Determining Sequentiality in Activity-based Models

Determining Sequentiality in Activity-based Models


Eric Petersen, RAND Europe, NL; Peter Vovsha, PB Consult, US


This paper will investigate the sequence of fundamental choices that fill in the framework of a daily activity-travel pattern in order to improve the state of activity-based modelling.


Despite the many recent advances made in activity-based models, certain fundamental issues remain unresolved. Perhaps none is greater than how to determine the sequence of fundamental choices that fill in the framework (or skeleton) of the daily activity-travel pattern. While some academic modelling exercises have successfully simultaneously estimated the joint and individual daily travel patterns of two household members (normally household heads in a nuclear household), a practical activity-based model must be able to cope with households of up to 10 members, requiring a different strategy for handling all possible permutations.

The activity-based microsimulation modeling framework used in the New York and Columbus, Ohio models depends upon a sequence of models where the behaviour of each individual is modeled sequentially (not simultaneously) and the information for each person is transmitted to the 2nd, 3rd, through Nth person in the household and typically affects these subsequent estimation results. This is particularly true for the individual level maintenance and discretionary travel models. The Columbus, Ohio model is one of a very few models to explicitly model joint household travel, though this is modeled after all mandatory activities (work and school) and associated travel has been accounted for through prior choices.

This paper will not attempt to definitively address the question of whether joint or individual travel needs should have the highest placement in the travel hierarchy. It will take on a narrower question about the sequencing of travel decisions to allow this question to be addressed in future research. This paper will return to the joint mode choice ? car allocation model framework developed for the Atlanta, Georgia metropolitan region. The specific issue that we will examine within this framework is whether joint travel decisions involving shared car use are made at the ?top? of the decision hierarchy or emerge towards the bottom of the hierarchy and are essentially an outcome that arises when sufficient cars are not available within the household to satisfy all the scheduled tours at a particular time slice.

We are estimating this new type of model on the basis of a travel diary-type survey for Atlanta. The Atlanta data covers two days, but is still essentially cross-sectional data. Cross-sectional data of this type is extremely difficult to use to determine the sequencing of choices, and one is basically left with the option of modelling two alternative structures to see which one has the best statistical fit. Ideally, one should introduce questions regarding causality and proper sequencing to the household surveys in order to clarify the order and conditionality of decisions as well as the formation of the choice set. However, this approach is not available to us for the Atlanta region.

Nonetheless, even cross-sectional data offers some opportunities to infer the underlying decision-making sequence through analysis and comparisons of adjustments made to certain choice dimensions in view of the actually made choices across other dimensions. For example, if one can confirm statistically that having joint non-mandatory activities and travel do not significantly affect duration of work activity for workers, we can conclude that work schedules are independent from non-mandatory parts of daily patterns. Alternatively, if presence of joint activities significantly affects work schedules we can conclude that work schedules are planned by workers having joint activities in mind. Similar types of analysis will be implemented for the sequence of choices associated with mode, car allocation, and car type choice. For example, the choice of public transport can be a consequence of better level-of-service (if car was available), or because of captivity (if household or person does not have access to cars) or as a result of intra-household allocation and constraints (if some other travel tours were implemented by the other household members by auto at the same time). In presence of several strong impacts, the underlying causality can be also explored by comparison of alternative hierarchical (nested) structures.

In this study of model sequentiality, we intend to pursue another avenue and draw on the experience of rule-based models, such as ALBATROSS and ILUTE. The modellers responsible for these models have considered the sequencing of choices, and have developed longitudinal datasets better suited to exploring tradeoffs within households, as the surveys can indicate what travel behaviour should be considered stable and what should be considered occasional within any particular household. We intend to investigate the six-week-long MOBIDRIVE data set to inform the discussion over model sequentiality.

The model structure can be briefly summarized as follows: all estimated tours are gathered and sorted at the household level including individual and joint tours. For each tour, the competing tours where there is time window overlap (i.e. the same car cannot be used for both) are identified. The number of available cars is derived from reported car ownership. For each processed tour, the binary choice of auto versus non-auto mode is considered. The non-auto mode category includes various public transport and non-motorized options. Thus, for households without car-allocation conflicts (households with high car ownership), mode choice will be processed independently for each tour without concern for competing tours? utilities. When there is meaningful competition over vehicles, each household member must take into account the utility of the other household members which creates linkages across individual binary models within a multi-level choice stucture. This structure allows for an integrative accounting of all vehicles and household tours with the further advantage that intra-household competition over vehicles has been explicitly added into the model framework. This work adds important insights into the intra-household structure of individual car preferences and car-allocation mechanisms into the mode choice modeling framework. Alternative structures of the model and linkages across elemental binary sub-models correspond to different assumptions about the sequence of decision making steps associated with mode choice, car allocation to household members, and car type choice. Our intention is to report on possibility of screening the underlying sequentiality.


Association for European Transport