Strategic Issues Facing Transportation, Volume 6: The Effects of Socio-Demographics on Future Travel Demand (2014)

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.

4An aim of this study was to help policy makers and planners in state and local transportation agencies gain an improved under- standing of the fundamental relationships between social and demo- graphic factors and travel demand, and how these relationships may change over the next 30 to 50 years. Such information is critical, because it is a basic element in the formulation of long-range trans- portation plans. Long-range transportation plans, with horizons of 15 years or greater, are an important part of defining a vision for the future and of establishing strategic transportation investment and system opera- tions directions for a metropolitan area. These plans are often viewed as a process for enabling decision makers to evaluate the strengths and weaknesses of various transportation alternatives. Decisions regarding future actions are based on implicit and explicit assumptions about the future state of the area in which the decisions will be implemented—for example, how the population in a region may change over time, how socio-demographic changes will affect how or where people will travel, and what kinds of transportation modes and infrastructure will be needed. Thus, transportation planners often are asked to predict socio-demographic trends that will affect future demand for transportation infrastructure. The greater the degree of uncertainty associated with these trends, the more problematic the resulting decisions will be. 2.1 Uncertainty in Forecasts Long-range transportation planning necessarily depends on uncertain forecasts. These fore- casts are generated from travel demand forecast models. Modeling and forecasting are related, but distinct, activities: modeling is about building and applying tools that are sensitive to the policies of interest and respond logically to change, while forecasting is an attempt to envision future conditions. In the current context, it usually involves predicting future travel demand and the resulting multimodal flows or changes in land-use patterns over time. The difference often becomes blurred because there is a tendency to think of anything that comes out of a numerical model as being a hard prediction. But in reality, as a model is run farther into the future, precision in data and forecasts becomes more challenging. Transportation travel demand models have evolved in recent years from four-step models, which average transport behavior over zones, to more sophisticated agent-based models based C H A P T E R 2 Long-Range Planning in an Uncertain World Chapter 2 Takeaways • Model predictions become less accurate over long time scales. • Model usefulness does not necessarily increase with complexity. • Scenarios are a well-researched way of handling uncertainty. • System dynamics models can be used to realistically illustrate different scenarios. “For all of its uncertainty We cannot flee the future.” Barbara Jordan, former member, U.S. House of Representatives

Long-Range Planning in an Uncertain World 5 on representations of actual populations. Therefore, the unit of analysis is shifted from rough aggregates to the level of the individual traveler. This development makes it possible for model- ers to incorporate detailed demographic data. Models can also reproduce nonlinear, dynamic feedbacks, leading to effects, such as congestion. One manifestation of this important distinction between modeling and forecasting can be seen in model complexity. From a pure modeling perspective, a model is often only considered to be realistic or complete if it incorporates all the necessary data, causative factors, etc., that may be considered to be relevant. The natural result of this process is that models become very detailed and complicated, and accumulate a large number of parameters that cannot be accurately measured from the available data. These are unrealistic desires to have one model that includes all available data to address all questions. The end result is the more data, causative factors, and assumptions that are placed into the model to ensure completeness, the greater the chances are that the added items may not be correct and may actually contribute to arriving at the wrong answer. As complex- ity increases, models also often become unstable and must be carefully tuned to give reasonable results. They therefore become poor at making predictions, or adapting to different scenarios. For this reason, the models favored by people who receive regular feedback on their predic- tions, such as those who work in business forecasting, tend to be quite simple. A detailed survey of forecasting models showed that—perhaps counter-intuitively—the simpler models consis- tently outperform more complicated models (Makridakis and Hibon 2000). This does not imply that agent-based models are inferior to aggregate models; the latter can be extremely compli- cated (as with large models of the economy), while agent-based models can be constructed to be quite parsimonious in terms of parameters (see, for example, Orrell and Fernandez 2010). But it is important to bear in mind that advances in modeling, and the creation of more elaborate and apparently realistic simulations, may not translate into advances in forecast accuracy, especially for long-range forecasts. 2.2 Accuracy of Travel Demand Forecasts Forecasting is by definition a forward-looking activity, but it is useful to also compare how past forecasts have compared with reality. This gives an idea of the nature and magnitude of expected forecast errors. Unsurprisingly, most such data are for forecasts over shorter time periods than that applies here. A summary of studies of travel demand forecast error is given in Parthasarathi and Levinson (2010). The largest available study of project-specific models, by Næss et al. (2006), presents results for more than 210 projects in 14 countries. It found large discrepancies between passenger forecasts and measured results. For rail projects, passenger numbers were overesti- mated in 90 percent of cases, with an average overestimation of 106 percent. Forecasts were more accurate for road projects, but half had a difference between actual and forecast traffic of more than ±20 percent and, in a quarter of cases, the difference was more than ±40 percent. Næss et al. (2006) also found that forecast accuracy has not improved with time, or with more advanced models or computer power. In fact, Road vehicle forecasts even appear to have become more inaccurate over time with large underestima- tions towards the end of the 30-year period studied. If techniques and skills for arriving at accurate traffic forecasts have improved over time, this does not show in the data. While these results are for individual projects over relatively short time frames, there is no reason to suppose that predictions will become more accurate over larger regions or longer time frames. Predictions usually deteriorate with time because of unforeseen effects. The forecast error is due to a number of factors. For rail projects, it seems that politics is important—passenger demand is overestimated because stakeholders, who want the project to

6 The Effects of Socio-Demographics on Future Travel Demand go ahead, favor optimistic forecasts over pessimistic ones (see also Flyvbjerg et al. 2009). Road projects do not show the same systematic bias, so the error is more likely due to model limita- tions, such as inaccurate estimates of trip generation (based on incomplete data) and land-use development (based on uncertain plans and projections). Parthasarathi and Levinson (2010) interviewed modelers responsible for making travel fore- casts in the Minnesota region. They found that the problem most frequently cited is, “the inability of the model to understand and predict fundamental societal changes,” such as increased female participation in the workforce. The locational distribution of forecast demographics was also a source of error. Predictions made in the 1970s for 1990s traffic did not anticipate such changes as a 40 percent increase in home-based work trip lengths, a 43 percent increase in per capita trips, a 39 percent increase in female labor force participation, highlighting the importance of demo- graphics. The tendency to maintain assumptions based on past trends, even after they have lost their validity, has been called “assumption drag.” As another example, Næss et al. (2006) note that the energy crises of 1973 and 1979 led to an abrupt, but temporary, decline in road traffic in Denmark. Danish traffic forecasters adjusted and calibrated their models accordingly on the assumption that they were witnessing an enduring trend. The assumption was mistaken. When, during the 1980s, the effects of the two oil crises and related policy measures tapered off, traffic boomed again, rendering forecasts made on 1970s’ assumptions highly inaccurate. To tie this discussion back to long-range planning, forecasting’s primary purpose is to generate information useful to decision makers for the specific types of decisions they are facing. The deci- sions are influenced by the degree of uncertainty associated with forecasts about the future. How many people will live in a region; in what types of households will they reside and by what modes will they travel; what will be the price of fuel; what are the rates of adoption of autonomous, self-driving vehicles? Good decisions (and good policies) should be robust across a wide range of socio-demographic futures. Therefore, to aid with this process, models should be viewed as tools for exploring scenarios, rather than providers of hard predictions, and should be designed to be flexible enough to explore scenarios, while avoiding (as much as possible) traps such as assumption drag. Models may have a poor track record at making precise numerical forecasts of the evolution of complex systems, such as the transportation network, but they are still invaluable for thinking about the future and comparing different possible outcomes (Orrell and Fernandez 2010). 2.3 Handling Uncertainty A goal of this study was to provide transportation planners and decision makers with an increased awareness of socio-demographic trends and how these may impact long-range trans- portation conditions or needs. With knowledge of the limitations of models to produce accurate long-range forecasts, the research team focused on developing a tool (Impacts 2050) that would help transportation planners and decision makers apply a scenario approach for handling uncer- tainty. Miller (2004) advocated for scenario planning as a method for addressing uncertainty in transportation forecasts: Scenario planning expands upon traditional planning techniques by focusing on major forces or drivers that have the potential to affect the future. By developing scenarios to tell a story of the future, planners are better able to recognize these forces and determine what planning activities can be done today and can be adapted in the future. The FHWA’s web site promotes scenario planning as an analytical tool. With the tool developed in this study, users should gain an overall understanding of how trends affect future travel demand; be in a position to test and account for these trends in

Long-Range Planning in an Uncertain World 7 projects, plans, and forecasts; and examine policy or other interventions that may offset or enhance these trends. The tool incorporates two elements: (1) scenarios representing visions of possible futures, considering basic demographic trends, globalization and immigration policy, economic growth, technology advances, transport funding, shifting social attitudes, etc.; and (2) a system dynamics model that represents regional links between population, land use, employment, transport supply, and travel behavior. With both elements, the objective is to provide a mechanism for dealing with uncertainty. The scenarios were developed to recognize a range of future outcomes, beyond what tradi- tional planning can create. The research team used four scenarios, not to cover up its inability to predict the future, but to help policy makers and planners think about the range of possibilities. The scenarios are multi-layered and complex, and are fundamentally distinct from each other. Titled Momentum, Technology Triumphs, Global Chaos, and Gentle Footprint, the scenarios are discussed in detail later in this report. The purpose of the system dynamics (SD) model is not so much to predict long-term travel behavior (since there is no evidence that models can perform this task), but to realistically illus- trate the different scenarios and provide a higher level of insight and understanding to policy makers and other interested parties. Using the model to gain a deeper understanding of the interaction of the elements of the decision has the possibility of helping planners generate sce- narios that provide the most value in considering possible futures. The SD model segments a region’s population by age, household structure, income, race/ ethnicity, acculturation, residence location area type, and work status. The model then “evolves” this population over time, simulating the population’s transitions from one category in each of these segments to another category over time (e.g., aging the population into different catego- ries: 0–15, 16–29, 30–44, 45–59, 60–74, 75 and older). The evolution of the population over time affects travel behavior. The impacts on travel behavior are indicated in terms of car ownership, trip rates and distance, and mode choice. Land use, employment, and transport supply sectors are present in the model in minimal detail, which enables the incorporation of feedback loops that represent the dynamics of the transportation system. Neither the use of the SD model in long-range planning nor the application of scenario plan- ning is new. What is new is the integration of both in a single tool that can aid long-range planning. 2.4 Improved Long-Range Transportation Planning The challenge for policy makers and planners is to make effective use of the tool and the new information that it will provide to actually improve decisions. This entails interacting with the tool itself, not just the outputs of the tool. According to Barabba (2011), no important decisions should ever be based solely on the results of a quantitative model. After extensive experience with models at the U.S. Bureau of the Census, Xerox, Kodak, and General Motors Corporation, he formulated Barabba’s Law: “never say, the model says.” The intent of the law was to remind modelers and decision makers that people make decisions. Models should not. So, why should transportation agencies go through all the trouble of using this new tool, based on a complicated SD model, to explore emerging trends and create possible futures? The reason is simple—to increase the chance of making better decisions, such as: • Supporting long-range plan development. • Supplementing the capabilities of existing planning models. • Formalizing the consideration of uncertainty in the planning process.

8 The Effects of Socio-Demographics on Future Travel Demand • Facilitating participation in the planning and decision-making processes. • Serving as a sketch-planning tool for providing quick and timely answers, as well as support- ing sensitivity and exploratory analysis. • Serving as a “utility” program for providing data inputs to models and the planning process. With all the uncertainty about the future, one thing is certain: the future will be very different from the present. If policy makers or planners get stuck in the present, let alone being stuck in the past, they will not be able to accommodate future trends. With all the uncertainty, the future reality is better understood by exploring multiple plausible future scenarios than by studying the present. As an example of this, when we were developing the Technology Triumphs scenario, we listed that sometime in the distant future society would see autonomous vehicles. So now, only two years later and Google has autonomous vehicles actually being tested on the street. Also several other “traditional” manufacturers have recently announced their intent to develop versions of these vehicles as well. The art of scenario planning lies in blending the known and the unknown into a limited number of internally consistent views of the future that span a very wide range of possibilities. This study blended what was known about current socio-demographic trends with the possible evolution of these trends to examine the influence on people’s future travel behavior in the future. In doing so, four possibilities for how this might unfold were constructed. The study team took a systems approach by speculating on how socio-demographics and travel behavior would interact with land use, employment, and transportation supply sectors to generate future scenarios. The outcome of this effort was the identification of key socio-demographic drivers related to population size and growth; population structure and composition; cultural and social diversity; household-based economic activity, geo-demographics, attitudes, and technology use; and incorporation of assumptions about these into the tool. A final focus was on impacts in terms of passenger travel and on travel by auto, transit, and nonmotorized modes. The next chapter summarizes eight socio-demographic trends that transportation agencies are already facing that will impact travel demand over the next 30 to 50 years. These trends have been drawn both from the team’s experience and expertise in this area of study and from a review of literature. They are important to this study, as they formed the conceptual framework for the development of the four scenarios and the structure of the SD model.