International Choice Modelling Conference, International Choice Modelling Conference 2015

Font Size: 
A Simultaneous Model of Intensity of Activity Opportunities on Supply Side and Destination & Departure Time-of-Day Choices on Demand Side
Rajesh Paleti, Naveen Eluru, Hsi-Hwa Hu, Guoxiong Huang

Last modified: 18 May 2015


Travel demand models (TDMs) use zonal land use and employment data as one of the inputs to predict the activity-travel patterns of residential population in the study region. These models implicitly assume that zonal land use and employment serve as indicators of opportunities that attract travelers to pursue different types of activities. The characterization of these decision processes in activity based models varies substantially.  Most Activity-based models (ABMs) have two main components. The first step is activity generation in which the model predicts the activities that each individual in the study region plans to undertake during the day. The second step is activity scheduling in which the spatial and temporal choices of the planned activities in the activity generation step are determined.  In ABMs, the supply side land use and employment information is predominantly used in three ways 1) as one of the inputs to zonal accessibility measures that affect daily planning choices in the activity generation step, 2) as one of the inputs to zonal accessibility measures that affect scheduling choices that are modeled prior to destination choices in the activity scheduling step, and 3) as zonal attraction size variables (which are usually linear combinations of different zonal employment variables) in destination choice models in the activity scheduling step.

            One common assumption across all these ABMs is that the supply side opportunities (i.e., zonal employment) do not vary across different hours of the day i.e., a constant zonal employment profile is assumed. However, in reality, a more reasonable hypothesis would be a bell-shaped temporal profile of zonal employment consistent with the expected opening and closing hours of most businesses (Paleti et al., 2014). This bell shaped temporal profile of opportunities on the supply side has significant implications to the way destination and departure time-of-day (TOD) choices are modeled. First, destination and departure TOD choices must be bundled together and viewed as simultaneous choices because zonal attractiveness (typically measured using size variables in destination choice models) changes significantly depending on the departure time. People are likely to compare and evaluate combinations of departure times and destinations instead of making these decisions in any pre-determined sequence. Second, the temporal profile of opportunities on supply side is not necessarily a collective decision of business establishments in the zone independent of the travel preferences of people in that region. For instance, businesses are open late night in Manhattan because they see people interested in pursuing activities during late hours. Similarly, people go shopping late night in Manhattan because they know shops are open for longer hours. The observed temporal profile of activity opportunities and the observed destination and departure TOD choices are most likely the outcomes of equilibrium between the supply and demand factors. So, these two systems (i.e., supply and demand) cannot be analyzed as two separate independent choices. A simultaneous model that captures the dynamic interactions between activity opportunities on supply side and destination/TOD choices on demand side is better suited for this choice context.

From a policy perspective, models that treat destination and departure time choices as sequential decisions can provide wrong policy implications. For instance, if a new business development that is open late hours comes up in the region, such models do not predict any changes in the departure time patterns but they predict many more people choosing the destination with the new business development during all hours of the day. So, the predicted origin destination flows by TOD and the implied vehicle-miles-travelled (VMT) estimates would be wrong. Also, models that do not account for simultaneity between the supply and demand side decisions can lead to inflated estimates of the zonal employment variables on the destination and TOD choices of travelers. From a methodological perspective, the proposed model in our study takes the form of a simultaneous choice model with two components – 1) a continuous component that models the intensity of activity opportunities (i.e., zonal employment density) during each time period, and 2) a discrete component that analyzes the destination and departure TOD choices as a combination alternative. The simultaneity between these two model components is accommodated using time-period and zone specific random error terms that enter both the continuous and discrete choice components. For the temporal dimension, we plan to use five time periods: morning peak (6:00 am to 9:00 am), mid-day (9:00 am to 3:00 pm), evening peak (3:00 pm to 7:00 pm), evening (7:00 pm to 9:00 pm), and night (9:00 pm to 6:00 am). For the location dimension, we chose to do the analysis at a spatial resolution of Traffic Analysis Zone (TAZ). Given that considering all TAZs can lead to explosion in the number of alternatives in the choice set (because of the combination of temporal and location dimensions), we will explore zonal sampling mechanisms that ensure consistent parameter estimates in mixed logit models (Guevara and Ben-Akiva, 2013). The resulting model will be estimated using Maximum Simulated Likelihood inference approach using quasi-Monte Carlo Halton sequences. The study will contribute to the existing literature on integrated modeling of continuous and discrete choice outcomes in empirical contexts with unwieldy choice sets by building upon the recent advances in alternative sampling (Guevara and Ben-Akiva, 2013). The proposed model will be used to analyze destination choice behavior of residents in the Southern California region. The two main data sources for estimating the model are the Southern California Association of Government (SCAG)’s 2010 household travel survey data (for demand side information) and the InfoUSA business establishment data (for supply side information).



Guevara, C.A. and M.E. Ben-Akiva (2013) Sampling of alternatives in Logit Mixture models, Transportation Research Part B, 58, 185–198.

Paleti, R., Vovsha, P., R. Picado, B. Aleksandr, H. Hu, and G. Huang (2014) Development of Time Varying Accessibility Measures: Application to the Activity-Based Model for Southern California Region, Technical Paper, Old Dominion University.


Conference registration is required in order to view papers.