Content uploaded by Michael Maness
Author content
All content in this area was uploaded by Michael Maness on Sep 23, 2017
Content may be subject to copyright.
A Theory of Strong Ties, Weak Ties, and Activity Behavior: Leisure
Activity Variety and Frequency
Michael Maness
Postdoctoral Research Associate
Oak Ridge National Laboratory
National Transportation Research Center
2360 Cherahala Boulevard
Knoxville, TN 37932
Phone: (865) 946-1288
Email: manessm@ornl.gov
Submission Date: August 1, 2016
Revision Date: November 29, 2016
Word Count: 6694 words (text) + 4 Tables + 1 Figure = 7951 words
TRR Paper Number: 17-06616
Maness 2
ABSTRACT
A social networks and personal social capital perspective have recently been applied in travel
and activity behavior. In this paper, a theory of activity behavior is developed inductively from
sociological principles about social capital and the creation of social networks. It focuses on the
quantity of strong social ties and the diversity of weak social ties to describe differences in
leisure activity variety and frequency. The theory is applied to a case study on leisure activity
participation in the US using the Personal Networks and Community Study. The size of
respondent’s core networks and the diversity of their contacts are found to be positively
correlated with activity variety and activity frequency. Additionally, endogeneity due to
correlations between network diversity and social personalities is accounted for with a two-step
estimation procedure. Results showed that including strong tie count and weak tie diversity
significantly increased model fit and support the theory’s hypothesis. The results also show the
biases that can exhibit in analyses that ignore social network effects. Including social networks
helps to account for additional heterogeneity in socioeconomic groups (race, income, marital
status, and education) in this case study.
Maness 3
INTRODUCTION
Existing modeling structures in regional planning describe society through distributions of
individual and household characteristics. While this system has served the transportation
planning community well in forecasting travel and answering mobility related inquiries, it is
insufficient for answering question of an interdisciplinary and intersectional nature. Current
models cannot answer such questions as: (1) How does transit accessibility affect social isolation
and what are the health implications for elderly populations? and (2) How does a parent’s spatial
dispersion of social contacts affect coordination of travel during emergency evacuations of
schools? These questions require modeling grounded in theories that consider social networks
and social interactions.
New social interactions research can promote this transition to incorporating social
network typologies into transportation planning models. Major new planning models are based
on activity-based modeling methodologies. These models begin with activity diaries. Activity
diaries have begun to ask respondents to explain who they perform activities with (1-3). This has
tended to be in the form of using a simple add-on question per activity: “With whom did you
participate in this activity with?” The answer choices tend to be simple – e.g. none, family,
friends. These efforts have spawned work to create models of social cooperation in the form of
activity coordination in activity selection and duration. In activity-based models, limited work
has been performed to explore models where cooperation and coordination of activities occur
within the household and sometimes between households (4).
But these measurement and modeling efforts have suffered from a lack of understanding
the social factors behind the participation in certain activities. To develop social network-based
behavioral modeling frameworks capable of answering policy and planning questions where
social network characteristics are relevant in sociotechnical systems, more theoretical support is
needed. Such a modeling framework must include a component that forms social connections
between individuals that it uses in the interactions between agents. But what factors and what
level of detail is needed to create such social networks?
In this paper, a theory of activity behavior is proposed that is developed from principles
about the nature of strong social ties and weak social ties (5). Specifically, the theory proposes
that increases in the number of strong social ties increases activity variety and frequency. But
additionally, the diversity and prestige of one’s social ties also increases activity variety and
frequency. This latter point has potential implications in the design of activity studies and models
as there has been limited focus on people’s loose social networks. A case study is performed
using the Personal Networks and Community Study where support for the theory is confirmed in
an econometric analysis. Respondents describe their social networks through a name generator
and position generator and these network indicators were found to be positively correlated with
activity variety and frequency.
This study contributes to the activity behavior literature with a clear and simple
explanation of important social network factors in out-of-home leisure (social) activity behavior.
Existing work has focused generally on strong ties via the use of a name generator, whereas this
study adds a position generator to account for weak tie characteristics. This study has
implications in the design of activity-based models looking to use social networks as model
components. Particularly, there has been limited emphasis on weak social ties, intrahousehold,
and interhousehold networks in these models (4). This theory and the supporting case study
encourages further research and may eventually lead to more behaviorally and socially realistic
models of activity and travel behavior.
Maness 4
SOCIAL CAPITAL AND ACTIVITY BEHAVIOR
Social capital essentially encapsulates the concept that social networks have value. Social
networks enable individuals to perform actions that could not be performed individually, would
be much more difficult without assistance, or would not be enjoyable alone. Strong ties are social
contacts with which an individual has a close connection. An individual tends to communicate
often with their strong ties and place a great deal of trust with these individuals. Examples of
strong ties include parent-child relationships, spousal relationships, sibling relationships, and
best friend relationships. Weak ties are social contacts with which an individual tends to have a
more loose connection (5). This looseness may be due to the short duration of the relationship,
infrequent interaction, or a personal feeling of a lack of closeness (6). Examples of weak tie
relationship are acquaintances, co-workers / colleagues, or family members that one is not too
close with.
In the field of travel and activity behavior, social capital can be extended in two ways: (1)
how networks bring value by enabling travel and activities and (2) how travel and activities
allow for the creation of social capital. Carrasco and Cid-Aguayo (7) analyzed the latter in the
area of social support (emotional and resources) in Chile. They focus on how measurable
characteristics of and relationships in one’s social network impact emotional support, small
money loans, carpooling, and job opportunities. They found that vehicle ownership impacted
emotional support more than resource support. Tilahun and Li (8) found correlations between the
face-to-face meeting frequency of strong ties and an individual’s age and gender (and their
differences between an individual and her strong ties). They also found that distance was not a
limiting factor within 50 miles (80 km) of respondent’s homes but limited contact at greater
distances.
Sadri, et al. (9) analyzed the characteristics of individuals’ strong ties on joint trip
frequency for different activity types. Their work found that using measures of network density,
homophily, and heterogeneity helped explain the frequency of joint trip. For example, greater
racial diversity among an ego’s alters was correlated with greater quantities of “eating out,”
study, and extra-curricular shared trips. Sadri, et al. based their choice of variables on concepts in
social capital pertaining to the sharing of resources. Network density was included because “A
dense personal network indicates close interpersonal contacts among alters, and helps to
promote the sharing of resources. In contrast, a personal network with many loose
connections (also known as structural holes) has been found to facilitate the flow of new or
unique information and resources.”
A Theory of Strong Ties, Weak Ties, and Leisure Activity Behavior
This theory depends on the relationships between social networks and activity behavior. The
theory works along four dimensions: (1) strong ties, (2) weak ties, (3) activity frequency, and (4)
activity diversity (e.g. activity type).
The basic tenets of this social capital theory of strong ties, weak ties, and activity
behavior are as follows:
1. As an individual’s social network grows in size, the individual’s frequency of leisure
activity participation will increase.
2. As an individual’s social network grows in size, the individual’s diversity of leisure
activity participation will increase.
Maness 5
The theory is based inductively on the relations between social safety, effectance, and status
seeking.
Social Safety
Social safety describes how individuals use their social network to nourish a sense of
community, affiliation, and trust (10). Social safety is built through strong ties in a person’s
social network. These strong ties tend to result in social structures that are dense and exhibit
homophily and propinquity. Homophily describes how individuals tend to associate with people
who are like themselves. Propinquity describes how individuals tend to associate with people
who it is easier to communicate with. This is typically described through spatial proximity but
can also be generalized to other forms, such as virtual propinquity.
Since strong ties are spawned from the motivation of social safety, it is theorized that this
would contribute to related activity patterns. Particularly, the safer someone feels with a
situation, the less risk and burden is associated with that situation which can lead to preference
for that situation over less safe ones. This manifests in the following hypothesis:
3. The size of an individual’s core network (strong ties) positively impacts the frequency of
leisure activity participation.
Because social safety results in strong ties, individuals are more likely to participate in activities
with their strong ties.
But this social safety also leads to homogenous groups due to the property of homophily.
Because an individual’s strong ties likely have similar interests and preferences to the individual,
an overreliance on strong ties in activity planning can results in less diverse (and more stable)
activity behavior patterns. For example, if an individual is a fan of local sports teams, then that
person tends to have friends who similarly are sports fans. The individual may often to go to
venues to watch sports with their friends rather than participating in other cultural activities
around their local area. This is related to the idea of “echo chambers.” The individual’s strong
ties have access to particular information about mostly sport activities around town that
reinforces their stable, less diverse set of activities.
Effectance and Status
Although social networks have a tendency towards social safety, effectance helps to broker
relations between different social groups to spread ideas, information, and resources. Effectance
describes how humans have a desire to explore the unknown (6). This is typically manifested
through weak ties between social groups. These weak ties are people whom are not contacted on
a regular basis for general advice, who are not very significant or important to an individual, but
are still considered acquaintances and are contacted from time to time. Due to relatively less
contact between an individual and their weak ties, this likely correlates with fewer activities
participated with these individuals. Thus, theoretically weak ties could have a weaker effect on
increasing the activity participation than strong ties.
Each weak ties itself tends to not be as important as the overall characteristics of the
weak ties in an individual’s social network. For example, studies of job access and social capital
concentrate on the diversity, status accessibility range, and upper reachability of status in an
individual’s social network (11). This concept can be extended to activity participation to look at
how the diversity of social contacts (weak ties in particular), impacts the diversity of activities
Maness 6
that an individual participates in. Since weak ties enable brokerage to new information, weak ties
provide the individual with new and different ideas for activities, thus varying the individual’s
activity patterns. This is theorized in the fourth tenet that diversity of weak ties increases activity
diversity more than the size of strong ties:
4. The diversity of an individual’s loose network (weak ties) positively impacts the diversity
and frequency of leisure activity participation.
5. Individuals with access to weak ties with higher social status have a greater diversity of
leisure activities than those less diversity.
As the occupational diversity of an individual’s contacts increases, the more likely the individual
is able to find a job when unemployed through a “networking” effect. The diversity of social
contacts aids the individual by increasing the likelihood of finding potential employers and
favorable job listings (12).
Lastly, status entails a ranking of the power and prestige of individuals and comparisons
thereof. Status can be created by organizational structures (e.g. job roles at work) and the
allocation of resources (e.g. money, authority, social connections). This can encourage social
interactions where individuals attempt to status seek – whether consciously or subconsciously –
in order to maintain their status or seek higher status (6). Because this higher status entails a
concentration of power and people strive for status, these higher status individuals tend to
establish large shares of contacts – particularly, weak ties. Being connected to these higher status
individuals could more likely place individuals in social circles unlike their own because these
higher status individuals interact with more people.
CASE STUDY: PERSONAL NETWORKS AND COMMUNITY SURVEY
To explore this theory using empirical data, the Pew Internet Personal Networks and Community
study (13) was used. This dataset includes questions on individual and household characteristics,
community involvement, internet usage, public spaces, and social networks. The social network
components included two question types for exploring social networks: a name generator for
strong ties and a position generator for weak ties.
Survey Design
The Personal Networks and Community study was conducted in July and August 2008 by the
Pew Internet and American Life Project. The survey was designed as an interviewer
administered telephone survey including both landline and mobile users. These survey design is
summarized in Table 1.
Measuring Activities
The survey provides self-reported data on the frequency of different activities. Eight activity
locations were provided and respondents provide frequency of visits for the last month. The
number of visits recorded is an integer between 0 and 6 (inclusive) – reported values greater than
6 are recorded as 6. The eight activity types included visits to: (1) coffee shop, (2) place of
worship, (3) library, (4) fast-food restaurant, (5) other restaurant, (6) community center, (7) park,
and (8) bar. Fast-food restaurant visits were excluded because it was assumed that individuals
were more likely to visit these places alone due to the quick service provided at these locations.
Maness 7
Measuring Strong Ties and Weak Ties
When collecting personal network data, name generators are typically used to elicit an
individual’s social contacts (11). The traditional name generator includes a generator prompt
which frames the types of people the respondent should attempt to recall. In the Personal
Networks and Community Survey, two traditional name generators were used in the survey. The
first name generator probed for individuals with whom the respondent discussed important
matters. Interviewers recorded up to five names, and if individuals submitted fewer than 5
names, the interviewer would attempt to probe for more names. This name generator used the
following question:
From time to time, most people discuss important matters with other people. Looking back over
the last six months — who are the people with whom you discussed matters that are important to
you? If you could, just tell me their first name or even the initials of their first AND last names.
The second name generator probed for individuals with whom the respondent felt were most
important to them. Interviewers recorded up to 5 new names and also recorded if names given
from the previous name generator were also repeated. This name generator used the following
question:
Now let’s think about people you know in another way. Looking back over the last six months,
who are the people especially significant in your life? [IF NECESSARY: By significant, I mean
just those who are MOST important to you.] If you could, just tell me their first name or even the
initials of their first AND last names. These may be some of the same people you just mentioned
or it may be other people.
As with the first name generator, if less than five new names are given, the interviewer probed
for more names. In total, up to ten names are possible for this name generator.
From the traditional name generator, we can gain an understanding of the size of the
person’s core network (i.e. their strong ties). In this study, the number of unique names
mentioned will be used as an indicator of the size of a respondent’s core network.
An alternative version of the name generator, the position generator, was proposed by Lin
and Dumin (12). In the position generator, generally respondents are asked if they know certain
types of people, but the focus is not on who exactly that person is. As Lin and Dumin (12)
originally used, they gave respondents a list of occupations of varying levels of status and
prestige. For each occupation, they then asked respondents if they knew an acquaintance who
was currently employed in that profession. In the Personal Networks and Community survey, the
following position generator prompt was used:
Next, I am going to ask about types of jobs and whether people you know hold such jobs. These
people include your relatives, friends and acquaintances. Do you happen to know someone who
is… [INSERT ITEM; RANDOMIZE]? What about…[INSERT]? [IF NECESSARY: Do you
know someone who is [INSERT]?]
Respondents then notified the interviewer whether they knew or did not know someone with the
given occupation. Twenty-two occupations were asked about:
Maness 8
1. a nurse
2. a farmer
3. a lawyer
4. a middle school teacher
5. a full-time babysitter
6. a janitor
7. a personnel manager
8. a hair dresser
9. a bookkeeper
10. a production manager
11. an operator in a factory
12. a computer programmer
13. a taxi driver
14. a professor
15. a policeman
16. a Chief Executive Officer (C-E-O) of a Large Company
17. a writer
18. an administrative assistant in a large company
19. a security guard
20. a receptionist
21. a Congressman
22. a hotel bell boy
From the position generator, an understanding of the respondent’s loose social network
(i.e. their weak ties) can be inferred. An indicator of the diversity of the respondent’s weak ties
can be obtained by summing up the number of “yes” responses. Additionally, since each
profession is associated with a certain level of status in society, the upper reachability of a
respondent’s loose network can be obtained. This measure equals the maximum status level of
the respondent’s acquaintances’ occupations. Occupational prestige levels were obtained from
the Standard International Occupational Prestige Scale ((15-16). Prestige is measured on a scale
between 0 and 100.
For the strong ties quantity using data from the name generators, the distribution of core
network size is shown in Figure 1. The median core network size was 3 alters with a mean of
3.34 alters. The 5th percentile core network size was 1 alter and the 95th percentile core network
size was 7 alters. The distribution of weak tie occupational diversity as measured by number of
occupations from the position generator is shown in Figure 1. The median network diversity was
10 occupations with a mean of 9.71 occupations. The 5th percentile network diversity was 1
occupation and the 95th percentile network diversity was 22 occupations.
Activity Variety Model Formulation
The model is formulated as a Poisson regression model where the dependent variable is the
number of different leisure activity types an individual participates in. The activity variety is
assumed to depend on individual and household factors and social network characteristics. The
model is specified through the following mathematical expectation:
(|,,,
,)=exp{+ln()+ln(+ 1)+
+}
Maness 9
Where:
≡ activity variety for individual n
≡ individual and household characteristics for individual n
≡ individual n’s core network size (number of strong ties)
≡ occupational diversity of individual n’s social network (normalized linearly
between 0 and 1)
≡ upper reachability of occupational prestige of individual n’s social network
≡ unobserved heterogeneity
, ≡ model parameters
Endogeneity was hypothesized to be possible due to correlation between the diversity of
occupations and unobserved factors. Specifically, it is possible that more sociable and outgoing
people may prefer to go out more often and have more friends. To account for this possibility, a
two-stage control function approach is used to control for the endogeneity (17). So under this
condition, we can rewrite weak tie diversity (i.e. the endogenous regressor) as:
ln(+ 1)=+
Where:
≡ individual and household characteristic that are exogenous to (with at least
one element different than )
≡ reduced-form errors
The model is closed by assuming that (,) is independent of such that:
=+, independent of and ()= 0
Thus, the expected value of the activity variety becomes:
(|,,,
)=exp(+ln()+ln(+ 1)+
++)
Consistent estimates for the parameters can be obtained through a two-step procedure (17):
1. Calculate residuals � from an OLS regression of weak tie diversity on
2. Replace the unknown with the residuals � in a Poisson regression of activity variety
where: (|,,,
)=exp{+ln()+ln(+ 1)+
+�}
Models were estimated using the MASS package in R (18). Overdispersion was tested for
using Cameron and Trivedi’s regression-based test for overdispersion (19). All models were
found to have overdispersion. Although this may imply that a negative binomial model should be
used, a Poisson regression is more robust with fewer assumptions. Wooldridge (17) explains that
the Poisson model is more efficient than the negative binomial model under its variance
assumption and that the Poisson model “has the edge over NegBin I [model] for estimating the
parameters of the conditional mean. If conditional probabilities need to be estimated, then a more
flexible model… is probably warranted” (p. 736-7). Because the emphasis of this study is on
Maness 10
testing for an increase in activity frequency and variety but not for predicting that activity
behavior, the robustness of the Poisson model is preferred. To deal with overdispersion,
sandwich standard errors are used to obtain correct standard errors. The reported standard errors
in all regressions are obtained using the lmtest (20) and sandwich (21) packages in R.
Activity Frequency Model Formulation
The model is formulated as a Poisson regression model where the dependent variable is sum of
the monthly participation frequencies over all activities except fast-food restaurant. The
formulation and correction for endogeneity and inconsistent standard error due to overdispersion
are identical to the specification in section 3.4.
MODEL RESULTS
In this section, the model results are presented for the activity variety and frequency models.
First, the OLS estimates used to obtain residuals for controlling for endogeneity are presented.
Then, both models have their results presented and the section concludes by comparing insights
from both models.
First Step Estimates to Control for Endogeneity
OLS estimates for the occupational diversity variable were performed and residuals were
obtained. These residuals were then inserted into the count models (activity variety and
frequency). This variable is the natural logarithm of a normalized measure of diversity (one plus
the quotient of the sum of weak tie occupations divided by 22). Table 2 provides the first step
estimation for both the activity variety and frequency models that handle endogeneity. The
residuals from this OLS estimation are inserted into each model as a covariate.
The major concern for endogeneity comes from the possibility that people who have
more diverse weak ties are also just likely to be outgoing people. The instruments used are
correlated with occupational diversity but are not correlated with the unobservable trait of being
outgoing. The age variable, for example, is included although being older may impact a person’s
health and working situation. But disability and retirement are included as covariates in the main
models so they are not part of the unobservable portion of the outgoing trait.
There is possible concern that the other social network measures are endogenous as well.
Core network size was not considered endogenous because having people to discuss important
matter and who were significant to you does not necessarily require out-of-home activities to
maintain contact. Particularly, these relationships generally involve close contact over many
years, such as being married or being family, and activity behavior would have some limited
effect on this. The process of meeting a variety of people seemed to involve more dependence on
going out and meeting people which creates the variety.
Activity Variety Results
The estimation of the activity variety model is provided in Table 3 for the non-social model and
the model with social capital measures (assuming exogeneity and endogeneity of weak tie
occupational diversity).
Individual Characteristics. Individuals with higher levels of education participated in a
greater variety of leisure activities than those with less education. Marital / intimate relationship
status impacted leisure activity variety where those who had been married exhibited similar
activity variety tendencies. Individuals who were currently single (not in any intimate
Maness 11
relationships) participated in less variety of activities, while those individuals who were in
intimate relationships but not married participated in a greater variety of activities.
Compared to white respondents, African American and Native American respondents had less
diverse leisure activity behavior, while Asian Americans had more diverse behavior. Compared
to full-time employees, part-time workers and retirees exhibited a greater variety of activity
behavior while disabled respondents exhibited less variety. This may be due to larger amounts of
non-mandatory work time for part-time workers and retirees, while disabled respondents may
have decreased mobility options.
Household Characteristics. Individuals who lived in apartments participated in a greater
variety of activities than individuals in detached houses. This may be due to differences in the
built environment. Apartments and townhouses tend to be in denser neighborhoods. Housing,
population, and business densities likely affect accessibility to activity location resources. Future
work with spatial data could help to explain if this is a plausible explanation for the residential
differences in leisure activity frequency behavior.
Higher family income tended to cause an increase in activity variety, but this effect reduced in
size and becomes statistically insignificant as social network characteristics were in included in
the model.
Social Network Characteristics. Adding the strong and weak tie information to the model
improved the model fit. The hypothesized effect of strong and weak ties was confirmed with core
network size, network occupational diversity, and upper reachability of occupational prestige
having a positive effect on activity variety. Handling the endogeneity in the occupational
diversity caused the effect from the core network size to decrease while increasing the effect
from the occupational diversity. According to the model correcting for endogeneity, someone
with maximum weak tie occupational diversity participated in about 5 times as many of the
measured activities types as someone with no diversity. The upper reachability parameter
maintained a similar effect size. Thus, the model provides support for the three hypotheses of:
(1) weak tie diversity increasing activity variety, (2) strong tie quantity increasing activity
variety, and (3) higher social prestige corresponds to greater activity variety.
Activity Frequency Results
The estimation of the activity frequency model is provided in Table 4 for the non-social model
and the model with social capital measures (assuming exogeneity and endogeneity of weak tie
occupational diversity).
Individual Characteristics. Individuals with higher levels of education participated in
more activities than those with less education. Marital / intimate relationship status impacted
leisure activity frequency where those who had been married exhibited similar activity frequency
tendencies. Individuals who were currently single (not in any intimate relationships) participated
in fewer activities, while those individuals who were in intimate relationships but not married
participated in more activities. Accounting for a respondent’s social network (strong and weak
ties) increased the explanatory power of many individual characteristics. Specifically, age,
gender, and race become statistically significant. Also, the effects of employment status became
less significant. This may be due to the social network characteristics disentangling
heterogeneity within these social groups and leading to less variation in parameter estimates.
Household Characteristics. Individuals who lived in apartments and townhouses
participated in more activities than individuals in detached houses. This may be due to
differences in the built environment – similarly to the results in the activity variety model. The
Maness 12
parameter for missing family income exhibited more variation when social network
characteristics were included in the model.
Social Network Characteristics. Adding the strong and weak tie information to the model
improved the model fit. The hypothesized effect of strong and weak ties was confirmed with core
network size, network occupational diversity, and upper reachability of occupational prestige
having a positive effect on activity frequency. Handling the endogeneity in the occupational
diversity caused the effect from the core network size to decrease and become less statistically
significant. According to the model correcting for endogeneity, someone with maximum weak
tie occupational diversity participated in about 8.6 times as many of the measured activities as
someone with no diversity. Thus, the model provides support for the hypothesis of weak tie
diversity increasing activity frequency. But the model provides weak support for the hypothesis
that the number of strong ties increases activity frequency.
Comparisons between Activity Variety and Frequency Models
Most of the social network, individual, and household affect both variety and frequency
similarly. The effects of age are non-linear with respect to frequency but linear with respect to
variety. Thus individuals tend to decrease their activity frequency at an increasing rate in older
age and but vary their activity at a consistent rate. Disability appears to negatively impact
activity variety and frequency but was statistically insignificant for frequency. Women tended to
have less activity variety and frequency than men, but this effect was only significant for
frequency. To control for childrearing responsibilities, including children in the models had no
significant effect on variety or frequency. Most of the activities included (aside from the bar) are
places that are typically considered family-friendly. At the household level, locating in an
apartment was positively correlated with activity variety and frequency. But income and living in
a townhouse only significantly affected activity frequency.
DISCUSSION AND FUTURE WORK
A theory of strong ties, weak ties, and activity behavior was presented in this paper. The
theoretical models starts by theorizing that the quantity of strong ties and diversity of weak ties
an individual has is correlated with both their activity frequency and activity variety. This theory
is supported by evidence from social network theories pertaining to social safety, effectance, and
status seeking. The theory is tested on the Pew Internet and Community Survey which features
both a name generator for measuring strong ties and a position generator for measuring weak tie
diversity. Count data models were estimated to determine how individual characteristics and
egocentric social network measure affect leisure activity variety and frequency. Results showed
that including strong tie count, weak tie diversity, and weak tie upper reachability significantly
increased model fit and support the theory’s hypothesis. The results also show the biases that can
exhibit in analyses that ignore social network effects. Excluding these factors magnified the
effects of race, income, marital status, and education in this case study. Including social networks
helps to account for heterogeneity in these socioeconomic groups. This is an important
consideration in the design of future research.
The results of the case study are somewhat limited by the survey design. The list of
activities is non-exhaustive which makes further inference and prediction difficult. This can be
solved by designing a study with an activity diary that included all leisure activities. Specifically,
indicators of network diversity and upper reachability of prestige / status have not seen use in
other studies using activity diaries (22). Additionally, accounting for endogeneity is limited due
Maness 13
to the cross-sectional nature of the study. Appropriate instrument are difficult to find in this case
and it is suggested that future work use panel data to aid in accounting for correlation between
social network factors and activity behavior-related unobservables.
This work has practical implications on the design of travel behavior models –
particularly, activity-based models. Social networks and their effect on social activity generation
bring social cooperation and social capital into the activity behavior research field. Including
social network information has the potential to increase the behavioral realism of activity-based
models, particularly for interhousehold activities. Additionally, overlaying spatial information on
social networks will strengthen social network-based activity-based modeling. From a policy
perspective, incorporating these factors allows for new concepts to be analyzed including social
cohesion and social isolation and the role of the build environment, community structure and
layout, and travel on the creation and maintenance of social ties and social capital. This has
particular relevance in the near to medium term as governments attempt to protect and aid
disadvantaged populations and as elderly populations begin to grow in developed countries.
In future work, the theory can be tested and extended to the case of activity diversity in
activity location. This could be accomplished through analyzing activity location data from
activity diaries and location-based social networks. Additionally, the theory is very general for
leisure activities. Some specific leisure activity types may not be affected by an individual’s
social network. Even in this dataset, fast-food restaurants visits were removed because of this
possibility (22).
The relevance of the position generator is currently unknown in a travel and activity
context. A possible concern with using a position generator in a study about activity behavior is
that it was originally designed for use in studies of access to new job opportunities. But it has
been applied in other areas such as political participation (23), civic participation (24), and public
libraries (25). A possible linkage is to study how career choice relates to personalities and
activities. Holland’s theory of career choice provides a basis where he describes how most
individual as well as work environments can be represented as a combination of personality
types (26). An important part of the theory, its development, and its operationalization has been
that individuals tend to search for and enter work environments that fit their personalities (27).
Additionally, it has been shown that Holland’s personality types correlate well with modern
personality classification systems such as the Big Five personality factors (28-29). To complete
the loop, Furnham (30) confirmed that personality is correlated with activity preferences. But
this theory has not been fully tested in the travel and activity behavior field.
There is no clear understanding of what degrees of diversity affect activity selection. The
occupation-based position generator in this study was chosen because it is empirically grounded
but its theoretical basis is still limited. Other measures of diversity, such as ethnicity, gender, and
religion, may have some relevance and the position generator can be modified to handle
variation in the characteristics of weak ties (e.g. asking about the gender of doctors that an
individual knows). These diversity measures likely also affect other aspect of activity behavior
such as location variety. This could have particular relevance in societies where people are
segregated residentially and socially (whether by government or individual choice) by these
factors.
Lastly, this dataset also was unable to shed light on some of the details between social tie
type and activity behavior. Particularly, it would be of interest to see the relative influence of
how strong ties impact activity behavior versus weak ties. That question was difficult to answer
in the study as there was no directly comparable unit of measure. There was data on the number
Maness 14
of close contacts and data on the diversity of acquaintances. It is suggested that future study also
measure the number of weak ties to provide a method to determine relative influence.
ACKNOWLEDGEMENT
This manuscript has been authored by UT-Battelle, LLC under Contract No. DE-AC05-
00OR22725 with the U.S. Department of Energy. The United States Government retains and the
publisher, by accepting the article for publication, acknowledges that the United States
Government retains a non-exclusive, paid-up, irrevocable, world-wide license to publish or
reproduce the published form of this manuscript, or allow others to do so, for United States
Government purposes. The Department of Energy will provide public access to these results of
federally sponsored research in accordance with the DOE Public Access Plan
(http://energy.gov/downloads/doe-public-access-plan).
REFRENCES
1. Habib, K. M. and J.A Carrasco. Investigating the role of social networks in start time and
duration of activities. Transportation Research Record: Journal of the Transportation
Research Board, Vol. 2230, 2011, pp. 1-8.
2. Van den Berg, P., T. A. Arentze, and H. J. P. Timmermans. Factors influencing the planning
of social activities. Transportation Research Record: Journal of the Transportation Research
Board, Vol. 2157, 2010, pp. 63-70.
3. Lin, T. and D. Wang. Social networks and joint/solo activity-travel behavior. Transportation
Research Part A: Policy and Practice, Vol. 68, 2014, pp. 18-31.
4. Castiglione, J., M. Bradley, and J. Gliebe. Activity-Based Travel Demand Models: A Primer,
Transportation Research Board, Washington, D.C., 2015.
5. Granovetter, M. S.. The strength of weak ties. American Jouranl of Sociology, Vol. 78, No.
6, 1973, pp. 1360-1380.
6. Kadushin, C. Understanding Social Networks: Theories, Concepts, and Findings. Oxford
University Press, Oxford, 2012.
7. Carrasco, J. and B. Cid-Aguayo. Network capital, social networks, and travel: An empirical
illustration from Concepción, Chile. Environment and Planning A, Vol. 44, No. 5, 2012, pp.
1066-1084.
8. Tilahun, N. and M. Li. Geography of Close Contacts and Face-to-Face Meetings. Presented
at 94th Annual Meeting of the Transportation Research Board, Washington, D.C., 2015.
9. Sadri, A. M., S. Lee, and S. V. Ukkusuri. Modeling Social Network Influence on Joint Trip
Frequency for Regular Activity Travel Decisions. Transportation Research Record: Journal
of the Transportation Research Board, Vol. 2495, 2015, pp. 83-93.
10. Kadushin, C., The motivational foundation of social networks. Social Networks, Vol. 24, No.
1, 2002, pp. 77-91.
11. Lin, N., Y. C. Fu, and R. M. Hsung. The position generator: Measurement techniques for
investigations of social capital. In: Social Capital: Theory and Research (N. Lin, K. Cook,
and R. S. Burt, eds.), Aldine de Gruyter, New York, 2001, pp. 57-81.
12. Lin, N. and M. Dumin. Access to occupations through social ties. Social Networks, Vol. 8,
No. 4, 1986, pp. 365-385.
13. Hampton, K. N., L. F. Sessions, E. J. Her, and L. Rainie. Social Isolation and New
Technology. 2009. www.pewinternet.org/2009/11/04/social-isolation-and-new-technology/.
Accessed Jan. 31, 2015.
Maness 15
14. Hampton, K. N. Comparing bonding and bridging ties for democratic engagement: Everyday
use of communication technologies within social networks for civic and civil behaviors.
Information, Comminication, and Society, Vol. 14, No. 4, 2011, pp. 510-528.
15. Treiman, D. J. Occupational Prestige in Comparative Perspective. Academic Press, New
York, 1977.
16. Ganzeboom, H. and D. Treiman. Three internationally standardised measures for
comparative research on occupational status. Advances in Cross-National Comparison (J. H.
Hoffmeyer-Zlotnik and C. Wolf, eds.), Kluwer Academic / Plenum Publisher, New York,
2003, pp. 159-193.
17. Wooldridge, J. M. Econometric Analysis of Cross Sectional and Panel Data. 2nd ed., MIT
Press, Cambridge, MA and London, 2010.
18. Venables, W. N. and B. D. Ripley. Modern Applied Statistics with S. 4th ed. Springer, New
York, 2002.
19. Cameron, A. C. and P. K. Trivedi. Regression-based tests for overdispersion in the Poisson
model. Journal of Econometrics, Vol. 46, No. 3, 1990, pp. 347-364.
20. Zeileis, A. and T. Hothorn. Diagnostic checking in regression relationships. R News, Vol. 2,
No. 3, 2002, pp. 7-10.
21. Zeileis, A.. Object-oriented computation of sandwich estimators. Journal of Statistical
Software, Vol. 16, No. 9, 2006, pp. 1-16.
22. Maness, M. Choice Modeling Perspectives on Social Networks, Social Influence, and Social
Capital in Activity and Travel Behavior. Ph.D. dissertation. Department of Civil and
Environmental Engineering, University of Maryland, College Park, MD, 2015.
23. Li, Y., M. Savage, and A. Warde. Social mobility and social capital in contemporary Britain.
The British Journal of Sociology, Vol. 59, No. 3, 2008, pp. 391-411.
24. Magee, M. P. Civic Participation and Social Capital: A Social Network Analysis in Two
American Counties. Social Capital: An International Research Program (L. Nan and B. H.
Erickson, eds.), Oxford University Press, Oxford, 2008, pp. 308-27.
25. Johnson, C. A. Access to Social Capital and the Structure of Inequality in Ulaanbaatar,
Mongolia. Social Capital: An International Research Program (L. Nan and B. H. Erickson,
eds.), Oxford University Press, Oxford, 2008, pp. 380-93.
26. Holland, J. L. A theory of vocational choice. Journal of Counseling Psychology, Vol. 6, No.
1, 1959, pp. 35-45.
27. Nauta, M. M. The development, evolution, and status of Holland’s theory of vocational
personalities: Reflections and future directions for counseling psychology. Journal of
Counseling Psychology, Vol. 57, No. 1, 2010, pp. 11-22.
28. Larson, L. M., P. J. Rottinghaus, and F. H. Borgen. Meta-analyses of Big Six interests and
Big Five personality factors. Journal of Vocational Behavior, Vol. 61, No. 2, 2002, pp. 217-
239.
29. Barrick, M. R., M. K. Mount, and R. Gupta. Meta‐analysis of the relationship between the
five‐factor model of personality and Holland's occupational types. Personnel Psychology,
Vol. 53, No. 1, 2003,pp. 45-74.
30. Furnham, A. Personality and activity preference. British Journal of Social Psychology, Vol.
20, No. 1, 1981, pp. 57-68.
Maness 16
TABLE 1 Summary of Personal Networks and Community Survey Methodology
Time Frame
July and August 2008
Target Population
Noninstitutionalized adults living in the United States, aged 18 and older
Sampling Frame
Households with landline phones and individuals with cellular phones
Sample Design
Random digit dialing of landline and cellular phones
Sample Size
2,512 adults
Response Rate
21% (landline), 22% (cellular phone)
Use of Interviewer
Interviewer administered
Mode of Administration
Phone interview
Computer Assistance
None by respondents
Reporting Unit
One person aged 18 or older per household reports for him/herself and the
entire household (landline), one person aged 18 or older reports for
him/herself (cellular phone)
Time Dimension
Cross-sectional survey
Frequency
One two-month phase of collecting responses
Levels of Observation
Person, Household
Note: Statistics come from (13) and (14).
Maness 17
TABLE 2 OLS Estimates of Logarithm of Occupational Diversity
Parameter Name
Estimate
Std. Error
Intercept
-0.035
0.030
Age / 100
0.734
0.119
Age2 / 100
-0.700
0.115
Education
0.014
0.002
Black
0.038
0.011
Married or Living with Partner
0.027
0.007
Internet User
0.037
0.010
Social Networking Site User
0.035
0.010
Frequent Home Internet User
0.053
0.009
Respondent Knows Some Neighbors
0.057
0.009
Core Network Size
0.012
0.002
R2
0.212
F-Statistic
50.6
Maness 18
TABLE 3 Activity Variety Models Estimations
Parameter Name
Models with Social Capital
Non-Social Model
Assumes Exogeneity
Handles Endogeneity
Estimate
Std. Error
Estimate
Std. Error
Estimate
Std. Error
Intercept
0.438***
0.140
0.054
0.141
0.003
0.140
Individual Characteristics:
Female Respondent
-0.004
0.023
-0.027
0.021
-0.033
0.021
Age
0.004
0.004
-0.002
0.004
-0.009**
0.004
Age2
0.000
0.000
0.000
0.000
0.000
0.000
Hispanic
-0.047
0.047
-0.040
0.045
-0.044
0.044
White
--
--
--
--
--
--
Black
-0.014
0.039
-0.044
0.034
-0.075**
0.035
Asian
0.043
0.072
0.147**
0.068
0.151**
0.067
Native American
-0.122
0.114
-0.190*
0.103
-0.181*
0.102
Other Race
0.046
0.069
0.039
0.061
0.042
0.061
Married
--
--
--
--
--
--
Living with Partner
0.126***
0.048
0.119**
0.046
0.131***
0.046
Divorced
-0.005
0.041
-0.007
0.037
0.026
0.038
Separated
-0.043
0.100
-0.030
0.087
0.002
0.087
Widowed
-0.010
0.053
-0.011
0.048
0.025
0.049
Never been Married
0.041
0.040
0.064*
0.036
0.095***
0.037
Single
-0.316***
0.105
-0.265**
0.104
-0.228**
0.103
Employed Full-time
--
--
--
--
--
--
Employed Part-time
0.083**
0.036
0.084**
0.034
0.093***
0.034
Retired
0.000
0.040
0.047
0.036
0.070*
0.037
Not Employed
-0.095***
0.037
-0.040
0.033
-0.023
0.033
Disabled
-0.324***
0.104
-0.276***
0.092
-0.250***
0.092
Student
0.079
0.126
0.135
0.092
0.136
0.094
Other Employment Status
-0.095
0.106
-0.079
0.109
-0.051
0.107
Education
0.096***
0.008
0.063***
0.007
0.042***
0.009
Household Characteristics:
Number of Household Kids
0.010
0.013
0.007
0.012
0.007
0.012
Number of Household Adults
0.003
0.019
0.004
0.018
0.006
0.018
ln(Income)
0.067***
0.022
0.040**
0.020
0.031
0.020
Income Missing
0.189**
0.090
0.116
0.080
0.087
0.080
Income Above $100k
-0.030
0.030
-0.041
0.028
-0.042
0.028
Detached House
--
--
--
--
--
--
Apartment
0.044
0.035
0.067**
0.031
0.077**
0.031
Townhouse
0.058
0.045
0.073*
0.043
0.066
0.043
Other Home Type
-0.098**
0.050
-0.075*
0.045
-0.068
0.045
Social Network Characteristics:
ln(Core Network Size)
--
--
0.099***
0.017
0.056***
0.020
ln(Diversity/22 + 1)
--
--
0.949***
0.091
1.953***
0.274
Upper Reachability / 100
--
--
0.442***
0.115
0.445***
0.114
Diversity Residuals
--
--
--
--
-1.061***
0.274
Log Likelihood
3485
3367
3362
AIC
7030
6801
6794
Number of Parameters
30
33
34
Note: *** denotes estimate p-value ≤ 0.01; ** denotes estimate p-value > 0.01 and ≤ 0.05;
* denotes estimate p-value > 0.10 and < 0.05; -- denotes parameter fixed to 0 in that model specification
Maness 19
TABLE 4 Activity Frequency Models Estimations
Models with Social Capital
Non-Social Model
Assumes Exogeneity
Handles Endogeneity
Parameter Name
Estimate
Std. Error
Estimate
Std. Error
Estimate
Std. Error
Intercept
1.564***
0.181
1.117***
0.183
1.051***
0.182
Individual Characteristics:
Female Respondent
-0.020
0.030
-0.043
0.027
-0.052*
0.027
Age / 100
-0.209
0.549
-1.054**
0.507
-1.940***
0.559
Age2 / 100
0.000
0.005
0.008
0.005
1.728***
0.006
Hispanic
-0.029
0.062
-0.022
0.056
-0.015
0.056
White
--
--
--
--
--
--
Black
-0.003
0.049
-0.043
0.045
-0.081*
0.045
Asian
0.080
0.100
0.198**
0.092
0.202**
0.090
Native American
-0.014
0.134
-0.096
0.121
-0.084
0.122
Other Race
-0.026
0.093
-0.032
0.080
-0.028
0.079
Married
--
--
--
--
--
--
Living with Partner
0.188***
0.060
0.183***
0.057
0.198***
0.057
Divorced
0.006
0.051
0.006
0.047
0.047
0.048
Separated
0.033
0.131
0.049
0.113
0.088
0.110
Widowed
-0.025
0.069
-0.022
0.063
0.023
0.065
Never been Married
0.131**
0.051
0.161***
0.047
0.200***
0.048
Single
-0.361**
0.150
-0.297**
0.150
-0.250*
0.149
Employed Full-time
--
--
--
--
--
--
Employed Part-time
0.126***
0.045
0.133***
0.042
0.145***
0.042
Retired
-0.004
0.048
0.060
0.044
0.090**
0.045
Not Employed
-0.130***
0.049
-0.059
0.044
-0.037
0.045
Disabled
-0.262**
0.134
-0.192
0.120
-0.161
0.119
Student
-0.185
0.150
-0.114
0.114
-0.111
0.116
Other Employment Status
0.008
0.157
0.044
0.169
0.081
0.165
Education
0.102***
0.010
0.064***
0.009
0.038***
0.012
Household Characteristics:
Number of Household Kids
0.021
0.016
0.018
0.015
0.018
0.015
Number of Household Adults
0.022
0.024
0.024
0.023
0.026
0.022
ln(Income)
0.093***
0.029
0.061**
0.026
0.050*
0.027
Income Missing
0.253**
0.117
0.166
0.106
0.129
0.105
Income Above $100k
0.021
0.040
0.008
0.037
0.007
0.037
Detached House
--
--
--
--
--
--
Apartment
0.086*
0.044
0.112***
0.041
0.124***
0.041
Townhouse
0.135**
0.059
0.153***
0.055
0.145***
0.055
Other Home Type
-0.085
0.065
-0.059
0.060
-0.052
0.060
Social Network Characteristics:
ln(Core Network Size)
--
--
0.092***
0.022
0.039
0.026
ln(Diversity/22 + 1)
--
--
1.274***
0.118
2.521***
0.359
Upper Reachability / 100
--
--
0.484***
0.147
0.488***
0.147
Diversity Residuals
--
--
--
--
-1.318***
0.356
Log Likelihood
7608
7010
6987
AIC
13705
12858
12832
Number of Parameters
30
33
34
Note: *** denotes estimate p-value ≤ 0.01; ** denotes estimate p-value > 0.01 and ≤ 0.05;
* denotes estimate p-value > 0.10 and < 0.05; -- denote parameter not used in that model specification
Maness 20
FIGURE 1 Distribution of Weak Ties Occupational Diversity (left) and Strong Ties
Quantity (right)
