You are on page 1of 34

American http://asr.sagepub.

com/
Sociological Review

Religion and Volunteering in Context: Disentangling the Contextual Effects of


Religion on Voluntary Behavior
Chaeyoon Lim and Carol Ann MacGregor
American Sociological Review 2012 77: 747
DOI: 10.1177/0003122412457875
The online version of this article can be found at:
http://asr.sagepub.com/content/77/5/747

Published by:
http://www.sagepublications.com

On behalf of:

American Sociological Association

Additional services and information for American Sociological Review can be found at:
Email Alerts: http://asr.sagepub.com/cgi/alerts
Subscriptions: http://asr.sagepub.com/subscriptions
Reprints: http://www.sagepub.com/journalsReprints.nav
Permissions: http://www.sagepub.com/journalsPermissions.nav

>> Version of Record - Sep 30, 2012


What is This?

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

457875
rican Sociological ReviewLim and MacGregor
2012

ASRXXX10.1177/0003122412457875Ame

Religion and Volunteering in


Context: Disentangling the
Contextual Effects of Religion
on Voluntary Behavior

American Sociological Review


77(5) 747779
American Sociological
Association 2012
DOI: 10.1177/0003122412457875
http://asr.sagepub.com

Chaeyoon Lima and Carol Ann MacGregorb

Abstract
This study examines whether religions effect on volunteering spills over to nonreligious
individuals through personal ties between religious and nonreligious individuals. We use
three different analytic strategies that focus on national, local, and personal network level
contexts to identify the network spillover effect of religion on volunteering. We find that
if nonreligious people have close friends with religious affiliations, they are more likely to
volunteer for religious and nonreligious causes. However, this network spillover effect cannot
be inferred from the relationship between volunteering and national or local level religious
contexta common approach in the literature. In fact, we find that the average level of local
religious participation is negatively associated with volunteering among the nonreligious in
the United States. This novel finding suggests that to fully understand religions civic role
in the wider community, we need to consider how religion might influence the civic life of
people outside religious communities, not just those within them. Our findings also suggest
that in spite of methodological advances, studies that purport to test mechanisms at one level
of analysis by using data at a larger level of aggregation run a high risk of committing an
ecological fallacy.

Keywords
contextual effects, network spillover, religion, volunteering

Numerous studies have documented a positive


relationship between various measures of religion as an independent variable and social and
civic outcomes such as philanthropic giving,
community group membership, and volunteering (e.g., Lam 2002, 2006; Wuthnow 1991).
Although these studies disagree about the specific mechanisms driving this relationship,
most suggest it is religions community, not
conviction, aspect that is most influential (e.g.,
Putnam and Campbell 2010). More recently,
scholars have built on this notion of religious
community and asked whether religion might
have spillover effects on the likelihood of volunteering among nonreligious people. These

studies ask whether secular people are more


likely to volunteer when they live in a highly
religious environment, in that they are likely to
be connected to religious individuals and their
communities through personal networks (Borgonovi 2008; Ruiter and De Graaf 2006; Traunmller 2009). In addressing this question, a
a

University of Wisconsin-Madison
Loyola University New Orleans

Corresponding Author:
Chaeyoon Lim, University of WisconsinMadison, 2446 Sewell Social Science Building,
1180 Observatory Drive, Madison, WI 53706
E-mail: clim@ssc.wisc.edu

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

748

American Sociological Review 77(5)

larger theoretical concern is at stake given


continuing debates about religions civic role,
especially whether religion is a dividing or
uniting force in civil society (Lichterman 2005;
Putnam and Campbell 2010). If religions
influence on civic engagement indeed spills
over to nonreligious individuals and encourages their participation in communities, it
would suggest a mechanism through which
religion could promote civic engagement and
thus serve as a uniting force in civil society.
This hypothesis is all the more relevant given
the growing number of nonreligious people in
the United States and the growing gaps
between the religious and nonreligious in political and social attitudes (Hout and Fischer
2002; Putnam and Campbell 2010).
Despite the significance of the issue, only a
few studies have examined effects of religious
contexts on nonreligious peoples civic behaviors. Moreover, data limitations have prevented
direct examination of religions potential network spillover effect. As a result, previous
studies have used a more general, contextual
analysis strategy and considered the effect of
average religiosity in a country or region on
the likelihood of an individual volunteering.
The result is a mismatch between the proposed causal mechanism (influence through
personal networks) and the empirical evidence
(effect of collective religiosity at an aggregated level). In this study, we attempt to
address this issue by looking at cross-national,
local, and personal network level contexts. In
doing so, our article offers the most comprehensive test of the network spillover effect to
date and a more thorough examination of
religions contextual effects on volunteering.
More specifically, our analysis consists of
three separate studies that examine the network spillover hypothesis. First, we revisit
Ruiter and De Graafs pioneering 2006 study
that uses cross-national survey data to articulate the mechanisms underlying religious
contextual effects and documents the relationship between religiosity in a country and
an individuals volunteering behavior. We use
a new comprehensive cross-national dataset
that includes almost 140 countries to examine
the relationship between national religious

contexts and volunteering. Contrary to Ruiter


and De Graaf (2006), we find little evidence
that individual religiosity matters less in
devout nations due to network spillover. Second, we examine spillover effects by measuring contextual religiosity at the local level
(i.e., county) in the United States. This is a
more proximate measure of religiosity in
individuals environments than average church
attendance in a country. Using another new
dataset that includes a sufficient number of
respondents to represent most of the populous
counties in the United States, we show that
nonreligious people in counties with a higher
level of average religiosity are less, not more,
likely to volunteer than their counterparts in
secular areas. This is opposite what the network spillover hypothesis predicts. Finally,
we examine the network spillover hypothesis
more directly by measuring contextual religiosity at the level of an individuals personal
networks. We ask whether nonreligious people are more likely to volunteer when they
report having a friend with a religious affiliation. Our findings suggest that people who are
not involved in religious congregations are
indeed more likely to volunteer when they
have religious friends.
Findings from these three studies offer not
only the most comprehensive test for religions
network spillover effect on volunteering, but
also the first systematic examination of how
different levels of religious context might be
related to an individuals civic behavior. As a
result, this study contributes to our understanding of how religion can shape the civic life of
the broader community. More generally, our
findings provide important lessons for researchers interested in the study of contextual effects.
Results highlight that scholars should carefully
consider how the level of aggregation may
influence the relationship between explanatory
and outcome variables.

Religion and
Volunteering
Considerable scholarly attention has been
paid to the relationship between religion and
various pro-social and civic outcomes1

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

Lim and MacGregor

749

(Ellison 1992; Hall 2005; Regnerus, Smith,


and Sikkink 1998). Evidence from a variety
of surveys conducted in different times and
places overwhelmingly suggests that religious individuals are more active volunteers
and community participants than their secular
counterparts (e.g., Lam 2002, 2006; Loveland
et al. 2005; Park and Smith 2000; Putnam and
Campbell 2010). Discussion of the mechanisms that link religion to volunteering highlights two important components of religiosity,
sometimes referred to as belief and belonging
(Bellah 1991) or conviction and community
(Wuthnow 1991). All major religions preach
some form of selflessness and place value on
helping others (Ellison 1992), but these convictions are primarily fostered in the context
of religious communities. Some scholars even
suggest that beliefs hold no power if they are
not ratified by a community of fellow believers (Stark and Bainbridge 1996; Wuthnow
1991). Consequently, most scholars examining this relationship focus on religious attendance as a key predictor. Congregations,
researchers suggest, are places where people
meet socially and with civically involved
friends and, as a result, religious attendance
increases the likelihood of learning about
volunteering opportunities or being asked to
volunteer (Musick, Wilson, and Bynum 2000;
Ruiter and De Graaf 2006). Additionally,
social ties, such as those established in faith
communities, help to generate trust, and trust
makes it easier for people to step forward and
donate their time (Wilson 2000).
Evidence suggests that congregational networks play a key role in recruiting religious
people into volunteerism (Cavendish 2000;
Park and Smith 2000; Putnam and Campbell
2010), but it is less well known whether
religiosity in ones surrounding environment
has an additional effect on volunteering over
and above the effects of ones own commitment and involvement. Given that people are
drawn into volunteerism through personal
networks, it seems logical to expect such contextual effects from religion, but few studies
directly examine this question. A less obvious
question, and one that has drawn even less
attention in the literature, is whether religious

contexts influence the volunteerism of people


who are not themselves involved in organized
religion. Does religions effect spillover to
people outside religious communities? Is
there any evidence of such spillover effects
through personal networks? These are the
questions we turn to now.

Contextual Effects of
Religion
Various structural and institutional factors
compose an individuals social environment,
but most contextual analyses in sociology
focus on effects of personal traits aggregated
over individuals in a geographic or organizational unit (Books and Prysby 1988; Stipak
and Hensler 1982). In the sociology of religion, religious context is often operationalized by aggregating personal religious
characteristics, most commonly religious
affiliation or frequency of church attendance.
Scholars have been interested in how religious contexts influence outcomes like suicide (Ellison, Burr, and McCall 1997),
teenage delinquency (Regnerus 2003; Stark
and Bainbridge 1996), and crime (Beyerlein
and Hipp 2005).
Most of these studies employ Rodney
Starks notion of moral communities (Stark
and Bainbridge 1996)a recapitulation of
Durkheims conceptions of social integration
and social regulation. Noting that the relationship between personal religiosity and teenage
delinquency varies across regions in the
United States, Stark and Bainbridge (1996:
164) suggest that teenagers personal religiousness reduces delinquency only when the
majority of their friends are religious and thus
religion enters freely into everyday interactions and becomes a valid part of the normative system. In other words, personal
religiosity must be ratified by ones social
environment to affect behavior. In this moral
communities thesis, Stark clearly invokes a
social interaction mechanism, which is the
most commonly proposed explanation of contextual analysis (Blalock 1984; Erbring and
Young 1979; Weatherford 1982). In this line
of argument, contextual effects are explained

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

750

American Sociological Review 77(5)

as a consequence of social interaction and


interpersonal influence among individuals
within a contextual unit. Aggregated personal
traits (e.g., average church attendance) serve
as proxies for personal networks and social
interactions.
Stark, however, also emphasizes that religion should be viewed as a group property,
not a personal trait. Regnerus (2003:527)
further develops this notion and argues that
religious context is not simply a set of personal relationships, but also an emergent
group property that can affect individuals
regardless of their status as immediate group
members. Viewed this way, contextual
effects might not require direct social interactions and instead might rely on indirect social
control and support, enabled by structural
features of social networks in a community,
such as network density (Regnerus 2003) or
network closure (Smith 2003). From this perspective, aggregated religiosity or affiliation
would represent structural features of networks in a contextual unit rather than serving
as a proxy for individuals ego-centric networks. It is also plausible that aggregated
religiosity is a proxy for global or institutional contextual factors, such as legal and
cultural institutions, rather than for social
interaction or macro network structure, especially when the contextual unit in question is
a country or large region (Books and Prysby
1988; Finke and Adamczyk 2008). In short,
although most studies of the contextual effects
of religion employ the moral communities
thesis, different theoretical mechanisms could
generate similar empirical contextual effects
of religion. It is difficult, however, to disentangle these mechanisms with a conventional
contextual analysis that uses aggregated personal religious characteristics, especially
when a large contextual unit is used.
Regardless of the mechanisms proposed,
previous studies focus almost exclusively on
how religious contexts affect religious individuals, what Regnerus (2003) calls the light
switch portion of the moral communities
thesis. Religious context matters, the light
switch hypothesis argues, because it turns
on the effect of personal religiosity. In other

words, it is only when personal religiosity is


combined with a high level of religiosity in
ones environment that religion influences
ones behavior. Consequently, effects of religious contexts should be concentrated among
individuals who are religious themselves, or
effects should be larger for religious than for
secular people. It remains unclear, however,
how nonreligious individuals might be
affected by religiosity in their environments.
In the next section, we discuss this question in
the context of civic engagement.

Religious Contexts and


Civic Engagement of the
Nonreligious
Only a few studies have examined contextual
effects of religion on civic engagement. Of
these, some have focused on religious affiliation and shown that the proportion of Protestants
in a country or region is positively related to
voluntary organization membership (Lam
2006; Woodberry 2012) and social trust
(Traunmller 2009). A few other studies feature
organizational context rather than national or
regional context (Schwadel 2005; Serow and
Dreyden 1990). Schwadel (2005) finds that
high levels of biblical literalism and friendships
within congregationscharacteristics often
associated with Evangelical Christianityare
negatively correlated with congregants
involvement in non-church organizations (see
also Beyerlein and Hipp 2005). Although these
studies offer rare insights into how religious
contexts influence voluntary behaviors, they
offer little understanding regarding the significance of religious contexts for people who are
not involved in organized religion.
Ruiter and De Graaf (2006) explicitly consider how religious contextmeasured as
average church attendance in a country
might affect the volunteering behavior of the
religious and nonreligious differently. They
suggest that devout countries have a higher
level of volunteering than secular countries,
not only because they have a larger number of
religious individuals who are embedded in
religious networks and thus volunteer more,

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

Lim and MacGregor

751

but also because in a highly religious context


even secular people will have active church
members among their social contacts who
will encourage them to volunteer. Due to this
spillover of religious influences via social
interaction between religious and secular
individuals, Ruiter and De Graaf predict personal religiosity will matter less in a highly
religious context than in a secular context.
They reason that in a highly religious context,
both religious and secular individuals will be
deeply embedded in religious social networks
and thus volunteer at a similar rate. In a secular context, however, nonreligious individuals
are less likely than religious individuals to
have religious social contacts and as a result,
they will have a lower level of volunteering.
Ruiter and De Graaf test this hypothesis using
the World Values Survey (WVS). Their analyses show that net of individual church attendance, people are more likely to volunteer in
devout countries. Moreover, as their spillover
hypothesis predicts, individual church attendance has a smaller effect, and perhaps no
effect at all, on volunteering in devout countries because nonreligious people volunteer as
much as the religious do, whereas it has a
significantly larger effect in secular countries.
Van der Meer, Te Grotenhuis, and Pelzer
(2010), however, challenge Ruiter and De
Graafs conclusions, showing that key findings of the study hinge on three African countries that are unusually religious and have an
exceptionally high rate of volunteering.
Ruiter and De Graaf (2010) reply to the criticism by replicating their key findings after
removing all potentially influential cases,2 but
this exchange suggests that the original findings are sensitive to which countries happen
to be included in the analysis. More importantly for our purposes, using average church
attendance in a country to test the network
spillover hypothesis assumes that the national
average is a reasonable proxy for religiosity
in individuals personal networks. This
assumption is common in contextual analyses
and is plausible considering that nonreligious
people should have a higher chance of meeting religious people when they are a small
minority in their communities (Blau and

Schwartz 1997; Olson and Perl 2011). Moreover, numerous studies in the United States
show that congregations often work closely
with secular nonprofit organizations to provide their members with opportunities to
engage in civic activities in the broader community, which makes the spillover from religious to secular more likely (Ammerman
2005; Chaves 2009).
Religiosity, however, is often not evenly
distributed geographically or socially within a
country; as a result, average religiosity in a
country could be a poor proxy for religiosity
in social networks. In fact, a recent study of
contextual effects of racial diversity on social
trust demonstrates that network effects are difficult to observe even when contextual unit is
as small as a two mile radius of a persons
residence (Dinesen and Snderskov 2011).
Even if religiosity is more or less evenly distributed, the premise that secular people in
devout countries or in highly religious local
communities would have to be more socially
engaged with religious people is questionable.
To be sure, a higher proportion of religious residents should increase the chances for the nonreligious to form ties with the religious, leading to
a lower level of baseline homophilythat is,
homophily created by the demography of the
potential tie pool (McPherson, Smith-Lovin,
and Cook 2001:419)among the nonreligious living in highly religious communities.
However, the propensity for inbreeding homophilyhomophily induced by social structures and personal preferencesmight be
stronger among the nonreligious in such communities. For example, studies have found
that people in racially diverse communities or
organizations show a stronger preference for
their own racial groups in friendship choices
than do individuals in less diverse environments (e.g., Moody 2001; Putnam 2007; Quillian and Campbell 2003). Similarly, religion
could be a more salient factor in social life in
highly religious countries and communities
and, as a result, we may see stronger inbreeding homophily based on religiosity in such
places. At least in the United States, some
evidence suggests that the level of segregation
by religious service attendance in social

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

752

American Sociological Review 77(5)

networks is on par with segregation by race


(DiPrete et al. 2010). Levels of segregation
might be higher in devout areas than in secular
ones, and thus spillover may be less likely to
take place where there is a high level of contextual religiosity. How the forces of baseline
and inbreeding homophily net out is an empirical question beyond the scope of this article,
but for now, it is sufficient to point out that
such conflicting tendencies would make it
even more difficult to identify network effects
with highly aggregated measures of contextual factors.
The network spillover hypothesis also presumes that interpersonal influence and volunteer recruitment through social networks
operate in much the same way between religious and nonreligious individuals as they do
among religious people. The social movement
literature, however, suggests that recruitment
efforts through personal networks often strategically target people who are more reachable
and have a higher propensity to participate
(Brady, Schlozman, and Verba 1999; Lim
2010). Because regular churchgoers can be
relatively easy to reach and tend to be active
volunteers, organizers may prioritize churchgoers in their recruitment efforts. This might
be particularly true in more religious communities where volunteering activities are likely
organized by religious organizations and
churchgoing potential volunteers are abundant. As a result, secular people living in
devout areas might be placed lower in organizers mobilization queues and are therefore
less likely to be asked to volunteer. These
individuals may also have a stronger incentive
to free-ride because they know their churchgoing neighbors will volunteer even if they do
not (Oliver 1984). The social movement literature also argues that interpersonal influence
tends to hinge on a shared identity that is relevant to the activism in question (Lim 2008;
McAdam and Paulsen 1993). This implies that
interpersonal influence might be more effective when individuals share a religious faith.
As a result, even when secular people are
contacted by religious acquaintances, they
might be less likely to respond positively than
would religious people.

Average church attendance of a country or


local area could also relate to individual volunteering through mechanisms other than network spillover. For example, countries with a
higher religious concentration may simply
have more organizations that provide volunteering opportunities, and thus nonreligious
people may have more opportunities to volunteer. Average church attendance could also
serve as a proxy for national religious culture,
which itself may influence individual behavior regardless of personal religiosity; this
could occur through public discourse, media,
and other social institutions, rather than
through interpersonal influence. Although this
could be called a contextual effect, it would be
a different mechanism than network spillover.
In short, Ruiter and De Graafs (2006)
study is illuminating, but evidence for the network spillover hypothesis appears to be inconclusive at best. Religiosity in highly aggregated
contextual units could be related to individuals volunteering behavior in ways that differ
from the mechanism suggested by the network
spillover hypothesis. Depending on which
mechanism prevails, it is plausible that religious contexts matter differently for the religious and the nonreligious. In certain contexts,
like the United States, contextual religiosity
may be unrelated or even negatively related to
volunteering among nonreligious individuals,
especially if effects of contextual religiosity
through other mechanisms run counter to network spillover effects or are stronger than
spillover effects. Given these potential tensions, the network spillover hypothesis needs
to be more carefully examined.

The Study
To study whether there is a network spillover
effect of religion on volunteering, we compare results from three different approaches
that measure individuals religious context
differently. We consider individuals religious
context at three different levels using three
different datasets. We use the Gallup World
Poll data to revisit Ruiter and De Graafs
(2006) key finding that nonreligious people in
devout countries are more likely to volunteer

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

Lim and MacGregor

753

than nonreligious people in secular countries,


due to the spillover effect. Second, we use the
Gallup Daily Poll data to examine the relationship between local religious contexts and
volunteering. If aggregated religiosity at the
country level predicts individuals volunteering due to the network spillover effect, we
would expect local level religiosity to have a
similar, or even stronger, relationship with
volunteering, because an individuals probability of having personal ties across religious
boundaries should be more closely related to
the religious composition of a local area than
that of an entire country. Finally, we use the
Faith Matters survey in the United States to
examine whether nonreligious people are
more likely to volunteer when they have close
friends who are religious.

Does Average Church Attendance in


a Country Predict Volunteering?
First, we examine whether national religious
context, measured as average church attendance in a country, predicts volunteering,
especially among the nonreligious. We use
Gallup World Poll data (henceforth World
Poll), a large survey series that Gallup has
conducted since 2005 in 156 countries (Gallup
Inc. 2011). Designed to be representative of
95 percent of the worlds adult population, the
World Poll has interviewed, either by phone
or face-to-face (the latter mostly in developing countries with telephone coverage lower
than 80 percent), at least 1,000 individuals
from each country, with a few exceptions
where the sample size is between 500 and
1,000 people. All interviews were conducted
using the major languages of each country,
based on nationally representative probability
samples (for more information on sampling
methods and other details of the surveys, see
Gallup Inc. [2011]). In most of the 156 countries surveyed, the survey has been conducted
more than once between 2005 and 2011 and,
as a result, the pooled sample size for each
country is often substantially larger than
1,000. Because of missing data, the final analytic sample includes 373,604 respondents
from 138 countries.3 The median sample size

per country in the analytic sample is 2,707.


The most important advantage of the World
Poll, for our purposes, is its extensive coveragenot only in terms of geography, but also
culturally and politically. Because these data
include a large number of countries, it is
unlikely that an observed relationship is
overly sensitive to inclusion or exclusion of a
few influential cases. In addition, as a result
of the large sample size in most countries,
national religious context can be estimated
reliably by aggregating individual responses
in each country (for the list of countries and
sample size in each country, see Part A, Table
S1 in the online supplement [http://asr.sage
pub.com/supplemental]).
The outcome variable in this study is
whether respondents volunteered their time to
any organization. To measure religious service
attendance, the World Poll asked whether
respondents had attended a place of worship
or a religious service within the past seven
days. We coded the volunteering and religious
service variables dichotomously. We adjust
for various individual- and country-level characteristics. At the individual level, we adjust
for age, gender, education, family status, and
religious preference. At the country level, control variables include gross national income
per capita, life expectancy, mean years of
schooling, democracy score, and religious culture (see Table A1 in the Appendix for additional information on these variables).
The key question is whether average
church attendance in a country predicts volunteering more strongly for individuals who
are not themselves religious than for people
who regularly attend religious services. We
examine this question using multilevel logistic regression, which allows individual- and
country-level covariates to predict an individuals volunteering. We start by fitting a
random-intercept model, which lets the intercept vary across countries. We then allow the
coefficient of individual religious service
attendance to vary across countries and examine the cross-level interaction between individual attendance and a countrys average
attendance to see whether average attendance
in a country is more or less strongly related to

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

754

American Sociological Review 77(5)

Figure 1. National Religious Context and Volunteering across Countries

volunteering for non-churchgoers than for


churchgoers. If the spillover effect proposed
by Ruiter and De Graaf exists, and if it is
identifiable using national religious context,
we should find that average attendance in a
country matters more for non-churchgoers
than for churchgoers. As a result, the gap in
volunteering between churchgoers and nonchurchgoers should be smaller in countries
with a higher level of average attendance.

Findings
Before looking at results of the multilevel
logistic regressions, we examine the bivariate
relationship between the proportion of respondents in a country who volunteered and average religious service attendance in that country
(i.e., the proportion of a countrys respondents
who attended religious services last week)
with a scatterplot (see Figure 1). Visually
examining the relationship is important
because Van der Meer and colleagues (2010)
show that the relationship could be sensitive to
a few influential cases. Figure 1 suggests that
although there are a few potential outliers, the
relationship does not hinge on a small number
of influential cases. However, Figure 1 also
indicates that the relationship between average

attendance and volunteering is curvilinear. The


rate of volunteering is higher among both secular and highly religious countries and lower
among moderately religious countries.
Table 1 presents results from multilevel
logistic regressions. Model 1 shows how
individual- and country-level religious service attendance is related to volunteering, net
of all control variables at both levels. Taking
into account the curvilinear relationship we
observe in Figure 1, we include a quadratic
term for the proportion of churchgoers in
each country. First, individual church attendance strongly predicts volunteering. Net of
individual and country characteristics, the
odds of volunteering for respondents who
attended religious services in the past week
are 1.87 times greater than the odds for those
who did not. This is similar to the difference
between respondents with an elementary education or less and those with at least four
years of education beyond high school.
In addition, Model 1 suggests that the proportion of churchgoers in a country is related
to volunteering. Consistent with what we
observe in Figure 1, the level of volunteering
is higher in secular and devout countries and
lower in moderately religious countries. In
Model 2, we add the interaction between

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

Lim and MacGregor

755

Table 1. Multilevel Logistic Regressions of Individual Volunteering on Individual- and


Country-Level Religious Service Attendance

Individual-Level Covariates
Age
Female
Elementary education or less
Secondary to three-year tertiary
education
Four years of tertiary education
or more
Married/domestic partner
Single (never married)
Divorced/separated
Widowed
Has children age 15 years or
younger
Catholic
Protestant
Orthodox
Islam/Muslim
Hinduism
Buddhism
Judaism
Secular/atheist/agnostic/no religion
Christian
Other religion
Religious service attendance (A)
Country-Level Covariates
Religious diversity
Religious culture: Protestant
Religious culture: Catholic
Religious culture: Orthodox
Religious culture: Muslim
Religious culture: Eastern

Model 1

Model 2

Model 3a

Model 4b

.001**
(.000)
.164***
(.009)

.001*
(.000)
.165***
(.009)

.001
(.001)
.123***
(.013)

.327***
(.011)
.634***
(.014)

.326***
(.011)
.634***
(.014)

.323***
(.018)
.641***
(.022)

.043***
(.011)
.023
(.018)
.213***
(.019)
.008*
(.004)

.044***
(.011)
.020
(.019)
.215***
(.019)
.008*
(.004)

.012
(.017)
.025
(.025)
.250***
(.030)
.013*
(.006)

.114***
(.015)
.109***
(.029)
.023
(.022)
.085
(.044)
.188***
(.034)
.118
(.085)
.106***
(.024)
.094***
(.023)
.054*
(.024)
.628***
(.010)

.110***
(.015)
.099**
(.029)
.016
(.022)
.074
(.045)
.204***
(.034)
.150
(.085)
.070**
(.025)
.091***
(.023)
.077**
(.025)
.570***
(.129)

.105***
(.026)
.010
(.040)
.019
(.041)
.044
(.082)
.191***
(.048)
.185
(.108)
.044
(.028)
.068
(.036)
.061*
(.031)

.002***
(.000)
.198***
(.012)

.336***
(.014)
.632***
(.020)

.065***
(.015)
.004
(.027)
.184***
(.025)
.005
(.005)

.114***
(.019)
.220***
(.045)
.030
(.027)
.077
(.053)
.221***
(.050)
.111
(.138)
.230***
(.065)
.113***
(.030)
.134**
(.045)

.179
(.172)
.791***
(.224)
.373
(.214)
.101
(.232)

.166
(.167)
.789***
(.217)
.438*
(.208)
.136
(.225)

.155
(.166)
.856***
(.217)
.498*
(.208)
.152
(.224)

.237
(.178)
.754**
(.233)
.400
(.221)
.200
(.240)
(continued)

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

756

American Sociological Review 77(5)

Table 1. (continued)
Model 1
Religious culture: other
Life expectancy at birth
Average years of schooling
Gross national income per capita
Democracy score
Proportion of churchgoers (B)
Proportion of churchgoers
squared (C)
A B

.312
(.288)
.011
(.010)
.052
(.034)
.000
(.000)
.021
(.040)
4.412***
(1.330)
4.281**
(1.378)

A C
Intercept

.294
(.634)

Level-2 var.: Intercept


Level-2 var.: (A)

.348

N (individual)
N (country)
Log-likelihood

373,604
138
180073

Model 2

Model 3a

Model 4b

.295
(.280)
.012
(.010)
.054
(.033)
.000
(.000)
.054
(.039)
5.069***
(1.294)
5.167***
(1.345)
.723
(.611)
1.115
(.625)
.043
(.617)

.303
(.278)
.012
(.010)
.054
(.032)
.000
(.000)
.065
(.038)
5.216***
(1.282)
5.321***
(1.333)

.373
(.298)
.010
(.010)
.041
(.035)
.000
(.000)
.012
(.041)
4.329**
(1.376)
4.223**
(1.423)

.139
(.613)

.154
(.654)

.067
.322

.315

.366

373,604
138
179812

187,161
138
80625

186,443
138
99227

Note: Logit coefficients with standard errors in parentheses.


a
Non-churchgoers only.
b
Churchgoers only.
*p < .05; ** p < .01; *** p < .001 (two-tailed tests).

individual attendance and proportion of


churchgoers in a country to examine whether
the gap between churchgoers and non-churchgoers is smaller in devout nations because
non-churchgoers in highly religious countries
volunteer more than non-churchgoers in secular countries. The interaction terms are not
statistically significant, suggesting that the
relationship between national context and volunteering does not vary significantly by individual attendance status. To make interpretation
of the interaction effect easier, we estimate
Model 1 separately for churchgoers and nonchurchgoers and calculate predicted probabilities of volunteering at varying levels of
average attendance in a country while fixing
other variables at their mean values. Results
are presented as Models 3 and 4 in Table 1 and

are also shown in Figure 1 as solid (churchgoers) and dashed (non-churchgoers) lines.
Results suggest that the curvilinear pattern
applies for churchgoers and non-churchgoers.
The gap between churchgoers and nonchurchgoers appears to become smaller among
the most religious countries, but only slightly.

Summary
Our analyses suggest that national religious
contexts do matter: there is a strong curvilinear
relationship between the proportion of churchgoers in a country and the level of volunteering.
The difference in volunteering between the most
secular countries (or the most devout countries)
and moderately religious countries, net of
individual- and country-level characteristics, is

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

Lim and MacGregor

757

larger than 10 percentage points. However, our


findings also suggest that contrary to what
Ruiter and De Graaf (2006) predict, national
context matters almost equally for churchgoers
and non-churchgoers. To be fair, measures of
religious service attendance and volunteering in
the World Poll do differ from those in the World
Values Survey, which could be responsible for
discrepancies between the two studies. However,
given that their findings are based on a smaller
number of countries and are highly dependent
on a handful of influential cases, we argue that
our findings, based on almost 140 countries,
carry more weight than findings from the WVS,
which includes one third as many countries.4
Given that national religious context matters regardless of personal religiosity, it seems
plausible that national religious culture, rather
than the network spillover effect, is responsible for the relationship between national context and volunteering. In addition, countries
on the left-hand side of Figure 1 tend to be
developed nations with democratic political
systems, whereas countries on the right are
mostly developing nations with less democratic or authoritarian political systems. This
suggests that the relationship between religion and volunteering may be conditioned by
a nations political system and civic culture in
addition to its religious culture. To fully
understand why national religious contexts
matter, it is necessary to explore these potential heterogeneities among countries. This
task, however, is beyond the scope of this
article. For our purposes, it is sufficient to
note that cross-national data offer little support for the network spillover hypothesis.

Does Average Church Attendance


in a Community Predict Individual
Volunteering?
Next, we turn to the religious context of local
areas within the United States. Using data
from a large survey collected by Gallup, we
examine whether average church attendance
in U.S. counties predicts religious and nonreligious individuals volunteering. One of the
difficulties of studying the network spillover
effect using national context arises from the

fact that individual religious context is measured at a highly aggregated level. To be sure,
average attendance in a local area could be a
coarse measure of an individuals religious
environment, and therefore analyses at the
county level may have similar issues as those
based on national context. Still, church attendance in a county should be a better approximation of an individuals religious
environment than average attendance in an
entire country. If a network spillover effect
exists and is identifiable with an aggregated
measure of contextual religiosity, we should
find a stronger relationship, or at least a similar positive relationship, between average
church attendance in a county and volunteering, especially for nonreligious individuals.
To examine how local religious contexts are
related to volunteering, we use data from the
Gallup Daily Poll (henceforth Daily Poll). The
Daily Poll interviews daily at least 1,000 U.S.
adults, age 18 years and older, using dual
frame (including landline and cell-phone users)
random-digit dial sampling (for more details of
the survey, see Gallup Inc. [2009]).5 Because
Gallup interviews 1,000 or more respondents
every day except for major holidays, the cumulative data between 2008 and 2011 contain
interviews from more than 1.2 million individuals, representing almost all populous geographical areas in the United States. This large
sample size offers a unique opportunity to
study effects of the local religious context on
volunteering. First, because the Daily Poll has
at least a few hundred respondents in most
populous counties, it can provide a relatively
reliable measure of religiosity for a large number of counties. For example, these data have
at least 200 respondents from 1,076 counties;
with a higher threshold of 500 respondents,
these data still cover more than 500 counties.
This is an important advantage because most
previous studies use indirect indicators, such
as congregational adherence rates reported by
denominations, to measure local level religiosity. Moreover, the Daily Poll includes a large
number of geographic units, which have substantial variations in terms of religiosity and
the level of volunteering. The large number of
areas represented in these data provides enough

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

758

American Sociological Review 77(5)

statistical power to adjust for various community-level characteristics in addition to individual traits and to examine the cross-level
interaction between individual- and countylevel attendance more reliablythe critical
test for the network spillover hypothesis.
Once again the key outcome variable is
volunteering. In 2008 and 2009, the Daily Poll
asked whether respondents had volunteered
time to an organization in the past month. We
coded responses dichotomously. The Daily
Poll also includes a measure of church attendance, which is measured in five categories
ranging from at least once a week to never.
We converted this to an interval variable by
approximating annual days of church attendance. Average church attendance in a county is
the mean of church attendance for all county
residents. We calculated average church
attendance only for counties with at least 200
respondents who answered the religious service attendance question.6
To ensure that the relationship between religious service attendance and volunteering is
not spurious, we adjust for various individualand county-level characteristics. Individuallevel controls include a number of demographic
variables and respondents religious preferences. We also include county-level factors
such as median household income, racial diversity, and poverty rate. Using the Religious
Congregation Membership Study (Jones et al.
2002), we include proportions of major religious groups (i.e., Evangelical Protestant,
Mainline Protestant, and Catholic) in a county
to examine whether religious compositions,
rather than average attendance, influences volunteering of the nonreligious. More information on all variables in this section is available
in Table A2 in the Appendix. Similar to Study
1, we use multilevel logistic regression to
examine the relationship between local religious contexts and individuals volunteering.

Findings
Model 1 in Table 2 shows results from the
random-intercept multilevel logistic regression.
Similar to what we find in Study 1, individuals

religious service attendance strongly predicts


volunteering. The coefficient suggests that net
of all covariates in Model 1, one additional
instance of church attendance per year is associated with a 2.8 percent increase in the odds of
volunteering. Average religious service attendance in a county also predicts volunteering,
but contrary to predictions of the network spillover hypothesis, the coefficient is negative. Net
of all individual and county characteristics in
Model 1, respondents in more religious counties are less, not more, likely to volunteer their
time for an organization.7 The negative relationship is counterintuitive, especially given the
positive association between attendance and
volunteering at the individual level.
Model 2 sheds some light on this puzzling
finding. Results suggest there is a significant
positive cross-level interaction between individual attendance and average attendance in a
county. To make the interaction effect more
interpretable, we calculate the predicted probability of volunteering, separately for weekly
churchgoers and non-churchgoers, at three different levels of average attendance in a county
with all other covariates fixed at their fullsample means. Figure 2 depicts these probabilities with their confidence intervals. The solid
line suggests there is no clear relationship
between local religious context and volunteering among weekly churchgoers. In other
words, the negative relationship we observe in
Model 1 is largely driven by sporadic or occasional churchgoers, and non-churchgoers in
particular, for whom there is a strong negative
relationship between contextual religiosity and
volunteering. The latter finding about nonchurchgoers directly contradicts the network
spillover hypothesis, which predicts a positive
relationship between contextual religiosity and
volunteering among the nonreligious. In addition, our results for weekly churchgoers are
inconsistent with the light-switch hypothesis.
According to this hypothesis, we should
observe a higher level of volunteering among
regular churchgoers in highly religious areas
than among those in secular areas.
One possible explanation for the patterns
in Figure 2 is that nonreligious people have

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

759

Individual-Level Covariates
Respondents age

Respondent is female

Respondents race (reference: White)


Black

Asian

Hispanic

Others

Marital status (reference: married/domestic partner)


Single (never married)

Separated/divorced

Widowed

Number of children under age 18 (reference: none)


One

Two

Three

Education (reference: high school degree)


Less than high school degree

Technical or vocational school/some college

College degree

Postgraduate education

.003***
(.000)
.214***
(.010)
.181***
(.022)
.464***
(.044)
.392***
(.023)
.037
(.029)
.101***
(.016)
.141***
(.016)
.177***
(.018)
.097***
(.017)
.256***
(.018)
.382***
(.023)
.376***
(.030)
.504***
(.015)
.777***
(.016)
1.133***
(.017)

(.010)
.180***
(.022)
.466***
(.044)
.394***
(.023)
.036
(.029)
.101***
(.016)
.142***
(.016)
.178***
(.018)
.096***
(.017)
.256***
(.018)
.381***
(.023)
.376***
(.030)
.504***
(.015)
.778***
(.016)
1.133***
(.017)

Model 2

.003***
(.000)
.214***

Model 1

(.016)
1.132***
(.017)

(.015)
.777***

.375***
(.030)
.504***

.097***
(.017)
.256***
(.018)
.382***
(.023)

(.016)
.177***
(.018)

.101***
(.016)
.141***

.180***
(.022)
.464***
(.044)
.392***
(.023)
.037
(.029)

(.010)

.003***
(.000)
.214***

Model 3

(.016)
1.132***
(.017)

(.015)
.777***

.374***
(.030)
.505***

.097***
(.017)
.256***
(.018)
.381***
(.023)

(.016)
.177***
(.018)

.102***
(.016)
.141***

.181***
(.022)
.464***
(.044)
.393***
(.023)
.038
(.029)

(.010)

.003***
(.000)
.214***

Model 4

(.027)
1.179***
(.028)

(.025)
.798***

.271***
(.051)
.522***

.096***
(.025)
.307***
(.027)
.432***
(.038)

(.023)
.134***
(.032)

.058*
(.024)
.123***

.172***
(.043)
.321***
(.062)
.205***
(.039)
.120**
(.043)

(.016)

.004***
(.001)
.295***

Model 5a

(.036)
1.093***
(.038)

(.034)
.821***

.333***
(.066)
.495***

.202***
(.035)
.329***
(.037)
.479***
(.046)

(.035)
.217***
(.042)

.036
(.037)
.092**

.187***
(.044)
.580***
(.099)
.411***
(.050)
.080
(.065)

(.023)

.000
(.001)
.213***

Model 6b

(continued)

(.026)
1.131***
(.028)

(.023)
.749***

.440***
(.042)
.499***

.031
(.030)
.168***
(.031)
.307***
(.036)

(.028)
.196***
(.027)

.211***
(.029)
.197***

.165***
(.033)
.624***
(.079)
.503***
(.035)
.106*
(.049)

(.017)

.004***
(.001)
.141***

Model 7c

Table 2. Multilevel Logistic Regressions of Volunteering on Individual- and County-Level Religious Service Attendance in the United States

760

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

.324***
(.014)
.114***
(.030)
.120***
(.015)
.029
(.032)
.045*
(.018)
.019***
(.001)

.326***
(.014)
.121***
(.030)
.121***
(.015)
.032
(.032)
.037*
(.018)
.028***
(.000)
.027**
(.009)
.002**
(.001)
.032
(.063)
.011**
(.004)
.006***
(.001)
.265***
(.065)

County-Level Covariates
Population density in county (logged)

% of population in urban area

Index of racial diversity

% of households in poverty in county

% of residents with college degree in county

Median household income in county (logged)

.029***
(.009)
.001*
(.001)
.031
(.063)
.011**
(.004)
.006***
(.001)
.266***
(.065)

.141***
(.018)
.266***
(.018)
.353***
(.019)
.163***
(.018)

Model 2

.142***
(.018)
.267***
(.018)
.355***
(.019)
.164***
(.018)

Model 1

Monthly household income (reference: less than $2,000)


$2,000 to $3,999

$4,000 to $7,499

$7,500 or higher

Dont know or refused to answer

Religious preference (reference: Protestant)


Catholic

Jewish

Other Christian

Other non-Christian

No religion

Religious service attendance (A)

Table 2. (continued)

.028**
(.009)
.002**
(.001)
.027
(.063)
.012***
(.004)
.005***
(.001)
.244***
(.072)

(.001)

(.018)
.020***

(.030)
.121***
(.015)
.028
(.032)
.046*

.326***
(.014)
.114***

.141***
(.018)
.265***
(.018)
.353***
(.019)
.163***
(.018)

Model 3

.028**
(.009)
.002**
(.001)
.021
(.062)
.009*
(.004)
.004***
(.001)
.213**
(.073)

(.001)

(.018)
.020***

(.030)
.121***
(.015)
.027
(.032)
.047**

.329***
(.014)
.113***

.141***
(.018)
.265***
(.018)
.353***
(.019)
.162***
(.018)

Model 4

.026*
(.012)
.002**
(.001)
.015
(.094)
.005
(.006)
.004*
(.002)
.197
(.108)

(.042)
.036
(.021)

(.038)
.052*
(.026)
.149***

.111***
(.023)
.150***

.110***
(.028)
.259***
(.028)
.337***
(.030)
.190***
(.029)

Model 5a

.032
(.017)
.001
(.001)
.050
(.120)
.014
(.007)
.006**
(.002)
.348*
(.141)

(.065)
.119***
(.031)
.001
(.073)
.101
(.085)

.283***
(.029)
.253***

.116**
(.040)
.248***
(.040)
.385***
(.043)
.231***
(.040)

Model 6b

(continued)

.027*
(.013)
.001
(.001)
.163
(.089)
.007
(.005)
.004*
(.002)
.077
(.108)

(.097)

(.023)
.227***
(.068)
.528***

.543***
(.021)
.084
(.089)
.181***

.181***
(.028)
.286***
(.029)
.362***
(.033)
.112***
(.027)

Model 7c

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

761

1,065
193,799
117563

1,065
193,799
117558

.006
.000

1.750*
(.784)

.127***
(.023)
.131***
(.022)
.130***
(.027)
.020***
(.002)
.004***
(.001)
.078*
(.031)
.068**
(.024)

Model 3

Note: Logit coefficients with standard errors in parentheses.


a
Non-churchgoers only (respondents who attend services seldom or never).
b
Occasional churchgoers only (respondents who attend services almost weekly or about once a week).
c
Weekly churchgoers only (respondents who attend services at least once a week).
*p < .05; ** p < .01; *** p < .001 (two-tailed tests).

1,065
193,799
117584

N (counties)
N (individuals)
Log-likelihood

.007
.000

1.894**
(.682)

1.699*
(.684)
.008
.000

.154***
(.022)
.157***
(.020)
.152***
(.026)
.022***
(.002)
.004***
(.001)

Model 2

.151***
(.022)
.150***
(.020)
.157***
(.026)
.014***
(.002)

Model 1

Level-2 variance (intercept)


Level-2 variance (religious service attendance)

Region (reference: South)


East

Midwest

West

Average service attendance in county (B)

A B

Religious organization density (logged)

Civic organization density (logged)

% Evangelical Protestant (logged)

% Mainline Protestant (logged)

% Catholic (logged)

Constant

Table 2. (continued)

1,065
193,799
117544

.006
.000

.008
(.033)
.083***
(.024)
.122***
(.028)
.018***
(.003)
.004***
(.001)
.029
(.044)
.046
(.025)
.089***
(.020)
.057*
(.023)
.000
(.013)
1.424
(.787)

Model 4

1,065
88,580
50523

1,065
38,127
24278

.007

.068
(.086)
.017
(.048)
.122**
(.038)
.036
(.044)
.018
(.025)
3.114*
(1.518)

.014
(.066)
.043
(.036)
.032
(.030)
.073*
(.035)
.028
(.020)
.901
(1.163)
.008

.021
(.063)
.151**
(.046)
.131*
(.052)
.014**
(.004)

Model 6b

.008
(.049)
.015
(.036)
.094*
(.039)
.020***
(.004)

Model 5a

1,065
67,092
42698

.005

.011
(.048)
.106**
(.035)
.148***
(.041)
.003
(.003)

.109
(.063)
.061
(.037)
.111***
(.029)
.075*
(.033)
.011
(.018)
1.237
(1.162)

Model 7c

762

American Sociological Review 77(5)

Figure 2. Local Religious Context and Volunteering: Regular Churchgoers and Nonchurchgoers

fewer volunteering opportunities in highly


religious areas, because in these places most
volunteering activities are organized through
churches and other religious organizations.
Although previous studies show that congregations frequently work with secular nonprofit organizations and government agencies
to provide social services (e.g., Ammerman
2005; Lichterman 2005), it is still plausible
that a strong presence of religion in the voluntary sector turns off nonreligious people. This
proposition is not easy to test without comprehensive data on types of volunteering
opportunities and organizations that provide
such opportunities in local areas. In Model 3,
we attempt to indirectly test this by including
densities of religious and nonreligious civic
organizations in a county. Both measures are
logged to normalize the distribution. Results
show that the number of civic organizations
does predict volunteering. Respondents in
a county with more civic organizations per
capita are more likely to volunteer. The number of religious organizations, on the other
hand, is negatively related to volunteering.
These patterns appear consistent with the
proposition about disparity in volunteering

opportunities between religious and secular


areas. However, adding these variables does
not affect either the coefficient of average
attendance in a county or that of the crosslevel interaction between individual attendance and average attendance.
Another possibility is that highly religious
areas are composed of large Evangelical Protestant populations. As previous studies show,
Evangelical Protestants are more likely to
volunteer for religion-related activities, but
not necessarily for nonreligious civic activities (Beyerlein and Hipp 2005; Wuthnow
1999; Wuthnow and Evans 2002). More
importantly, some studies suggest that Evangelical congregations tend to focus on
spreading the faith in their outreach activities and are less likely to collaborate with
nonreligious organizations than are Mainline
and Catholic congregations, which frequently
engage in activities serving the community
in collaboration with secular organizations
(Ammerman 2005:115). As a result, there
could be fewer bridging ties between religious and nonreligious people in the civic
sector and fewer volunteering opportunities
for the nonreligious in local areas with a large

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

Lim and MacGregor

763

Evangelical population. Similarly, public perception about the connection between Evangelical Protestants and right-wing politics
may make secular peoplewho tend to be
socially and politically liberalmore reluctant to join activities organized by religious
organizations and heavily attended by Evangelical Christians (Hout and Fischer 2002).
We test these possibilities by including the
number of Evangelical Protestants, Mainline
Protestants, and Catholics, each as a percentage of a countys total population.8 Consistent
with what previous studies suggest, a strong
presence of Evangelical Protestants is negatively associated with volunteering, whereas
the percentage of Mainline Protestants is positively related. Once these variables are
included, neither of the organization density
variables are statistically significant, indicating that relationships in Model 3 are explained
by a countys religious composition. However, adding these variables does not affect
the relationship between average attendance in
a county and volunteering, or the interaction
between individual and average attendance.
Although a countys religious composition is
related to volunteering in ways predicted by
previous studies, this does not explain why
nonreligious people in highly religious areas
are less likely to volunteer than the nonreligious in more secular areas.9
In Models 5 through 7, we estimate the
specification from Model 4 separately for
weekly churchgoers, occasional churchgoers
(i.e., attended services almost every week
or about once a month), and non-churchgoers (attended services seldom or never).
In addition to checking the robustness of the
cross-level interaction effect, this approach
allows us to investigate whether effects of
other religion variables at the county level
also vary by the level of individual attendance
without including a number of cross-level
interaction terms. As expected, average
church attendance in a county is negatively
related to volunteering among non-churchgoers (Model 5)10 but there is no relationship
among weekly churchgoers (Model 7). There
is also a strong negative relationship between

average attendance and volunteering among


occasional churchgoers, who attend religious
services almost every week or about once
a month (Model 6). In other words, it is not
just secular people in highly religious areas
who are less likely to volunteer. Even moderately religious people who do not attend a
religious service weekly are less likely to
volunteer if they live in a highly religious
area. This finding is puzzling, because we
would not expect occasional churchgoers to
be turned away by a strong presence of religion in the voluntary sector. We return to this
finding later in the article.
In addition, the percentage of Evangelical
Protestants is negatively related to volunteering among weekly and occasional churchgoers, but not among non-churchgoers. These
findings also seem strange, because we would
expect secular people, not regular churchgoers, to be more likely to turn away from volunteering activities when there is a strong
presence of Evangelical Protestants. We suspect these findings have to do with the compositional effect. Although we control for
individuals religious affiliation in the models, we do not distinguish Evangelical Protestants from Mainline Protestants because the
Daily Poll does not ask about specific denominations. As a result, the stronger negative
relationship between the percentage of Evangelical Protestants and volunteering for
churchgoers may capture cross county compositional differences among Protestants.

Summary
Using the Gallup Daily Poll data, we find that
average religious service attendance in a local
area is either unrelated or negatively related
to volunteering, depending on individuals
level of attendance. Contrary to what the network spillover hypothesis predicts, our findings suggest that nonreligious individuals
living in more religious areas are significantly
less likely to volunteer than are their counterparts in secular areas. Furthermore, occasional churchgoers are also less likely to
volunteer when they live in highly religious

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

764

American Sociological Review 77(5)

areas. Only weekly churchgoers volunteering


is unrelated to contextual religiosity. Although
not shown here, we find similar patterns with
the Social Capital Community Benchmark
Survey (SCCBS) data (Roper Center for
Public Opinion Research 2001). Our analyses
of these data suggest that the negative relationship between average attendance in a
local area and volunteering among the nonreligious applies to religious and nonreligious
types of volunteering.11
This study suggests that the negative relationship is not driven by a strong presence of
Evangelical Protestantism in an area. Even
though the number of Evangelical Protestants
in a county is negatively related to volunteering, it does not change the relationship between
average attendance and volunteering. Indeed,
the relationship between Evangelicalism and
volunteering is stronger among weekly churchgoers than among non-churchgoers. We also
find no evidence that the density of religious or
civic organizations in a county explains the
negative relationship between contextual religiosity and volunteering.
The findings in this section do, however,
directly challenge the spillover hypothesis,
which predicts that nonreligious people in
religious areas will volunteer at a higher rate
than their counterparts in secular communities. Our findings are also inconsistent with
the light-switch hypothesis, in that contextual
religiosity is not related to volunteering
among weekly churchgoers. We will return to
these findings about the role of local religious
context in our discussion. Next, however, we
consider the role of religious context measured as religiosity among an individuals
close ties.

Does Having Religious Friends


Predict Individuals Volunteering?
In the previous two sections, we attempted to
infer the existence of a network spillover
effect from correlations between national or
local religious contexts and individuals volunteering. This section tests the network spillover hypothesis more directly by looking at

the relationship between religious compositions of intimate social networks and volunteering. Using the Faith Matters surveys (FM),
we examine whether nonreligious individuals
are more likely to volunteer when they have
friends who are religious. If having religious
friends increases the likelihood that nonreligious people will volunteer, it will offer the
most direct evidence for the spillover effect.
Faith Matters is a nationwide panel study
conducted between 2006 and 2011 to examine
the connection between religion and civic life
in the United States (Putnam and Campbell
2010). The first wave of data was collected
through phone interviews with a large national
representative sample of respondents (N =
3,109).12 Respondents were contacted for the
second interview approximately a year later
and 61.6 percent (N = 1,915) of the original
respondents participated. In 2011, all original
respondents were contacted again for the third
interview. A total of 1,810 respondents were
reached and 1,685 of them (93 percent) participated in the survey, yielding a re-interview
rate of 54.2 percent.13 In addition, the 2011
study interviewed a fresh sample of respondents; when combined with the panel respondents and properly weighted, the full sample (N
= 2,646) is representative of the U.S. population in 2011. In this study, we use the 2006 and
2011 data for cross-sectional and panel analyses.14 This dataset is useful for examining the
network spillover hypothesis for several reasons. First, these data include measures of
religious compositions of respondents intimate social networks. Most relevant to this
study are questions that ask whether any of a
respondents closest friends belong to one of
the four major religious traditions: Mainline
Protestant, Evangelical Protestant, Catholic,
or other non-Christian religions. These questions were repeated in 2011. In addition, the
2011 survey asked whether any friends with
religious affiliations attended religious services regularly. As a result, we can examine
whether having churchgoing friends, rather
than only nominally religious friends,
increases the probability of volunteering for
secular people. Second, by using the panel

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

Lim and MacGregor

765

data, we can attempt to address the issue of


causality in a way that improves upon previous studies based on cross-sectional data.
Finally, a relatively large sample size, which
includes a substantial number of respondents
who are not religiously observant, is a great
advantage because we focus on individuals
who are not involved in organized religion.
The key outcome variable in this section is
whether respondents volunteered in the past
12 months. Because we are interested in
whether having religious friends increases the
probability of volunteering among individuals who are not regular members or participants of religious congregations, our analysis
focuses on respondents who report that they
do not attend religious services on a regular
basis. Ideally, our analysis would focus on
individuals who never attend religious services and claim no religious affiliation, but
this would limit the sample size to only a
couple hundred respondents. Instead, we use
a more generous threshold of about once or
twice a year to identify nonreligious respondents.15 Given our interest in volunteering
behaviors among people who are not regular
participants in congregations, the threshold of
once or twice a year seems reasonable.16
We control for various personal characteristics that may influence both friendship with
religious people and volunteering. In addition
to sociodemographic characteristics and personal religiosity, we control for the number of
close friends and the level of social engagement to ensure that the relationship between
religious friendship and volunteering is not
driven by a high level of social involvement
among nonreligious people with religious
friends. For a similar reason, we control for
the number of racial groups in which respondents have close friends. For nonreligious people, friendship with a religious person might
be considered a bridging friendship; therefore, the relationship between religious
friendship and volunteering could be due to
personal characteristics (e.g., a high level of
tolerance or civic mindedness) of people who
have more bridging friendship ties in general.
This is particularly a concern because our

measure of religious friendship counts the


number of religious traditions among which
respondents have close friends rather than the
total number of religious friends. In addition,
we use these panel data to examine whether
the change in the number of religions among
close friends between 2006 and 2011 predicts
volunteering in 2011, net of volunteering in
2006 and other individual characteristics (see
Table A3 in the Appendix for a complete list
of variables used in this section).

Findings
Table 3 presents results from logistic regression analyses predicting the likelihood of
volunteering in either 2006 or 2011. Model 1
shows how the number of religious traditions
in which respondents have close friends is
related to volunteering after controlling for
personal characteristics.17 Controlling for
respondents own religiosity and levels of
social involvement, individuals who never or
rarely attend a religious service but have
friends in more religious groups are significantly more likely to volunteer. When all
variables in Model 1 are set to their mean
values, 35.5 percent of people who have no
close friends in any of the four religious
groups are predicted to have volunteered in
the past year. In comparison, the model predicts that 42.6 percent of people with close
friends in one religious group and 64.5 percent of respondents with friends in all four
religious groups have volunteered.18
Even though our analyses in Model 1
focus on people who never or rarely attend
religious services, many of these respondents
do have religious preferences. One may therefore wonder whether having close friends in
ones own religious group is more influential
than having friends only in other religions.
One possibility is that bonding ties may be
more effective than bridging ties in recruiting
people for volunteer activities. Model 2
examines this question by comparing respondents who have friends in their own religious
groups with those who have friends only in
different religious groups. Because most

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

766

American Sociological Review 77(5)

Table 3. Logistic Regressions of Volunteering on Religious Friendship Networks (respondents who attend religious services once or twice a year or less)
Volunteering in 2006

Model 1

Age
.004

(.006)
Male
.283

(.160)
Race (reference = White)
Hispanic
.842

(.716)
Black
.137

(.337)
Other race
.515

(.590)
Education (years)
.149***

(.040)
Family income ($)
.243

(.124)
Married
.130

(.182)
Has children under age 18
.207

(.189)
Religious traditions (reference = no religion)
Catholic
.016

(.253)
Mainline Protestant
.116

(.277)
Evangelical Protestant
.099

(.265)
Black Protestant
.894

(.849)
Jewish
.443

(.475)
Mormon
.015

(.958)
Other non-Christian religions
.493

(.468)
Religiosity index
.111

(.117)
Region (reference = South)
Northeast
.243

(.226)
Central
.200

(.229)
West
.001

(.218)
Number of close friends
.047

(.030)
Social involvement index
.372**

(.122)

Volunteering in 2011

Model 2

Model 3

Model 4

Model 5

.002
(.006)
.290
(.160)

.016**
(.005)
.274
(.158)

.028**
(.009)
.230
(.217)

.028**
(.009)
.241
(.222)

.868
(.722)
.157
(.340)
.495
(.590)
.149***

.336
(.496)
.285
(.338)
.127
(.527)
.125***

(.040)
.245*
(.124)
.142
(.182)
.194
(.190)
.102
(.272)
.153
(.283)
.073
(.265)
.991
(.858)
.518
(.481)
.100
(.965)
.476
(.466)
.096
(.117)
.257
(.226)
.181
(.229)
.036
(.217)
.043
(.030)
.389**
(.122)

(.037)
.121
(.118)
.129
(.180)
.205
(.184)

.860
(1.056)
.642
(.554)
1.417
(.913)
.031
(.055)
.203
(.175)
.218
(.248)
.202
(.264)

.520
(1.064)
.726
(.576)
1.744
(.929)
.004
(.057)
.185
(.178)
.117
(.255)
.266
(.271)

.147
(.259)
.168
(.280)
.273
(.271)
.181
(.684)
.350
(.436)
1.730
(1.165)
.066
(.412)
.071
(.123)

.032
(.341)
.256
(.364)
.043
(.367)
1.373
(1.187)
.090
(.605)
1.617
(1.472)
.044
(.562)
.063
(.149)

.094
(.349)
.270
(.370)
.052
(.374)
1.310
(1.208)
.042
(.624)
1.810
(1.487)
.131
(.580)
.058
(.152)

.582**
(.219)
.416
(.213)
.688**
(.224)
.046
(.032)
.226
(.125)

.376
(.294)
.120
(.300)
.407
(.321)
.008
(.043)
.406*
(.173)

.451
(.304)
.166
(.309)
.447
(.329)
.021
(.044)
.477**
(.177)
(continued)

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

Lim and MacGregor

767

Table 3. (continued)
Volunteering in 2006

Model 1

Model 2

Number of races among friends


.380***
(2006)
(.097)
Number of religions among friends
.319***
(2006)
(.089)
Friends in own versus other religions
(2006)
No friends with religion
Friends only in own religion

Friends only in a different religion

Friends in two or more religions

Number of religions among churchgoing friends (2011)


Change in number of religions
among friends (2011 to 2006)
Volunteering in 2006

Number of types of group involved


in 2006
Change in number of types of group
involved (2011 to 2006)

Constant
5.371***

(1.253)

.408***
(.096)

Number of observations
Log-likelihood
Pseudo R2

778
473.2
.114

Volunteering in 2011
Model 3
.215*
(.094)

Model 4
.108
(.136)
.105
(.152)

(omitted)
.673*
(.340)
.553*
(.257)
.832***
(.246)

5.576***
(1.253)
778
473.8
.113

.170*
(.086)

3.354**
(1.178)
788
495.6
.093

Model 5
.106
(.139)
.098
(.156)

.307**
(.112)
1.514***

.233*
(.115)
1.462***

(.227)
.367***
(.090)

(.233)
.665***
(.117)
.457***
(.105)

2.872
(1.779)
510
280.3
.207

2.456
(1.818)
510
270.3
.235

Note: Logit coefficients with standard errors in parentheses.


*p < .05; ** p < .01; *** p < .001 (two-tailed tests).

respondents who have friends in more than


one religious group have both bonding and
bridging ties, the most meaningful comparison would be between respondents who have
friends in only one religious group. Model 2
shows there is not much difference between
respondents with only bonding ties and those
with only bridging ties in terms of their probability of volunteering. Respondents who
have friends in more than one religious group
are more likely to volunteer than those with
friends in one religious group, whether or not
that religion is same as their own.19

An important caveat of the analyses so far


is that the religious friendship variable is
based on friends religious affiliations, rather
than how religiously observant they are.
Given the proposed mechanism of the network spillover effect, it is critical to know
whether close friends with religious affiliations actually attend religious services regularly. The 2011 FM survey asked respondents
whether any of their close friends in each
religion attend religious services regularly. In
Model 3, we include the number of religious
traditions in which respondents have at least

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

768

American Sociological Review 77(5)

one close friend who is a regular churchgoer.


The coefficient for the variable is positive and
statistically significant, suggesting that people who have churchgoing friends in more
religious traditions are more likely to volunteer than those who do not.20
Although adjusting for various individual
traits offers some assurance against the issue
of unobserved heterogeneity, we still cannot
rule out the possibility that people with religious friends differ from those without in
ways that matter for volunteering. Another
concern is that a nonreligious person may
befriend religious people through, rather than
prior to, their volunteer work. In Models 4 and
5, we attempt to address these issues using the
panel sample in the 2006 and 2011 FM data.
Model 4 examines whether the change
between the two surveys in the number of
religions among which respondents have close
friends predicts volunteering in 2011, net of
previous volunteering, religious friendship,
and other personal characteristics in 2006.
Results suggest that respondents who have a
higher value on the religious friendship variable in 2011 than they did in 2006 are significantly more likely to volunteer. Model 5 adds
a variable that measures the change in civic
group involvement between the two surveys,
to examine whether this makes the relationship between the religious friendship variable
and volunteering disappear. Gaining more
religious friends might lead to an increase in
group involvement, and thus we may underestimate the effect of religious friendship on
volunteering by including this variable in the
model. As a result, this model offers a conservative test for religious friendships effect
on volunteering. Results suggest that the
change in the religious friendship variable still
predicts volunteering in 2011, net of group
involvement in 2006 and the change in group
involvement between the two surveys, which
also predict volunteering significantly.

Summary
Findings in this section support the network
spillover hypothesis that religions effect on

volunteering can spread through personal


networks to nonreligious people. Our analyses show that people who rarely or never
attend religious services themselves are significantly more likely to volunteer if they
have religious friends. The relationship holds
in cross-sectional and longitudinal analyses.
However, our findings in the earlier sections
show that this spillover effect cannot be identified by looking at the relationship between
aggregated religiosity, in the community or in
the country, and individual volunteering.

Conclusions
When considered together, results of these
analyses advance our understanding of how
religions influence may reach beyond the
boundaries of religious congregations. In particular, we focus on how religions effect on
volunteering may spillover to nonreligious
individuals through personal ties between the
religious and the nonreligious. Our findings
suggest that nonreligious individuals who do
not regularly participate in religious congregations but have religiously observant friends
are more likely to volunteer. However, we
find that national and local religious contexts
are related to individuals volunteering in
ways that contradict predictions made by the
network spillover and light-switch hypotheses. Our analyses of the Gallup World Poll
suggest there is a strong U-shaped relationship between national religious context and
volunteering, regardless of an individuals
own religiosity. These findings contradict
previous studies, especially Ruiter and De
Graaf (2006), that argue that national religious context positively affects volunteering
among nonreligious individuals through network spillover. Moreover, we find that local
religious context, at least in the United States,
is negatively related to volunteering among
individuals who do not participate in a congregation regularly, whereas there is no relationship among regular churchgoers.
Our findings concerning national religious
contexts suggest researchers cannot assume
that religion will be related to civic participa-

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

Lim and MacGregor

769

tion in a similar way across countries.


Researchers need to more seriously consider
potential heterogeneity in this relationship. In
particular, we need to pay more attention to
how religions civic role may vary depending
on religious cultures and political systems.
Future research should also consider whether
volunteering itself has the same meaning in
different countries. More generally, our study
reminds researchers that any cross-national
study based on a nonsystematic sample of
countries needs to be cautious when generalizing findings beyond countries in the dataset.
This is particularly pertinent when a study
examines the relationship between variables
at the country level, as findings could be
largely dependent on which countries happen
to be in the data or on a small number of
influential cases. Finally, findings at the
national level, especially in light of evidence
for the network spillover effect at the individual level, suggest we cannot assume that
social interactions between the religious and
nonreligious are determined by national religious contexts. Religiosity is often unevenly
distributed geographically and socially within
a country. National religious contexts could
affect individual outcomes through other
mechanisms; as a result, making an inference
about network effects based on nationally
aggregated data could be misleading.
Our analysis at the county level in the
United States indicates that researchers should
be cautious about inferring network effects
even when using local level data. Our findings
at the county level contradict predictions from
network-based accounts of religious contextual effects, both the network spillover and the
light-switch hypotheses. How do we explain
this negative association between local religious context and volunteering among people
who do not attend religious services regularly,
especially in light of the finding that nonchurchgoers are more likely to volunteer when
they have religious friends? Earlier we discussed several different possibilities. Although
we cannot fully explore these potentialities
due to constraints of available data, our analyses do offer some useful insights.

In results not shown here, we find no significant relationship between average church
attendance in a county and religious friendship of nonreligious individuals in the FM
data. This result, although preliminary, suggests that a common premise in the literature
on contextual effects of religion needs to be
reconsidered, because nonreligious individuals in more religious areas may not be more
likely to have ties to the religious. One possibility is that in highly religious areas, religion is a more salient factor in social life and
thus religiosity-based inbreeding homophily
is stronger in such areas, off-setting the baseline homophily induced by a populations
religious composition. This is plausible, at
least in the United States, given the negative
public perception of atheists and the nonreligious (Edgell, Gerteis, and Hartmann 2006).
In the early twentieth century, Weber observed
that in the United States, membership in a
congregation was a prerequisite to being a
full member of the community (Lichterman
and Potts 2009), and this might still be the
case in some parts of contemporary America.
Some of our findings are also consistent
with the mobilization queue hypothesis,
which posits that organizers of volunteering
activities may target regular churchgoers
before reaching out to non-churchgoers,
because churchgoers could be easier to
recruit. In our county-level analyses, we find
that the negative relationship holds not only
for people who never attend religious services, but also for those who go to church
occasionally. One possibility is that in counties with many regular churchgoers, there
might be fewer volunteering opportunities for
occasional churchgoers, simply because
opportunities are taken mostly by regular
churchgoers. Chaves (2009) suggests that a
congregations main role in the local voluntary sector is to provide volunteers, and social
service organizations are well aware of the
unique resources congregations can provide
and actively pursue them. Thanks to the rich
pool of regular churchgoers, organizations in
highly religious areas may simply have less
need to reach out to non-churchgoers.

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

770

American Sociological Review 77(5)

There are other possibilities that seem


inconsistent with our findings. For example,
given that the negative relationship between
contextual religiosity and volunteering also
applies to occasional churchgoers and that a
strong presence of Evangelicals predicts a
lower level of volunteering for regular
churchgoers but not for non-churchgoers, it
seems unlikely that this negative relationship
is driven by secular people in religious areas
who are reluctant to participate in activities
associated with certain religious congregations. We also find that the number of
religious or nonreligious civic organizations
does not predict volunteering significantly
and fails to explain the negative relationship
between contextual religiosity and volunteering. Although this is indirect evidence at best,
it suggests that a simple supply-side account
that is, fewer secular volunteering opportunities in highly religious areasmay not be
sufficient to explain the negative association.
Although these findings shed useful light on
why non-churchgoers and occasional churchgoers in religious areas are less likely to volunteer, future research needs to examine this
negative relationship more thoroughly. Scholars should consider more fully how religious
contexts may affect nonreligious individuals
with respect to their social and civic life more
generally. Our inconsistent findings about contextual effects of religion at the personal network and local level lend support to Lichterman
and Pottss (2009) claim that religions civic
potential is probably more complicated than the
popular notion that religion is either uniformly
good or bad for civil society. On the one hand,
we find that religions influence on civic
engagement can spillover to nonreligious individuals and thus vitalize the civic sector beyond
the bounds of religious congregations. On the
other hand, we also find that a high level of

religiosity in local areas can isolate nonreligious individuals civically. To fully understand
the civic role of religion, researchers need to
pay attention to this complex relationship
between religion and civic life among the nonreligious, not just among the religious.
Another important implication of this
studys findings is that future research on any
kind of contextual effect should be cautious
of how context is defined and measured.
Researchers need to clearly articulate the
theoretical mechanisms underlying contextual effects and develop measures that directly
reflect these mechanisms. Across different
substantive areas of sociology, contextual
effects are commonly proposed as explanations for a variety of individual and social
outcomes. Context, however, is often chosen
on the basis of data availability, rather than on
the basis of congruence with a proposed theoretical mechanism. Even though many studies
explain contextual effects as being the result
of social interaction and interpersonal influence through social networks, such studies
usually measure contextual factors by aggregating personal traits in a large geographic
unit. Our results suggest that how context is
operationalized is consequential for adequately testing explanatory mechanisms. As
our case shows, depending on which contextual unit is chosen, the relationship between
two variables could be very different or even
in opposite directions. Different mechanisms
might be at work to link variables at different
levels of context. Researchers need to more
carefully consider the match between a proposed mechanism and the way context is
actually operationalized in analyses. Advances
in methodology, such as hierarchical linear
modeling, do not solve the problem of the
ecological fallacy unless due attention is
given to choosing the correct contextual unit.

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

Lim and MacGregor

771

Appendix
Table A1. Description of Variables Used for Analyses in Table 1
Variable
Individual-Level Variables
Volunteering

Description
Have you done any of the following in the
past month? How about: Volunteered your
time to an organization? (no = 0; yes = 1)
Respondents age (years)
Respondent is female (male = 0; female = 1)
Respondents educational achievement
(Reference category)

Mean

SD

.20

.40

Age
39.54
Female
.53
Education
Elementary education or less
.34
Secondary to three-year tertiary
.53
education
Four years of tertiary education or more
.14
Marital Status
Respondents marital status
Married/domestic partner
(Reference category)
.58
Single (never married)
.30
Divorced/separated
.05
Widowed
.06
Number of children under age 15
Number of children under age 15 years living
1.21
in household (capped at four)
Respondents religion
Respondents religious affiliation
Catholic
(Reference category)
.23
Protestant
.12
Orthodox
.07
Islam/Muslim
.32
Hinduism
.04
Buddhism
.06
Judaism
.01
Secular/atheist/agnostic/no religion
.04
Christian
.08
Others
.04
Religious service attendance
Have you attended a place of worship or re.49
ligious service within the last seven days?
(no = 0; yes = 1)
Country-Level Variables
Religious culture
Predominant religious culture in country
(source: Alesina et al. 2003)
Protestant
(Reference category)
.11
Catholic
.27
Orthodox
.10
Muslim
.35
Eastern
.14
Other
.04
Life expectancy at birth
Life expectancy at birth in 2008 or the nearest 69.91
year available (source: UNDP 2011)
Average years of schooling
Average years of schooling in 2008 or the
7.63
nearest year available (source: UNDP 2011)
Gross national income per capita
Gross national income per capita in 2008 or 13504.69
the nearest year available (source: UNDP
2011)
Democracy score
Democracy score in 2008 (source: Economist
5.57
Intelligence Unit 2010)
Proportion of churchgoers
Proportion of churchgoers in country (source:
.50
Gallup World Poll data)

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

16.96
.50

.47
.50
.34

.49
.46
.22
.25
1.36

.42
.32
.26
.47
.19
.23
.09
.21
.27
.20
.50

.31
.44
.30
.48
.35
.19
9.44
2.87
14715.22

2.13
.21

772

American Sociological Review 77(5)

Table A2. Description of Variables Used for Analyses in Table 2


Variable

Description

Individual-Level Variables
Volunteering

Age
Female
Race
White
Black
Asian
Hispanic
Others
Marital status
Single (never married)
Married/domestic partner
Separated/divorced
Widowed
Number of children under age 18
None
One
Two
Three
Education
Less than high school degree
High school diploma
Technical or vocational school/
some college
College degree
Postgraduate education
Monthly household income
Less than $2,000
$2,000 to $3,999
$4,000 to $7,499
$7,500 or higher
Dont know or refused to answer
Religious preference
Protestant
Catholic
Jewish
Other Christian
Other non-Christian
No religion
Religious service attendance

Have you done any of the following in


the past month? How about volunteered your time to an organization?
Respondents age (years)
Respondent is female (male = 0; female
= 1)
Respondents race
(Reference category)

Mean

SD

.41

.49

54.57
.50

16.97
.50

.82
.06
.01
.07
.03

Respondents marital status


(Reference category)

Number of children under age 18 years


in household
(Reference category)

.15
.59
.13
.12

.71
.12
.11
.06

Respondents educational attainment


(Reference category)

.05
.20
.31
.23
.21

Respondents monthly family income


.17
.21
.23
.18
.21
Respondents religious preference
(Reference category)

.36
.26
.03
.19
.03
.13
Respondents approximated frequency of 22.43
annual attendance

County-Level Variables
Average service attendance in county Mean of respondents religious service
attendance in each county
Region
Region where respondent lives
(Reference category)
South

.38
.25
.12
.25
.17

.36
.49
.34
.32

.45
.32
.31
.24

.22
.40
.46
.42
.41

.38
.40
.42
.39
.41

.48
.44
.17
.40
.16
.33
22.05

22.44

4.49

.27

.45
(continued)

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

Lim and MacGregor

773

Table A2. (continued)


Variable

Description

East
Midwest
West
Religious organization density

Mean

SD

.20
.28
.25
1.88

.40
.45
.43
.35

Number of religious organizations per


10,000 residents (logged) (source:
Rupasingha and Goetz 2008)
Civic organization density
Number of nonreligious civic organiza1.25
tions per 10,000 residents (logged)
(source: Rupasingha and Goetz 2008)
% Evangelical Protestant
Evangelical Protestant as % of county
2.27
population (logged) (source: Religious
Congregational Membership Study
[Jones et al. 2002])
% Mainline Protestant
Mainline Protestant as % of county
1.89
population (logged) (source: Religious
Congregational Membership Study
[Jones et al. 2002])
% Catholic Protestant
Catholic as % of county population
2.25
(logged) (source: Religious Congregational Membership Study [Jones et al.
2002])
Population density in county (logged) Source: U.S. Census Bureau 2000
6.21
% of population in urban area
Source: U.S. Census Bureau 2000
81.93
Index of racial diversity
Index of racial diversity is based on
.38
seven racial categories (source: U.S.
Census Bureau 2000)
% of households in poverty in county Source: U.S. Census Bureau 2000
8.20
% of residents with college degree in Source: U.S. Census Bureau 2000
25.85
county
Median household income in county Source: U.S. Census Bureau 2000
10.69
(logged)

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

.36

.64

.56

.67

1.46
20.21
.19

3.97
9.14
.22

774

American Sociological Review 77(5)

Table A3. Description of Variables Used for Analyses in Table 3


2006
Variable
Volunteering

Description

Volunteered in the past 12


monthsa
Age
Respondents age
Male
Respondent is male (male = 1;
female = 0)
Race
Respondents race
White
(Reference category)
Black
Black = 1; else = 0
Hispanic
Hispanic = 1; else = 0
Other race
Other race = 1; else = 0
Education
Years of formal education
Family income
Income (dollars)
Married
Married or in a marriage like
relationship (married = 1;
else = 0)
Children
Have children under age 18 at
home
Number of close friends
Number of close friends
Social involvement index
Social involvement factor
score indexb
Voluntary group involvement Number of types of voluntary
groups involved inC
Respondents religious preference
Catholic
Catholic = 1; else = 0
Mainline Protestant
Mainline Protestant = 1; else
=0
Evangelical Protestant
Evangelical Protestant = 1;
else = 0
Black Protestant
Black Protestant = 1; else = 0
Jewish
Jewish = 1; else = 0
Mormons
Mormon = 1; else = 0
Other non-Christian
Other non-Christian religion =
1; else = 0
No religion
(Reference category)
Religiosity index
Religiosity factor score indexd
Region
Census region where respondent lives
South
(Reference category)
Northeast
Northeast = 1; else = 0
Central
Central = 1; else = 0
West
West = 1; else = 0
Number of religions among
Number of religious groups
friends
with at least one friend (0
to 4)
Friends in own versus other
Whether respondent has
religions
friends in own or other
religions
No friends with religion (Reference category)
Friends only in own
Only in own religion = 1; else
religion
=0

2011

Mean

SD

Mean

SD

.46

.50

.51

.50

47.38
.53

16.00
.50

49.76
.51

17.48
.50

.84
.05
.09
.02
14.12
57464
.51

.36
.22
.28
.14
2.48
37786
.50

.82
.06
.07
.05
14.48
61372
.50

.38
.24
.25
.21
2.46
38672
.50

.35

.48

.30

.46

4.90
.05

3.14
.68

4.75
.07

2.96
.66

1.52

1.36

1.66

1.34

.22
.15

.41
.35

.17
.14

.37
.35

.21

.41

.21

.40

.04
.04
.01
.04

.19
.18
.09
.19

.04
.04
.00
.04

.19
.19
.07
.20

.31
.69

.46
.91

.37
.68

.48
.87

.29
.23
.20
.27
1.50

.46
.42
.40
.44
1.05

.30
.23
.26
.21
1.64

.46
.42
.44
.41
1.09

.19
.12

.39
.33

(continued)

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

Lim and MacGregor

775

Table A3. (continued)


2006
Variable

Description

Mean

Friends only in a different


religion
Friends in two or more
religions
Number of races among
friends
Number of religions among
churchgoing friends

Only in different religion = 1;


else = 0
Friends in two or more religions = 1; else = 0
Number of racial groups with
at least one friend
Number of religious groups
with at least one regular
churchgoer friend (0 to 4)

2011
SD

Mean

SD

.20

.40

.49

.50

.88

.95

1.33

1.12

The original question is: Some people volunteer, others dont. Did you happen to volunteer in the past
12 months? By volunteering, I mean any unpaid work youve done to help people besides your family
and friends or people you work with.
b
This index is based on the following frequency variables: visiting family or relatives, having friends at
home, and visiting a neighbors home.
c
The types include (1) hobby/sports/arts/music/other leisure activity group; (2) service/social welfare/
fraternal organization; (3) a youth/parent/school-support organization; (4) professional/trade/farm/
business association; (5) neighborhood/ethnic/political association; and (6) support group/self-help
program.
d
This index is based on the following variables: importance of religion in daily life, frequency of church
attendance, importance of religion to self-identity, belief in god, and strength of religious affiliation.

Acknowledgments
The authors thank Robert D. Putnam, Nan Dirk De Graaf,
Robert Wuthnow, Pamela Oliver, Thomas Sander, the
participants of the Politics, Culture, Society Workshop at
the University of Wisconsin-Madison, and the anonymous reviewers of ASR for their helpful comments. We
also thank the Gallup organization for making their data
available for this research.

4.

5.

Notes
1. This is not to say religions influence is universally
positive. Scholars have found that some religious
groups are intolerant (Putnam and Campbell 2010;
Reimer and Park 2001; Wilcox and Jelen 1990) and
U.S. congregations remain highly segregated along
racial lines (Emerson and Smith 2001). See also
Beyerlein and Hipp (2005) and Blanchard and colleagues (2008).
2. However, Ruiter and De Graaf do not replicate the
cross-level interaction between individual attendance and average attendance in a country, the key
test of the spillover effect.
3. In a few countries (e.g., China) the World Poll did not
ask the religion questions. For other countries, we
could not find some of the country-level covariates.
To ensure that our findings are not sensitive to exclusion of these countries, we replicated our analyses

6.

7.

without the country-level control variables. Results


based on 148 countries were almost identical with
what we report in Table 1.
Our own analyses of the WVS suggest that the finding on the cross-level interaction is extremely
sensitive to a few influential cases and also to which
countries are included in analyses. We also found
that the relationship between national religious context and volunteering is nonlinear, consistent with
what we found in the World Poll.
On average, 31 percent of all eligible respondents
contacted agreed to participate and 90 percent of
them completed the interview, yielding a response
rate below 30 percent. This is a typical response rate
for telephone surveys. Although a higher response
rate would be preferred, a number of studies demonstrate that low response rates do not necessarily
reduce the accuracy of survey estimates (Holbrook,
Krosnick, and Pfent 2007).
As a robustness check, we experimented with different thresholds (e.g., 500 respondents per county
rather than 200). Results were not sensitive to
which threshold we chose. We also examined
whether the key findings held when we used zip
code area rather than county as contextual unit.
Again, results were similar. These results are available from the authors upon request.
We examined whether the relationship is curvilinear
by visually inspecting the relationship in the

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

776

8.
9.

10.

11.

12.

13.

14.
15.

16.

American Sociological Review 77(5)

scatterplot and formally by including the quadratic


term of average attendance in a county. We found
no evidence for a nonlinear relationship.
Because all three variables have highly skewed distributions, we log-transformed them.
We examined the models with the interaction terms
between percentages of Evangelical and Mainline
Protestants and average attendance in a county.
Both interaction terms were negative, indicating
that the negative association between average attendance and volunteering is stronger with a higher
percentage of Protestants, regardless of their type.
These findings offer additional support for the
claim that the negative association between local
religiosity and volunteering is not driven by a
strong presence of a specific religion. Results are
available upon request.
We also estimated the model separately for never
and seldom churchgoers and results were similar.
In fact, the negative coefficient of average attendance in a county was somewhat larger for seldom
churchgoers, suggesting that the negative association is not limited to the most secular respondents.
We also found a negative relationship between local
religious contexts and the number of organizations
respondents volunteered for in SCCBS. These
results are available upon request.
The FM surveys response rate was 53 percent
based on formula 3 recommended by the American
Association for Public Opinion Research. The
response rate based on formula 5 was 69 percent
(Lim and Putnam 2010), comparable to the 2010
General Social Surveys response rate.
Although the re-interview rate of 54.2 percent is
less than ideal, this response rate is comparable to a
high quality survey such as the American National
Election Studies (ANES), despite the fact that the
two waves are five years apart. For example, in
2010, the ANES Panel Recontact Survey re-interviewed 54 percent of respondents from the first
wave conducted a year and a half earlier. For further
discussion on potential bias due to panel attrition,
see Part B in the online supplement.
We do not use 2007 data because they do not
include some of the control variables.
For a robustness check, we experimented with different thresholds (never, less than once a year,
and once or twice a year). Results were similar
regardless of the threshold we used, although some
coefficients were statistically insignificant when we
used a more restrictive threshold (e.g., never) due
to a small sample size.
Our analyses of FM and GSS panel data also suggest
the boundary of never churchgoers is porous.
Among never churchgoers in FM 2006, only 63 percent were still never churchgoers in 2007, even though
almost none of them said they had experienced any
religious change. Most of them switched to less than
once a year or about once a year.

17. We also estimated a model in which the number of


religions among which respondents have close
friends is a set of dummy variables rather than a
single continuous variable. Results suggest there is
indeed an approximately linear relationship between
the number of religions among ones friends and
volunteering. The strength of the relationship is
comparable to that in Model 1. In addition, we
experimented with a binary version of the variable
(no friends with religion versus any friend with religion) and found a similarly positive relationship.
These results are available upon request.
18. We also examined whether including the religious
friendship variable in the same models with a measure of local religious context changes the negative
association between the latter and volunteering that
we found in Study 2, using the 2006 FM data as
well as the SCCBS data. In both datasets, we found
that local religious contexts negatively predict volunteering with or without the religious friendship
variable.
19. We also examined whether having friends in any
specific religious tradition matters. Given the findings in Study 2, we might expect stronger spillover
effects when respondents have Mainline Protestant
friends than when they have Evangelical friends.
Results suggest that nonreligious respondents with
Mainline Protestant friends volunteer more than
those without and that people with only Evangelical
Protestant friends volunteer less than those with
friends in other religious groups. These differences,
however, are statistically insignificant.
20. Note that the coefficient for the churchgoing friendship variable in Model 3 is smaller than that for the
religious friendship variable in Model 1. This
appears to be due to random sampling variability. In
general, we found that relationships between religious friendship variables and volunteering tend to
be weaker (but still statistically significant) in the
2011 FM data than in the 2006 FM data even when
we use the exact same variable.

References
Alesina, Alberto, Arnaud Devleeschauwer, William Easterly, Sergio Kurlat, and Romain Wacziarg. 2003.
Fractionalization. Journal of Economic Growth
8:15594.
Ammerman, Nancy Tatom. 2005. Pillars of Faith: American Congregations and their Partners. Berkeley and
Los Angeles: University of California Press.
Bellah, Robert Neelly. 1991. Beyond Belief: Essays on
Religion in a Post-Traditional World. Berkeley: University of California Press.
Beyerlein, Kraig and John R. Hipp. 2005. Social Capital, Too Much of a Good Thing? American Religious
Traditions and Community Crime. Social Forces
84:9951013.

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

Lim and MacGregor

777

Blalock, Hubert M. 1984. Contextual-Effects Models:


Theoretical and Methodological Issues. Annual
Review of Sociology 10:35372.
Blanchard, Troy C., John P. Bartkowski, Todd L. Matthews, and Kent R. Kerley. 2008. Faith, Morality
and Mortality: The Ecological Impact of Religion on
Population Health. Social Forces 86:15911620.
Blau, Peter Michael and Joseph E. Schwartz. 1997.
Crosscutting Social Circles: Testing a Macrostructural Theory of Intergroup Relations. New Brunswick, NJ: Transaction Publishers.
Books, John and Charles Prysby. 1988. Studying Contextual Effects on Political Behavior. American Politics Research 16:21138.
Borgonovi, Francesca. 2008. Divided We Stand, United
We Fall: Religious Pluralism, Giving, and Volunteering. American Sociological Review 73:105128.
Brady, Henry E., Kay Lehman Schlozman, and Sidney
Verba. 1999. Prospecting for Participants: Rational
Expectations and the Recruitment of Political Activists. American Political Science Review 93:15368.
Cavendish, James C. 2000. Church-based Community
Activism: A Comparison of Black and White Catholic Congregations. Journal for the Scientific Study of
Religion 39:6477.
Chaves, Mark. 2009. Congregations Significance to
American Civic Life. Pp. 6981 in The Civic Life
of American Religion, edited by P. Lichterman and
C. Brady Potts. Palo Alto, CA: Stanford University
Press.
Dinesen, Peter and Kim Mannemar Snderskov. 2011.
Ethnic Diversity of the Micro-Context and Generalized Trust: Evidence from Denmark. Unpublished
manuscript
(http://www.ecprnet.eu/MyECPR/proposals/reykjavik/uploads/papers/2167.pdf).
DiPrete, Thomas A., Andrew Gelman, Tyler McCormick,
Julien Teitler, and Tian Zheng. 2010. Segregations
in Social Networks Based on Acquaintanceship and
Trust. American Journal of Sociology 116:123483.
Economist Intelligence Unit. 2008. Democracy Index
2008. The Economist (http://graphics.eiu.com/PDF/
Democracy%20Index%202008.pdf).
Edgell, Penny, Joseph Gerteis, and Douglas Hartmann.
2006. Atheists as Other: Moral Boundaries and
Cultural Membership in American Society. American Sociological Review 71:21134.
Ellison, Christopher G. 1992. Are Religious People
Nice People? Evidence from the National Survey of
Black Americans. Social Forces 71:41130.
Ellison, Christopher G., Jeffrey A. Burr, and Patricia L.
McCall. 1997. Religious Homogeneity and Metropolitan Suicide Rates. Social Forces 76:27399.
Emerson, Michael and Christian Smith. 2001. Divided by
Faith: Evangelical Religion and the Problem of Race
in America. Oxford, UK: Oxford University Press.
Erbring, Lutz and Alice A. Young. 1979. Individuals and
Social Structure. Sociological Methods & Research
7:396430.

Finke, Roger and Amy Adamczyk. 2008. Cross-National


Moral Beliefs: The Influence of National Religious
Context. Sociological Quarterly 49:61752.
Gallup Inc. 2009. Gallup-Healthways Well-Being Index:
Methodology Report for Indexes (http://well-beingindex.com/files/Gallup-Healthways%20Index%20
Methodology%20Report%20FINAL%203-25-08.
pdf).
Gallup Inc. 2011. Gallup Worldwide Research Methodology and Codebook. Unpublished document.
Hall, Peter Dobkin. 2005. Religion, Philanthropy and
Civic Engagement in Twentieth Century America.
Pp. 15984 in Gifts of Money and Time: The Role of
Charity in Americas Communities, edited by A. C.
Brooks. Lanham, MD: Rowman and Littlefield.
Holbrook, Allyson L., Jon A. Krosnick, and Alison Pfent.
2007. The Causes and Consequences of Response
Rates in Surveys by the News Media and Government
Contractor Survey Research Firms. Pp. 499528 in
Advances in Telephone Survey Methodology, edited
by J. M. Lepkowski, C. Tucker, J. M. Brick, E. D. De
Leeuw, L. Japec, P. J. Lavrakas, M. W. Link, and R.
L. Sangster. Hoboken, NJ: John Wiley & Sons, Inc.
Hout, Michael and Claude S. Fischer. 2002. Why More
Americans Have No Religious Preference: Politics
and Generations. American Sociological Review
67:16590.
Jones, Dale E., Sherri Doty, Clifford Grammich, James
E. Horsch, Richard Houseal, Mac Lynn, John P. Marcum, Kenneth M. Sanchagrin, and Richard H. Taylor. 2002. Religious Congregations and Membership
in the United States 2000. Nashville, TN: Glenmary
Research Center.
Lam, Pui-Yan. 2002. As the Flocks Gather: How Religion
Affects Voluntary Association Participation. Journal
for the Scientific Study of Religion 41:405422.
Lam, Pui-Yan. 2006. Religion and Civic Culture: A
Cross-National Study of Voluntary Association Membership. Journal for the Scientific Study of Religion
45:17793.
Lichterman, Paul. 2005. Elusive Togetherness: Church
Groups Trying to Bridge Americas Divisions.
Princeton, NJ: Princeton University Press.
Lichterman, Paul and C. Brady Potts. 2009. The Civic
Life of American Religion. Palo Alto, CA: Stanford
University Press.
Lim, Chaeyoon. 2008. Social Networks and Political
Participation: How Do Networks Matter? Social
Forces 87:96182
Lim, Chaeyoon. 2010. Mobilizing on the Margin:
How Does Interpersonal Recruitment Affect Citizen
Participation in Politics. Social Science Research
39:34155.
Lim, Chaeyoon and Robert D. Putnam. 2010. Religion,
Social Networks, and Life Satisfaction. American
Sociological Review 75:91433.
Loveland, Matthew T., David Sikkink, Daniel J. Myers,
and Benjamin Radcliff. 2005. Private Prayer and

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

778

American Sociological Review 77(5)

Civic Involvement. Journal for the Scientific Study


of Religion 44:114.
McAdam, Doug and Ronnelle Paulsen. 1993. Specifying the Relationship between Social Ties and Activism. American Journal of Sociology 99:64067.
McPherson, Miller, Lynn Smith-Lovin, and James M.
Cook. 2001. Birds of a Feather: Homophily in Social
Networks. Annual Review of Sociology 27:41544.
Moody, James. 2001. Race, School Integration, and
Friendship in America. American Journal of Sociology 107:679716.
Musick, Marc A., John Wilson, and William B. Bynum
Jr. 2000. Race and Formal Volunteering: The Differential Effects of Class and Religion. Social Forces
78:153970.
Oliver, Pamela. 1984. If You Dont Do it, Nobody
Else Will: Active and Token Contributors to Local
Collective Action. American Sociological Review
49:601610.
Olson, Daniel V. A. and Paul Perl. 2011. A Friend in
Creed: Does the Religious Composition of Geographic Areas Affect the Religious Composition of
a Persons Close Friends? Journal for the Scientific
Study of Religion 50:483502.
Park, Jerry Z. and Christian Smith. 2000. To Whom
Much Has Been Given...: Religious Capital and
Community Voluntarism among Churchgoing Protestants. Journal for the Scientific Study of Religion
39:27286.
Putnam, Robert D. 2007. E Pluribus Unum: Diversity
and Community in the Twenty-first Century; The
2006 John Skytte Prize Lecture. Scandinavian Political Studies 30:13774.
Putnam, Robert D. and David E. Campbell. 2010. American Grace: How Religion Divides and Unites Us.
New York: Simon & Schuster.
Quillian, Lincoln and Mary Campbell. 2003. Beyond
Black and White: The Present and Future of Multiracial Friendship Segregation. American Sociological
Review 68:54066.
Regnerus, Mark D. 2003. Moral Communities and
Adolescent Delinquency. Sociological Quarterly
44:52354.
Regnerus, Mark D., Christian Smith, and David Sikkink.
1998. Who Gives to the Poor? The Influence of Religious Tradition and Political Location on the Personal
Generosity of Americans toward the Poor. Journal
for the Scientific Study of Religion 37:48193.
Reimer, Sam and Jerry Z. Park. 2001. Tolerant (In)civility? A Longitudinal Analysis of White Conservative
Protestants Willingness to Grant Civil Liberties.
Journal for the Scientific Study of Religion 40:735
45.
Roper Center for Public Opinion Research. 2001. Social
Capital Benchmark Survey, 2000 (http://www.
ropercenter.uconn.edu/misc/usmisc2000-soccap/
usmisc2000-soccap.PDF).
Ruiter, Stijn and Nan Dirk De Graaf. 2006. National
Context, Religiosity, and Volunteering: Results

from 53 Countries. American Sociological Review


71:191210.
Ruiter, Stijn and Nan Dirk De Graaf. 2010. National
Religious Context and Volunteering: More Rigorous
Tests Supporting the Association. American Sociological Review 75:17984.
Rupasingha, Anil and Stephan J. Goetz. 2008. U.S.
County-Level Social Capital Data, 19902005. The
Northeast Regional Center for Rural Development,
Penn State University, University Park, PA.
Schwadel, Philip. 2005. Individual, Congregational, and
Denominational Effects on Church Members Civic
Participation. Journal for the Scientific Study of
Religion 44:15971.
Serow, Robert C. and Julia I. Dreyden. 1990. Community Service among College and University Students:
Individual and Institutional Relationships. Adolescence 25:55366.
Smith, Christian. 2003. Theorizing Religious Effects
among American Adolescents. Journal for the Scientific Study of Religion 42:1730.
Stark, Rodney and William Sims Bainbridge. 1996. Religion,
Deviance, and Social Control. New York: Routledge.
Stipak, Brian and Carl Hensler. 1982. Statistical Inference in Contextual Analysis. American Journal of
Political Science 26:15175.
Traunmller, Richard. 2009. Individual Religiosity,
Religious Context, and the Creation of Social Trust in
Germany. Schmollers Jahrbuch 129:35765.
United Nations Development Programme (UNDP). 2011.
The Human Development Report 2011 (http://hdr.
undp.org/en/humandev/).
United States Census Bureau. 2000. The United States
2000 Census (http://www.census.gov/main/www/
cen2000.html).
Van der Meer, Tom, Manfred Te Grotenhuis, and Ben
Pelzer. 2010. Influential Cases in Multilevel Modeling. American Sociological Review 75:17378.
Weatherford, M. Stephan. 1982. Interpersonal Networks
and Political Behavior. American Journal of Political Science 26:11743.
Wilcox, Clyde and Ted Jelen. 1990. Evangelicals and
Political Tolerance. American Politics Quarterly
18:2546.
Wilson, John. 2000. Volunteering. Annual Review of
Sociology 26:21540.
Woodberry, Robert D. 2012. The Missionary Roots of
Liberal Democracy. American Political Science
Review 106:24474.
Wuthnow, Robert. 1991. Acts of Compassion: Caring
for Others and Helping Ourselves. Princeton, NJ:
Princeton University Press.
Wuthnow, Robert. 1999. Mobilizing Civic Engagement:
The Changing Impact of Religious Involvement. Pp.
33163 in Civic Engagement in American Democracy, edited by T. Skocpol and M. P. Fiorina. Washington, DC: Brookings Institution Press.
Wuthnow, Robert and John H. Evans, eds. 2002. The
Quiet Hand of God: Faith-Based Activism and the

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

Lim and MacGregor

779

Public Role of Mainline Protestantism. Berkeley and


Los Angeles: University of California Press

and economic well-being, and the effects of formal education on civic engagement in East Asian countries.

Chaeyoon Lim is an Assistant Professor of Sociology at


University of Wisconsin-Madison. His current research
includes civic and political outcomes of religious involvement, dynamics of religious identification and behavior,
civic consequences of the Great Recession, civic health

Carol Ann MacGregor is an Assistant Professor of Sociology at Loyola University New Orleans. Her research
focuses on religion and civic engagement, socialization,
and educational attainment. She is currently working on a
book manuscript that looks at the causes and consequences of the decline of Catholic education in the United
States.

Downloaded from asr.sagepub.com at CARLETON UNIV on November 28, 2014

You might also like