Activity-Based Modeling System for Travel Demand Forecasting
Chapter 2: A Critical Review of the Trip-Based, Four-Step Procedure of Urban Passenger Demand Forecasting
Practically all tools currently available for passenger travel demand forecasting and policy analysis are based on the four-step procedure. The procedure was developed in the 1950s and 1960s during the post-war expansion period, when:
- Urban population was rapidly growing,
- Motorization was progressing, and
- Suburban sprawling was starting.
The emphasis in transportation planning at that time was infrastructure development. The issue at hand was where to build a new freeway and how many lanes were needed. Because of such straightforward planning contexts, coarse forecasting procedures sufficed at that time. In fact, it is not difficult to see that when the population of a metropolitan area doubles, the total number of trips will approximately double and increases in trips can be relatively easily forecast once one can determine in which parts of the metropolitan area increases in residential and work populations will take place.
Planning emphasis has changed substantially since then. In the 1970s Transportation Systems Management (TSM) was promoted, while in the 1980s Travel Demand Management (TDM) was proposed. Currently the transportation planning community embraces a more inclusive concept of Transportation Control Measures (TCM). The measures being considered are extensive and increasingly more sophisticated and are fine-tuned to target specific traveler segments. The trip-based four-step procedure, developed to serve the planning needs of decades ago, is not best suited to address these new transportation measures.
2.1 Advantages
The simplification incorporated into the four-step procedure made urban passenger travel demand forecasting practicable using standard survey methods, census and other existing data, and computational capabilities that had been available. The simplifying assumptions adopted in the procedure facilitated quantitative analysis of travel demand, which is a result of complex (to analyze) travel behavior. In particular, the development of a standard analysis package, Urban Transportation Planning System (UTPS), led to the development of PC-based transportation planning packages, which in turn have made the forecasting procedure affordable to practically any MPO.
2.2 Internal Inconsistencies
The procedure, however, contains several well acknowledged internal inconsistencies. For example, the area-wide totals of zonal trip productions and attractions normally do not coincide with each other, requiring some adjustment; zone-to-zone travel times used as input to trip distribution and modal split are not necessarily consistent with travel times that are derived from the network assignment; and trips are assigned to different time periods of the day (e.g., peak vs. off-peak) prior to network assignment, usually using heuristic procedures. For additional issues involved in the application of the four-step procedure, see Table 2.1.
Table 2.1: Sample of Recognized Issues Involved in the Application of the Four-Step Procedure
|
2.3 Data Inefficiency
When disaggregate choice models were proposed in the 1970s, it was argued that the aggregate four-step procedure was not data-efficient. This is mainly because the procedure was developed when available computational capabilities were very limited and costly and statistical theory for model estimation was not well advanced. As a result, model calibration procedures adopted inefficient data use (especially the aggregation of household survey results into zonal averages) led to an inefficient parameter estimation (e.g., trip distribution models).
2.4 Lack of Behavioral Foundation
More problematic are the implicit assumptions in the four-step model components which lack behavioral foundation. For example, consider trip generation models. Implicit in typical linear-regression or cross-classification models of trip generation is the assumption that the number of trips generated by a household is a function of the number of its members and the number of vehicles available. This assumption does not reflect the well known behavioral fact that employment status affects travel behavior. Therefore, the number of workers in the household affects trip generation.
2.5 Resulting Problems as a Policy Tool
Suppose parking pricing is implemented in the downtown area. This event may cause some travelers to choose suburban destinations. This result, however, is not accounted for by the four-step procedure because the total number of trips attracted to the downtown area is determined in the trip generation phase, which typically does not incorporate parking cost. The procedure would indicate no change in the number of trips attracted to the downtown area before and after the implementation of parking pricing. Likewise, effects of congestion on travel demand cannot be fully accounted for by the four-step procedure because trip generation models are typically insensitive to travel time (this problem cannot be alleviated by incorporating feedback loops).
Trip-Based: The four-step procedure treats each trip as an independent entity for analysis. This assumption is central to the four-step procedure in the sense that its model structure hinges on it. This dependence, however, leads to a number of serious limitations, especially when its application to TCMs is considered. The problems stem from the fact that trips made by an individual are linked to each other and the decisions underlying the respective trips are all inter-related.
Example of Travel Mode Choice for Multi-Stop Trip Chains: Consider a home-based trip chain (a series of linked trips that starts and ends at the home base) that contains two or more stops. The four-step procedure looks at each trip at a time and determines the best mode for it. Let h be the home base and i and j be the destination zones visited in a trip chain. There are three trips, (h, i), (i, j), and (j, h). When a trip-based, post-distribution mode choice model is applied while comparing the alternative modes available between each pair of zones, it is entirely possible that bus is assigned for (h, i), drive alone for (i, j), and carpool for (j, h). This contains two major problems. First, the result violates the modal continuity condition. Mode choice for a trip with non-home origin is regulated by the mode selected for the first home-based trip; if one leaves home by bus, it is normally not possible to choose the drive-alone mode in subsequent trips. On the other hand, once one leaves home by driving alone, all subsequent trips tend to be made by driving alone. Second, the result ignores the behavioral fact that one will most likely plan ahead and choose a mode while considering the entire trip chain, not just each individual trip. One may decide to take the auto even when good bus service is available between h and i and between j and h, but because no bus service is available between i and j.
Treating each individual trip in isolation becomes a problem on many occasions. For example, commuters who make trips on the way to or from work (e.g., dropping off/picking up children) are less likely to switch from the drive-alone mode when TDM measures such as congestion pricing are implemented. What is termed "activity re-sequencing" in this study is another example. Suppose a drive-alone commuter stops by at a grocery store on the way home from work. Faced with congestion pricing, this commuter may choose to take the bus to commute, and go shopping by auto at a grocery store near home after returning home by bus. The trip-based four-step procedure is not capable of addressing such secondary and tertiary changes brought about by the primary commute mode change.
Over-Predicted Mode Shift: Because its trip-based structure does not recognize the mode continuity condition, it is logically expected that the procedure over-predicts mode changes. The problem is multiplied by the fact that the modal split phase tends to be most sensitive to changes in the travel environment because it often incorporates disaggregate choice models. As a result, the four-step procedure may grossly over-estimate mode shift, when in fact travel mode may be the last thing travelers wish to change.
No Time Dimension: The fact that the four-step procedure does not incorporate the time-of-day dimension is curious when congestion -- which has been the single most important concern of transportation planning -- occurs with the concentration of demand in the same area at the same time. The absence of the time dimension is behind some of the recognized issues listed in Table 2.1. In addition, it implies that departure time choice cannot be incorporated into the forecasting procedure (without introducing ad hoc assumptions). This in turn implies that the four-step procedure cannot be effective in the analysis of peak spreading in general and congestion pricing in particular.
The time dimension is crucial in air quality analysis. Because air quality is a function of complex meteorological relationships, it is important to be able to predict when within the day pollutants are emitted, not just the total amount of emissions. Determining the split between hot and cold starts in any consistent manner would also require the introduction of the time dimension into the analytical scope. Furthermore, recent interest in Intelligent Transportation Systems (ITS) technologies calls for the ability to predict traffic dynamics on the network.
Vehicle Ownership: An area where very little effort has been directed at the Metropolitan Planning Organization (MPO) level is vehicle ownership modeling. This may not be a problem if the number of vehicles available to the household is the only concern (which in fact was the case at the time when motorization was progressing at fast rates). Recent concerns with air quality and fuel consumption, however, imply that increased importance is assigned to which types of vehicles are chosen by households and how much and where each type of vehicle tends to be used. This calls for the implementation of vehicle type choice models, and development of vehicle allocation models that predict which vehicle will be used for which trip.
Representing Accessibility and Land-Use: The state-of-the-art has not advanced enough to incorporate into the forecasting process:
- Impact of new highway and transit facilities on land-use,
- Impact of travel patterns (materialized demand) on land-use,
- Impact of accessibility (congestion) on trip generation and attraction, and
- Impact of multiple-activity land-use development (e.g., shopping malls) on travel demand.
2.6 Summary
In summary, the following can be listed as the limitations of the four-step procedure in the current policy contexts:
- Trip-based, sequential structure,
- Lack of the time-of-day dimension,
- Limited sets of explanatory variables,
- Limited behavioral responses,
- Consequently unresponsive to most TDM measures,
- Trip generation unresponsive to congestion and pricing,
- Consequently the trip distribution phase is not fully responsive to system change,
- Inability to address vehicle fleet mix evolution, and
- Totally exogenous land-use, economic and socio-demographic input.
While some of the problems discussed in this section may be resolved for certain situations by introducing new model elements, the problems stemming from its atemporal, trip-based structure are difficult targets for improvement within the framework of the four-step procedure.
Before closing this section, it is emphasized that no single model system is suited for all study objectives. The trip-based, four-step procedure continues to be an effective demand forecasting procedure for certain types of problems. Yet, current policy contexts call for alternative models. The array of transportation planning tools available to policy makers needs to be expanded.
