AMOS Model Framework

• Overall
• Synthetic Population Generator
• Fixed Activity Location Choices
• Vehicle Ownership
• Fixed Activity Prism Generator
• Child Daily Status and Allocation
• Adult Daily Status
• Activity Skeleton Reconciliation
• Activity Travel Pattern Simulator
• Activity Travel Reconciliation
• Time Use Utility

UrbanSim - AMOS - MALTA Interface

UrbanSIM - AMOS - MALTA Framework

AMOS Components

Time-Space Prism Generator

AMOS generates time space prisms for each individual based on mandatory work and school activities that they undertake outside the home. These time space prisms capture the latest and earliest departures or arrivals at key locations and activities along the course of the day. These time space prisms should not be violated, although the process in AMOS now allows for infrequent random violations of time- space prisms to reflect the fact that these constraints are not always hard constraints that can never be violated. The time-space prism vertices are estimated using stochastic frontier models that are capable of estimating unobserved vertex conditions based on observed attributes that are influenced by the unobserved prism boundaries.

Child Dependency Allocation Model

Children have mandatory activities that may include school or other after-school activities. The locations and timings of such activities are often fixed. Children may take the school bus, use public transit, or walk or bicycle to these activities. In such instances, they may be assumed to operate reasonably independently without the specific requirement that they be accompanied by an adult of the household. However, in instances where the child is not using an independent mode of transport and is dependent on the parents or adults in the household for transport, then a child dependency allocation model will allocate the child’s mandatory activity-travel to an adult in the household. This happens up-front as part of the “skeletal activity agenda” formation process. Thus, the adult time-space prisms will be modified to reflect the presence of mandatory child activities that must be served. The child dependency allocation model is a simple multinomial logit model that allocates each mandatory activity to an adult in the household. The feasible choice set will always include the option for the child to undertake the activity-travel “independently” without accompaniment of an adult of the household.

Vehicle Allocation Model

AMOS includes a vehicle allocation model that generally assigns a vehicle to each driver in the household, in cases where there are multiple adults in the household. Even in cases where there are fewer vehicles than adults, a vehicle allocation model is applied to represent the broad vehicle allocation pattern in the household. This model helps associate each driver with a primary vehicle and each vehicle with a primary driver. In general, it has been found that drivers in households tend to be associated with specific household vehicles and drive a vehicle on a much more regular basis than other vehicles in the household. This vehicle allocation model, applied at the vehicle level and employing simple multinomial logit specifications, will associate a primary driver to each vehicle in the household. It is to be noted that vehicle ownership and type choice are considered longer term choices and are included in the land use microsimulation model system as opposed to the activity-travel model system. Thus, the fleet of vehicles owned by a household will be determined at the population synthesis stage as part of the longer term choice model system. The multiple discrete continuous extreme value (MDCEV) model will be used to determine vehicle ownership by type and vintage.

Joint Activity Type – Duration Model

Within each open period depicted by a time-space prism, an individual may engage in discretionary activities. After the start of an open period, an individual will decide what type of activity to undertake the approximate amount of time that will be allocated to this activity. By jointly determining activity type and duration and considering the simultaneity in the choice process, it is possible to capture unobserved factors that may impact activity type choice (activity generation) and time allocation to different activity types. The project team will have to make specific decisions regarding the activity type categorization to be included in the model, but the current plan is to incorporate as disaggregate an activity purpose categorization as possible subject to limitations of the model estimation data set. It is to be noted that one of the elemental alternatives in the choice set may include the choice to undertake an in-home activity. The project team has developed a new probit-based discrete-continuous modeling approach to estimate parameters in simultaneous discrete-continuous model systems without having to make arbitrary distributional transformations or employ limited information sequential estimation approaches.

Joint Destination – Mode Choice Model

After the activity type and duration are estimated, the destination and mode choice are determined for the activity. The mode-destination combinations included in the choice set are only those alternatives that are feasible. The destinations include only those that can be reached by the fastest mode within the constraints of the time-space prism. Only those modes that are actually available to the individual are include in the choice set so that modal constraints and schedules are explicitly recognized and addressed. The joint destination-mode choice model takes the form of a classic nested logit model with modal alternatives nested within destination choices. If a nested logit model with intuitively appealing coefficients cannot be estimated, then a joint logit model may be considered, although due consideration must be given to the fact that the alternatives are inevitably correlated with one another leading to violations of the IIA property.

Activity Accompaniment Model

The new AMOS will incorporate an explicit model component to reflect joint activity engagement. If the mode choice model suggests that a multi-occupant vehicle mode is going to be used, then this model component is invoked. Otherwise, this module remains dormant and the person is assumed to move forward with activity engagement without consideration of accompaniment arrangement. Note that mandatory serve-child activities and trips are preset at the beginning to reflect child accompaniment for such activities. With respect to discretionary activities, if the mode choice model predicts that multi-occupant mode may be used, this model uses a simple multinomial logit to identify whether the other occupants are solely household members, non-household members, or a combination thereof. If household members are to be present on the trip, this model will run a heuristic check to identify household members who are capable of accompanying the individual on the activity- trip. If no household member is “available” to accompany the individual, then this option is rejected and only non-household members are allowed to accompany the person on the trip. Several heuristic rules will be applied within the context of this model component to ensure that the accompaniment arrangement does not defy basic logic and coupling constraints.

Vehicle Choice Model

Although an individual may be primarily associated with a certain vehicle in the household, it is always possible for individuals to switch vehicles among one another – particularly when multiple household members are going jointly for an activity or trip. The vehicle choice model is invoked every time multiple vehicles of the household are available at the location and time that the next activity is going to be undertaken. The model takes into account the primary vehicle associated with the driver and uses that information, together with other trip attributes, to predict the specific vehicle that will be chosen for the trip. As an initial simplification, the model system will allow this module to be called only when the activity-trip is going to commence from the home location and multiple vehicles are parked at the home location at the instant that the choice is made.

Activity Duration Adjustment Model

The activity duration is initially estimated as part of the joint activity type – duration model. However, the actual activity duration is not known until the person arrives at the destination and then must adjust the activity duration in such a way that time space prism boundaries are not violated (occasional violations may be allowed). A heuristic activity duration adjustment module is being developed and incorporated into the activity based model system. This duration adjustment module takes into account the actual arrival time at a destination, the remaining time available in the time-space prism, and the initial estimate of duration to estimate a revised activity duration. A hazard-based duration model coupled with heuristic decision rules is employed to make the activity duration adjustments.

Consistency and Feasibility Checks

One of the powerful features of AMOS is that it combines quantitative econometric model specifications with heuristic decision rules that are usually incorporated in computational process models. There are numerous consistency checks and routines that must be carried out throughout the activity-travel simulation process. Mandatory serve-child activities must be undertaken and children cannot be abandoned at will. Departure times and arrival times must be consistent and follow a natural sequence. At the end of a discretionary activity, a determination needs to be made as to how much time is available in the time-space prism to engage in an additional activity. A multinomial logit model can be applied to determine if an individual will continue the same activity to fill up the prism, proceed to a new activity (then invoke the joint activity type – duration model system), or simply proceed to the next fixed activity early. Consistency checks and routines will be applied to ensure that no physical and coupling constraints are violated. If a multi-occupant trip is going to be undertaken, then accompanying household members must be in a position to accompany the individual. Moreover, the activity duration may have to be modified to reflect the influence of the other accompanying person’s time-space prism constraints. AMOS currently incorporates numerous heuristic rules and conditions that ensure the simulation of feasible and internally consistent activity-travel patterns. In this project, these routines will be further enhanced and strengthened to provide for robust simulation of activity-travel patterns. Within this project, two types of consistency will be ensured:

• Within-household consistency
• Within-person consistency

Time Use Utility Measure of Welfare

The notion of quality of life is gaining increasing attention in the travel behavior arena. There is work that is beginning to look at people’s subjective well- being in the context of their travel patterns and activity engagement routines. The project team has developed a simple utility formulation that offers a measure of the welfare or satisfaction that an individual derives from his or her activity-travel pattern. The measure incorporates time spent traveling, time spent at activities, and time spent inside the home. Based on the time allocation patterns associated with activity-travel engagement, this measure can be computed and used to determine the loss (or gain) in utility due to the implementation of a policy or change in system capacity and conditions. This time-use based utility measure is a very convenient mechanism for performing environmental justice analysis and to easily quantify the impacts of various projects or alternative policies on traveler welfare, both at a disaggregate level and at a more macro level for different market segments. It is envisioned that this time use utility measure will be used more in the context of alternatives analysis and policy evaluation than in the standard forecasting mode.

-- PaulWaddell - 12 Nov 2009