International Choice Modelling Conference, International Choice Modelling Conference 2017

Font Size: 
Household Daily Non-Mandatory Activity Participation and Duration Modeling Accounting for Person Level Budget Constraints
Rajesh Paleti, Abdul Pinjari

Last modified: 28 March 2017

Abstract


A key methodological and behavioral innovative component in recent Activity-Based Models (ABMs) (more specifically, the ABM for Southern California) is the household-level non-mandatory activity participation model. This model currently takes the form a ‘Multiple Discrete Continuous Extreme Value (MDCEV)’ model. While traditional ABMs use a series of simple models to predict non-mandatory activity participation decisions in a sequential manner (which is often not correct), the MDCEV model predicts all the following decisions in a household in single step:

1)      Individual non-mandatory activity participation decisions in different out-of-home activities of all household members

2)      Amount of time spent in different individual non-mandatory activities in different out-of-home activities by all household members

3)      Joint non-mandatory activity participation decisions in different out-of-home activities (i.e., the composition of household members who partake in the same activity jointly)

4)      Amount of time spent in joint non-mandatory activities in different out-of-home activities

The key advantage of the MDCEV framework is that it accounts for complex intra-household interactions among different household members by allocating the total household time available in a day to different household members in a utility-consistent manner.

 

The MDCEV model assumes that each household optimally allocates time among different alternatives subject to a budget constraint. The budget constraint ensures that the total time spent in all non-mandatory activities by all household members must be equal to the total time available in the household. The alternatives correspond to the all possible combinations of participating people and non-mandatory activity purposes. An alternative with only one participating member corresponds to individual non-mandatory activity and an alternative with multiple participating members corresponds to joint non-mandatory activity. The MDCEV model output is used as an input in all scheduling models downstream consistent with the core assumption of ABMs that activity participation decisions must determine travel (i.e., tours and trips). In some cases, the MDCEV outputs serve as key explanatory variables in the scheduling model components. For instance, activity participation and duration variables are used as explanatory variables in tour frequency models with the assumption that high activity participation rates result in higher tour frequency. In other cases, the MDCEV output significantly constraints the choice set of the scheduling model components. For example, if MDCEV model predicts that a person participates in shopping and maintenance activities during the day, only shopping and maintenance activity purposes constitute the choice set of the tour and stop activity purpose model components. So, it is extremely critical to ensure the prediction accuracy of the MDCEV model because any errors in the MDCEV model can potentially propagate all through the scheduling modules. In this context, it is important to understand key issues concerning the MDCEV model. These include:

(1)               Single Household Level Budget Constraint: The current version of MDCEV model works with a single household level budget constraint. So, the model ensures consistency of time predictions with the total household available time but it does not ensure consistency at person level. To see this, consider a household with two people P1 and P2 with 2 hours and 3 hours of available time, respectively. The current version of MDCEV model ensures that total time spent in all non-mandatory activities (individual and joint) is equal to 5 hours ( = 2 hours of P1 + 3 hours of P2). However, it is possible that either time allocated to P1 exceeds 2 hours or time allocated to P2 exceeds 3 hours. Alternatively, it is possible that some of the available time for certain individuals is left without being allocated to any non-mandatory activity. For instance, Table 1 shows some of the possible MDCEV predictions that are inconsistent with individual available times. While the first and third MDCEV predictions violate the time availability constraint of P1, the second set of predictions not only violate the time constraint of P1 but also leave 30 minutes of P2’s time unallocated.

Table 1 Inconsistent MDCEV Predictions due to Household Level Budget Constraint

Possible MDCEV Predictions

Inconsistencies

5 hours shopping by P1

Violates P1 time constraint

5 hours joint shopping by P1 and P2

Violates P1 time constraint (because currently MDCEV treats 5 hours of joint shopping as 2.5 hours by P1 and 2.5 hours by P2)

0.5 hours of P2 time remains unallocated

1 hour shopping by P1

2 hours shopping by P2

Violates P2 time constraint

(2) Extremely Low Participation Durations: It is currently possible that MDCEV predicts extremely low activity participation durations. For instance, one possible prediction could be 1 minute of shopping which does not seem feasible from a practical standpoint. These low predictions occur because there are currently no constraints to enforce minimum time allocation. These low predictions have implications to prediction accuracy because it can result in over-prediction of tours and stops (within tours) without sufficient time to accommodate all predicted activities. To see this, consider a scenario where the MDCEV model predicts one minute in shopping and one hour in social activities. If indicator variables for activity participation in shopping and social activities are used as explanatory variables in tour frequency models, both these variables assume a value equal to 1 in this example, resulting in more tours. In reality, there might not be sufficient time to schedule multiple tours to accommodate the additional social activity predicted by the MDCEV model.

 

In the light of previous discussion, it is clear that are several avenues for improvement in the existing version of the MDCEV model for activity generation and allocation model. The primary objective of the proposed study is to enhance the behavioral and prediction accuracy of the non-mandatory activity generation and allocation model along all these dimensions by developing an improved MDCEV model that

(1)   handles multiple person level budget constraints,

(2)   ensures that the time predictions are always greater than certain minimum duration levels.


Conference registration is required in order to view papers.