In addition to the immediate recommendations, the peers recommended that many other significant improvements be made to the model in the short- (2002) and long-terms (2005). Representatives of the Sierra Club were informed in an October 1999 meeting that improvements 2 to 7 would be completed for the final environmental impact statement (FEIS) analysis of the proposed Legacy Parkway and I-15 North Corridor projects. However, in a telephone interview with Mick Crandall of WFRC in March 2000 for this report, we were informed that less that half of the immediate improvements would be incorporated in the WFRC model used in the FEIS. We requested documentation of the revised WFRC model and were told that this documentation would be made available by the middle of March 2000. Unfortunately, as of the end of March, we have only been provided with copies of overheads that roughly describe some of the changes in the revised WFRC model. Therefore, this report can only include a partial, preliminary assessment of the revised model.

The revised WFRC model includes the following immediate improvements recommended by the peers (2 and 5 above):

An auto ownership submodel model that reflects urban design and accessibility variables and affects the trip generation and mode choice submodels and

A merged Salt Lake City, Ogden, and Provo trip distribution submodel, but not a merged trip generation submodel.

The revised WFRC model will not include the following (1, 3, and 4 above):

A Delphi review of land use allocations that would have provided a separate set of land use projections to accompany each transportation alternative;

AM Peak, PM Peak, and midday trip tables that would have greatly improved the accuracy of the travel time estimates in the model and facilitated improved feedback mechanisms for trip distribution; and

A feedback mechanism to trip distribution using the successive averages approach for convergence; instead, the direct method of feedback was used, which is known not to always converge.

A travel demand model steering committee was convened. At that meeting (January 2000), new documentation of the WFRC model was provided. However, this documentation lacks goodness-of-fit statistics for the model, where applicable. As mentioned above, complete documentation for the revised WFRC model has not yet been provided.

Because less than half of the immediate improvements recommended by the peer review committee have been made to the revised WFRC model, we can only conclude that the model is insufficiently accurate for plan or project analysis. More specifically, the use of a single set of land use forecasts for each transportation alternative biases analyses conducted with the revised WFRC model in favor of highway expansion alternatives by underestimating vehicle miles traveled (VMT) and by overestimating congestion reduction compared to the no-build scenario. Furthermore, the use of a flawed feedback mechanism for trip distribution renders suspect any analysis performed by the model; that is, convergence must be demonstrated for each alternative for the results to be considered valid. Finally, WFRC has not provided goodness-of-fit statistics for their model, where possible, and thus its statistical strength cannot be assessed.

In the next section, a more detailed discussion of the implications of the improvements and the omitted improvements with respect to the results of the revised WFRC model will be provided.

EVALUATION OF IMMEDIATE IMPROVEMENTS AND OMMITTED IMMEDIATE IMPROVEMENTS

Conduct a Delphi Review of Land Use Allocations
Land use forecasts are key inputs to travel demand models, and the failure to represent land use changes resulting from new highway capacity is likely to result in an underestimation of auto trips, VMT, and congestion. The I-15 North Corridor and Legacy Parkway projects each significantly expand highway capacity in the region and connect an urbanized area to a relatively undeveloped suburban area. As a result, their effect on land use over twenty years may be significant. The best way to address this issue is to use a land use simulation model (e.g., the Urban Sim model) that is linked to a greatly improved WFRC travel model or an integrated land use and transportation model (e.g., MEPLAN or TRANUS). However, the use of a formal Delphi or expert panel review of land use allocations is an inexpensive and quick method to account for the interaction of land use with transportation alternatives. Such a method could be used to account for the land use and transportation interaction while work on the Urban Sim land use model is completed and the WFRC travel model is improved to meet the recommendations of the peer review committee. The peer review committee recommended that a Delphi review of land use be done immediately.

The land use changes that result from new highway capacity can be a significant effect of induced travel, which is defined as the increase in VMT that can be attributed to expansion of highway capacity. Recent research has provided persuasive evidence for the existence of induced travel. The U.S. Environmental Protection Agency (EPA) recently conducted a review of the induced travel literature for their Science Advisory Board and concluded that this research "has not only built a strong case for the existence of induced travel effects, but in some cases suggests that a large fraction of growth in VMT is directly attributed to increases in road capacity" (EPA, 2000). In addition, leading transportation researchers in the Transportation Research Board's 1995 report, Expanding Metropolitan Highways, have concluded that:

Highway capacity additions that reduce travel time and the day-to-day variability in travel time will induce increased highway use as long as travel times are shorter and the reliability of motor vehicle time is improved, all else being equal.

Additions to highway capacity that reduce the cost of travel have decentralizing effect on urban development. This effect is most pronounced when the added capacity provides improved access to developable land in outlying areas.

One of the difficulties of testing the induced travel hypothesis has been controlling for confounding economic activity variables such as population, income, and other demographic trends. Much of the recent induced travel research, however, has controlled for these variables and has not been able to reject the hypothesis of induced travel (Goodwin, 1996; Hansen and Huang, 1997; Noland and Cowart, 2000; Chu, 2000; Fulton et al., 2000; Noland, 2000). The results of this research have yielded fairly consistent long-term elasticities of VMT with respect to roadway lane miles. See Table 1 below.

Table 1. Long-term elasticities of VMT with respect to roadway lane miles.
Source
Geographic region
Elasticity range

Hansen and Huang, 1997
County and metropolitan area
0.3 to 0.7 (county)
0.5 to 0.9 (metropolitan)

Noland and Cowart, 2000
Metropolitan area
0.8 to 1.0

Fulton et al., 2000
County
0.5 to 0.8

Noland, 2000
State
0.7 to 1.0

Thus, for example, a 1% change in highway lane miles would be expected to results in a 5% to 10% increase in VMT for a metropolitan region in a twenty year time horizon.

Induced travel effects include changes in land development and the location of households and employment, the number of trips made, destination choice, mode choice, route choice, and departure time choice (Transportation Research Board, 1995; Transportation Research Circular, 1998). In the long run, changes in land use can have a significant effect on total VMT (EPA, 2000).

In a recent study, Rodier, Johnston, and Abraham (2000) evaluate the contribution that the representation of the land use and transportation interaction makes to travel and vehicle emissions analyses of transportation scenarios in the Sacramento, California, region over a 25-year time horizon. One of the more theoretically consistent and practical operational land use and transportation models, MEPLAN, was used to simulate a trend base case, high occupancy vehicle lanes (HOV), and beltway freeways. These transportation scenarios were simulated, first, with the full MEPLAN model to represent the land use and transportation interaction of the scenarios and, second, with the distribution of activities held constant from the future base case scenario so that the interaction was not represented. The failure to represent the land use and transportation interaction from the HOV and Beltway scenarios significantly altered the magnitude of change for both travel and emissions results and the rank ordering of scenarios for emissions results. Travel speeds were overestimated, and vehicle hours of travel, VMT, and emissions were underestimated when the land use and transportation interaction was not represented in the HOV and Beltway scenario. Emissions were projected to decrease for some pollutants in the HOV and Beltway scenarios when land use changes were not simulated, and when they were, emission of all pollutants were projected to increase. In most cases, the error due to the absence of the interaction was greater than the percentage change from the trend base case to the HOV and Beltway scenarios when the interaction was not represented.

The stated goal of the proposed Legacy Parkway and I-15 North Corridor projects is to reduce congestion. Research on induced travel, however, suggests estimates of congestion resulting from new highway projects may be underestimated if forecasting methods do not adequately account for induced travel (EPA, 2000). A U.S. District Court Case in the Chicago region recently held that the National Environmental Policy Act requires the consideration of land development changes when a new freeway segment is analyzed in an EIS (Sierra Club, et al. v. U.S. DOT et al., No. 96 C 4768, U.S. Dist. Ct. for the N. Dist. of Illinois, E. Div., Jan. 16, 1997).

Our evaluation of the revised WFRC model and the results of the old WFRC model (which should not be much different from the revised model because very few changes were made), suggests that this model does not adequately represent the effects of induced travel. The model forecasted a decrease in VMT for the highway alternatives examined (I-15 North Corridor DEIS, 1998, pg. 4-19: Legacy Parkway DEIS, 1998, pg. 4-55). In light of the recent evidence on induced travel, this renders the model's results highly suspect.

Auto Ownership
The addition of an auto ownership submodel to the WFRC travel demand model is an important first step in its overall improvement. This auto ownership submodel has been borrowed from Portland, Oregon and includes an urban design variable, accessibility variables (retail employment within one mile and jobs within 30 minutes of transit access), and other socioeconomic variables (household size, income, and workers per household). This auto ownership step affects the trip generation and mode choice steps. Because of this improvement, the revised WFRC model would tend to more reasonably represent increased auto trips due to highway project alternatives that reduce the travel time and cost of auto use. However, because of other problems with the revised WFRC model (described below) changes in accessibility among alternatives simulated in the model are not likely to be very accurate or significantly different.

Trip Distribution
Joint Salt Lake City, Ogden, and Provo model. The peer review committee recommended that the Salt Lake City, Ogden, and Provo travel demand models be merged into one unified model. The revised WFRC model has merged the trip distribution model, but they have not merged the trip generation model.

AM Peak, PM Peak and midday trip tables. To estimate peak period travel times, the revised WFRC model still assigns a daily trip table to the highway network and uses a single peak hour factor, 12% of average daily traffic, to all links in the region. The peer review committee recommended as an immediate improvement that AM peak, PM peak, and midday trip tables be included in the WFRC model. Such an improvement would also facilitate improvement of the feedback mechanism from traffic assignment to trip distribution.

In their manual of best modeling practices for metropolitan planning agencies, Harvey and Deakin (1993) state that the use of 12% average daily traffic is a "crude estimate of total bi-directional peak hour travel." Furthermore, "this procedure yields only a rough approximation of link-or corridor-level peaking" and should only be used by smaller MPOs "where the duration and intensity of congestion are limited." Despite this recommendation, WFRC continues to use the 12% factor in a medium-sized region that is using their model for the purpose of addressing growing congestion. Therefore, the use of the 12% factor is clearly not appropriate for use in the revised WFRC model, and AM peak, PM peak, and midday trip tables should be developed.

Implement a feedback mechanism to trip distribution using the successive averages approach for convergence. WFRC has not implemented this peer review recommendation in the new version of their model; instead, they continue to use the direct method of feedback, which is known not always to converge.

The trip distribution model in the travel demand modeling process requires estimates of travel time in order to estimate the destination of trips. However, travel times depend on the level of congestion on streets in the network. The level of congestion is not known until later model steps are executed (i.e., assignment of traffic on the roadway network). Thus, it is important to feed travel times from the traffic assignment step back to the trip distribution step. This process of recycling travel times once from traffic assignment to trip distribution is commonly called an iteration. Many iterations of a travel model may be required to ensure that the input travel times in trip distribution are consistent with the output travel times from traffic assignment. Modelers must employ criteria to determine when to stop iterating the model. They are called convergence criteria. The incorporation of feedback into the travel modeling process will tend to decrease average travel times, average travel distance, average volume/capacity ratios, and total VMT. If a travel model is not fully equilibrated on travel times, then its simulations may be biased in favor of roadway expansion scenarios by exaggerating VMT and congestion in the no-build scenario.

The peer review committee recommended that the successive averages approach for convergence be implemented as the feedback mechanism to trip distribution because it is known to consistently converge and it is an efficient feedback method. In this approach, the volumes on each link and each origin destination flow are averaged with previous results. The direct method used by WFRC is known not to always converge. We have reported on WFRC's failure to apply formal criteria to determine whether their model has converged (Rodier and Johnston, 1998). We applied several convergence criteria recommended by the Federal Highway Administration's (FHWA) Incorporating Feedback in Travel Forecasting (1996) and found that the model did not meet 2 of 3 convergence criteria (Rodier and Johnston, 1998). FHWA recommends the application of more than one criterion to demonstrate convergence. When a model does not converge, its theoretical validity, and thus its results, are questionable. We asked Mick Crandall of WFRC what the convergence criteria for the revised model was, and he was unable to tell us (Crandall, 2000). The convergence criteria must be documented to verify the validity of the results of the revised WFRC model.

We have also reported that the direct feedback method has also been incorporated in the WFRC model in an unsystematic fashion (Rodier and Johnston, 1998). The model was programmed to iterate only once; however, this iteration required an input network that contained the results of traffic assignment from a previous model run. This input network could be the product of anywhere between one to four model iterations. If this loaded network is not a product of sufficient runs, then it is likely that the model will not have converged. We asked Mick Crandall of WFRC how the feedback mechanism had been formalized in the revised WFRC model, and he was unable to answer this question (Crandall, 2000). This information must be provided in the documentation of the revised model.

Model Documentation
As recommended by the peer review committee, a travel demand model steering committee was convened in January 2000. At that meeting new documentation of the old version of the WFRC model was provided. However, this documentation lacks goodness-of-fit statistics for the models estimated locally. Goodness-of-fit statistics tell us how well the estimated model fits actual values; that is, whether the fit is statistically significant. The trip generation model was developed with local data and documentation of goodness-of-fit statistics is needed for this submodel. The auto ownership and mode choice submodels are transferred from other regions. Goodness-of-fit measures (e.g., comparison of modeled vs. observed results) and documentation of adjustments to constants are needed for these calibrated models. WRFC has provided documentation of goodness-of-fit measures for their mode choice model but has not provided a documentation of the adjustment to the model constants.

Complete documentation of the revised WFRC model has not yet been provided. We were told that it would be provided to us in the middle of March, but instead we received only copies of overheads with brief descriptions of some of the improvements. These overheads provide goodness-of-fit measures for the transferred auto ownership submodel and the new trip distribution submodel, but here is no discussion of the adjustments made to the auto ownership constants or whether k-factors were used in trip distribution. It is disturbing that complete documentation of the model has not been released because the FEIS, in which the revised model is used, is scheduled for release in April.

EVALUATION OF SHORT- AND LONG-TERM RECOMMENDATIONS

The peer review recommended that many other significant improvements be made to the WFRC model in the short-term and long-term. The WFRC model required major improvements, and only the most important could be made immediately. However, many of the short- and long-term improvements could significantly affect the results of the evaluation of competing alternatives in the FEIS. The proposed projects could have significant long-term regional consequences. The communities affected deserve a sound and fair evaluation of the alternatives. To ensure such an evaluation, additional time to incorporate model improvements should be taken.

Moreover, all the improvements recommended by the peer review committed are necessary in the WFRC travel demand model to iterate it with the new land use model, Urban Sim, which is currently being calibrated to the region. This would enable the representation of the effect of changes in the transportation system on land development and the location of employment and households. These improvements are needed to adequately represent modal time and costs throughout the WFRC travel model hierarchy. The current model's representation of time and cost is so limited that there will be little change in zone-to-zone impedance (or the travel time and cost from trip origins and destinations) from one transportation alternative to another. As a result, there will be very little difference among land uses for transportation alternatives.

Trip Distribution
The peer review committee suggested that in the long-term (by 2005) WFRC should explore the inclusion of a logit trip distribution model in the travel demand model. Trip distribution in the WFRC model uses auto travel time and distance as its measure of ease of access from origin to destination locations. As a result, this model would be insensitive to changes in travel patterns that may results from significant expansion of modes of travel other than the auto, for example, transit. The model may predict an overly wide distribution of trips and overestimate VMT for transit alternatives. WFRC should consider the use of a logsum of the denominator from the mode choice model as its measure of impedance in the trip distribution model. This would capture the travel time and cost of all available modes in trip distribution. Because this improvement would improve the model's sensitivity to alternative modal travel time and costs, we would argue that this should be included as an immediate improvement to the model.

Mode Choice
In the short-term, the peer review committee suggested that WFRC should consider using a transferable mode choice model calibrated to local data. In addition, the model should include socio-economic variables (e.g., number of workers in a household and income), urban design variables, and walk and bike modes. In the long-term, they recommend that results from stated preference surveys be used to improve the sensitivity of the model to modes that do not currently have a large share in the region (e.g., bus, light rails, and bike) and that rail be established as a separate mode.

All of these improvements would greatly enhance the sensitivity of the mode choice model to alternative policies that change the quality of the auto, transit, walk, and bike modes, and thus would more fairly simulate scenarios other than highway alternatives. Moreover, improvements in the representation of travel time and cost by mode (or accessibility) for competing scenarios are key to valid integration of the WRFC model with the Urban Sim model. The Urban Sim model will use modal accessibility by zone pair from the mode choice model as the accessibility inputs to the land use model (Crandall, 2000).

Traffic Assignment
The peer review committee recommended that in the short-term HOV assignment capabilities and peak spreading analysis should be included in the traffic assignment submodel.

The WFRC model is used to evaluate HOV lane alternatives; however, HOV lanes are not coded separately in the network. Instead, assumptions are made about the reduction in VMT as a result of the construction of HOV lanes. To verify these assumptions, use of HOV lanes needs to be simulated in the model by including HOV lanes in the network and by using an HOV lane use model.

A peak spreading component is needed in the traffic assignment submodel to represent how traffic congestion eases and disappears over time. The WFRC model would currently tend to overestimate congestion in the no-build scenario and underestimate it in the build scenario because peak spreading is not represented. A peak spreading model has been implemented by the Metropolitan Planning Commission for the San Francisco Bay Area in the Tri-Valley sub-area study.

CONCLUSIONS

Because less that half of the immediate improvements recommended by the peer review committee have been made to the revised WFRC model, we can only conclude that the model does not meet the recommended standard for model accuracy established by that committee. More specifically, the use of a single set of land use forecasts for each transportation alternative biases analyses conducted with the revised WFRC model in favor of highway expansion alternatives by underestimating VMT and by overestimating congestion reduction compared to the no-build scenario. Furthermore, the use of a flawed feedback mechanism for trip distribution renders suspect any analysis performed by the model; that is, convergence must be demonstrated for each alternative for the results to be considered valid. Finally, WFRC has not provided sufficient goodness-of-fit statistics for their models, and thus its statistical strength cannot be assessed.

The U.S. EPA has concluded that recent research on induced travel provides strong evidence for its effects and that a significant portion of growth in VMT results from highway construction (EPA, 2000). Our evaluation of the WFRC travel model indicates that this model does not adequately represent the induced effects of changes in land use, changes in trip making, or changes in trip lengths resulting from expansion of highway capacity in the roadway network. The results of the old WFRC model (which should not be much different from the revised model because very few changes were made) confirm our evaluation. This model forecasted a decrease in VMT for the highway alternatives examined (I-15 North Corridor DEIS, 1998, pg. 4-19: Legacy Parkway DEIS, 1998, pg. 4-55). We can only conclude that the WFRC model is inadequate to evaluate the travel effects of the proposed I-15 North Corridor and the Legacy Parkway.

REFERENCES

Chu, Xuehao. Highway Capacity and Areawide Congestion. Preprint for the 79th Annual Meeting of the Transportation Research Board. National Research Council, Washington, D.C. 2000.

COMSIS Corporation. Incorporating Feedback in Travel Forecasting: Methods, Pitfalls and Common Concerns. Office of Environmental Planning, Federal Highway Administration, DOT-T-96-14, Washington, D.C. 1996.

Crandall, Mick. Telephone interview of Mick Crandall of WFRC by Caroline Rodier. March 2000.

Fulton, Lewis M., Daniel J. Meszler, Robert B. Noland, and John V. Thomas. A Statistical Analysis of Induced Effects in the U.S. Mid-Atlantic Region. Preprint for the 79th Annual Meeting of the Transportation Research Board. National Research Council, Washington, D.C. 2000.

Goodwin, Phil B. Empirical evidence of induced traffic, a review and synthesis, Transportation, 23:35-54. 1996.

Hansen, Mark and Yuanlin Huang. Road supply and traffic in California urban areas, Transportation Research A, 31:205-218. 1997.

Harvey, G. and E. Deakin, A Manual of Regional Transportation Modeling Practice for Air Quality. National Association of Regional Councils. 1993.

Noland, Robert B. Relationships between highway capacity and induced vehicle travel. Transportation Research A (forthcoming). 2000.

Noland, Robert B. and Cowart, William A. Analysis of Metropolitan Highway Capacity and the Growth in Vehicle Miles of Travel. Preprint for the 79th Annual Meeting of the Transportation Research Board. National Research Council, Washington, D.C. 2000.

Rodier, C., R. Johnston, and J. Abraham. Air quality analysis of transportation plans: is it important to model the land use effects? Department of Environmental Science and Policy, University of California, Davis, CA. 2000. Sacramento Area Council of Governments (1996). 1995 Regional Housing, Population, and Employment Projections. Sacramento, CA.

Transportation Research Board. Expanding metropolitan highways: implications for air quality and energy use, Special report 245, National Research Council, National Academy Press, Washington, D.C. 1995.

Transportation Research Circular. Highway Capacity Expansion and Induced Travel: Evidence and Implications. Transportation Research Board. National Research Council. Number 481. February 1998.

U.S. Environmental Protection Agency. Induced Travel: A Review of Recent Literature with a Discussion of Policy Issues. Office of Policy. Energy and Transportation Sectors Division, Washington, D.C. 2000

Wasatch Front Regional Council. Travel Demand Model Peer Review. Summary of Proceedings. May 18 and 19, 1999.