3. Using Integrated Models in Policy Analysis: An Assessment
|
|
3. Using Integrated Models in Policy Analysis: An Assessment |
|
Recognizing an inevitable lag between latest theory and best practice, if we were to evaluate currently operational models on the basis of their collective ability to incorporate latest theories within their frameworks, they would get quite high marks. Taken as a set, the previously reviewed integrated models have advanced in a number of important directions since the 1960s. They have managed to combine the minimum effort and locational accessibility premises inherent in spatial interaction theory with the statistical and information theoretic notions of entropy and the economically rational notions of utility maximization. Methodologically, they make use of nonlinear mathematical programming methods as well as the latest developments in econometric and microsimulation modeling of the demand for travel, residences, and employment. The more comprehensive models also tackle demographic change in the urban population, and some also model physical stocks other than transportation infrastructure, notably the aging and renewal process associated with the urban housing market. Finally, they model these events using an extensive database, resulting in the allocation of traffic volumes and speeds over detailed link-node representations of multi-modal urban transportation networks.
But this, of course, is not the test in which we are most interested. How well such theories stand up in practice is the true test. Here we are currently at something of an impasse. In contrast to the considerable effort made to develop the theoretical aspects of the relationships between transportation and spatial structure, the practical application of models has been relatively neglected. This conclusion is mirrored, with respect to U.S. practice, in the review by Cambridge Systematics and Hague Consulting (1991), which found that only a handful of the top 18 metropolitan areas were using integrated models in their planning processes. In their relatively brief history, the land use-transportation models reviewed in Sect. 2 have been subjected to a good deal of criticism (see Batty, 1980, for an early historical review; see also the Winter, 1994 edition of the Journal of the American Planning Association for a retrospective). Past criticisms have tended to revolve around (1) conceptual issues of model realism and hence usefulness; (2) practical issues of data availability and quality, as well as computational requirements and ease of use; and, as something of an offshoot from these two issues, (3) the role such models are to play in the planning process. While recent computational advances have done much to remove concerns over both computer costs and computer run times, the other issues remain. Each is discussed below, highlighting some concerns frequently voiced in the recent literature.
As Lee (1994) points out, the role of large scale urban land use-transportation simulation models remains a cause for debate. Should they be considered as tactical or as strategic planning tools? If used as tactical planning tools, their most common application would probably be to evaluate travel policies along specific urban corridors, with an eye to an environmentally influenced benefit-cost ratio being realized within a suitable time period. Even so, such an evaluation period might cover as long as 15 or 20 years depending on the TCM being proposed (i.e., up to the expected lifetime of a typical urban highway pavement, if the addition of new infrastructure is involved).
However, a danger with using models solely to analyze individual travel reduction projects is the potential for disjointed, piecemeal planning. Ideally, and central to the aims of this present review, we need to find a way to embed such project evaluations within more strategically developed, areawide transportation plans. If these plans are to make the sort of contributions to petroleum savings and CO2 reductions which have come from more efficient engine and fuel technologies, areawide impacts will almost certainly be required. We also need to think in terms of longer planning horizons. Watterson (1993) concludes that even a planning horizon of 30 years may not be long enough to capture the true impacts of a plan which contains significant transportation infrastructure investments. He notes that such plans may go on to influence urban form, and therefore urban travel activity, for many years into the future.
Lee (1994) argues that in searching for such a strategic role we may be trying to get too much detail into our models. As we add more detail and functionality to what are already rather ambitious models, we loose flexibility in their application and increase expensive data requirements. In contrast, Harris (1994) prefers to view such efforts as an aid to comprehensiveness of understanding, rather than comprehensiveness in forecasting. This second argument meshes well with Wilson's (1984) perspective on the use of integrated models as tools for evaluating the robustness and resilience, rather than the details, of alternative urban and regional plans. As Owens (1989, p. 233) puts it; "In the end, perhaps, accurate prediction matters less than flexible normative planning, based on an intelligent assessment of the most likely directions of certain trends."
To carry out such planning, mathematical, computer-based models would seem to be our only realistic alternative if we wish to apply, and properly test the results of applying, a formally developed logic behind our planning decisions. Without reasonably comprehensive models, we cannot hope to simulate the often nonintuitive effects of combining a wide range of policy options within any single plan. Our ability to determine the general magnitude and direction of policy-generated effects seems well worth the effort. This, however, raises the issue of how we gain confidence in our model-based results. Such a question moves us on to issues of model validation.
Validation means carrying out checks to establish how well a model did in forecasting a future situation by comparing the model's results with observed data. As Wegener (1994, p. 25) points out, "remarkably few validation exercises are reported in the modeling literature." Travel data availability constitutes the major constraint on validation exercises to date, especially data covering time intervals long enough to capture some of the important changes in urban infrastructure and land use.
An obvious problem for cross-sectionally calibrated models is that they are using the many parameters established in their base year calibrations to predict changes over time. In doing so, they may be placing an overreliance on the behavioral implications of spatial variability in traveler and land owner responses to differing conditions. Of greater interest is the temporal variability in such responses for a suitable range of geographically as well as socioeconomically varying urban environments. To understand and model such behavioral responses, we need to make more and better use of time series data. Webster et al. (1988) briefly describe the results of using seven of the nine models covered by the international study group on land use-transportation interaction (ISGLUTI) to project both zonal employment and population totals, using data for intervals from 3 to 12 years into the future. This includes versions of MEPLAN used in the studies of Sao Paolo, Brazil, and Bilbao, Spain, which apparently used specially developed follow-up survey data for the purpose (Echenique, 1985). All results reported R2 values > 0.95 when comparing absolute values, using from 30 to 148 zonal observations, depending on the particular model and its application. The 3-time period, 12-year, incremental forecasts produced by the Dortmund model gave particularly high R2 values. However, R2 values took on much wider ranges, from 0.98 to 0.59,when comparing observed versus modeled rates of change in these same variables.
Similar R2 values are reported by Prastacos (1986b), using the POLIS model to predict changes in the number of households as well as employees per zone within two basic and two nonbasic sectors for the period from 1975 to 1980. This involved regressions on 107 observations (i.e., land use zones) within the nine county San Francisco Bay Area. Noticeable improvements in these coefficients occurred when aggregating the results to county totals or when using such county totals to control the subsequent allocation of employment to zones within a county. Some checks were also made on the resulting interzonal private and public trip matrices produced by POLIS, but with synthetic rather than observed 1980 flows for comparison. Recent recalibrations of the trip distribution and modal choice (auto versus transit) submodels using 1990 journey-to-work data from the Census Transportation Planning Package (Caindec and Prastacos, 1995) produced R2s for auto trips around 0.80 and for transit around 0.77, with model averaged travel times within 10% of expected results. In general, however, producing similar comparisons of modeled versus observed non-work trip matrices is problematic, with little in the way of consistent historical data for guidance.
Hunt (1994) also describes an extensive series of model validation tests carried out as part of the application of MEPLAN to the city of Naples, Italy. Maps and graphs are used to show the generally good fit between the model generated versus observed number of households as well as the private residential floorspace rents per zone (26 zones). Also examined were (1) expenditures on travel; (2) floorspace and other purchases by each of five household types; (3) average trip distances for four trip purposes (work, shopping, school, other); and (5) selected modeled versus observed average weekday morning peak period traffic cordon counts. He also describes the considerable time and effort required to calibrate, or "fit" the model, including the definition of suitable household classes and the too often experienced problem of having land use data for one year and transportation systems data for another, somewhat earlier or later one.
Whether using data sets from two or more periods to forecast or "backcast", using a cross-sectionally calibrated set of model parameters, ideal requirements for such tests would include use of the same set of traffic analysis zones as well as the same trip purpose definitions from one period to the next. In the past this has often meant considerable data reconciliation efforts. Wegener (1994) suggests that a model's performance should be based on its ability to forecast the essential system dynamics over a past period at least as long as the forecasting period to which it is being applied. He goes on to note that only Dortmund and MEPLAN, among currently operational models, appear to have followed this philosophy. In both cases these models are only partially calibrated by statistical estimation techniques and partially by manual fine-tuning as part of a long, interactive process. Often, as Hunt (1994) points out, it's difficult to distinguish data problems from errors in a model's formulation or in its underlying assumptions.
Clearly, greater emphasis on validating the models is required, including the establishment of procedures to track the major data sources necessary to calibrate them. This constitutes the most significant obstacle to model validation and, by implication, further useful model development. More comprehensive models mean more demanding data requirements.
Given current data limitations, how are we to assess the value of such models in a strategic context? Here the ideas expressed by Cowing and McFadden (1984) and restated by Hensher et al. (1992) are apropos. When an analysis task involves forecasting over a long period of time with substantial deviation from historical experience to be expected, they suggest that assessment of a simulation model is best focused on realism in process. This contrasts with more direct assessment of a model's predictive capability, involving the above discussed comparison of model results against a known, and empirically observed, reality; a validation process they term realism in performance. At the present time any discussions of current model weaknesses and associated research needs are necessarily focused heavily on such realism in process. However, more realism in process suggests that we also use more behaviorally based (i.e., more realistic) models. That is, it suggests that we focus more attention on how travelers behave and, for the purposes of policy impact assessments, how such behavior changes over time once policies are implemented which act upon it. This in turn suggests that more attention be given to the collection and use of longitudinal data sets. In particular, multiwave traveler panel surveys, collecting information from the same group of travelers at discrete time intervals, are discussed below as an important data collection option. A concerted effort will be required to design, collect, and maintain such temporally anchored databases. A first step is to determine which are the major variables of interest to such longitudinal analyses and (since cost of data collection remains the major constraint) which data we can effectively relegate to less regular data collection activities. To do so, we need to better understand the causes of current variability in travel demand.
The traditional four-step transportation planning model described in Sect. 2.3 of this review (Fig. 3) has been the focus of a good deal of criticism for many years. Within the United States, the need for metropolitan planning organizations to address the vehicle travel reduction requirements of the 1990 Clean Air Act Amendments (CAAA) is now leading to a new round of model development, known as the Transportation Model Improvement Program (see Texas Transportation Institute, 1993). Much of the criticism within the modeling literature argues that we need to place both household and company-based travel decisions within more behaviorally realistic decision-making frameworks. Treatment of travel as a good composed of separately modeled attributes of frequency, mode, destination, and route choices is being challenged. While energy, economic, and environmental impact analyses may require that we translate the demands for travel into numbers of temporally and spatially explicit vehicular trip volumes, the current methods we use for getting there are proving increasingly restrictive. Frequently voiced criticisms of the traditional Urban Transportation Planning (UTPS) process are described on the following pages. Cumulatively, these weaknesses act to obscure the relationship between cause (including policy-induced cause) and effect.
The Relationship Between Trip Frequencies and Travel Costs
An appropriate feedback mechanism between the trip generation model and the rest of the four-step urban transportation modeling procedure continues to elude modelers. The dashed arrow in Fig. 3 shows the desired (hypothesized) linkage. While travel speeds and costs are often interactively solved for within the destination, mode, and route choice steps, traffic generation remains inelastic with respect to such travel cost changes. To date, no sound and generally reproducible basis has been found for such a linkage. Similarly, empirical efforts at direct incorporation of the effects of cost-determined locational and modal accessibility within existing trip generation models have met with almost universally poor results (see Kitamura, 1994, for a recent discussion).
It may be the case that where trip frequencies are concerned, even among the more discretionary forms of travel, transportation costs or traditional forms of cost-based accessibility are in many cases not the only, or even the most important, determinants of daily or weekly travel activity schedules. However, one difficulty associated with obtaining a relatively simple functional relationship between trip frequency and trip length or cost may be the nature of past survey data. Cross-sectional, single day trip sampling may not contain the information required to fathom a behaviorally sensible and statistically consistent relationship. Implicit in nearly all past efforts to simulate urban travel activity patterns is the treatment of transportation as a separable good to be purchased independently of other household needs. However, once we place our analysis within a longer-term perspective, other nontravel cost factors become important. That is, housing, food, health, education, and other costs may compete with travel costs for the household budget in ways which may affect trip frequencies every bit as much as urban accessibility surfaces do.
An argument for the use of integrated land use-transportation models is that they currently offer the only means of getting the costs of travel back into the trip generation process; albeit via a rather complex series of modeling processes. However, while a residential allocation submodel is used to link housing rents to travel costs within a number of the models reviewed in Sect. 2, these models don't go any deeper into the trade-offs between travel and other goods which take place within budget-constrained households. For strategic planning purposes it may be sufficient to model travel versus housing costs in this manner, as long as a household's share of its income spent on travel and activities remains reasonably constant (see below). However, many short term decisions by household members may reflect a wide range of responses to daily or weekly time as well as monetary travel budgeting. The cumulative variability in such responses may be an important reason why no simple empirical relationships between daily trip frequencies and travel accessibilities or costs appear to repeat themselves across different studies.
Trip Chaining and Destination Choice
Making the relationship between trip frequency and travel cost more difficult to assess, many trip destinations in urban areas occur within multipurpose, multistop daily travel chains such as the home to work to shop to home type of travel circuit (see, for example, the travel data described by Hummon and Burns, 1981; Kitamura and Kermanshah, 1983; O'Kelly and Miller, 1984). By ignoring such trip chaining activity, the traditional transportation planning model fails to capture the time and cost savings offered by multistop travel activity patterns. This in turn means that integrated models of land use and transportation which use traditional, single destination spatial interaction models also fail to provide support for the analysis of land use policies which might take advantage of such mileage saving options.
The destination choice set problem, a frequently revisited technical problem within the travel demand literature, further exacerbates the problem of destination choice. Spatial interaction models, whether calibrated at the zonal level or fitted to a sample of individual traveler responses, require a prespecified set of alternative destinations to choose from. Removal of a possible destination from the available choice set within a logit model changes the absolute probabilities of selecting each of the remaining options. The behavioral dilemma results from not knowing what the choice set really is or how it differs across individual travelers at different originating locations and by different trip purposes. While a number of approaches to the problem have been tried (Stopher and Meyburg, 1976; Richardson, 1982; Recker et al., 1986), the usual approach is to allow all traffic zones in the system or all zones chosen by survey respondents (where such data is available) to be in the choice set. This approach recognizes that more distant and less attractive locations will receive few or no trips and, hopefully therefore, will affect the results only marginally and well within the bounds of modeling error. Significantly, the way in which multidestination trip chaining opportunities affect such choice sets has not been thoroughly researched to date (but see Recker et al., 1983 for some interesting work on simulating feasible, including multistop, activity programs for specific household members).
Discretionary and Off-Peak Travel Activity Modeling
Both time of day and time within the week need to be recognized and modeled as important travel options. Nonwork, and frequently non-peak, trips are now responsible for 78% of annual trip starts and 73% of total vehicle miles traveled (VMT) in the United States (Hu and Young, 1994). Also, since many daily trip chains combine peak period commutes with more "discretionary" forms of off-peak travel (e.g., shopping, social and recreation, and personal business trips), this needs to be recognized in some fashion if we wish to understand the effects of such transportation control measures as staggered work hours and compressed work weeks, or the potential for mixed use urban activity centers to encourage midday walk, paratransit, or public transit use for personal business and shopping.
Collectively, the above weaknesses render policy analysis for specific highway impact projects rather suspect. For example, the potential for a particular highway capacity expansion project to lead initially to less congested, less polluting travel may in reality erode over time as the greater ease of travel over this highway induces some travelers to change their daily or weekly movement patterns, resulting in a revised form of multidestination trip chaining activity and shifts in the temporal distribution of traffic. Among other effects, such temporal adjustments may affect existing vehicle availability within multidriver households. For example, the introduction of an high-occupancy vehicle (HOV) lane as part of a highway capacity expansion may induce ride-sharing, which in turn leads to a rearrangement of vehicle utilization within the household. Vehicles left at home by the ride sharing commuter may then be used by a spouse or other family member. In total, some of the new weekly household travel activity patterns which form may, on balance, involve more VMT, consume more fuel, and put more pollutants into the atmosphere than before.
It's also not obvious how to introduce the effects of emerging telecommunications options, such as telework or teleshopping, into such a rigid trip-oriented approach. Are such options considered trip generation or travel mode options, and more importantly, how does the adoption of frequent telecommuting affect the travel activity patterns of other household members? The recent reports by DOT (1993) and DOE (Greene et al., 1994) discuss this topic in some depth, and recognize our currently limited understanding of what to expect. Indeed, the whole area of in-vehicle as well as in-home real-time information systems and their effects on travel patterns raises questions not well suited to a single destination, separable trip purpose approach.
Single destination trip based models also run into problems in evaluating such low energy, and potentially low emissions options as electric or hybrid fueled vehicles. Should petroleum prices rise sharply in the future, it would help to know what percentage of household travel could be supported by a single daily vehicle charge, given a particular land use arrangement within which trip chaining is an option.
If we are to make current models more behavioral, or replace them with new models, the following areas warrant further study and possible unification:
Analysis of Multi-Day Household Travel Activity Schedules
An increasingly popular approach to travel demand modeling is to look for ways to link travel decisions more closely to such lifestyle factors as intrafamily obligations, leading to jointly organized trips. Such considerations place us firmly within what is termed the "travel activity analysis" literature (see Carpenter and Jones, 1983, and Kitamura, 1988, for extensive literature coverage). This empirical and modeling literature suggests that shifts in travel behavior may only be properly understood within the wider context of how people organize their lives over a series of planning horizons and, notably, over multiday rather than single day periods (see, for example, Hirsh, Prashkea and Ben-Akiva, 1986; Kitamura and Van Der Hoorn 1987; Pas, 1988).
An important aspect of such an approach is the study of how households make use of their various automobiles (and, increasingly in the United States, of their light trucks and minivans) to carry out their activity schedules. Hensher et al. (1992) provide a review of past literature on this topic, as a prelude to describing an econometric modeling approach and supporting empirical study of the various dimensions of household-based automobile demand. In the United States, this recent literature on vehicle utilization includes the nested logit modeling of vehicle class, vintage, and fleet size by Train (1986) and the ordered probit modeling of household vehicle ownership decisions by Golob (1990). It also includes the analysis of gasoline price effects on vehicle use in multivehicle households by Greene and Hu (1984) and the statistical modeling of multiday vehicle utilization levels by Greene (1985). Greene (1985, p. 350-351) pointed out that "in particular, while single-day surveys of a large sample of households have been extensively studied and modeled, there is a lack of information and analysis of how usage of a particular vehicle varies from day to day over a long period of time."
With a limited but growing collection of multiday travel surveys for the past decade, notably in the form of panel surveys (see below), this situation is beginning to change. Much greater use of longitudinal data on travel activity patterns, including vehicle utilization patterns, is needed if we are to understand how policies intended to discourage low occupancy vehicle travel actually affect behavior. Currently, as Bhat and Koppleman (1993) point out, even conceptualizing the activity scheduling framework within which travel and other weekly activity decisions are made constitutes a new and challenging task.
Microsimulation appears to be a natural candidate for making operational such ideas as the simulation of multiday household travel activity patterns, and efforts such as the STARCHILD model (Recker et al., 1986) and TRANSIMS (Morrison and Loose, 1994; Shunk, 1994) fall into this category. Such efforts have a growing, if rather varied theoretical and empirical literature on the modeling of multistop travel chains to begin from, as evidenced by the review of trip chaining research by Thill and Thomas (1987). This literature includes the empirical work on vehicle use by Hummon and Burns (1981); the empirical and theoretical work with logit-based models of destination choice by Kitamura (1984); the suggestions for using nested logits to identify and capture the empirical linkages between primary versus secondary destinations within such trip chains by Wilson et al., (1981); the use of microsimulation linked to logit demand functions to investigate the effects of chaining on locational accessibilities (Southworth, 1985b); the extensive empirical and theoretical work on multistop, multipurpose shopping trips by, among others, Narula, Harwitz and Lentnek (1983) and O'Kelly and Miller (1984); and the use of Markov models to assess the effects of trip chaining on the location specific demands for retail facilities (O'Kelly, 1983). However, these and similar ideas have yet to find their way into actual use within the integrated urban modeling systems described in Sect. 2 of this review.
Multi-Wave Panel Analysis of Household Travel Behavior
A new development in recent years has been the emergence of travel panel analysis. By surveying the same group of travelers or households at two or more intervals in time, with successive surveys separated by a few months or years, we are now beginning to acquire data on how travelers actually responded to a specific event. Hensher and Raimond (1992) provide a summary of the major transportation panels studies to date. These include panels used to analyze household vehicle ownership and utilization decisions, and the effects of staggered work hours, an HOV lane, and telecommuting on household travel behavior. Greater use of such panels should allow us to get away from the always suspect use of disaggregate travel demand models whose calibrations are based on limited size, single cross-sectional samples of urban residents.
However, travel panel analysis is still in its early stages. The first text devoted exclusively to the topic was compiled only recently (Golob, Kitamura, and Long, 1994). The recent theoretical work of Jiang et al. (1992) and the related methodological and empirical work by Hensher and Raimond (1992) are particularly interesting. By treating the speed at which potential travelers change their travel behavior as a process of adaptation, Hensher and Raimond embed a stochastic process within their differential equations model. In this way they can affect the timing (including instantaneous adoption) at which such changes occur from one state to another (in their case a change in route from a free to a new toll road). Here the considerable literature on hazard response and survival models becomes very useful and is likely to be visited more often in future travel-related research.
These last authors also describe the problems involved in translating data from a series of discrete time panels into a continuous time stochastic model of the real world process. Currently, we are at a relatively early stage in the design and use of such models and also in our design of the panel data sets which can best support them. In one of the best-documented research efforts to date, Hensher et al. (1992) collected four waves of panel data, spanning a 70-month period, from households in metropolitan Sydney, Australia. They use this data to develop an automobile market share and energy demand forecasting system based on a combined discrete-continuous econometric model of household automobile choice. Nested logits are used to model the discrete choices of household fleet size (i.e., ownership of 0, 1, 2, or 3+ cars), vehicle type/vintage, and vehicle body mixes. These are linked to a series of continuous vehicle utilization models. Lagged operators and other devices are used to render these models dynamic in the sense of capturing what the authors refer to as "experience effects" and "expectation effects" within the multidimensional choice process. The approach shows what can be done today to improve the behavioral basis of vehicular travel demand modeling given a suitable longitudinal database. The authors used their modeling system to generate a range of automobile demand and fuel use scenarios at 5- to 7-year intervals for up to a 20-year period (1985-2005). Policy variables analyzed were vehicle and fuel prices, advances in vehicle fuel saving technology, and socioeconomic changes which affect household demands.
Use of Household Budgets to Bound Travel Activity Estimates
A number of researchers have hypothesized that the amount of travel we undertake is highly constrained on the individual level. Proponents of this approach argue that the total amount of travel people engage in is strongly constrained by either time or money budgets. Such budgets, the latter strongly related to available income, are claimed to either remain quite stable over time for any given city and population subgroup, or to change in clearly recognizable directions as a function of a few independent variables. Such constraints allow the amount of travel that a person or household engages in to be determined by appealing to a simple utility maximizing model, subject to budget constraints. For example, the longer-term decisions faced by the household as to how much total travel to engage in (that is, number of trip generations กม trip distances) is modeled in utility terms (U) by Golob, Beckmann, and Zahavi (1981), as:
where x refers to the amount of travel, measured as travel distance; m
refers here to one of m =1,2,....M travel mode options; Y is household
income; T is the time available to the household members for the
completion of their activities, such as an "averaged" travel
day; M* and T* are equal to the optimal fixed
travel time and money budgets of the household, given by M* =
k
cm.xm and T* =
m
xm/vm ; c and v refer respectively to mode
specific average costs and velocities (speeds); and a, b1 and
b2 are the model parameters associated, respectively, with
modal share weights, income, and time budget constraints.
The idea here is that by maximizing over the expenditures on time and money themselves, the longer term relationship between the travel and nontravel budgets (leisure time, consumption of other goods) of the household can be explored. If stable relationships between such budgets can be shown over a number of years, then we would have a very useful approach for placing reasonably tight bounds on the total amount of travel consumed. Zahavi and colleagues used such an approach to develop the Unified Mechanism of Travel (UMOT) to investigate empirically such hypothesized relationships (Zahavi, Beckmann and Golob, 1981; Zahavi, 1982). While conceptualized at the level of the individual household, the supporting empirical modeling is carried out on an aggregate, urban areawide scale.
However, the empirical evidence to date has been less than conclusive, and no operational models based on household travel budgets have been generally adopted (although as noted in Sect. 2.5, the Dortmund model does incorporate a travel budget constraining procedure within its transportation modeling process). Household stratification along at least car ownership and income lines would appear to be required if reliable forecasts of future budgets are to be made. Also, to be strictly applicable, nonmotorized modes of travel (walking, cycling) need to be included in the analysis. It may also be argued that people actually seek to maximize their accessibility to opportunities, rather than seeking to maximize their (budget constrained) distance traveled. Since maximizing accessibility to a set of spatially diverse opportunities need not involve minimizing distance or cost of travel, the latter is really a special case of the former. Various other pros and cons of a travel budget based approach are reviewed by Gunn (1981), Wigan and Morris (1981) and others in the same volume of Transportation Research.
In the most general terms, the notion of dealing with travel distance (VMT) as the result of a budgetary process appears to have considerable behavioral merit. Further empirical study is needed to determine if a travel budget-based approach can be developed directly into an effective forecasting mechanism. Rather, it may offer a check on the economic realism implied by otherwise unconstrained modeling approaches. To be most useful such investigations should, however, be time-series as well as cross-sectional in nature.
A second area of deficiency in current practice is the underdeveloped treatment of urban freight modeling. Attention to the behavioral aspects of urban goods movement (i.e. to the logic behind shipper and carrier operations) has seen little application at the fully urban scale (UMTA, 1982). If the behavioral waters of personal travel demand analysis are murky, then those associated with freight generating business practices appear downright obscure. Limited effort to date has gone into determining the relationships between company logistics and management practices and their effects on either the daily scheduling and use of multivehicle fleets or on the longer term decisions of where to (re)locate factories and offices with respect to customers and existing freight terminals (including some quite large break-bulk terminals).
Urban trucking research has, as we might expect, dominated what literature there is in urban goods movements. The most comprehensive attempt to come to terms with this area to date was carried out by Transport Canada (1979), which produced a multivolume report on various aspects of urban freight movement. This work includes calibration (to Vancouver data) of an urban truck transport model which links traditional forms of trip generation, distribution, and shortest path-based traffic assignment models to a truck load consolidation model which "consigns" freight to trucks of different sizes. Such consignments are based on effective vehicle weight capacity, the maximum daily hours of operation (industry regulated), and the expected dwell times at pick up and delivery sites. To allocate this consigned traffic between inner and outer city traffic zones, an optimization based model was used to determine whether to route this traffic directly or via a freight consolidation terminal. At the strategic, urban areawide level, Southworth, Lee, and Zavattero (1986) also examined the efficiencies involved in the use of alternative primary truck route designations and the clustering of freight terminals within the Chicago metropolitan area. Their approach and its empirical application embeds circuit based measures of locational accessibility within spatial interaction models. They also propose a method for using the resulting flows within a mixed person-freight traffic assignment model which in turn could be used to compute fuel use and emissions. Recently, Oppenheim (1993) has proposed an interesting and improved urban-areawide formulation of a combined personal and freight equilibrium traffic assignment model. Among other useful studies, the recent work on freight logistics by Daganzo (1991) and Hall (1993) consider the design of local area freight networks. However, such efforts again reflect a series of largely independent studies, focused on very specific aspects of urban freight travel. A suitably rewarding conceptual framework for urban goods movement analysis remains to be defined.
As with personal travel, much intraurban trucking is also known to involve highly circuitous, multistop daily routing activity (Southworth, 1982a). Circuit-based transportation costs have been used within logit models of destination choice similar in form to those used in passenger travel modeling (Southworth, 1982b, using data from the Chicago region). An interesting possibility, discussed by Wigan and Morris (1981), is the application of a travel budget approach to freight movements. This notion has appeal, since it is the activities at pick-up or delivery sites that often dominate the urban trucker's daily time budget, and since we might expect very similar allocations of travel time to such goods movement services across cities of similar size, given the highly competitive nature of the industry.
Our major hope for projecting aggregate levels of freight-creating industrial activity lies in the belief that businesses follow recognizable profit-maximizing or cost-minimizing development paths. As with the above developments in household travel, freight movement models based on the individual firm have moved into, among other directions, logistic demand based (Sheffi, Eskandari, and Koutsopoulos, 1988) as well as constraint based, multi-criteria mathematical programming forms (McGinnis, 1989). These models again place transport costs (including freight rates) as one among a number of important decision variables in activity pattern (notably, truck route, and schedule) generation. McGinnis (1989), for example, found that carrier reliability, transit time, and shipment loss and damage experience could be more important to shippers than freight rates when selecting a particular carrier.
Finally, in addition to the above issues, we now also need to address the impacts of just-in-time inventory/delivery systems, electronic message transfers, and the increasing interfirm as well as intrafirm coordination of logistics apparently taking place within today's information society. A major shift away from expensive warehousing costs to just-in-time parts and product deliveries clearly has the potential to increase vehicle miles traveled within a number of industries.
Looking at longer-term responses in the form of site (re)locations, nontransportation factors again loom larger than traditional industrial location theory suggests. Both intraurban personal and freight travel patterns are affected by the location of such companies; the former through the necessary daily journeys to and from work and, less directly but with increasing significance, through the effects such demands place on the rest of a household's typical weekly activity patterns. As reviewed by Giuliano (1989), much of the recently available empirical evidence supports the view that transportation is at best only one of many determinants in both a household's or a company's location decisions, sometimes acting as a constraint on subsequent economic and related land development without alone being a sufficiently motivating reason to cause a change in current location. Giuliano argues that most large U.S. cities now have such dense transportation networks that the perhaps once more obvious relationship between new highway infrastructure and land development is much less straight-forward, at least within the boundaries of urbanized areas.
In an assessment of the influence of road investment on economic development, Forkenbrock et al. (1990) reviewed a number of studies that reached the following conclusions (also listed by Parker, 1991):
In an often quoted national study of beltway (circumferential highway) impacts, Payne-Maxxie Consultants (1980) found no consistent relationship between the presence of such beltways and land use. Rather, land use impacts were dependent on (1) overall local economic conditions, (2) access to medium income or high income residential areas, (3) availability of developable land, and (4) favorable local zoning ordinances.
However, the empirical evidence suggests care in reaching to too general a set of conclusions. For Texas cities with over 4000 population, Buffington, et al. (1992) found significant correlations between 67 bypass, loop and radial highway improvements and the growth in employment and wage rates for the period from 1954 to 1988. They also cite a number of other studies reporting positive relationships between highway investment and employment growth. Their results may reflect the small size of many of the areas defined as urban, plus a starting point in 1954, when our urban and transportation systems were far less developed. However, even though much of the nation's basic urban highway infrastructure may now be in place, if we are going to use our models to project a similar distance into the future, then we should at least recognize the possibility of similarly large (if in practice very different types of) changes in transportation's future relation to economic development. How we make use of our built structures is changing, if gradually, with each important new laborsaving technology to come along.
Of the above nontransportation factors of significance, the presence of suitably trained labor pools has become an important concern for companies looking to locate, or relocate a factory or office. A recent survey of 504 manufacturers in North Carolina (Hartgen et al., 1991) provides an informative empirical study. Transportation related accessibility measures (state-to-state highway access, access to input materials, and local access by road) were generally ranked below labor factors (notably worker attitudes, availability, and trainability), while other important factors included quality of life (public schools, quality of area for raising children), site and utility costs (electricity costs and supply), and local tax rates. A question facing model developers, therefore, is how to better incorporate or recognize such nontransportation, nonaccessibility based factors within future urban models.
Factors in New Urban Center Formation
As Berechman and Small (1988) point out, many of our newly emerging urban places are different in structure from the classical city containing a radial highway network focused on a centrally located CBD. As a companion, and apparently necessary, corollary to the automobile-induced urban sprawl, the location of suburban centers, their rates of growth, and their mix of traffic generating land uses now represent a central concern for urban land use planning (Orski, 1985; JHK Associates, 1989; Garreau, 1991; Southworth and Jones, 1995). Traffic congestion within and between industrial as well as commercial and mixed use suburban activity centers is now also a problem. The very benefits of location and agglomeration of activities offered by a city's CBD, and which led to its subsequent traffic congestion problems, are now causing the more peripheral urban subcenters to experience their own version of traffic related negative externalities; encouraging us to ponder what the solution to such agglomeration diseconomies might be and where this process is leading us (Cervero, 1989).
Such "polycentric" urban development appears to be occurring at a number of scales and is having effects on travel speeds, trip distances, and total travel mileage. What are today seen as an expanding metropolitan area's suburban centers may tomorrow become small satellite cities in their own right. It is possible also that the functional ties between these satellite cities and the long established CBDs will be fewer and different than they have been in past decades. What is currently lacking in our operational models is any in-depth analysis of how such subcenters originate, develop, and perhaps eventually become smaller cities in their own right.
Despite the now quite long and active history of urban economic analysis, current operational models shed little light on this process; using simple incrementalism or random event generation coupled with spatial accessibility measures to produce alternative development scenarios. Traffic congestion would here seem to be an important indicator of when, if not where, a new industrial park or mixed use urban activity center is likely to be needed. Just where they spring up, or which existing centers will continue to compete successfully, is currently much less obvious.
More effort appears warranted here in at least two directions. First, more work needs to go into understanding the locational influences on basic sector industries, including both heavy and light manufacturing industries. Second, a more in-depth understanding of both intraindustry and interindustry dynamics is required. To move this process forward properly requires that we recognize the influence of locationally induced economies of scale on the site selections of such "basic" industries. The POLIS model discussed previously has made one start in this area. More in-depth analysis is necessary.
Such economies of scale arise from placing relevant resources in close spatial proximity to each other, thereby improving the productivity of participating firms. Henderson (1988) distinguishes between scale economies internal to each industry and urbanization economies resulting from the general increases in economic activity which occur as a result of locating within a large city. Both are industrial production economies of the types discussed by Mills (1967), who suggests that cities form in an economy because of scale economies resulting from
As Henderson points out, scale economies which result from the interactions between different but related industries are particularly difficult to identify, because spatial agglomeration may occur without their presence. Transportation cost savings have been cited in the past as the major reason for such spatial clustering.
More generally, locations sharing a number of traits desired in common by a variety of firms may lead to the formation of a mixed use suburban center. Such tendencies are evident in the commercial and retail as well as the industrial employment sectors. Within the retailing sector, what Berechman and Small (1988) term "shopping" agglomeration economies are clearly important. The development of multistore shopping malls recognizes the attraction to consumers of one-stop locations. These tendencies have been extensively modeled over recent years, notably by Wilson and colleagues (see, for example, Harris and Wilson, 1978; Wilson et al., 1981) whose experiments in applying quite rudimentary dynamics to spatial interaction models quickly throw up complex temporal shifts in the locational advantage of retail stores. The wider applicability of such ideas on dynamics to the commercial/office building sector (see Pivo, 1990, for example) or, with adaptations, to the manufacturing sector also needs to be looked into (see Wilson, 1987, for some ideas on this).
In addressing this question of how activity centers originate and subsequently grow, we also need to allow for the long recognized transition of our economy from manufacturing towards both service-based and, increasingly, information-based industries. In tomorrow's cities the never entirely satisfactory distinction between basic and nonbasic sectors is likely to become less useful. With many locationally footloose industries emerging, just what constitutes the benefits of a particular locational choice may be quite different from what it was just twenty years ago. A search for better theories suggests a search of the wider literature on the nature of both intraindustry and interindustry contacts, their types, frequencies, and impacts on firms' locations. For example, the emergence of network forms of organization both within and between firms is discussed by Cooke and Morgan (1993), who consider it to be a significance development in terms of not only corporate strategy but also in terms of regional development potential.
Our urban system model-based explorations into these and industrial linkages have been largely theoretical to date, and our efforts to make urban center formation endogenous to the modeling process are still highly theoretical in nature (see Berechman and Small, 1988). Clapp (1984), for example, adapted the new urban economics bid-rent model to include the effects of business contacts by a single agent, such as the corporate headquarters of a single company, on the rise of suburban centers. However, much more work is needed in this area, with potentially significant payoffs in terms of model realism.
Factors Affecting Travel Within and Between Urban Centers
There is also a need for a more normative approach to the problem, which might lead eventually to more prescriptive modeling efforts. As Dyett (1991) points out, neither current urban economic nor locational accessibility based theories provide much insight into how to best configure land uses at the neighborhood and community scale. However, such designs may prove to be an important source of personal travel reduction. He suggests more work be done to establish whether suburban mixed use centers can be designed to take advantage of cost effective urban designs (or in older suburbs, redesigns) which support walk, cycle, park-and-ride, transit and paratransit options. The location of public buildings (police, fire, city hall) as well as urban parks and other open spaces would also be important components in such designs. We must also pay more direct attention to the role played by land developers within this process. They can become key players in the creation of effective private-public partnerships. For example, they have been active in the adoption of a range of trip reduction ordinances (TROs) through their participation on local Transportation Management Associations in states such as California (Ferguson, 1990). They have also been identified as important players in the development of employer-based rideshare-supporting schemes (Southworth, 1985a). However, our current land use models contain little if anything to reflect the actual role and motivations behind these developer activities (see Levy, 1990, for an interesting discussion).
Some useful research on practical design specifications for mixed use urban centers was recently sponsored by DOT (Snohomish County Transportation Authority, 1989; Middlesex Somerset Mercer Regional Council, 1993) and by the 1000 Friends of Oregon (Cambridge Systematics et al., 1992a). Each of these U.S. studies key their discussions to a specific type of land use arrangement known as transit-oriented development (TOD), which is proposed as an integral part of a neighborhood or urban center's planned growth strategy.
Once urban centers have formed, movements between them will naturally become increasingly important. In doing urban center planning then, the type, location, and areal extent of the suburbs surrounding these commercial or mixed use centers, from which they draw their workers and customers, should be explicitly recognized and accounted for. According to the 1990 National Personal Transportation Survey (Hu and Young, 1994), while average personal trip lengths (one way, averaged over all purposes) increased from 8.7 to 9.5 miles between 1983 and 1990, increases in travel speeds kept average trip travel times relatively stable. A possible explanation for such increased speeds is the growth in intersuburb, as oppose to suburb-to-CBD, trips. Such intersuburban travel patterns are now an important component of urban development and should be further investigated.
A recent empirical study of the travel patterns of Chicago residents by Prevedouros and Schofer (1990, 1991) contains some interesting findings. Using aggregate census data to classify suburbs into growing versus stable suburbs, they surveyed 1420 respondents to compare two low-density, growing outer-ring suburbs with two suburbs selected for their higher density, stability, and inner-ring location. Among their findings (Prevedouros and Schofer, 1991): for both types of suburb, average speeds for automobile work trips were statistically similar for all but trips to the CBD. However, average trip distances were noticeably higher on average for those from the growing suburbs, resulting in residents from these suburbs staying in traffic some 25% longer than their counterparts from the stable areas. Among their other findings of interest were the high dependence of suburban females on the automobile, the substantial amount of off-peak travel being engaged in, and the possible, if not entirely clear, effects of an aging population on trip rates in the coming decade.
What the above suggests is that an integration of detailed, possibly design-oriented, models of suburban mixed use center formation with a more spatially extensive, and highly structured socioeconomic analysis of intersuburban linkages offers a useful approach to consider for further theoretical and empirical development.
As discussed in Sect. 2.5, the simulation of increasingly comprehensive urban dynamics is already quite evolved, and for multiyear forecasting the use of static-recursive approaches may be sufficient for most strategic policy making. However, further study of system dynamics is both warranted and arguably necessary for the following reasons. First, a number of studies indicate that failure to consider such dynamics explicitly may cause us to misinterpret the actual processes of urban change. Second, intriguing possibilities for more direct representations of detailed traveler behavior now exist than at any time in the past. By making use of microsimulation methods in conjunction with massively parallel or vectorized computers its now possible to generate tens of thousands of daily activity patterns in a surprisingly short turnaround ("wall clock") time. To take advantage of this opportunity the travel activity analysts needs to develop explicitly dynamic equation sets when trying to represent the behavioral responses of travelers.
Over the past two decades a number of research efforts have addressed the issue of introducing dynamics more explicitly into our urban systems models. These include a number of efforts focused on the evolution of all or parts of complete urban systems, notably (1) the work by the Leeds group in the United Kingdom based on catastrophe theory (Wilson, 1981), (2) the work by the Brussels Group in Belgium based on uses of self-organizing system's theory and micro-simulation (Allen, Engelen and Sanglier, 1986), and (3) a number of efforts, including research in France (Fournier, 1986) and Italy (Bertuglia et al., 1981), to adapt the urban dynamics approach proposed by Forrester (1969) to real world cities.
Wilson (see, e.g.,Wilson, 1987) offers the following approach to tracking the change over time in the size of a facility at location j, Wj, as a function of profit accrued at that location. Let Rj = the revenue attracted to that location, then profit at j is given as:
where kj = the unit cost of floorspace in location j for the given facility type. Assuming a desire to maximize such profits among facility suppliers (e.g. developers) a suitable hypothesis for change in facility size is taken to be
for some constant 
.
Spatial equilibrium is then assumed to be reached in a given time period
t, when the value of Eq. (50) equals zero. A more general functional form
is proposed for applications, as is the conversion from a differential to
difference equation form for computational purposes, i.e.
which for an exponent n=1 produces logistic growth. Rearrangement then gives:
Experiments with these sorts of equations for retail activity location
systems (Harris and Wilson, 1978; Beaumont, Clarke and Wilson, 1981) show
the potential for significant oscillations in facility sizes, including
possible jumps back to zero floorspace in some Wj values when
the value of (1+Rj)
in Eq. (52) is greater than 2.
One conclusion from such findings is the need for caution in oversimplifying the assumptions involved in detailed travel pattern and associated land use forecasts. A more positive view of the picture is that such discoveries will allow us to experiment with the robustness of alternative urban land use and transportation infrastructure plans if we can find a way to bring them effectively into our operational models. Wilson (1987) describes a beginning in this process by formulating a dynamical version of the Lowry model based on the above ideas. These developments also help to tie the above described spatial interaction approach more closely to economic concerns. For example, consider the mathematical programming version of the shopping model discussed in section 2.4.2 above, and repeated here for convenience:
subject to:
and
where Sij = the flow of shopping revenues from shoppers in zone i visiting shops in zone j, Wj = the size of shopping facilities in zone j in units of floorspace, and the two terms in the objective function measure consumer surplus and the spatial dispersion of destination choice (spatial entropy). Now if we let:
Where the unit floorspace costs kj =
/µ,
and where µ is the lagrangian multiplier associated with the total
floorspace supply constraint (55) above, we have
placed a supplier's floorspace requirement as a condition of the
optimization. The resulting equilibrium pattern of Wj's is
therefore based on a supplier profit motive as well as on consumer
benefits maximization.
To be of practical use such developments will need to incorporate the effects of a variety of other factors, notably the effects of spatially varying prices. A search for ways to bring such ideas of market driven, profit-induced change into modeling, via the use of modern day highly computer intensive microsimulation techniques appears worth pursuing. Of particular interest are methods which can use difference equations to simulate the dynamics involved in mixed use urban activity center formation and decline. For example, can we use explicitly dynamic equations to microsimulate the alternative temporal paths available to individual (that is, synthetically constructed) companies as well as synthetically constructed travelers? In particular, can we simulate the effects of spatial agglomeration of activities based on the mutual locational benefits discussed above by microsimulating the passage of information as well as goods and people between such companies?
A strong argument can be made that as far as land use-transportation modeling efforts to date are concerned, toolmaking is more advanced than theory. It would be difficult to find an area of research that has drawn on a greater variety of mathematical, statistical, and computational methods in its search for empirical validation and subsequent practical applicability. Yet the application of many of these techniques is much less widespread within the planning profession than might be expected. Few practicing regional or metropolitan planners calibrate their own multinomial logit models or experiment with alternative land availability or density constraints as part of a nonlinear mathematical programming exercise. Nor is the issue simply one of technical training. In order to encourage practicing planners to make greater use of the models which do exist, the models need to be made easier to use.
If planners from more than one jurisdictional level (local, metropolitan, statewide, or regional) can be brought together by use of a common, easy to use modeling, possibly game-playing software, then improved models could possibly be transformed into consensus building tools, rather than the seemingly arcane components of a planning process in which only one or possibly two experts within any metropolitan planning agency have anything to do with them directly.
Researchers have already developed a range of user interfacing capabilities, including graphical interfaces, for commercially available models such as MEPLAN. However, we can go much further here, as evidenced by the use of highly interactive, multimedia, decision-support aids in other fields. The emergence of reasonably priced and generally accessible geographic information system (GIS) software is the latest step in this development of decision-support tools. By linking a relational database management tool to software programs for manipulating spatial primitives (points, lines, polygons), adding the land use and transportation modeling subroutines themselves, and building around all of these an easy to use, map-based interface, we have the principal components of a spatial decision support system (an SDSS). Ongoing developments in the SDSS arena promise more effective manipulation of both spatial and nonspatial data elements, in the short term through the more efficient selection of which computations to carry out via database manipulations and which to continue to model through the more context-specific algorithms (see, e.g., Lolonis, 1993). The field of urban transportation modeling is only now beginning to make use of such GIS tools (Prastacos, 1991; Hartgen et al, 1993; Anderson, Kanaroglou and Miller, 1994; Spiekermann and Wegener, 1994).
Current developments in database encapsulated software systems and object-oriented programming languages also suggest a move towards more flexible software systems (Stevens, Tonn, and Southworth, 1994). Again, we are just beginning to make use of such advances in the field. For example, the SUSTAIN model, an offshoot of the TOPAZ efforts at CSIRO in Australia, is currently being developed as an object-oriented programming approach specifically directed at the linkages between transport, urban form, and energy consumption (Roy and Marquez, 1993).
A key component of such decision-support systems will be their ability to help the planner resolve often competing energy, environmental, fiscal, social, and economic goals. Here the use of multicriteria decision-making methods are also worth further explorations. Three increasingly popular decision aides that have been applied recently to transportation project assessments include Saaty's Analytic Hierarchy Process (see Zahedi, 1986 for a review), Roy's ELECTRE III method (see Roy, Present, and Silhol, 1986 for an application to Paris metro station locations) and Concordance Analysis (see Giuliano, 1986b, for an application to the ranking of alternative highway, bus transit, ride sharing and commuter rail investment projects in Orange County, California).
Each of these approaches can be applied to, actually on top of, the outputs of any of the above reviewed land use-transportation models as a further aid to strategic as well as project specific decision making. In particular, they can enhance the evaluation of trade-offs between the costs of plan implementation and the energy savings, emissions reductions, and the economic and social impacts (including intercommunity equity impacts) of proposed travel reduction strategies. Any realist hope for the acceptance of transportation plans which significantly reduce petroleum based fuel use and greenhouse gas emissions will certainly require plans which also address such trade-off issues (see Bae, 1993).
There are a number of additional benefits to developing a highly interactive, multimedia approach to decision support. It's much easier to remain skeptical about a batch driven process in which the user's only interaction of note is data input than it is to feel the same way about a process in which he or she is an active component. Highly interactive user-centered planning tools can prove to be very powerful decision-support aids. This is particularly true for spatially explicit problems. Here an interactive game of consequence analysis is appealing; a game in which the analyst gets to experiment with different starting points, parameter values, land use controls, pricing schemes, fiscal and other constraints, and a range of differentially weighted plan objectives. A major benefit from the use of such interactive systems is likely to be the knowledge both gained from and added back to the system by the analysts. Tomorrow's decision-support systems are likely to combine text, and geo-graphics with sound, and video, including animation (see Wiggins and Shiffer, 1990, for a discussion). Software which allows the user to place aerial photographs behind model constructs, such as network upgrades and new land use arrangements, can also serve the process of bringing the art of modeling closer to the planner. With the arrival of global communications in the form of the Internet and World Wide Wed, a considerable increase in the use of interactive, map-based editing tools should be expected to inform the land use and transportation planning process.(6) Indeed, in coming years, whenever a major metropolitan planning proposal is transmitted in digital form its likely to encourage a significant number of similarly transmitted responses from a growing number of interested parties. This constitutes a type of information which our present models and supporting databases are perhaps ill-equipped to handle.
The development of such interactive, model-encapsulated analysis tools constitutes a challenging research and development task. Well-designed decision-support system require as much thought (and at least as much money to render operational) as do the computer models of land use and transportation interaction on which they are based. The tools now exist with which to build such software. However, current commercial GIS packages are still some way from being the spatial decision-support tools we need. Experience with such software in the field of urban and regional transportation modeling has been quite limited to date. Education in how to construct, adapt, and use such software tools is now required within the transportation planning profession.
6. Early efforts in this area currently include those at the University of California at Davis (Johnston, 1995).
This
site will be under constant change. If you have any suggestions for
improving the site, please send a message to
Lynette Engelke.
This page was updated July 14, 1998.
Top
of Page
On
to References
Back
to Table of ContentsBack
to Land Use Compendium Table of ContentsBack
to Clearinghouse Main PageBack
to TMIP Homepage
