Content uploaded by Juan Antonio Carrasco
Author content
All content in this area was uploaded by Juan Antonio Carrasco on May 06, 2014
Content may be subject to copyright.
1
Exploring the propensity to perform social activities: A social network approach
JUAN ANTONIO CARRASCO* and ERIC J. MILLER
Department of Civil Engineering, University of Toronto, Toronto, Canada M5S1A4
(*Author for correspondence, E-mail: j.carrasco@utoronto.ca)
Transportation (2006), in press
Key words: social networks, activity-travel behavior, social activities, information and
communication technologies, structural equation modeling.
Abstract. Conceptual and empirical models of the propensity to perform social activity-travel
behaviour are described, which incorporate the influence of individuals’ social context, namely
their social networks. More explicitly, the conceptual model develops the concepts of egocentric
social networks, social activities, and social episodes, and defines the three sets of aspects that
influence the propensity to perform social activities: individuals’ personal attributes, social
network composition, and information and communication technology interaction with social
network members. Using the structural equation modelling technique and data recently
collected in Toronto, the empirical model tests the effect of these three aspects on the
propensity to perform social activities. Results suggest that the social networks framework
provides useful insights into the role of physical space, social activity types, communication and
information technology use, and the importance of “with whom” the activity was performed with.
Overall, explicitly incorporating social networks into the activity-travel behaviour modelling
framework provides a promising framework to understand social activities and key aspects of
the underlying behavioral process.
2
1. Introduction
1.1. Overview and motivation
Although metaphors from physics and other natural sciences have been useful in the past
(Harvey 1969), they are now not enough to understand the rich complexity of travel behavior
(Pas 1990). In this context, activity-based approaches incorporate more truly behavioral
explanations which recognize travel as a derived demand, triggered by the desire to perform
activities with others (Pas 1990). While this recognition has existed for some time, the need to
complement the dominant econometric-based approach is still an important research challenge.
More specifically, models that explain the generation of trips (“why” travel is performed) still
heavily rely on the individual socioeconomic characteristics of travelers, without considering the
importance of the individual’s social context in this process. A potential approach to better
understand the generation of individual activities and travel in general, and social activities and
travel specifically, is looking at the propensity to perform them, especially those elements less
measurable in terms of costs and socioeconomics, as recognized long ago by Chapin (1974). In
this context, a key hypothesis is that individuals’ social network characteristics are relevant for
their propensity to perform social activities and that these effects can be appropriately measured
and used to understand the underlying decision making processes.
The study of social networks in activity-travel behavior responds to “the need to underpin
our travel models with a better understanding of the social structures of daily life and, as we
implicitly forecast/speculate about them when we predict travel behavior over long time
horizons, anyway…” as Axhausen (2002: p.3), argues. This requirement is even more patent
when a series of “possible transport questions” are considered, such as “physical spatial-
temporal coherence / overlap (constraints), replacement of physical and telecommunication-
based contact, interaction frequency and spatial reach, and interaction and information /
knowledge transfer” (2002: p.10)
3
In addition, the focus in social activities is particularly interesting since interactions
intuitively play a “motivator” role in the behavioral processes that lead to the generation of those
activities. The study of these activities has been a neglected area in travel behavior research,
although some attempts have been undertaken recently (Mokhtarian et al 2003; Schlich et al
2004). In addition, social networks can potentially help capture the propensity to perform social
activities in a new context, such as the role of information and communication technologies
(ICT) in activity-travel generation. This link between social activity-travel and ICT has been
discussed (Mokhtarian et al 2003; Senbil and Kitamura 2003), but with no explicit inclusion of
social networks characteristics.
Although the interest in social interactions in the activity-travel urban context has a long
tradition (e.g., Stutz 1973, Kemper 1980), recent literature in this area is scarce. Some
exceptions are theoretical discussions about long-term effects of social networks and travel
(Axhausen 2006), and insights about social influence and travel (Dugundji and Walker 2005;
Páez and Scott 2006). However, no dedicated data collection effort, and very few empirical
analyses have been undertaken recently. In response to this need, the objective of this paper is
to present a conceptual and an empirical statistical model to study the propensity to perform
social activities, explicitly incorporating social networks concepts. The main underlying
hypothesis is that studying social networks provides new insights to understand the social
activity generation process. More explicitly, it is expected that this analysis incorporating the
social network perspective will enrich the behavioral components of operational agent-based
activity-travel demand models, such as TASHA (Miller and Roorda 2003) and integrated land-
use models, such as ILUTE (Salvini and Miller 2005). The rest of this section further elaborates
the social networks concept in activity-travel behavior; section two presents the conceptual
framework used in the empirical models; section three discuss the main results of these models;
and section four summarizes some conclusions and prospective future work.
1.2. Social networks and activity-travel behavior
4
“Social network analysis is the study of social structure and its effects. It conceives social
structure as a social network, that is, a set of actors (nodes) and a set of relationships
connecting pairs of these actors” (Tindall and Wellman 2001: 265-266). Two key components
define this paradigm: actors, who represent different entities, such as groups, organizations,
nations, as well as persons; and relationships, which represent flows of resources that can be
related with aspects such as control, dependence, cooperation, information interchange, and
competition.
The core concern of the social network paradigm is “to understand how social structures
facilitate and constrain opportunities, behaviors, and cognitions”. Social network analysis
conceives overall behavior as more than the sum of individual behaviors, and contrasts with
“explanations that treat individuals as independent units of analysis” (as traditionally used in
travel behavior research). Thus, behavior is explained not only through personal attributes but
also by using social structure attributes that incorporate the interaction among the different
social network members. In this vision the whole is more than the sum of its parts; that is, social
phenomena cannot be understood solely by individual characteristics (such as socio-economic
attributes), but also by the social structure attributes that emerge from the interaction between
those individuals.
A key link with travel behavior is that ties among people may be interpreted not only as
mere interactions but also as links that indirectly represent potential activity and travel where
these actors are involved. Analysis and modeling these ties not only requires understanding
these interactions, but also what are the potential activities and trips involved in them. As a
consequence, the structural characteristics – and the underlying individual or actor attributes –
can be potentially sources of explanation of activity and travel, as the following conceptual
framework presents.
2. Conceptual framework
5
The purpose of this section is to sketch the conceptual framework which serves as a
background for the operational definition of the propensity to perform social activities, and the
empirical analysis developed later.
2.1. Social networks
Egocentric approach. The general definitions about social networks outlined in the previous
section need to be further operationalized in order to collect data and conduct empirical
analyses into the phenomenon. Two kinds of studies can be done with social networks: whole or
egocentric networks (Wellman 1988). Whole network studies assume that the entire set of
actors and their relationships is known, forcing the analyst to know or at least make
assumptions about all the individuals relevant to the phenomenon of interest. On the other
hand, egocentric network studies concentrate on one specific individual and those who are
related with him/her. Concretely, since generally the interest in travel behavior research is about
large populations in urban areas, the egocentric network approach constitutes the only feasible
way to study explicit interactions. Egocentric networks thus become “samples” of the entire
urban social network. The social network definition below uses this framework.
Social network definition. Each individual (called ego) has a social network, defined as a set of
actors or alters who have relationships or ties with the ego, and who may or may not have ties
with each other.
Network composition. A key characteristic of social networks is their composition, that is, which
alters constitute the network and what are their characteristics. As previously discussed, this is
an important aspect since it can be hypothesized that the network composition constitutes a
potential source of explanation for the propensity to perform social activities. In this paper, the
influence of the roles of the alters, their distance, and their gender homophily (having the same
gender) with respect to the ego are analyzed.
Tie characteristics. Each tie may have several characteristics that define the relationship
between the ego and each alter. In this paper, two tie attributes are explored. The first is tie
6
strength, defined as the degree of closeness between the ego and alter. Ego-alter ties can be
“strong” or “weak” depending on how emotionally close the ego feels to the alter. Strong tie is
operationalized as “people you discuss important matters with, or regularly keep in touch with,
or there for you if you need help”, and weak tie is operationalized as “more than just casual
acquaintances, but not very close people”. These definitions also define the social network’s
boundary, explicitly excluding acquaintances. The second tie characteristic is the frequency
and media of interaction, which measures the intensity and type of ego-alter interaction.
2.2. Social interaction and episodes
Social interactions. A social interaction can be generally defined as an activity or a set or
activities performed by two or more individuals primarily for recreational or support purposes,
that can be performed face-to-face or virtually (telephone or the Internet in the latter case).
Social interactions are conceptualized as “projects”, in the way Miller (2005) uses the concept;
i.e., as a “coherent, logically interconnected sets of actions” (p.22). Social interactions comprise
a primary project because they represent a major generic activity type within the personal and
household agenda (Miller 2005).
Social episodes. A social project generates a series of activity and travel episodes. Three types
of episodes can be differentiated: travel, provision, and social episodes. Travel episodes are
trips that start and end in a provision, social or another travel episode; provision episodes are
shopping or another secondary activity necessary to perform the social activity; social
episodes are the core of the social interaction project, and are the focus of this paper. Social
episodes are undertaken using different kinds of media: face-to-face, telephone (cellular and
regular), and Internet (email and instant message). At the same type, each episode has
duration, start time, and location. Social episodes’ locations can be concurrent (i.e. the same
place for all the members interacting) or non-concurrent (i.e. different places). Furthermore, if
the location is concurrent, for the purposes of this paper two kinds of places are differentiated:
7
the ego’s or alter’s home (hosting and visiting social activities), or institutional or public places
other than homes (e.g. social activities at pubs or restaurants).
2.3. The decision to perform a social project
The decision to perform a social project can be characterized by the individual’s propensity and
opportunity to engage in a social project, inspired by Chapin’s general activity patterns model
(Chapin 1974). For the specific purposes of this paper, personal and network attributes are
explored to measure this propensity.
Propensity to engage in a social project. The propensity to engage in social projects, and more
specifically, to engage in social face-to-face episodes, potentially depends on:
- Personal attributes, such as age, gender, income, lifecycle, personality, and household
characteristics,
- Social network attributes, specifically the ego’s network composition, and
- Social episodes performed by the ego with other media (telephone and Internet) by strength
of the tie (strong / weak) and frequency of interaction.
Finally, the propensity to perform social projects is postulated as a “latent” attribute, not
directly observable from individuals’ activity patterns, that is measured in this paper as the
intensity of face-to-face social episodes by tie strength (strong / weak), social activity type
(hosting and visiting / bar and restaurants), and frequency of interaction.
Opportunities to engage in a social project. These mainly refer to the individual’s time and space
constraints and opportunities (Hägerstrand 1970; Chapin 1974). Although not explicitly
considered in this paper, these latter aspects have a strong social network component,
especially considering Hägerstrand’s coupling constraints, and the fact that the locations of ego
and alters are in general fixed in the short and medium-run (e.g. homes for visiting social
activities).
3. Empirical models
8
3.1. Data: The Connected Lives Study
A main challenge to effectively incorporate social networks in an activity-travel framework is the
collection of adequate data that accounts for the interactions among individuals. The data used
to calibrate the empirical models in this section corresponds to the Connected Lives Study, a
broader study about people’s communication patterns, conducted by the NetLab group at the
Centre for Urban and Community Studies, at the University of Toronto. The study was occurred
between May 2004 and April 2005, and consisted of 350 surveys of people randomly selected in
the East York area of Toronto, with more detailed follow-up interviews and observations of a
subsample of 87 from the original sample. The East York area is located east of downtown
Toronto, and is fairly representative of the overall inner city characteristics regarding
sociodemographics and general transportation level of service. For more details about the data
collection process, the reader is referred to Carrasco et al (2006) and Wellman et al (2006).
The data used in this section corresponds to the initial survey part of the study. The
method used to gather the characteristics of the respondent’s social networks is known as the
summation method (see McCarty et al 2000 for details), and consists of eliciting the number of
alters who have specific characteristics, such as role, gender, distance, and frequency and
media of interaction. In addition, standard questions about personal and household
characteristics were gathered.
Table 1 presents the list of variables considered in the analysis, conforming to the
previous conceptual framework. The dependent variables – which serve as an indicator of the
propensity to perform social activities – correspond to the number of people in the social
network by tie strength (weak / strong) with whom the ego usually performs social activities
(hosting or visiting / going to pub or restaurants) at a certain frequency (less than once a week /
between a week and a month). These three dimensions (tie strength, social activity type, and
frequency) imply eight dependent variables according to each tie strength / frequency
9
combination for each social activity type. Regarding the independent variables, three sets are
analysed:
- Personal and household characteristics, including socioeconomic and lifecycle attributes.
- Network composition attributes, including roles of each alter, their distance, and their gender
homophily (having the same gender) with respect to the ego. Role composition includes how
many people of each relationship compose the network (close family, other relatives,
friends, co-workers or classmates, and people from organizations). Distance of the
networks’ members with respect to the ego (number of people living in Canada at more than
an hour’s travel away, and number of people living outside Canada). Gender homophily is
defined as the number of people from the network who have the same gender as the ego.
- Interaction through information and communication technology use, i.e., how many people
the ego usually communicates with using each medium (cell phone / regular phone / email /
instant message), by tie strength (strong / weak) and by frequency (at least once a week /
between once a week and once a month). These three dimensions imply sixteen different
independent variables according to each tie strength / frequency combination by media.
Finally, since the number of members in a network can be very high for a few cases (a
“long-tailed” distribution), the models in this paper have censored the network variables in the
tenth higher percentile; this technical constraint is explicitly considered in the models and does
not add bias to the results.
(TABLE 1 ABOUT HERE)
3.2. Method: structural equation models
The statistical method used in this paper is structural equation modeling (SEM), which consists
of a series of linear equations that relate observed exogenous and endogenous variables, and
latent variables. This method has been extensively used in the social sciences for decades, and
is increasingly a standard tool in travel behavior research (for a more in depth review of SEM
10
and applications in the area see Golob 2003 and the references therein). The SEM used here
consists of two equations:
Structural equation:
ς
η
η
+
Γ
+
=
XB (1)
Measurement equation:
ε
η
+
Λ
=
Y (2)
where
η
is the vector of latent variables,
X
is the vector of observed independent variables,
Y
is the vector of observed dependent variables,
ς
is the vector of unobserved dependent
variables affecting the latent variables,
ε
is the vector of measurement errors; and
B
,
Γ
and
Λ
are the coefficient matrices that reflect the causal relationships among the variables. The effect
of the independent variables
X
on the latent variables
η
can be direct (measured by
Γ
) and
also indirect (measured by
B
); thus, the total effect of
X
on
η
corresponds to the sum of both
effects, measured in the reduced form equations. The measurement relationship between
observed and unobserved independent variables is represented by
Λ
. It is assumed that there
are no measurement relationships and errors at the level of the endogenous variables.
The SEM calibrated in this paper (see the path diagrams in Fig. 1) comply with the
conceptual framework presented before: three sets of independent variables (personal and
household attributes, ego’s social network composition, and social network ICT use) influence
the propensity to perform social activities by each type (hosting / visiting and going to pubs /
restaurants) and strength of tie (weak / strong). These propensities are latent variables,
measured by observed dependent variables, defined as the number of people with whom the
ego socializes by tie strength, frequency, and activity type. In addition, the structure presented
in the path diagrams allows models representing the influence of the propensity to perform
social activities between both activity types, and both tie strengths for the given set of
independent variables, capturing the indirect overall network effect. For example, although the
number of strong tie friends directly affects the number of strong tie people hosting and visiting,
the path structure allows models to explore the indirect effect of this variable on the number of
11
weak tie people hosting / visiting. Three models were estimated (one for each independent
variable set) rather than a single one due to small sample sizes, which did not allow statistically
reliable results for combined models to be obtained.
(FIGURE 1 ABOUT HERE)
3.3. Empirical models
The results from the three SEM are presented in this section. Tables 2, 3 and 4 show the
structural equations coefficients (representing the direct effects), and the reduced form
equations (representing the total effects, direct and indirect); all models were calibrated using
the statistical package LISREL (Jöreskog and Sörbom 2001). Blank spaces in the tables
indicate coefficients with a t-statistic lower than 1.20 (p value lower than 0.885), and grey
spaces indicate coefficients not considered in the conceptual model sketched in Fig. 1. In
general, the goodness of fit of the three models was adequate, according to standard criteria
used in the literature, such as a ratio between 2
χ
and degrees of freedom lower than 3, a Root
Mean Square Approximation (RMSA) confidence interval which includes 0.05, and a
Comparative Fit Index (CFI) greater or equal than 0.95 (for more details, see Washington et al
2003 and the references therein).
(TABLES 2, 3 AND 4 ABOUT HERE)
The first model tested whether personal and household characteristics are factors that
influence the propensity to perform social activities (Table 2). In addition, these characteristics
can be understood as “systematic effects” with respect to the social networks’ influence in this
propensity. A first interesting result is the significant positive coefficient of income for all four
models, that is, a positive relationship between higher income and more people socializing. In
the case of bar and restaurants, this positive relationship is consistent with other findings in the
transportation literature (Lu and Pas 1999; Schlich et al 2004). The case of hosting and visiting
is less clear, since the dependent variable mixes in-home and out-of-home activities, and low
12
income groups seem to have higher propensity to perform in-home social activities, and high
income to engage in out-of-home activities (Lu and Pas 1999). A second intriguing effect is the
presence of children in the household which is different between hosting / visiting (positive
effect) and bar / restaurants (negative effect for strong ties). The negative effect in bar /
restaurants is expected according to time-pressure hypotheses, which assume that the
presence of children implies more maintenance activities and thus less time for social /
recreational (Lu and Pas 1999). However, the positive effect opens some research questions:
first, whether the mix between in-home and out-of-home activities in hosting / visiting influences
the final positive result; second, whether measuring the number of individuals with whom social
activities are performed induces a positive effect in this case (e.g. families with children will tend
to have more relationship with other people with children and, in general, bigger groups); and
third, whether the effect is mainly due to the activity type (e.g. with children, bar / restaurants are
less “convenient” or comfortable places to perform social activities than homes).
A similar difference in signs can be found for the case of whether the ego lives with
partner: positive for hosting / visiting and negative for bar / restaurants, an intuitive result in that
people living with a partner would have more propensity to perform more activities at homes
compared with single people. In the case of gender, the negative effect of being female is
consistent with other results (Lu and Pas 1999). Regarding the negative effect of being
employed and the positive effect of working at home, time pressure explanations intuitively
support this result; being employed generates a much more fixed schedule than working at
home, potentially making people have lower propensities to perform social activities. Finally, the
results show an interesting statistically significant and positive effect of years in the household
and years in the city for strong tie people in hosting / visiting. A low residential and urban
mobility can be a proxy for a more stable and settled social network, and thus more intimate
people with whom to socialize. Further, there could be a neighborhood and urban effect that
13
could trigger a positive social effect: the more time people have spent in their neighborhood and
city, the more close people they know, and the more potential for social events.
Regarding network composition (Table 3), a first interesting result is the positive
relationship between the number of people living in Canada at more than an hour’s travel away
and the propensity to socialize with both strong and weak ties and both hosting / visiting and bar
/ restaurants. This is an interesting result which apparently contradicts the intuition that distance
is a barrier to performing social activities. However, from a network perspective, a potential
explanation is that, if respondents report many people living at further distances in their network,
it is very likely that these egos actively work more to “maintain” their relationships. In other
words, these egos may have more “propensity” for performing social activities compared with
those that have less people living further away. A second complementary potential explanation
is that individuals with more network members living further from them compensate their
socializing needs with network members living closer. This network effect in distance has
however some limit; in fact, the more strong tie people living outside Canada, the less likely the
individual hosts or visits strong tie people. This last result also shows that having more people in
the social network does not directly translate into necessary having more propensity to perform
social activities. Regarding the number of people of the same gender as the ego, results show a
significant negative effect in the case of weak ties for both hosting / visiting and bar / restaurant
propensities to perform social activities. These results intuitively seem appropriate since social
activities with weak ties tend to be in large groups of friends, and generally in couples or in
family (in the case of family, this is reaffirmed by the significant effect of weak ties immediate
family).
Also, the role composition within the network has different effects comparing hosting /
visiting and bars / restaurant social activities, and strong and weak ties. First, the positive and
significant direct effect of neighbors in both strong and weak ties in hosting / visiting suggests
that they are still an important part of urban social life. Interestingly, this result also reaffirms the
14
continuing importance of local space and short distances for socializing activities, mainly related
with home, also with an indirect positive effect in bar / restaurants. Second, work and student
mates have a positive significant effect only for strong ties, direct for hosting and visiting social
activities, and indirect for bar and restaurant social activities. Thus, the relationship between
work and student mates and the propensity to perform social activities is a positive relationship
mainly when the intimacy is higher. The propensity effects in activity types are different in the
case of people from organizations, which are only direct for bar / restaurants for both strong and
weak ties, and indirect for hosting / visiting. Third, the results also suggest the importance of the
number of immediate family in performing social activities, especially hosting and visiting, but
also bar and restaurants for weak ties. As intuitively expected, the other role as important as
immediate family is that of friends, whose directs effects are positively significant for both tie
strengths and social activity types.
The previous results suggest an interesting intertwined effect between with whom
individuals socialize and the distance for these social activities. First, local space remains
important, judging for the effect of neighbors, and also by the results in the personal and
household characteristics model, regarding the number of years in the household and the
number of years in the city. At the same time, further distances also remain important, as the
effect of the people living at more than one hour’s travel suggests. Although the previous results
are not conclusive, they seem to confirm a “glocalization” effect in the context of social
activities (Wellman 2001), that is, heavy interaction intensity at both far and close distances.
Second, the differences found between hosting / visiting and restaurants / pubs, suggest a
specialization of both spaces, which reinforces the argument that social activities conform to a
broad set of activities and episodes, which needs to be analysed in their specific context. In
other words, with whom social activities are performed, is a relevant aspect to understand the
specific characteristics of each activity type and their propensity to perform them.
15
The conceptual framework presented in section 2 explicitly incorporates the potential
role of information and communication technologies (ICT) in social episodes, defining them as
different media, which can supplement, complement or be neutral to social face-to-face
interactions. As it was discussed before, this media use variable is measured as the usual
number of people with whom the ego communicates by tie strength and frequency in each
media (cell phone, regular phone, email, and instant messaging) (see Table 4). First, a key
general observation is that all significant estimated direct and total effects are positive for
communicating by cellular phone, regular phone, and email; that is, from this perspective,
communicating with more people in each of these three media shows a complementary (if
significant) or neutral effect (if not significant) in the number of people with whom individuals
socialize, but never a substitution effect. This complementary effect is consistent with previous
discussions in the travel behavior literature (Mokhtarian et al 2003; Senbil and Kitamura 2003).
Exceptions of this result are the effects of frequent instant messaging, where very frequent
communication has a negative direct and indirect effect; further investigation needs to be done
considering that less frequent instant message communication has a positive or neutral effect,
that individuals who use instant message correspond to only a 15% of the entire sample and are
mainly restricted to the young cohort (less than 29 years old), and that instant message seems
to be less important in the social activity planning process than other media (Hogan 2005).
Second, different ICT media seem to have different stimulation effects based on the
nature of the tie, the frequency of communication, and social activity type. In the case of regular
phone, there is a consistent positive direct effect in most of the strong / weak tie combinations
for both hosting / visiting and bar / restaurants. The effects seem to be more specific for the
other media communication patterns, depending on tie strength and the frequency. For
example, the major direct effects in cell phone seems to be for strong tie / very frequent and
weak tie / less frequent combinations, and the importance of email communication varies
according to tie strength and activity type. This apparent specialization of frequency, tie
16
strength, and media raises interesting research questions about the influence of ICT and the
propensity to perform social activities. For example, this specialization effect may illustrate the
influence of ICT in the social activity planning process (as described using the same data set in
Hogan 2005); however, more disaggregated data accounting at the level of each ego-alter is
needed to obtain more solid conclusions about this issue.
4. Conclusions
This paper presents a model that incorporates the concept of propensity to perform social
activities, using a social networks approach. In general, propensity helps to link a set of different
potential causes of the generation of social activities, such as social networks, socioeconomic,
and individual attributes, as well as the ego’s communication patterns with his/her network by
other means, such as telephone and Internet-based media. The explicit incorporation of social
network concepts provides a useful way to describe the complexity of social activities, which not
only depend on the individuals’ scheduling and time use decision process, but also on their
social context, that is, “with whom” individuals perform those activities. Social network theory
provides a natural way of incorporating the intrinsic interactions that occur in social activities,
and also provides a potentially useful way of understanding aspects such as the influence of
information and communication technologies.
A first step towards testing and further understanding these ideas was made through an
empirical model that studied the propensity to perform social activities, measured as the number
of people with whom individuals socialize. Overall, results suggest that the effect of personal
and other characteristics on social activities cannot be generalized, and depends on the tie
strength (“with whom” the social activity is made), and specific social activity type (hosting /
visiting versus bar / restaurant). Further analysis needs to be done to disentangle the effect of
income and gender that suggest different – or at least complementary – explanations from those
traditionally found in the existing literature. Another interesting result is the positive effect of the
17
number of years in the household that seem to uncover a set of interesting processes, such as
the neighborhood effect in social activities, complemented with the importance of neighbors
network members on the propensity to perform hosting / visiting social activities. Also in the
case of role composition, the results of this paper suggest that family is an important segment to
consider as well as friends, but their effect varies according to tie strength and social activity
type. Regarding network composition, another interesting result is the positive relationship
between the number of people with whom individuals socialize and variables such as the
number of people living more than one hour’s travel away, and the number of neighbors in the
social network. This relationship suggests that the propensity to socialize cannot be explained
only by physical distance but needs to consider aspects of the individuals’ social behavior, more
explicitly who composes their social networks.
This distance effect, combined with the importance of neighbors suggests a potential
“glocalization” effect in face-to-face social activities (Wellman 2001); that is, intense social
activities both at near and far spaces, although the linkage with other ways of social interaction
remains to be seen. Precisely the effect of most ICT media on the overall propensity to perform
social activities suggests a complementary effect at most, but not a supplementary effect for the
overall sample. The exception is instant messaging, which needs further cohort study. Finally,
results indicate differences in the effects of different media, tie strength, and frequency
combinations, which suggest interesting venues to further analyze the effect of ICT in social
activity-travel behavior. Overall, the exercise of exploring these several attributes and their
effect in social activities show that studying the social context can help to better understand
behavior in physical space and the individuals’ propensity to perform social activities.
Although the empirical model has not rejected the ideas elicited from the conceptual
model, a number of aspects need further research. First, the effects discussed here need to be
further controlled by personal and socioeconomic characteristics, and possibly time use and
scheduling context, in order to assess whether they can be generalized or depend on specific
18
personal contexts. Second, personality needs to be explicitly incorporated; in fact, the
Connected Lives Study includes indicators of extroversion that can be incorporated in further
analysis. Third, the opportunities to engage in the social project need to be explicitly studied,
exploring the differences in the behavioral processes between hosting / visiting and bar /
restaurant activities; this exercise would require a more explicit and detailed consideration of
space. Finally, since the empirical model discussed here uses aggregated measures of the
social network’s composition and ICT communication patterns, a more disaggregated data
analysis needs to be developed, which could include in more detail the specific characteristics
of each ego-alter interaction. This kind of data were collected in the interview stage of the
Connected Lives Study, and can serve to illuminate and further test the questions raised here.
Overall, explicitly incorporating social networks into activity-travel modeling provides a
rich set of insights into social activities and their embedded behavioral processes, potentially
helping to better understand the propensity to perform social activities in particular, and the
general activity-travel behavioral process in general.
Acknowledgements
The authors would like to thank Barry Wellman, Bernie Hogan, Jeffrey Boase, Kristen Berg,
Jennifer Kayahara, and Tracy Kennedy, members of the NetLab group at the Centre for Urban
and Community Studies, University of Toronto, with whom the data used in this paper were
collected. Thanks also to Ilan Elgar, K. Nurul Habib, Antonio Páez, Matthew Roorda, and three
anonymous referees for their comments on previous versions of this work. Finally, the authors
would like to acknowledge the financial support received from the Social Sciences and
Humanities Research Council of Canada (SSHRC), Major Collaborative Research Initiative
(MCRI), and Regular Research Grant.
References
19
Axhausen, K.W. (2002) A dynamic understanding of travel demand: A sketch. Arbeitsberichte
Verkehrs- und Raumplanung, 119. Institut für Verkehrsplanung, Transporttechnik, Strassenund
Eisenbahnbau (IVT), ETH Zurich, Zurich.
Axhausen, K.W. (2006), Social networks and travel: Some hypotheses. In Donaghy, K Social
Aspects of Sustainable Transport: Transatlantic Perspectives, Ashgate, Aldershot.
Carrasco, J.A, Hogan, B, Wellman, B. and Miller, E.J. (2006) Collecting social network data to
study social activity-travel behavior: An egocentred approach. 85th Annual Meeting of the
Transportation Research Board, January 22-26, 2006, Washington D.C.
Chapin, F.S. (1974) Human Activity Patterns in the City: Things People Do in Time and in
Space, Wiley, New York.
Dugundji, E. and Walker, J. (2005) Discrete choice with social and spatial network
interdependencies: An empirical example using Mixed GEV models with field and “panel”
effects. Transportation Research Record 1921 70-78.
Golob, T. (2003) Structural equation modeling for travel behavior research, Transportation
Research 37B 1-25.
Hägerstrand, T. (1970) What about people in regional science? Papers of the Regional Science
Association 24 (7) 7-21.
Harvey, D. (1969) Explanation in Geography, Edward Arnold, London.
Hogan, B. (2005) Managing to connect: Linking social structure to personal network
management. XXV International Sunbelt Social Network Conference, February 16-20, 2005,
Redondo Beach, California.
Jöreskog, K. and Sörbom, D. (2001) LISREL 8.51. Scientific Software International.
Kemper, F.J. (1980) Social contacts of an urban-population within an activity-space framework.
Geographische Zeitschrift 68 (3) 199-222.
20
Lu, X. and Pas, E. (1999) Socio-demographics, activity participation and travel behavior.
Transportation Research 33A 1-18.
McCarty, C, Killworth, P.D, Bernard, H.R, Johnsen, E.C. and Shelley, G.A. (2000) Comparing
two methods for estimating network size. Human Organization 60 28-39.
Miller, E.J. (2005), Propositions for modelling household decision-making. In Lee-Gosselin, M.
and Doherty, S Integrated Land-Use and Transportation Models: Behavioural Foundations,
Elsevier, New York.
Miller, E.J. and Roorda, M. (2003) A prototype model of household activity/travel scheduling.
Transportation Research Record 1831 114-121.
Mokhtarian, P, Salomon, I. and Handy, S. (2003) A taxonomy of leisure activities, with
implications for the choice of ICT alternatives. 10th International Conference on Travel Behavior
Research, August 10-15, 2003, Lucerne.
Páez, A. and Scott, D. (2006) Social influence on travel behavior: A simulation example of the
decision to telecommute. Environment and Planning A, forthcoming.
Pas, E. (1990), Is travel demand analysis and modelling in the doldrums? In Jones P
Developments in Dynamic and Activity-based Approaches to Travel Analysis. Gower, Aldreshot.
Salvini, P. and Miller E.J. (2005) ILUTE: An operational prototype of a comprehensive
microsimulation model of urban systems. Networks and Spatial Economics 5 217-234.
Schlich, R, Schönfelder, S, Hanson, S. and Axhausen, K.W. (2004) Structures of leisure travel:
Temporal and spatial variability. Transport Reviews 24(2) 219-237.
Senbil, M. and Kitamura, R. (2003) Simultaneous relationships between telecommunications
and activities. 10th International Conference on Travel Behavior Research, August 10-15, 2003,
Lucerne.
21
Stutz, F.P. (1973) Intra-urban social visiting and leisure behavior. Journal of Leisure Research 5
(1) 6-15.
Tindall, D. and Wellman, B. (2001) Canada as social structure: Social network analysis and
Canadian Sociology. Canadian Journal of Sociology 26 (2) 265-308.
Washington, S, Karlaftis, M. and Mannering, F. (2003) Statistical and Econometric Methods for
Transportation Data Analysis, Chapman & Hall, Washington D.C.
Wellman, B. (1988), Structural analysis: from method and metaphor to theory and substance. In
Berkowitz, S. and Wellman, B Social Structures: A Network Approach, Cambridge University
Press, New York.
Wellman, B. (2001) Physical space and cyberspace: the rise of personalized networking.
Journal of Urban and Regional Research 25 (2) 227-252.
Wellman, B, Hogan, B, Berg, K, Boase, J, Carrasco, J.A, Côté, R, Kayahara, J, Kennedy, T.
and Tran P. (2006), Connected Lives: The project. In Purcell, P The Networked Neighbourhood.
Springer, London.
About the authors
Juan Antonio Carrasco, a PhD candidate in Civil Engineering at the University of Toronto,
holds a MSc degree in Transportation Engineering from the Pontificia Universidad Católica de
Chile. His doctoral research explores the relationships between social networks, activity-travel
behavior, and ICTs. His research interests also include microsimulation, land use-transportation,
and econometric modeling.
Eric J. Miller is Bahen-Tanenbaum Professor of Civil Engineering at the University of Toronto
where he is also Director of the Joint Program in Transportation. His research interests include
integrated land-use / transportation modeling, activity-based travel modeling, microsimulation
and sustainable transportation planning.
22
Table 1. Independent and Dependent Variables
Personal and household characteristics
INCOME Household income (categorical variable)
AGE Age (categorical variable)
CHILD IN HOUSEHOLD Presence of children at home
FEMALE Ego is female
LIVE WITH PARTNER Ego lives with partner
EMPLOYED Ego is employed
WORKS AT HOME Ego works at home
YEARS IN THE HOUSEHOLD Number of years the ego lives in the same household
YEARS IN THE CITY Number of years the ego lives in Toronto
Network Composition
IMMEDIATE FAMILY Number of social network members who are immediate family
NEIGHBORS Number of social network members who are neighbors
WORK / STUDENT MATES Number of social network members who are work or student mates
FROM ORGANIZATIONS Number of social network members who are from other organizations (e.g. sport or
social clubs)
FRIENDS Number of social network members who are friends not included above
WITH THE SAME GENDER Number of social network members who have the same ego’s gender
IN CANADA > 1 HOUR OF TRAVEL Number of social network members who live in Canada at more than an hour’s
travel away with respect to the ego
OUTSIDE CANADA Number of social network members who live outside Canada
Network interaction through ICT use
CALL BY CELL PHONE Number of social network members with whom the ego calls by cell phone: by tie
strength (strong, weak), and frequency (typically at least once a week and between
once a week and once a month)
CALL BY REGULAR PHONE Number of social network members with whom the ego calls by regular phone: by
tie strength (strong, weak), and frequency (typically at least once a week and
between once a week and once a month)
EMAIL Number of social network members with whom the ego emails: by tie strength
(strong, weak), and frequency (typically at least once a week and between once a
week and once a month)
USE INSTANT MESSAGE Number of social network members with whom the ego communicates by instant
message: by tie strength (strong, weak), and frequency (typically at least once a
week and between once a week and once a month)
Dependent variables
HOST / VISTING Number of social network members with whom the ego visits or hosts: by tie
strength (strong, weak), and frequency (typically at least once a week and between
once a week and once a month)
BAR / RESTAURANTS Number of social network members with whom the ego meets in places such as bar
or restaurants: by tie strength (strong, weak), and frequency (typically at least once
a week and between once a week and once a month)
23
Table 2. Model 1 (personal characteristics)
HOST / VISTING BAR / RESTAURANTS
From / To Effect Strong ties Weak ties Strong ties Weak ties
Total - - 0.22 (1.66) - -
HOST / VISTING, Strong ties a Direct - - 0.22 (1.66) - -
Total 0.22 (1.66) - - - -
HOST / VISTING, Weak ties a Direct 0.22 (1.66) - - - -
Total 0.23 (3.72) 0.29 (5.60) 0.12 (2.70) 0.30 (6.75)
INCOME Direct 0.17 (2.49) 0.23 (3.37) 0.15 (3.24) 0.29 (4.18)
Total -0.38 (-5.75) -0.34 (-6.64) -0.11 (-2.33) -0.28 (-6.85)
AGE Direct -0.32 (-4.43) -0.25 (-3.78) -0.15 (-3.31) -0.28 (-4.05)
Total 0.04 (1.68) 0.13 (2.95) -0.09 (-2.66)
CHILD IN HOUSEHOLD Direct 0.12 (2.85) -0.09 (-2.44)
Total -0.06 (-1.62) -0.27 (-6.33) -0.29 (-7.89)
FEMALE Direct -0.25 (-4.42) -0.28 (-5.81)
Total 0.04 (1.29) 0.09 (1.93) -0.17 (-4.51)
LIVE WITH PARTNER Direct 0.08 (1.86) -0.16 (-4.48)
Total -0.66 (-5.45) -1.14 (-9.8) -1.03 (-9.95)
EMPLOYED Direct -0.40 (-2.57) -0.97 (-4.99) -1.00 (-4.93)
Total 0.62 (5.57) 1.02 (9.54) 0.93 (9.88)
WORKS AT HOME Direct 0.39 (2.79) 0.86 (4.86) 0.91 (4.98)
Total 0.11 (1.86)
YEARS IN THE HOUSEHOLD Direct 0.11 (1.85)
Total 0.10 (1.68)
YEARS IN THE CITY Direct 0.09 (1.67)
Minimum Fit Function Chi Square = 110.86; Degrees of Freedom = 55; RMSEA 90% confidence = [0.037, 0.067]; CFI = 0.98
Note: “-“ are coefficients omitted in conceptual model, blank spaces = t < 1.20, a,b,c,d = coefficients set equal in the SEM
Table 3. Model 2 (network composition)
HOST / VISTING BAR / RESTAURANTS
From / To Effect Strong ties Weak ties Strong ties Weak ties
Total - - 0.13 (1.71) 0.18 (4.73) - - HOST / VISTING, Strong ties
b, c Direct - - 0.13 (1.71) 0.18 (4.73) - -
Total 0.18 (4.73) - - - - BAR / RESTAURANTS, Strong
ties c Direct 0.18 (4.73) - - - -
Total 0.23 (4.83) 0.03 (1.89) 0.08 (1.89) IMMEDIATE FAMILY, Strong
ties Direct 0.21 (4.97) - - - -
Total 0.15 (3.35) 0.02 (1.52) 0.02 (2.77)
NEIGHBOURS, Strong ties Direct 0.11 (3.42) - - - -
Total 0.07 (1.57) 0.01 (1.50) WORK / STUDENT MATES,
Strong ties Direct 0.08 (2.00) - - - -
Total 0.02 (2.08) 0.01 (1.37) 0.10 (2.27) FROM ORGANIZATIONS,
Strong ties Direct 0.09 (2.26) - -
Total 0.29 (6.30) 0.04 (1.66) 0.26 (5.87)
FRIENDS, Strong ties Direct 0.28 (6.64) - - 0.20 (4.66) - -
Total 0.06 (1.29) 0.12 (2.52)
IN CANADA > 1 HOUR OF
TRAVEL, Strong ties Direct 0.06 (1.43) - - 0.10 (2.35) - -
Total -0.16 (-3.53) -0.02 (-1.5) -0.03 (-2.85)
OUTSIDE CANADA, Strong
ties Direct -0.16 (-3.79) - - - -
Total 0.13 (1.71) - - - - 0.11 (1.64) HOST / VISTING, Weak ties
b, d Direct 0.13 (1.71) - - - - 0.11 (1.64)
Total - - 0.11 (1.64) - - BAR / RESTAURANTS, Weak
ties d Direct - - 0.11 (1.64) - -
Total 0.02 (1.47) 0.14 (3.27) 0.12 (2.82)
IMMEDIATE FAMILY, Weak
ties Direct - - 0.13 (3.06) 0.10 (2.47)
Total 0.01 (1.34) 0.09 (1.89) 0.01 (1.31) 0.01 (1.24)
NEIGHBOURS, Weak ties Direct - - 0.08 (1.88) - -
Total 0.01 (1.44) 0.12 (2.89)
FROM ORGANIZATIONS,
Weak ties Direct - - - - 0.12 (2.89)
Total 0.03 (1.65) 0.11 (2.41) 0.20 (4.47)
FRIENDS, Weak ties Direct - - 0.09 (1.90) - - 0.19 (4.28)
Total 0.03 (1.65) 0.20 (4.27) 0.25 (5.56)
IN CANADA > 1 HOUR OF
TRAVEL, Weak ties Direct - - 0.17 (3.58) - - 0.23 (4.99)
Total -0.02 (-1.36) -0.13 (-2.71) -0.20 (-4.47)
WITH THE SAME GENDER,
Weak ties Direct - - -0.10 (-2.24) - - -0.19 (-4.28)
Minimum Fit Function Chi Square = 193.02; Degrees of Freedom = 86; RMSEA 90% confidence = [0.044, 0.067]; CFI = 0.95
Note: “-“ are coefficients omitted in conceptual model, blank spaces = t < 1.20, a,b,c,d = coefficients set equal in the SEM
24
Table 4. Model 3 (ICT contact in social network)
HOST / VISTING BAR / RESTAURANTS
From / To Effect Strong ties Weak ties Strong ties Weak ties
Total - - - -
HOST / VISTING, Strong ties Direct - - - -
Total - - - - 0.13 (1.89) BAR / RESTAURANTS, Strong
ties a Direct - - - - 0.13 (1.89)
Total 0.22 (5.90) 0.15 (3.24) 0.02 (1.76) CALL BY CELL PHONE < 1
week, Strong tie Direct 0.21 (4.86) - - 0.12 (2.34) - -
Total CALL BY CELL PHONE 1 week -
1 month, Strong tie Direct - - - -
Total 0.34 (8.36) 0.15 (3.24) 0.01 (1.39) CALL BY REGULAR PHONE < 1
week, Strong tie Direct 0.32 (7.73) - - 0.06 (1.24) - -
Total 0.25 (5.83) 0.08 (1.43) CALL BY REGULAR PHONE 1
week - 1 month, Strong tie Direct 0.24 (5.66) - - 0.05 (1.26) - -
Total 0.09 (1.58)
EMAIL < 1 week, Strong tie Direct - - 0.09 (1.57) - -
Total 0.06 (1.53) 0.14 (2.38) 0.01 (1.49) EMAIL 1 week - 1 month, Strong
tie Direct 0.05 (1.21) - - 0.13 (2.26) - -
Total -0.05 (-1.35) -0.09 (-1.97) -0.01 (-1.41) USE INSTANT MESSAGE < 1
week, Strong tie Direct - - -0.08 (-1.85) - -
Total USE INSTANT MESSAGE 1
week - 1 month, Strong tie Direct - - - -
Total - - - -
HOST / VISTING, Weak ties Direct - - - -
Total - - 0.13 (1.89) - - BAR / RESTAURANTS, Weak
ties a Direct - - 0.13 (1.89) - -
Total CALL BY CELL PHONE < 1
week, Weak tie Direct - - - -
Total 0.09 (2.00) 0.02 (1.64) 0.13 (1.76) CALL BY CELL PHONE 1 week -
1 month, Weak tie Direct - - 0.09 (1.81) - - 0.13 (2.81)
Total 0.06 (5.13) 0.01 (1.35) 0.10 (1.81) CALL BY REGULAR PHONE < 1
week, Weak tie Direct - - 0.25 (4.94) - - 0.09 (1.47)
Total 0.29 (5.64) 0.02 (1.56) 0.13 (2.39) CALL BY REGULAR PHONE 1
week - 1 month, Weak tie Direct - - 0.29 (5.33) - - 0.12 (1.91)
Total 0.02 (1.55) 0.15 (2.76)
EMAIL < 1 week, Weak tie Direct - - - - 0.15 (2.77)
Total 0.06 (1.24) 0.02 (1.55) EMAIL 1 week - 1 month, Weak
tie Direct - - 0.06 (1.20) - -
Total -0.02 (-1.69) -0.17 (-3.54) USE INSTANT MESSAGE < 1
week, Weak tie Direct - - - - -0.16 (-3.54)
Total 0.01 (1.39) 0.10 (2.01) USE INSTANT MESSAGE 1
week - 1 month, Weak tie Direct - - - - 0.09 (2.01)
Minimum Fit Function Chi Square = 218.31; Degrees of Freedom = 89; RMSEA 90% confidence = [0.048, 0.072]; CFI = 0.96
Note: “-“ are coefficients omitted in conceptual model, blank spaces = t < 1.20, a,b,c,d = coefficients set equal in the SEM
25
Figure 1. Conceptual causal structure represented in the structural equation models
Bar / Restau
r
ants
Strong ties
Hosting / Visiting
Weak ties
Hosting / Visiting
Strong ties
Bar / Restaurants
Weak ties
Hosting / Visiting
# socializing strong tie people
usually between once a week – once a month
Independent
variables
Dependent variable (social activities
with ego’s social network)
Latent variables:
Propensity to perform
social activities by tie
Hosting / Visiting
# socializing strong tie people
usually less than once a week
Hosting / Visiting
# socializing weak tie people
usually between once a week – once a month
Hosting / Visiting
# socializing weak tie people
usually less than once a week
Bar / Restaurants
# socializing strong tie people
usually between once a week – once a month
Bar / Restaurants
# socializing strong tie people
usually less than once a week
Bar / Restaurants
# socializing weak tie people
usually between once a week – once a month
Bar / Restaurants
# socializing weak tie people
usually less than once a week
Independent variables:
1. Personal
characteristics
2. Social network
composition by tie
strength
3. ICT use with network
members by frequency
and by tie strength
ε
1
ε
2
ε
3
ε
4
ε
5
ε
6
ε
7
ε
8
