International Choice Modelling Conference, International Choice Modelling Conference 2015

Font Size: 
Inter-person or intra-person variation: the value of multi-day data in modeling travel behavior decisions
Kathleen Elizabeth Deutsch-Burgner

Last modified: 18 May 2015


Inter-person or intra-person variation: the value of multi-day data in modeling choices 

The importance of understanding variability in choices related to daily behavior is of the utmost importance in travel behavior modeling.  Traditionally, this variation has mostly been assessed using travel diaries encompassing one day.  However, conclusions reached from analyzing data from a one-day observation period could be incorrectly attributing variation seen in the sample to inter-person variation (across people) rather than to possible intra-person variation (same person behaving differently across days) due to data limits.  This leads to biased estimates and a bloated error term.  Although the need for multi-day and multi-period analysis has been discussed within the travel behavior analysis and travel demand modeling community for over forty years (see for example Shapcott, 1978; Hanson and Hanson, 1981; Pas, 1988; Hirsch, 1986; Kitamura, 1987; or Pendyala and Pas, 2000 for an overview), the research is limited.  This is primarily due to the limited datasets that include observations across multiple days or multiple periods.  However, with the improvement in technology over the last thirty years, and survey methodologies presently available, the topic of multi-day data and increased understanding of decision making by analyzing observed data across multiple days can be again discussed, this time in new light.  The increased ubiquity of technological devices and the relatively low cost have enabled large-scale data collection efforts to occur with longer durations of data collection.

The research presented in this paper examines the existence of intra-person variation in travel behavior choices, and the importance of this type of data and modeling in travel behavior.  Using a sample of Bay Area residents who participated in the 2012 California Household Travel Survey, an individual’s behavior across three days is examined.  Variations in observed day-to-day travel behavior is explored using data from respondents who carried a personal, wearable GPS tracker for three consecutive days.  In this paper, variation in behavior will be examined in several ways.  First, an exploratory analysis will be presented to examine the change in aspects of travel and activity decision-making by comparing the observations from each day with the other days.  Findings show, for example, that while a majority of the sample has a change in trips of less than 20 trips from one day to the next, there is still quite a large variety in this change.  For instance, 65% of the respondents in the sample have a change of between 2 and 11 trips for each day-to-day comparison.  In addition, only 15% of the sample has no change in trips from one day to another for at least one permutation of the three days of the sample, and only 30% have a change of one trip or fewer for any day-to-day comparison.  Similar statistics for the sample for the change in total distance, change in average distance per trip (to reflect length of trips), and change in standard deviation of the trip lengths (to reflect the distributions of trip length for each respondent) are also examined.    

Following this descriptive analysis, a latent class cluster analysis of the sample is preformed.  For each respondent, the day-to-day change across the sample days is used.  With this latent class cluster analysis, respondents are broken into groups or clusters based on similarity in the behaviors that they exhibit.  Six distinct clusters of people manifest through this analysis, identifying different “variability types” in travel decisions and thus behaviors.  These clusters range in attributes of change between days in total distance traveled, change from day to day in average distance per trip, change in standard deviation in trip distance, and change in total number of trips.  Clear groups of individuals emerge, for instance, some findings include: clusters comprised of those who have very little change in trip attributes from day to day, those who have drastic changes from day to day due to long distance travel, those who have medium sized changes in travel from day to day due to changes in trip length or frequency, and those who have medium changes in day to day behavior due to the presence of days with no trips.  Covariates are used to explain cluster membership, and clear patterns are revealed in the variation and thus cluster membership.  These clusters also provide further meaning to inter-person variability, through a cluster-by-cluster comparison of travel and destinations choices compiled for all three days rather than the change in travel attributes across days.

Throughout this analysis, it becomes very clear that current models of behavior and observed choice using one-day travel diaries are limited in their ability to depict behavior accurately.  The notion of a “typical day” is founded more on convenience than on theory.  The work of this paper illustrates the need for multi-day data, and the possibilities for model improvement.  A latent class cluster model indicates that individuals can be categorized into “variability types,” that highlight intra-person variability.


Acknowledgements: This research is conducted using the Transportation Secure Data Center housed in the National Renewable Energy Laboratory.  Funding for this project is provided by the Federal Highway Administration.



Pendyala, R.M. and E.I. Pas (2000) Multiday and Multiperiod Data for Travel Demand Modeling. Invited Resource Paper in Transport Surveys: Raising the Standard, Proceedings of an International Conference on Transport Survey Quality and Innovation. Transportation Research Board E-Circular Number E-C008, Transportation Research Board, National Research Council, pp. II-B/1 II-B/22

Hanson S., and P. Hanson (1981).“The Travel-Activity Patterns of Urban Residents: Dimensions and Relationships to Sociodemographic Characteristics. Economic Geography, 57,332-47

Hirsh, M., Prashker, J. and M. Ben-Akiva (1986) Dynamic model of weekly activity pattern. Transporation Science 20, 24-3

Kitamura, R.( 1987) An analysis of weekly activity patterns and travel expenditure, in Golledge, R. and H. Timmermans (Eds.) Behavioral modeling approaches in geography and planning. 

Pas, E. (1988) Weekly travel-activity behavior. Transportation 15, 89-109.

Shapcott, M. (1978) Comparison of the use of time in Reading, England with time use in other countries. Transactions of the Martin Centre for Architectural and Urban Studies 3:231-257



Conference registration is required in order to view papers.