand, including field data used for model calibration and validation: good data reduces the calibration effort. It should also be emphasized that field data for calibration and validation must be collected for key facilities and in general with good coverage of the study area or corridor.
  2. Freeways should be fully coded with all lane detail from the beginning; this provides a common and solid reference framework for individual corridor studies as needed.
  3. Basic calibration (not detailed calibration) should be done on the full system once it is completely coded.
  4. See the following question for more related to this topic.

6.5.2 What is the suggested frequency of model updates, given the dynamic nature of traffic operations and travel pattern in analysis corridors? It is wise to perform detailed calibration then validate the model in the study corridor at the time of the study; this allows use of the most recent data in the corridor. In relation to the previous point about model coverage expansion strategies, note that adding minor network detail should not require major re-calibration. It would be wise to craft institutional arrangements so that TRPC is notified of major operational changes that would require larger-scale updates and more detailed re-calibration (see San Diego's arrangement for distributed system data entry: the TRB Planning Applications conference in Reno had a presentation on it by Joaquin Ortega). Updates can then be data-driven on an as-needed basis rather than on a fixed, and thus potentially arbitrary, schedule.

6.5.3 In general, the panel endorsed the TRPC past and proposed uses of DTA as appropriate and useful.

6.6 Other Panel Observations/Recommendations

6.6.1 Are there any other low hanging fruit or significant improvements that our draft scope of work missed, and those we should consider including? The panel did make recommendations beyond the scope of the original questions in the answers listed above; these are repeated here for easy identification. An optional idea would be to replace trip generation submodels with a population synthesizer to enable use of distributed values-of-time, complemented with an expansion of user classes to match those of, for example PSRC, to allow assignment of more value-of-time groups. This will add appropriate sensitivity for pricing analysis. Consider enhancing the current work trip production and appropriate parts of the existing destination choice model with a usual workplace location choice submodel. Even at a coarse geography, this coupled with the "internalization" of geographies that produce or attract what are now considered to be "external" flows, would much improve the model's treatment of the major north-south flows that are now so challenging. Consider adding a time-of-day choice model. This would usefully complement the proposal of assigning peak periods instead of just peak hours by building in actual sensitivity to traveler and system factors that influence peak spreading and traveler response to pricing. Consider adding full treatment of Park and Ride (PNR) travel, including fixed route transit service, as a means to strengthening the mode choice model, complementing the desired treatment of vanpool/carpool travel, and properly internalizing the transit-using flows now considered "external" (the geographic expansion discussed above will enable this treatment).

[8] Walker, J.L., "Making Household Microsimulation of Travel and Activity Accessible to Planners," Transportation Research Record, 2005, No. 1931, pp. 86-98. See

[9] Direct communication with Mr. Dennis Bloom of Intercity Transit on 7/24/12

Updated: 10/20/2015
HEP Home Planning Environment Real Estate
Federal Highway Administration | 1200 New Jersey Avenue, SE | Washington, DC 20590 | 202-366-4000