International Choice Modelling Conference, International Choice Modelling Conference 2017

Font Size: 
Analysis of Daily Activity-Travel Pattern Focusing on Inter- and Intra-Personal Heterogeneity
Lei GONG, Toshiyuki Yamamoto, Ryo Kanamori

Last modified: 28 March 2017


Daily activity-travel patterns (activity and travel-related behavior such as the number of trips, the number of trip chains, daily travel distance, daily travel duration, time of first trip starts, and time of last trip ends) vary day-to-day and person-to-person. Traditional personal trip (PT) survey randomly selects one day to collect data as a representative of people’s activity-travel behavior, based on which travel demand modeling is proceeded. However, single day data could lead to biased estimates and a bloated error term (1). Existing research results have already shown that multi-day PT data which could reflect behavioral variations are necessary in the travel demand modeling. Several decades see the discussion of needs for multi-day data collection and using multi-day data has both advantages in travel behavior analysis and statistical benefits (2).


Variability of one or several aspects of activity-travel pattern is studied in the literature. Beltman used Mode Variation Index to indicate intrapersonal variability in mode choice behavior (3). He finds that trip distance, trip purpose, rain during the trips and gender show significant effects. Dharmowijoyo et al. selected five characteristics of activity-travel pattern (including number of trips, share of motorized modes, number of trip chains, daily travel time, and departure time) and analyzed their variability in a separate way using a hierarchical Structural Equation Model (SEM) with four consecutive days of activity diary data collected in Jakarta, Indonesia (4). They found that workers and students fit the model well; household and individual characteristics are significant variables shaping day-to-day interaction.


On one side, most existing research only select one or several aspects of activity travel pattern in a separate way which may not be able to completely reflect the activity-travel pattern as a whole. Results by Dharmowijoyo et al. show the interaction between different aspects of activity-travel patterns, such as more trips lead to more motorized trips which further lead to more trip chains (4). This result indicates that it is a better way to deal with different aspects of activity-travel pattern as a whole rather than separately. On the other side, most existing research use a data set collected less than a week which might conflict with findings by Beltman who found two-week data show a proper quantity to capture intrapersonal variability in mode choice (3). It indicates the necessity of collecting data in a longer period with the intention of capturing variability of activity-travel pattern.


Consequently, the objectives of this paper are to consider different aspects of activity-travel pattern as a whole by clustering activity-travel patterns into different types, then to analyze changes among patterns involving intra- and inter-person heterogeneity.


Several studies have achievements on activity-travel pattern clustering. However, they either use one-day pattern representing a person’s pattern which neglect intra-person heterogeneity or use limited features which may not fully demonstrate characteristics of the pattern. Přibyl developed a modified version of k-means algorithm, called K-medoids algorithm to cluster activity patterns using a probabilistic search algorithm that simulates natural evolution; It intends to separate individuals into groups with similar activity patterns (5). Activity type at out-of-home stop and time of the day are used as primary and secondary attribute respectively to describe and classify activity-travel patterns (6; 7).


Several studies also found inter- and intra- person heterogeneity in activity-travel patterns analysis.  However, they either focus on recurrence structures of daily patterns or randomly use the data of one or two days which may not fully demonstrate respondents’ pattern variability. Alexander used Markov chain models to analyze the recurrence structure of daily travel pattern (2). Lee et al. analyzed the relationship between travel behavior change and household composition variables, land use variables etc. during a much longer period from 1989 to 2002 using two-day diary data in each selected year (8).


This paper focuses on clustering activity-travel patterns and which factors influence people’s pattern decision-making with consequences whether a daily pattern changes from one type to another. Using a data set collected by mobile phone in Japan (done by 16 respondents for eight months in Hakodate city), activity-travel pattern alteration across 10 month is analyzed and compared at inter- and intra- person levels. At first, descriptive analysis on change in aspects of activity-travel pattern is presented. Following descriptive analysis, a latent class cluster analysis (9) of daily activity-travel patterns is conducted. Using latent class cluster analysis methods, daily activity-travel patterns are clustered into groups with inner similarity of their primary features. Then transition/change of activity-travel pattern over days is analyzed by longitudinal Mixed Markov Latent Class models (10). The time-constant latent variables are used to deal with unobserved heterogeneity in the change process, whereas the time-varying discrete latent variables are used to correct for measurement error in the observed response. Characteristics grouped into individual, activity, trip and environment are used to check whether/how they influence the respondents to make a choice of daily activity-travel pattern. Environment-related variables include weather data and respondents’ residence which change over time or person.



1          Deutsch-Burgner, K. E. Inter-person or intra-person variation: the value of multi-day data in modeling choices Presented at International Choice Modelling Conference 2015, Austin, Texas, USA, 2015.

2          Alexander, B. On variability and heterogeneity of day-to-day travel (Doctor thesis). Kyoto University, 2007.

3          Beltman, J. Intrapersonal variability in mode choice behavior: a research based on data from the Dutch mobile mobility panel (Master thesis). University of Twente, Enschede, 2014.

4          Dharmowijoyo, D. B., Y. O. Susilo, and A. Karlström. Day-to-day variability in travellers’ activity-travel patterns in the Jakarta metropolitan area. Transportation, 2015, pp. 1-21.

5          Přibyl, O. Clustering of activity patterns using genetic algorithms.In Soft Computing: Methodologies and Applications, Springer, 2005. pp. 37-52.

6          Pas, E. I. A flexible and integrated methodology for analytical classification of daily travel-activity behavior. Transportation science, Vol. 17, No. 4, 1983, pp. 405-429.

7          Mo, H., and M. Yun. Classifying and Modeling Activity-Travel Pattern.In Logistics: The Emerging Frontiers of Transportation and Development in China, ASCE, 2008. pp. 3354-3359.

8          Lee, J. H., A. W. Davis, and K. G. Goulias. Triggers of behavioral change. Presented at International Choice Modelling Conference 2015, Austin, Texas, USA, 2015.

9          Vermunt, J. K., and J. Magidson. Latent class cluster analysis. Applied latent class analysis, Vol. 11, 2002, pp. 89-106.

10        Vermunt, J. K., B. Tran, and J. Magidson. Latent class models in longitudinal research. Handbook of longitudinal research: Design, measurement, and analysis, 2008, pp. 373-385.

Conference registration is required in order to view papers.