The following text summarizes a point-by-point response to the topics of interest posed by ADOT at the commencement of the peer review. The comments and recommendations are provided in order of the questions posed by ADOT and the summary of this discussion follows the panel's final presentation to ADOT. After the topics of interest portion of this section, a summary is included containing the comments and recommendations of the peer panel on the other modeling topics discussed at the review.
The panel commented that the use of an abstraction method to estimate local transit in a statewide model was acceptable practice since the model would be used to test regional transit demand and not estimating local ridership or local transit route choice. Employing a transit abstraction method would reduce network coding at a statewide scale. The panel recommended pursuing applying the local bus abstraction method in the Arizona statewide travel models.
The panel also concurred with the practice of only using similar transit service types in an abstraction set when doing transit abstraction. Transit services with different characteristics, such as local bus vs. express bus, should not be combined in the same abstraction. The panel recommended following a pre-established structure when developing the local transit abstraction methods.
In applying the abstraction process, the panel suggested the consideration of variables, such as employment density as a factor for out of vehicle time. For reference, the California statewide travel model uses employment density as a parameter in the out of vehicle time function as well as population, HOV3 distance and transit LOS. The panel also commented that there may be a need for the representation of other services, such as dial-a-ride, which may be part of the local system.
Statewide Cordon Count Data
Much of the discussion centered on the merits of an intercept cordon survey vs. an approach that uses license plate capture and a mail in survey. The panel found that using a license plate survey with a follow up questionnaire did not have high response rates and the method was typically seen by the public as intrusive resulting in bad publicity. Also, the panel raised concern that in a license plate survey semi trucks with trailers would have the license plate obscured. This could prevent a mail in survey to be sent to the right address if trucks were to be included in the response survey.
The panel had better experiences using an intercept survey to collect cordon data. Michigan DOT notifies the public safety department to make them aware that the DOT is performing an intercept survey but does not ask to have police present while performing the survey. This is done to reinforce the idea that participation in the survey is voluntary. It was believed that having a police presence would cause drivers to think the study was mandatory. Ohio DOT also works with the public safety department in performing an intercept survey. The panel also commented that an effectively structured survey can minimize respondent burden. For instance, Ohio DOT has been able to survey respondents in under a minute, on average. The panel also shared advice on administering paper-based vs. electronic survey collection methods indicating that the paper-based works just fine and may be preferred. The panel also suggested that intercept surveys can more easily be obtained at international border crossings.
The panel recommended that if ADOT is allowed to stop vehicles then they do so. For low volume roadways, ADOT could stop traffic in the lane. For interstate or higher volume facilities, ADOT may want to consider only stopping a sample at a designated area such as a rest stop. If ADOT is not allowed to stop vehicles, then the panel recommended conducting a license plate survey. However, when conducting either an intercept or a license plate survey, the panel strongly recommended that ADOT not conduct a mail-in follow-up survey due to the panel's previous experience resulting in inconsistent data and bad public relations. The panel also recommended ADOT review potential new technologies, such as cell phone or bluetooth, as these may provide an alternate method to collect key data if vehicles cannot be stopped.
The panel also suggested that the timing of when a cordon survey is administered will be important to minimize seasonal bias, as the model represents an average weekday condition. The panel also suggested that combining various data collection methods could correct for this and other biases and validate average trip lengths as necessary.
Rural Household Travel Survey
The panel concurred that there was a need to obtain better household data in the rural areas of the state as well as better data for long distance trips. The panel recommended that ADOT collect this data but acknowledged that it would most likely not be available for the current model development deadline of March.
The panel also recommended stratifying the sampling plan by geography, such as by county, as well as by hard to reach groups, such as minorities or long distance trips. The panel suggested that ADOT could conduct a preliminary sample design study as a low cost way of getting a sense of the number of samples needed.
The panel also recommended ADOT look for opportunities to partner with local MPO's to share in their data collection efforts. Partnering with the MPO's would provide for more efficient data collection and greater consistency between data sources.
The panel commented that commodity-based mode choice modeling for freight can be difficult. Typically it is driven by a need, such as for distribution through major rail yards or for looking at economic viability of different modes. The panel recommended that ADOT begin by defining the questions they would like to answer with a freight model, such as:
The panel recommended scaling the complexity of the truck mode choice model to meet the need based on the questions being asked of the model. For instance, would the mode choice model be built to simply answer questions about truck volumes on highway or is the need to answer economic questions? If the need is to simply answer questions about trucks on the highway system, the panel suggested ADOT may want to consider implementing simpler mode choice freight model, such as a rule-based mode choice freight model, or a simplified network with GIS tagged rail/non-rail access links. Economic answers could require a more complex solution. The key is to understand the type of questions that need to be answered by the truck model.
The panel also commented that ADOT should be able to take advantage of the Transearch commodity database purchased regardless of the complexities chosen for the development of the freight mode choice model. The panel also cautioned that in the past there has been an issue with the TAZ level disaggregation of the Transearch data. The panel acknowledged that the issue may have been corrected in more recent releases of the data but recommended a check of the data to ensure usability at that scale.
The panel thought building a hybrid statewide-local model was a good idea and strongly recommended that ADOT pursue this development. Most of the panel members have similar functionality in the statewide models they use or maintain. The panel concurred with ADOT's recommended approach of maintaining data at a more disaggregate level. The panel also recommended ADOT look at creating a "window out" tool that could be used to perform the sub-area extraction from the statewide model.
The panel recommended that ADOT develop sub-area models that could be used for project-level forecasts. The panel's experience has shown that it is best to plan ahead with design-level results in mind as this is often the reason these sub-area models are run. The panel also suggested considering a combination of a sub-area travel demand model and a micro-simulation model in the solutions for these areas.
The panel expressed their accommodations to ADOT for the progress made by ADOT staff in establishing ADOT's modeling program and the development made to date on the current statewide model. The panel encouraged ADOT to continue to build from this good place and acknowledged that ADOT is definitely within the state of the practice in statewide modeling. The panel did want to caution against being on the bleeding edge of model development unless there was a specific need. They recommended that ADOT tie the development of advanced modeling techniques to specific needs moving forward.
Of the advanced modeling specifically mentioned by ADOT in the peer review, the panel made the following recommendations:
The panel recommended ADOT consider ties to both state and national economic models when looking at linkages between the statewide and economic models. Ohio, Maryland, Michigan and California statewide models have an economic component. The panel commented that ADOT would need to decide if the economic models tied to the front or back end of the statewide model and if/how the Transearch commodity flows would be used by the economic models.
The panel commented that true dynamic traffic assignment (DTA) implementation, which is to drill down to operational level fidelity, is only appropriate for subarea studies and is not computationally feasible for statewide model application. However, aspects of DTA may be useful for dealing with long trip lengths spanning model periods. A statewide DTA application would require development of a simplified approach that extracts the salient portions, such as time of day. The University of Maryland is experimenting with a low fidelity DTA which does not have signal controls but would have network performance by time of day. The panel recommended that ADOT consider why advanced models are needed, and then develop them according to available resources and ease of application.
The panel recommended looking into dynamic traffic assignment (DTA) due to the long distances in state travel. However, the panel commented that ADOT needed to specify the functions that would be performed by a dynamic traffic assignment and to be clear on the functional specifications. The panel also recommended ADOT consider collecting more detailed highway network data, such as signal locations and timing plans, in anticipation of implementing a dynamic traffic assignment.
The panel felt they needed more information in order to answer the activity based modeling (ABM) question. The panel wanted to better understand the need to move to an ABM platform. The panel wondered if there was something missing from the current models that a tour model would address. For instance, was the need to get a better assignment or was the need to better understand items such as transit, hot lanes or peak spreading. The panel felt these issues needed to be addressed first. The panel recommended that ADOT consider an open architecture to implement any ABM platform if ADOT decides to use an ABM. The panel also commented that an activity based approach would most like also entail an investment in resources and acceptance of longer runtime until technology caught up.
Along with the topics of interest presented, ADOT invited the peer panel to comment on the current model and the proposed model development plans. The comments made by the peer panel are summarized below by topic.
Markets
The panel commented that non-resident travel did not seem to be fully addressed in the statewide model. The panel recommended an assessment of the relative importance of including non-resident, visitor or tourist trips in the model.
Trip Generation
The panel commented that home-based school (HBS) and home-based university (HBU) trip purposes were unusual for a statewide model, though there are instances where this is the case such as the California statewide model which has "school" as a trip/tour purpose.
Long Distance Person Travel
It was proposed that the long distance personal travel in AZTDM3 will be based in part on county-to-county flows and employment (FHWA traveler analysis framework method). The panel suggested that county to county flows from the latest Census may not be available in time to meet the AZTDM3 development schedule deadline.
Freight Model
The panel liked that the AZTDM2 borrowed MAG's short distance freight model and recommended ADOT use MAG's freight data to complement the FAF3 data in recalibrating their freight models. The panel also commented that a rule of thumb threshold for long distance trucks is about 250 miles.
Time of Day
The AZTDM3 transit model will include peak, off-peak and a separate overnight period when limited service is provided. The panel suggested ensuring that the model is reconciled between the highway peak periods, the peak periods when transit operators provide service, and the peak periods when people actually use the transit system.
A decision has not been yet made as whether AZTDM3 will use the midpoint of the travel time for a trip or a trips-in-motion method for defining the trip's period (AZTDM2 uses trips in motion). The panel acknowledged that this is an important issue and commented that using midpoint may not be as useful given the long duration expected in statewide model trips. The panel further commented that a trips-in-motion method seems to make sense if dynamic traffic assignment is being considered. The panel wondered if trips could be parsed so that they appear in multiple time periods.
The panel also commented on the proposed peak spreading model in AZTDM3. The peak spreading model is expected to be a multinomial logit model that breaks up the peak periods into 30 minute intervals. The model would use a variety of household, person and trip variables that are specific to each period. The panel questioned if local data existed to measure the shift due to congestion and commented that ADOT may need to look for outside data sources to support this model. The panel also commented that ADOT consider employing a measure of congestion, such as congested travel time, as an explanatory variable. There was some discussion regarding the use of speed versus time as the explanatory variable. Though there was no consensus reached, the panel did recommend using time as the measure.
Mode Choice Model
AZTDM3 will use mode choice logsums in distribution. There was much discussion on whether mode choice should come before or after distribution and how that might be implemented. The question was asked if the mode choice model was applied at the daily level or if it was planned to split the trip tables into periods before mode choice.
The panel offered several suggestions as to the implementation order of the time of day, mode choice and distribution models. The panel commented that applying time of day factors before distribution or mode choice would be needed as skims are performed by period. The panel also suggested that time of day factoring could be split into two parts, one up front in the model for micro-level assessments and the second at the back end similar to the adjustments made in dynamic traffic assignment or activity based models. Alternatively, the panel suggested beginning the modeling process with preloaded congested networks to reduce the number of iterations.
The panel was in favor of having mode choice come after distribution; however, acknowledged that the decision would ultimately depend on what the estimation results say about the chosen hierarchy. Decisions more sensitive to travel conditions should be later on in lower nests of the model. Mode is often more sensitive than destination, but not always. Poor estimation results can indicate an incorrect ordering of model steps. Estimation results may also vary by socio-economic category.
Socioeconomic Forecasts
The panel commented that the review did not include much discussion of the socioeconomic forecasts being used by the statewide model. The panel suggested the need to pay attention to the socioeconomic forecasts to ensure a certain comfort level as it pertained to reasonableness of the data and the process by which the forecasts were derived.
Source Data
ADOT plans to use the recent release of the American Community Survey (ACS) in AZTDM3 model development. The panel commented that the 3 year data may have missing data due to sample size suppression and recommended using both the 3 year and 5 year ACS data sources.
