Activity-based Models: a Comparison of Approaches Used to Achieve Integration Among Trips and Tours Throughout the Day
J L Bowman, Transportation Systems & Decision Sciences, US; M Bradley, MB Research & Consulting, US
We compare various integration techniques used by five activity-based models that have been used for travel forecasting in the US, providing a conceptual understanding and a reasoned discussion of their strengths and weaknesses.
This paper examines the so-called activity-based models implemented to date in the United States, explaining and comparing the various techniques that have been used to achieve model integration. These models integrate the representation of activities and travel conducted by an individual, and in some cases an entire household, over the course of an entire day. Such integration is what distinguishes these models from earlier trip-based and tour-based models. Three techniques of integration are typically used. First, a model is developed that simultaneously represents outcomes spanning the tours in a day and, in some cases, the persons in a household. Sometimes called an ?activity pattern? model, it provides what could be called horizontal integration across all the dimensions of choice. Second, since the outcomes that need to be modeled are more complex than can be represented in a single activity pattern model, additional aspects of choice are modeled by breaking the outcome into a conditional model hierarchy or a chain of models. Models lower in the hierarchy (or later in the chain) take as given the outcomes higher in the hierarchy. This achieves what has been referred to as downward vertical integrity. Done properly, it assures that lower level models adhere to constraints imposed at higher levels, and makes the lower level models indirectly sensitive to all variables that directly affect the upper level outcomes. Just as important as downward integrity is upward vertical integrity. Upward integrity comes from making the upper level models appropriately sensitive to variables that affect the upper level outcome, but can?t be measured directly because they differ among the undetermined lower level model outcomes.
In practice, a variety of techniques have been developed and used to achieve horizontal, downward and upward integrity. This paper examines and compares the techniques employed by five model systems that have been used for travel forecasting and policy analysis in the United States (in Portland, Oregon; San Francisco; New York City; Columbus, Ohio; and Sacramento.) Among the techniques considered are:
--individual pattern models that explicitly identify the tours in a day and the intermediate stops that occur on the way to and from the tour?s primary destination,
--individual pattern models that identify the purposes for which tours and extra stops occur, without associating the purposes and stops to specific tours,
--models implemented without an activity pattern model, using instead a cascade of tour generation models by purpose,
--household pattern models that identify the primary activity of the day for all persons in the household,
--household pattern models implemented as a hierarchy, identifying the presence of joint tours conducted together by household members and tours conducted by individuals to achieve household maintenance activities,
--time window techniques that enforce realistic time constraints: conditional tours and stops are limited to windows of time not occupied by higher priority tours and stops,
--half-tour models that identify the number and purpose of intermediate stops on each half-tour, given the purposes for which stops are to be made in the day and the stop purposes already included on higher priority half-tours,
--half-tour models that simulate stop locations, timing and mode in a temporal sequence emanating from the tour?s origin or destination,
--long-term models of usual work and school locations that condition the daily activity pattern, tour and stop locations,
--disaggregate tour mode and mode-destination logsums used to measure accessibility in various higher level models within the model system, relying on simulated or assumed time-of-day choice in the logsum calculation,
--tour mode and mode-destination logsums calculated for aggregate market segments and representative conditional alternatives, requiring less computation time than disaggregate logsums, and
--aggregate trip destination logsums that measure the attractiveness of locations along the path between a tour?s origin and its primary destination.
The techniques employed are not always directly comparable, the models are still relatively new, and the situations in which they are used vary considerably. Therefore, a meaningful empirical comparison of the techniques is not feasible. Accordingly, the emphasis in this paper is on providing a comparative conceptual understanding of the techniques themselves, including a reasoned discussion of the potential strengths and weaknesses of the various approaches.
Association for European Transport