National Academies Press: OpenBook

Integrated Transportation and Land Use Models (2018)

Chapter: Chapter 7 - Choosing a Model

« Previous: Chapter 6 - Spatial Input-Output Land Use Models
Page 56
Suggested Citation:"Chapter 7 - Choosing a Model." National Academies of Sciences, Engineering, and Medicine. 2018. Integrated Transportation and Land Use Models. Washington, DC: The National Academies Press. doi: 10.17226/25194.
×
Page 56
Page 57
Suggested Citation:"Chapter 7 - Choosing a Model." National Academies of Sciences, Engineering, and Medicine. 2018. Integrated Transportation and Land Use Models. Washington, DC: The National Academies Press. doi: 10.17226/25194.
×
Page 57
Page 58
Suggested Citation:"Chapter 7 - Choosing a Model." National Academies of Sciences, Engineering, and Medicine. 2018. Integrated Transportation and Land Use Models. Washington, DC: The National Academies Press. doi: 10.17226/25194.
×
Page 58
Page 59
Suggested Citation:"Chapter 7 - Choosing a Model." National Academies of Sciences, Engineering, and Medicine. 2018. Integrated Transportation and Land Use Models. Washington, DC: The National Academies Press. doi: 10.17226/25194.
×
Page 59
Page 60
Suggested Citation:"Chapter 7 - Choosing a Model." National Academies of Sciences, Engineering, and Medicine. 2018. Integrated Transportation and Land Use Models. Washington, DC: The National Academies Press. doi: 10.17226/25194.
×
Page 60
Page 61
Suggested Citation:"Chapter 7 - Choosing a Model." National Academies of Sciences, Engineering, and Medicine. 2018. Integrated Transportation and Land Use Models. Washington, DC: The National Academies Press. doi: 10.17226/25194.
×
Page 61
Page 62
Suggested Citation:"Chapter 7 - Choosing a Model." National Academies of Sciences, Engineering, and Medicine. 2018. Integrated Transportation and Land Use Models. Washington, DC: The National Academies Press. doi: 10.17226/25194.
×
Page 62

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.

56 This chapter is designed to help agencies considering the implementation of an integrated land use/transport model to select the modeling approach best suited for their particular requirements. Section 7.1 summarizes general considerations for developing integrated land use/transport models. Section 7.2 compares various model types and attempts to identify major strengths and weaknesses of each type. 7.1 Model Implementation Criteria Table 7-1 summarizes topics mentioned in the interviews. The summaries are relevant to all model types. Many of the comments apply to transportation modeling in general; given that integrating land use and transport models tends to make the modeling structure more complex, these topics are even more relevant. Findings are presented in no particular order. 7.2 Model Type Evaluation Matrix Ultimately, the goal of this synthesis report is to provide information for agencies to use in implementing an integrated land use/transport model. Table 7-2 summarizes the strengths and weaknesses of four options: (1) the “no land use model,” (2) sketch planning models, (3) micro- simulation discrete choice models, and (4) spatial input-output model. This section, which is based on the interviews, presents considerations in selecting the modeling approach, given the agency’s needs for analysis. This chapter does not recommend a specific land use model. No one model will fit every agency’s needs. Available models are diverse, and each has strengths and weaknesses. Once a model type has been selected, it is advisable to discuss the models’ costs, data requirements, scenario analysis capabilities, and support options with each model developer in this model type segment. Costs are particularly difficult to assess, because model development and maintenance expen- ditures vary widely among agencies. Early adopters, such as Ohio DOT and Oregon DOT, have spent significantly more on implementing integrated land use/transport models than agencies that started using such models after they became more mature. Varying model specifications make costs paid by different agencies difficult to compare. The level of staff involvement sub- stantially affects model implementation and maintenance costs. Agencies with sufficient staff members who are well trained may pay significantly lower consulting fees for modeling than agencies that have to operate with very few staff members. Typical cost elements of integrated land use/transport modeling include • Hardware. • Software licensing fees, if not open source. C H A P T E R 7 Choosing a Model

Choosing a Model 57 Table 7-1. Important factors in selecting an integrated land use/transport model. Topic Relevance Considerations Open source Relevant for agencies that intend to revise code themselves and for agencies that want the flexibility to switch consultants. Open source also helps users understand what is happening in the model when documentation is sparse. Many land use models (and, increasingly, transport models) are provided as open source. Even if an agency does not intend to change code, using open-source tools helps open black boxes and allows review of code to better understand the internal functioning of a model. Developer dependence Most successful model implementations were done in cooperation (usually in the form of a consulting contract) with the model developer. Too often, only the model developer, or someone who has worked closely with the developer at some point, can fully comprehend the complexity of the model. This is particularly an issue if the model developer retires or ends support for other reasons. Although working with the model developer creates a certain level of dependence, the initial implementation has rarely been successful without involving the model developer. In subsequent iterations of model updates and improvements, agencies may be able to work without the model developer, particularly if the agency has in-house programming expertise. For models that are more frequently applied, it will be easier to find staff qualified to work with the model. Modeling expertise A common challenge in modeling is to build the expertise to use models in house and reduce dependency on consultants. Consulting support is useful for model development and implementation, but, in the long run, agencies should build the capacity to run models independently. It was encouraging to hear in the in-depth interviews that all seven surveyed agencies had the capability and capacity to run scenarios in house. This turns out to be a key success factor. If the agency can operate the model without outside help, more scenarios at lower costs can be run and knowledge is built on how to use the model. This is relevant expertise that agencies should not outsource. Programming expertise Ideally, staff members who work on integrated land use/transport modeling are also trained in programming. In reality, however, most agencies have no or limited programming expertise. Agencies who have staff with programming expertise have built some of the most renowned models (for example, MetroScope in Portland or GreenStep in Oregon). Lack of programming expertise increases dependency on external consultants to provide model updates and revisions. For continuity, most agencies would benefit from in-house programming expertise. This requires upper management to provide training, staff members to be willing to learn new skills, and universities to train future modelers in programming. Model runtime The runtime of the model defines how many times the model can be run for calibration and testing, as well as how many scenarios can be modeled. In this regard, faster models are preferable. More powerful models typically offer greater precision but are more complicated and so tend to come with longer runtimes. Model developers tend to focus on improving model features and pay less attention to model runtime. For model users, long runtimes may severely affect the usability of the model. Before selecting a model, agencies need to consider the runtime of various models. Agencies may want to specify a maximum acceptable runtime at the outset and build a model that works as well as possible within this runtime constraint. Agencies using a consultant to set up a model are advised to agree on an approximate runtime in advance. Expert panel A panel of experts who advises on model implementation and reviews model results likely increases model acceptance throughout the region. Such a panel may also help to discover flaws in model design or input data. Although the panel should be staffed with experts who understand how to use model results, some diversity in backgrounds is helpful to add various viewpoints. Members could include land use planners, real estate experts, highway planners, transit agency representatives, environmentalists, academics, and utilities providers, among others. Upper management Support from upper management is generally needed at every institution engaged in land use/transport modeling, both in terms of providing funding and managerial support. Interviews showed that the most successful implementations of integrated land use/transport models happened where upper management provided two things: funding and the trust that modelers will make the right decisions. Micromanagement usually limits model advancement. Although upper management should provide guidance to staff members, agencies tend to be more successful if staff members are authorized to make model design and implementation decisions. Transparency Most agencies that conduct integrated land use/transport modeling (including all agencies interviewed for this research effort) report their model results to several member agencies (usually counties, but also transit agencies and planning departments or sewage providers, among others). If member agencies understand the input data and how the model works, model results are likely to be accepted more easily. Sharing model design, model results, data collection methods, and estimation approaches helps build trust in the model as a useful decision support tool. Long-term modeling vision Implementing an integrated land use/transport model is an endeavor that occupies management and staff members over several years. Having agreed on a long-term vision increases the likelihood of consistency in funding and support in the long run. Providing staff members with a long-term modeling vision helps to build trust that work will be supported in the long run. Closely related is staff members taking ownership of their modeling work. The vision also helps in securing funding for staff, data purchases, and, if needed, consulting support.

58 Integrated Transportation and Land Use Models Topic Options Implications Motivation No land use Not implementing a land use model is still the most common approach for transportation agencies. In part, this is because most agencies that run transport models are not authorized to do land use planning. Also, land use modeling often is perceived as more complex than transport modeling, in part because of its long-term nature. However, agencies that skip land use modeling severely limit their transport model’s sensitivity for many scenarios. Furthermore, forecasts often are provided at coarse spatial resolutions (such as counties) that need to be disaggregated to TAZ. The most common approach for this disaggregation is a more or less manual process that is laborious, error-prone, often uses nebulous methods, and would most likely lead to different results if conducted by a different analyst. Sketch planning Sketch planning models are particularly useful for assessing the land area needed to accommodate growth under varying density assumptions. Because these models tend to run faster, they are often used in public meetings to demonstrate the effects of scenarios developed in such meetings. Such models have also been used to successfully disaggregate forecasts to finer geographies, such as from counties to TAZ. Using such models makes this disaggregation step transparent and repeatable. Microsimulation discrete choice Microsimulation discrete choice models aim at representing location choice behavior and demographic transitions realistically. Based on psychological behavioral theory, utilities of different choices are compared against each other. The random utility theory concept allows accounting for many detailed attributes, if data and theory can be provided. In theory, microsimulation allows for unlimited market segmentation. For example, if the analyst wants to distinguish the behavior of government workers from other workers, this attribute could be added to the utility evaluation. The flexibility of these models is a particular strength, but brings also the risk of making such models overly complicated. Spatial input- output Spatial input-output models are particularly good at representing equilibrium between land use and prices. In a free-market economy where redevelopment – if profitable – may happen within a few years, the spatial input-output model is strong at representing this evaluation of who is willing to pay how much to develop a certain site. This benefit comes with the shortcoming that the real world is not at equilibrium, but, at best, moving toward equilibrium. Both the demographics of a given household and the environment in the form of accessibilities, job market, and prices change over the years. The spatial input- output model commonly simplifies reality and assumes immediate development of sites and relocation of users. These models commonly also provide flows of labor (which may be translated into commute trips) and flows of commodities (which may be translated into freight flows). Scenario analysis No land use Transport scenarios should not be run to affect choices in households and firms. In theory, every transportation project or policy could affect location choice. The model user needs to be aware that the pure transportation model may miss important land use developments. For example, a new highway to mitigate congestion may fill up quickly because the added accessibility entices more households to move near this new highway. As a consequence, congestion may end up being as bad as before building the highway. The analysts need to be aware that the model may miss this kind of land use/transport feedback. Sketch planning Sketch planning models are particularly strong at assessing the effects of various density assumptions. Creating housing for 1,000 people in single- family homes may require ten times as many acres as providing housing for the same population in multi-family buildings. Such effects can be pictured impressively with sketch planning models. Zoning scenarios can easily be implemented. Such models also have been used to show the attractiveness of living near transit stations. However, the sensitivity to transit proximity usually is set heuristically and does not relate to the actual transit use of residents in that area. For high-level analysis, however, this may be sufficient. Microsimulation discrete choice The number of scenarios that can be modeled with microsimulation discrete choice models is theoretically infinite, because the individual behavior can be represented in great detail. In reality, models have to be built to be sensitive to certain scenarios—this limits the number of scenarios for each individual model substantially. Typical scenarios include zoning, development subsidies, housing subsidies, and varying population and employment growth assumptions. Table 7-2. Comparison of four land use modeling approaches.

Choosing a Model 59 Topic Options Implications “side product” of the input-output model used. This might allow for additional model sensitivities, such as a scenario that increases costs for transport of hazardous goods. Time needed for model development No land use No time needs to be allocated if a land use model is not used. However, time needs to be spent to disaggregate forecasts of socioeconomic data to TAZs. If TAZs change or if a new forecast year is analyzed, this disaggregation needs to be done again from scratch. Given that this disaggregation step often is done manually, if no land use model is available, the time required may be substantial. Sketch planning A reasonably sensitive sketch planning model can be set up in as little as 1 year, depending on the status of the input data. It is not uncommon, though, that model implementation takes 2 to 3 years until the model is calibrated, staff is trained, and reasonable sensitivities have been confirmed. Microsimulation discrete choice It took MTC about 5 years and WFRC about 3 years to implement their models. These numbers are not necessarily representative because MTC had to coordinate and reconcile parcel-level data that was more readily available for WFRC. Model estimation may take as much as a year. If planning to use a microsimulation discrete choice model, it is not uncommon to budget 5 years for model implementation to ensure model results are stable and defendable. Depending on data and staff availability and the modeling expertise of staff members, however, model implementation may take 2 to 3 fewer or more years. Spatial input- output Development time is similar to that for microsimulation discrete choice models. ARC and the Metropolitan Council were comparatively fast with approximately 3 years. In contrast, Ohio DOT had to spend much more time on model development as an early adopter. Generally speaking, models that have been on the market for a longer time tend to be faster to implement. Similar to microsimulation discrete choice models, development time predominately depends on (1) availability of data in the necessary format and (2) availability and expertise of staff. Development costs No land use Although no model is without development costs, the costs of manually disaggregating sociodemographic data to TAZs can be substantial. Sketch planning Costs for implementing sketch planning models range widely. First of all, costs depend on how many tasks can be completed in house and how much work needs to be done by consultants. The amount of data purchased further affects prices. The complexity of the study area is another factor. Simple models implemented in small study areas may be built for as little as US$ 50,000 in consulting fees, while more complex implementations in large regions may cost ten times as much. Software licenses tend to be inexpensive, with What If? and I-PLACE3S being available at no fee and CommunityViz being available for an annual license fee of US$ 1,000. Generally speaking, sketch planning models are less expensive to implement than microsimulation discrete choice or spatial input-output models. Microsimulation discrete choice Prices for discrete choice models range even more widely than for sketch planning models. A model developed completely in house, such as MetroScope in Portland, Oregon, may be less expensive to implement than a model that requires a lot of consulting support. MTC has an annual budget of US$ 150,000 to support land use modeling, plus one staff member, and US$ 15,000 being spent annually on data updates. MTC has attracted strong model developers as staff members. Other agencies with less modeling expertise may need to invest substantially more to achieve an equally comprehensive model setup. Spatial input- output The same price range applies to spatial input-output models as to microsimulation discrete choice models, because complexity and data needs are comparable. ARC spent roughly US$ 1 million over 5 years, plus two staff members who work almost full time on the land use model. The implementation for the Metropolitan Council was roughly US$ 250,000 over 3 years and about two staff members. Ohio DOT is an outlier—several million dollars were spent on implementing PECAS. This was due in part to the early development stage of the model in the mid-2000s. The SEAM/SLUM model was implemented at almost no cost for Ohio DOT by the consultant as a placeholder. The latest model update of SEAM/SLUM to reflect newer input data cost approximately US$ 70,000 in consulting fees plus staff time. Spatial input- output The diversity of scenarios that can be modeled with spatial input-output models is essentially identical with scenarios that can be modeled with microsimulation discrete choice models. Proponents of this model type claim that spatial input-output models are better at representing the competition between different land uses, because prices are represented more realistically, but this claim has not been proven. One benefit of spatial input-output models could be that they sometimes provide commute trips and commodity flows as a (continued on next page) Table 7-2. (Continued).

60 Integrated Transportation and Land Use Models Topic Options Implications Developer dependency No land use model None. Sketch planning The case study of TJCOG shows that developer dependency can be avoided once the model has been established in the region. SACOG, on the other hand, depended on the developer whenever they wanted to change something in the model. For that reason, SACOG is exploring a different sketch planning model now. Operating a sketch planning model without the developer was feasible for both SACOG and TJCOG. Microsimulation discrete choice No case is known where an agency successfully applied a discrete choice model without strongly involving the model developer. These models tend to be too complex for anyone who has not closely worked with the model developer to implement the model. This dependency becomes irrelevant if the model developer works for the agency where the model is applied, as is the case in Portland, Oregon, and as was the case in Oahu, Hawaii, before their model developer retired. Maintaining a discrete choice model that is operational and fully calibrated, on the other hand, is possible without the model developer, if the agency’s staff is well trained. Spatial input- output The same considerations for developer dependency apply to spatial input- output models as to microsimulation discrete choice models. A series of failed implementations suggest that if the model developer is not heavily involved in model implementation, reaching a properly implemented model is highly unlikely. Once the model is set up, agencies can operate spatial input-output models without support from the model developer, if trained staff is available. Training requirements No land use Manual disaggregation of socioeconomic forecasts to TAZs requires little training. Intimate knowledge of the study area is required, and GIS expertise is helpful. Sketch planning Staff members using sketch planning models need to have a good knowledge of the model’s method. Many sketch planning models are GIS-based, which makes GIS expertise important. For making changes to the model, usually some programming knowledge is required. Microsimulation discrete choice Staff members working with microsimulation discrete choice models ideally have a background in discrete choice theory, model estimation, and programming. Once the model is implemented and calibrated, requirements for the model user are smaller. However, to fully understand which scenarios can be analyzed with the model and how to read model results, staff members need a good understanding of the model design. Spatial input- output Staff members working on spatial input-output models ideally should have a background in input-output theory, model estimation, and programming. Once the model is implemented and calibrated, requirements are reduced the same way as for microsimulation discrete choice models. Table 7-2. (Continued). • Input data, which vary widely depending on model specifications, but may include parcel data- bases, zonal population and employment data, regionwide population and employment growth forecasts, zoning restrictions, floor space vacancy rates, and zonal development capacity. • Consulting fees for model implementation, which vary widely depending on model specifica- tions, level of agency staff involvement, data availability, and expectations of model accuracy. • Consulting fees for model updates, which vary widely depending on required model changes, level of staff involvement, and amount of data processing needed. In some cases, model updates were implemented by agencies without consultant support. Model revisions, however, commonly require involvement of a consultant. Defining fixed data requirements for the three model types is impossible. Different imple- mentations of the same model may require different data items. All models need population and employment by zone for the base year as a starting point, but, beyond that, data requirements vary widely based on the chosen model and the model specifications for a particular study area. It is common that the largest share of resources is spent on purchasing, processing, reviewing, and cleaning up data. Crowdsourcing data, which can be downloaded from online sites such as Google, Foursquare, Twitter, or Geofabrik, could reduce the burden of data processing. It is

Choosing a Model 61 unknown how complete crowdsourcing data are, but many modelers have had good experience using such data instead of conventional data. Some model developers argue that data limitations should not limit the model design. Avail- able data are collected as far as possible, and other data can either be synthesized or replaced with good theory. Synthetic populations used in activity-based models, for example, are synthesized from aggregate data where micro data are unavailable. Theories on travel time, budget, or housing location choice can be used if data on observed behavior are not available. Although data development is expected to require a significant amount of resources for the foreseeable future, online sources and the use of behavioral theory may reduce the burden of data development. Choosing among no model, sketch planning models, and advanced mathematical models is comparatively easy (Figure 7-1). Not implementing a land use model means manually dis- aggregating land use forecasts, while sketch planning models provide a systematic approach to automatically disaggregate data based on rules that can be modified. Mathematical models offer behavioral detail that greatly enhances the number of scenarios that can be analyzed. Not using a model has the disadvantage that the disaggregation of forecasts of socioeconomic data will be arbitrary to some degree and cannot produce the same result if repeated by a different staff member. Sketch planning models are excellent tools for automated disaggrega- tion and allow running simple scenarios with regard to zoning, housing densities, and access to transit stations. For agencies not interested in land use scenario analyses (but interested in simple zoning scenarios) or agencies that need to keep land use forecasts static for political reasons, a sketch planning model appears to be a reasonable choice. Sketch planning models tend to be easier to implement, but they also tend to have limitations with respect to the various scenarios that can be modeled. Agencies interested in representing the land use/transport feedback cycle and/or conducting more complex scenario analysis are likely to prefer an advanced mathematical model. Sketch planning models are particularly limited for scenarios that try to assess behavioral change after policies have been introduced. For example, it requires far-reaching assumptions to model the effect of VMT tax in a sketch planning model. A microsimulation discrete choice or a spatial input-output model, which can represent behavioral responses based on economic and psychological theory, is better positioned to endogenously represent the change in behavior after the VMT tax has been introduced. Choosing between microsimulation discrete choice and spatial input-output models, unfortunately, is much more difficult. Both model types offer advanced scenario analysis. This report found success stories for either approach, data requirements are comparable, and costs for implementation appear to be in the same range. Some agencies noted that discrete choice theory is more commonly taught at universities than input-output theory. However, modeling expertise in different agencies varies widely, and the experience of current staff members may be more relevant than what is taught at universities. Regardless, new staff members could be trained in either modeling paradigm. Table 7-3 summarizes common tasks and scenarios and shows which modeling approaches are most suited for each task. Without an integrated model, none of the example analyses shown in Table 7-3 will be possible to analyze. Furthermore, lacking the integration with a land use model will make the capture of induced demand of new roads built or the densification in developed areas unlikely. Table 7-3 does not distinguish between microsimulation discrete choice models and spatial input-output models because both types of models are used for essentially the same analyses. Proponents of discrete choice models argue that the world is never at equilibrium, and thus, No Land Use Model Sketch Planning Model Mathematical Model Discrete Choice Model Bid-Rent Approach Model Figure 7-1. Decision hierarchy.

62 Integrated Transportation and Land Use Models spatial input-output models create a distribution of socioeconomic data that is too perfect. Proponents of spatial input-output models argue that discrete choice models are inherently poor at predicting prices, which are a major driver of urban development. Martínez (1992) concluded that the differences between the two modeling paradigms are theoretical, and he showed that these two approaches, in practice, generate similar results. Task No Model Sketch Planning Model Microsimulation Discrete Choice Model Spatial Input- Output Model Disaggregate forecasts of socioeconomic data to TAZ X X X Quickly run scenarios in public meetings X Visualize effect of different density assumptions X X X Run zoning scenarios X X X Run access to transit or highway entrances scenario X X X Represent land use/transport feedback cycle (shown in Figure 2-1) X X Run TOD scenario (with proximity assumptions) X Run TOD scenario (with behavioral model) X X Run housing subsidy scenario X X Analyze sociodemographic composition of population in future year X X Analyze retail market areas X X Analyze effect of telework on housing choice and firm location choice X X Analyze effect of autonomous vehicles on housing choice and firm location choice X X Analyze effects of congestion or cordon pricing on housing choice and firm location choice X X Table 7-3. Sample tasks and scenarios that can be run with different land use model types.

Next: Chapter 8 - Conclusions »
Integrated Transportation and Land Use Models Get This Book
×
MyNAP members save 10% online.
Login or Register to save!
Download Free PDF

TRB's National Cooperative Highway Research Program (NCHRP) Synthesis 520: Integrated Transportation and Land Use Models presents information on how select agencies are using sketch planning models and advanced behavioral models to support decision making. The synthesis describes the performance of these models and the basic principles of land use/transport integration.

  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!