You are on page 1of 9

Analysis on Commuter’s Activity Chain Choice Behavior

Mingjun Liu1, Baohua Mao2, Feng Gao3, Jinyi Guo4 and Liping Gao5

1
PhD Candidate, State Key Laboratory of Rail Traffic Control and Safety, Beijing
Jiaotong University, P.R.China. E-mail: alexlius@yahoo.cn
2
Professor, PhD, State Key Laboratory of Rail Traffic Control and Safety, Beijing
Jiaotong University, P.R.China. E-mail:bhmao@china.com
3
Master, 4 PhD Candidate, 5 PhD Candidate, State Key Laboratory of Rail Traffic
Control and Safety, Beijing Jiaotong University, P.R.China

Abstract

Basing on single trip, the traditional travel demand analysis approaches such as
four-stage method is inadequate to depict the activity sequence and difficult to reflect
individual, family factors which influence to travel behavior. Its limiting has already
been subjected to a lot of queries. To a commuter, work is the most important activity,
so according to the commuter travel’s chaining complexion, this paper analyzes
activity chain patterns. By considering personal socioeconomic characteristics,
household socioeconomic characteristics, family member’s interactions and temporal
characteristics, this paper presents an activity-chaining choice model, employing the
discrete choice theory, for the analysis of commuter travel behavior. At last, an
empirical analysis is adopted to specify and estimate commuter travel behavior using
the data from Beijing Household Travel Survey 2005 (BHTS, 2005).

Key Words: activity chain; travel behavior; discrete choice model; travel survey

Introduction

To attend more fixed activities and obtain maximum utility of these activities during
a limited time, travelers incline to arrange these activities as a chained pattern. The

222
1
Copyright ASCE 2008 Traffic and Transportation Studies Congress 2008
Downloaded 13 Feb 2009 to 59.108.199.16. Redistribution subject to ASCE license or copyright; see http://www.ascelibrary.org
Traffic And Transportation Studies 223

traditional transportation demand analysis, for example, four-stage model has long
been practically modeled from a trip-based perspective. Trips, rather than activities,
have served as the analysis units. The conventional modeling approach, which is lack
of depicting the activity sequence and difficult to reflect individual and family factors
influence to travel behavior, has been criticized (Feng et al., 2006). Particularly, the
conventional method, which ignores such a basic principal that travel demand is
derived from the activities that individuals and households need or wish to conduct,
leave how a trip relates to a specific activity, and when, where, with whom, and for
how long the activity last out of consideration. Then, despite some key factors which
influences the individual’s choice have changed, it is unable to reflect how great the
impact is.
Thus, a substantial amount of analysis has been done to study the relationship
between individual interaction, temporal and spatial constraints, household and its
background and activity’s generation. According to utility maximization principle,
Kitamura (1984) suggested that traveler incline to participate in a higher linkage
activity instead in a lower linkage one. After earlier research on activity chain,
Kitamura (1988) considered that the basic theory of trip chain was composed by
activity participation and schedule, household structure and relationship, and travel
decision behavior etc. Goulias et al. (1991) developed a model system through a
recursive model structure representing home-based trip generation and trip chaining
by purpose, however, the model system is unable to be used to predict the sequence
in which trips for different purposes are linked. Bowman and Ben-Akiva (2001) in an
analysis of integrated activity-based discrete choice model system with activity and
travel schedules define 3 prototypes of the activity patterns as the travel from home
to one or more activity locations and back home again. Vadarevu and Stopher (1996)
presented that there are four types factors such as social demographic, individual’s
characters, role allocation, and resource constraints influence the traveler’s activity
schedule. Kumar and Levinson (1995) investigated the allocation of time and
trip-making across time-of-day, and believed the time of day patterns of shop and
other trips for workers and non-workers are both rational. Lu and Pas explored the
relationships among socio-demographics, activity participation and travel behavior
by Structural Equation Modeling (SEM).Chen et al. (2006) simulated the railway
line passenger’s travel choice process by considering various impact factors; Zhang
et al. (2007) analyzed the travel patterns of the elders in Beijing, and just as they said
in the paper, “this is only a first step for Beijing’s travel behavior research.”
In this paper, we preliminarily focus on the Beijing commuters’ activity chain choice
behavior by considering the influencing factors such as individual socioeconomic
characteristics, household socioeconomic characteristics, interaction of members,
and temporal characteristics to build an activity chain choice behavior model. In the

2
Copyright ASCE 2008 Traffic and Transportation Studies Congress 2008
Downloaded 13 Feb 2009 to 59.108.199.16. Redistribution subject to ASCE license or copyright; see http://www.ascelibrary.org
224 Traffic And Transportation Studies

model, the total utility of activity chaining is taken into consideration. At last, an
empirical analysis is adopted to specify and estimate commuter travel behavior using
the data from BHTS 2005.

Activity Chain Pattern

Activity chain represents an individual’s choice of activities in a day’s, which reflects


the traveler’s sequence of activities in temporal or spatial. The demand for travel is
derived from the demand for activities, thus, for the sake of attending more activities
to get a higher utility during a given period, the individuals have to reduce the travel
time while the dwelling time is incompressible (Adler et al., 1979; Liao, 1997). Then,
individuals are apt to link the activity one by one to decrease disutility caused by
travel. For instance, individual A have to attend 3 activities such as work, server
passenger, and shopping, then he will link these activities by a trip chain: leaves
home, goes to a day care center, continues to work, then leaves work, stops at a store,
and goes back home. The activity chain can be shown by Home (H)-Work
(W)-Shopping (S)-Home (H). Just as in Figures 1 and 2, during attending the
activities in the same numbers, the trips will decrease if the activities were chained.
Of course, the activities will be influenced by each other for their chain form, as the
former activity was postponed, the latter one will be delayed, even canceled
(Ben-Akiva et al, 1985). For example, if individual A prolongs his working time for
1 hour, then the scheduled activity (shopping) will be delayed at least 1 hour, even
canceled for his supper with family at home.

Figure 1. No chained activities Figure 2. Chained activities

As commuter’s activity chaining patterns become increasingly complex and involve


interactions with other household and non-household members and as time is a finite
resource, it may be conjectured that trip chains are likely to be increasingly complex
over time. In this paper, the term activity chaining refers to a sequence of activities
that begins at home, involves visits one work and one or more other places, then ends
at home. Depending on the number of home visited within chain, activity chaining

3
Copyright ASCE 2008 Traffic and Transportation Studies Congress 2008
Downloaded 13 Feb 2009 to 59.108.199.16. Redistribution subject to ASCE license or copyright; see http://www.ascelibrary.org
Traffic And Transportation Studies 225

may be classified into two categories, i.e. simple and complex. A chain with double
stops or activities at home is defined as a “simple” chain, whereas a chain with more
than two stops or activities at home is defined as a “complex” one. Thus, an activity
chaining in the form of: home → work → home is considered a simple chain while a
chain in the form of: home → work → home → shop → home is considered a
complex one. According to the sample of Beijing Household Travel Survey 2005
(BHTS 2005), the commuter’s activity chaining share is shown in Table 1.

Table 1. Activity chain patterns of Beijing commuters


Patterns Percent of chains Notes
H-W-H 92.5% Simple tour from home to work and back
H-O-W-H 0.94% Stops between home to work.
H-W-O-H 3.29% Stops between work to home.
H-O-W-O-H 0.41% Stops between home to work and back.
With at least a stop at home, for example
Complex chain 2.86%
H-W-H-O-H
Data source: BHTS 2005.

Where H stands for Home, W is short for Work, O stands for non-work activities
such as shopping, dropping or personal business etc.

Methodology

Supposing U ni to be the utility of activity chain i selected by individual n , J n is


choice set, then i ∈ J n , U ni = Vni + ε ni , and Vni = β ' X ni k . X nik is the k-th factor for
individual n , β ' is vector of estimated coefficient and ε ni is the error term. Thus,
the probability of any alternative i being selected by person n from choice set J n
is given by the following equation:

Pn (i ) = Prob(U n i > U n j , j ∈ J n , i ≠ j )
= P rob(Vni + ε ni > Vnj + ε nj , j ∈ J n , i ≠ j )
= Prob(Vn i + ε ni ≥ max(Vnj + ε nj )) (1)
j∈J n

Let U n* = max(Vnj + ε n j ) , U n* = Vn* + ε n* , and under the assumption the ε n* obey to


j∈ J n

Gumbel distribution with (0, 1), then:


Pn (i ) = Pr ob(Vni + ε n i ≥ Vn* + ε n* )

= Prob((Vn* + ε n* ) − (Vni + ε n i ) ≤ 0) (2)

Assuming that ε ni − ε nj is logistically distributed (Ben-Akiva et al., 1985), the


probability of choosing alternative i is:

4
Copyright ASCE 2008 Traffic and Transportation Studies Congress 2008
Downloaded 13 Feb 2009 to 59.108.199.16. Redistribution subject to ASCE license or copyright; see http://www.ascelibrary.org
226 Traffic And Transportation Studies

β ' X njk
1 eVn i e
Pn (i ) = = = ,i ≠ j (3)
∑ j∈J
* Vnj β ' X n jk
1 + e(Vn −Vn i ) e
n
∑ j∈J n
e

Before we estimate the parameters of the model above, we can define a


selection scheme as a reference. Other selection scheme can be compared with it
and the i − 1 Logit model can be constructed. Here activity chain 1 is assumed
to be the reference, then:
Pn (i )
logit = α1 + β11 x1 + " + β1k xk , i = 2" J n . (4)
Pn (1)

In order to estimate β' , an indicator variable defined


⎧1, if person n chose alternative i
as δ ni = ⎨ , then, we obtain the likelihood function
⎩0, otherwise
N
L = ∏∏ Pn (i )δ n
i
(5)
n =1 i∈J n

where Pn (i ) is a function of β11 ," , β1k . According equation (4), we can rewrite
equation (5) as follows:
N
β ' X nj
L = ∑ ∑ δ ni ( β ' X ni − ln ∑e ) (6)
n =1 i∈J n j∈J n

By differentiating it with respect to each of the β =0, we obtain the necessary


first-order conditions:
β ' X nj
∂L N
= ∑ ∑ δ ni ( xnik −
∑ j∈J n e xnjk
) = 0 , for k = 1," K (7)
∂ β k n =1 i∈J n
' β 'Xnj
∑ j∈J n
e

we can obtain the estimate value of β ' , thus we get the commuter’s activity chain
choice model.
Empirical Study

The primary data source used in this research is drawn from the Beijing Household
Travel Survey 2005 (BHTS, 2005), a representative multiyear travel survey financed
by the Beijing Municipal Government. The survey, covering the 18 districts of
Beijing, includes 82 000 households and 184 000 individuals. A sample from BHTS
2005 is used in the estimation. The data file contains the demographic and
socioeconomic attributes of 1339 person. In addition, records of all activities made
by commuters in one day are included. In the paper, for convenience, we take
home-based one day activity chaining as analysis unit and make a hypothesis that
commuter’s one day’s first activity starts at home and the last one also ends at home.

5
Copyright ASCE 2008 Traffic and Transportation Studies Congress 2008
Downloaded 13 Feb 2009 to 59.108.199.16. Redistribution subject to ASCE license or copyright; see http://www.ascelibrary.org
Traffic And Transportation Studies 227

The explanatory variables of commuter’s choice behavior include in the model, the
selection of which was guided by a review of literature and the survey data, can be
conceptually grouped into four broad categories: individual attributes, household
socio-demographic characteristics, interaction of members, and temporal
characteristics. Individual socio-demographic are a set of personal characteristics
including age, gender, license, etc., while household income per month, car
ownership, motor bike ownership, bike ownership, etc. are classified into household
socio-demographic characteristics. Family interaction factors include household size
and child under 6 in household. The temporal characteristic is depicted by the time
leaving home at the first trip in one day, arriving time of work, off-working time and
the last time home. The values for all variables are self-reported; descriptive statistics
and definition are presented in Table 2.
Table 2. Definitions and descriptive of variables use in model
Variable Definition Mean Std. Dev.
Gender 1 (if male), 0 (otherwise) 0.568 0.495
Age Age of respondent 40.630 10.226
License 1 (if have driving license), 0 (otherwise) 0.524 0.500
Child Number of children under 6 in household 0.078 0.272
Bike Number of bikes owned by household 1.529 1.023
Motor Number of electric bikes owned by household 0.052 0.239
Car Number of cars owned by household 0.512 0.544
Household income per month, 0: ¥0-3500;
Income 0.565 0.564
1: ¥3501-10000; 2: >¥10000
Population People in household 2.764 0.888
T1 First leaving home time in one day 7:57 1.785
T2 Arriving working place time 8:44 1.840
T3 Off-working time 17:10 2.244
T4 Last arriving home time in one day 18:24 1.947

All the models are estimated using a maximum likelihood method as described in
part 3. Table 3 shows parameter estimates and goodness-of-fit statistics for all
models. The goodness-of-fit is measured through the use of ρ 2 , which is defined as
1-L(β )/L(0) . In the paper, the models fit the data very well and produce virtually the
same overall goodness of fit (an adjusted ρ 2 of 0.330 in the case).
According to the estimations in Table 3, we can obtain the H-O-W-H choice model
as follows:

logit(PHOWH / PHWH) =−0.87−0.18× Age +0.27×Gender +0.68×License+0.01×Income


+1.21×Car +0.48×Motor −0.09×Bike−0.01×Population+0.73×Child
−9.01×T1 +6.87×T2 +5.22×T3 −3.59×T4

6
Copyright ASCE 2008 Traffic and Transportation Studies Congress 2008
Downloaded 13 Feb 2009 to 59.108.199.16. Redistribution subject to ASCE license or copyright; see http://www.ascelibrary.org
228 Traffic And Transportation Studies

Table 3. Parameter estimations


H-O-W-H H-W-O-H H-O-W-O-H H-W-H-O-H
Variable
β t β t β t β t
Constant –0.87 –0.53 –1.23 –1.62 1.71 1.01 –3.85 –3.25
Age –0.18 –0.89 0.24 2.44* –0.12 –0.57 0.12 0.74
Gender 0.27 0.97 0.01 0.06 0.17 0.63 0.03 0.11
License 0.68 1.96* 0.47 2.73* 1.76 4.59* –0.30 –1.01
Income 0.01 0.04 0.14 1.07 0.37 1.56 –0.10 –0.44
Car 1.21 4.02* –0.04 –0.23 0.86 2.86* –0.11 –0.41
Motor 0.48 0.98 –0.08 –0.27 0.80 1.92* 0.75 1.88
Bike –0.09 –0.68 –0.08 –1.09 –0.24 –1.80 0.14 1.14
Population 0.35 2.37* 0.13 1.47 0.31 2.08* 0.03 0.22
Child 0.78 1.85 0.18 0.59 1.29 3.27* 0.38 0.82
T1 –3.51 –10.27* 1.10 6.04* –3.48 –10.09* 3.08 7.90*
T2 3.28 10.35* –1.00 –5.29* 3.11 9.81* –2.94 –7.24*
T3 2.97 7.87* –0.44 –4.49* 3.87 9.20* –1.11 –10.58*
T4 –3.08 –7.99* 0.43 4.22* –4.06 –9.51* 1.17 10.57*
N = 1339 L(0) = -1907 L( β ) = -1276 ρ 2 = 0.33 Hit ratio=0.755
β =Estimated Model Coefficient, * significant at α = 0.05 .

Similarly, we can get the choice model of logit(PHWOH / PHWH ) , logit(PHOWOH / PHWH ) , and
logit(PHWHOH / PHWH ) through Table 3. According the models, we can explain on
commuter’s activity chain choice behavior as follows:
Individual socioeconomic characters, for example, different gender travelers have
different linkage habit. Male commuters are more likely to form complex work-trip
activity chains than female, stopping to participate in other activities at multiple
destinations to and from work (the coefficients of the variable “gender” are all plus).
This is a little dissimilar with western country: According to McGuckin and
Murakami (1999),women consistently made slightly more stops between home to
work or work to home. The gap may be due to culture difference, to a Chinese man,
he always dominate in his family and should do more social activities to sustain the
life, however, the western women pursue for feminism, they eager for attending
activities as many as men.
Just as McGuckin and Murakami (1999) presented, men and women stopped for the
same purposes. The most common activities for stops in the commute to work or
work to home are to drop off or pick up a passenger or child, to do family or personal
errands, or to buy a meal. Thus, whether the family has got a child or not is a key
factor to influence the commuters’ activity chain choice behavior. As the parameter
estimations show, the coefficients of the variable “child” are all positive affection and
it reaches 1.29 for the H-O-W-O-H model. It indicates that commuter with children
need to combine work with household and family responsibilities makes it likely that
they will make stops on the way between home and work. In other words, they will

7
Copyright ASCE 2008 Traffic and Transportation Studies Congress 2008
Downloaded 13 Feb 2009 to 59.108.199.16. Redistribution subject to ASCE license or copyright; see http://www.ascelibrary.org
Traffic And Transportation Studies 229

choose the activity chain like H-O-W-O-H.


When commuters make a choice of an entire day’s schedule, they will take the
activity plastic into account first, then they will weight the dwelling time and travel
time for attending the activity. Working is the major activity for a commuter, or
working related activities have a prior claim on his time. The temporal characteristics,
especially per- or post- working, get a positive influence on activity chain choice.
The earlier the commuter departs for work, or the later he reaches work, the more
likely to link non-work activities.

Conclusions

This paper presents a discrete choice model of activity chain that can be specified
and estimated from available diary survey and transportation system level of service
data. The model is designed to capture interactions among a commuter’s decisions
throughout a 24h day by explicitly representing tours and their interrelationships in
an activity pattern. It considers personal socioeconomic characteristics, household
socioeconomic characteristics, family member’s interactions and temporal
characteristics and finds some factor as age, license, car, motor, bike, population,
child and all temporal attributes are significant in the model. A prototype
demonstrates the model concept and statistically verifies the basic structure.
However, an operational implementation would require further empirical tests and
model refinements.

Acknowledgments

This research was funded by the program for New Century Excellent Talents in
University of China (NCET-05-0094), the National Basic Research Program of
China (No.2006CB705500), the National Natural Science Foundation of China
(70571005, 70631001) and BJTU Science Foundation of China (2007RC103). The
authors alone are responsible for any errors.

References

Adler, T., and Ben-Akiva, M. (1979). “A theoretical and empirical model of trip
chaining behavior.” Transportation Research B, 243–257.

Ben-Akiva, M, and Bowman, J. L. (2001). “Activity-based Disaggregate Travel


Demand Model System with Activity Schedules.” Transportation Research A, 35(1),
l–28.

Ben-Akiva, M., and Lerman, S. R. (1985). “Discrete Choice Analysis–Theory and


Application to Travel Demand.” The MIT Press, Cambridge, MA, 70–71.

8
Copyright ASCE 2008 Traffic and Transportation Studies Congress 2008
Downloaded 13 Feb 2009 to 59.108.199.16. Redistribution subject to ASCE license or copyright; see http://www.ascelibrary.org
230 Traffic And Transportation Studies

Chen, T. S., Mao, B. H., Gao, L. P., and Yue, F. (2007). “Research about Passenger
Travel Choice Behavior of Dedicated Passenger Railway Line.” Journal of the
Railway Society, 8–12.

Feng, X. S., Zhang, J. Y., Fujiwara, A., Senbil, M., and Lee Backjin. (2006). “A
Travel Demand Prediction Model with Feedback for Jabodetabek Metropolitan Area,
Indonesia.” In: Mao Baohua, Tian Zongzhong, Gao Ziyou, and Huang Haijun. Traffic
and Transportation Studies Proceeding of ICTTS 2006, 535–546.

Goulias, K., Pendyala, R. M., and Kitamura, R. “Practical method for the estimation
of trip generation and trip chaining.” Transportation Research Record,1285, 7–56.

Kitamura, R. (1988). “A sequential history dependent approach to trip-chaining


behavior.” Transportation Research Record, 944, 2–38.

Kitamura, R. (1984). “Incorporating trip chaining into analysis of destination


choice.” Transportation Research B, 18B(1), 67–81.

Kumar, A., and Levinson, D. “Temporal variations on allocation of time.”


Transportation Research Record 1493, 118–127.

Liao, Y. C. (1997). “Trip chaining in Urban Travel. Ph.D Dissertation. Universtiy of


Southern California.” Los Angeles, CA. Published by UMI.

Lu, X., and Pas, E. T. “Socio-demographics, activity participation and travel


behavior.” Transportation Research A., 33(1), 1–18.

McGuckin Nancy, Murkami Elaine. (1999). “Examining Trip-Chaining Behavior:


Comparison of Travel by Men and Women.” Paper presented at 78th Annual
Meeting of the Transportation Research Board, Washington, DC.

Vadarevu, R. V., and Stopher, P. R. “Household activity, life cycle and role
allocation.” Transportation Research Record 1556, 15–26.

Zhang, Z., Mao, B. H., Liu, M. J., Chen, J. C., and Guo, J. F. (2007). “An analysis of
travel patterns of the elders in Beijing.” Journal of Transportation Systems
Engineering and Information Technology, 7(6), 11–20.

9
Copyright ASCE 2008 Traffic and Transportation Studies Congress 2008
Downloaded 13 Feb 2009 to 59.108.199.16. Redistribution subject to ASCE license or copyright; see http://www.ascelibrary.org

You might also like