Académique Documents
Professionnel Documents
Culture Documents
A Dissertation
Presented to
In Partial Fulfillment
Doctor of Philosophy
Elizabeth R. B. Grossman
December, 2013
AN EXAMINATION OF PUTNAM’S, COLEMAN’S, AND BOURDIEU’S
Elizabeth R. B. Grossman
Dissertation
Approved: Accepted:
______________________________ ______________________________
Advisor Department Chair
Dr. Rebecca J. Erickson Dr. Matthew Lee
______________________________ ______________________________
Committee Member Dean of the College
Dr. Cheryl Elman Dr. Chand Midha
______________________________ ______________________________
Committee Member Dean of the Graduate School
Dr. Juan Xi Dr. George R. Newkome
______________________________ ______________________________
Committee Member Date
Dr. Clare L. Stacey
______________________________
Committee Member
Dr. Karl Kaltenthaler
ii
ABSTRACT
This dissertation explores the form and function of social capital, both theoretically and
based on race, gender and class. Specifically, this study (1) examines major theories of social
capital to identify points of comparison and contrast; (2) operationalizes social capital in
ways that are theoretically consistent with each approach; and (3) examines each theoretically
structured empirical model to determine the extent to which it captures the meaningful
experience of social capital across race, gender, and class. The study uses data from the 2000
Social Capital Community Benchmark Survey to create and examine three distinct models of
social capital drawn from the theories of Robert Putnam, James Coleman, and Pierre
Bourdieu. Confirmatory factor analysis is used to test how well the theoretical models fit the
data and multi-sample structural equation modeling (SEM) is used to test hypotheses
regarding significant differences in the structure of social capital across race, gender, and
class groups. Structural differences in social capital were found for all three socio-
demographic groups. However, the findings were not consistent across models. For example,
although differences by race were found across all three social capital models, class and
gender did not generate similar results. Theoretical and empirical implications for future
iii
ACKNOWLEDGEMENTS
First and foremost I need to thank and extend all my love to my husband, Michael
and my daughter Rebekah. Michael’s unfailing support, love, and home cooked meals
are what helped me get through this incredible journey. He is the first person who
believed and encouraged me to pursue my doctorate in Sociology, and support was key to
my completing this degree and dissertation. Thank you seems so small for all that you
have done, but nonetheless, thank you, sweetheart for sharing your strength and faith in
me, for always believing that I could finish, and for giving me the space and time I
needed to write.
arrival she has been a beacon of light and joy in my life. Rebekah has sacrificed much in
her young years for “Mommy’s Dissertation”, including but not limited to: bedtime
stories, time with mom, and dealing with stressed out mom. It is because of her that I not
once gave up on finishing this project. For you my darling, I would do anything.
Beautiful girl, follow your dreams and never give up, even when it gets hard. You can do
anything you focus on and whatever your dream is, I believe in you, always. I owe a
their part in guiding my efforts, without their expertise and unfailing support this
iv
dissertation would not be what it is. Thank you to Dr. Rebecca J. Erickson, Dr. Cheryl
Elman, Dr. Juan Xi, Dr. Clare L. Stacey, and Dr. Karl Kaltenthaler. While Dr. Plagens
had to leave for another institution, I would also like to thank him for his comments and
short shrift for the long road you have walked with me. You were always there to
support me, thank you for pushing me to reach further and go forward regardless of how
worried or scared I was. Thank you for sharing your brilliance and amazing energy with
me. I have learned so much from you. You are an accomplished and inspiring teacher
and I feel damn lucky to have had the opportunity to learn from you. Thank you for
everything that you did as my advisor, acting as a mentor, and being a friend when life
was throwing me boulders and expecting me to catch. Finally, when I first brought you
this project, I argued that it should be purely theoretical, thank you for making me do an
empirical dissertation. Yes, you were right, it is better this way. You have earned my
eternal thanks and support many times over. Thank you for being a true mentor,
For Cheryl, there is no end to your brilliance and positivity. Whenever I would
feel like the end would never come, you would point out how far I had come and your
positive outlook did more for me than I can ever truly express. Thank you so much for
believing in me, for being a mentor throughout my tenure at the University. I cannot tell
you how much I value our conversations and your input on this project. I wish you the
v
very best in your next journey at Duke, they have no idea how lucky they are to have you
there.
Joann I want to thank you so much for helping me to make sense of SEM and my
statistical models. Thank you for always finding time to meet with me when I needed
guidance, you are a very talented academic and a joy to work with, thank you so much. I
sincerely hope to work with you again. Thank you to Clare who stayed with me through
my comprehensive exam in inequality, the proposal, and then also the dissertation. Your
continuous support and advice have been irreplaceable during this process, thank you.
You were right, quality and reliable childcare is the key! Thank you to Dr. Kaltenthaler
from political science who helped me back at the very beginning of my doctorate work
when I thought I would study terrorist networks. That particular interest did not pan out
for me, but I thank you for your time and knowledge on the topic. Thank you for your
time and input on this project as well, I cannot tell you how much I appreciate it.
entire process and I would like to express my thanks to professors whose mentorship and
teaching over the years helped to make me who I am today as a Sociologist. First, to Dr.
Andre Christie-Mizell, thank you so much for your support, guidance, and friendship- I
learned so much in your classes and appreciate the opportunity you provided to do
research and publish together. Most importantly for this project, thank you so much for
taking time out of your busy schedule to teach me multi-sample SEM long-distance.
Thank you for your help, guidance, knowledge, and support throughout my doctoral
scholarship.
vi
Then also, to Dr. Kathy Feltey, thank you for always believing in me and giving
advice when I was clearly in need of it! I am grateful to know you and grateful for the
chance to learn from you. Dr. Matt Lee, it has been a true pleasure to work for someone
who takes a humanistic approach to life and academics. Thank you so much for the time
and guidance that you gave, and for remembering me even though many might have
written me off, thank you. For Dr. Peralta, thank you so much for your advice and
feedback on the process. Your summary of where I was in the process and how close I
was to finishing was a beacon of light when things seemed so dark. Dr. Mark Tausig,
thank you so much for always offering support and advice, it was helpful to know that
should I need your help, I had only to ask. In truth I want to thank the faculty at the
University of Akron who over the years I have had the distinct pleasure of getting to
know each of you and value your advice, research, and teaching. Finally, to my first
mentor in sociology, Dr. Shelley Smith, who believed in me and supported me continuing
on to earn my doctorate and to Dr. John Skvoretz, for peaking my interest in social
networks and reflecting back to me that I had a proclivity for Sociology, or as you once
For my family, who are always a wellspring of love and support, I am eternally
grateful. Thank you so much to my mom, I love you with all of my heart. It is not just
for all of the times I had to put this project first, it is for all the love and support you have
given for every project in my life, and believing in me. Thank you to her husband, Jim,
and my brother Brooks for their unfailing support and belief that I can do this thing also.
vii
To my dad and Peggy, thank you so much for your understanding, constant faith,
and love. I adore and love both of you and appreciate everything you do. Thanks to my
sister Dawn and her family for their support and love as well. Thank you to my
Grandma, my Aunt Lois and Julie, Aunt Dottie and Uncle Carlos, Ali, Anna, Eric, and
Uncle John and Aunt Charlotte. Your light and love have helped to light my way.
Grandma, thank you so much for your words of advice, I carry with me always. Thank
you to my cousin Mark and Jean Anne, to my cousins and extended family as well.
Thank you to my husband’s family, most especially my mother-in-law, Irina. Thank you
to my fictive kin, Shyla Vesitis, Lymy (Jennifer Lymneos), and the Divers’ family. I am
forever thankful to have you in my life, you have brought humor, love, and perspective
when it was dearly needed, thank you. To all of my family, thank you for sharing in my
To all of my friends and colleagues, what a ride! Thank you to Dr. Rachel Stein, I
cannot imagine going through grad school without you. Thank you for watching the
beasts, sharing a smoke, sharing a laugh, hug, drink, advice, and smile, and most of all
for being my friend. Thank you so much to my friends for being there when I needed
you most, to Aya Kimura Ada, Jodi Ross, Stacye Blount, Jean Anne Sutherland, Jeanine
Gailey, Ariane Prohaska, Wendy Grove, Michelle Bemiller, Rachel Schneider, Suzanne
Slusser, Angela Atkins, Daysha Lawerence, Lori Tuttle, Martina Sharp-Grier, Dani Jauk,
Monica, Nic, Peter, Kelsey, and Marcella. Jodi, thank you for the great conversations
and time spent discussing theory, I am grateful to you for our time together. To my
office buddies, Richelle Dykstra and Dave Skubby, thank you so much for being such
viii
great roomies and friends. Thank you to the office staff and support staff at the
University of Akron, beginning with Jill and now finishing with Tammy Dixon. I would
also like to thank Dee for his constant support. A big thank you to Cindy Saylor Steinel,
the web Goddess and friend, without you I would have been lost technologically.
To all of my Cottey sisters, your love and friendship has been a sustaining force in
my life and I am and always will be grateful to you all, “friends we are and friends we’ll
Thank you to those whose love I hold in my heart, my Granny and Pa, Rita and
Bob Brooks, my Grandfather, George Baseler, my ‘Aunt’ Lizzie, my cousin Eric, and my
ix
TABLE OF CONTENTS
Page
LIST OF TABLES…………..………………………………………………………..…xii
CHAPTER
I. INTRODUCTION………………………..…………………………………………….1
Social Capital……....…...…………………………………………………………4
II. THEORY…………………………..............................................................................10
Missing Data…...………………………………………………………………...48
Measurement…….……………………………………………………………….51
Analytic Strategy………..……………………………………………………….66
Descriptive Statistics………...…………………………………………………...73
x
Coleman’s Social Capital Model…………...……………………………………91
Contributions……..…………………………………………………………….191
REFERENCES………..………………………………………………………………..198
APPENDICES………………………………………………………………………….210
xi
LIST OF TABLES
Table Page
4.3 Covariance Structure Analysis: Maximum Likelihood Estimation for the CALIS
Procedure Confirmatory Factor Analysis – Fit Statistics – Putnam Social Capital
Model…………………………………………………………………………….86
4.4 EFA for Coleman’s Three Forms of Social Capital – Rotated Factor Pattern,
Standardized Results……………………………………………………………..92
4.5 CFA – Covariance Structure Analysis: Maximum Likelihood Estimation for the
CALIS Procedure Confirmatory Factor Analysis – Fit Statistics – Coleman’s
Social Capital Model……………………………………………………………..97
4.6 EFA for Bourdieu’s Social Capital – Rotated Factor Pattern, Standardized
Loadings………………………………………………………………………...104
4.7 CFA – Covariance Structure Analysis: Maximum Likelihood Estimation for the
CALIS Procedure Confirmatory Factor Analysis – Fit Statistics – Bourdieu’s
Social Capital Model……………………………………………………………108
5.1 Fit Statistics for Putnam’s Social Capital Model Using Multi-Sample SEM Social
Class in the Sample, 2000……………………………..……………………….115
5.2 Fit Statistics for Putnam’s Social Capital Model Using Multi-Sample SEM of
Race in the Sample, 2000………………………………………………..……..126
5.3 Fit Statistics for Putnam’s Social Capital Model Multi-Sample SEM with Gender
and Social Capital………………………………………………………………134
5.4 Fit Statistics for Coleman’s Social Capital Model, Multi-Sample Structural
Equation Modeling for Social Class………………………………..…………..136
5.5 Fit Statistics for Multi-Sample SEM of Race and Social Capital in the Sample,
2000………………………………………………………………………..……138
xii
5.6 Fit Statistics for Multi-Sample SEM of Gender and Social Capital in the Sample,
2000………………………………………………………………………….….146
5.7 Fit Statistics for Bourdieu’s Social Capital Model, Multi-Sample SEM for Social
Class, U.S., 2000………………………………………………………………..148
5.8 Fit Statistics for Multi-Sample SEM of Race and Bourdieu’s Social Capital in the
Sample, 2000…………………………………………………………...……….157
5.9 Fit Statistics for Multi-Sample SEM of Gender and Bourdieu’s Social Capital in
the Sample, 2000………………………………………………………….....….165
xiii
LIST OF FIGURES
Figure Page
4.3 CFA Measurement Model for Putnam’s Social Capital (unstandardized beta, [t-
score], standardized Beta)……………………………………………………….90
4.6 CFA Measurement Model for Coleman’s Social Capital (unstandardized beta, [t-
score], standardized Beta)……………………………………………………...100
5.1 Putnam’s Social Capital Model for Lower Social Class (unstandardized beta, [t-
score], standardized Beta) Fit Indexes for Lower Class: SRMSR 0.07, GFI 0.96,
NFI 0.84…………………………………………………………………….…..117
5.2 Putnam’s Social Capital Model for Middle Social Class (unstandardized beta, [t-
score], standardized Beta) Fit Indexes Middle Class: SRMSR 0.07, GFI 0.96, NFI
0.79……………………………………………………………………..……….120
5.3 Putnam’s Social Capital Model for Upper Social Class, (unstandardized beta, [t-
score], standardized Beta) Fit Indexes Upper Class: SRMSR 0.07, GFI 0.96, NFI
0.84……………………………………………………………………………...125
5.4 Putnam’s Social Capital Model, Whites in the Sample, 2000 (unstandardized beta,
[t-score], standardized Beta) Fit Indexes: SRMSR 0.04, GFI 0.98, NFI
0.92……………………………………………………………………………...128
xiv
5.5 Putnam’s Social Capital Model, Non-Whites in the Sample, 2000 (unstandardized
beta, [t-score], standardized Beta) Fit Indexes: SRMSR 0.07, GFI 0.94, NFI 0.73
…………………………………………………………………………………..132
5.6 Coleman’s Social Capital Model Multi-Sample SEM for Whites in 2000,
(unstandardized beta, [t-score], standardized Beta) Fit Indexes: SRMSR 0.04, GFI
0.99, NFI 0.94.……………………………………………………….…………140
5.7 Coleman’s Social Capital Model Multi-Sample SEM for Non-Whites in 2000,
(unstandardized beta, [t-score], standardized Beta) Fit Indexes: SRMSR 0.06, GFI
0.96, NFI 0.80……………………………………..……………………………143
5.8 Bourdieu’s Social Capital Model for Lower Social Class, model 1,
(unstandardized beta, [t-score], standardized Beta) Fit Indexes: SRMSR 0.08, GFI
0.96, NFI 0.81……………………………………..………………………...….150
5.9 Bourdieu’s Social Capital Model for Middle Social Class, model 1,
(unstandardized beta, [t-score], standardized Beta) Fit Indexes: SRMSR 0.05, GFI
0.98, NFI 0.92……………………………………..……………………………153
5.10 Bourdieu’s Social Capital Model for Upper Social Class, model 1,
(unstandardized beta, [t-score], standardized Beta) Fit Indexes: SRMSR 0.10, GFI
0.93, NFI 0.75……………………………………..……………………………156
5.11 Bourdieu’s Social Capital Model for Whites, model 1, (unstandardized beta, [t-
score], standardized Beta) Fit Indexes: SRMSR 0.05, GFI 0.98, NFI 0.93……160
5.12 Bourdieu’s Social Capital Model for Non-Whites, model 2, (unstandardized beta,
[t-score], standardized Beta) Fit Indexes: SRMSR 0.09, GFI 0.95, NFI 0.79....163
5.13 Bourdieu’s Social Capital Model for Women, model 1, (unstandardized beta, [t-
score], standardized Beta) Fit Indexes: SRMSR 0.03,GFI 0.99, NFI 0.95…….167
5.14 Bourdieu’s Social Capital Model for Men, model 2, (unstandardized beta, [t-
score], standardized Beta) Fit Indexes: SRMSR 0.04, GFI 0.98, NFI 0.94...…169
xv
CHAPTER I
INTRODUCTION
Social capital is a “force majeure” within the social sciences. It has emerged as a
(educational, income, social mobility) as well as health and well-being, civic engagement
and social solidarity (Coleman 1989; Field 2003; Lin1999; Lin 2007; Parks-Yancy,
DiTomaso, and Post 2008; Putnam 2000; Putnam et al. 2004; Putnam 1993, 2000).
Despite these strengths, a basic question remains: What is social capital? Although
scholars continue to make claims about its importance, different theoretical approaches
related to social capital would benefit from a systematic examination of how key
consistent measures to capture the experiences of different racial, gender, and class
groups would advance social scientific theory and research on social capital in
Researchers frequently use a global measure of social capital that emerges from
the tendency to view it as a universal concept, that is, one that is true for all groups at all
1
times (Coleman 1988; Putnam 2000). In contrast, I concur with others (Lin 2001;
Nieminen, Martelin, Koskinen, Simpura, Alanen, Harkanen, and Aromaa 2008; Parks-
Yancy, DiTomaso, and Post 2008) who suggest that different groups living in different
contexts with various experiences are likely to experience social capital differently. For
example, Parks-Yancy et al. (2008) note that while “[m]ost discussions of social capital
take for granted that group members are culturally similar”, they argue that social capital
should include socio-demographic variables such as class, race and gender (240). This
view is consistent with that of Lin (2001) who calls for more research on social capital
and the work of Nieminen et al. (2008) whose study found important differences in social
Although studies have drawn on various theories of social capital and some have
even focused on measurement issues, none have attempted to: (1) examine major theories
of social capital to identify points of comparison and contrast; (2) operationalize social
capital in ways that are theoretically consistent with each approach; and (3) examine each
theoretically structured empirical model to determine the extent to which they capture the
race, class, and gender)1. In order to better understand social capital within the context of
such socio-demographics as race, gender and class, the current study explores the form
and function of social capital both theoretically and empirically. In so doing, I address
two main questions. First, are Putnam’s, Coleman’s, and Bourdieu’s conceptualizations
of social capital supported empirically? Theory and method are intricately intertwined
and tied to empirical data (Cohen 1989). The three theorists’ conceptualizations of social
1
While there are other socio-demographic aspects that could be included, the available literature shows that
(aside from age), these are the three major points of interest with regard to inequality in our society (Parks-
Yancy et al. 2008).
2
capital are used across disciplines and in many varied ways throughout the literature.
However, researchers have tended to assume that the theories are being used in
reflective of observable reality, theories are nonetheless abstract. In contrast, data are
observable facts that are collected and studied. It would be easy to put the two side by
side and assume that the theory reflects the data. In this dissertation, however, I do not
assume such consistency but instead assess the extent to which empirical survey data can
be used to create sound theoretical models of social capital for Putnam, Coleman and
Bourdieu. The current dissertation research also addresses a second question: Does the
structure of social capital operate differently along class, race and gender lines for
The three potential contributions of this research, which comes from a mainly
adding to the understanding of what social capital is, and finally the findings herein
There are four major parts to this dissertation. First, in the remaining sections of
chapter one, I introduce the concept of social capital, develop a working definition for it,
and review the general social capital literature. In chapter two, I provide a detailed
outline of the three major theoretical frameworks of social capital that emerge from the
work of James Coleman, Robert Putnam, and Pierre Bourdieu. In this section, I further
develop the argument that theorists and researchers using the concept of social capital
need to be careful about making assumptions that treat actors’ social structural positions
3
as equivalent or assuming that social statuses do not impact how social capital operates in
people’s lives. To provide support for this argument, I review the literature that examines
the operation of social capital across class, race, and gender groups. In chapter three, I
introduce the data, the measures, and discuss the analytic strategy. I also provide a
thorough discussion of how missing data were handled as well as other methodological
and analytic concerns. In chapter four the three models (drawn from Putnam, Coleman,
and Bourdieu) are presented and the results of the confirmatory factor analysis used to
address Research Question one are reported for each theorists’ model. That is I examine
analysis, the findings from this analysis directly impact the analysis in the following
chapter by socio-demographic groups. Should the model(s) not fit for the confirmatory
factor analysis, then it would not be used for further analyses. Chapter five reports the
results of the multi-sample structural equation analyses for each social capital model and
by class, race, and gender where appropriate. These results test the hypotheses
introduced at the end of chapter two. Finally, chapter six summarizes the empirical
results, the limitations of the study, and the implications of my findings for current and
Social Capital
Some scholars argue that social capital has become increasingly popular because
it captures our sociological imaginations and allows us to socialize the economic concept
of capital (Lichterman 2006). Others suggest that social capital is a way for us to refocus
our attention away from the individual and toward social constructs, and to do so in a
4
way that does not demote agency (i.e., individual activity) (Field 2003). Still others
argue that its popularity is due to its operating as the “WD-40” of society (Putnam 2000:
One concern emerging from the concepts popularity is that social capital will be
transformed into a “catch-all” concept (Field 2003). Many argue in fact that social
capital is becoming just that (Fine 2001; Huber 2009). However, simply because social
capital crosses disciplines, accounts for important aspects of social life, and is useful for
understanding a multitude of outcomes does not mean that it is or will become a “catch-
all” concept. Instead, I would suggest that the “fears” associated with social capital
becoming irrelevant and/or losing its meaning are misplaced. The real issue concerning
social capital that may underlie both its popularity and concerns over its becoming
“dark side” of social capital (Cloud and Granfield 2008; Demant and Jarvinen 2011;
Field 2003; Putnam 2000). For example, Cloud and Granfield (2008) propose in their
study that social capital should be conceptualized in terms of both a positive and negative
scale. In their study they examined at the importance of social capital to the recovery
from substance abuse. For their purposes, positive social capital would help a person to
overcome an addiction, whereas a negative store of social capital would work against
recovery. Although this “dark side” approach presents an interesting avenue to explore in
future research, it is beyond the scope of the current study and is only mentioned briefly
one more time in relation to Putnam’s (2000) discussion on bonding social capital.
5
Instead, the current research focuses on more common conceptualizations of the term and
its correlates.
Social capital is related to the idea of human and economic capital, but it is much
more difficult to “see” and thus harder to define and measure. However, regardless of the
theorist or discipline, it is widely acknowledged that social capital has to do with people,
1977,1980,1984,1986, 2002; Coleman 1988, 1990; Field 2003; Lin 1999; Putnam1993,
2000). In most cases, scholars have conceptualized social capital in terms of “resources.”
Resources, and one’s access to them, have been of paramount interest and importance
within sociology since the writings of Weber (1946; also see Corcoran 1995). As part of
a theory of resources, social capital has become critical for understanding how resources
Building on these ideas, social capital can be defined as the social resources that
exist in the relationships between individuals and groups wherein such resources can be
accessed and used to reach individual or collective goals (Bourdieu; Coleman 1988; Lin
1999; Putnam 1993, 2000). This interpretation of social capital incorporates examples
from finding a good hairdresser through word of mouth to being promoted because of a
pre-existing relationship, or finding a job because your friend knows someone. The key
relationships. Seemingly simple at first (Field 2003), social capital has emerged over
time as a complex, abstract construction including such terms as social skills, trust,
6
reciprocity, exchange, obligations, in-group/out-groups, and norms (Alder and Kwon
2002; Bourdieu 1980, 1986; Coleman 1988; DeGraaf and Flap 1988; Elster 2003;
Herreros 2004; Huber 2008; Knudsen et al. 2007; Lin 1999, 2001, 2004; Portes 1998;
Putnam 2000). For example, a person’s social skills lie with them, but those social skills
help them to engage in interactions which can lead to the creation, maintenance and use
of social capital. Are these social skills then also social capital? Knudsen et al. (2008)
provide further support for the argument that social capital is not as simple as it may
seem, and that its definition, understanding and meaning can vary depending on the
bonding or bridging, Knudsen et al. (2008) report that the experience of social capital
resources in that they exist between people and help people to attain goals. For example,
someone who is traveling may ask a perfect stranger to watch their suitcases because they
need to use the restroom. Yet, the discussion of trust and whether it is social capital, or
whether it simply underlies it, or both, is not resolved (Field 2003). As these few
examples illustrate, a seemingly simple concept quickly becomes complex when applied
characteristics important to status, health, mental health, and inequality (e.g. income,
education), have sometimes been included in studies using social capital not outcomes,
but as moderating or mediating variables (Knudsen et al. (2007); Nieminen et al. 2008;
Parks-Yancy et al. 2008). Stack (1974) found, for instance, that social resources were
7
utilized differently than previously thought within a poor African American community.
Respondents in her study shared resources whenever possible, sometimes at the expense
of their own immediate family, and in many cases for the betterment of the community.
Her work illustrated that respondents used social capital in creative ways to manage their
lack of economic capital and in the face of structural constraints (Field 2003; Stack
1974).
Social capital’s complexities can be seen through the lens of gender as well.
Traditionally women tend to have more ties to family and personal friends whereas men
tend to have more work related ties (Parks-Yancy et al. 2008). Because these bonds have
fundamental differences, they tend to produce different social resources. Thus both men
and women tend to suffer from a deficit of certain types of social capital. While race,
class and gender have been included in models with social capital2, there is relatively
little research exploring how these three dimensions of stratification might combine to
Although the growing body of literature related to social capital has increased
social scientific appreciation for the concept and has succeeded in raising awareness of its
conceptualization and measurement of social capital has also placed limits on its
theoretical and empirical value. One reason for this may lie in the fact that researchers
are drawing on conceptualizations of social capital stemming from at least three different
2
Parks-Yancy et al.’s (2008) theoretical piece argues that race, class and gender matter to social capital and
specifically to reciprocity within exchange interactions.
8
operationalizing social capital within empirical research, one must first understand these
different traditions.
9
CHAPTER II
THEORY
people’s lives. But are we really “seeing” social capital? While it is safe to assume that
people have the potential to create, maintain and use social capital, the assumption of
many researchers (whether explicit or implied) is that social capital is social capital,
regardless of one’s position in the social structure (DeGraaf and Flap 1988; Coleman
1988, 1990; Friedkin 1980; Putnam 1993, 2000)3. However recent research suggests that
economic status, class, race, gender, and age (Aguilera 2002; Knudsen et al. 2007;
Nieminen et al. 2008; Parks-Yancy et al. 2008). The findings of researchers parallel
activism, and solidarity Campbell 2000; Gittell et al. 2000; Maloney et al. 2000; O’Neill
and Gidengil 2006; Norris and Inglehart (2006); Putnam 2000; Stack 1974; Sretzer
2000). Whether it was their intention or not, these researchers also found that social
3
Bourdieu would be the one major social capital theorist that from the beginning considered social capital
according to one’s position. However, even Bourdieu generalized that social capital should in theory
operate in the same ways for all groups. His point was that some people had more of it than others
based on their position in social space (2002).
10
As such, a more complete understanding of social capital requires an empirical
dimensional concept that has the potential to be influenced by the characteristics of the
capital theoretical and empirical knowledge about social capital, and to do so in the
context of allowing for differences across socio-demographic groups. First, however, the
following section details and differentiates the theoretical approaches taken by the three
most prominent theorists of social capital: Robert Putnam (1993, 2000), James Coleman
(1988, 1990), and Pierre Bourdieu (1977, 1980, 1984, 1986, 2002).4
and civic action as well as democracy and community. His treatment of social capital is
one reason so many scholars have latched onto social capital as a concept and some as a
theory in itself. Specifically, he argues that communities with higher social capital have
an easier time with volunteerism due to norms of reciprocity within the networks of those
who are active in the community. He generally defines social capital in regard to
characteristics of organizations, for instance “…trust, norms, and networks that can
167).
4
Others, whose work has informed the discussion on social capital include, Granovetter 1973; Lin 1999,
2004; McPherson et al. 2006; Marsden and Hulbert 1988; Temkin and Rohe 1998. However, I have
chosen to limit my analysis to Coleman, Putnam and Bourdieu because researchers base their theoretical
and empirical measurements of social capital upon these three major theorists. These three
conceptualizations of social capital make up the foundation for most researchers focusing on and/or using
social capital.
11
For Putnam (2000), social capital is located in the social tie between two actors or
is inherently found within a community network of relations (as noted below, this is also
the case for Coleman). Thus, social capital lies within one’s network and more
specifically, in the relationships therein. Putnam (1993) likens social capital to other
forms of capital in that social capital will grow with use and dissipate with disuse (he also
notes that trust, social networks and social norms operate similarly). Putnam (2000) sees
a decline in social capital overall, he notes that there are fewer organizational
memberships, loosening ties across the country, as well as a sense of lost community. In
his 2013 article, Putnam uses his own hometown in Ohio as a prime example of the
deleterious effects the loss social capital has on community. Specifically, he compares
the experience of his classmates to the lack of community today. His classmates noted
that growing up they were poor and did not realize it at the time, but Putnam elaborates
that what they lacked in monetary resources they made up with social ties and support.
For example, a family that could not afford a car was able to hitch rides from neighbors
to work and school. Thus, the lack of a car did not stop the father from making every one
of his son’s games. However, in that same town today one does not find those kind of
Consistent with the preceding, Putnam (1993) views social capital as a public
good and not privately held (also see Coleman 1988, 1990). For Putnam (1993), social
capital is formed as a by-product of some other social event. Putnam (1993, 2000) sees
trust as a key indicator of social capital, because of its link to the norm of reciprocity as
well as the broad and localized benefits that are derived from social or generalized trust.
The latter is defined “as a ‘standing decision’ to give most people- even those whom one
12
does not know from direct experience- the benefit of the doubt” (Putnam 2000:136). For
Putnam (2000:134), the trust that underlies reciprocity is essential to social capital. The
benefits that can come from a network, community or society built on a solid foundation
of trust are numerous and may include, more volunteerism, higher levels of giving, higher
Putnam (2000) further delineates between “thin” and “thick” trust. “Thin” trust is
between acquaintances or strangers. This “thin” trust is directly tied to the social or
generalized trust that Putnam (2000) writes about as declining and imminently important
to community social capital. “Thick” trust in contrast is the trust that one builds with a
strong tie (i.e. family, close friend, life-time acquaintance). For example, a young
woman in Virginia divorced her husband. She decided to move back home and raise her
young daughter around family. She worked as a teacher and when she moved she needed
a new place to live. Her mother and step-father gave her the land she needed to build a
new house. Not only could she rely on her thick trust with family, but she was living
again in the small town she had grown up in which meant that she could find people she
knew and had built relationships with over time to help her build her home.
Conversely, a strange man on the street whose car had broken down approached a
young female student late one evening in the university parking deck. He asked for her
help and she explained that she could not take him to an undisclosed location to pick up
his car or get a tow as she was a woman alone and he was a stranger. However, she did
loan him the use of her cell phone to call a friend. While there was an absence of “thick”
trust, there was an element of “thin” trust in the example because she did help him find a
13
way home. According to Putnam (2000), however, “thin” trust is in decline both in urban
and rural areas. Illustrating some of the effects of this decline, Putnam points to the drop
in people who are willing to volunteer or pick up hitchhikers, that drivers on the road are
less civil, and people drive over the speed limit more often and at higher levels than in the
past.
Losing levels of “thin” trust in our society and in our communities is a great loss.
It is important because “thin” trust is one of the key foundations for social capital and as
[i]n short, people who trust others are all-round good citizens, and those more
engaged in community life are both more trusting and more trustworthy…[while]
the civically disengaged believe themselves to be surrounded by miscreants and
feel less constrained to be honest themselves (137).
The tie and network provide the structure necessary for social capital, but generalized
trust provides the foundation for beneficial norms to emerge and, most importantly for
social capital, the norm of reciprocity without which social capital would lose its
There are four important functions of social capital for Putnam (2000). First,
social capital reduces the complexity of solving “collective problems.” Putnam (2000)
identifies networks and social norms as the mechanisms that allow for ease in solving
community issues. In other words, social norms and social interactions are the important
aspects of how social capital is created, stored, and used. Second, social capital, “greases
the wheels” of society. Here Putnam suggests that when people are trustworthy and
willing to trust others “social transactions are less costly” (Putnam 2000: 288).
Next, Putnam (2000) argues that social capital allows for people to achieve a
common consciousness, where we realize that our lives and consequences are linked.
14
Essentially what he is referring to is the feeling that we are all in the same proverbial
“boat”. Here, Putnam draws again on social ties that are characterized by trust and which
“develop or maintain character traits that are good for the rest of society” (2000: 288).
Thus he asserts that people who are “joiners” will learn to be more tolerant and
empathetic because they are willing to listen to other viewpoints, while an absence of ties
will mean that there is no one to stop someone from acting on “their worst impulses”
(Putnam 2000: 289). Of course, he does not argue that if you have fewer social ties you
capital. He references John Dewey’s distinction between “doing with” and “doing for”
(p. 116) when talking about actions that are charitable and/or altruistic. According to
Dewey’s conceptualization, the people who actually stay and participate in person are
“doing with”. While those who (for whatever reason) cannot participate in person, but
instead give money (or things) to a charity or organization are “doing for”. Putnam
(2000) argues that “[s]ocial capital refers to networks of social connection – [and
connects social capital to] doing with” (116). Thus, participating with others toward a
Putnam (2000) further develops the concept of social capital to include bonding
capital and bridging capital. One can think of the main difference between bonding and
bridging social capital in terms of the type of tie or relationship that exists between two
actors. The type of tie dictates the type of social capital. In what follows, I will describe
each form of capital in turn. I will then contrast the two in order to clearly illustrate that
15
Bonding Social Capital
According to Putnam, bonding social capital refers to social resources that lie
within ones’ community and close network. Putnam writes, “[b]onding social capital is
good for undergirding specific reciprocity and mobilizing solidarity” (2000: 22).
Accordingly, bonding social capital is an exclusionary form of social capital that helps to
bind people together based on their similarities.5 Bonding social capital is characterized
by strong ties and close personal relationships. It references those in-groups that feed off
Another way to think of bonding social capital is social resources that lie with ties
to people who are “like” you. For Putnam (2000), bonding social capital is the
social capital. An example of bonding social capital would be when someone in your
family dies and you need help and support. If you are a member of a church or place of
worship then you expect that the minister and congregants will provide you with this
that are exclusive (where some restrictions apply to membership, for example, based on
race, gender, class, or kinship (Knudsen et al. 2007; Putnam 2000)). Therefore, bonding
social capital includes the ties between family members and close friends. It can also
refer to memberships such as fraternities or sororities, ethnic group ties, and religious
group membership.
It has been said that bonding social capital can encapsulate the darker “side” of
social capital, also noted by Weber (1945) as social closure, where bonding social capital
5
This is reminiscent of Durkheim’s (1997) ideas related to solidarity and specifically mechanical solidarity, one aspect
being collective consciousness.
16
can have deleterious effects on social inequality, especially where issues already exist
“bonding practices can reinforce the practices of nepotism, ethnic hatred, and
sectarianism, as well as sexism” (Norris and Inglehart 2006: 77). As noted earlier, this
facet of bonding social capital, while important, is not considered in the current study.
Bridging social capital refers to ties with resources that extend outside of one’s
external assets and for information diffusion” (Putnam 2000: 22). Furthermore, ties that
can be construed as bridging social capital tend to “generate broader identities and
reciprocity, whereas bonding social capital bolsters our narrower selves” (Putnam 2000:
23). Putnam equates bridging social capital to the social lubricant or “WD-40” of
relationships. In fact, most of what Putnam refers to as “thin” trust underlies bridging
social capital.
Putnam (2000), it is this form of capital that underlies people’s decision to join the
Chamber of Commerce and attend conferences or other events that promote social
networking opportunities that reach beyond their immediate local ties. Drawing on these
examples, groups that are considered inclusive (where membership is not restrictive) are
linked to bridging social capital (Putnam 2000; Knudsen et al 2007). Underlying these
Granovetter (1973) found that weak ties, also known as acquaintances, should not be
17
dismissed. These ties tended to help people find jobs and linked them to the broader
Whereas bonding social capital helps to create loyalty to the group and strengthen
distrust, bridging social capital lies in the diversity of relationships and promotes the
between the two types of social capital, where having more bonding social capital means
one will have less bridging social capital. However, within his discussion on bridging
versus bonding social capital, Putnam (2000) emphasizes that neither is exclusive of the
other. Although he does not go into great detail about the process through which this
happens, Putnam (2000) notes that groups (his unit of analysis is at the community level)
have the ability to create bonding and bridging social capital simultaneously.
Finally, Putnam (2000) argues that people tend to have more bonding social
capital than bridging. However, bridging social capital is important to branching out and
increasing the breadth of one’s social resources. He goes on to note that bonding social
capital is easier to build and maintain over bridging social capital, but that both are
important, one to the well-being of one’s emotional and immediate social network, and
The most common definition of social capital comes from James Coleman who
defines it as existing inside the “structure of relations between actors and among actors”
(1988:S98). He further identifies social capital as relations that bring about social action
18
(see Coleman 1988, 1990). In his effort to integrate an agentive, rational actor with
structuralist constraint, Coleman was led to the concept of social capital. Social capital
was thus his answer to solving the sociological question of agency, structure, or both? In
social capital both agency and structure are of vital importance and neither can
completely be ignored. For Coleman (1988, 1990) social capital was able to bridge the
the creation of social capital as well as what it means to have social capital. For him,
social capital resides in the ties between people, (i.e. in their networks) and is not
privately held (Coleman 1988, 1990). Coleman (1988) considers having relationships
with others as a social form of “capital” because these ties contain resources. He also
For Coleman (1988, 1990), the family is the natural birthplace for social capital.
organization which is distinguished by the fact that its origins lay ‘in the relationships
specifically the benefits that parents share with their children, for social capital (Abbas,
2002; Francis and Archer, 2005; Goulbourne and Solomos, 2003; Thapar and Sanghera
As a result, Coleman views the decline of the family as a major blow to the
creation and maintenance of social capital within society. For it is through our
experiences in our families that we are able to create, sustain and use social capital.
19
capital (Coleman 1988; Field 2003). It follows then that the loss of or lessening of the
family unit means that the fundamental understanding and operation of social capital will
article focusing on the creation of human capital through social capital. Here, he clearly
links the creation of human capital in children through the social capital that they hold
with their family members, specifically parents. Children’s relationships to their parents
are an important aspect of this process. However, it is how their parents’ social capital
operates outside the home to connect their children with important actors in the education
system, as well as in regard to the creation of social closure within the broader
community that are integral to the children’s educational success and, ultimately, to the
Coleman (1988) defines social capital by its function, allowing those using his
theory to apply the concept quite broadly. By function, Coleman (1990) is referring to
the fact that social capital “is not a single entity” (302). As described below, social
capital can take different forms. In each case, social capital is seen as existing within
social structures and as enabling actors within social structures to take action (Coleman
1990). While social capital deals with social relations, the unit of analysis is not limited
to individual actors, but can include corporate or group actors as well. When talking
about corporate actors Coleman (1990) treats them similarly to individuals in that they
are also seen as rational and having an internal structure. With these considerations in
20
Expectations/Obligations
The first form of social capital has three parts: obligation, expectation and trust.
Coleman (1988, 1990) writes of trust in a social structure or society (depending on the
nature of the capital being studied) and emphasizes its importance to the building of
social capital. For Coleman, trust is an essential building block of social capital.
The following example illustrates how these three facets combine to operate as social
capital. An expectation is established when A provides a service (action) for B and trusts
that B will reciprocate the favor. Consequently, A expects something in the future from
B and B is now obligated to A. The relationship now has a new dimension wherein both
agree that B owes a favor to A in return for the one performed by B. According to
Coleman (1988), for this form of capital to work it depends on two things: (1) actor A
needs to place trust wisely (that B will return the favor), and (2) the extensiveness of
Although researchers debate whether trust is, in and of itself social capital,
Coleman (1988, 1990) assumes that trust is integral to social capital and includes it in his
analysis of not only expectations/obligations, but also in his discussion of social norms.
Social Norms
Coleman (1988) also considers social norms as providing social capital. The
example Coleman gives is of norms in the communities of Jerusalem. There, they have
norms that allow children to walk safely and, in so doing, to reflect a diffuse and valuable
form of social capital. In Jerusalem, for example, when a child is wandering alone it is
the responsibility of every adult surrounding that child to care for him or her. Elster
21
(2003) provides a different example of norms as social capital: When there is a social
norm against littering, a person who has a piece of trash may have to take the extra step
and time to find and deposit his or her trash in a designated receptacle and, as a result,
may grumble or feel put out. However, when followed, this particular social norm
provides a clean environment for all. While not explicitly stated, the example of the
family in Jerusalem who can let their child go into the city unattended without fear of the
child being hurt or injured versus a major United States city where children are feared for
if not supervised by a known and trusted adult illustrates Coleman’s assertion that trust
We can however, think of trusting as a norm in and of itself. Even with the
advent of new technologies people are finding that certain values and “shared
Demant and Jarvinen (2011) conceptualize social capital as comprised by social norms
and resources. Their study analyzes the interaction between the social capital embedded
in peer groups and alcohol use. They cite work showing the importance of the peer group
22
to the individual influences how much a person will adhere to a norm promoting the
consumption of alcohol. Demant and Jarvinen (2011) find that norms, or “‘the rules by
which people coordinate their actions along with systems of sanctions and incentives that
ensure consistence in those actions’”, are important for understanding the use and
sometimes abuse of alcohol when respondents are in the company of peer groups (93).
Demant and Jarvinen (2011) use Coleman’s conceptualization of social capital to inform
their research by “defin[ing] social capital as a ‘resource for action’ related to certain
collectively stipulated norms” (92). However, they do not explicitly acknowledge the
other forms of social capital that Coleman highlights (e.g., information channels and
expectations/obligations).
the norm of reciprocity6 as the operating force that allows for the creation of social
capital. It is through the enforcement of social norms that people can reap benefits that
otherwise would not be available to them. One can think of the expectation/obligation
reciprocity, versus the generalized notion of social norms being a form of social capital.
6
According to Malinowski (1932), reciprocity is where there is a “mutual dependence…realized in the
equivalent arrangement of reciprocal services…” (p. 55). Reciprocity refers to structures of duties,
obligations and expectations within relationships (Gouldner 1960). Putnam (2000) writes that
“[n]etworks involve…mutual obligations…[and they] foster sturdy norms of reciprocity: I’ll do this for
you now, in the expectation that you (or perhaps someone else) will return the favor” (p. 20). Putnam
(2000) further specifies a “norm of generalized reciprocity: I’ll do this for you without expecting
anything specific back from you…[with the] expectation that someone else will do something for me
down the road.” (p. 20-21).
23
Finally, Coleman (1988) points to the norm of altruism7 within communities –
that is, the expectation that actors will in certain situations act morally without
consideration for themselves and motivated by internal social controls to illustrate the
importance of this form of social capital. According to Schneider (2000), “ [i]t is the
interrelationship of individuals, held together through social ties, that Coleman saw as the
conduit through which norms could be transmitted and sanctions imposed” (376). Based
on Demant and Jarvinen (2011), as well as Coleman (1988), social norms in this study,
refers to the shared expectations of behavior for any given social circumstance.
Information Channels
Social capital also lies in what Coleman (1988) calls “information channels”.
These refer to the relations that help people find and collect information. For example, a
graduate student married to an expert in the field of political science goes to him for
information and analysis for both domestic and international politics. Their connection
reduces the need for the graduate student (working very hard on her dissertation) to
friends. They did not know about it as a vacation destination until a friend of theirs met
someone whose family owns a home on the island. In this instance, that social tie
became an information channel and led to a family identifying this island as a vacation
destination (action). While there is no real obligation, these relations can be beneficial in
7
According to Peterson (1980), altruism is considered a “prosocial behavior occurring without external
reward…and internal rather than external control has been suggested in the form of self-congratulations
and increases in self-esteem” (p. 830).
24
that they provide information that could later influence, or lead to, action (Coleman
1988).
used among researchers (Field 2003). More specifically, many refer to his description of
social capital as residing in the ties between people, or in their networks, and in the form
of social capital focuses more on the means used to create social capital and what social
capital is more broadly, rather than being limited to the content of the relation. He
centers his conceptualization of social capital on the structure and function of the creation
According to Coleman (1988, 1990), social capital’s creation comes about through
actors’ actions. Once created, social capital can then take on the different forms outlined
It is easy to get lost in the different forms of social capital, and to lose the idea
that Coleman (1988S) conceives of social structures as social capital. As he notes, ”[t]he
function identified by the concept of ‘social capital; is the value of these aspects of social
structure to actors as resources that they can use to achieve their interests” (S101). Social
capital can take on different forms that are characterized by two parts: First, that social
capital exists within social structures and second, that social capital enables actors within
social structures to take action (Coleman 1990). Writers very often will add the famous
two parts of social capital to the discourse, but then focus on the tie and what lies within
as the unit of analysis. When studying Coleman’s examples and use of social capital it is
clear that simply being a member of a social structure or organization is considered social
25
capital. The actor’s self is not important. In fact, this is why social capital does not lie
with the actor. It is also why it is important that Coleman assumes a rational actor,
because then each actor will act according to the dictates of structural position(s). Taken
to its conclusion, Coleman’s theory of social action leads one to say that without social
structures there can be no action (Coleman 1988S). In fact, Coleman (1988) argues that
social capital is a public good and not privately held, “[t]he public goods quality of most
purposive action than are most other forms of capital” (S118) (see also p.116S).
In sum, Coleman states, “[t]he preceding pages… [focused on the forms of social
capital were] directed toward defining and illustrating social capital in general…[b]oth
social capital in the family and social capital in the community play roles in the creation
of human capital in the rising generation” (1988: S109). Coleman (1988) cited social
capital as one piece of a “family’s background,” along with human capital and financial
capital. For the purposes of the current dissertation project, social capital was the sole
focus, while what Coleman terms human capital (parents education) and financial capital
(income) were linked to the concept of social class. It is important to note that Coleman
includes social class as a part of his larger model, adding to the argument that one should
While the various definitions and writings on social capital have many similarities
(see Bourdieu 1986; Lin 1999; Putnam 2000), the distinguishing qualities of Coleman’s
work are: First, that he brings agency into a field largely dominated by structure; second,
he defines social capital using a structural functionalist’s perspective; and third, that he
conceptualizes social capital in terms of its creation and its role in creating other forms
26
of capital (e.g., human capital in the form of education) (Coleman 1988). As was shown
above, Coleman’s work has and continues to have a large impact on the concept of social
capital and its uses (Graaf and Flap 1988; Alder and Kwon 2002; Putnam 2000).
Both Coleman and Putnam look at resources, trust, and the norm of reciprocity as
integral parts of social capital, but their definitions are distinct in that Coleman’s is less
specified and Putnam takes more time to clearly delineate social capital based on the type
of social tie in terms of content and structure. While Coleman’s strength lies in
understanding the creation of social capital, defining it through its function(s), Putnam’s
strength lies in his conceptualization of social capital formed and shaped by the social
relationship. Putnam distinguishes himself within the social capital literature through his
delineation of social capital in the formation of bonding and bridging respectively. While
Coleman’s three-pronged approach may seem more varied, it is in fact not. Information
channels and expectation/obligation relationships are not that dissimilar and may in fact
happen within the same relationship. Both also focus not on the structure of the network,
not ignored, the focus is on the resources in the network, and how the relationships
perform differently. Bonding, for example, operates from exclusive membership ties and
according to the norm of reciprocity, but with the caveat that it is related to specific, well-
defined reciprocities. In Putnam’s efforts to define these two aspects of social capital, he
looks at it not just from a functional viewpoint, but with a clear idea of what social capital
does for a group and the larger society (and by extension individuals) in regard to
27
more vague and malleable in terms of goals. Bonding and bridging leaves no doubt as to
the role each concept is intended to play. Coleman’s social capital includes the creation
and the action that is involved in producing social capital, whereas Putnam assumes that
an actor (group or individual) is already plugged into a network and has at least some
form of bonding social capital. This works well given that Putnam’s research targets the
Both Putnam and Coleman look at social capital in terms of resources, its
structure, and to some extent, each attempts to assert a typology of social capital. Both
theorists are searching for the framework that will allow them to “see” social capital in all
its various permutations. The most critical difference between the two is that Coleman
focuses more on function and is less specific, whereas Putnam focuses on the content as
well as the structure of the relationship and the relationship’s place in the larger network.
focus was on the social hierarchy embedded in society and understanding how people use
four specific capitals. Bourdieu (1977, 1980, 1984, 2002) conceptualizes social capital
along with three other forms of capital: economic8, cultural9 and symbolic10. The
8
Economic capital refers to the ability of an individual to exchange resources for economic or monetary gain. All
forms of capital are ultimately reducible to this form. Still, all forms of capital work together to reproduce inequality in
society (Field 2003; Bourdieu 1986: p. 252).
9
Cultural capital has three parts, objectified, institutionalized and habitus. Objectified cultural capital refers to material
objects, such as cars, houses, etc. While institutionalized cultural capital are the credentials that one holds, such as
president of the university or professor of sociology. Habitus is a system of internalized dispositions, what Bourdieu
calls tastes or style, through which one perceives and deals with their social world, what is good or bad, right or wrong,
in fashion or not, etc. (Bourdieu 1984, 1980).
28
operation of these forms of capital is framed in the context social space11, fields (see
below) and practice. It is through social practices that one can mobilize social capital or
While all forms of capital are important, this study remains focused solely on
social capital. Consistent with Bourdieuian theory, the concept of fields will be used to
situate social capital in that it is mobilized within particular social contexts or fields.
Bourdieu’s treatment of social capital was not as thorough as it could have been, and the
concept itself is certainly not as fully conceptualized as cultural capital. Therefore, the
current study may be of use to scholars interested in adding more complexity and
capital. Although Bourdieu theorized that the different forms of capital were linked to
one another, the purpose of the current project is to delve deeper into the
conceptualization of social capital and not to serve as a general test Bourdieu’s theory of
inequality.
The reason that Bourdieu is included in the current study is that his approach to
social capital brings with it a refined theory of inequality and a clear understanding of the
ways in which social context may influence how social capital influences the
10
Symbolic capital refers to the honor or prestige that one has and is measured by how much social, economic and
cultural capital one holds. For example, if one has high economic capital, high social capital and high cultural capital,
then one can be said to have high symbolic capital (Bourdieu 1984, 1986, 2002).
11
Social space is a methodological tool whereby one can analyze the stratification of society. Depending on where a
position is in relation to another will predict how they relate to one another. The way in which they interact or relate
depends on their ability to exchange capital. The important benefit of social space is the ability to measure the distance
between positions. This distance informs us whether and how much positions will interrelate (Bourdieu 1984, 2002).
However, social space is composed of cultural and economic capital. If anything, social space can frame social capital,
not the other way around. Moreover, there is no way to fully measure the rich and multilevel concept of cultural capital
with the proposed dataset. Therefore, social space is not included. Although, the current study has the potential to add
to the larger idea of social space by showing that socio-demographic variables (two of which were not directly
addressed in Bourdieu’s theory- race and gender) do impact social capital. While habitus encompasses race, gender
and class in terms of experiences that are similar or different depending on one’s position in social space, race and
gender are not formally included in Bourdieu’s theory of inequality.
29
reproduction of inequality. Further, one of the weaknesses within Bourdieu’s theory of
inequality is his conceptualization of social capital. Scholars such as Field (2003) and
Portes (1998) raise questions about the fact that so much of the published work on social
capital does not include a consideration of Bourdieu. This may be due to his brevity on
the subject. While Bourdieu (1986) writes that social capital is an important form of
capital and should be included, he provides little more in the way of detailed
conceptualization.
Bourdieu (1984) views social capital in the context of the position that one holds
in a field within social space. Fields are patterns of relations between positions (e.g.,
political field, religious, economic, educational, health, family, etc.). Whether or not a
form of capital can be exchanged largely depends on the field. A field provides a context
to interactions and thus becomes a part of its meaning. Thus, fields play an important
role in the market place of relations (Bourdieu 1986). Moreover, Bourdieu (1980) argues
that social capital is people’s key to obtaining varying degrees of economic and cultural
friends, acquaintances, school and business contacts that enables actors to transform
social capital into a resource that will lead to achieving personal and social goals
(Bourdieu 1980).
In a general sense, capital for Bourdieu (1986) refers to the possessions, qualities
and attributes of a person or position that are exchangeable within a field. Social capital
refers to the relations between positions that are meaningful (Bourdieu 1986). In his later
30
[s]ocial capital is the sum of resources, actual or virtual, that accrue to an
individual or a group by virtue of possessing a durable network of more or less
institutionalized relationships of mutual acquaintance and recognition (Bourdieu
and Wacquant 1992: 119).
Bourdieu’s theory has roots in a Marxist view of economic capital and labor as
well as Weberian leanings with regard to power. The former can be seen in Bourdieu’s
belief that “economic capital is at the root of all other types of capital” (Bourdieu 1986:
252). Accordingly, it was through people’s labor that social capital (and all capital) was
accumulated (Field 2003). However, focusing solely on economic capital was not
enough, like Weber he was concerned with more than that. Bourdieu (1986) was
particularly interested in how the combination of the four forms of capital helped to
of the different capitals that we can see how inequality happens and continues (Bourdieu
1986).
Bourdieu (1977, 1980, 1984, 2002) delved deeply into the conceptualizations of
cultural capital, economic capital and symbolic capital. While he did use and talk about
the idea of social capita, only one piece of his writing focuses on it specifically; (see
Bourdieu 1986). Because of this, the treatment and conceptualization of social capital is
less thorough and sophisticated than with other forms. Social capital was more of an
offshoot from cultural capital, a tertiary piece of his puzzle (Field 2003). Cultural capital,
particularly habitus, and economic capital played much bigger roles than social capital in
Bourdieu’s (1980, 1984) larger theoretical framework (Field 2003). However, where
Bourdieu was ahead of his time was in terms of treating social capital as a concept that is
influenced by context (both social capital itself (form) and its individual owner
(content)).
31
In Bourdieu’s (1984, 1986, 2002) theory of inequality, which he sees being
resources available to obtaining it. The value of social capital is thus somewhat
dependent upon the actors with whom it is associated. If, for example, someone has a
high social status, a lot of money, and higher education from a highly ranked institution,
(e.g., Harvard University) then that person has many resources of value at their disposal.
Contrast that life situation with that of someone who has no higher education, lives pay-
check to pay-check, and thus has a lower social status. While the latter actor may have
many social ties, and even some social capital, it is qualitatively different from that of the
former actor whose social resources, according to societal standards, have more value.
The basic idea here is that pre-existing conditions affect the quality of one’s social capital
which, in turn, leads to a lower position in the social hierarchy (Bourdieu 1986).
Both Coleman (1988) and Bourdieu (1980, 1986, 2002) view the context or the
structure of relationships as important. As Coleman puts it, certain forms of social capital
are not useful and can indeed be harmful depending on the circumstance. The network of
relationships and how these operate is explicated in Coleman’s theory of social action
(1990). The ways in which individuals within a network of relations can have power
over another is based on social closure.12 Overall, when isolating the concept of social
12
The inequality in positions is only exacerbated by what Weber (1978) called “social closure.” Social
closure refers to the means undertaken by those in higher, more prestigious positions to insulate their
positions. Specifically, it refers to the “social groups formed around positions in the technical division of
labor [that] create social and legal barriers that restrict ‘access to resources and opportunities to a limited
32
capital, it is clear that both Putnam (2000) and Coleman (1988) are more explicit
conceptually than Bourdieu (1988). Coleman places the actor within a structure of
networks and institutions that inform actors’ relations and thus their social capital while
Putnam clearly outlines two forms of social capital (i.e. bonding and bridging) that are
based on the content of the relation itself. In fact, Putnam’s clear outline of social capital
micro actor. Finally, both Coleman and Putnam’s more detailed theoretical descriptions
of social capital can be operationalized more easily than Bourdieu’s. These similarities
and differences are illustrated in how each approach to social capital is operationalized
Bourdieu’s theory of inequality takes into account much more of the social world
than either Coleman’s or Putnam’s. As Silva and Edwards (2007:3) notes, one of the key
Economic capital does not work like cultural capital or social capital. We
accumulate and invest in all forms of capital, yet the effects of accumulation and
investment are not the same throughout. In addition, they must be thought of not
simply in terms of accumulation or investment processes because power and
control are conferred and legitimised through particular possessions of capital
(Silva and Edwards 2007: 3).
Power is linked to the position an actor occupies in the social space. It is an important
difference to note because with the inclusion of such concepts as social space, field,
capital, and the fungibility of capitals from one to another, Bourdieu simultaneously
circle of eligibles’” (Weeden 2002: p. 57). There are four different mechanisms that are used to create
social closure. First, the labor supply for an occupation is limited (this can be done by making it difficult to
obtain the credentials necessary e.g. becoming a college professor). Second, working to increase the
demand for the services that an occupation provides. Third, legitimizing that a particular occupation is the
only one that can truly provide that service. And finally, communicating to consumers that the services are
of a certain quality (Weeden 2002).
33
acknowledges the import of social networks and, at the same time, that how these
networks come to affect actors are not equal because of differential access to other forms
of capital. So, simply having a network or belonging to an organization does not mean
that each actor shares the same choices, power, or resources as another (Bourdieu; Silva
the “invisible force” that influences one’s occupational outcomes (Lin 2004). Social
capital, its possibilities, drawbacks and influences, are of critical importance for
understanding how one can not only “climb” the social strata, but also combat the
deleterious effects of inequality tied to poverty, race, and gender. However, the question
that has not yet been explored with regard to these three theoretical approaches to social
capital is how they may vary once potential differences across demographic groups are
The term social capital is found throughout the community and network
literatures (e.g. Coleman 1988; and Putnam 2000) as well as within the work of social
inequality theorists interested in social class (Bourdieu 1980; Lin 2001). More recently
researchers have called for contextual variables to be considered when looking at social
capital (Aguilera 2002; Briggs 1998; Campbell 2000; Gittell et al 2000; Goulbourne and
Solomos 2003; Norris and Inglehart 2006). If so, then race and gender (Campbell 2000;
(2008:240) observe:
34
Certainly, studies have found that gender, class or race do often define a group’s
status, access to social capital resources, and thus, the extent of interactions and
resource exchanges...
In what follows I will show how class, race, and gender are important components for
and mobility (Burt 1992; Gabbay and Zuckerman 1998), to finding a job (Granovetter
1973, 1995; Lin and Dumin 1996), and within the workplace social capital has also been
associated with decreased turn-over (Krackhardt and Hanson 1993). Further, researchers
(Blau 1994; Burt 1976; Friedkin 1980, 1981, 1982; Granovetter 1973, 1995; Prell and
Skvoretz 2008, etc.) have discussed the importance of resources within ties, between
nodes, and the structural importance of a network, including the distance between nodes.
Much of this discussion is enmeshed in studies on bridges in networks, weak ties, and
gaining status either through work, birth, and friendships, or obtaining a higher status by
aligning oneself with someone who already has a higher status in the community.
Within the class literature, two of the main areas that use social capital are studies
focused on occupations/occupational prestige and social mobility (De Graaf and Flap
1988; Lin 2004; Marsden and Hulbert 1988). One other major area in which both class
and social capital are used concurrently is the educational literature (Coleman 1988).
Regardless of the area, scholars agree that social capital has a part to play in the social
35
Studies have shown that social capital positively influences occupational prestige,
although not always income (De Graaf and Flap 1988; Lin 2004; Marsden and Hulbert
1988). For example, De Graaf and Flap (1988) point to the importance of one’s social
resources in finding a job in the Netherlands, West Germany, and the United States. In
this study, the authors define social capital as “[s]omeone’s resources [which] are the
result of the number of people who want to help him/her, the resources that he/she can
mobilize in this indirect way, and the extent to which others are prepared to give support”
(De Graaf and Flap 1988: 453). Of particular interest is that the occupational prestige of
the person who serves as the social resource impacts the prestige of the position one
finds. This finding is more pronounced in the U.S. than in the Netherlands. However, the
impact of the occupational prestige of the social contact on income was shown to be non-
Marsden and Hulbert (1988) also report that the occupational prestige of a contact
is associated with the prestige of the respondents. Furthermore, both the person’s
previous educational attainment and occupational prestige impact the kind of contacts
that he/she will have. This suggests that mobility through social contacts, or social
capital may be difficult on a larger scale (e.g. going from a lower occupational
Along the same lines as Coleman (1988), Granovetter (1973) looks at social
capital through the ties that one has with others and specifically he looks at the
differences in tie strength. His major finding is that weak ties, those characterized by less
time and energy (what Putnam would refer to as bridging social capital), can lead to
finding a job. The ability of acquaintances, or weak ties, to help one find a job tends to
36
be simply through the interaction and exchange of information. The reason that
Granovetter (1973) gives for this finding is that weak ties can lead to broader networks.
However, Granovetter (1973) does not address whether weak ties will lead to higher
occupational prestige or status. McPherson et al. (2006) find in their study that these ties
are actually shrinking. This could have interesting and possibly negative consequences
on people’s ability to not only find jobs, but jobs that have higher occupational prestige.
The focus of most studies looking at class inequalities and social capital surrounds
issues of social mobility (e.g. education, status, and/or income). Variations exist both in
the quality and quantity of social capital that falls along class lines (Parks-Yancy et al.
2008). However, few acknowledge that these differences (in quantity or quality, form
experience or meaning and, as a result, such differences are not currently accounted for in
who argue that there are fewer exchanges between higher and lower classes because
those within higher social classes tend to be more exclusionary (protecting their status in
society) and to hoard their resources. By contrast, those in the lower social classes have
(Coleman 1988) relationship and the potential investment of social capital even if a tie
does exist. As Willis (1977) observed, people in a lower social class have less social
capital and their social capital is not as good in terms of the quality of resource that they
can offer. In sum, to obtain a complete picture of social capital, measures of social class
37
Racial Differences and Social Capital
less social capital and gain fewer rewards from the social capital that they possess
(Castilla 2008; Ibarra 1993, 1995; Petersen et al. 2000). Putnam (2003) also reports
finding lower levels of social capital and economic resources in communities with a
higher level of ethnic diversity (also cited in Silva and Edward, 2007).
Without using the term social capital explicitly, Stack’s (1974) work on kin and
community speaks to this topic, particularly with respect to bridging and bonding social
The model of a cooperative life style built upon exchange and reciprocity as
described in the present study represents one dimension of the multivalued
cultural system, the value-mosaic of the poor…black urban poor, assuming a
cooperative life style, are simultaneously locked into an intimate, ongoing bond
with white culture and white values.
To clarify, in her research Stack (1974) finds that the relationships minorities in
the inner city have with one another are characterized by a social norm of sharing
resources (social, economic, and so on). The norm of sharing resources is important to
the survival of the individuals in the community and to their children’s well-being.
However, there can be negative consequences with regard to this norm. In fact, those
couples or individuals who were able to “get out” of the inner-city tended to remove
One example Stack (1974) provides is of a married couple who came into a bit of
money (enough to possibly buy a house or make a good investment for their retirement).
However, within no time at all, this information had been disseminated throughout the
community and neighbors were coming to access bonding social capital in the form of
38
help with rent, shoes for their children, food, and so on. Within a couple of months, the
entire inheritance was gone. Again, this highlights the importance of looking at context
in the form of structural location (class and race) as well as at the social norms within
communities that help to shape the type of social capital that is formed within networks.
It also supports the assertion that social capital is different depending on such socio-
Smith (2005) reports similar findings to Stack (1974) when looking at the effect
of weak and strong ties on social mobility among the African American working poor.
Whereas most research shows that weak ties (Granovetter 1973, 1995) are more useful
than strong ties for finding better jobs, Smith (2005) finds that neither strong nor weak
ties aid working poor African Americans in “climbing” the social ladder. Using Putnam’s
(2000) bonding and bridging terminology, neither bonding nor bridging social capital is
useful.
class, race and kin networks, does not find support for what Stack (1974) found in her
study, that minorities draw on their networks for support to offset poverty-related effects.
These findings are particularly relevant to the current discussion of social capital in that
they raise questions about the assumption that simply being in a social network means
that you have access to social capital. Further, it is clear in Roschelle’s (1997) study that
age, race, gender, and class all play a part in network participation and social capital. It
seems then that race does affect social capital and new insights might be gained were this
to be tested directly.
39
In his work on social capital and labor participation, Aguilera (2002) tested for
and diversity. Because he utilized the 2000 SCCBS he was also able to examine social
capital in regard to racial and gender differences. He found that friendship networks
were generally positive forces for jobs and workforce participation. Aguilera (2002) also
reported both racial and gender differences in social capital. For race, he found that
blacks with friends who are leaders are nearly “four times more likely to be employed as
compared to those without leader friends” (Aguilera 2002: 869). Moreover, blacks with
more than six friends worked 17 percent more hours than those with fewer friends.
Aguilera also tested for the impact of social capital by race and found that for blacks who
experience discrimination in the labor market there was a “higher return” in the number
of hours worked and having social capital via ‘leader friends’ when compared to whites.
Gender also impacts social capital with regard to the type of social capital and the
amount (Parks-Yancy et al. 2008). Research shows that men generally have more access
to social capital than women and also experience more for their social capital “buck” than
women (Campbell et al. 1988; DiTomaso et al. 2007; Lin 2001). For example, Wellman
and Frank (2001) state that women as opposed to men are more likely to participate in
exchanges of social support13. Moreover, if more women are present in a network, then
those actors in the network (men and women) will receive more social support. Women
are also more likely to “deal” in social capital that is characterized by or related to
13
Lin (2001) uses the terms social capital and social support interchangeably, but it is not clear that these
two terms are equivalent and most researchers using social capital do not use both terms or see them as
the same.
40
emotions. In contrast, men are more likely to have, sustain, and access social capital that
is more instrumental in nature. As Wellman and Frank (2001: 253) note, women tend to
relate “by exchanging emotional support while men interact ‘side by side’ by exchanging
Several authors in the edited volume, Gender and Social Capital (O’Neill and
Gidengil, 2006), focus on refuting Putnam’s (2000) claims that women’s entry into the
workforce has led to a decrease in civic participation (i.e., social capital). These chapters
bring in gender as a component of civic life, specifically with regard to social capital.
important component of the model, but do not conceive of “gendered” social capital.
One exception is Norris and Inglehart (2006). Norris and Inglehart (2006) look at
Putnam’s treatment of gender and social capital (i.e., that women’s move into the work
force greatly affected civic participation) and report three important findings. First, that
sex-segregation in key civic groups does influence the social capital that women can
build in the political field. Second, there is a “gender gap” in memberships with regard to
associations. The authors cite this finding as evidence that gender should be included
when considering social capital and that it is not a “gender-neutral phenomenon” (Norris
and Inglehart 2006: 93). And finally, they report that informal networks influenced the
types of social capital that women had compared to men (this finding came when
considering the gender gap in formal associations). The point here is that men and
women experience social capital very differently (Norris and Inglehart 2006). Therefore,
41
Aguilera (2002) found, with the SCCBS, that there were some gender differences
when using a general conceptualization of social capital. For one, women who tend to
have more diverse friendships also work more hours. For the most part social capital had
a positive impact on hours worked for both men and women. However, white women
with leader friends were less likely to have jobs. Further, white women with friends who
own their own business worked fewer hours than white women who did not have that
type of friendship. Finally, women in this study were found to have benefits from having
gender and found that black women with friends who were on welfare tended to not have
jobs when compared to black women without friends on welfare. White women who
were engaged in a social group were six times more likely to have a job than those who
were not. White males benefitted most from having a friend who owned their own
business. The only groups that had any positive effects from the number of hours they
worked were black men and white women. There were positive effects for black men
who knew manual workers. White women who had more than six friends worked longer
hours, and those who had diverse networks also tended to work more hours. However,
white women who had a personal friend that owned their own business worked less
hours.
social capital is not the same for everyone. It varies across class, race, and gender
groups. Thus the distinctions should be reflected in social capital’s conception as well as
42
Research Questions and Hypotheses
social capital, the first empirical issues I address concerns the extent to which the three
theoretical frameworks are supported empirically and whether social capital differs across
Many assume that these conceptualizations of social capital are not only
essentially similar, but can be modeled empirically. In contrast, the current study
truism to be assumed. Thus, it remains to be seen whether the three theories of social
capital can each be modeled empirically. In order to answer ‘yes’ to Research Question
One, each social capital model should be able to be specified and each of the models
Research Question Two: Does the structure of social capital operate differently
As shown in the previous section, researchers have shown that social capital can
operate differently across class, race, and gender groups. Most have focused on the
differing levels of social capital for these groups, however, the current study focuses on
the structure of social capital. In expecting that the “structure” of social capital operates
differently across class, racial, and gender groups, I am suggesting that the underlying
structure varies as well. Focusing on the structure of social capital reveals for each group
how social capital is captured for each group. If there are group differences, then it can
43
be argued that for the group with larger parameter estimates that aspect of social capital is
more important. Another way to interpret these results is to think of that aspect of social
capital manifesting differently for one group versus another group. This structural
analysis will be conducted for class, race, and gender groups. In order for any of the
following hypotheses to be supported, a Likelihood ratio test will be used. If the test is
does not relate directly to social capital, Coleman and Bourdieu argue that in different
ways different forms of capital can be used to attain other forms. Bourdieu (1997) in
particular views all forms of capital as fungible and ultimately tied to economic capital.
Putnam focuses less on economic resources, but does not completely disregard them.
Moreover, the literature examining social capital in general terms identifies potentially
important differences by income. Not only does social capital appear to operate
differently depending on income, but there is substantial evidence for both racial and
gender variation in social capital as well (Stack 1974; Rochelle 1997; Smith 2005;
Aguilera 2002; Norris and Inglehardt 2006; O’Neill and Gidengil, 2006; Wellman and
Frank 2001;Parks-Yancy et al. 2008). Therefore, I hypothesize the following for each of
the three social capital models (i.e., that of Putnam, Coleman, and Bourdieu):
Hypothesis One: The structure of social capital operates differently across lower,
Hypothesis Two: The structure of social capital operates differently across White
44
Hypothesis Three: The structure of social capital operates differently for men and
women.
45
CHAPTER III
University (2000). Robert Putnam was a member of the team that compiled the survey
instrument and measures therein. It is possible then that his involvement may bias how
the questions were worded in the survey as well as the types of questions asked. The
major strength and what makes the SCCBS truly unique was the heavy emphasis on
social capital. Its measures were specifically designed to capture the different aspects of
related to class, race and gender were gathered. Therefore, it was uniquely suitable to
To date there were no other datasets, including the GSS, which could compare to
the SCCBS in terms of its ability to measure different facets of social capital (Messner,
Baumer, and Rosenfeld 2004). Messner et al (2004) do point out however, that one
limitation of the SCCBS was its low response rate. About 42% of those who were
reached via telephone actually completed the survey, “which reflects 29 percent of all the
46
The data for the survey were gathered using telephone surveys across the
continental United States in 2000 and included respondents 18 and older. Proportionate
random sampling was used in the selection of households. Certain questions or indicators
were asked for a subset of the sample randomly selected. Due to the random variation
and the fact that the indicators changed slightly from model to model, the actual sample
size (N) for the following three social capital models varied. Using this dataset meant
using a cross-section for analysis. As such the data is tied to a “moment” in time and
space.
The sample for the entire SCCBS, including regional selections, was 26,000.
However, the national sample size was 3,003 respondents and included an over-sampling
of black and Hispanic respondents (500 people from each group). The goal of this study
was to look at broader theoretical topics, not look across specific regions, thus the
national sample was used in the analysis. A full listing of the questions and responses
used in the study can be found in Appendix A, separated by theorists, for ease of
comparison.
The SCCBS dataset has grown in its use by researchers. For example, Aguilera
(2002) used the SCCBS for 2000 in order to assess the relationship between people’s
networks, specifically friendship networks, and work. His study helped to elucidate
racial differences in the relationship between different groups’ labor (hours worked) and
social capital as well as provide support for bringing in more programs with important
information on jobs to disadvantaged communities. Rahn and Rudolph (2005) also used
the SCCBS for their study on public trust, specifically looking at trust in the government
47
at various levels. Messner, Baumer and Rosenfeld (2004) used Putnam’s
Missing Data
amount of bias (Allison 1999). As Allison (1999) notes, randomly distributed missing
data does lead to as many problems of bias compared to that which is systematically
related to the variables being examined. However, it matters whether the missing in
random (MNAR). The latter two introduce significant bias, whereas the first does not
(Allison 1999).
Missing completely at random means that the missing was not related to any other
variable. Thus, we can say that while there was missing, because it was missing
completely at random, it does not bias the results. Listwise deletion, pairwise deletion
and missing values replacements are common ways to address MCAR (Bollen 1989).
The worst that MCAR could do when used in statistical analysis would be to lessen the
power in the model, especially if the MCAR were to be on the dependent variable(s)
(Howell 2007).
here the missing would be seen as random after controlling for another variable. MAR,
while not as good as MCAR, can still be considered ignorable missing (Howell 2007).
2007), here a clear pattern exists. MNAR must be carefully managed because it causes
48
biased parameter estimates and affects the power of the model (Allison 1999; Bollen
1989; Howell 2007). One of the most common ways to address MNAR would be to use
a Heckman selection model or variant that creates “a new ‘selection equation’ to model
which cases are included or excluded from the sample [where] [t]he equation is
There are many different ways to deal with missing data (Allison 1999; Bollen
1989; Howell 2007). Listwise deletion, for example, removes any cases with missing
information from the analysis. As such it is also known as complete case analysis
pairwise deletion or available case analysis, which allows for partial use of the case.
However, Allison (1999) discourages researchers from using this method or replacing the
missing values with the means (Allison 1999). Other benefits of using listwise deletion
include the use of complete cases, and the production of unbiased parameter estimates
when the missing are completely at random (MCAR). The major limitation to this
method is the potential decrease to sample size. Decreasing the sample size has an
impact on the statistical power, however, as long as the sample size is still substantial,
In the current study, tests were conducted to see if the missing data was random
across cases or were patterned in some way. Three variables had more than 30% of the
sample missing, they were, perception of community trust (missing 1505), worked with
others (missing 1497), and coworker contact (missing 985). The first two variables were
part of a larger skip pattern within the dataset.14 The pattern was administered
14
Due to budget constraints and in an effort to shorten the interview length, certain sections and in some
cases questions within the SCCBS were “administered to randomly selected halves of the sample…the
49
completely at random and consistently throughout the sample, thus the missing can be
Dummy variables indicating “missing” for coworker contact were created and
then cross-tabulated with income, race, and gender using the chi-square significance test
to indicate if the pattern was missing completely at random (MCAR) or missing not at
random (MNAR). When coworker contact was combined with income the chi-square
was 43.6277 with 8 degrees of freedom and was significant at the p< 0.0001 level; with
race the chi-square was 9.6373 with 4 degrees of freedom and was significant the 0.05
level; and finally when coworker contact was combined with gender, the chi-square was
16.7609 with 4 degrees of freedom and was statistically significant at p<.01 level.
Based on chi-square tests, combined with the data instructions for skip-patterns
performed at random in collection, it was determined that the missing data were not
MCAR. A significant chi-square indicates that there was some association between the
row’s and column’s variables that were selectively missing. While the study cannot claim
that the missing were MCAR, following Allison (1999), listwise deletion was used for
the study. 15 Some bias may have been introduced because the missing was not MCAR.
The sample size varied for the analysis depending on the social capital model as well as
the type of analysis being conducted, however, quality analysis in Structural Equation
Modeling in SAS needed a minimum of 200 complete cases and that was easily achieved
(Hatcher 1994).
probability was 50% that any respondent would be administered each section” (SCCBS 2000: p. 9).
There were six different forms and each respondent was randomly given one of the versions of the
survey to answer. Each form took essentially the same amount of time to administer.
15
Listwise deletion allows for the use of complete cases and does not bias parameter estimates when
MCAR (Howell 2007; Bollen 1989).
50
Measurement
This section operationalizes the indicators used in each model. In the following
sections I detail the measures used for each domain and the independent variables.
Descriptive statistics for the measures detailed below can be found in Chapter 4. Finally,
I present the analytical strategies used in the two results chapters (Chapter 4 and Chapter
5).
The indicators used to measure social capital were selected based on the
capital theorists Robert Putnam (1993, 2000), James Coleman (1988, 1990), and Pierre
data, and exploratory factor analysis (see Chapter 4 for results). The theoretical
corresponding empirical models that were then used to test the model fit for each
capital was comprised of trust (thin and thick) and social ties (the social resources that lie
therein), in the form of memberships, friendships, and altruistic actions. For Putnam
(1993, 2000), having people who volunteer, were involved in many groups and
organizations, and people willing to make things better for those they knew or their
51
Putnam (2000) argued that those with social capital will be “joiners” or people
who are active in groups. Being active in groups included mere membership or
participation in group activities, or both.16 Following are the specific indicators for first
bonding social capital, then bridging social capital, and finally trust.
(ties) that have certain characteristics, such as in-group, specific reciprocity, strong ties,
close relationships, and one where people are more similar than different (Putnam 2000).
At the empirical level this translates into family members, close friends, memberships in
shared confidences, and family contact. Close friends was measured by asking the
respondents how many close friends they currently have. Specifically, respondents were
asked “Now, how about friends? About how many CLOSE FRIENDS do you have these
days? These are people you feel at ease with, can talk to about private matters, or call on
for help. Would you say that you have no close friends, one or two, three to five, six to
ten, or more than that?” Responses were coded one for no close friends, two for one to
16
The idea of joining was important to both bonding and bridging social capital, but in different ways. For
bonding, joining was based on similarities between members and on stronger ties. However, bridging
focused more on weak ties built on diverse relationships (Putnam 2000).
17
When the EFA was run using Knudsen’s conceptualization of inclusive versus exclusive groups, or
separating out group membership between bonding and bridging social capital the Cronbach’s Alpha
were very weak suggesting that neither scale was reliable. Moreover, when run in the EFA they both
loaded on the same factor, showing no support for the argument that they are measuring different
aspects of social capital. Thus, the groupinvolvment index was used and ended up measuring bridging
social capital. Perhaps if measures denoting more specifics, time spent, feelings of closeness, one could
make a distinction between those ties that are bonding within groups and those that are bridging.
52
two close friends, three for three to five close friends, four for six to 10 close friends, and
Share confidences looked at the number of ties that a respondent had with whom
they felt comfortable sharing intimate or personal information about themselves or their
experiences. Respondents were asked “Right now, how many people do you have in
your life with whom you can share confidences or discuss a difficult decision – nobody,
one, two, or three or more?” Response categories were coded (1) nobody, (2) for one, (3)
Finally, family contact asked respondents how often they go to visit their family.
The continuous variable was used in all analysis ranging from 0 (none) to 53
(respondents who stated more than 53 were coded as 53). For this question respondents
were asked “(How many times in the past 12 months have you) visited relatives in person
or had them visit you?” The interviewer was asked to give the respondent time to
Bridging social capital focused on the less close ties, or “weak” ties19 (Putnam
2000). According to Putnam (2000), by focusing on ties that are not as integrated in
one’s personal network, then one is able to connect to other networks and branch out.
Putnam views diversity as a way to see these bridging ties (2000). The domain of
bridging social capital used three indicators, diversity, group involvement scale, and
worked with others. The diversity scale was comprised from a series of questions that
asked respondents the following broad prompt followed by specific questions, “[t]hinking
now about everyone that you would count as a PERSONAL FRIEND, not just your
18
In the SCCBS all responses of “don’t know” were coded as 8, 88, or 888; and all refusals were coded as
9, 99, or 999.
19
See Granovetter’s seminal work or previous discussion in chapter 2.
53
closest friends- do you have a personal friend who…” “owns their own business”, “is a
manual worker”, “has been on welfare”, “owns a vacation home”, “has a different
religion than you”, “is white”, “is Latino or Hispanic”, “is Asian”, “is Black or African
American”, “is Gay or Lesbian”, “you would describe as a community leader”. Each of
the 11 questions in the series (see Appendix A) were dichotomous, yes (1) or no (0) and
combined into a diversity scale that ranged from 0 to 11 where 0 was no diversity in the
network. The Cronbach’s Alpha of reliability is a test for measurement error and
reliability on a scale or index that ranges from zero to one, and for this scale the alpha
was 0.73.20 According to Allison (1999) an alpha of reliability above 0.70 is considered
an acceptable alpha of reliability for the social sciences. Essentially what this statistic
tells us is that 30 percent of the variance for this scale was measurement error, or stated
another way, 70 percent of the variance for this scale was “true” (Allison 1999).
The group involvement index used the question 33 series which measured the
different group memberships of respondents. Each respondent was prompted with the
following and then asked about specific groups and organizations, “Now I’d like to ask
about other kinds of groups and organizations. I’m going to read a list; just answer YES
if you have been involved in the past 12 months with this kind of group.” The indicators
included on the group involvement scale were as follows: “An adult sports club or league,
or an outdoor activity club”, Besides your local place of worship, any organization
affiliated with religion, such as the Knights of Columbus or B’nai B’rith, or a bible study
group”, “a youth organization like youth sports leagues, the scouts, 4-H clubs, and Boys
& Girls Clubs”, “a parents’ association, like the PTA or PTO, or other school support or
20
The standardized results and raw Cronbach’s alphas did not vary greatly for any index or scale in the
sample, however, all results reported are standardized.
54
service groups”, “a veteran’s group”, “a neighborhood association, like a block
organization that provides services in such fields as health or service to the needy”, “a
fraternal organizations such as the Lions or Kiwanis or a local women’s club or a college
National Organization for Women, the Mexican American Legal Defense or the
NAACP”, “other public interest groups, political action groups, political clubs, or party
group”, “any other hobby, investment, or garden clubs or societies”, “a support group or
self-help program for people with specific illnesses, disabilities, problems, or addictions,
or for their families”, “are you involved in any group that meets only over the Internet”,
“and do you belong to any other kinds of clubs or organizations?” Each of the responses
for the former indicators were coded (0) for no, and (1) for yes. The scale ranged from
zero to 18 where the higher the score, the more involved the respondent was in groups.
The alpha of reliability for this scale was 0.72, this again was considered acceptable and
thus the indicator was retained for this domain (Allison 1999).
Finally, worked with others asked respondents whether or not they had worked
respondents were asked “In the past two years, have you worked with others to get people
55
Finally, for the trust domain, perception of community trust was one of three
questions used to measure trust. This variable was measured by asking respondents
whether they felt that people in their community would restrict their use of water or other
shortage. Respondents could reply with very likely (5), likely (4), depends (3), unlikely
(2), or very unlikely (1). Whereas, rate your community was a general measure of the
respondent’s view of their community from poor, only fair, good, or excellent (0 – 3).
The social trust index was composed of six trust questions that included general
trust, trust of one’s neighbors, trust of co-workers, trust of fellow congregants, trust of
store employees where they shopped, and trust of the local police. This was an index
created by the SCCBS that ranged from -2.49 to 1.01 and included the Question 7 series
that asked respondents whether they felt that people in different areas of their life could
be trusted or not, such as at work, places one shops, in general, or in one’s neighborhood.
A minimum of three answers from the six indicators had to be present in order for a
calculated score.
The survey first asked respondents “Now, I want to ask you some questions about
how you view other people. Generally speaking, would you say that most people can be
trusted or that you can’t be too careful in dealing with people?” Possible responses for
this measure included, people can be trust (1), you can’t be too careful (2), and depends
(3). After which respondents were prompted with “[n]ext, we’d like to know how much
you trust different groups of people. First, think about (GROUP). Generally speaking,
would you say that you can trust them a lot, some, only a little, or not at all?” Then,
respondents were asked about trust related to specific groups, including “people in your
56
neighborhood”, “people you work with”, people at your church or place of worship”,
“people who work in the stores where you shop”, and “the police in your local
community”. The index was constructed using standardized responses to the questions.
The alpha of reliability for this index was fairly strong at 0.81, meaning that only 19
percent of the variance shown in the analysis was measurement error (Allison 1999).
Coleman’s social capital model. The following is a list of indicators from the
Social Capital Community Benchmark Survey (SCCBS) that fit the three parts of
Coleman’s theory of social capital. Nine specific questions from the SCCBS are included
interactions within social structures and more specifically social exchanges based on trust
and reciprocity. Of the three domains of social capital that Coleman (1988, 1990)
described, this form was the most frequently discussed in the literature (E/O). Trust
underlies all three forms of social capital, but was included as a proxy for social norms
(Coleman 1988). This section details the measures used for the domains of
(SN).
33 series) was created using 18 questions that asked respondents to answer yes (1) or no
(0) to being a member of certain types of groups in the past 12 months. Each respondent
was prompted with the following and then asked about specific groups and organizations,
“Now I’d like to ask about other kinds of groups and organizations. I’m going to read a
57
list; just answer YES if you have been involved in the past 12 months with this kind of
group.” The indicators included on the group involvement scale were as follows: “An
adult sports club or league, or an outdoor activity club”, Besides your local place of
worship, any organization affiliated with religion, such as the Knights of Columbus or
B’nai B’rith, or a bible study group”, “a youth organization like youth sports leagues, the
scouts, 4-H clubs, and Boys & Girls Clubs”, “a parents’ association, like the PTA or
group”, clubs or organizations for senior citizens or older people”, “a charity or social
welfare organization that provides services in such fields as health or service to the
clubs or fraternal organizations such as the Lions or Kiwanis or a local women’s club or a
the National Organization for Women, the Mexican American Legal Defense or the
NAACP”, “other public interest groups, political action groups, political clubs, or party
group”, “any other hobby, investment, or garden clubs or societies”, “a support group or
self-help program for people with specific illnesses, disabilities, problems, or addictions,
or for their families”, “are you involved in any group that meets only over the Internet”,
“and do you belong to any other kinds of clubs or organizations?” Each of the responses
for the former indicators were coded (0) for no, and (1) for yes. The scale ranged from
zero to 18 where the higher the score, the more involved the respondent was in groups.
58
The Cronbach’s Alpha of reliability for this index was .72, which was deemed acceptable
(Allison 1999).
The second measure for the E/O domain was other church participation. This
indicator looked at whether or not respondents had participated in any church functions
categories were yes (1) or no (0). Specifically participants in the study were asked “In
the past 12 months, have you taken part in any sort of activity with people at your church
or place of worship other than attending services? This might include teaching Sunday
The final measure for the E/O domain was club meetings. This asked respondents
how often in the past 12 months they had attended a club meeting, specifically, “(How
many times in the past twelve months have you) attended a club meeting?” The
responses ranged from 0 to 53, anything over 53 times was capped at 53.
questions were included that were in line theoretically with Coleman’s second form of
social capital, information channels (IC). He specifically wrote about IC’s working as a
tie where the actor had access to information they might otherwise not have time to find
or any access to without that particular connection. Any type of information that would
be considered useful was included in the conceptualization, including but not limited to
political, social, medical, etc. (Coleman 1988). Therefore, the measures chosen for the
IC domain were based on ties that encompassed specific information, such as fictive kin,
59
Fictive kin measures the number of other adults that the respondent can name who
s/he considers to be “like family”, “How many other adults, if any, do you treat as
members of your family even though they are not related to you? (IF NECESSARY:
These are people who are regularly included in family gatherings and celebrations, and
who may be called "uncle" or "aunt" although they are not.” The responses were coded 0
to 10 where 0 means there was no one to fill that role and 10 was a category for 10 or
Respondents were asked in two separate questions whether they had a personal
friend who (1) owned their own business (business owner) “has personal friend who own
a business” or (2) owned their own vacation home “has a personal friend who owns a
Lastly, for the social norms domain, Coleman (1988) alluded to trust while
discussing social norms as a form of social capital, but did not explicitly write that trust
equals social norms. This project viewed norms as the shared expectations for any given
social circumstance. In choosing measures for this form of social capital, it was clear that
rate your community and perception of community trust were two of the few measures
from the SCCBS that came close to approximating social norms. The first, rate your
community asked respondents to rate their community on a scale from poor, only fair,
good, or excellent (0 to 3), and looked at how the respondent viewed their community in
general. This was a good measure for social norms, because norms are generally
considered tied to expectations and one can argue that those who saw their community as
excellent would expect positive social norms to be present in the community. The
specific question asked of respondents was “Overall, how would you rate your
60
community as a place to live — excellent, good, only fair, or poor?” with the response
categories of “poor”, “only fair”, “good”, or “excellent”. The responses were reverse
coded from how they were asked and coded initially so that an increase meant a positive
While the second indicator on the social norms domain, perception of community
trust, focused on whether respondents felt it was likely that people would cooperate
together to save water or electricity, “Now I’d like to ask you a few questions about the
local community where you live. If public officials asked everyone to conserve water or
electricity because of some emergency, how likely is it that people in your community
would cooperate — would you say it is very likely, likely, unlikely, or very unlikely?“
This measure was coded very unlikely (1), unlikely (2), neither/depends [volunteered]
(3), likely (4), and very likely (5). Perception of community trust was related to social
norms because it directly asked whether or not respondents expected others in their
community to act in a certain way. Essentially this measure asked respondents about a
specific social norm in their community: what they expected from their neighbors should
there be a need to conserve resources for the good of the community. Sanctions also
anchor normative behaviors such as water conservation and neighborhood effects that
tend to reflect underlying rules as well as sanctions for breaking them (Demant and
Jarvinen 2011).
Because trust was an important component of both measures, and trust or trusting
is linked to social norms, a third measure of trust, the social trust index, was included on
the SN domain as well. This was an index created by the SCCBS that ranged from -2.49
to 1.01 and included the Question 7 series that asked respondents whether they felt that
61
people in different areas of their life could be trusted or not, such as at work, places one
shops, in general, or in one’s neighborhood. A minimum of three answers from the six
The survey first asked respondents “Now, I want to ask you some questions about
how you view other people. Generally speaking, would you say that most people can be
trusted or that you can’t be too careful in dealing with people?” Possible responses for
this measure included, people can be trust (1), you can’t be too careful (2), and depends
(3). After which respondents were prompted with “[n]ext, we’d like to know how much
you trust different groups of people. First, think about (GROUP). Generally speaking,
would you say that you can trust them a lot, some, only a little, or not at all?” Then,
respondents were asked about trust related to specific groups, including “people in your
neighborhood”, “people you work with”, people at your church or place of worship”,
“people who work in the stores where you shop”, and “the police in your local
community”. The index was constructed using standardized responses to the questions.
The alpha of reliability for this index was fairly strong at 0.81, meaning that only 19
percent of the variance shown in the analysis was measurement error (Allison 1999).
must be understood in context, thus the items for social capital under Bourdieu were
divided into different fields. For Bourdieu (1977, 1980, 1984, 2002), social capital is
found in the relations between positions as well as with the actor. However, in order to
understand the currency of social capital one should assess the interactions/relations
62
between individuals and groups. These do overlap some with Coleman and Putnam, but
While not all aspects of Bourdieu’s theory of inequality have been included, the
current study was able to examine class, race, and gender patterns across models to assess
structural position effects in social capital. Research using Bourdieu has in the past
singled out one or more aspects of his theory, i.e. cultural capital (see Lamont and Lareau
1988; Lamont 2012). It was therefore important to understand how Bourdieu interpreted
theoretical concepts and how his view affected measurement. For Bourdieu concepts
were “‘polymorphic, supple and adaptive, rather than defined calibrated and used rigidly’
(Bourdieu and Wacquant 1992: 23). The measures included for Bourdieu were therefore
chosen using his definition of social capital and the concept of field from a dataset
For the Religious Field domain, volunteer religious, religious membership and
religious groups were used. Volunteer religious was measured by asking respondents if
they had volunteered in the past 12 months for their “place of worship”, yes (1) or no (0).
The specific question asked of respondents was “I’m going to list some of the types of
organizations where people do volunteer work. Just tell me whether you have done any
Also on the religious domain was the measure religious membership. This was
also a dichotomous variable where yes (1) meant one was a member of a church,
synagogue or other religious organization and no (0) meant they were not. The
interviewer would ask participants in the study “[a]re you a member of a local church,
63
The last indicator for the religious domain was religious groups. This was part of
a larger series of questions asking whether respondents had any involvement in the past
12 months with various groups. Religious groups specifically asked “(Besides your local
place of worship,) Any organization affiliated with religion such as the Knights of
Columbus or B’nai B’rith…or a bible study group?” and was coded 0 for no and 1 for
yes.
The Community Field domain was derived from friend contact, hanging with
friends, and coworker contact. These variables were coded with the number of times one
visited with friends, were hanging out with friends socially, or spending time with
coworkers (truncated at 53). Specifically, respondents were asked “(How many times in
the past twelve months have you) had friends over to your home?”, “(How many times in
the past twelve motnhs have you) hung out with friends at a park, shopping mall, or other
public place?”, and “(How many times in the past twelve months have you) socialized
Finally, for the Political Field domain, rally, public interest, petition, and
neighborhood association were used. Rally asked whether one had attended a political
rally in the past 12 months, specifically “[a]ttended a political meeting or rally?“; coded
yes (1) or no (0). Public interest asked respondents if they were a member of any “other
public interest groups, political action groups, political clubs, or party committees”,
where ‘yes’ was coded (1) and ‘no’ was coded (0). Petition asked respondents “have you
signed a petition?”, coded (1) yes, and (0) no. Neighborhood association measured
whether one had been a member of a neighborhood association in the past 12 months, yes
(1) or no (0). The question asked of participants was worded “[a] neighborhood
64
association, like a block association, a homeowner or tenant association, or a crime watch
group.”
Independent Variables
The following section looked at the independent variables including class, race
and gender. Originally class was to be measured using income, education and
occupation. However, there were no real measures for occupation in the SCCBS and
when the variable for income was combined with education in the multi-sample SEM, the
sample size dipped below 200 respondents, 200 being the cut-off for running a model in
SEM (Hatcher 1994). In order to ensure the integrity of the method being used, income
was used as a measure for social class. While class is a multi-dimensional concept
(Weber 1946), this was not the first time the data did not allow for a deeper measure of it.
Other research has used household yearly income as a measure for one’s wealth and as a
proxy for social class has been done in the past (Fritzell, Nermo, and Lundberg 2004;
Wamala, Ahnquist and Mansdotter 2009). Therefore, the question asking for a
respondent’s yearly household income was used and recoded as lower (under
$30,000/year), middle ($30,000 and over, but less than $75,000), and upper ($75,000 but
less than $100,000 and over $100,000 were combined) income groups.
some other race?” The interest of this particular study focused on any major differences
between whites and non-white. Thus, this measure was coded 0 for non-whites and 1 for
whites. This measure did not include those who self-identified as Hispanic.
65
Gender was recorded by the interviewer, and if needed the interviewer was told to
prompt the respondent with “I am recording your gender as male/female.” It was recoded
0 for men and 1 for women. It was understood by the researcher that although the prompt
was labeled gender, there are many layers and other categories beyond male and female.
However, for the purposes of looking at general differences between the larger “norms”
of male and female in American society, they were used for the following analysis.
Analytic Strategy
There were two parts to the analytical strategy, corresponding to the two research
questions posed earlier. In order to answer the first research question, are Putnam’s,
empirically, it was necessary to model each theorist’s approach to social capital using the
SCCBS. Based on the theories, different items were chosen and presented above (please
see chapter 4 for descriptive of each indicator). These were inputted first into an
communality estimates. A principle factor method was used to extract factors; followed
by a promax (oblique) rotation (Hatcher 1994). The main purpose of the EFA was to aid
in the creation of the final model for each theorist. At no point was an EFA used to test
directly any hypotheses or research question. Its sole purpose for the current study was to
aide and find support for how the indicators were arranged by domain in each model.
factors. One can think of these factors linked to the domains for each model. When a
66
here on called domain (Hatcher 1994). Thus, when a variable only loads at .40 or higher
on a factor, and on only one factor, we can argue that it supports that the indicator(s) is
measuring a specific latent domain. The EFA for each theorist helped in assessing how
well the items had been distributed based on the theoretical variables they were meant to
measure. Thus, EFA was the first step toward building a model. In combination with the
EFA, a dataset specifically meant to measure social capital (SCCBS) and social capital
The second step to being able to answer Research Question one, and the actual
test for research one, was to employ confirmatory factor analysis (CFA)21. CFA is well-
suited to this task because it relies heavily on theory and allows for a test of the model fit
on empirical data. Proc Calis in SAS was used to develop appropriate measurement
models. A measurement model allows one to search for empirical support that one’s
items are measuring the underlying, abstract constructs under consideration. Moreover,
the measurement model does not focus on causal relationships and the latent domains are
free to correlate to one another its sole focus is whether the model has been specified well
(Hatcher 1994). Using model fit statistics, a model was determined to have a “good”
model fit or not. A “good fit” meant that the theoretical model approximated the actual
empirical data. If a “good fit” was found, then we were able to provide support for
21
This dissertation utilized Benter’s (1989) standard of noting manifest variables with a“V” for variable (as
shown in Figure 4.2), however, it uses D for domain as opposed to F for factor because of the nature of
the research project. For example, the figure showed that the E/O construct (D1) was measured using
manifest variables V15, V16, and V21, the IC construct (D2) was measured utilizing manifest variables
V12, V27, and V28, and so on. The measurement model posits that latent variables covary. Therefore a
covariance is estimated to connect each latent variable with every other latent variable in the model.
These are easily identifiable in the measurement model figure by a curved, two-headed arrow
connecting the F variables to one another. Stated another way, “a measurement model is equivalent to a
confirmatory factor analysis model in which each latent construct is allowed to covary with every other
latent construct” (Hatcher 1994: p. 415).
67
Research Question one. CFA differs from EFA in many ways, one of which is that with
EFA it is not always known how many latent variables will emerge, if any. However, in
CFA, one needs to have a clear vision and understanding of the theory because the
researcher must input a formed theoretical model using empirical measures. The idea
behind both CFA and SEM in fact, is that they are grounded in theory, and being that this
project is a test of three theorists’ conceptualizations of social capital, these methods were
In order to answer Research Question two and test the hypotheses associated with
it, structural equation modeling (SEM) was employed using multi-sample method to test
whether there were differences in the structure of social capital based on the social
inequalities of class, race, and gender. It was important to look for group differences on
social capital because “if parameter values differ across groups, we run the risk of
making serious errors” (Bollen 1989: 355). The examples that Bollen (1989) used
compared female and male job satisfaction, if one assumed that job satisfaction operated
the same for men and women, then one ran the risk of mis-interpreting females for
example as having had lower job satisfaction than men. Using the SEM multi-sample
method also provided the use of model fit statistics in order to test all parts of the model
and latent variables along specific group lines and see if there were indeed differences. A
chi-square differential test showed whether one model was statistically better than
another and if differences that existed between the different groups were significant
(Hatcher, 1994).
The second step in the analysis, multi-sample SEM, used Proc Calis in SAS. The
multi-sample SEM procedure was optimal for testing these three hypotheses because it
68
allowed use of the three social capital models generated and tested in the first step of
models for women and men completely separate, using multi-sample SEM allowed for a
test of the overall fit of the model(s) across men and women at the same time. Thus we
could look at the form of the model and whether there were structural differences across
the different groups in the same run (Bollen 1989; Kline 2011; Peng 1992).
Whether to accept or reject a particular model is considered less black and white
in SEM than grey, “[i]f we consider the X2 test as a statistical method for evaluation
models, and the fit indexes as more descriptive than statistical, it is seen that both
approaches can work well and be trusted under some conditions” (Hu and Bentler 1995:
98). Therefore, it is important to not only consider one test when either accepting or
In order to determine whether the null hypothesis was rejected for a particular
model, the chi-square and chi-square ratio were used. As with any SEM, a significant
chi-square meant that we could not reject the null hypothesis which states that the data fit
the model. Therefore, we want to have a non-significant chi-square in order to keep the
null hypothesis, because we want to show that the model fits the data. Because the chi-
square is sensitive to sample size, other fit statistics and something known as a chi-square
ratio is often used to ascertain whether to reject the null hypothesis (Hu and Bentler
While the chi-square was used as a sharp test of the hypothesis, it was not the
only test used to determine whether to keep a model (Bollen 1989; Hoyle 1995; Hu and
Bentler 1995; Kline 2011). Other fit statistics used were Bentler’s comparative fit index
69
(CFI), which has shown again and again to be a good index to use, as well as the
Goodness of Fit Index (GFI) (Kline 2011). All fit indexes used in the analysis for
goodness of fit, or how well the data fits the model overall, can be considered to show a
good fit when over 0.90, where 1.0 is best (Hu and Bentler 1995). RMSR, and root mean
square error of approximation (RMSEA). The latter works in the reverse from the other
fit indexes where the closer to zero the RMSEA is, the better the fit. Any finding for
For the purposes of this project, a chi-square ratio of 2:1 or smaller is considered
best, but larger ones will be considered if the fit indexes along with RMSR and RMSEA
show a good fit. Fit indexes of 0.90 or higher are considered supporting a model ‘fit’,
Using multi-sample SEM it was possible to test different parameters using the
social capital model forms across groups. First, datasets were created for each group, e.g.
for gender a male dataset and female dataset were created respectively. The same model
created in the CFA was used in this step of analysis, however, it was run on the newly
created group datasets in the same model. The overall fit was reported (Peng 1992) and
showed whether or not the model fit the groups when separated. This was done for the
model when the parameters were allowed to estimate freely for each group, as well as
when they were constrained to be the same across groups. In order to know which model
was the better model, whether there were differences or not, a chi-distribution was used.
If the findings in the simpler, constrained model were significantly different from the
more complex, unconstrained model, then we could say that there were differences across
70
social capital based on those particular groups. Thus, the more complex model would be
model are the indicators (see Figures in Chapters 4 and 5 for where the arrows point to
the indicators). Therefore, the latent construct or domain(s) are considered the
independent variable(s) in the analysis. Because of the abstract nature of the latent
domains, (it is difficult for example to determine what a unit increase in bonding social
capital domain is) the standardized betas were reported for each individual model (group)
while the unstandardized parameters were used to look between groups (models).
In sum, the first step of analysis looked at the exploratory factor analysis and
models that also fit the data. Secondly, those same models were then used in multi-
sample SEM to test whether there were significant differences between the various
income, race, and gender groups for each theorist’s model of social capital.
71
CHAPTER IV
This chapter reports results of tests for the social capital models for Putnam,
Coleman and Bourdieu based on Research Question one. As such, the chapter also
presents the constructed social capital models for Putnam, Coleman, and Bourdieu. In
order to create each of the three final models, indicators were entered in an exploratory
factor analysis. Combined with the theoretical arguments presented earlier, a model was
created to reflect each theorists’ ideas and a confirmatory factor analysis (CFA) was run
to assess the fit of each and provide a test for Research Question one. The CFA shows
the extent to which each model has been measured well and if the configuration and
model structure are sound (Hatcher 1994). The convergence criterion was satisfied for
every model reported below (Hatcher 1994). While important, a model could potentially
not be identified that satisfied the convergence criteria. Therefore, for each model both
the general requirements and the requirements for CFA identification were met.22 Before
22
Identification refers to the ability of the structural equation model to be able to produce a unique solution,
“whether it is theoretically possible for the computer to derive a unique set of model parameter
estimates” (Kline 2011: 124). The general requirements for meeting identification in SEM are as
follows: 1) the degree of freedom for the model has to be no less than zero; 2) The latent domains must
be given a metric. In the current study each model had df that was over zero and in every model the
domains were set to 1, providing a metric (Kline 2011). There are further requirements for CFA
models, including that each domain has a minimum of three indicators and should the model have only
two latent domains, each domain should have a minimum of two indicators. The models provided in
the current study follow the former where each domain has a minimum of three indicators (Kine 2011).
There are other empirical checks that one can perform to test for identification, but for most fulfilling
the requirements listed as well as having the convergence criteria met indicates that the model has been
identified. Therefore, for the current study each model was identified using these criteria (Kline 2011).
72
presenting the first wave of results, descriptive statistics for the indicators used in each of
the social capital models are presented together below. To avoid redundancy, the
Descriptive Statistics
As shown in Table 4.1, respondents from the SCCBS averaged three friends; this
was closely related to the three people they would feel comfortable sharing confidences.
In general, participants in the national study visited with family [contact] in 2000 around
23 times. On the scale of diversity,23 overall, respondents scored a six where the scale
ranged from 0 to 11, the latter being the most diverse and had a standard deviation of 2.7.
Therefore, we can say that 68.26 percent of the sample scored between 3.3 and 8.7 on the
diversity scale. For the group involvement index24, the average index was 3.10 and the
23 As stated previously in chapter three, The diversity scale was comprised from a series of questions that
asked respondents the following broad prompt followed by specific questions, “[t]hinking now about
everyone that you would count as a PERSONAL FRIEND, not just your closest friends- do you have a
personal friend who…” “owns their own business”, “is a manual worker”, “has been on welfare”, “owns a
vacation home”, “has a different religion than you”, “is white”, “is Latino or Hispanic”, “is Asian”, “is
Black or African American”, “is Gay or Lesbian”, “you would describe as a community leader”. Each of
the 11 questions in the series (see Appendix A) were dichotomous, yes (1) or no (0) and combined into a
diversity scale that ranged from 0 to 11 where 0 was no diversity in the network. The Cronbach’s Alpha of
reliability is a test for measurement error and reliability on a scale or index that ranges from zero to one,
and for this scale the alpha was 0.73.
24
For the social trust index, as detailed in chapter three and Appendix A, each respondent was prompted
with the following and then asked about specific groups and organizations, “Now I’d like to ask about other
kinds of groups and organizations. I’m going to read a list; just answer YES if you have been involved in
the past 12 months with this kind of group.” The indicators included on the group involvement scale were
as follows: “An adult sports club or league, or an outdoor activity club”, Besides your local place of
worship, any organization affiliated with religion, such as the Knights of Columbus or B’nai B’rith, or a
bible study group”, “a youth organization like youth sports leagues, the scouts, 4-H clubs, and Boys & Girls
Clubs”, “a parents’ association, like the PTA or PTO, or other school support or service groups”, “a
veteran’s group”, “a neighborhood association, like a block association, a homeowner or tenant association,
or a crime watch group”, clubs or organizations for senior citizens or older people”, “a charity or social
welfare organization that provides services in such fields as health or service to the needy”, “a labor union”,
a professional, trade, farm, or business association”, “service clubs or fraternal organizations such as the
Lions or Kiwanis or a local women’s club or a college fraternity or sorority”, “ethnic, nationality, or civil
rights organizations, such as the National Organization for Women, the Mexican American Legal Defense
or the NAACP”, “other public interest groups, political action groups, political clubs, or party committees”,
73
standard deviation was 2.78. The index ranged from 0 to 18. Thus we can say that 68.26
percent of the sample scored between 0.32 and 5.88 for group involvement. Under half,
social trust index with a standard deviation of 0.73. Therefore we can say that 68.26
percent of the sample fell between -0.78 and 0.68 on the social trust index. For the most
part then, this sample had an average level of trust, not extremely high (closer to 1.01),
but not weak or distrustful either (closer to -2.49). Overall the sample converged around
very likely to perceive their community as trustworthy, with a mean of 4.25 and a
standard deviation of 0.89 where 5 was most likely and 1 was least likely on perception
of community trust. Overall, respondents rated their community as good with a mean of
2.21 and a standard deviation of 0.76 where 3 was excellent and 0 was poor.
Respondents from the SCCBS in 2000, had on average three (3.44) people they
felt were like family or fictive kin. The standard deviation for this indicator was 3.76 and
ranged from zero to 10. Thus, for this sample, 68.26 percent of the sample had between
zero and seven (7.2) fictive kin. Forty-six percent of the sample reported being active in
other church participation and overall respondents attended one club meeting in the past
12 months. The mean for club meeting was 1.01 with a 1.55 standard deviation in an
indicator that ranged from zero to six. Sixty-three percent of the sample reported knowing
“a literary, art, discussion or study group or a musical, dancing, or singing group”, “any other hobby,
investment, or garden clubs or societies”, “a support group or self-help program for people with specific
illnesses, disabilities, problems, or addictions, or for their families”, “are you involved in any group that
meets only over the Internet”, “and do you belong to any other kinds of clubs or organizations?” Each of
the responses for the former indicators were coded (0) for no, and (1) for yes. The scale ranged from zero
to 18 where the higher the score, the more involved the respondent was in groups. The alpha of reliability
for this scale was 0.72
74
someone who owned their own business (business owner) while, 44 percent of the sample
Therefore, the results indicate that over half of all respondents have some religious
affiliation in the form of religious membership and a large percentage (29) who also
contact with friends, coworkers or spent time with friends, the average for friend contact
was 20 times in the past year. The mean for this indicator was 20.47, the standard
deviation was 19.24, and the range was from zero to 53 times. Therefore, 68.26 percent
of the sample contacted friends between 1.23 and 39.71 times. For coworker contact the
mean was 13.01 with a standard deviation of 16.65 and ranged from zero to 53. Thus,
68.26 percent of respondents in 2000 contacted their coworkers between zero and 26.66
times. On average, respondents contacted their friends more than their coworkers. Those
from the sample reported hanging with friends on average 15.15 times in the past 12
months. The standard deviation was 18.18 and again it ranged from zero to 53 times.
Consequently, we can say that 68.26 percent of the sample hung out with their friends
between zero and 33.33 times in 2000. For this sample, only 17 percent reported
participating in a rally, and only nine percent responded being involved in public interest
75
groups. In contrast, 35 percent of the sample reported signing a petition, and 22 percent
Finally, for the independent indicators, the average for income was 2.08, with a
standard deviation of 0.90 and a range of 1 to 3. What this means is that most
respondents fell in the lower and middle income category or less than $75,000/year
(55.41 percent). When breaking down this number, however, the majority (36.4 percent)
of the 55 percent was in the lower income category of less than $30,000/year. For this
sample 44.59 percent reported yearly income in the upper income category. The sample
76
Table 4.1. Descriptive Statistics for Social Capital Measures
Mean N Std Dev Minimum Maximum
Friends 3.28275402992 1.0932439 1 5
Share confidences 3.46809222993 0.8419965 1 4
Family contact 23.86059162941 20.3483343 0 53
Diversity 6.08358313003 2.6555550 0 11
Group involvement index 3.09923413003 2.7817831 0 18
Worked with others 0.30942901506 0.4624118 0 1
Social trust index -0.05300032993 0.7246622 -2.49 1.01
Perception of community trust4.25166891498 0.8871234 1 5
Rate your community 2.20847232998 0.7616747 0 3
Fictive kin 3.44032922916 3.7634432 0 10
Other church participation 0.46127692647 0.4985925 0 1
Club meetings 1.01137502989 1.5466435 0 6
Business Owner 0.62969172984 0.4829682 0 1
Vacation home 0.44342822952 0.4968735 0 1
Volunteer religious 0.29433462983 0.4558195 0 1
Religious membership 0.66439652643 0.4722904 0 1
Religious groups 0.17160953001 0.3771035 0 1
Friend contact 20.47020772937 19.2399842 0 53
Coworker contact 13.01238852018 16.6462612 0 53
Hanging with friends 15.15032022967 18.1819210 0 53
Rally 0.17400003000 0.3791727 0 1
Public interest groups 0.09200003000 0.2890741 0 1
Petition 0.34984932987 0.4770020 0 1
Neighborhood association 0.22385083002 0.4168927 0 1
Income 2.08187132736 0.8963996 1 3
Race 0.74786502459 0.4343267 0 1
Gender 0.60039963003 0.4898978 0 1
supporting the assertion of three important factors for Putnam, not two. Then a
confirmatory factor analysis (CFA) was performed to empirically test Research Question
one for Putnam. Stated here again, Research Question one asked: are Putnam’s,
77
For each of the three theorists it was possible that the model would not fit the data and
thus be rejected. Following are the results from the EFA and CFA analyses for Putnam.
and family contact; the domain of bridging as represented by the indicators diversity,
group involvement index, and worked with others; and the domain of trust as represented
by the social trust index, perceptions of community trust, and rate your community were
In order to extract meaningful factors, the principal factor method was utilized; which
was then followed by a promax rotation (Hatcher 1994). The means with which the
domains were accepted was based on the interpretability criteria presented by Hatcher
(1994). The scree test was used, based on the eigenvalues generated by principal
factors. Where there is a drop, it is customary to use that as the cut-off. The scree test
and eigenvalues showed that the first factor should be retained, but was less clear on
those following. Thus, the interpretability criteria were used. These included having
three indicators that load significantly on each factor, shared conceptual meaning
between indicators on a factor, and support that each factor was measuring a distinct
construct.
78
‚
1.75 ˆ
‚
‚
‚ 1
‚
1.50 ˆ
‚
‚
‚
‚
1.25 ˆ
‚
‚
‚
‚
E 1.00 ˆ
i ‚
g ‚
e ‚
n ‚
v 0.75 ˆ
a ‚
l ‚
u ‚ 2
e ‚
s 0.50 ˆ
‚
‚
‚
‚ 3
0.25 ˆ
‚
‚
‚
‚
0.00 ˆ 4
‚
‚ 5 6
‚
‚ 7 8
-0.25 ˆ 9
‚
----------------------------------------------------------------------------------
0 1 2 3 4 5 6 7 8 9
Number
Figure 4.1. Scree Plot of Eigenvalues for Putnam’s Social Capital Model
Table 4.2 showed that instead of two factors, three distinct factors emerged,
bonding, bridging, and trust. Putnam’s theory for social capital focused on two major
forms of social capital, bonding and bridging (Putnam 2000). For Putnam, trust was
important for understanding the basis of both types of social capital. Bonding capital
involves “thick” trust and bridging capital involves what Putnam referred to as “thin”
79
trust. However, as shown in the following analysis, trust, while correlated with bonding
To interpret the rotated factor pattern, one had only to look for indicators that
loaded at .40 or higher (denoted with an *) and be sure that it only loaded on one factor at
that level, or presented with a “simple” structure. Using this criterion, three indicators
clearly loaded for bridging social capital, and trust respectively. However, on bonding
social capital, friends loaded at 0.5825, shared confidences loaded well at 0.57, but family
contact loaded under 0.40 at 0.19. Family Contact clearly does not load as the theory
suggests, but it does load the highest on the bonding factor and theoretically it fits very
well with Putnam’s conceptualization of bonding social capital, where one’s family and
frequency of interaction should lead to bonding social capital (See Table 4.1 and
involvement loaded strongly at 0.55, and worked with others loaded at 0.43. The
strongest loadings were on the trust factor, the social trust index and perception of
community trust both loaded at 0.52, and rate your community loaded at 0.58. Thus, the
scree plot, theory, and EFA indicator loadings, showed support for retaining three factors
for further analysis for trust, bonding, and bridging social capital.
25 The standardized loadings were reported for all three social capital models.
26 The EFA use in the current study was to empirically look for support for measures that were cohevsive
or hung together. Those that loaded on a factor with a “simple” structure provided support, but were not
the only criteria for indicators inclusion. Any and all indicators that were removed were not only due to
a loss of loading on a factor, but because conceptually it made sense.
80
Table 4.2 Exploratory Factor Analysis (EFA) - Putnam’s Social
Capital Model
As suggested by the results from the EFA, three indicators could be found for
each factor for the confirmatory factor analysis (we used the Proc Calis procedure in SAS
(SAS Institute Inc. 1989)). The following CFA provides a test for the first research
question, whether or not Putnam’s theoretical model for social capital could be
empirically measured. The models tested were covariance structure models with multiple
empirical indicators for the latent indicators.27 The standardized parameter estimates for
each pathway as well as the corresponding t-score were reported in Figure 4.3. The t-
scores were all above 2.0 and ranged between 6 and 20, indicating that the loadings were
significant (p< .001). The t-scores provided support for the convergent validity of the
27 A minimum of three indicators were used for each domain throughout the analysis, the method of using
the 3 indicator rule promotes robustness (Bollen 1989).
81
indicators (Anderson and Gerbing 1988). Essentially, that means that each domain was
the relationship between the latent domains and the manifest indicators measuring them.
In the current study, the model was comprised of three latent domains corresponding to
the three forms of social capital: bonding, bridging, and trust. Each of the latent domains
was measured using a minimum of three manifest indicators, displayed in Figure 4.2.
82
Friends (V2)
E2
Bonding
Share confidences
(D2)
(V3) E3
Bridging
(D3) Diversity (V9)
E9
Perception of
communitytrust(V22 E22
The fit indices. The t-score for all parameters were significant at the p<.001 level
(please see Figure 4.3 below) and thus a review of the model fit statistics was provided to
28 This dissertation utilized Benter’s (1989) standard of noting manifest indicators with a“V” for indicator
(as shown in Figure 4.2), however, it uses D for domain as opposed to F for factor because the nature of
this research project focuses on domains . For example, the figure showed that Trust (D1) was
measured using manifest indicators V9, V120, and V25, the Bonding construct (D2) was measured
utilizing manifest indicators V2, V3, and V27, and so on. The measurement model posits that latent
domains covary. Therefore a covariance is estimated to connect each latent domain with every other
latent domain in the model. These are easily identifiable in the measurement model figure by a curved,
two-headed arrow connecting the D variables to one another. Stated another way, “a measurement
model is equivalent to a confirmatory factor analysis model in which each latent construct is allowed to
covary with every other latent construct” (Hatcher 1994: p. 415).
83
test Research Question one. Stated again, the first research question asks whether or not
CFA provides a unique test of model fit on empirical data to answer this question.
Before reporting the findings from the confirmatory factor analysis and analyzing other
forms of fit, however, support needed to be obtained from the fit statistics for either
accepting the model as fitting or rejecting the model as not fitting the data. The first test
As shown in Table 4.3, the CFA for Putnam’s social capital model had a Hoelter’s
(1983) Critical N of 523. Hoelter’s Critical N is a fit statistic that focuses on the size of
the sample. A good benchmark is anything more than 200, with a Hoelter’s (1983)
Critical N of 523, the sample size was clearly large enough to give a chi-square statistic
(Byrne 2010).
The chi-square was significant at the p<0.0001 level. As stated before, for CFA
one actually wants the chi-square to not be significant because the null hypothesis is that
the model fits the data. The chi-square ratio of 4:1 was clearly above 2, but lower than
the CFA chi-square ratio of 6:129. Although a p-value larger than 0.05 or a chi-square
ratio of less than 2:1 would indicate a good fit, with a large sample size, the chi-square
test is sensitive to even trivial deviation from a perfect model. Therefore, alternative fit
29 A first run for Putnam produced two factors, but was based on the theoretical assumption by Knudsen et
al that memberships went along the lines of bonding and bridging using the idea of exclusive and
inclusive groups. The Cronbach’s Alpha for these scales were below .40, showing that they were not
reliable measures, and they both loaded on the same factor, showing clearly that they were measuring
the same abstract concept and not distinct. In the CFA run with the model based off of that first EFA,
there was a high chi-square ratio of 6 and lower indexes of fit, supporting the need to recalibrate the
membership indicators into one and build a different model.
84
In many cases the sample size is so large that it is not mathematically possible to
have a non-significant chi-square (Hatcher 1994), which also in turn affects the chi-
square ratio. According to Hatcher (1994) “it has been shown that the chi-square/df ratio
is affected by sample size, and that the same model may display significantly different
ratios with small samples than with large samples…we advise that this criterion be used
only as a very rough rule of thumb, if at all, and be supplemented with other criteria that
are not affected by sample size” (290). Moreover, Bentler and Bonett (1980) state that
“[i]n large samples virtually any model tends to be rejected as inadequate, and in small
At this point, it would be easy to reject the Putnam social capital model and
simply deem it unfit. If one were to only use the chi-square statistic that would be the
verdict. However, based on my reading of the statistical literature (Bollen 1989; Bentler
and Bonett 1980; Hatcher 1994) and because of the sensitive nature of the chi-square in a
A major benefit of the Bentler and Bollen indexes was that they are not as closely
tied to sample size (Bentler and Bonett 1980). All of these should be as close to 1 as
possible, indicating that the structure of the theoretical model fits the actual empirical
model, and in fact, as seen in the table below, they were all above 0.90. The comparative
fit index (CFI) was close to 1 at 0.9489, while the goodness-of-fit index (GFI) was even
closer at 0.9843. Moreover, Bentler’s and Bonett’s non-normed index was at 0.9234 and
Finally, one other important fit statistic was evaluated, the root mean square error
85
value of zero indicates the best fit (Kline 2011: 205). It is not tied to sample size per se,
and is not based on a central chi-square approximation, but rather on a non-central chi-
square distribution. While useful in showing how well the theoretical model
approximates the actual model, it is not a test for “perfect” fit. Generally a RMSEA of
0.08 or lower is considered “good” (Kline 2011). The RMSEA for Putnam’s model was
0.0471, well below the 0.08 cut-off, and thus also showed support for a good model fit.
Along with the fits statistics, the standardized residuals were consulted to see if
the projected model varied from the actual model greatly.31 Friends did have a large
standardized residual with diversity at 4.58 and shared confidences with the social trust
index at 4.08 (please see Table 4.3 below). It may be that bonding, bridging, and trust
work together and do not have distinct boundaries. For example, for this finding it makes
sense that respondents who had higher levels of trust would be more likely to share
confidences with others. It could be just as likely that one’s bonding social capital and/or
86
bridging social capital influence one another, perhaps having more bonding social capital
leads to less bridging or vice versa. Another possibility is that social capital begets more
social capital and thus, the more on one domain, the more likely you are to build on a
different dimension(s). These questions, however important, were not directly tested
Family contact was also associated with some very high standardized residuals,
4.58, -3.91, and -3.10, -2.96, and -2.8232. The largest standardized residual was between
worked with others and group involvement at 6.29. The group involvement index, along
with worked with others had some high residuals with friends and diversity (-3.91; -3.26).
The second largest residual was between rate your community and perception of
community trust at 5.02. According to Hatcher (1994), high residuals do not necessarily
mean that one should reject the model. Instead, he encourages researchers to use all data
available including other fit statistics to determine whether to accept or reject the model.
Furthermore, although there were some large residuals, the pattern of residuals in the
matrix also showed support for the model fit with the lower residuals being under 2.0.
While some of the standardized residuals for indicators in the Putnam model were
considered high, these indicators were not deleted because the measures were the best for
modeling the theory available in the dataset and other fit statistics (reported above)
showed support for them. The residual matrix was only one of many results used to
assess model fit and to choose the best theoretical model to data fit. Thus, based on the
32 With every confirmatory factor analysis or SEM that uses covariance matrixes a residual matrix and
normalized residual matrix are created. These numbers come from the predicted model matrix
produced while the analysis is done on the covariance matrix. If the theoretical model fits the actual
data, then the residuals will be near zero, anything over 2.0 is considered high for standardized
residuals. If they are low or close to zero, then the causal relationships between the indicators are being
accounted for correctly (Hatcher 1994).
87
EFA, and other CFA fit statistics listed above that showed a good fit for the Putnam
social capital model, the current model (see Figure 4.1) stood as the final specified
model. All future analysis for Putnam’s social capital was based off of this model.
Finally in answering Research Question one, finding a good model fit for the Putnam
social capital model provided support that Putnam’s theory of social capital can be
modeled empirically.
Putnam’s results. The covariances for the latent domains are reported here as
unstandardized estimates where the bonding domain covaried with the trust domain with
an estimate of 0.50, significant with a t-score of 13.06. Whereas the trust domain
covaried with the bridging domain with an estimate of 0.30 which was significant with a
t-score of 7.52; finally the bonding domain covaried with the bridging domain with an
Due to the fact that the SEM model for Putnam had latent domains for which the
unit of analysis was abstract, the following analysis focused on the standardized Beta
results. As shown in Figure 4.3, all indicators significantly and positively loaded on the
bonding social capital domain. For respondents in this sample, a one standard deviation
increase in bonding social capital was associated with a 0.66 standard deviations increase
in friends, a 0.63 standard deviations increase in shared confidences, and a 0.20 standard
friendships, the number of people they could share confidences with, and the number of
times they had contact with family in 2000. Of all indicators loading on the bonding
88
domain, family contact had the weakest loading. Furthermore, family contact had a weak
R-square of 0.04 where only four percent of the variance in family contact was explained
by bonding social capital. However, friends and shared confidences had R-squares of
0.43 and 0.40. Thus, 44 percent of the variance in friends and 40 percent of the variance
in shared confidences was explained by bonding social capital for respondents in the
sample.
Diversity, group involvement index, and worked with others also loaded
significantly on bridging social capital (t-scores going no lower than 10.04). Thus, a one
standard deviation increase in bridging social capital was correlated with a 0.72 standard
0.57 standard deviations increase in the group involvement index, and a 0.33 standard
deviations increase in worked with others. Clearly diversity had the highest loading and
was therefore the strongest indicator for bridging social capital for those in the sample.
This was followed by group involvement index, whereas, working with others had the
weakest return on an increase in bridging social capital. Not surprisingly, the R-squares
for each indicator followed the same pattern. Fifty-one percent of the variance in
diversity was explained by bridging social capital. Thirty-two percent of the variance in
the group involvement index for those in the sample was explained by bridging social
capital. Finally, only 12 percent of the variance in worked with others was explained by
Lastly, all indicators loaded positively and significantly on the dimension of trust
with t-scores ranging from 14.77 to 20.78. For a respondent the sample, a one standard
deviation increase in trust was associated with a 0.60 standard deviations increase in rate
89
your community, a 0.46 standard deviations increase in perception of community trust,
and a 0.71 standard deviations increase in the social trust index. The largest domain
loading was with the social trust index and the lowest with the perception of community
trust. The social trust index had the most variance explained by the trust dimension with
an R-square of 0.50 or 50 percent. While accounting for less variance, the perception of
community trust had 22 percent of variance explained by trust and 36 percent of the
variance in rate your community was explained by the trust dimension for
respondents.
Bridging
(D3) (1.87, [18.38], Diversity (V9)
0.72) E9
(1.58, [16.11],
0.57) Group involvement E20
(0.15 [10.04] index (V20)
0.33)
Worked with others E25
(V25)
Trust (D1) (0.45, [18.46],
0.60) Rate your
community(V31) E31
(0.41, [14.77],
0.46)
Perception of
communitytrust(V22 E22
(0.51, [20.78],
0.71)
Social Trust Index(7) E7
Figure 4.3. CFA Measurement Model for Putnam’s Social Capital (unstandardized beta,
[t-score], standardized Beta)33
90
Coleman’s Social Capital Model
For the purposes of this project, the three forms of social capital theorized by
Coleman were used as latent domains and measures from the SCCBS were used using the
written works of Coleman, EFA, and face validity to choose indicators for each form of
social capital.
Responses to nine indicators for the domains E/O, I.C., and S.N., including group
involvement index, other church participation, club meetings, rate your community,
perception of community trust, social trust index, fictive kin, business owner, and
vacation home34, were run through an exploratory factor analysis (EFA) using squared
multiple correlations as prior communality estimates. The principal factor method was
4.4 below, for the E/O domain all three indicators loaded positively and significantly
above 0.40. Group involvement index loaded at 0.58, other church participation at 0.45,
and club meetings at 0.50. On the I.C domain business owner (0.58) and vacation home
(0.50) loaded positively and significantly. However, fictive kin loaded just below the
0.40 cut-off at 0.39. Because fictive kin could potentially provide important information
networks (Stack 1974), it was retained for further analysis. Finally, rate your community
(0.56), perception of community trust (0.51), and the social trust index (0.54) loaded
V22 (0.22), V31 (0.36). RMSEA = root mean square error of approximation; CFI = comparative fit
index; GFI = goodness-of-fit index.
34 All indicators came from the Social Capital Community Benchmark Survey from 2000.
91
Table 4.4. EFA for Coleman’s Three Forms of Social Capital- Rotated Factor
Pattern, Standardized Results
Not
I.C. S.N. E/O Sig.
Factor Factor Factor Factor
1 2 3 4
Expectation/Obligation
v15-group involvement index 17 -5 58 * -2
v16-other church participation 0 10 45 * -4
v21-club meetings -6 -2 50 * 5
Social Norms
v8-rate your community 7 56 * -1 -1
v9 -perception of community
trust -6 51 * -1 -3
v10-Social trust index 1 54 * 6 5
Information Channels
v12-fictivkin 39 ^ 1 9 -26
v27-business owner 58 * -2 0 -3
v28-vacation home 50 * 2 -1 3
*Significant loading at .40 or greater. Values marked with a ‘^’ are kept in model for
further analysis because theoretically sound. Printed values are standardized and multiplied
by 100 and rounded to the nearest integer.
A scree test supported three clear meaningful factors and a possible fourth based
on the loadings (see Figure 4.4). However, the loadings on the fourth factor were not
significant and the scree plot showed no support for retaining the fourth factor.
Moreover, the theory supported three latent constructs, therefore, only the first three
factors were retained.35 Using these criteria, three indicators loaded for E/O, SN, and IC
35 As stated earlier, an indicator was said to load on any given factor if it was .40 or higher (see Table 4.4).
36 A “simple” structure refers to the indicators only loading significantly on one factor, not across factors.
37 Prior to running the final EFA for Coleman’s three-pronged model, an EFA to assess the measures and
show support for their allotment was used as an extraction method. Three EFA’s were run before the
final one shown below to confirm the measures did load on one factor for each. Because each EFA
produced one-factor solutions, it was not necessary to perform a promax (oblique) rotation.
92
‚
1.75 ˆ
‚
‚
‚
‚ 1
1.50 ˆ
‚
‚
‚
‚
1.25 ˆ
‚
‚
‚
‚
E 1.00 ˆ
i ‚
g ‚
e ‚
n ‚
v 0.75 ˆ
a ‚
l ‚ 2
u ‚
e ‚
s 0.50 ˆ
‚
‚
‚
‚
0.25 ˆ 3
‚
‚
‚
‚
0.00 ˆ 4 5
‚
‚
‚ 6 7
‚ 8
-0.25 ˆ 9
‚
Š----------------------------------------------------------------------------------------------------------------------------------------------
0 1 2 3 4 5 6 7 8
In order to test whether or not Coleman’s theoretical model for social capital
could be empirically measured, the results from the EFA and his theory were used to help
build the model shown in figure 4.2 below. This measurement model was then entered
into a confirmatory factor analysis (CFA). CFA using the Proc CALIS procedure in SAS
(SAS Institute Inc. 1989) was used. The model tested used covariance structure with
93
multiple empirical indicators for the latent domains. The parameter estimates for each
pathway as well as the corresponding t-scores and standardized Betas are reported in
Figure 4.6. The t-scores were all above 2.0 and ranged between 5.42 and 20.85,
indicating that the factor loadings were significant (p< .001) and also provided support
for the convergent validity of the indicators, meaning that the domains were accurately
the relationship between the latent domains and the manifest indicators included to
measure the latent factors. In the study, the model was comprised of three latent domains
information channels (IC), and social norms (SN). Each of the latent domains was
94
Group involvement
index (V15) E15
Expectatio
n/Obligati Other church
participation (V16) E16
on (E/O)
D1 Club meetings (V21) E21
Social
Norms Rate your
(SN) D3 community(V8) E8
Perception of
community trust(V9) E9
The fit indices. The t-scores for all parameters were significant at the p<.001 level
and thus a review of the model fit statistics was provided to answer Research Question
one. The measurement model was estimated using the maximum likelihood method.
The chi-square value for the model was statistically significant p<.0001 level. As stated
previously, the chi-square statistic may be used to test the null hypothesis that the model
fits the data. However, in practice, the statistic is sensitive to sample size as well as
departures from multivariate normality. While the Hoetler’s Critical N was at 612 for the
95
model (see Table 4.5), which was above 200 and supported that the chi-square statistic
was acceptable to use, it does not change the fact that because of the chi-square’s reaction
to a large sample size, using it can often result in the rejection of a model that fits.
Therefore, the model chi-square statistic (also called a chi-square ratio) was used as a
goodness of fit index, with smaller chi-square values (relative to the degrees of freedom)
as a better test of model fit (James, Mulaik, and Brett 1982; Joreskof and Sorbom 1989).
The chi-square ratio for this measurement model was 2.95, where 2 or lower is best.
Because the chi-square ratio was above 2, and due to the sensitive nature of the chi-
square, other goodness of fit indices were also consulted in assessing the fit for this
model.
Coleman’s CFA measurement model had goodness of fit values greater than .90
for the non-normed-fit index (NNFI) and the comparative fit index (CFI) as well as the
Bollen Normed Index Rhol, Bentler and Bonett’s (1980) NFI, and the Bollen (1988)
Non-normed Index Delta2, meaning that there was an acceptable fit, (see Table 4.4)
(Bentler and Bonett 1980; Bentler 1989). Moreover, there were very few high residuals
in the Coleman model. The covariance between other church participation and social
trust index had a standardized residual of 4.60, and also the covariance between the social
trust index and rate your community was high at -4.06. However, there were more
standardized residuals reporting at lower than 2.0, therefore supporting that the predicted
model shadowed closely the actual model. The average standardized residual for
A few of the standardized residuals for indicators in the Coleman model were
high, however, these indicators were not deleted because the measures were the best for
96
modeling the theory available in the dataset and many other fit statistics (reported above)
showed support for them. The residual matrix was only one of many results used to
assess model fit and to choose the best theoretical model to fit the data. Thus, based on
the EFA, and the other CFA fit statistics listed above that showed a good fit for
Coleman’s social capital model, the current model (see Figure 4.3) was accepted as the
final specified model. All later analyses for Coleman were based on this model. Finally
in answering Research Question one, finding a good model fit for Coleman’s model
Table 4.5. CFA-Covariance Structure Analysis: Maximum Likelihood Estimation for the
CALIS Procedure Confirmatory Factor Analysis - Fit Statistics – Coleman’s Social
Capital Model
Fit Statistic
Goodness of Fit Index (GFI) 0.9899
Chi-Square 58.5610
Chi-Square DF 24
Pr > Chi-Square38 <.0001
RMSEA Estimate 0.0339
Bentler’s Comparative Fit Index 0.9732
Benter’s & Bonett’s (1980) Non-normed Index 0.9597
Bentler & Bonett’s (1980) NFI 0.9558
Bollen (1988) Normed Index Rho1 0.9336
Bollen (1988) Non-normed Index Delta2 0.9734
Hoelter’s (1983), Critical N 783
N 1257
Coleman’s results. The covariances for the latent domains are reported here as
the information channel (IC) domain with an estimate of 0.54, significant with a t-score
of 13.56. Whereas the E/O domain covaried with the social norms (SN) domain with an
97
estimate of 0.22 which was significant with a t-score of 5.50; finally the IC domain
covaried with the SN domain with an estimate of 0.32 and a t-score of 7.12.
The following figures show Coleman’s social capital model across racial groups.
As shown below in Figure 4.6, all indicators significantly loaded on the E/O dimension.
Thus we can say for this sample, a one standard deviation increase in E/O social capital
was associated with a 0.86 standard deviations increase in the group involvement index, a
0.42 standard deviations increase in other church participation, and a 0.47 standard
deviations increase in club meetings. Of all indicators loading on E/O social capital,
other church participation and club meetings both loaded fairly strongly, although the
former was the weakest loading for E/O social capital. The group involvement index,
As with Putnam, the R-square statistic was used to assess how well the model and
its parts predicted for the sample. Only 18 percent of the variation in other church
participation was explained by E/O social capital. However, 22 percent of the variation
in club meetings and 74 percent of the variation in group involvement index was
explained by E/O social capital, which were fairly strong and very strong results for R-
square. For respondents in the sample, it was clear that E/O social capital was a strong
All indicators loaded at a statistically significant level for the information channel
(IC) dimension. Here a one standard deviation increase in IC was associated with a 0.19
standard deviations increase in fictive kin, a 0.63 standard deviations increase in business
owner, and a 0.61 standard deviations increase in vacation home. The strongest loadings
were business owner and vacation home. Interestingly fictive kin had a very weak
98
loading and using R-square the IC accounted for only four percent of the variation in
fictive kin. In contrast, 40 percent of the variation in business owner and 37 percent of
the variation in vacation home was explained by the IC for the sample. These findings
Vacation home had 37 percent and business owner had 40 percent of its variance
explained by IC social capital, however, only four percent of the variance in fictive kin
was explained by IC social capital. The latter finding suggests that for this sample, IC
social capital had less to do with the people one sees as “family” and more to do with ties
to people with specific resources or knowledge (e.g. owning one’s own business).
All indicators for the SN social capital dimension loaded significantly. For the
sample, a one standard deviation increase in SN social capital was correlated with a 0.66
increase in rate your community, a 0.46 increase in the perception of community trust,
and a 0.64 increase in the social trust index. Therefore, for the sample, increasing SN
social capital meant that how they rate their community, how they perceive others’
willingness to conserve for the greater good, and their overall trust in others increased.
This was strongest for rate your community, where 43 percent of the variance was
explained by SN social capital. The social trust index followed closely with 41 percent
99
Group involvement
E15
(2.42, [20.85],0.86 ) index (V15)
Social
(0.49,[17.16], 0.66) Rate your
Norms
(SN) D3 community(V8) E8
(0.40,[13.22], 0.46 )
Perception of
community trust(V9) E9
(0.46,[16.91], 0.64)
Social Trust E10
Index(V10)
Figure 4.6. CFA Measurement Model for Coleman’s Social Capital (unstandardized beta,
[t-score], standardized Beta)39
For Bourdieu, social capital can be thought of as social resources that exist in the
39 Note: Model Fit Statistics: Chi-square = 58.5610, 24 df (p<.0001), RMSEA = 0.0339, CFI = 0.9732,
GFI = 0.9899. Adjusted R2 : V8 (0.43), V9 (0.21), V10 (0.41), V12 (0.04), V15 (0.74), V16 (0.18),
V21 (0.22), V27 (0.40), V28 (0.37). RMSEA = root mean square error of approximation; CFI =
comparative fit index; GFI = goodness-of-fit index.
100
[s]ocial capital is the sum of resources, actual or virtual, that accrue to an
individual or a group by virtue of possessing a durable network of more or less
institutionalized relationships of mutual acquaintance and recognition (Bourdieu
and Wacquant, 1992: 119).
theory, the following analysis uses the concept of field to add context to Bourdieu’s
social capital model. It was difficult to come to a final measurement model for
sophisticated, rich, and also hard to measure (Silva and Edwards 2007). EFA was used
initially on the measures to explore the loadings and see which measures hung together,
or on which Field they loaded. EFA was used to support domains or fields, of religious,
with three indicators, community with three indicators, and political with four indicators.
Altogether they comprised 10 measures and I explored whether they loaded on the field
Due to the fact that most studies found in a review of the literature concentrated
approach Bourdieu’s social capital model from an exploratory standpoint (Bourdieu and
Wacquant 1992, 1999; Lareau 2000; Lamont and Lareau 1988; Lamont 2012). This was
where having a dataset created specifically to measure social capital was very useful.
The EFA loadings (see Table 4.6) and scree plot (see Figure 4.7) showed that there were
40 One limitation for the Bourdieu model in this project was that a pre-existing dataset was used (SCCBS).
101
three significant factors with a minimum of three indicators loading discreetly on three
As opposed to Putnam and Coleman, where the scree plot and factor loadings
were needed to “see” the underlying latent structure,” for Bourdieu there was a clear
break in the scree plot showing three domains. Supporting the scree finding, the loadings
shown in Table 4.7 present a “simple” structure and aside from one that loads slightly
under 0.40 (neighborhood association) all load positively, significantly, and distinctly on
three domains.
For the religious field, volunteer religious loaded at 0.66, religious membership
loaded at 0.63 and religious groups loaded at 0.46. The first two indicators were
especially high loadings for this domain. All three indicators were measuring different
aspects of religious interaction, membership, participation, and whether one was located
spatially in the religious field. It was harder to “see” the resources within each measure,
thus it was assumed that volunteering, being a member of a religion, and being a part of
religious groups also meant that one had, accrued and could access various social
102
1.50 ˆ 1
‚
‚
‚
‚
‚
1.25 ˆ
‚
‚
‚
‚
‚
1.00 ˆ 2
‚
‚
‚
E ‚
i ‚
g 0.75 ˆ
e ‚
n ‚
v ‚
a ‚
l ‚
u 0.50 ˆ 3
e ‚
s ‚
‚
‚
‚
0.25 ˆ
‚
‚
‚
‚
‚
0.00 ˆ 4
‚ 5
‚ 6
‚ 7
‚ 8
‚ 9
-0.25 ˆ 0
Š--------------------------------------------------------------------------------
0 1 2 3 4 5 6 7 8 9 10
Number
Friend contact (0.65), coworker contact (0.55), and hanging with friends (0.62)
all loaded significantly and positively on the community domain. These indicators
measured different aspects of one’s personal community, the people and connections that
one would spend time with and share resources. The first and last focusing on spending
time and being connected to one’s friends and the second indicator including time spent
103
Finally, rally (0.52), public interest groups (0.44), petition (0.44), and
neighborhood association (0.37) loaded on the political domain. The only one to load
below 0.40 was neighborhood association. Again, it was assumed that by being
connected in these different ways and through these measures that there was with each an
opportunity to create, maintain and use the social resources that develop within the tie
and are held by the actors. This assumption is in line with Bourdieu’s definition of social
41 *Printed values were multiplied by 100 and rounded to the nearest integer.
Values greater than 0.4 were flagged by an ‘*’; values that were used in further
analysis but which did not load were flagged by a’^’.
104
Confirmatory Factor Analysis: Bourdieu
As with both Putnam and Coleman, confirmatory factor analysis (CFA) was used
to answer the first research question of whether or not Bourdieu’s social capital model
can be empirically modeled using SEM. The factors presented below were based on the
below, was based on his definition of social capital combined with his notion of field and
the results from the EFA above. This measurement model was then entered into a
confirmatory factor analysis. CFA using the Proc CALIS procedure in SAS (SAS
Institute Inc. 1989) was used. The model used covariance structure with multiple
empirical indicators for the latent domains. The parameter estimates for each pathway as
well as the corresponding t-scores and standardized Betas are reported in Figure 4.6. The
t-scores were all above 2.0 and ranged between 10.64 and 25.29, indicating that the
loadings were significant (p< .001) Bourdieu’s concept of social capital related to the
idea that actors’ social capital was embodied in them and their social position. He saw
organizations that were tied to specific social structures was used to operationalize the
105
Volunteer Religious
(V2) E2
Religious Religious
Field (D1) Membership (V19) E19
Community
Field (D2) Friend Contact E42
(V42)
CoWorker Contact
(V43) E43
Rally (V16)
E16
Public Interest
Groups(V27) E27
Petition (V15)
E15
Neighborhood Assoc.
(V5) E5
The fit indices. The t-scores for all parameters were significant at the p<.001 level
and thus a review of the model fit statistics was provided to answer Research Question
one (see Bourdieu’s results below). The measurement model was estimated using the
maximum likelihood method. The chi-square value for the model was statistically
significant p<.0001 level. As stated previously, the chi-square statistic may be used to
test the null hypothesis that the model fits the data. However, in practice, the statistic is
106
sensitive to sample size as well as to departures from multivariate normality. While the
Hoetler’s Critical N was at 1076 for the model (see Table 4.7), which was above 200 and
supported that the chi-square statistic was acceptable to use, it does not change the fact
that because of the chi-square’s reaction to a large sample size, using it can often result in
the rejection of a model that fits. Therefore, the chi-square ratio was used (James,
Mulaik, and Brett 1982; Joreskof and Sorbom 1989). The chi-square ratio for this model
Because the chi-square ratio was slightly above 2, and due to the sensitive nature
of the chi-square, other goodness of fit indices were also consulted in assessing the fit for
this model. Bourdieu’s CFA measurement model had goodness of fit values close to 1
for the non-normed-fit index (NNFI) (0.9737) and the comparative fit index (CFI)
(0.9813) as well as the Bollen Normed Index Rhol (0.9542), Bentler and Bonett’s (1980)
NFI (0.9674), and the Bollen (1988) Non-normed Index Delta2 (0.9814), meaning that
there was an acceptable fit, please see Table 4.7 (Bentler and Bonett 1980; Bentler 1989).
There were some high residuals in this model. The covariance between coworker
contact and religious membership had the highest standardized residual of -3.48, and also
the covariance between the religious membership and friend contact was high at -3.22.
However, there were many standardized residuals reporting at lower than two, some
under 1. These supported the argument that the predicted model was close to the actual
model (see Table 4.6). The average standardized residual for Bourdieu’s social capital
Some of the standardized residuals for indicators in the Bourdieu model were
high, however, as with the previous two theorists, these indicators were not deleted
107
because the measures were the best available for modeling the theory in the dataset and
many other fit statistics (reported above) showed support for them. The residual matrix
was only one of many results used to assess model fit and to choose the best theoretical
model to data fit. Thus, based on the EFA, and other CFA fit statistics listed above that
showed a good fit for Bourdieu’s social capital model, consequently the current model
(see Figure 4.5) was accepted as the final specified model. All analysis for Bourdieu in
this study was based on this model. Finally in answering Research Question one, finding
a good model fit for Bourdieu’s model showed support for his theory of social capital
modeled empirically.
Table 4.7. CFA-Covariance Structure Analysis: Maximum Likelihood Estimation for the
CALIS Procedure Confirmatory Factor Analysis - Fit Statistics –Bourdieu’s Social
Capital Model
Fit Statistic
Goodness of Fit Index (GFI) 0.9915
Chi-Square 73.1932
Chi-Square DF 32
Pr > Chi-Square42 <.0001
RMSEA Estimate 0.0275
Bentler’s Comparative Fit Index 0.9813
Benter’s & Bonett’s (1980) Non-normed Index 0.9737
Bentler & Bonett’s (1980) NFI 0.9674
Bollen (1988) Normed Index Rho1 0.9542
Bollen (1988) Non-normed Index Delta2 0.9814
Hoelter’s (1983), Critical N 1076
N 1704
Bourdieu’s results. The covariances for the latent fields are reported here as
unstandardized estimates where the religious domain covaried with the community
domain with an estimate of 0.12, significant with a t-score of 3.30. Whereas the religious
42 Note: The Chi-Square Ratio was 2.3:1 for Bourdieu’s social capital confirmatory factor analysis model
using a covariance structure analysis with maximum likelihood estimation for the CALIS procedure in
SAS.
108
domain covaried with the political domain with an estimate of 0.35 which was significant
with a t-score of 10.26; finally the community domain covaried with the political domain
All indicators loaded positively and significantly for the sample with t-scores
ranging from 10.64 to 25.29. The results indicate that a one standard deviation increase
in the religious field was associated with a 0.88 standard deviations increase in volunteer
standard deviations increase in religious groups. Therefore, in 2000, we can say that as
the religious social capital increased, there was a corresponding increase in volunteering
in religious activities for those in the sample; they also had higher loadings for religious
membership and for participation in religious groups. The strongest Beta was for
volunteer religious, and the weakest loading was on religious groups. In fact, only 12
percent of the variation in religious groups was explained by the religious field.
All indicators loaded positively and significantly for the sample on the
community field. Here a one standard deviation increase in the community field was
associated with a 0.77 standard deviations increase in friend contact, a 0.55 standard
hanging with friends. The strongest loading for this group along the community
dimension was friend contact, followed closely by hanging with friends, and a fairly
strong showing by coworker contact. In fact, 59 percent of the variation found in friend
109
contact, 30 percent of the variation in coworker contact, and 40 percent of the variation
Finally, all the indicators for political field loaded positively and significantly for
respondents in 2000. A one standard deviation increase in the political dimension was
associated with a 0.62 standard deviations increase in rally, a 0.54 standard deviations
increase in public interest groups, a 0.41 standard deviations increase in petition, and a
2000, respondents whose political social capital increased spent more time on rallies, in
public interest groups, signing petitions, and working on their neighborhoods. The
strongest loading for political social capital was rally. Not surprisingly, the R-square for
rally was fairly strong at 0.39. Therefore, 39 percent of the variation seen in rally was
in public interest groups, 17 percent of the variation in petition, and 11 percent of the
110
(0.42,[23.14], 0.88) Volunteer Religious
(V2) E2
Community (14.82,[25.29],0.77)
Field (D2) Friend Contact E42
(8.94,[19.73],0.55) (V42)
CoWorker Contact
(V43) E43
Rally (V16)
E16
(0.16,[16.85],0.54)
Public Interest
Groups(V27) E27
(0.20,[13.28],0.41)
Petition (V15)
E15
(0.14, [10.64],0.33)
Neighborhood Assoc.
(V5) E5
Figure 4.9. Bourdieu’s Social Capital Measurement Model, (unstandardized beta, [t-
scores], unstandardized Beta).43
43 Note: Model Fit Statistics: Chi-square = 73.1932, 32 df (p<.0001), RMSEA = 0.0275, CFI = 0.9813,
GFI = 0.9915. Adjusted R2 : V2 (0.77), V19 (0.37), V22 (0.12), V42 (0.59), V43 (0.30), V44 (0.40),
V16 (0.39), V27 (0.29), V15 (0.17), V5 (0.11). RMSEA = root mean square error of approximation;
CFI = comparative fit index; GFI = goodness-of-fit index.
111
CHAPTER V
In order to address the proposed hypotheses, each model was run using the multi-
sample SEM method. For example, the first hypothesis predicted social capital would
vary according to social class. As a result, the model was run across lower, middle and
upper class groups. The second hypothesis expected the structure of social capital to
operate differently across White and Non-White racial groups. The final hypothesis
tested below proposed that the structure of social capital would operate differently for
men versus women. In conducting these tests, the current chapter presents the results for
Putnam, Coleman, and Bourdieu’s social capital models respectively across social class,
race, and gender groups. Standardized Betas were reported for each individual model,
while the unstandardized betas were used to compare across models (groups). The
convergence criterion was satisfied for all models reported (Hatcher 1994).
Putnam’s social capital model was derived mainly from his work in Bowling
Alone (2000) where he attempted to understand why there was a decrease in social ties
and interactions among Americans. While the following does not constitute an
exhaustive test, it did allow for the analysis of Putnam’s social capital model with
112
bonding, bridging, and trust dimensions among important socio-demographic statuses
Two models were compared using Putnam’s social capital model, one that
allowed all coefficients to freely estimate and another that constrained all coefficients to
be the same. The likelihood ratio (LR) test44 in Table 5.1 showed that for Putnam’s
social capital model, social capital varied by income. What that means was that there
Moreover, the overall fit for the Putnam social capital model was better when the
parameter estimates were allowed to freely estimate with a chi-square ratio of 2.2:1.
However, most fit indexes ranged between 0.80 and 0.90. The CFI (0.8967) and NNFI
(0.8937) were both close to 0.90. The only index falling above 0.90 at 0.9620 was the
GFI. The RMSEA and SRMR (both looking at how well the data fit the theory) were at
0.05 and 0.06 respectively, well below 0.8 (Bollen 1989). There was enough evidence
with the chi-square ratio, GFI, RMSEA, and SRMSR to accept the model. Given the
evidence, we accept the unconstrained model as being better than the constrained. Thus,
44
The LR statistic is also known as a chi-square difference statistic and allows a researcher to do a statistic
of significance between models (Bollen 1989). It can be written as: LR = (N – 1) Fr – (N - 1)Fu
= (N-1) (Fr – Fu)
45
A standard .05 level of significance was used as a measure of statistical significance. If the LR statistic
was significant, then we accepted the more complex (freely estimated) model over the simpler
(constrained) model. Worded another way, if it was significantly different then the LR was significant
and thus worth losing for example 18 degree of freedom because there were significant differences
found.
113
the following presents results of an analysis that examined the multi-sample
unconstrained model for Putnam across lower, middle and upper social class groups.46
Other models were considered (see footnote three below), specifically models
where different groups were constrained and one left free to estimate. For example, one
run had the lower and upper classes constrained to estimate the same (by giving the path
the same name for both groups), but the middle class was allowed to estimate freely. It
was determined that these models did not add any important information to the analysis
and in the interest of parsimony where there were significant socio-demographic group
differences, the unconstrained model was reported. The unconstrained model is actually
the more complex model because all groups (e.g. lower, middle and upper class) are
allowed to estimate freely, thus it takes up more degrees of freedom and is more
complicated than the model where they are all assumed to work the same, or constrained.
Thus, when the LR test was statistically significant, the loss of degrees of freedom was
considered a worthy loss and the more complex model showing those differences was
reported. Where there was no significant difference, the findings for the original model
in chapter four were supported (please see chapter four for those results).
46
Three other constrained models were run as well. Here two groups at a time were constrained and one
allowed to freely estimate. Because there were significant differences found, the model showing the
differences between all groups was reported.
114
Table 5.1 Fit Statistics for Putnam's Social Capital Model Using
Multi-Sample SEM Social Class in the Sample, 2000.
*Constrained Unconstrained
All Free
chi-square 263.8587 234.6390
chi-square DF 114 105
Pr>Chi-Square 0.0000 0.0000
Chi-Square Ratio 2.3146 2.2347
Goodness of fit (GFI) 0.9575 0.9620
Bentler's CFI 0.8785 0.8967
Bentler & Bonett's NFI (1980) 0.8032 0.8278
Bollen Non-normed Index Delta 2 (1988) 0.8779 0.8937
Hoelter's (1983), Critical N 701 732
RMSEA 0.0547 0.0530
Standardized (SRMSR) 0.0806 0.0679
subtraction of chi-square/DF 29.2197 9
Likelihood Ratio (LR) Test 0.0006
Total N 1322
Lower 466
Middle 263
Upper 593
Reported here are the covariances for the latent domains’ standardized estimates.
The trust domain covaried with the bonding domain with an estimate of 0.47, significant
with a t-score of 11.45. Whereas the trust domain covaried with the bridging domain
with an estimate of 0.25 which was significant with a t-score of 5.74; finally the bonding
domain covaried with the bridging domain with an estimate of 0.45 and a t-score of
10.32.
The following figures show Putnam’s social capital model across social class
groups. For model 1, the lower class, all indicators significantly loaded on the bonding
social capital dimension (t-statistics no lower than 5.06). For those in the sample who
were lower class, one standard deviation increase in bonding social capital was associated
with a 0.67 standard deviations increase in friends, a 0.63 standard deviations increase in
115
shared confidences, and a 0.19 standard deviations increase in family contact. Of all
variables loading on bonding social capital domain, family contact had the weakest
loading. More than that family contact had an R-square of 0.04 which was low. This
suggests that four percent of the variance in family contact was explained by bonding
social capital. However, friends and shared confidences had R-squares of 0.44 and 0.40
which for social science is considered fairly strong. Thus, 44 percent of the variance in
friends was explained by bonding social capital domain and 40 percent of the variance in
shared confidences was explained by bonding social capital domain for the lower social
class group in the sample. These findings are consistent with Putnam’s (2000) theoretical
expectation that there is a positive correlation between close friends, more intense
relationships and social situations where exclusionary behavior leads to bonding social
capital.
Diversity, group involvement index, and worked with others also loaded
significantly on bridging social capital (t-statistics going no lower than 7.97). Thus, a
one standard deviation increase in bridging social capital was correlated with a 0.70
corresponding 0.55 standard deviations increase in the group involvement index, and a
0.31 standard deviations increase in worked with others. Clearly diversity had the
strongest loading and was therefore the strongest indicator for bridging social capital for
those in the sample from a lower social class. This was followed by group involvement
index, whereas, working with others had the weakest relationship on an increase in
bridging social capital. Not surprisingly, the R-squares for each indicator followed the
same pattern. Forty-nine percent of the variance in diversity was explained by bridging
116
social capital. Thirty percent of the variance in the group involvement index for those in
the sample in the lower social class was explained by bridging social capital. Finally,
only 10 percent of the variance in worked with others was explained by bridging social
capital.
Bonding (0.51,[15.88],
Share confidences
(D2) 0.63)
(V3) E3
(3.90, [5.06], 0.19)
Family contact (V27) E27
Bridging
(D3) (1.83, [15.56], Diversity (V9)
0.70) E9
(1.52, [13.24],
0.55) Group involvement E20
(0.14, [7.97], index (V20)
0.31)
Worked with others E25
(V25)
Trust (D1) (0.46, [16.68],
0.61) Rate your
community(V31) E31
(0.45, [13.78],
0.49)
Perception of
communitytrust(V22 E22
(0.51, [18.71],
0.71)
Social Trust Index(7) E7
Figure 5.1 Putnam’s Social Capital Model for Lower Social Class
(unstandardized b, [t-statistic], standardized Beta)Fit Indexes for Lower Class: SRMSR
0.07, GFI 0.96, NFI 0.84
Finally, all indicators loaded significantly on the dimension of trust with the
lowest t-statistic at 13.78 and the highest at 18.71. For those in the lower class, a one
117
rate your community, a 0.49 standard deviations increase in perception of community
trust, and a 0.71 standard deviations increase in the social trust index. The largest
increase was with the social trust index and the lowest with the perception of community
trust. The social trust index had the most variance explained by the trust dimension with
an R-square of 0.50 or 50 percent. While accounting for less variance, the perception of
community trust had 24 percent of variance explained by trust and 37 percent of the
variance in rate your community was explained by the trust dimension for respondents in
a lower social class. The lowest yield for perception of community trust may have been
due to the fact that most of the sample perceived their community in a positive light, it
may also have indicated that trust is related to one’s perception of their community only
so much.
When looking at the middle class from the SCCBS with Putnam’s social capital
model, covariances were the same as with the lower class model: The trust domain
covaried with the bonding domain with an estimate of 0.4747, significant with a t-score of
11.45; the trust domain covaried with the bridging domain with an estimate of 0.25 which
was significant with a t-score of 5.74; and finally the bonding domain covaried with the
Both the bridging and trust factorial dimensions had indicators with loadings that
were statistically significant. However, for the bonding domain, friends and share
confidences were significant, whereas family contact with a t-score of 1.33 was not.
Therefore, for the middle class in this sample, a one standard deviation increase in
bonding social capital was associated with a 0.62 standard deviations increase in friends
and a 0.57 standard deviations increase in shared confidences. The R-square for friends
47
Reported here are the covariances for the latent domains’ standardized estimates.
118
was 0.38 and for shared confidences it was 0.33. Thus, we can say that bonding social
capital explained 38 percent of the variance in friends, while bonding social capital
among the middle class explained 33 percent of the variance in shared confidences. That
the family contact loading was not statistically significant was surprising because
according to Putnam’s (2000) theory, bonding social capital comes from those ties where
respondents would have spent more time and energy and contact with, in other words,
family. Even with a weak finding, it was expected to be statistically significant. Thus,
we can say that for the middle class group, family contact was not a meaningful indicator
Those in the middle class had an increase of 0.65 standard deviations in diversity,
a 0.55 standard deviations increase in the group involvement index, and a 0.36 standard
deviations increase in worked with others for a one standard deviation in bridging social
capital. As with the lower class, the strongest indicator for bridging social capital was in
percent of the variation in the group involvement index was explained by bridging social
capital. Only 13 percent of the variation in worked with others was explained by bridging
social capital.
A one standard deviation increase in trust brought about 0.52 standard deviations
community trust, and a 0.68 standard deviations increase in the social trust index. Stated
another way, rate your community correlated with the domain of trust at 0.27, this was a
somewhat weak correlation, closer to zero than one, but it was positive and significant.
Thus as one’s bonding domain increased, so did their rating of the community in which
119
they lived. For the social trust index, which had the strongest correlation with the trust
The R-squares showed that for those from the middle class, 50 percent of the
variance in the social trust index, 37 percent of the variance in rate your community, and
24 percent of the variation in perception of community trust was explained by the trust
dimension. Therefore, not only did all three indicators load, but the trust dimension was
a fairly strong predictor for each one, although the index being the strongest based on the
standardized Beta.
Bridging
(D3) (1.60, [8.72], Diversity (V9)
0.65) E9
(1.50, [7.35], 0.55)
Group involvement E20
index (V20)
(0.17, [4.63], 0.36)
Worked with others E25
(V25)
Trust (D1)
(0.36, [6.91], 0.52) Rate your
community(V31) E31
(0.22, [3.39], 0.27)
Perception of
communitytrust(V22 E22
(0.46, [9.62], 0.68)
Social Trust Index(7) E7
Figure 5.2 Putnam’s Social Capital Model for Middle Social Class
(unstandardized b, [t-statistic], standardized Beta) Fit Indexes Middle Class: SRMSR
0.07, GFI 0.96, NFI 0.79
120
As shown above, the model worked along similar lines to that of the lower class
worked with others (0.17 slightly stronger than 0.14), were weaker. Also, the t-statistics
for the middle class were weaker without exception. Based on these findings, we can
argue that the lower class had stronger indicators for social capital than the middle class
Furthermore, for the middle class, visiting one’s family had no bearing on
bonding social capital whatsoever. In terms of bonding social capital however, it was
expected that family would be an important component and that it would be a part of
one’s bonding social capital. This was expected given the idea these were ties that one
would spend more time and effort on maintaining, creating stronger and more intense
relationships which exclude others who are not family members (Putnam 2000). It would
be interesting to see if this finding holds if applied to a different sample, and with other
methods. However, the current results indicate that for the middle class, it was their
friends and people they shared confidences with that mattered most.
When one reviews the upper class model, as with the lower and middle class
models, the trust domain covaried with the bonding domain with an estimate of 0.4748,
significant with a t-score of 11.45; the trust domain covaried with the bridging domain
with an estimate of 0.25 which was significant with a t-score of 5.74; and finally the
bonding domain covaried with the bridging domain with an estimate of 0.45 and a t-score
of 10.32.
The indicators for the bonding domain, friends, share confidences, and family
contact loaded significantly for the upper class model. Therefore, a one standard
48
Reported here are the covariances for the latent domains’ standardized estimates.
121
deviation increase in bonding social capital domain was correlated with a 0.67 standard
confidences, and a 0.19 standard deviations increase in family contact. The R-square for
friends was 0.44, for shared confidences it was 0.40, and for family contact it was 0.04.
Thus, the bonding social capital domain explained 44 percent of the variance in friends,
while the bonding social capital domain among the upper class explained 40 percent of
the variance in shared confidences, and only four percent of the variance in family
On the bridging domain, the upper class had an increase of 0.70 standard
index, and a 0.31 standard deviations increase in worked with others for a one standard
deviation increase in bridging social capital domain. As with the lower and middle
classes, the strongest indicator for bridging social capital was in the diversity of
friendships. Stated in a different way, the diversity of one’s friendships correlated with
the bridging domain fairly strongly at 0.70. Forty-nine percent of the variation in
diversity and 30 percent of the variation in the group involvement index was explained by
bridging social capital domain. Only 10 percent of the variation in worked with others
was explained by bridging social capital domain. The last was three percent smaller than
The percent of variance explained for worked with others was surprising as some
researchers have emphasized the need to include action measures of social capital along
with structural or network variables (see Demant and Jarvinen 2011). For example,
Norris and Inglehart (2006) argue that social capital “needs to gauge activism as well as
122
formal membership” (80). However, the findings here suggest that working with others
in one’s community toward a common goal may not be a part of bridging social capital
for those in the middle and upper classes for this sample.
For the upper class, all indicators loaded significantly on the dimension of trust
with the lowest t-score at 13.78 and the highest at 18.71. For those in the upper class, a
community trust, and a 0.71 standard deviations increase in the social trust index. As
with the lower class, the largest increase was with the social trust index and the lowest
with the perception of community trust. The social trust index had the most variance
accounting for less variance, the perception of community trust had 24 percent of
variance explained by trust and 37 percent of the variance in rate your community was
explained by the trust dimension for respondents in a lower social class. Again, the
lowest yield for perception of community trust may have been related to the fact that most
of the sample perceived their community in a positive light, it may also have indicated
that trust is associated with one’s perception of their community only so much.
lower class and upper class with regard to the dimensions of bonding and bridging social
capital or trust.49 Interestingly, the only significant differences lay with the middle class.
Not only was the structure or loadings of social capital diminished for the middle class,
the tests of significance were also, to the point that family contact was not at all
significant under the bonding social capital factor. It can be argued that there was little
49
The R-squares for the upper class were also identical to those of those in the lower class.
123
difference between class groups using Putnam’s model, and that therefore, the original
results from chapter four should stand regardless of the LR test being significant.
However, the finding, though small, shows a potentially important difference for the
middle class and challenges a basic assumption for the role of the family in our society.
The results presented here are not conclusive and it would make sense to do further
testing using various methods and other samples to confirm and support the results found
in this model. Also, the N for the middle class group was lower at 263 than either that of
the lower or upper social class groups (see Table 5.1), therefore, this finding was
potentially associated with a loss of statistical power due to a smaller sample size
124
(0.73, [16.49], Friends (V2)
E2
0.67)
Bonding (0.51,[15.88],
Share confidences
(D2) 0.63)
(V3) E3
(3.91, [5.06], 0.19)
Family contact (V27) E27
Bridging
(D3) (1.83, [15.56], Diversity (V9)
0.70) E9
(1.52, [13.24],
0.55) Group involvement E20
(0.14, [7.97], index (V20)
0.31)
Worked with others E25
(V25)
Trust (D1) (0.46, [16.68],
0.61) Rate your
community(V31) E31
(0.45, [13.78],
0.49)
Perception of
communitytrust(V22 E22
(0.51, [18.71],
0.71)
Social Trust Index(7) E7
Figure 5.3 Putnam’s Social Capital Model for Upper Social Class (unstandardized b, [t-
statistic], standardized Beta) Fit Indexes Upper Class: SRMSR 0.07, GFI 0.96, NFI 0.84
Race was divided into two categories, White and Non-white. The hypothesis
tested here was that race would be significant for Putnam’s model of social capital. In
order to test this hypothesis, as with social class, the LR test was used to compare the
constrained model with the unconstrained model which allows all factor loadings to be
different for Whites and Non-Whites. As shown below in Table 5.2, that test was highly
significant, supporting the claim that there were statistically significant differences in the
125
structure of social capital for Whites and Non-whites. Because the test supports the
hypothesis, the unconstrained or free model is reported which shows how social capital
varies when using Putnam’s social capital model. This model was considered more
complex than the constrained, but because the test was significant it was worth losing 9
degrees of freedom to understand the difference in the structure of social capital for racial
groups. The chi-square ratio for this model was 2.4:1, which was slightly higher than 2.0,
and most indexes reported at 0.90 or above. Both the RMSEA and SRMSR were less
than .05 supporting a model fit. The only index that dipped below 0.90 at 0.8611 was
Bentler & Bonett’s NFI (1980). Overall, this model fits the data well and as such was
accepted.
Table 5.2. Fit Statistics for Putnam’s Social Capital Model using
Multi-Sample SEM of Race in the Sample, 2000.
Constrained* Unconstrained
ALL Free
chi-square 192.7500 146.7604
chi-square DF 69 60
Pr>Chi-Square 0.0000 0.0000
Chi-Square Ratio 2.7935 2.4460
Goodness of fit (GFI) 0.9268 0.9739
Bentler’s CFI 0.8743 0.9119
Bentler & Bonett’s NFI (1980) 0.8176 0.8611
Bollen Non-normed Index Delta 2 (1988) 0.8747 0.9129
Hoelter’s (1983), Critical N 545 634
RMSEA 0.0553 0.0497
Standardized (SRMSR) 0.0694 0.0459
Total N 1175
White N 883
Non-White N 292
subtraction of chi-square/DF 46 9
Likelihood Ratio (LR) Test 0.0000
The following figures show Putnam’s social capital model across the racial
groups of Whites and Non-Whites. The trust domain covaried with the bonding domain
126
with an estimate of 0.4050, significant with a t-score of 8.34; the trust domain covaried
with the bridging domain with an estimate of 0.26 which was significant with a t-score of
5.36; and finally the bonding domain covaried with the bridging domain with an estimate
For model 1, Whites, all indicators significantly loaded on the bonding social
capital dimension. For Whites in the sample, a one standard deviation increase in
bonding social capital was associated with a 0.69 standard deviations increase in friends,
a 0.50 standard deviations increase in shared confidences, and a 0.19 standard deviations
increase in family contact. The strongest correlation was between friends and the bonding
domain for Whites. The R-square followed the same pattern with a R-square of 0.48 for
friends, 0.25 for shared confidences, and 0.04 for family contact. Therefore, we can say
that bonding social capital accounts for 48 percent of the variance in friends, 25 percent
of the variance in shared confidences, and only 4 percent of the variance in family
contact. Of all variables loading on bonding social capital, family contact had the
weakest loading. While unexpected, the theoretical writings of Putnam (2000) argued
that ties with close friends and those one feels closest to (e.g. feel comfortable sharing
confidences with) were a part of bonding social capital. However, these findings suggest
that when using Putnam’s social capital model on Whites as a group, family contact was
less important to one’s bonding social capital than friends or those with whom one shares
confidences.
capital for Whites. A one standard deviation increase in bridging social capital was
associated for Whites with an increase of 0.67 standard deviations in diversity, 0.52
50
Reported here are the covariances for the latent domains’ standardized estimates.
127
standard deviations in group involvement index, and 0.32 standard deviations in worked
with others. For Whites, bridging social capital explained 45 percent of the variance in
diversity, 27 percent of the variance in the group involvement index, and 10 percent of the
variance in worked with others. The strongest indicator for Whites along the dimension
of bridging social capital was their diversity of friendships, while the weakest loading
was Whites working with those in their community toward a common goal. Across all
models, worked with others, appears to be the weakest indicator for bridging social
capital.
Bonding (0.37,[10.81],
Share confidences
(D2) 0.50)
(V3) E3
(3.79, [4.33], 0.19)
Family contact (V27) E27
Bridging
(D3) (1.68, [13.49], Diversity (V9)
0.67) E9
(1.35, [11.20],
0.52) Group involvement E20
(0.15, [7.38], index (V20)
0.32)
Worked with others E25
(V25)
Trust (D1) (0.39, [12.46],
0.57) Rate your
community(V31) E31
(0.39, [10.37],
0.46)
Perception of
communitytrust(V22 E22
(0.35, [12.05],
0.56)
Social Trust Index(7) E7
Figure 5.4 Putnam’s Social Capital Model, Whites in the Sample, 2000 (unstandardized
b, [t-statistic], unstandardized Beta)Fit Indexes: SRMSR 0.04, GFI 0.98, NFI 0.92
128
Finally, using Putnam’s social capital model rate your community, perception of
community trust and the social trust index loaded significantly on the dimension of trust
for Whites in 2000. A one standard deviation increase in trust was associated with a 0.57
standard deviations increase in rate your community, a 0.46 standard deviations increase
in perception of community trust, and a 0.56 standard deviations increase in the social
trust index for Whites in the sample. The trust dimension explained 32 percent of the
trust, and 32 percent of the variance in the social trust index. For Whites then, rate your
community and the social trust index had the same amount of variance explained by the
trust dimension. Interestingly, all three indicators for trust were similar in strength,
although the social trust index was the strongest indicator for Whites when looking at the
standardized Betas.
When Putnam’s social capital model was run using the Non-White sample, the
trust domain covaried with the bonding domain with an estimate of 0.4051, significant
with a t-score of 8.34; the trust domain covaried with the bridging domain with an
estimate of 0.26 which was significant with a t-score of 5.36; and finally the bonding
domain covaried with the bridging domain with an estimate of 0.49 and a t-score of
10.58.
For Non-Whites in the sample, two indicators, friends and share confidences,
loaded significantly on the dimension of bonding social capital. However, family contact
was not statistically significant. In contrast, all indicators loaded significantly for Non-
51
Reported here are the covariances for the latent domains’ standardized estimates.
129
A one standard deviation increase in bonding social capital for Non-Whites was
correlated with a 0.73 standard deviations increase in friends, and a 0.67 standard
deviations increase in shared confidences. The friends correlation to the bonding domain
was fairly strong at 0.73. Family contact did not load on the dimension of bonding social
capital for Non-Whites. This was further supported by a nearly non-existent R-square of
0.005. Whereas the R-square for friends was 0.53 and for shared confidences was at
0.45. Therefore, the results show that 53 percent of the variance in friends and 45 percent
of the variance in shared confidences was explained by bonding social capital for Non-
social capital was associated with a 0.73 standard deviations increase in diversity, a 0.68
standard deviations increase in group involvement index, and a 0.36 standard deviations
increase in worked with others. The R-square for diversity was 0.53, for the group
involvement index was 0.46 and for worked with others it was at 0.13. Both diversity and
the group involvement index were strongly predicted by bridging social capital in the
model, however, worked for others was weakly predicted by bridging social capital for
Non-Whites.
The dimension of trust for Non-Whites also had three significant indicators. Here
a one standard deviation increase in trust translated to a 0.58 standard deviations increase
trust, and a 0.71 standard deviations increase in the social trust index for Non-Whites.
The R-squares for the trust indicators for Non-Whites were 0.34 for rate your community,
0.34 for perception of community trust, and 0.51 for the social trust index. The strongest
130
loading was for the social trust index, where 51 percent of the variation in it was
explained by the trust domain. However, rate your community and perception of
community trust were fairly strong predictors for this domain as well.
While family contact was not significant for Non-Whites on the bonding social
capital dimension, it was significant for Whites. This suggests that family contact was
structurally more important for Whites than Non-Whites in 2000. However, when
looking at the unstandardized loadings, except for family contact, Non-Whites had a
stronger loading of social capital across bonding, bridging and trust. Non-Whites tended
to have stronger loadings for bonding social capital in terms of the number of close
friends reported and the number of people they felt comfortable sharing confidences.
This means that having close friends and people one can share confidences with were
stronger indicators for Non-Whites on the bonding domain than for Whites.
131
(0.78, [11.61], Friends (V2)
E2
0.73)
Bonding (0.58,[10.89],
Share confidences
(D2) 0.67)
(V3) E3
(1.46,[1.04]NS 0.07)
Family contact (V27) E27
Bridging
(D3) (2.01, [12.08], Diversity (V9)
0.73) E9
(2.07, [11.28],
0.68) Group involvement E20
(0.17, [5.35], index (V20)
0.36)
Worked with others E25
(V25)
Trust (D1) (0.41, [9.11],
0.58) Rate your
community(V31) E31
(0.55, [9.33],
0.59)
Perception of
communitytrust(V22 E22
(0.53, [12.08],
0.71)
Social Trust Index(7) E7
Figure 5.5 Putnam’s Social Capital Model, Non-Whites in the Sample, 2000
(unstandardized b, [t-statistic], unstandardized Beta)Fit Indexes: SRMSR 0.07, GFI 0.94,
NFI 0.73
Whites had weaker loadings for bridging social capital in comparison with Non-
Whites and structurally they had weaker loadings for diverse friendships and group
involvement. Non-Whites had stronger loadings for working with others in their
communities toward a common goal. Along the trust dimension, Whites and Non-Whites
were very similar structurally in rating their communities, however, Non-Whites had
stronger loadings of both perception of community trust and social trust index. These
were not indicators that were race specific (see Chapter 3 for more details on variables).
132
In general, the data showed that Whites in the sample had weaker loadings for the
indicators of bonding and bridging social capital as well as trust. The only dimension
that varied from that was bonding. Here, family contact was significant for Whites in the
sample and not at all for Non-Whites. However, Non-Whites had stronger loadings of
both friends and shared confidences when compared with Whites in 2000.
Gender was divided into two groups, women and men. The hypothesis tested
here was that gender would have a significant association with the structure of Putnam’s
model of social capital. In order to test this hypothesis, as with social class and race, the
LR test was used to compare the constrained model with the unconstrained model. As
shown below in Table 5.3, that test was not significant (0.109). Because the LR test was
not significant we cannot argue there were any significant differences in the structure of
social capital for women and men using Putnam’s model. Therefore, it did not make
sense to report the unconstrained model and lose more degrees of freedom.
Consequently, the findings reported in Chapter 4 hold for both women and men’s social
133
Table 5.3. Fit Statistics for Putnam’s Social Capital Model
Multi-Sample SEM with Gender and Social Capital.
Constrained Unconstrained
chi-square 181.3202 166.9233
chi-square DF 69 60
Pr>Chi-Square 0.0000 0.0000
Chi-Square Ratio 2.6:1 2.8:1
Goodness of fit (GFI) 0.9722 0.9745
Bentler’s CFI 0.9271 0.9306
Bentler & Bonett’s NFI (1980) 0.8875 0.8964
Bollen Non-normed Index Delta 2 (1988) 0.9272 0.9311
Hoelter’s (1983), Critical N 713 685
RMSEA 0.0475 0.0497
Standardized (SRMSR) 0.0506 0.0467
Total N 1445
Female N 866
Male N 579
subtraction of chi-square/DF 14.3969 9
The following reports multi-sample SEM analysis using Coleman’s social capital
model for class, race, and gender. These findings were based off of the findings from the
As with Putnam, two models were compared using Coleman’s social capital
model, one that allowed all coefficients to freely estimate and another that constrained all
coefficients to be the same. The likelihood ratio (LR) test52 in Table 5.4 showed that
52
The LR statistic is also known as a chi-square difference statistic and allows a researcher to do a statistic
of significance between models (Bollen 1989). It can be written as: LR = (N – 1) Fr – (N - 1)Fu
= (N-1) (Fr – Fu)
134
social capital did not vary by income for Coleman’s model. What that means is that there
were no meaningful differences in the structure of social capital by income groups53. The
overall fit for the Coleman’s social capital model was better when the parameter
estimates were constrained with a chi-square ratio of 1.15:1. However, most fit indexes
fell above 0.90 and the RMSEA and SRMR (both looking at how well the data fit the
theory) were at 0.04 for each, well below 0.8. There was enough evidence with the chi-
square ratio, GFI, RMSEA, and SRMSR to accept the models. However, given that the
LR test was not significant showing no support for the first hypothesis, the original model
analysis holds for different social class groups and the structure of social capital when
53
A standard .05 level of significance was used as a measure of statistical significance. If the LR statistic
was significant, then we accepted the more complex (freely estimated) model over the simpler
(constrained) model. Worded another way, if it was significantly different then the LR was significant
and thus worth losing for example 18 degree of freedom because there were significant differences
found.
54
Three other constrained models were statisticed as well, where two groups at a time were constrained and
one allowed to freely estimate. Because there were significant differences found, those models were not
reported.
135
Table 5.4. Fit Statistics for Coleman’s Social Capital Model,
Multi-Sample Structural Equation Modeling for Social Class.
*Constrained Unconstrained
All Free
chi-square 130.7839 118.9576
chi-square DF 114 96
Pr>Chi-Square 0.1346 0.0562
Chi-Square Ratio 1.1472 1.2391
Goodness of fit (GFI) 0.9850 0.9863
Bentler’s CFI 0.9909 0.9875
Bentler & Bonett’s NFI (1980) 0.9328 0.9389
Bollen Non-normed Index Delta
2 (1988) 0.9908 0.9876
Hoelter’s (1983), Critical N 2029 1911
RMSEA 0.0153 0.0195
Root Mean (RMSR) 0.1245 0.1135
Standardized (SRMSR) 0.0421 0.0381
subtraction of chi-square/DF 11.8263 18
Likelihood Ratio (LR) Test 0.856074397
N 1898
Lower 642
Middle 745
Upper 511
The two models compared for race differences using Coleman’s social capital
model had one freely estimated and one constraining, Whites and Non-Whites to be the
same. The likelihood ratio (LR) test55 (0.000) in Table 5.5 showed that for Coleman’s
social capital model, social capital varied by race. Therefore we can say there were
55
The LR statistic is also known as a chi-square difference statistic and allows a researcher to do a statistic
of significance between models (Bollen 1989). It can be written as: LR = (N – 1) Fr – (N - 1)Fu
= (N-1) (Fr – Fu)
56
A standard .05 level of significance was used as a measure of statistical significance. If the LR statistic
was significant, then we accepted the more complex (freely estimated) model over the simpler
(constrained) model. Worded another way, if it was significantly different then the LR was significant
136
Moreover, the overall fit for the Coleman social capital model was better when the
parameter estimates were allowed to freely estimate with a chi-square ratio of 1.7:1.
Most fit indexes were close to 1 for the unconstrained model, whereas the NFI dipped
below 0.90 for the constrained model. Further, the RMSEA was lower for the
unconstrained model at 0.04 and the SRMSR was also lower at 0.04, showing support
that the data fits the theoretical model. Therefore, there was enough evidence with the
chi-square ratio, fit indexes, RMSEA, and SRMSR to accept the unconstrained model.
Furthermore, the statistically significant LR test showed support for the second
unconstrained model for Coleman across Whites and Non-Whites in the sample.
and thus worth losing for example 18 degree of freedom because there were significant differences
found.
137
Table 5.5 Fit Statistics for Multi-Sample SEM of Race
and Social Capital in the Sample, 2000.
Coleman
Constrained Unconstrained
For the covariances among domains, for Whites, the E/O domain covaried with
the IC domain with an estimate of 0.53,57 significant with a t-score of 11.87; the E/O
domain covaried with the SN domain with an estimate of 0.21 which was significant with
a t-score of 4.38; and finally the SN domain covaried with the IC domain with an
The following figures show Coleman’s social capital model across racial groups.
(E/O) dimension for model 1 (Whites). Thus we can say for those in the sample who
57
Reported here are the covariances for the latent domains’ standardized estimates.
138
were White, a one standard deviation increase in E/O social capital was associated with a
0.81 standard deviations increase in the group involvement index, a 0.43 standard
deviations increase in other church participation, and a 0.52 standard deviations increase
in club meetings. Of all variables loading on E/O social capital, other church
participation and club meetings both loaded fairly strongly, although the latter was the
weakest loading for E/O social capital. The group involvement index had a very strong
loading at 0.81 for Whites. Stated another way, the correlation between the group
involvement of those from the sample with the E/O domain was both positive and strong
As with Putnam, the R-square statistic was used to assess how well the model and
its parts predicted for Whites. Only 18 percent of the variation in other church
participation was explained by E/O social capital for Whites. However, 27 percent of the
variation in club meetings and 65 percent of the variation in group involvement index was
explained by E/O social capital, which were fairly strong and very strong results for R-
square. It was clear that for Whites in 2000, E/O social capital was a strong predictor for
group involvement.
(IC) for Whites. Here a one standard deviation increase in the IC domain was associated
with a 0.18 standard deviations increase in fictive kin, a 0.63 standard deviations increase
in business owner, and a 0.60 standard deviations increase in vacation home. The
strongest loadings were business owner and vacation home. Interestingly fictive kin had a
very weak loading and using R-square the IC accounted for only 3 percent of the
variation in fictive kin. In contrast, 39 percent of the variation in business owner and 36
139
percent of the variation in vacation home was explained by the IC for Whites. These
findings suggest that fictive kin was not all that important to Whites in the context of IC
Group involvement
E15
(2.07, [16.57],0.81 ) index (V15)
Social
(0.40,[11.38], 0.58) Rate your
Norms
(SN) D3 community(V8) E8
(0.39,[9.50], 0.46 )
Perception of
community trust(V9) E9
(0.34,[10.65], 0.56)
Social Trust E10
Index(V10)
Figure 5.6. Coleman’s Social Capital Model Multi-Sample SEM for Whites in 2000,
(unstandardized beta, [t-statistic], unstandardized Beta)Fit Indexes: SRMSR 0.04, GFI
0.99, NFI 0.94.
All indicators for the social norm dimension (SN) of Coleman’s social capital
loaded at a statistically significant level. For Whites, a one standard deviation increase in
SN was correlated with a 0.58 standard deviations increase in rate your community, a
140
0.46 standard deviations increase in perception of community trust, and a 0.56 standard
deviations increase in the social trust index. While they were not identical, it was
interesting how close in strength all three indicators loaded. Similarly, 34 percent of the
trust, and 31 percent of the variation in the social trust index was explained by SN. Thus
we can say that Whites overall have a fairly strong sense of social norms in the form of
trust.
The other racial group under study was Non-White; this group included anyone
who did not self-identify as White. For the covariances among domains, the E/O domain
covaried with the IC domain with an estimate of 0.53,58 significant with a t-score of
11.87; the E/O domain covaried with the SN domain with an estimate of 0.21 which was
significant with a t-score of 4.38; and finally the IC domain covaried with the SN domain
For the E/O dimension of Coleman’s social capital model, all indicators loaded
significantly. Therefore we can say for Non-Whites, a one standard deviation increase in
E/O social capital was associated with a 0.88 standard deviations increase in the group
involvement index, a 0.35 standard deviations increase in other church participation, and
a 0.52 standard deviations increase in club meetings. While other church participation
and club meetings were not weak, clearly group involvement loaded strongly for Non-
Whites on the E/O dimension of social capital. In terms of the E/O predicting these
indicators, for Non-Whites only 12 percent of the variance in other church participation
was explained by the E/O. While 27 percent of the variance in club meetings and 77
percent of the variance in group involvement index were explained by the E/O dimension.
58
Reported here are the covariances for the latent domains’ standardized estimates.
141
For the IC dimension, all indicators loaded significantly for Non-Whites. A one
standard deviation increase in IC social capital was associated with a 0.27 standard
deviations increase in fictive kin for Non-Whites. While a one standard deviation
increase in IC social capital was associated with a 0.67 standard deviations increase in
business owner and a 0.54 standard deviations increase in vacation home. IC social
capital explained the most variance for business owner, 45 percent. Vacation home had
29 percent of its variance explained by IC social capital, however, only 7 percent of the
variance in fictive kin was explained by IC social capital. The latter finding suggests that
for Non-Whites, IC social capital had less to do with the people one sees as “family” and
more to do with ties to people with specific resources or knowledge (e.g. owning one’s
own business).
142
Group involvement
E15
(2.81, [16.11],0.88 ) index (V15)
Social
(0.41,[8.62], 0.59) Rate your
Norms
(SN) D3 community(V8) E8
(0.46,[7.51], 0.52)
Perception of
community trust(V9) E9
(0.53,[11.34], 0.72)
Social Trust E10
Index(V10)
Figure 5.7. Coleman’s Social Capital Model Multi-Sample SEM for Non-Whites in 2000,
(unstandardized beta, [t-statistic], unstandardized Beta)Fit Indexes: SRMSR 0.06, GFI
0.96, NFI 0.80.
All indicators for the SN social capital dimension loaded significantly. For Non-
Whites, a one standard deviation increase in SN social capital was a correlated with a
0.59 standard deviations increase in rate your community, a 0.52 standard deviations
increase in the perception of community trust, and a 0.72 standard deviations increase in
the social trust index. Therefore, for Non-Whites in the sample, there were positive,
significant, and fairly strong correlations between the indicators and the SN domain.
This was strongest for the social trust index where 52 percent of the variance was
143
explained by SN social capital. The other two indicators had fairly strong R-squares also,
0.35 for rate your community and 0.27 for perception of community trust. These findings
support that for Non-Whites there was a fairly strong amount of trust, here measured as a
When analyzing the unstandardized betas a more complex picture was revealed
for the differences between Whites and Non-Whites using Coleman’s social capital
model. Non-Whites had structurally stronger (2.81) group involvement than Whites
(2.07) in the sample. However, Whites (0.21) had a slightly stronger loading for other
more about E/O social capital for Non-Whites than Whites; and other church
participation tells us slightly more for Whites than Non-Whites. For club meetings both
For the IC dimension of Coleman’s model, Non-Whites (1.06) had stronger fictive
kin loadings than Whites (0.67), both groups were similar on business owner (0.29 and
0.33), but Whites loaded slightly stronger on vacation home. Along the SN dimension,
both groups were basically the same for rate your community (0.40 and 0.41), however,
Non-Whites (0.46) loaded slightly stronger on perception of community trust than Whites
(0.39). Finally, Non-Whites (0.53) in the sample had a stronger loading for trust than
Whites (0.34).
In sum, the E/O domain indicators were stronger for Non-Whites than Whites,
due in part to their high loading of group involvement. Non-Whites also had stronger
loading on IC social capital, especially seen in the structure of non-relations whom they
considered part of the family. Moreover, Non-Whites had stronger loadings of trust than
144
Whites, here as part of social norms. These findings could be understood through the
lens of privilege. Being White in America carries with it some privilege (Doane 2003;
Feagin and McKinney 2003), and as such, it may be that Whites do not have to work as
hard along the aspect of social norms for their social capital. However, it was interesting
that Non-Whites had structurally stronger loadings for fictive kin, although Stack (1974)
did write about the need for strong social networks beyond one’s immediate family. It
may be that those in the minority have greater need of these forms of social capital and
thus put more time and energy into creating and maintaining it, as seen with the
As with Putnam, two models were compared using Coleman’s social capital
model, one that allowed all coefficients to freely estimate and another that constrained all
coefficients to be the same. The likelihood ratio (LR) test59 (0.14) in Table 5.6 showed
that social capital did not vary by gender for Coleman’s social capital model. Therefore
we can say that there was no support for Hypothesis three, and that there were no
Coleman’s social capital model. The overall fit for Coleman’s social capital model was
essentially the same for the constrained and unconstrained models. Both models fit well
with a chi-square ratio of 1.4:1 for each. The fit indexes fell well above 0.90 and the
59
The LR statistic is also known as a chi-square difference statistic and allows a researcher to do a statistic
of significance between models (Bollen 1989). It can be written as: LR = (N – 1) Fr – (N - 1)Fu
= (N-1) (Fr – Fu)
60
A standard .05 level of significance was used as a measure of statistical significance. If the LR statistic
was significant, then we accepted the more complex (freely estimated) model over the simpler
(constrained) model. Worded another way, if it was significantly different then the LR was significant
and thus worth losing for example 18 degree of freedom because there were significant differences
found.
145
RMSEA and SRMR (both looking at how well the data fit the theory) were at 0.03 for
each, well below 0.8. There was enough evidence with the chi-square ratio, GFI,
RMSEA, and SRMSR to accept the models. However, given that the LR test was not
significant, the original model analysis for Coleman holds for men and women.
Bourdieu’s social capital was a small, but a potentially important piece of his
larger theory on inequality. While the following analysis did not include the three other
forms of capital from Bourdieu, it was based on his theory and incorporates, as much as
the data would allow, for context in the form of fields. The following analysis follows
the same course as Putnam and Coleman, presenting the multi-sample SEM results for
146
Bourdieu’s Model: Social Class
Two models were compared using Putnam’s social capital model, one that
allowed all coefficients to freely estimate and another that constrained coefficients to be
the same across groups. The first hypothesis was supported for this model using the
likelihood ratio (LR) test61 (0.0000) in Table 5.7, which showed that for Bourdieu’s
social capital model, social capital varied by income. In other words, there were
However, most fit indexes ranged between 0.80 and 0.90. The only index falling
above 0.90 at 0.9635 was the GFI. The RMSEA and SRMR (both looking at how well
the data fit the theory) were at 0.06 and 0.07 respectively. While the fit of the model
itself was not as good as the original (see chapter 4), there was enough evidence with the
chi-square ratio (2,6:1), GFI, RMSEA, and SRMSR to accept the model. Given the
evidence, we accept the unconstrained model as being better than the constrained and
showing important structural differences in social capital along class groups. Thus, the
following analysis examines the multi-sample unconstrained model for Bourdieu across
61
The LR statistic is also known as a chi-square difference statistic and allows a researcher to do a statistic
of significance between models (Bollen 1989). It can be written as: LR = (N – 1) Fr – (N - 1)Fu
= (N-1) (Fr – Fu)
62
A standard .05 level of significance was used as a measure of statistical significance. If the LR statistic
was significant, then we accepted the more complex (freely estimated) model over the simpler
(constrained) model. Worded another way, if it was significantly different then the LR was significant
and thus worth losing for example 18 degree of freedom because there were significant differences
found.
63
Three other constrained models were statisticed as well, where two groups at a time were constrained and
one allowed to freely estimate. Because there were significant differences found, those models were not
reported.
147
Table 5.7. Fit Statistics for Bourdieu’s Social Capital Model,
Multi-Sample SEM for Social Class, U.S., 2000.
*Constrained Unconstrained
All Free
chi-square 382.8819 320.4031
chi-square DF 142 122
Pr>Chi-Square 0.0000 0.0000
Chi-Square Ratio 2.6964 2.6263
Goodness of fit (GFI) 0.9555 0.9635
Bentler’s CFI 0.8771 0.8988
Bentler & Bonett’s NFI (1980) 0.8172 0.8471
Bollen Non-normed Index Delta
2 (1988) 0.8767 0.8994
Hoelter’s (1983), Critical N 716 745
RMSEA 0.0564 0.0552
Root Mean (RMSR) 6.4982 3.5584
Standardized (SRMSR) 0.0890 0.0697
subtraction of chi-square/DF 62.4788 20
Likelihood Ratio (LR) Test 0.0000
N 1605
Lower 467
Middle 790
Upper 348
Using Bourdieu’s social capital model for the lower class, the religious domain
covaried with the community domain with an estimate of -0.06,64 with a non-significant
t-score of -1.64; the political domain covaried with the religious domain with an estimate
of -0.17 which was significant with a t-score of -4.48; and finally the community domain
covaried with the political domain with an estimate of 0.18 and a t-score of 4.34.
For model 1, the lower class, all indicators significantly loaded on the religious
social capital dimension. For those in the sample who were lower class, one standard
deviation increase in the religious field was associated with a 0.84 standard deviations
64
Reported here are the covariances for the latent domains’ standardized estimates.
148
membership, and a 0.36 standard deviations decrease in religious groups. Therefore, as
significant decrease in the amount they volunteered for religious groups, their religious
social capital, religious groups had the weakest loading. Moreover, religious groups had
an R-square of 0.13 which was low. That meant that only that 13 percent of the variance
in religious groups was explained by the religious field. Alternatively, the strongest
loading for the religious field was -0.84 for volunteer religious. The R-square for
volunteer religious supports the argument that this was the strongest indicator, 70 percent
of its variance was explained by the religious field for those in the lower class. Religious
membership had 41 percent of its variance explained by the religious field. For social
Friend contact, coworker contact, and hanging with friends also loaded
significantly on the community field dimension (t-statistics going no lower than 12.54).
Thus, a one standard deviation increase in community social capital was correlated with a
0.76 standard deviations increase in friend contact, a 0.63 standard deviations increase in
the coworker contact, and a 0.61 standard deviations increase in hanging with friends.
Clearly friend contact had the highest loading and was therefore the strongest indicator
for community social capital for those in the sample from a lower social class. For those
in the lower class, 58 percent of the variation in friend contact, 40 percent of the variation
in coworker contact, and 37 percent of the variation in hanging with friends was
65
The coding for these indicators were checked and re-checked for error, the analysis was re-run to check
for possible error in the first run. No problems were found with the coding and the findings remained.
149
Further, all indicators for the political field loaded at statistically significant
levels, albeit lower with t-scores ranging from 3.89 to 6.14. A one standard deviation
increase in the political field dimension of social capital for the lower class was
correlated with a 0.44 standard deviations increase in rally, a 0.34 standard deviations
increase in public interest groups, a 0.38 increase in petition, and a 0.28 standard
Community (14.76,[16.29],0.76)
Field (D2) Friend Contact E42
(V42)
(10.87,[13.20],0.63)
CoWorker Contact
(V43) E43
Rally (V16)
E16
(0.09,[4.62],0.34)
Public Interest
Groups(V27) E27
(0.18,[5.38],0.38)
Petition (V15)
E15
(0.12, [3.89],0.28)
Neighborhood Assoc.
(V5) E5
Figure 5.8. Bourdieu’s Social Capital Model for Lower Social Class, model 1,
(unstandardized b, [t-statistic], standardized Beta)Fit Indexes: SRMSR 0.08, GFI 0.96,
NFI 0.81
150
However, aside from rally, the indicators’ loadings were relatively weak. This
corresponded to low R-squares in that 20 percent of the variation in rally was explained
by the political field. However, only 11 percent of the variation in public interest groups,
Using Bourdieu’s social capital model for the middle class, the religious domain
covaried with the community domain with an estimate of -0.06,66 with a non-significant
t-score of -1.64; the political domain covaried with the religious domain with an estimate
of -0.17 which was significant with a t-score of -4.48; and finally the community domain
covaried with the political domain with an estimate of 0.18 and a t-score of 4.34.
Again, for the middle class using Bourdieu’s social capital model, all indicators
loaded negatively and significantly on the religious field dimension. A one standard
deviation increase in religious field was associated with a 0.86 standard deviations
membership, and a 0.34 standard deviations decrease in religious groups. As with the
lower class from the sample, for those in the middle class the strongest loading was for
volunteer religious and the lowest was for religious groups. In fact, 75 percent of the
variation in volunteer religious was explained by religious social capital. With not as
much explained variance, but still a substantial amount, 38 percent of the variance in
percent of the variance in religious groups was explained by the religious field. It was
somewhat surprising that both the lower and middle classes in the sample had a negative
66
Reported here are the covariances for the latent domains’ standardized estimates.
151
relationship between these three indicators and the religious field. One would think that
the correlation between social capital in the religious field and the religious indicators
All indicators loaded positively and significantly for the community field domain.
The structure of social capital for those in the middle class meant that a one standard
deviation in community social capital was associated with a 0.75 standard deviations
increase in friend contact, a 0.52 standard deviations increase in coworker contact, and a
0.62 standard deviations increase in hanging with friends. For those in the middle class,
56 percent of the variance in friend contact was explained by the community field. While
27 percent of the variance in coworker contact was explained by the community field
As with the lower class, all indicators for the political field dimension loaded
positively and significantly. A one standard deviation increase was associated with a
0.54 standard deviations increase in rally, a 0.52 standard deviations increase in public
interest groups, a 0.37 standard deviations increase in petition, and a 0.31 standard
deviations increase in neighborhood associations for those in the middle class in 2000.
For the middle class, 29 percent of the variation in rally and 27 percent of the
variation in public interest groups was explained by the political field dimension of
Bourdieu’s social capital model. However, only 14 percent of the variation in petition
social capital. The strongest loading for the political field dimension was clearly rally,
152
but all indicators loaded fairly strongly, although less of the variation in the indicators
Community (14.40,[19.09],0.75)
Field (D2) Friend Contact E42
(8.19,[13.05],0.52) (V42)
CoWorker Contact
(V43) E43
Rally (V16)
E16
(0.14,[10.52],0.52)
Public Interest
Groups(V27) E27
(0.18,[7.61],0.37)
Petition (V15)
E15
(0.13, [6.31],0.31)
Neighborhood Assoc.
(V5) E5
Figure 5.9. Bourdieu’s Social Capital Model for Middle Social Class, model 2,
(unstandardized b, [t-statistic], standardized Beta)Fit Indexes: SRMSR 0.05, GFI 0.98,
NFI 0.92
Using Bourdieu’s social capital model for the upper class, the religious domain
covaried with the community domain with an estimate of -0.06,67 with a non-significant
t-score of -1.64; the political domain covaried with the religious domain with an estimate
67
Reported here are the covariances for the latent domains’ standardized estimates.
153
of -0.17 which was significant with a t-score of -4.48; and finally the community domain
covaried with the political domain with an estimate of 0.18 and a t-score of 4.34.
Conversely, for those in the upper class, all indicators loaded positively and
significantly for the religious field dimension. A one standard deviation increase in
religious social capital was associated with a 0.88 standard deviations increase in
0.35 standard deviations increase in religious groups. Based on the findings above for
lower and middle class groups, we can say that the upper class has a positive return
structurally on religious social capital, while the lower and middle classes have a
As with the other class groups, the most variance explained by the religious field
membership was explained by religious social capital, where only 12 percent of the
variance in religious groups was explained by the religious field. While these predictions
were similar in strength to that of the lower and middle classes, the direction of the
For those in the upper class, all indicators loaded positively and significantly for
the community field as well with t-scores ranging from 10.57 to 14.32. A one standard
deviation increase in community social capital was correlated with a 0.76 standard
contact, and a 0.66 standard deviations increase in hanging with friends. Therefore, for
those in the upper class in 2000, we can say that an increase in community social capital
was associated with contacting more friends, coworkers, and hanging out with friends
154
more in a social setting. The strongest loading for this dimension of social capital was
friend contact, although all three loaded fairly strongly. The most variance explained by
the community field was friend contact, with an R-square of 0.57. Community social
capital explained 57 percent of the variance in coworker contact, and 44 percent of the
Finally, all indicators loaded positively and significantly for the political field. A
one standard deviation increase was associated with a 0.72 standard deviations increase
in rally, a 0.78 standard deviations increase in public interest groups, a 0.38 standard
association. Both rally and public interest groups had very strong loadings. It did not
seem as though petition and neighborhood association had weak loadings. However,
only 15 percent of the variation in petition and 9 percent of the variation in neighborhood
association was explained by political social capital; whereas, 52 percent of the variation
in rally, and 61 percent of the variation in public interest groups was explained by the
political field.
155
(0.44,[16.22], 0.88) Volunteer Religious
(V2) E2
Community (14.62,[14.32],0.76)
Field (D2) Friend Contact E42
(9.79,[10.57],0.59) (V42)
CoWorker Contact
(V43) E43
Rally (V16)
E16
(0.30,[15.86],0.78)
Public Interest
Groups(V27) E27
(0.19,[6.45],0.38)
Petition (V15)
E15
(0.13, [5.06],0.31)
Neighborhood Assoc.
(V5) E5
Figure 5.10. Bourdieu’s Social Capital Model for Upper social class, model 3,
(unstandardized b, [t-statistic], standardized Beta)Fit Indexes: SRMSR 0.10, GFI 0.93,
NFI 0.75
As with social class, an unconstrained and constrained model along the racial
groups of White and Non-White were compared using the LR test. For Bourdieu’s social
capital model, the LR test was statistically significant at 0.0002, supporting hypothesis
two stating that there were differences in the structure of social capital along racial
groups. Thus, it made sense to lose 10 degrees of freedom because important information
156
on the differences could be analyzed. The fit of the unconstrained model was overall
good. The chi-square ratio was a bit high at 2.7:1, but the fit indexes were mostly over
0.90, with the exception of the NFI at 0.8855. Further, the RMSEA was under 0.08 at
0.05 and the SRMSR was also good at 0.06. Therefore, the following analysis examines
the structure of Bourdieu’s social capital model across the two models, one for Whites
Using Bourdieu’s social capital model for Whites in the sample, the religious
domain covaried with the community domain with an estimate of 0.004,68 with a non-
significant t-score of 0.101; the political domain covaried with the religious domain with
68
Reported here are the covariances for the latent domains’ standardized estimates.
157
an estimate of -0.13 which was significant with a t-score of -3.39; and finally the
community domain covaried with the political domain with an estimate of 0.18 and a t-
score of 4.43.
Unexpectedly, all indicators loaded negatively and significantly for Whites on the
religious field with t-scores ranging from -8.60 to -16.67. We found that a one standard
deviation increase was associated with a 0.86 standard deviations decrease in volunteer
standard deviations decrease in religious groups. There was in fact a negative return
structurally in their religious social capital. The strongest correlation was for volunteer
religious, and the weakest loading was on religious groups. In fact, only 10 percent of
the variation in religious groups was explained by the religious field. Conversely, 73
percent of the variation in volunteer religious, and 38 percent of the variation in religious
All indicators loaded positively and significantly for Whites on the community
field. For this group, a one standard deviation increase was associated with a 0.75
coworker contact, and a 0.62 standard deviations increase in hanging with friends. The
strongest loading for this group along the community dimension was friend contact,
followed closely by hanging with friends, and a fairly strong showing by coworker
contact. In fact, 56 percent of the variation found in friend contact, 28 percent of the
variation in coworker contact, and 39 percent of the variation in hanging with friends was
158
Finally, all the indicators for political field loaded positively and significantly for
Whites in 2000. A one standard deviation increase was associated with a 0.63 standard
deviations increase in rally, a 0.56 standard deviations increase in public interest groups,
a 0.40 standard deviations increase in petition, and a 0.23 standard deviations increase in
neighborhood association. Thus, for Whites in the sample whose political social capital
increased spent more time on rallies, in public interest groups, signing petitions, and
working on their neighborhoods. The strongest loading for political social capital for this
group was rally. Not surprisingly, the R-square for rally was fairly strong at 0.44.
political social capital. While not as strong, 26 percent of the variation in public interest
159
(-0.41,[-16.67], - Volunteer Religious
0.86) (V2) E2
Community (14.08,[19.55],0.75)
Field (D2) Friend Contact E42
(8.63,[14.76],0.53) (V42)
CoWorker Contact
(V43) E43
Rally (V16)
E16
(0.17,[13.22],0.56)
Public Interest
Groups(V27) E27
(0.20,[9.84],0.40)
Petition (V15)
E15
(0.10, [5.65],0.23)
Neighborhood Assoc.
(V5) E5
Figure 5.11. Bourdieu’s Social Capital Model for Whites, model 1, (unstandardized b, [t-
statistic], standardized Beta)Fit Indexes: SRMSR 0.05, GFI 0.98, NFI 0.93
Using Bourdieu’s social capital model for Non-Whites in the sample, the religious
domain covaried with the community domain with an estimate of 0.004,69 with a non-
significant t-score of 0.101; the political domain covaried with the religious domain with
an estimate of -0.13 which was significant with a t-score of -3.39; and finally the
community domain covaried with the political domain with an estimate of 0.18 and a t-
score of 4.43.
69
Reported here are the covariances for the latent domains’ standardized estimates.
160
However, for Non-Whites in this sample all indicators loaded positively and
significantly on the religious field. This was in stark contrast to Whites from the sample.
For Non-Whites, a one standard deviation increase in religious social capital was
associated with a 0.86 standard deviations increase in volunteer religious, a 0.59 standard
religious groups. The strongest loading for Non-Whites on the religious field was for
volunteer religious was explained by religious social capital. While the other loadings
were not what one would deem weak, they were not as predictive as volunteer religious.
variation in religious groups was explained by religious social capital. The main point of
interest for this particular dimension was that all indicators for Non-Whites were positive,
community field, t-scores ranged from 9.54 to 14.81. For Non-Whites a one standard
deviation increase in community social capital was associated with a 0.77 standard
contact, and a 0.67 standard deviations increase in hanging with friends. Therefore, we
can say that Non-Whites in the sample whose community social capital increased led to
contacting more friends and coworkers, and spending more time with friends socially.
All three loadings were fairly strong, the strongest being friend contact. In fact, 58
percent of the variation in friend contact was explained by the community field for Non-
161
community social capital. Lastly, 45 percent of the variation found in hanging with
Finally, all indicators loaded both positively and significantly for Non-Whites on
the political field. A one standard deviation increase in political social capital was
correlated with a 0.66 standard deviations increase in rally, a 0.51 standard deviations
increase in public interest groups, a 0.41 standard deviations increase in petition, and a
loading for this group, as with Whites, was rally. In terms of predicting, the weakest
predicted indicator was petition, where only 16 percent of the variation was explained by
political social capital. The variation in public interest groups and neighborhood
associations was explained by the political field at 0.26 and 0.22 respectively. However,
the most variation, 44 percent, was explained by political social capital for rally.
162
(0.42,[14.44], 0.86) Volunteer Religious
(V2) E2
Community (14.92,[14.81],0.77)
Field (D2) Friend Contact E42
(8.59,[9.54],0.53) (V42)
CoWorker Contact
(V43) E43
Rally (V16)
E16
(0.15,[8.42],0.51)
Public Interest
Groups(V27) E27
(0.20,[6.55],0.41)
Petition (V15)
E15
(0.21, [7.83],0.47)
Neighborhood Assoc.
(V5) E5
Figure 5.12. Bourdieu’s Social Capital Model for Non-Whites, model 2, (unstandardized
b, [t-statistic], standardized Beta)Fit Indexes: SRMSR 0.09, GFI 0.95, NFI 0.79
There were significant, but few differences in the structure of social capital
between Whites and Non-Whites using Bourdieu’s social capital model. For both the
community and political fields, there were slight differences in the structure of social
capital for each group. For example, Whites (14.08) had a slightly weaker loading for
contact with friends when compared with Non-Whites (14.92) and a slightly weaker
loading (11.06) for time spent socially with friends than Non-Whites (12.51). However,
Whites (8.63) had a slightly stronger loading for contact with coworkers than Non-
163
Whites (8.59). Within the field of politics Whites and Non-Whites were essentially the
same on rally (0.24, 0.26), public interest groups (0.17, 0.15), and petition (0.20, 0.20).
However, there was a slightly stronger loading for neighborhood association for Non-
Whites (0.21) when compared to Whites (0.10). Therefore this indicator tells us more
The biggest difference in the structure of social capital for Bourdieu’s social
capital model between Whites and Non-Whites was in the religious field. Here Whites
had a negative relationship between the religious field and the three indicators, volunteer
religious, religious membership, and religious groups. Therefore we can say that for
Whites in this sample, when compared with Non-Whites in 2000, there was a negative
return structurally on religious social capital. Worded another way, there was an inverse
relationship for Whites between the religious domain and volunteering in religion, being
a member of a religion, and being a part of religious groups. This finding alone was
unexpected and interesting, but made even more so because the exact opposite was true
for Non-Whites. The latter’s relationship along the religious field indicators was positive
and significant, in nearly the same strength, but with a differing structure.
Using Bourdieu’s social capital model, two models were run along gender lines
for women and men one unconstrained and one constrained. The LR test was statistically
significant at 0.0008, supporting hypothesis three stating that there were differences in
the structure of social capital along gender groups. Thus, it made sense to lose 10
164
The fit of the unconstrained model was overall very good. The chi-square ratio was
below 2 at 1.6:1, and the fit indexes scored over 0.90, without exception. Further, the
RMSEA was clearly under 0.08 at 0.02 and the SRMSR was also good at 0.03.
Therefore, the following analysis examines the structural differences between men and
Using Bourdieu’s social capital model for women in the sample, the religious
domain covaried with the community domain with an estimate of 0.11,70 with a
significant t-score of 3.32; the political domain covaried with the religious domain with
an estimate of 0.36 which was significant with a t-score of 10.39; and finally the
70
Reported here are the covariances for the latent domains’ standardized estimates.
165
community domain covaried with the political domain with an estimate of 0.20 and a t-
score of 5.42.
For women, all indicators loaded positively and significantly on the religious
field. A one standard deviation increase in religious social capital was associated with a
groups. The most variance explained by the religious field was volunteer religious, 77
religious social capital, where only 12 percent of the variance in religious groups was
All indicators loaded positively and significantly for the community field as well
with t-scores ranging from 14.10 to 20.69. A one standard deviation increase in
community social capital was associated with a 0.75 standard deviations increase in
friend contact, a 0.51 standard deviations increase in coworker contact, and a 0.67
standard deviations increase in hanging with friends. The strongest loading for this
dimension of social capital was friend contact, although all three loaded fairly strongly.
The most variance explained by the community field was friend contact, with an R-
Finally, all indicators loaded positively and significantly for the political field for
women. A one standard deviation increase was associated with a 0.62 standard
deviations increase in rally, a 0.51 standard deviations increase in public interest groups,
a 0.41 standard deviations increase in petition, and a 0.31 standard deviations increase in
166
neighborhood association. Rally had the strongest loading, followed closely by public
interest groups. The R-squares for how well each was predicted by the political field
followed the same pattern as the standardized Beta loadings. Thirty-eight percent of the
variation in rally was explained by one’s political social capital. While, 26 percent of the
variation in public interest groups was explained by the political field. However, only 17
Community (14.19,[20.69],0.75)
Field (D2) Friend Contact E42
(7.99,[14.10],0.51) (V42)
CoWorker Contact
(V43) E43
Rally (V16)
E16
(0.01,[12.25],0.51)
Public Interest
Groups(V27) E27
(0.20,[10.13],0.41)
Petition (V15)
E15
(0.13, [7.57],0.31)
Neighborhood Assoc.
(V5) E5
Figure 5.13. Bourdieu’s Social Capital Model for Women, model 1, (unstandardized b, [t-
statistic], standardized Beta)Fit Indexes: SRMSR 0.03,GFI 0.99, NFI 0.95
167
Using Bourdieu’s social capital model for men in the sample, the religious
domain covaried with the community domain with an estimate of 0.11,71 with a
significant t-score of 3.32; the political domain covaried with the religious domain with
an estimate of 0.36 which was significant with a t-score of 10.39; and finally the
community domain covaried with the political domain with an estimate of 0.20 and a t-
score of 5.42.
For men in the sample, all indicators loaded significantly and positively on the
religious field, t-scores ranged from 8.05 to 19.26. A one standard deviation increase in
religious social capital was associated with a 0.86 standard deviations increase in
0.33 standard deviations increase in religious groups. The most variance explained by
the religious field, as with women, was volunteer religious, 75 percent. Forty percent of
the variance in religious membership was explained by religious social capital, where
only 11 percent of the variance in religious groups was explained by the religious field.
All indicators loaded positively and significantly for the community field as well
with t-scores ranged from 15.70 to 19.98. A one standard deviation increase in
community social capital was correlated with a 0.76 standard deviations increase in
friend contact, a 0.61 standard deviations increase in coworker contact, and a 0.62
standard deviations increase in hanging with friends. The strongest loading for this
dimension of social capital was friend contact, although all three loaded fairly strongly.
The most variance explained by the community field was friend contact, with an R-
coworker contact, and 38 percent of the variance in hanging with friends. Slightly less
71
Reported here are the covariances for the latent domains’ standardized estimates.
168
was explained by community social capital for men than women on hanging with friends,
however, slightly more was explained by community social capital for men when
Community (14.93,[19.98],0.76)
Field (D2) Friend Contact E42
(10.48,[15.90],0.61) (V42)
CoWorker Contact
(V43) E43
Rally (V16)
E16
(0.19,[14.55],0.61)
Public Interest
Groups(V27) E27
(0.20,[9.38],0.42)
Petition (V15)
E15
(0.15, [7.76],0.35)
Neighborhood Assoc.
(V5) E5
Figure 5.14. Bourdieu’s Social Capital Model for Men, model 2, (unstandardized b, [t-
statistic], standardized Beta) Fit Indexes: SRMSR 0.04, GFI 0.98, NFI 0.94
Finally, all indicators loaded positively and significantly for the political field for
men. A one standard deviation increase was associated with a 0.61 standard deviations
increase in rally, a 0.61 standard deviations increase in public interest groups, a 0.42
169
neighborhood association. Therefore an increase in political social capital was
associated with men in the sample attending more rallies, public interest groups, signing
more petitions, and participating more in their neighborhood. Rally had the strongest
loading, followed closely by public interest groups. Both rally and public interest
groups loaded at the same strength (0.61). The R-squares for how well each was
predicted by the political field showed that 37 percent of the variation in rally was
explained by one’s political social capital. While, 38 percent of the variation in public
interest groups was explained by the political field. However, only 17 percent of the
For indicators of the religious field there were very few differences between men
and women in the structure of social capital for that context. There was a slightly
stronger loading for women (0.43) on volunteer religious than for men (0.40) which
correlated to a 2 percent increase for women more than men in the percent of variation
explained by religious social capital for this indicator. The largest difference was for
religious membership, women (0.27) had slightly weaker loading than men (0.31). Using
the R-square for this sample we can say that 5 percent more of the variation in religious
membership was explained by the religious field for men when compared to women.
Both men and women were essentially the same on the religious groups indicator.
When looking at community social capital in the sample for men and women
there were some important differences. First, men (14.93) and women (14.19), loaded
very similarly along friend contact. This was supported by the R-square which showed a
slight increase in explained variation for men (2%) more than women. However, men
170
(10.48) had a higher loading on coworker contact versus women (7.99). Further, the R-
square comparison showed that 10 percent more of the variation in coworker contact was
explained by community social capital for men than women. Conversely, women (12.37)
loaded stronger for spending more time hanging out socially with their friends than men
Finally, for the political field, men and women were strikingly similar, with one
exception: men (0.19) had a significantly stronger loading for activity in public interest
groups than women (0.01). The R-square showed that 12 percent more of the variation in
the indicator public interest groups was explained by the political field for men than for
women.
171
CHAPTER VI
across class, race, and gender in regard to the structure of social capital for each model.
These empirical evaluations were conducted using the national sample from the SCCBS.
Each theoretical approach was modeled using latent domains and manifest indicators.
The benefit of using these methods was that the abstract nature of social capital could be
differences across class, race, and gender with regard to social capital (Aguilera 2002;
Parks-Yancy et al 2008), none have modeled all three theorists separately, and used CFA
and multi-sample SEM to evaluate the clear theoretical and latent aspects associated with
Using CFA, evidence was found to support (see Chapter 4) three separate models
of social capital for Putnam, Coleman and Bourdieu. For each theorist, indicators for
each domain and model loaded positively and significantly with t-tests much higher than
2.0. While each model had a significant chi-square, Coleman and Bourdieu had better
chi-square ratios and each model was found through fit indexes to fit. Thus, support was
found for Research Question one, that each theorists’ approach to social capital could be
modeled empirically.
172
The second Research Question is: Does the structure of social capital operate
differently along class, race and gender lines for Putnam’s, Coleman’s, and Bourdieu’s
empirically supported social capital models? The following hypotheses were used to test
and answer this question for each social capital model created in Chapter 4, first, that
there would be differences in the structure of social capital across class groups; second,
that there would be differences found in the structure of social capital across race groups;
and third, that differences in the structure of social capital would be found across gender
groups. However, when each model was run across socio-demographic categories using
multi-sample SEM, the LR test was not always significant across. As a result, some
hypotheses were supported and some were not depending on the model. Specifically,
significant differences were found for class and race, but not gender for Putnam’s social
capital model. However, for Coleman’s model the only socio-demographic group
showing any structural differences in social capital was race. Finally, all socio-
The following chapter discusses these results, their implications, and the
limitations of each model by theorist. After which I discuss the contributions of the
current study, followed by the limitations across all models, and ending with suggestions
Putnam’s social capital model included domains for bonding social capital,
bridging social capital and trust. According to Putnam (2000), trust is important in that it
173
underlies the norm of reciprocity and at the community level allows for people to give
others “the benefit of the doubt” (Putnam 2000:136). However, Putnam (2000) does not
conceive of trust as a separate concept per se, but an indicator for social capital. It was
therefore interesting that trust did not load in the EFA on either bonding or bridging
social capital domains, but emerged instead as its own latent domain. Further, while
Putnam (2000) spoke of trust in terms of “thick” or “thin”, only one domain emerged for
trust, not two. This finding is based on the social trust index which had a general and
then more specific contextual instances of trust included in the measure. Beyond that,
two other measures of more specific trust were used as indicators on the trust domain,
perception of community trust and rate your community. While the fact that trust
emerged as its own domain does not fully answer the question of where trust fits into the
picture of Putnam’s social capital fully, it does add to our understanding. This finding
supports conceptualizing and measuring trust with social capital, but as a separate
domain, not as part of bonding or bridging social capital. It also suggests that while the
two are connected, (trust and social capital), they are not the same. Thus this finding
Using Putnam’s model for social capital, there were class differences in the
structure of social capital showing support for the first hypothesis under Research
Question two, that there will be class differences in the structure of social capital. The
major difference in the structure of social capital was in the middle class results, where
not only was the strength of the social capital indicators diminished, but the tests of
significance were as well. The most surprising difference for the middle class was that
the family contact indicator on the bonding domain was not significant. Putnam
174
(2000:277) cites the family as a “key form of social capital” and links the loss of the
Further he remarks upon how families spend less time together vacationing, watching TV
together, going to religious services as a family, and spending less time together relaxing.
Here for the middle class, it was their friends and the people they shared confidences with
who were important for the bonding social capital domain. This finding could be an
indicator of the mobility of the middle class in 2000 (Putnam 2000), it is possible that
because of job relocation or schooling that those in the middle class were not close
enough to have in person contact with family members. Increasingly the family needs
two incomes to survive and thrive economically and this could also be influencing the
time that families have to make contact or visit their family. This finding for the middle
class suggests that bonding and bridging social capital do not operate in the same way
capital on racial categories, supporting the second hypothesis. The results showed that
Non-Whites in the sample had higher levels of social capital on the bonding, bridging,
and trust domains when compared with Whites, with one important difference. For
Whites family contact loaded significantly on the bonding domain, however, it was not
significant at all for Non-Whites. Therefore, for Non-Whites in the sample, family
contact was not important to the structure of the bonding social capital domain. For this
group, the most important structural aspects include their friends and the people with
whom they share confidences. This finding supports to some extent the findings of Stack
175
It was expected that there would be significant differences in the structure of
social capital for gender groups (see Hypothesis three, Chapter two). However, for
Putnam’s model, the LR test was not significant, showing no support for this hypothesis.
This is counter to findings in the literature showing differences in social capital by gender
(Demant and Jarvinen 2011; DiTomaso et al. 2007; Noris and Inglehart 2006; Parks-
social capital in their study and found gendered effects. They looked at membership and
and volunteerism among groups, it was found that there were gender differences in most,
but not all groups, showing some “sex segregation.” The only types of groups that did
not show significant differences or a gender gap were education, arts, music, or cultural
activities. There was a small, but significant gender gap when looking at membership
and trust, however, the gender gap for membership was smaller in postindustrial
societies.
Putnam (2000), however, does not conceive of a gendered social capital, although
he does test social capital along socio-demographic lines. In fact, for his health and
connections with social capital persisted even when the studies examined
other factors that might influence mortality, such as social class, race,
gender, smoking, and drinking, obesity, lack of exercise, and
(significantly) health problems…what these studies tell us is that social
engagement actually ahs an independent influence on how long we live
(Putnam 2000:331).
It is not that he thinks gender is unimportant, nor that there could not be differences along
gender lines (or class or race, etc.) for social capital, but instead, he has yet to include
176
class, race, and gender in his theory of social capital. The model in this study focused on
his theoretical writings for social capital and attempted to model specifically his bonding,
bridging and trust domains. It is possible that part of the reason that there are not any
significant differences in the structure of social capital for gender is in the structure of the
model.
Does this mean that there are no gender differences in the structure of social
capital? Absolutely not, in fact gender differences were found when using Bourdieu’s
social capital model (see below for discussion). However, what it does mean is that for
this sample there are no structural differences in social capital between men and women
While trust emerged as its own domain for Putnam, it did not do so as “thin” and
“thick” trust. That only one latent domain emerged may be because the measures
themselves were too broad and encapsulated both “thick” and “thin” trust within them.
Another possibility is that it is not so much what kind of trust one holds, but that
generalized trust is held by the group and therefore there is only one latent domain related
to trust. A study using data with measures that can fully capture the conceptualization of
trust as “thin” and “thick” should test to see if these load on one domain or two. Due to
concepts. In part, the difficulty in measuring Putnam may be because this study went off
of his theory on bonding and bridging social capital and attempted to model them as
177
distinct domains. While Putnam theorizes about these aspects of social capital, he does
not see them as completely distinct and thus did not attempt to measure them as separate
concepts (Putnam 2000). Unfortunately, the one important distinction, exclusive versus
inclusive groups,72 was not used because of issues with reliability of the scales. Knudsen
et al (2007) found through EFA that they could have two factors on the group
However, when this study attempted to use their scales for exclusive groups versus
inclusive groups, the Alpha of reliability for each was around 0.30, meaning that 70
percent of the variance was measurement error. And further, when the inclusive and
exclusive group scales were included with the trust domain, both scales loaded on the
same latent domain. Thus, it was concluded that their findings were lost when other
aspects of social capital were introduced and their measures were deemed unreliable and
For the current study, the best third variable for bonding social capital domain
was family contact. However, this variable loaded weakly and is considered a limitation
for the Putnam model. In future studies it is recommended that a different measure or
measures be used in its place. In particular, time spent having a family meal, vacation
time with family, doing activities with family, and going to family for support (both
emotional and instrumental (Lin 1999)) would be useful to measure family in terms of
social capital.
It would make more sense, if this is the crux of Putnam’s social capital model, to
distinguish between the two sufficiently that one can measure each. It stands to reason
72
Putnam (2000) uses the idea of inclusive groups being reflective of bonding social capital and exclusive
(hard to attain membership or tied to a particular identity) group membership as a way to identify
bridging social capital.
178
that one could lead to another, and that they are intimately connected, but for this
capital both theoretically and empirically, reliable and valid measures need to be included
that are clearly one versus the other. For example, creating measures that include
whether the tie is exclusive or inclusive in nature, (e.g., anyone can join this group versus
only certain people can join). A deeper examination would include measuring on which
aspect the exclusion is based, such as race, gender, age, region, or income.
It can be argued that there was in truth very little difference between class groups
using Putnam’s model, and therefore, the original results from chapter four should stand
in its stead, regardless of the LR test being significant. However, the finding that for the
middle class family was not a significant indicator for bonding social capital, though
small, shows a potentially important difference for the middle class and challenges a
basic assumption that the family is a universal wellspring of social capital. The results
presented here, however, are not conclusive. Further testing using various methods and
other samples to confirm and support the results found in this model are needed. Also,
the sample size for the middle class group was lower at 263 than either that of the lower
or upper social class groups (see Table 5.1), therefore, this finding was potentially
associated with a loss of statistical power due to a smaller sample size (Allison 1999;
Bollen 1989). In any case, it is interesting that for the middle class in this sample their
main relationship focus was on non-family members. Future research should test for this
179
relationship instead of assuming the importance of family on this dimension of social
For race, it was found that for Non-Whites, bonding and bridging social capital as
well as trust were more important structurally. The only aspect for which this did not
hold was for family contact. This indicator was more important to Whites, it would be
interesting in future research to look at this finding at the intersection of race and class to
see if there are further distinctions in the structure of the bonding social capital domain
and specifically family contact. Putnam (2000) argues that the bonds of the family are
“loosening.” But whether that is the case or the way in which family is bonding has
changed, one cannot determine accurately with the current data. Future studies may
want to explore how families stay linked, communicate, and bond. For example,
technology plays an important role in the way in which we stay connected, Facebook,
LinkedIn, and Skype are examples. The latter could be a part of why certain groups are
not visiting family as often. Simply because the structure of the family no longer reflects
a 1950s era does not mean that its role in social capital has diminished. These other ways
whether the family is truly “loosening” or simply shifting and adapting to new
As will be discussed below, there is only one model in the study that showed
significant differences in the structure of social capital for gender. Therefore, it is not
that there are no differences in the structure of social capital by gender, but that these
differences are contextual. This argument is supported by the fact that the one model that
did show structural differences by gender was Bourdieu’s social capital model, which
180
included the use of the field concept. By definition field is tied to context. Parks-Yancy
et al. (2008) for example found that men and women had different kinds of ties and
assumed then that they also had different resources. Men in general had more work ties,
whereas women tended to have more ties to family and personal friends. Putnam (2000)
also noted that women and men had different ties, but that when women started working
The fact that no significant differences were found for gender groups means that
for this socio-demographic group bonding, bridging, and trust domains were neither more
important for men than women and vice versa. This does support the argument that
researchers examining gender as an important variable along with social capital should
consider using a different model and theoretical framework. Based on the findings,
Bourdieu’s model would be best because it was the one model that showed important
(E/O), information channels (IC), and social norms (SN) (Coleman 1988). An argument
was made to include different trust indicators for the SN domain, the same ones in fact
that were used to measure trust for Putnam’s model. This makes theoretical and
empirical sense when one considers that Putnam (2000) also talks about social capital in
terms of trust and social norms. The Coleman model fit well and was subjected to multi-
sample SEM analysis for class, racial, and gender groups (see Chapters 4 and 5).
181
Interestingly, only one significant LR test was found for Coleman’s model
supporting the second hypothesis that there were differences in the structure of social
capital for race. What this means is that Coleman’s model captures social capital
differently for Whites and Non-Whites. Therefore, his model allows us to examine
which aspects of social capital were more important for Non-Whites and those that were
relationships and group involvement tells us about E/O social capital more so for Non-
Whites than for Whites. This means that having diverse relationships and being involved
in groups was more important for Non-Whites’ social capital than for Whites’. However,
other church participation was more important for Whites than for Non-Whites. Both
shared similar structure on the club meetings indicator. In sum, fictive kin was more
important for Non-Whites than Whites on the IC domain, but both groups were similar in
terms of knowing someone who owns their own business. The latter is interesting
because Aguilera (2002) found important differences on race specifically for black men
and white women. It may be that if the interaction between race and class were examined
for Coleman those differences would emerge. Finally, having a friend who owns their
own vacation home was more important for Whites than Non-Whites. These findings
support the argument that it matters who you know and your connections can help
structurally with one’s social capital; in this case, one’s social capital along the IC
domain.
community trust than Whites and had stronger loadings on the trust domain than Whites.
182
Thus, the structure of one’s E/O varied by race, and specifically, perception of trust in
one’s community was more important for Non-Whites than Whites in the sample. Non-
Whites had stronger loadings on social capital on the IC domain than Whites and thus
structurally were advantaged in terms of that domain. The same was true for the SN
As evidenced by the findings for Coleman’s model, hypotheses one and three did
not find support, however, evidence was found to support differences in the structure of
social capital by racial groups (hypothesis two). These findings support those of Aguilera
(2002) who found that race and class both mattered for Whites and Non-Whites when
considering social capital. But in particular that Non-Whites and Whites were able to
model of social capital because his empirical analysis takes into account material
resources. However, his analysis looked at the transformation of social capital into
economic or human capital and the subsequent consequences therein, e.g. parent
current analysis does not have indicators to measure the transference from one capital to
another and that is why there were not differences. For example, Coleman (1988)
examined the social capital that children drew from their parents; part of the resources
that their parents provided to their children was in the form of their human capital (e.g.,
183
education). However, it was not possible to use education as a measure for class because
the sample size dipped below 200 in the multi-sample SEM model.
Clearly human capital is omitted from the analysis. One could argue that the class
analysis at the very least does bring in income as a material resource, and thus it is
particularly surprising result. It is also possible that the measures were so broad (e.g., the
measures for income did not include interval level data, but categories) and thus the
differences were masked. There was no problem with the sample size for his model, and
it fit very well in the CFA analysis, moreover, when one looked at the overall fit indexes
One serious limitation to Coleman’s social capital model was the lack of family
measure(s) in the dataset. The one available to include, family contact, did not load
significantly on any of the three domains in the EFA73 and when included in the CFA to
test, the fit was worse than without it. Theoretically, it makes sense to include a measure
of family within Coleman’s model, especially since he views the family as the nexus of
social capital. However, the measures available did not approximate his measures used,
time spent with children doing homework as one example (1988), and the closest
approximation to family that fit with the model was fictive kin.
73
When family contact was included in the analysis it loaded at 0.30 on a fourth domain. Unfortunately
this domain did not have any indicators load significantly (0.40 or above) and combined with the fact
that the scree plot did not show support, it was not included in the analysis. Moreover, there were no
other useful family measures to include to examine if their inclusion would change the results and
potentially allow us to see how family as social capital would work or not in the model. The CFA chi-
square ratio was 3.5:1 and other indices were below 0.90. The better fit was without the indicator
included, but it is only because of its lack of holding with the other indicators on the EFA combined
with its influence on the model fit in the CFA that this indicator was not included in the Coleman model
for social capital.
184
Implications of Results for Future Research: Coleman
Neither class, nor gender had significant LR tests and those findings were thus
excluded from the analysis. Instead the original findings in chapter four stand for both
class groups and gender groups for the structure of social capital using Coleman’s model.
Coleman’s conceptualization of social capital is one of the most common in the social
sciences and while these findings are not conclusive, they do suggest that Coleman’s
model of social capital does not tell us more about class and gender groups. It is
therefore important for future research concentrating on social capital and class that
Coleman’s model potentially does not capture differences on social capital for income
groups. The same holds true for studies where social capital and gender are the main
focus. However, these findings do suggest that Coleman’s model of social capital does
provide important information on the structural differences of social capital for racial
groups. Future studies concentrating on social capital and race may find his
Bourdieu’s social capital model was based not only on his concept of social
capital within his larger theory of inequality, but also on his concept of field. This is
where using a dataset that examines social capital specifically was very useful. The three
fields that emerged from the analysis discussed Chapter 4 were religious, community, and
political. Throughout the dissertation field and domain were used interchangeably.
The LR test was significant for each socio-demographic group, supporting the
three hypotheses that there were significant differences in the structure of social capital
185
across class, race, and gender groups. Bourdieu’s social capital model was the one social
capital model upon which the socio-demographic group’s differences were found to be
significant for all three socio-demographic groups: class, race, and gender. This meant
that there were significant differences on the structure of social capital for each group. It
also suggests that his model provides us with more information about the structural
For the lower and middle classes, the indicators for the religious domain loaded
negatively and significantly. However, for the upper class, the predictions were similar
in strength to the lower and middle classes, but were positive in direction. The middle
class in particular had a stronger negative loading on all three indicators (volunteer
religious, religious membership, and religious groups) than the lower class.74 It is
interesting that for the lower and middle classes there would be a negative association for
religious indicators on the religious field. Is there in fact a negative return structurally on
social capital in the religious field for those who are in the lower and middle classes? Is
this because the people who gain the most from the religious field are those who have
more material resources? The latter is possible considering that recent research has
shown (Greenfield 2013) that those who make more money actually give proportionately
less to charity than those who make less (when one looks at the percent of income gifted).
It is also possible that those at the upper income levels give more dollars and thus gain
social capital from that aspect, having to participate less with more return on their social
capital. These findings show support for including context and specifically the use of
74
This finding was unexpected, therefore, the models were checked, rechecked (in terms of indicator
coding and model coding), and then subjected to multi-sample SEM on class again to confirm the
results.
186
The middle class had the weakest loading for social capital structurally on the
community field versus both upper and lower classes. Whereas the lower class had the
strongest loadings for social capital structurally on the community field with regard to
going out with friends, coworkers, or contact with friends. It would be interesting to look
at these results based on hours worked or having a job at all and break it down further to
see if that had something to do with the strength of loadings for this type of social capital
and the lower class, perhaps it was a simple issue of time. It is possible that those in the
middle class spend so much time at work that hanging out after work with friends or
otherwise was difficult to fit in (Aguilera 2002; Putnam 2000). It is also possible that this
social capital and specifically family contact. Americans like to move upward, promoted
by credit card companies and our consumer culture, and thus many people are living
outside of their means. Further that with a need to keep up with others in one’s group
(Field 2008), and a competitive nature for those who have made great gains in their
professional careers, or are able to have professional careers as opposed to jobs and it
makes some sense that the middle class would have potentially less social capital
middle class where family contact is not significant at all to the bonding domain has any
bearing on the finding that the middle class has weaker social capital structurally on the
community domain.
There were significant, but very few differences in the structure of social capital
between Whites and Non-Whites on Bourdieu’s social capital model. Whites had slightly
weaker loadings for contact with friends when compared with Non-Whites and slightly
187
weaker loading on time spent socially with friends than Non-Whites. What this means is
that using Bourdieu’s social capital model captured structural differences for Whites and
Non-Whites in social capital. Here for example, contact with friends and spending time
socially with friends was more important for Non-Whites than Whites in the sample.
However, contact with coworkers was more important for Whites than Non-Whites. For
the political field Whites and Non-Whites were essentially the same, however, there was
meaning that neighborhood association tells us more about Non-Whites than Whites in
the sample.
The most striking difference in the structure of social capital between Whites and
Non-Whites for Bourdieu’s model was on the domain of religion. Here again the study
found negative relationships on all indicators for the religious domain, but only for
Whites in the sample. Conversely, Non-Whites in the sample had positive and significant
relationships for all indicators on the religious domain. The strength of the relationship
was similar, but the direction was opposite, suggesting a negative relationship for Whites
on social capital within the religious domain. Possible reasons for this finding may
include that there were more Whites in the sample who were either lower or middle class
and there may have been an embedded interaction effect that was not tested in this study.
It is also possible that Whites’ experience in the religious field based on these indicators
leads to weaker returns in their social capital because of another status that is not
accounted for in this study. A concept not used explicitly in this study, but important to
Bourdieu, was that of social space and the importance of status or one’s position. While
racial status was accounted for, perhaps there were others that interact with race and may
188
have some bearing on this finding. It may be a conflation of the type of religious
institution(s) one belongs to, Catholic versus Protestant, or even the density of the
membership therein that is affecting these results. In any case, it is very interesting that
Non-Whites, for this model and sample, were able to garner stronger loadings in this field
than Whites. These findings support that overall social capital was more important
Interestingly, the only model for which differences between men and women were
significant was for Bourdieu’s social capital model. There were very few differences
between men and women in the structure of social capital for the religious domain. For
women, volunteer religious, was more important than for men. The largest difference
was for religious membership which told us more about men than women. For men
being a member of a religion was more important to their social capital. Is it possible that
the reason behind this finding is that the power structure of most mainstream religions
favors men over women (Shaw and Lee 2012: 621)? Putnam found that married people
and especially those with children were more likely to be involved in religion and
religious organizations, but very little was said about whether men gained more for the
When looking at the community domain, the indicator of friend contact was
similar for both men and women. However, there were some important differences on
this domain for gender groups. For example, having contact with coworkers was more
important for men than for women in the sample. This is not surprising as most research
shows that men tend to go out more socially with their work contacts (Norris and
Inglehart 2006). Also, the results showed that for women time spent hanging with their
189
friends was more important than for men. This finding was somewhat different from
Norris and Inglehart’s (2006) findings that men overall spent more time outside of the
home than women. They attributed the difference to the constraints on most women from
traditional gender roles and the work of home and family. Norris and Inglehart (2006)
used global data from multiple societies and thus this may be a difference in nationality
and gender roles, showing that America (2000) was slightly less traditional in terms of
One indicator on the community domain for Bourdieu, coworker contact, had
missing that was more than 30 percent. When the missing was compared against
dummied variables for class, race, and gender, all three chi-squares were significant
showing that there was some selection bias for the missing. The current study used
listwise deletion, but this is a limitation because it introduced bias into the model.
underdeveloped was a limitation. However, while a limitation, the fact that Bourdieu was
could be used which produced a model that allowed for the examination of differences in
Future studies should test the negative relationships found when using Bourdieu’s
social capital model. In particular, it was odd that the religious field had negative returns,
190
not to mention the negative returns to Whites. What these findings suggest is that social
capital had a negative return, or it took away from, these groups versus others. Further
research needs to be conducted to corroborate these findings before any real conclusions
can be made.
social capital combined with field provides a fertile framework for understanding and
considering that his theory focuses more on cultural capital. However, it is clear using
the current study as a point of reference, that his conceptualization of social capital when
used with field reveals important differences in the structure of social capital for all three
socio-demographic groups examined here. While his writings did not center on social
capital, he did value both agency and structure and above all the concept of context in
Contributions
There are three important contributions that this dissertation provides. First, the
findings help with the greater question of what is social capital. Context has been shown
to be important (Knudsen et al. 2008; Lin 2001; Nieminen et al. 2008; Parks-Yancy et al.
2008), this study illustrates how the structure of social capital differs across class, race,
and gender groups. Moreover, the findings reported here show that the differences are in
fact influenced by the way in which one conceives and models social capital. For
example, neither Putnam’s model nor Coleman’s model showed significant differences
for gender groups, however, Bourdieu’s model did. Therefore, if one wishes to
191
understand structural differences along gender groups in social capital then one should
utilize Bourdieu’s conception of social capital combined with field. Bourdieu’s model
performed best in representing a framework that allowed for the various structural
differences to be seen for gender groups. That is not to say that Putnam’s and Coleman’s
models were not useful, but the models were not consistent. For example, Bourdieu’s
model was the only model that found differences for all three socio-demographic groups.
In sum, the findings reported also support the assertion made early on in the dissertation
that social capital operates differently across socio-demographic groups and further
Second, this project takes a foundational approach to social capital. It does not
assume that the definitions being used for social capital are the same and instead begins
with conceptualizing social capital using established theoretical threads within the study
used to create three separate empirical models, where rigorous analysis and testing were
social capital using confirmatory factor analysis and multi-sample structural equation
modeling guided by specific theorists. Applying theorists individually to frame the data
revealed the nuanced differences between the three theorists which are often ignored in
the literature.
192
Third, and finally, this dissertation provides multi-dimensional models based on
theory for use potentially across samples. For example, the SCCBS collected data from
various years and thus analysis can be extended to include those years as well as change
models examining differences from one year to the next. How far reaching these models
can be used has yet to be determined, however, these three models potentially can be
One limitation of the current study was that some of the variables included for
analysis were dichotomous. However, it was deemed more important to have three
indicators to load per domain (Hatcher 1994; Bollen 1989); for the current study the
Further, secondary data was used in the analysis. The SCCBS was uniquely
suited to the task, but cross-sectional survey data by definition only allows a small
snapshot in time (Allison 1999). Quantitative analysis also does not allow for as much
explanatory analysis as does qualitative data. The benefit to using survey data and the
SCCBS in particular is that the sample size is large (3003) and when dealing with a study
that had skip patterns in it, a loss in some data did not adversely affect the sample size.
Also, the fact that this study was created for the sole purpose of measuring social capital
lent itself to the goals of this study. It would also have been both interesting and useful to
further the understanding of social capital and Putnam’s, Coleman’s, and Bourdieu’s
conceptualization of it; to use both quantitative and qualitative data to examine the
193
It also must be mentioned that some bias in terms of the ability to measure the
three approaches of social capital was introduced by the fact that Robert Putnam was on
the committee that created the survey instrument and measures therein. While the
empirical evidence did not support a bias, Coleman and Bourdieu’s models fit better than
prestige to the measure of social class. The best available measure for social class in the
current dataset was income. There was a measure for education, but none for
occupational prestige. Unfortunately, when education was added to create the class sub-
samples in the multi-sample SEM, the sample size dipped below 200 for each model.
The sample size for SEM needs to have a minimum of 200 observations to be considered
(Hatcher 1994). Thus, income was used as a measure for social class in the current study,
but it is recommended that both measures for education and occupational prestige are
As shown in the limitations, survey research has its drawbacks, as does using
SEM. Future studies may wish to use only interval-ratio data when modeling social
capital in CFA and SEM in general. Whenever possible, scales, indexes, continuous,
and/or likert scale style questions were used for the current study. However, in order to
overcome the various drawbacks to survey data in general, a future study on social capital
may wish to use both qualitative and quantitative data. Specifically I think it would be of
use to collect in-depth interviews from all the subgroups examined in the current study.
194
These interviews could ask questions in person with follow-up probes that seek to find
the answer to why social capital matters more to one groups than another. Included in the
interview should be questions or prompts attempting to understand the role that context
and need play with regard to the importance of social capital to one group over another.
For example, in this way we could find the reasons behind the finding that social capital
was more important for Non-Whites in the sample than for Whites. In order to do a true
comparison both Non-Whites and Whites would need to be included in the qualitative
data. While focus groups might give us a larger picture, it would not be as in depth and I
would argue that a more in depth analysis of why religious membership is more
important to men than women is needed as opposed to the broader picture which was
Finally, it would behoove any qualitative research following this line of inquiry to
ask questions specifically seeking further elucidation on the fundamental question, what
is social capital. Each interview could ask a series of questions that could be tailored for
the respondent’s identity and position in social space. Some examples of questions that
might help us to understand social capital at a deeper level would be: When your child is
sick or has a day off from school how do you manage that with a full-time job? Who do
you go to for advice in your job [life] [family] [relationships]? What would you consider
a social resource? What does it mean to feel safe in your community? Where do you go
for support when you need it? Does the kind of support you are looking for influence
who you go to? Why? Why did you move into the community that you live in now?
What are your dreams for yourself and your family? How do you plan to achieve those
dreams? What sort of resources do you have to do that? The questions should be open-
195
ended and delve into the way in which people actively seek and give support to one
another. Other types of qualitative analysis that would be useful would be a content
analysis (of local papers across the country, what sorts of stories headline locally? How
many community events are being held throughout the year? What sort of language is
being used when talking about the community, events, etc.); and ethnography. Again, the
purpose here would not be to generalize, but to understand social capital at a deeper level.
It is interesting to note that without fail, the structure of social capital varied for
race on every model: Putnam, Coleman, and Bourdieu. Aguilera’s (2002) study further
supports these findings and the assertion that race in America still produced differences
between groups in 2000. It would be interesting for future research to examine what
types of differences exist within these groups as well and extend the current analysis to
The findings from this study were revealing, however, in order to have more
confidence in the results, it would make sense to do validation. This can be done by
testing for the results using different methods and other samples. It is possible that
changing the sample or method used could produce a different result(s). Thus, it is
recommended that further research be conducted on the issues raised to find support or
elucidate the differences across class, racial, and gender groups on the structure of social
capital.
While the current study is limited to the measures available to it in the SCCBS,
future studies may want to look at the differences in gender in a more specific context by
expanding the definition and categories available to the concept of gender. This study
focused on the male/female division, but future studies may wish to dissect the structural
196
differences in social capital along gender lines further by including transsexuals, or even
197
REFERENCES
Abbas, Tahir. 2002. “The Home and the School in the Educational Achievements of
South Asians’.” Race Ethnicity and Education. 5: 291 – 316.
Aguilera, Michael Bernabe. 2002. “The Impact of Social Capital on Labor Force
Participation: Evidence from the 2000 Social Capital Benchmark Survey.” Social
Science Quarterly. 83: p.853-874.
Alder, Paul S. and Seok-Woo Kwon. 2002. “Social Capital: Prospects for a New
Concept.” Academy Management Review. 27:17-40.
Allison, Paul D. 1999. Multiple Regression, A Primer. Thousand Oaks, CA: Pine Forge
Press.
Bassani, Cherylynn. 2007. “Five Dimensions of Social Capital Theory as They Pertain to
Youth Studies.” Journal of Youth Studies. 10: 17-34.
Bentler, P.M. and Douglas G. Bonett. 1980. “Significance Tests and Goodness of Fit in
the Analysis of Covariance Structures.” Psychological Bulletin. 88: 588-606.
Bentler, P.M. and Douglas G. Bonett. 1989. EQS Structural Equations Program Manual.
Los Angeles: BMDP Statistical Software.
Beugelsdijk, S., & Smulders, S. 2003. Bridging and Bonding Social Capital: Which Type
Is Good for Economic Growth?, in W. Arts, J. Hagenaars, & L. Halman (Eds.),
The cultural diversity of European unity, pp. 147-184. Leiden: Brill.
198
Blau, Peter M. 1994. “The Process of Stratification.” Pp. 317-346 in Social
Stratification: Class, Race, and Gender in Sociological Perspective, edited by
David B. Grusky. Boulder, CO: Westview Press.
Bollen, Kenneth A. 1989. Structural Equations with Latent Variables. New York: John
Wiley & Sons, Inc.
-----1986. “Forms of Capital.” Pp. 241-58 in Handbook Theory and Research for
the Sociology of Education, edited by John G. Richardson. Westport, CT.:
Greenwood Press.
-----1997. “The Forms of Capital.” Pp. 46-58 in Education: Culture, Economy, Society,
edited by A. Halsey, H. Lauder, P. Brown, and S. Wells. Oxford: Oxford
University Press.
Bourdieu, Pierre and Loïc Wacquant . 1999. “On the Cunning of Imperialist
Reason.”Theory, Culture & Society. 16: 41-58.
Briggs, X. S. 1998. “Brown Kids in White Suburbs: Housing Mobility and the Many
Faces of Social Capital.” Housing Policy Debate. 9: 177-221.
199
Byrne, Barbara M. 2010. Structural Equation Modeling With AMOS: Basic Concepts,
Applications, and Programming, 2ed. New York, NY: Rutledge.
Campbell, K.E., P. V. Marsden, and J. S. Hurlbert. 1988. “Social Resources and Socio-
Economic Status.” Social Networks. 8:97-116.
Cohen, Bernard P. 1989. Developing Sociological Knowledge: Theory and Method, 2nd
Ed. Chicago, Il: Nelson-Hall Inc.
-----1986. “Social Theory, Social Research, and a Theory of Action.” The American
Journal of Sociology.91: 1309-1335.
Cook, Karen S. (ed.). 2001. Trust in Society. New York: Russell Sage Foundation.
Cook, Karen S., Eric R. W. Rice, and Alexandra Gerbasi. 2004. “The Emergence of Trust
Networks under Uncertainty: The Case of Transitional Economies—Insights from
Social Psychological Research.” Pp. 193-212 in Creating Social Trust in Post-
Socialist Transition, edited by Janos Kornai, Bo Rothstein, and Susan Rose-
Ackerman. New York: Palgrave Macmillan.
Corcoran, M. 1995. “Rags to Rags: Poverty and Mobility in the United States.” Annual
Review of Sociology. 21:237-267.
Cote S, Healy, T. (2001) The Well-being of Nations. The role of human and social
capital.Organisation for Economic Co-operation and Development, Paris.
Curtis, James, William McTeer, and Philip White. 2003. “Do High School Athletes Earn
More Pay? Youth Sport Participation and Earnings as an Adult.” Sociology of
Sport Journal. 20: 60-76.
200
Demant, Jakob and Margaretha Jarvinen. 2011. “Social Capital As Norms and Resources:
Focus Groups Discussing Alcohol.” Addiction Research and Theory. 19: p. 91-
101.
De Graaf, Nan Dirk and Hendrik Derk Flap. 1988. “‘With a Little Help from My
Friends’: Social Resources as an Explanation of Occupational Status and Income
in West Germany, The Netherlands, and the United States.” Social Forces. 67:
452-472.
Doane, Woody. 2003. “Rethinking Whiteness Studies.” Pp. 3-18 in White Out: The
Continuing Significance of Racism, edited by Ashley W. Doane and Eduardo
Bonilla-Silva. New York: Routledge.
Durkheim, Emile. 1973. Moral Education: A Study in the Theory and Application of the
Sociology of Education. New York: Free Press.
Eastis, Carla M. 1998. “Organizational Diversity and the Production of Social Capital.”
American Behavioral Scientist. 42: 66-77.
Elman, Cheryl. 1996. “Old Age, Economic Activity, and Living Arrangements in the
Early-Twentieth-Century United States.” Social Science History. 20:439-468.
Elster, Jon. 2003. “Coleman on Social Norms.” Revue francaise de sociologie. 44: 297-
304.
Farr, James. 2004. “Social Capital: A Conceptual History.” Political Theory. 32: p. 6-33.
Feagin, Joe R. and Karyn D. McKinney. 2003. The Many Costs of Racism. New York,
NY: Rowman and Littlefield Publishers, Inc.
Fine, Ben. 2001. Social Capital Versus Social Theory: Political Economy and Social
Science at the Turn of the Millennium. New York: Routledge.
Fischer, Claude, Micahel Hout, Martin S. Jankowski, Same Lucas, Anne Swidler, and
Kim Voss. 1996. Inequality by Design: Cracking the Bell Curve Myth. Princeton,
N.J.: Princeton University Press.
201
Francis, B. & Archer, L. (2005) “British-Chinese pupils’ and parents’ constructions of
the value of education.” British Educational Research Journal. 31:89-108.
Fritzell, J., M. Nermo, and O. Lundberg. 2004. “The Impact of Income: Assessing the
Relationship Between Income and Health in Sweden.” Scandinavian Journal of
Public Health. 32: 6-16.
Gittell, Marilyn, Isolda Ortega-Bustmante, and Tracy Steffy. 2000. “Social Capital
and Social Change: Women’s Community Activism.” Urban Affairs
Review. 36:123-147.
Goulbourne, H. and J. Solomos. 2003. “Families, Ethnicity and Social Capital.” Social
Policy and Society. 2:329-338.
Greenfield, Patricia. 2013. “The Changing Psychology of Culture From 1800 to 2000.”
Psychological Science. 24: 1722-1731.
Grootaert, Christiaan, Deepa Narayan, Veronica Nyhan Jones, and Michael Woolcock.
2004.“Measuring Social Capital: An Integrated Questionnaire.” World Bank
Paper no. 18.
202
Halpern, David. 2005. Social Capital. Malden, MA: Polity Press.
Harpham, Trudy, Emma Grant, and Elizabeth Thomas. 2002. “Measuring Social Capital
Within Health Surveys: Key Issues.” Health Policy and Planning. 17: 106-111.
Hatcher, Larry. 1994. A Step-by-Step Approach to Using SAS for Factor Analysis and
Structural Equation Modeling. Cary, NC: SAS Publishing.
Herd, Pamela and Madonna Harrington Meyer. 2002. “Care Work: Invisible Civic
Engagement.” Gender and Society. 16:665-688.
Herreros, Francisco. 2004. The Problem of forming social capital: Why Trust? New
York, NY: Palgrave Macmillan.
Horton, Nicholas J. and Ken P. Kleinman. 2007. “Much Ado About Nothing: A
Comparison of Missing Data Methods and Software to Fit Incomplete Data
Regression Models.” The American Statistician. 61: 79-90.
Howell, David C. 2007. “The Analysis of Missing Data.” In Handbook of Social Science
Methodology, edited by W. Outhwaite and S. Turner. London: Sage.
Hoyle, Rick H. ed. 1995. Structural Equation Modeling: Concepts, Issues, and
Applications. Thousand Oaks, CA: Sage Publications.
Hu, L., and P. M. Bentler. 1995. “Evaluating Model Fit.” Pp. 76-99, in R. H. Hoyle, Ed.,
Structural Equation Modeling: Concepts, Issues, and Applications. Thousand
Oaks, CA: Sage.
James, L.R., S. A. Mulaik, and J. M. Brett. 1982. Causal Analysis. Beverly Hills, CA:
Sage.
James, Sherman A., Amy J. Schulz, and Juliana van Olphen. 2001. “Social Capital,
Poverty, and Community Health: An Exploration of Linkages.” Pp. 165-188 in
Social Capital and Poor Communities, edited by Susan Saegert, J. Phillip
Thompson and Mark R. Warren. New York, NY: Russell Sage Foundation.
Joreskog, K.G. and D. Sorbom. 1989. LISREL 7: A Guide to the Program and
Application, 2nd Ed. Chicago, Il.: SPSS Inc.
203
Kaufman, Jason and David Weintraub. 2004. “Social-Capital Formation and American
Fraternal Association: New Empirical Evidence.” Journal of Interdisciplinary
History. 35:1-36.
Kline, Rex B. 2011. Principles and Practice of Structural Equation Modeling, 3rd Ed.
New York, NY: The Guilford Press.
Krais, Beate. 1993. “Gender and Symbolic Violence: Female Oppression in the Light of
Pierre Bourdieu’s Theory of Social Practice,” Pp. 156-177 in Bourdieu: Critical
Perspectives, edited by Craig Calhoun, Edward LiPuma and Moishe Postone.
Chicago: University of Chicago Press.
Knudsen, Brian, Richard Florida, and Denise Rousseau. 2007. “Bridging and Bonding: A
Multi- Dimensional Approach to regional Social Capital.”
http://www.rotman.utoronto.ca/userfiles/prosperity/File/Bridging_and_Bonding.
w.cover.pdf (2008-09-23).
Kolankiewicz, George. 1996. “Social Capital and Social Change.” The British Journal of
Sociology. 47:427-441.
Kooistra, Paul. "Bend it Like Bourdieu: Class, Gender and Race in American Youth
Soccer." Paper presented at the annual meeting of the American Sociological
Association, Marriott Hotel, Loews Philadelphia Hotel, Philadelphia, PA, Aug
12, 2005. 2009-03-05 http://www.allacademic.com/meta/p21892_index.html.
Lamont, Michele. 2012. “How Has Bourdieu Been Good To Think With? The Case of
the United States.” Sociological Forum. 27: 228-237.
Lamont, Michele and Annette Lareau. 1988. “Cultural Capital Allusions, Gaps, and
Glissandos in Recent Theoretical Developments.” Sociological Theory. 6:153-
168.
Lareau, A. 2000. Home Advantage. New York: Rowman and Littlefield Publishers,
Inc.
Lin, Nan. 1999. “Social Networks and Status Attainment.” Annual Review of Sociology.
25: 467-487.
-----2001. Social Capital: A Theory of Social Structure and Action. New York, NY:
Cambridge University Press.
204
-----2004. “The Invisible Hand of Social Capital.” Conference Paper, American
Sociological Association: San Franscico, Ca.
Lin, Nan and M. Dumin. 1996. “Access to Occupations Through Social Ties.” Social
Networks. 8:365-385.
MacDonald, John and Robert J. Stokes. 2006. “Race, Social Capital, and Trust in the
Police.” Urban Affairs Review. 41:358-375.
MacLeod, Jay. 1995. Ain’t No Makin’ It: Aspirations and Attainment in a Low-Income
Neighborhood. Boulder, CO: Westview Press.
Malinowski, Bronislaw. 1932. Crime and Custom in Savage Society. London: Paul,
Trench, Trubner.
Maloney, William A., Graham Smith, and Gerry Stoker. 2000. “Social Capital and
Associational Life.” Pp. 212-225 in Social Capital: Critical Perspectives, edited
by Stephen Baron, John Field, and Tom Schuller. New York: Oxford University
Press.
Marsden, Peter V. and Jeanne S. Hulbert. 1988. “Social Resources and Mobility
Outcomes: A Replication and Extension.” Social Forces. 1038-1059.
Messner, Steven F., Eric P. Baumer and Richard Rosenfeld. 2004. “Dimensions of Social
Capital and Rates of Criminal Homicide.” American Sociological Review. 69:
882-903.
Mitsztal, Barbara A. 1996. Trust in Modern Societies: The Search for the Bases of Social
Order. Cambridge: Polity Press.
Mouw, Ted. 2003. “Social Capital and Finding a Job: Do Contacts Matter?” American
Sociological Review. 68:868-898.
Nieminen, Tarja, Tuija Martelin, Seppo Koskinen, Jussi Simpura, Erkki Alanen, Tommi
Harkanen, and Arpo Aromaa. 2008. “Measurement and Socio-Demographic
Variation of Social Capital in a Large Population-Based Survey.” Social
Indicators Research. 85:405–423.
Norris, Pippa and Ronald Inglehart. 2006. “Gendering Social Capital: Bowling in
Women’s Leagues?” Pp. 73-98 in Gender and Social Capital, edited by Brenda
O’Neill and Elisabeth Gidengil. New York: Routledge.
205
O’Neill, Brenda, and Elisabeth Gidengil (ed.). 2006. Gender and Social Capital. New
York: Routledge.
Parcel, Toby L. and Elizabeth G. Menaghan. 1993. “Family Social Capital and Children’s
Behavior Problems.” Social Psychology Quarterly. 56:120-135.
-----1994. “Early Parental Work, Family Social Capital, and Early Childhood Outcomes.”
American Journal of Sociology. 99:972-1009.
Peng, Yusheng. 1992. “Wage Determination in Rural and Urban China: A Comparison of
Public and Private Industrial Sectors.” American Sociological Review. 57: 198-
213.
Portes, Alejandro. 1998. “Social Capital: Its Origins and Applications in Modern
Sociology.” Annual Review of Sociology. 24:1-24.
Portes, Alejandro and Patricia Landolt. 2000. “Social Capital: Promise and Pitfalls of its
Role in Development.” Journal of Latin American Studies. 32:529-547.
Prell, C. and J. Skvoretz. (2008). “Looking at Social Capital Through Triad Structures.”
Connections 28, 2: 4-16.
----- 2001. Social Capital Benchmark Survey, 2000. Storrs, CT: Roper Center.
Putnam, Robert D. (Ed.). 2002. Democracies in flux: The evolution of social capital in
contemporary society. New York: Oxford University Press U.S.
206
-----2013. “Crumbling American Dreams.” The New York Times. Accessed on November
12, 2013 at http://opinionator.blogs.nytimes.com/2013/08/03/crumbling-
american-dreams/?_r=1.
Rahn, Wendy M. and Thomas J. Rudolph. 2005. “A Tale of Political Trust in American
Cities.” Public Opinion Quarterly. 69: p. 530-560.
Raju, Das. 2004. “Social Capital and Poverty of the Wage-Labour Class: Problems with
the Social Capital Theory.” Transactions of the Institute of British Geographers.
29: 27-45.
Roschelle, Anne R. 1997. No More Kin: Exploring Race, Class, and Gender in Family
Networks. London: Sage.
Rothstein, Bo. 2005. Social Traps and the Problem of Trust. New York: Cambridge
University Press.
SAS Institute Inc. 1989. SAS/STAT Users Guide, Version 6, Fourth Ed., Volume 1. Cary,
NC: SAS Institute Inc.
Shaw, Susan M. and Janet Lee. 2012. Women’s Voices Feminist Visions: Classic and
Contemporary Readings, 5th ed. McGraw Hill.
Smaje, Chris. 1997. “Not Just a Social Construct: Theorising Race and Ethnicity.”
Sociology. 31: 307-327.
Smith, Sandra S. 2000. “Mobilizing Social Resources: Race, Ethnic, and Gender
Differences in Social Capital and Persisting Wage Inequalities.” Sociological
Quarterly. 41:509-537.
Stack, Carol B. 1974. All Our Kin: Strategies for Survival in a Black Community. New
York: Harper & Row.
Stephen, Morgan L. and Aage B. Sorensen. 1999. “Parental Networks, Social Closure,
and Mathematics Learning: A Test of Coleman’s Social Capital Explanation of
School Effects.” American Sociological Review. 64:661-681.
207
Thapar, Suruchi and Gurchathen S. Sanghera. 2010. “Building Social Capital and
Education: The Experiences of Pakistani Muslims in the UK.” International
Journal of Social Inquiry. 3: 3-24.
Wamala, Sarah, Johanna Ahnquist, and Anna Mansdotter. 2009. “Research Article: How
Do Gender, Class and Ethnicity Interact To Determine Health Status?” Journal of
Gender Studies.18: 115-129.
Warren, Mark R., J. Phillip Thompson, and Susan Saegert. 2001. “The Role of Social
Capital in Combating Poverty.” Pp. 1-28 in Social Capital and Poor
Communities, edited by Susan Saegert, J. Phillip Thompson and Mark R. Warren.
New York, NY: Russell Sage Foundation.
Weber, Max. [1922] 1978. Economy and Society: An Outline of Interpretive Sociology
edited by Guenther Roth and Claus Wittich. Berkeley and Los Angeles:
University of California Press.
-----[1921] 1968. Economy and Society. 3 Vols. Totowa, NJ: Bedminster Press.
-----1946. “Class, Status, and Party.” Pp. 180-195 in From Max Weber: Essays
in Sociology edited by H. H. Gerth and C. Wright Mills. New York: Oxford
University Press.
-----1949. The Methodology of the Social Sciences, edited by Edward Shils and Henry
Finch. New York: Free Press.
Wellman, Barry and Kenneth A. Frank. 2001. “Network Capital in a Multilevel World:
Getting Support from Personal Communities,” Pp. 233-273 in Social Capital:
Theory and Research, edited by Nan Lin, Karen Cook, and Robert S. Burt. New
York: Aldine De Gruvter.
White, Michael J. and Gayle Kaufman. 1997. “Language Usage, Social Capital, and
School Completion Among Immigrants and Native-Born Ethnic Groups.” Social
Science Quarterly. 78: 385-398.
Willem, Annick and Harry Scarbrough. 2006. “Social Capital and Political Bias in
Knowledge Sharing: An exploratory study.” Human Relations. 59:1343-1370.
Willis, P. 1977. Learning to Labour: How Working Class Kids Get Working Class Jobs.
Westmead, England: Saxon House.
Wilson, William Julius. 1996. When Work Disappears: The World of the New Urban
Poor. New York: Vintage Books.
Woolcock, M.. (2001) The place of social capital in Understanding Social and Economic
Outcomes. ISUMA Canadian Journal of Policy Research 2 (1) 11-17.
208
Yee, Julie L. and Debbie Niemeier. 1996. “Advantages and Disadvantages: Longitudinal
vs. Repeated Cross-Section Surveys, A Discussion Paper.” Project Battelle 94-16.
http://ntl.bts.gov/lib/6000/6900/6910/bat.pdf
Yip, Winnie, S.V. Subramanian, Andrew D. Mitchell, Dominic T. S. Lee, Jian Wang,
Ichiro Kawachi. 2007. “Does Social Capital Enhances Health and Well-being?
Evidence From Rural China.” Social Science & Medicine. 64: 35-49.
Zhang, Saijun, Steven G. Anderson, and Min Zhan. 2011. “The Differential Impact of
Bridging and Bonding Social Capital on Economic Well-being: An Individual
Level Perspective.” Journal of Sociology and Social Welfare. 38:119-142.
Zhou, M. and C. Bankston, III. 1994. “Social Capital and the Adaptation of the Second
Generation: The Case of Vietnamese Youth in New Orleans.” International
Migration Review. 28: 821-845.
Zippay, Allison. 2001. “The Role of Social Capital in Reclaiming Human Capital: A
Longitudinal Study of Occupational Mobility among Displaced Steelworkers.”
Journal of Sociology and Social Welfare. 28: 99-119.
209
APPENDICES
210
APPENDIX A
PUTNAM
BONDING SOCIAL CAPITAL DOMAIN
Friends:
Q53. Now, how about friends? About how many CLOSE FRIENDS do you have these
days? These are people you feel at ease with, can talk to about private matters, or call on
for help. Would you say that you have no close friends, one or two, three to five, six to
ten, or more than that?
<FRIENDS>
1 No close friends
2 1-2 close friends
3 3-5 close friends
4 6-10 close friends
5 More than 10 close friends
8 Don’t know
9 Refused
Shared confidences:
Q. 54. Right now, how many people do you have in your life with whom you can share
confidences or discuss a difficult decision – nobody, one, two, or three or more?
(INTERVIEWER NOTE: INCLUDES FAMILY)
<CONFIDE>
1 Nobody
2 One
3 Two
4 Three or more
8 Don’t Know
9 Refused
Family Contact:
Q. 56D. (How many times in the past 12 months have you) visited relatives in person or
had them visit you? (GIVE RESPONDENT A MOMENT TO THINK ABOUT THE
ANSWER.
211
IF RESPONSE IS 53 OR GREATER, ENTER 53)
<CFAMVISI>
VALID RANGE 0 to 53
__
98 Don’t Know
99 Refused
BRIDGING SOCIAL CAPITAL DOMAIN
Diversity scale:
Q.55. Thinking now about everyone that you would count as a PERSONAL FRIEND,
not just your closest friends—do you have a personal friend who…
Q.55E. [coded to signify, has personal friend with a different religious orientation]
(IF <RELIG>=1) Is not Protestant?
(IF <RELIG>=2) Is not Catholic?
(IF <RELIG>=3) Has a different religion than you?
212
(IF <RELIG>=4) Is not Jewish?
(IF <RELIG>=5) Has a different religion than you?
(IF <RELIG>>5) You consider to be very religious?
<BREL>
1 Yes
0 No
8 Don’t know
9 Refused
Q. 55F. Is White?
<BWHT>
1 Yes
0 No
8 Don’t know
9 Refused
Q. 55H. Is Asian?
<BASN>
1 Yes
0 No
8 Don’t know
9 Refused
213
1 Yes
0 No
8 Don’t know
9 Refused
Group involvement scale:
Q.33. Now I’d like to ask about other kinds of groups and organizations. I’m going to
read a list; just answer YES if you have been involved in the past 12 months with this
kind of group. (BEGIN LIST)
(NOTE: SCHOOL YEAR INVOLVEMENT/SCHOOL ORGANIZATIONS SHOULD
BE INCLUDED)
Q.33A. (IF Q30=1 DISPLAY: Besides your local place of worship,) Any organization
affiliated with religion, such as the Knights of Columbus or B’nai B’rith (BA-NAY
BRITH), or a bible study group?
<GRPREL>
1 Yes
0 No
8 Don’t know
9 Refused
Q.33B. (How about) An adult sports club or league, or an outdoor activity club.
<GRPSPORT>
1 Yes
0 No
8 Don’t know
9 Refused
Q.33C. (How about) A youth organization like youth sports leagues, the scouts, 4-H
clubs, and Boys & Girls Clubs.
<GRPYOUTH>
1 Yes
0 No
8 Don’t know
9 Refused
Q.33D. A parents’ association, like the PTA or PTO, or other school support or service
groups.
<GRPPTA>
1 Yes
0 No
8 Don’t know
9 Refused
214
0 No
8 Don’t know
9 Refused
Q.33H. A charity or social welfare organization that provides services in such fields as
health or service to the needy.
<GRPSOC>
1 Yes
0 No
8 Don’t know
9 Refused
Q.33K. Service clubs or fraternal organizations such as the Lions or Kiwanis or a local
women’s club or a college fraternity or sorority. (NOTE: Includes Alumni Organizations)
<GRPFRAT>
1 Yes
215
0 No
8 Don’t know
9 Refused
Q.33M. Other public interest groups, political action groups, political clubs, or party
committees.
<GRPPOL>
1 Yes
0 No
8 Don’t know
9 Refused
Q.33N. A literary, art, discussion or study group or a musical, dancing, or singing group.
<GRPART>
1 Yes
0 No
8 Don’t know
9 Refused
Q.33P. A support group or self-help program for people with specific illnesses,
disabilities, problems, or addictions, or for their families.
<GRPSELF>
1 Yes
0 No
8 Don’t know
9 Refused
Q.33Q. Are you involved in any group that meets only over the Internet.
<GRPWWW>
1 Yes
0 No
216
8 Don’t know
9 Refused
Q.7. Next, we’d like to know how much you trust different groups of people. First, think
about (GROUP). Generally speaking, would you say that you can trust them a lot, some,
only a little, or not at all?
(GROUP:)
Q.7B. (How about) People you work with (would you say that you can trust them a lot,
some, only a little, or
217
not at all?)
(CLARIFY IF NECESSARY: How about in general?)
<TRWRK>
1 Trust them a lot
2 Trust them some
3 Trust them only a little
4 Trust them not at all
5 (VOLUNTEERED) Does not apply
8 Don’t Know
9 Refused
218
Q.14. Overall, how would you rate your community as a place to live — excellent, good,
only fair, or poor?
<QOL>
1 Excellent
2 Good
3 Only Fair
4 Poor
8 Don’t Know
9 Refused
Perception of community trust:
Q.11. Now I’d like to ask you a few questions about the local community where you live.
If public officials asked everyone to conserve water or electricity because of some
emergency, how likely is it that people in your community would cooperate — would
you say it is very likely, likely, unlikely, or very unlikely?
<COOP>
1 Very likely
2 Likely
3 (VOLUNTEERED) Neither/Depends
4 Unlikely
5 Very Unlikely
8 Don’t Know
9 Refused
COLEMAN
EXPECTATION/OBLIGATION DOMAIN
Group involvement scale:
PROGRAMMING: RANDOM ORDER A-O, KEEPING K-M TOGETHER, KEEPING
N-O TOGETHER, ITEM R SHOULD ALWAYS BE LAST
Q.33A. (IF Q30=1 DISPLAY: Besides your local place of worship,) Any organization
affiliated with religion, such as the Knights of Columbus or B’nai B’rith (BA-NAY
BRITH), or a bible study group?
<GRPREL>
1 Yes
0 No
8 Don’t know
9 Refused
Q.33B. (How about) An adult sports club or league, or an outdoor activity club.
<GRPSPORT>
1 Yes
0 No
8 Don’t know
9 Refused
Q.33C. (How about) A youth organization like youth sports leagues, the scouts, 4-H
clubs, and Boys & Girls Clubs.
219
<GRPYOUTH>
1 Yes
0 No
8 Don’t know
9 Refused
Q.33D. A parents’ association, like the PTA or PTO, or other school support or service
groups.
<GRPPTA>
1 Yes
0 No
8 Don’t know
9 Refused
Q.33H. A charity or social welfare organization that provides services in such fields as
health or service to the
needy.
<GRPSOC>
1 Yes
0 No
8 Don’t know
9 Refused
220
<GRPLAB>
1 Yes
0 No
8 Don’t know
9 Refused
Q.33K. Service clubs or fraternal organizations such as the Lions or Kiwanis or a local
women’s club or a college fraternity or sorority. (NOTE: Includes Alumni Organizations)
<GRPFRAT>
1 Yes
0 No
8 Don’t know
9 Refused
Q.33M. Other public interest groups, political action groups, political clubs, or party
committees.
<GRPPOL>
1 Yes
0 No
8 Don’t know
9 Refused
Q.33N. A literary, art, discussion or study group or a musical, dancing, or singing group.
<GRPART>
1 Yes
0 No
8 Don’t know
9 Refused
221
<GRPHOB>
1 Yes
0 No
8 Don’t know
9 Refused
Q.33P. A support group or self-help program for people with specific illnesses,
disabilities, problems, or addictions, or for their families.
<GRPSELF>
1 Yes
0 No
8 Don’t know
9 Refused
Q.33Q. Are you involved in any group that meets only over the Internet.
<GRPWWW>
1 Yes
0 No
8 Don’t know
9 Refused
222
Q.49. How many other adults, if any, do you treat as members of your family even
though they are not related to you? (IF NECESSARY: These are people who are
regularly included in family gatherings and celebrations, and who may be called "uncle"
or "aunt" although they are not.) (INTERVIEWER NOTE: IF RESPONSE IS 10 OR
GREATER, ENTER 10)
<NONFAM>
VALID RANGE 0-10
__
98 Don’t know
99 Refused
Business owner:
223
4 Unlikely
5 Very Unlikely
8 Don’t Know
9 Refused
Q.7. Next, we’d like to know how much you trust different groups of people. First, think
about (GROUP). Generally speaking, would you say that you can trust them a lot, some,
only a little, or not at all?
(GROUP:)
224
Q.7C. People at your church or place of worship
(CLARIFY IF NECESSARY: How about in general?)
<TRREL>
1 Trust them a lot
2 Trust them some
3 Trust them only a little
4 Trust them not at all
5 (VOLUNTEERED) Does not apply
8 Don’t Know
9 Refused
BOURDIEU
RELIGIOUS FIELD DOMAIN
Volunteer religious:
Q.59. I’m going to list some of the types of organizations where people do volunteer
work. Just tell me whether you have done any volunteer work for each in the past twelve
months.
PROGRAMMING RANDOMIZE A-F
225
8 Don’t Know
9 Refused
Religious membership:
Q.30. Are you a MEMBER of a local church, synagogue, or other religious or spiritual
community?
<RELMEM>
1 Yes
0 No
8 Don’t know
9 Refused
Religious groups:
Q.33A. (IF Q30=1 DISPLAY: Besides your local place of worship,) Any organization
affiliated with religion, such as the Knights of Columbus or B’nai B’rith (BA-NAY
BRITH), or a bible study group?
<GRPREL>
1 Yes
0 No
8 Don’t know
9 Refused
COMMUNITY FIELD DOMAIN
Friend contact:
Q.56F. (How many times in the past twelve months have you) had friends over to your
home?
(GIVE RESPONDENT A MOMENT TO THINK ABOUT THE ANSWER.
IF RESPONSE IS 53 OR GREATER, ENTER 53)
<CFRDVIST>
VALID RANGE 0 to 53
__
98 Don’t Know
99 Refused
Hanging with friends:
Q.56I. (How many times in the past twelve months have you) hung out with friends at a
park, shopping mall, or other public place?
(GIVE RESPONDENT A MOMENT TO THINK ABOUT THE ANSWER.
IF RESPONSE IS 53 OR GREATER, ENTER 53)
<CFRDHANG>
VALID RANGE 0 to 53
__
98 Don’t Know
99 Refused
Coworker contact:
Q.56H. (How many times in the past twelve months have you) socialized with coworkers
outside of work?
(GIVE RESPONDENT A MOMENT TO THINK ABOUT THE ANSWER.
IF RESPONSE IS 53 OR GREATER, ENTER 53)
<CJOBSOC>
226
VALID RANGE 0 to 53
__
98 Don’t Know
99 Refused
POLITICAL FIELD DOMAIN
Rally:
Q.26B. Attended a political meeting or rally?
<RALLY>
1 Yes
0 No
8 Don’t know
9 Refused
Public Interest:
Q.33M. Other public interest groups, political action groups, political clubs, or party
committees.
<GRPPOL>
1 Yes
0 No
8 Don’t know
9 Refused
Petition:
Q.26A. Have you signed a petition?
<PETITION>
1 Yes
0 No
8 Don’t know
9 Refused
Neighborhood association:
Q.33F. A neighborhood association, like a block association, a homeowner or tenant
association, or a crime watch group.
<GRPNEI>
1 Yes
0 No
8 Don’t know
9 Refused
227
APPENDIX B
V27 0.10858 0.12350 1.00000 0.14074 0.08157 0.04814 0.07765 0.01385 0.05820
<.0001 <.0001 <.0001 0.0019 0.0673 0.0031 0.5989 0.0270
V20 0.13257 0.13059 0.08157 0.39689 1.00000 0.27545 0.13192 0.02633 0.12554
<.0001 <.0001 0.0019 <.0001 <.0001 <.0001 0.3173 <.0001
V25 0.03870 0.06669 0.04814 0.20034 0.27545 1.00000 0.05528 0.00930 0.04897
0.1415 0.0112 0.0673 <.0001 <.0001 0.0356 0.7239 0.0628
V22 0.15150 0.09478 0.01385 0.03642 0.02633 0.00930 0.30847 1.00000 0.33163
<.0001 0.0003 0.5989 0.1664 0.3173 0.7239 <.0001 <.0001
V31 0.15235 0.15269 0.05820 0.12868 0.12554 0.04897 0.41181 0.33163 1.00000
<.0001 <.0001 0.0270 <.0001 <.0001 0.0628 <.0001 <.0001
228
V8 V9 V10 V12 V15 V16 V21 V27 V28
V10 0.41170 0.29510 1.00000 0.02443 0.12994 0.17534 0.14392 0.15608 0.16293
<.0001 <.0001 0.3867 <.0001 <.0001 <.0001 <.0001 <.0001
V12 0.05158 0.02750 0.02443 1.00000 0.12796 0.06642 0.07852 0.12802 0.08672
0.0675 0.3299 0.3867 <.0001 0.0185 0.0053 <.0001 0.0021
V15 0.11948 0.00418 0.12994 0.12796 1.00000 0.35903 0.43482 0.29341 0.27810
<.0001 0.8822 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001
V16 0.10902 0.08317 0.17534 0.06642 0.35903 1.00000 0.20381 0.14724 0.12284
0.0001 0.0032 <.0001 0.0185 <.0001 <.0001 <.0001 <.0001
V21 0.10561 0.04441 0.14392 0.07852 0.43482 0.20381 1.00000 0.17353 0.18921
0.0002 0.1156 <.0001 0.0053 <.0001 <.0001 <.0001 <.0001
V27 0.12396 0.00363 0.15608 0.12802 0.29341 0.14724 0.17353 1.00000 0.38628
<.0001 0.8978 <.0001 <.0001 <.0001 <.0001 <.0001 <.0001
V28 0.14479 0.05100 0.16293 0.08672 0.27810 0.12284 0.18921 0.38628 1.00000
<.0001 0.0707 <.0001 0.0021 <.0001 <.0001 <.0001 <.0001
229
V2 V19 V22 V42 V43
230