ArticlePDF Available

A Dimensional Approach to Measuring Social Capital: Development and Validation of a Social Capital Inventory

Authors:

Abstract and Figures

While social capital is emerging as a theory rich in its potential for understanding the relationships between societal norms and values, and community outcomes, clarification of its measures remains unresolved. This article attempts to contribute to this measurement issue by presenting a reliable self-report instrument for measuring social capital in societal environments. The instrument is grounded in the theoretical and measurement literature of social capital, and proposes an evolving conceptual framework of social capital's dimensions, determinants and outcomes. The instrument was empirically validated using data collected in the African Republics of Ghana and Uganda. The article presents results of exploratory and confirmatory factor analyses that substantiate a number of robust dimensions of social capital, prominent at the household and aggregate levels, and across the two country data sets. Both recommended and suggested survey questions are documented for use in subsequent research relevant to measuring social capital. Regression analyses supporting the validity of the measures are included, as are reliability measures.
Content may be subject to copyright.
Deepa Narayan and Michael F.
Cassidy
A Dimensional Approach to
Measuring Social Capital:
Development and Validation of a
Social Capital Inventory
Introduction
S
ocial capital has gained wide acceptability as a fruitful theoretical per-
spective for understanding and predicting the norms and social relations
embedded in the social structures of societies. It is these patterns of social
interrelationships that enable people to coordinate action to achieve desired
goals (Putnam, 1993).
Bourdieu, a French sociologist, was one of the first authors to analyze
systematically the properties of social capital, defining it as ‘the sum of
resources, actual and virtual, that accrue to an individual or a group by virtue
of possessing a durable network or less institutionalized relationships of
mutual acquaintance and recognition’ (Bourdieu, 1980).
James Coleman, a sociologist interested in the role of social capital in
human capital creation and educational outcomes, defined social capital by
its function. ‘It is not a single entity, but a variety of different entities having
two characteristics in common: they all consist of some aspect of social struc-
ture and they facilitate certain actions of individuals who are within the struc-
ture’ (Coleman, 1988). Emphasizing social capital’s function in different
contexts, Portes (1998) defines social capital ‘as the ability of actors to secure
benefits by virtue of memberships in social networks or other social struc-
tures’. These socialization processes, in turn, lead to internalization of a par-
ticular set of values and norms that can then be taken advantage of by others.
Current Sociology, March 2001, Vol. 49(2): 59–102 SAGE Publications
(London, Thousand Oaks, CA and New Delhi)
[0011–3921(200103)49:2;59–102;020037]
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 59
The practical implications of social capital are broad and profound, with
consequences that may be beneficial or pathological. Positive outcomes
operate through and include social control or norm observance, family
support and benefits mediated through extra-familial networks. These have
been demonstrated to have an impact on income outcomes (Burt, 1997;
Montgomery, 1991; Narayan and Pritchett, 1999; Grootaert, 1999; Robin-
son and Siles, 1997); collective action at the community level (Narayan, 1995;
Molinas, 1998); as well as others. The inherent value of social controls is that
they render formal or overt controls unnecessary. The way social capital is
embedded in social structures may contribute to the public good (Narayan,
1998). Conversely, the negative impact of social capital embedded in power-
ful, tightly knit social groups, not accountable to citizens at large, is
evidenced, for example, in corruption and cronyism in political and govern-
ment institutions (see, for example, Evans, 1989; Mauro, 1995; World Bank,
1997).
An intrinsic characteristic of social capital is that it is relational.
Whereas economic capital is in people’s bank accounts and human capital is
inside their heads, social capital inheres in the structure of their relationships.
To possess social capital, a person must be related to others, and it is these
others, not himself, who are the actual source of his or her advantage. (Portes,
1998)
Simply, social capital exists only when it is shared.
While social capital is relational, its influence is most profound when
relationships are among heterogeneous groups. From an economic perspec-
tive, several recent studies conducted as part of the World Bank’s Local Level
Institutions Study (Grootaert and Narayan, 2000) confirm the importance of
heterogeneity in group membership (a gauge of positive social capital) and
economic outcomes. This pattern of results is found in rural Tanzania
(Narayan and Pritchett, 1999), in Indonesia (Grootaert, 1999) and in rural
Bolivia (Grootaert and Narayan, 2000). It is not simply an issue of the extent
to which people are connected to others, but the nature of those connections.
Other studies, particularly from Latin America, consistently demon-
strate that despite high ratings in community solidarity in indigenous com-
munities, communities with high concentrations of indigenous people remain
poor if they have few connections to the powerful within or outside the com-
munity. While they may manage to attract government-provided basic social
infrastructure, this does not result in production opportunities. Indeed, there
is little evidence that indigenous social organizations are providing the foun-
dation for indigenous groups to mobilize either for fundamental rights or for
greater access to economic and political participation (Junho Pena and Lindo-
Fuentes, 1998; Gacitua-Mario, 1998). In the absence of outside allies, indigen-
ous social capital of poor communities remains a substitute for the resources
and services provided by the state.
60 Current Sociology Vol. 49 No. 2
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 60
While there is high consistency in the definitions of social capital at a
general level, including the forms and dimensions it embraces, at an opera-
tional level the interpretations of what social capital is and is not are diverse.
Correspondingly, methods used to measure social capital are varied, reflect-
ing the diversity of its interpretations. Paxton (1999), for example, noted the
‘wide gap between the concept of social capital and its measurement’. A
worthwhile contribution to the growing body of social capital literature,
therefore, is one that advances the reliability and validity of its measures. It
is to this end that this article is principally directed. In particular, our primary
purpose is to provide researchers with a set of statistically validated survey
questions for measuring social capital in developing communities. This effort
at measurement should, as well, further refine the theoretical constructs. A
secondary goal is to document the use of these measures in two African
republics: Ghana and Uganda.
We begin with a brief description of select methodological studies on
social capital, and proceed with the results and conclusions of our own work.
Our principal tool is factor analysis, a multivariate statistical technique for
isolating subsuming factors or dimensions. We also present regression results
that partially validate the measures. We conclude with a recommended set of
core questions for measuring social capital.
The Measurement of Social Capital
Measurement in the social sciences is an inevitably tricky business. The
iconoclastic Nobel laureate in physics, Richard Feynman, suggested that he
chose a career in physics over the social sciences because social science prob-
lems are more difficult (Feynman, 1988); this difficulty stems in part from the
problem of measurement. Theories such as social capital comprise constructs
that are inherently abstract and require subjective interpretation in their
translation into operational measures. Such operational measures are invari-
ably indirect surrogates of their associated constructs. An intermediate step
in defining what social capital is and is not is to unbundle the theory into its
dimensions. We turn next to a brief review of what some others have done in
this pursuit. The review is intended as illustrative, not exhaustive.
Select Social Capital Measurement Studies
World Values Survey
Ronald Inglehart conducted the earliest cross-country work on dimensions
of social capital. Over the last decade he and his collaborators collected data
from 43 societies in the World Values Survey to understand the role of
Narayan and Cassidy: Measuring Social Capital 61
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 61
cultural factors in political and economic development. The first surveys were
conducted in 1981 and the last round in 1995 (Inglehart, 1997).
The most well-known aspects of the World Values Survey among econ-
omists are the variables most directly linked to social capital, namely trust
and membership in associations. Several researchers have used questions
addressing generalized trust. Knack and Keefer (1997), for example, demon-
strate a strong relationship between generalized trust and levels of investment
in 29 countries.
The World Values Survey also includes two questions on memberships
and associations (‘Do you belong to different types of associations?’ and ‘Are
you actively engaged in them?’). Inglehart (pers. comm., 1998) agrees that the
measure on associations is partial since it does not include characteristics of
associations, nor does it include many traditional organizations in develop-
ing countries. In response, the instrument we developed, the Global Social
Capital Survey (GSCS), included questions addressing both of these issues.
Inglehart found no correlation between economic growth and group
membership. However, he did find that the relationship changes with the
level of economic development. The correlation between cumulative
membership in 16 types of organizations was negative in societies with a
GNP/capita above US$8333 (r = –.35) but positive among the less developed
countries or those with GNP/capita below US$8300 (r = .24). Although,
neither correlation is very strong, Inglehart interprets his results to support
Putnam’s thesis that voluntary organizations play positive roles in the early
stages of economic development (Putnam, 1993).
New South Wales Study Onyx and Bullen (1997) developed a practical
measure of social capital for community organizations to assess themselves
as well as the impact of their work in building civic engagement. Using data
drawn from five Australian communities, they identified one general under-
lying factor and eight primary independent or orthogonal factors that col-
lectively account for approximately 50 percent of the variance of social
capital. The eight factors in order of their contribution to the underlying
factor were: participation in local community; proactivity in social context;
feelings of trust and safety; neighborhood connections; family and friends
connections; tolerance of diversity; value of life; and work connections.
Those questions that did not appear to be related to social capital were
those that concerned government institutions and policy. However, the ques-
tions were general, and not specifically related to the quality of interactions
with government agencies. In the GSCS, we attempted to measure the quality
of interactions with such agencies.
The Barometer of Social Capital, Colombia John Sudarsky (1999), drawing
in part on the World Values Survey, has developed and tested an instrument
62 Current Sociology Vol. 49 No. 2
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 62
in Colombia that empirically resulted in two dimensions: social capital and
‘faith in unvalidated sources of information’ (FUSI). Using factor analysis,
Sudarsky reported that the social capital factor accounted for 38 percent of
the variance, while FUSI accounted for 12 percent of the variance. Again
using factor analysis, Sudarsky identifies eight dimensions subsumed within
the social capital factor. The dimensions he identifies are: institutional trust;
civic participation; mutuality and reciprocity; horizontal relationships; hier-
archy; social control; civic republicanism; and political participation.
Index of National Civic Health, USA Concerned with the decline of civic
engagement in the USA, the National Commission of Civic Renewal identi-
fied five equally weighted dimensions in the Index of National Civic Health
(National Commission on Civic Renewal, 1996). The five dimensions are:
political engagement; trust; associational membership; security and crime;
and family stability and integrity. The political dimension includes voting in
elections and other political activities such as signing a petition and writing a
letter to a newspaper, behaviors we measured as well in the GSCS. Trust
includes trust in others, and trust and confidence in the federal government.
Associational membership includes membership in groups or church atten-
dance, charitable contributions, local level participation and serving as an
officer in local groups. Security and crime encompasses murder rates in the
youth population, fear of crime and survey-reported crime per population.
Finally, family stability and integrity comprise such elements as divorce rates
as well as non-marital birth rates. The study tracks change since 1974 and
concludes that although the trend lines are different for different dimensions,
overall there has been a consistent decline in civic participation over the last
three decades in the USA. A similar conclusion is documented in Putnam’s
recent work, Bowling Alone (Putnam, 2000).
A summary of some of the key dimensions across studies is presented in
Table 1. As described later, our view on what we considered to be the under-
lying dimensions changed as the study progressed. The Table, however,
reflects our current thinking. We have taken liberties in broadly defining the
dimensions across investigations. We also acknowledge that there are quali-
tative differences in the ways in which these dimensions have been opera-
tionalized across studies. A tick simply indicates an attempt to measure the
construct. Table 1 is presented primarily to give the current investigation a
methodological context.
Table 1 reveals the strong consistency across researchers in the dimen-
sions conjectured to be subsumed within the social capital construct. Trust
and membership, for example, are included in all the studies. Safety, connec-
tion with family and friends, reciprocity and social proactivity dominate the
studies. The uniqueness of our research rests on several issues. It strives for
comprehensiveness in the dimensions and measures it employs. Second, it
Narayan and Cassidy: Measuring Social Capital 63
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 63
64 Current Sociology Vol. 49 No. 2
Table 1 Comparison of Social Capital Dimensions Across Studies
Studies
————————————————————————————————
Underlying World New Barometer Index of Present
Dimensions of Values South Social of National Study
Social Capital Wales Capital Civic Health (GSCS)
Trust ✓✓Institutional ✓✓
Includes Trust, including
institutional institutional trust
(government)
trust
Memberships in ✓✓Horizontal ✓✓
associations/ relationships Included in the
participation in dimension ‘group
local community characteristics’
Proactivity in Social control Empowerment
social context
Crime and safety ✓✓(Outcome)
Neighborhood Horizontal Asking for help
connections relationships
Family and friend Horizontal Everyday
connections relationships Divorce/ sociability
non-marital
birth rates
Tolerance of (Outcome)
diversity
Reciprocity ✓✓ Included in the
dimension
‘generalized norms’
Political ✓✓(Outcome)
engagement
Subjective
well-being Variables related to
trustworthiness of
people; how well
people get along, etc.
are subsumed under
the social capital
dimensions
‘generalized norms’
and ‘togetherness’.
Variables specifically
related to self-
reported happiness,
satisfaction with life,
etc. are defined here
as outcome variables
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 64
attempts to address some of the problematic issues that emerged in previous
research. Third, it attempts to differentiate among determinants, dimensions
and outcomes of social capital. Finally, it presents a reliable set of measures
for use by other researchers, thus permitting more reliable comparisons
across future empirical investigations.
The Conceptual Basis for the Current Study
We postulated several key dimensions against which social capital should be
measured and their relationship to determinants and outcomes. A simplified
version of this framework, derived from Narayan (1999), is shown in Figure
1. The framework reflects our point of departure. It changed over time,
largely as a function of our analyses of the two data sets. While results varied,
certain patterns remained consistent and strong. It is these patterns, discussed
later in the article, that are ultimately included in what we propose as viable
social capital measures. They are reflected in Figure 3, a revision of Figure 1
presented near the end of the article.
Note that neither determinants nor outcomes constitute exhaustive sets
in Figure 1. Note as well that empowerment, a gauge of the perceived posi-
tive impact one can have on a community, is represented as both a determi-
nant and an outcome. While psychometrically vexing, certain variables can
Narayan and Cassidy: Measuring Social Capital 65
Illustrative Proximate
Determinants
of Social Capital
Dimensions of
Social Capital
Illustrative Social, Political and
Economic Outcomes
of Social Capital
Memberships in informal groups,
and networks with particular
characteristics
Everyday sociability
Community participation and
neighborhood connections
Family connections
Trust and fairness norms
Crime and safety
Subjective well-being
Political engagement
Sense of
belonging
Empowerment
Community
solidarity
Governance;
political
engagement
Safety and
security
Empowerment
Social
cohesion
Figure 1 A Simplified Measurement Framework
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 65
be defined as both determinants and outcomes. Consider, for example, the
intercorrelations for four variables selected for illustration in Table 2, and
drawn from the Ghana EA (enumeration area) data. Each column heading
includes a designation of whether we perceive the variable to be an outcome
(O) measure or a possible outcome or determinant (O/D). Spearman’s rho
correlations are shown. Emboldened correlations are significant at the .01 or
.05 level.
Nearly 70 percent of the correlations are statistically significant, a finding
that substantiates a strong interrelationship between determinants and out-
comes. Those correlations that are significant are, as well, sensible. For
example, happiness is significantly correlated with a sense of identity and per-
ceived impact. (The negative and non-significant correlation between happi-
ness and voting provides an interesting counterpoint.) Determinants and
outcomes are difficult to differentiate into mutually exclusive categories.
Methods
Development of the Global Social Capital Survey: Overview
Figure 2 details the relationship among the hypothesized dimensions of social
capital and the questions employed to measure each dimension, and served
as a blueprint against which the questionnaire was organized. (Note that
everyday sociability refers to questions regarding with whom and with what
frequency people do such things as visit one another, eat outside the home,
shop and play games together.) Building on a review of the literature and 25
existing questionnaires and qualitative data collection instruments focused on
social capital, one of the authors (Narayan) developed a draft instrument. In
turn, a workshop with multidisciplinary participation (sociologists, anthro-
pologists, political scientists, economists and so on) was held at the World
66 Current Sociology Vol. 49 No. 2
Table 2 Intercorrelation Matrix for Select Variables
Ghana – EA-Level Data Impact
(Empowerment) Happiness Identity
(O/D) (O) (O/D)
How much impact can you have on your
community? 1.0
How happy are you? .35** 1.0
Sense of identity .31** .40** 1.0
Vote – last state election .53** –.002 .23
* Correlation is significant at .05 level (two-tailed)
** Correlation is significant at .01 level (two-tailed)
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 66
Bank on 24–25 June 1998 to review and suggest modifications to the ques-
tionnaire.
1
The approach meets the two principal standards for ensuring
content validity insofar as it was both ‘sensible’ and resulted in a representa-
tive collection of questions (Nunnally and Bernstein, 1994).
The questionnaire was piloted in the Republic of Ghana in summer 1998.
Narayan and Cassidy: Measuring Social Capital 67
Figure 2 The Dimensions of Social Capital
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 67
Data were extensively analyzed, with exploratory factor analysis of particu-
lar relevance. What we found supported our a priori suspicions; emerging
from the data was a number of stable factors, or dimensions, of social capital:
trust, everyday sociability, generalized norms and so forth.
A second pilot study was conducted in the Republic of Uganda in late
1998. Our intent in this phase of the investigation was to progress from
exploration to confirmation. Would the same dimensions that emerged so
clearly in Ghana emerge as well in a population so demographically dis-
similar? They did. Exploratory factor analyses resulted in outcomes highly
consistent with what was found in the Ghana. Subsequent confirmatory
factor analyses were also performed on the Ugandan data using structural
equation modeling (SEM) to test for the hypothesized dimensions. In sum,
the hypothesized dimensions of social capital are largely stable and consist-
ent across data sets, and the questions used to measure social capital are
demonstrably reliable and valid.
Administration of the Survey – Ghana and Uganda
In this section, we first describe some issues pertinent to the administration
of the survey in the two countries, and then describe some salient differences
in the two samples.
Ghana The questionnaire was piloted by a trained team from the Ghana
Statistical Services in 1471 households in four regions, three rural and one
urban (greater Accra), in Ghana during August and September 1998. The
household sample was drawn from clusters established in previous research
in Ghana (Ghana Statistical Service, 1997). In that investigation, a two-stage,
stratified sampling procedure was employed using the National Sampling
Frame of EAs. The frame was first stratified into coastal, forest and savannah
zones, and then into urban and rural EAs.
The largest ethnic group represented in the sample are the Akan (N =
839; 57 percent of the respondents). The Gan (N = 173; 11.8 percent) are the
next largest group represented, followed by the Dagomba (N = 113; 7.8
percent). Most respondents identified themselves as Christians (N = 1103; 75
percent), with Islam represented as the next largest group (N = 245; 17
percent). Respondents’ employment industries are predominately split
between those working in agriculture (N = 658; 45 percent) and in non-agri-
cultural industries (N = 810; 55 percent).
Uganda Prior to its administration in Uganda, several questions in the
questionnaire were reworded and response scales modified. The field experi-
ence in Ghana revealed, for example, two questions in which the precoded
categories were not broad enough to reflect anticipated variability. In
addition, several categorical level scales were revised to ordinal level scales.
68 Current Sociology Vol. 49 No. 2
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 68
These in turn were treated as interval level scales, a practice neither uncom-
mon nor indefensible (see, for example, Cliff, 1993).
Likert-like scales, such as those used in the GSCS, are obviously not con-
tinuous and as such may violate assumptions of univariate or multivariate
normality (Bollen, 1989). To reduce the probability of potentially spurious
results and conclusions, the Uganda data were analyzed in parallel using non-
linear principal components analysis, an approach appropriate to analysis of
categorical data. In general, the results were largely consistent with analyses
performed on both the Ghana and Uganda data sets. Emergent components
(factors) emerged, and, within these components, patterns related to trust,
safety and so forth emerged as well.
The revised questionnaire was administered to a sample of 950 men and
women from impoverished, urban communities in Kampala, Uganda in
November 1998.
Circumstances precluded the same level of sampling rigor employed in
the Ghana study. In addition, the Uganda sample is more homogeneous in
several regards than its Ghana counterpart. The Uganda sample, for example,
was predominately from an economically depressed urban area. Finally, given
the representation of a small number of EAs in the sample (N = 10), analysis
of the Uganda data was limited to the individual household level.
Ghana and Uganda – Some Descriptive Comparisons
To provide the reader with a sense of similarities and differences between the
two data sets, several comparisons follow. We present these data primarily to
establish context for a key finding of the article: despite the substantial demo-
graphic and psychographic differences between the two groups, the dimen-
sions emerging from the factor analyses are remarkable alike.
General Demographic Differences Table 3 compares the two data sets on
several key demographic variables. Chi square tests of independence were
conducted on each categorical variable to test the null hypothesis that the two
variables are statistically independent. In all cases, the null hypothesis was
rejected (p < .001). Averages reported were compared using t tests; again, all
were significant (p < .001).
In sum, the Ghana sample represents smaller sized families, fewer
younger children, lower unemployment, greater diversity in employment
industry and lower crime than the Uganda sample.
General Social Differences Differences between the two sample groups
extend beyond demographics, however, as reflected in Table 4.
The 3232 memberships reported by the 1471 respondents in Ghana rep-
resents an average of about 2.2 memberships per person. In contrast, the
average membership per respondent in Uganda is 0.5. Approximately 92
Narayan and Cassidy: Measuring Social Capital 69
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 69
70 Current Sociology Vol. 49 No. 2
Table 4 Group Membership
Ghana Uganda
Total number of 3232 Represents an average 477 Represents an average
memberships of 2.2 memberships of 0.5 memberships
reported per respondent per respondent
Respondent belongs to a: Yes No Yes No
Religious or spiritual 1350 119 112 835
group 91.9% 8.1% 11.8% 87.9%
Cultural, social group 741 728 80 865
50.3% 49.4% 8.4% 91.1%
Sports group 135 1333 70 865
9.2% 90.5% 8.3% 91.1%
Ethnic-based group 384 1086 47 897
26.1% 73.7% 4.9% 94.4%
Table 3 Demographic Comparisons
Demographics Ghana Uganda
Number of children under In general, the numbers By contrast, only 20%
16 years are much smaller. For (N = 189) of the respondents
example, 33% (N = 471) in Uganda responded that
of the respondents in they had no children in the
Ghana indicated no household less than 16
children under 16 years years.
of age.
Average number of 7; Std. Error = 0.14 12; Std. Error = 5.87
household members
Average age of respondent 45; Std. Error = 16 31; Std. Error = 10
Employment status 9% unemployed (N = 133) 13% unemployed (N = 126)
Industry of employment Agriculture = 44.7% Agriculture = 7.6% (N = 50)
(N = 658) Non-agriculture = 92.4%
Non-agriculture (N = 609)
= 55% (N = 810)
Experience with violent 93% (N = 1364) 76% (N = 714)
crime in previous responded that they had answered similarly
12 months not experienced violent
crime
Non-violent crime in 15% reported one or 40% reported one or more
previous 12 months more non-violent crimes non-violent crimes
(N = 223) (N = 388)
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 70
percent of the Ghana respondents, for example, indicated membership in a
religious or spiritual group. The percentages in Uganda are a near mirror
reversal. While not as dramatic a difference across other types of groups, with
the exception of sports groups, memberships in Ghana are substantially
higher than those in Uganda.
Table 5 compares the two data sets across several attitudinal variables. As
with the demographic data, chi square tests of independence were conducted
to test the null hypothesis that the particular variable is statistically indepen-
dent of country origin. In all cases the null hypothesis was rejected (p < .001).
In summary, data from the Ghana sample reflect a people high in opti-
mism, self-efficacy, security and personal power. Uganda, by contrast, is a
largely mirror image. Setting aside differences between the two samples, one
provocative similarity is worth mentioning here because of its relationship to
a fundamental aspect of social capital: the reasons given for why groups are
active or inactive.
Narayan and Cassidy: Measuring Social Capital 71
Table 5 Attitudinal Comparisons
Variable Ghana Uganda
Extent to which Relatively high. For Relatively low. For
respondents perceive that example, ~27% (N = 401) example, ~16% (N = 155)
they can have an impact on reported ‘big impact’, reported ‘big impact’,
community and ~13% (N = 198) whereas nearly 25%
indicated ‘no impact’. (N = 232) reported ‘no
impact’.
Power to change one’s life Relatively high sense of Relatively low self-efficacy.
self-efficacy. For example, For example, ~32%
~ 50% (N = 733) (N = 301) answered
responded ‘somewhat ‘somewhat powerful’ or
powerful’ or ‘very ‘very powerful’.
powerful’.
Perception of how the Relatively high optimism. Relatively low optimism.
household will be in the For example, the
future percentage of those
responding ‘much better
off’ (~35%; N = 518) is
seven times that of the
Ugandan responses (5%;
N = 47).
Perceived safety of Relatively high sense of Relatively low sense of
household from crime and safety. For example, those safety. For example, those
violence answering ‘very safe’ was answering ‘very safe’ was
1012 (~69%). 124 (~13%).
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 71
In both Ghana and Uganda, the same principal reasons given for why
groups are active or inactive are identical. Strong leaders are why groups are
active; lack of strong leaders is why groups are inactive. A strong sense of
community engenders active groups; a lack of community cohesion impedes
the creation or maintenance of active groups. The third and fourth most fre-
quently occurring responses in both countries for groups being active are: to
make life better and to advance economically. The percentages, however, are
comparatively small. For example, a strong sense of community was men-
tioned by 36 percent of the Ghana respondents, whereas economic advance-
ment was cited by only 9 percent of these respondents. Groups have,
presumably, some intangible, inherent value to their members. While group
involvement may contribute to economic growth under some circumstances,
it is not perceived by respondents as an explicit means for achieving this end.
Community involvement appears to stimulate involvement in groups, which,
in turn, reinforces community involvement.
Factor Analyses
General Factor Structures
All scales in the GSCS were examined and recoded where necessary to
achieve consistency in direction: that is, the higher the value for any given
question the greater the social capital, or the higher the value on the outcome
variable. For example, several questions used to measure social interactions
were recoded to assign a higher value to greater heterogeneity of social inter-
actions. Several individual questions were combined into additive indices.
Group membership, for example, originally segmented into 13 variables
(religious, cultural, political and so forth), was collapsed into two indices –
group membership (a measure of the number of different groups to which
respondents belong) and number of groups (a measure of the number of
groups to which respondents belong).
Preliminary analysis of the data to access suitability for factor analysis
was performed. In particular, the Keiser-Meyer-Olkin (KMO) measure of
sampling adequacy was .77, and the Bartlett test of spericity was significant
at p < .001 for the Ghana data. Both measures support the appropriateness of
performing factor analyses on the data. Similar results were obtained for
Uganda. The primary approach was maximum likelihood with Varimax rota-
tion.
While the issue of the number of factors to retain is ultimately judg-
mental (Green, 1978), we selected a ten-factor model as the best fit for the
Ghana data, and a more parsimonious four-factor model for Uganda. Virtu-
ally all eigenvalues (which define the proportion of variance accounted for
by each of the factors extracted) exceeded unity. In Ghana, the combined
72 Current Sociology Vol. 49 No. 2
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 72
factors account for 47.8 percent of the variance. In Uganda, the combined
factors account for 63.8 percent of the variance.
2
What is common to the two factor structures? First, the principal factor
in both solutions is group characteristics (GC; detailed in Table 7), account-
ing for the largest percentage of variance in each solution (Ghana: 9.2 percent;
Uganda: 41.4 percent). (Table 6 presents the variances, cumulative variances
and eigenvalues associated with each factor for the two data sets.) Indeed, the
first factor in the Uganda factor structure includes all GC variables included
in the analysis.
Second, while the Ghana factor structure disperses the remaining GC
variables across two additional factors, it keeps intact the three indices of
group involvement (participation; money contributed; and involvement in
decision-making) within one factor.
3
In other words, the clustering of vari-
ables remains largely similar across the two factor structures. Third, trust
emerges as a unique factor in both structures. Fourth, volunteerism emerges
as a unique and independent factor in the Ghana structure, while neighbor-
hood connection questions cluster together into a unique factor for Uganda.
While everyday sociability (ES) variables were eliminated from the Uganda
analysis for reasons described in note 3, they do cluster together, albeit across
several factors, in the Ghana structure. Fifth, factor loadings are generally
quite high; all exceed a .4 threshold. In addition, there are no cross-loadings.
In sum, there are strong similarities between the two factor structures, and
Narayan and Cassidy: Measuring Social Capital 73
Table 6 Eigenvalues and Total Variance Explained, Ghana and Uganda
Ghana Uganda
————————————————————– ———————————————————–
(Rotated Sum of Squared Loadings) (Rotated Sum of Squared Loadings)
————————————————————– ———————————————————–
% of Cumulative % of Cumulative
Factor Eigenvalue Variance % Factor Eigenvalue Variance %
GC 3.03 9.19 9.19 GC 7.86 41.36 41.36
Trust 1.95 5.91 15.11 GN and 1.59 8.40 49.77
togetherness
ES 1.84 5.58 20.69 NC 1.49 7.88 57.65
ES 1.65 5.01 25.71 Trust 1.17 6.19 63.85
Volunteerism 1.63 4.94 30.66
ES 1.60 4.85 35.51
GC 1.25 3.80 39.32
Togetherness 1.18 2.59 42.90
ES 0.857 2.59 45.49
GC 0.772 2.34 47.83
GC = Group characteristics; ES = Everyday sociability; GN = Generalized norms; NC =
Neighborhood connections.
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 73
the dimensions depicted in Figure 2 are supported in either one or both of
the two factor structures. What emerged empirically supports what was
speculated theoretically.
Hypothesized Dimensions
We factor analyzed each separate, hypothesized dimension of social capital.
Why? The overall factor structures have integrity as demonstrated in the pre-
ceding section. Since a primary intent of this article, however, is to recom-
mend a stable set of survey questions for measuring the multiple dimensions
of social capital, analyzing each separate dimension permits us to assess the
relative homogeneity of each dimension and to select the ‘best’ items for
measuring the dimension. Our selection guidelines were to choose individual
items that: explain the largest percentage of the variance; have relatively large
and statistically significant intercorrelation coefficients; demonstrate a stable
pattern across the two data sets; and have relatively high loadings on the
factor(s).
In recommending items for inclusion in the final instrument we employ
the following coding scheme. Variables marked with a single asterisk (*) are
strongly recommended for inclusion in subsequent research. Those with two
asterisks (**) are variables that might be eliminated should the researcher
wish a briefer instrument, or retained to improve reliability. Those items
74 Current Sociology Vol. 49 No. 2
Table 7 Group Characteristics – Ghana and Uganda Factor Loadings
Variable Factor 1 Factor 2
*Same education/income heterogeneity index .829
.967
*Same family/kin group heterogeneity index .812
.985
*Same gender heterogeneity index .724
.954
**Same religion heterogeneity index .667
.939
**Same tribe/caste heterogeneity index .654
.965
*Number of memberships .573 .371
.783
**Same neighborhood/community heterogeneity index .541
.927
*Participation index .744 .745
*Decision-making index .810 .587
*Money-contributed index .783 .523
Note: Ghana values are in plain type; Uganda values are in bold type.
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 74
without an asterisk are suggested for elimination. A summary of all variables
and scales is documented in the Appendix.
Group Characteristics Factor loadings for both Ghana and Uganda are
shown in Table 7. Consistent with previous analyses, there is greater vari-
ability in the Ghana results than in the Uganda results. All variables load on
one factor in Uganda. In Ghana, a second factor emerges with one variable
(number of memberships), cross-loading on the two factors. Despite these
minor differences, the pattern across the two country data sets is remarkably
consistent. For example, the relative magnitude of the loadings is relatively
similar, as is the relative order. Note, for example, that the first two variables
in the table have the highest loadings for each country. Note as well the clus-
tering of the last three variables in the table.
Some items intended to measure this dimension were discarded on the
basis of the overall factor analysis. For example, funding source failed to load
meaningfully in either the Ghana or Uganda analyses and was consequently
eliminated.
Generalized Norms It was not technically feasible to extract a factor struc-
ture from the Ghana data for those items related specifically related to
generalized norms (GN). As such, results from Uganda alone are presented
in Table 8. The three items intended to measure GN load on a single factor.
All items are moderately and significantly intercorrelated, and underlie the
recommended (*) vs suggested (**) designations. For subsequent research,
investigators have two options for these three questions. One might legiti-
mately choose to use only one of the questions to measure the construct,
specifically the question related to trust because it has the highest loading on
the factor. An equally defensible option would be to use one or both of the
additional questions. The tradeoff is between time and inconvenience to the
respondents and reliability. In general, the larger the number of questions
asked that defensibly measure the same construct, the higher the reliability.
Narayan and Cassidy: Measuring Social Capital 75
Table 8 Factor Matrix: Generalized Norms – Uganda
Factor
Variable Loading
*Generally speaking, would you say that you can’t be too careful in
dealing with people or that most people can be trusted? .739
**Would you say that most of the time people are just looking out for
themselves, or they are trying to be helpful? .658
**Do you think that most people would try to take advantage, or would
they try to be fair? .648
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 75
Trust Trust emerged as a single factor in the overall factor structures of both
Ghana and Uganda, as presented in Table 9. In comparing the two data sets,
however, note that several variables included in the Ghana questionnaire were
eliminated in the Uganda version (namely, people in your tribe; people in the
same clubs; and politicians). In turn, several variables were added, specific-
ally questions related to trust of community/ward government officials;
judges/courts/police and so on. What conclusions can be drawn from these
data?
First, the factor structure for Ghana is elegantly unidimensional. All vari-
ables load onto the same, single factor. In addition, all intercorrelations
are generally moderate; the exception being the not surprising relatively
low correlation between trust in family members and trust in politicians.
Second, the Uganda data present a less clear pattern. Three factors
emerge rather than one, although the first factor in Uganda includes only
those variables absent in the Ghana questionnaire. With two exceptions
(trust of business owners and trust of community/ward officials), the
remaining variables load on a single factor paralleling Ghana’s one factor.
Correlations are virtually all very low in Uganda, particularly those
associated with community/ward officials. The simplest explanation for
problems with the variable is one of ambiguous interpretation among
respondents.
Finally, since trust is empirically a strong and consistent dimension of
social capital, researchers may want to include or exclude specific vari-
ables pertinent to the community being studied. For example, in a com-
munity with a very strong NGO presence, it may be meaningful to
76 Current Sociology Vol. 49 No. 2
Table 9 Trust – Factor Structures: Ghana and Uganda
Ghana Uganda Uganda Uganda
Variable Factor 1 Factor 1 Factor 2 Factor 3
*Trust in people in your tribe/caste .786 NA
**Trust in people in other tribes .757 .533
**Trust in people in your village .718 .736
**Trust of people in same clubs .697 NA
**Trust of business owners .645 .427
*Trust of politicians .585
*Trust of family members .534 .369
*Trust of government service providers NA .719
*Trust of local/municipal government NA .593
*Trust judges/courts/police NA .447
Trust community/ward officials NA .612
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 76
inquire about trust in the NGO. We believe that the foci of the trust
questions are less important than the inclusion of questions relevant to
trust.
Before examining the dimension ‘togetherness’, a comment about the
relationship between GN and trust is in order. GN it will be recalled includes
one variable related specifically to trustworthiness and two additional ones
related to perceived helpfulness and fairness. In general, the correlations
among the three variables comprising GN and those comprising trust are
generally very low; many are not statistically significant. While superficially
the variables (trustworthiness in particular) seem conceptually related to
trust, they are measuring a different construct.
Togetherness This dimension presents an interesting dilemma. It is techni-
cally not viable to perform a factor analysis on the two variables for either
data set. In the overall factor analysis performed on the Ghana data, both
variables load singularly on the same factor. In Uganda, only one of the two
questions, togetherness, was ultimately included. The question of how well
people get along was very weakly and not significantly correlated (r = .024;
p = .22). In contrast, in Uganda the two variables are moderately correlated
(r = .529; p .0001). Given the conflicting results, we recommend that both
Narayan and Cassidy: Measuring Social Capital 77
Table 10 Factor Loadings: Everyday Sociability – Ghana
Factors
———————————————
12 34
Variables (17%) (16.2%) (10.9%) (7.6%)
*How often do you get together to do arts,
crafts, etc.? .972
**With whom do you do arts, crafts, etc.? .821
*With whom do you play cards, games, etc.? .968
**How often do you get together to play
games? .780
*With whom do you spend time doing chores? .653
**Who visits you at home? .596
**With whom do you eat meals outside the
home? .537
How often do you do chores with people? –.533
How often do you eat meals with others outside
the home? –.453
How often do people visit you at home? .338
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 77
items be included in future research. In addition, we recommend that one or
more additional items directed at measuring the togetherness dimension be
piloted in subsequent surveys.
Everyday Sociability None of the measures of everyday sociability was ulti-
mately included in the overall Uganda factor analysis for reasons mentioned
earlier. While considering the results of the separate analysis of everyday
sociability for Uganda in our recommendations for variables to be included
in subsequent research, we rely primarily on the Ghana results, shown in
Table 10. The percentage of variance explained by each of the four factors is
included in parentheses in the appropriate columns of Table 10.
Neighborhood Connections The two items targeted at measuring neighbor-
hood connections were eliminated from the overall factor solution for Ghana.
We suspect their failure to load meaningfully in the solution is partly a func-
tion of lack of variability in response. The questions were asked with binary
(yes/no) response options; (‘If you were sick, would you ask your neighbors
to care for your children for a few hours?’; ‘If you were sick, would you ask
your neighbors for help?’). The percentage of those responding ‘yes’ to each
question respectively was 88 percent and 90 percent. For the Uganda
administration, the scale was changed to a five-point likelihood scale, result-
ing in greater response variability. In addition, in the overall factor analysis
for Uganda, the two items loaded together on the third of the four-factor
solution. While not technically viable to conduct a separate factor analysis on
the dimension, the two items are strongly and significantly correlated (r= .75;
p .0001). We recommend, therefore, including both items in future survey
research instruments. To improve scale reliability, it is advisable to add one
or more items directed at measuring the same construct.
78 Current Sociology Vol. 49 No. 2
Table 11 Volunteerism – Ghana
Factors
———————
12
*In your community/neighborhood, is it generally expected
that people will volunteer or help in community activities? .758
*Are people who don’t volunteer or participate in community
activities likely to be criticized or fined? .583
**Do you think most people in your community/neighborhood
make a fair contribution? .514
**On average how often do you volunteer in community
activities? .461
How often have you helped someone in the past six months? .995
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 78
Volunteerism Six items were used to measure volunteerism. In the Ghana
data set, five remained in the final factor solution (Table 11). Factor 1 accounts
for 28 percent of the variance in the factor solution, while factor 2 accounts
for approximately 20 percent.
In Uganda, only the second item, addressing negative consequences for
not volunteering, loads with a meaningful value (.998), and on the first factor
from that solution; all other factor loadings are < .23. Indeed, no variable
related to volunteerism remained in the overall Uganda factor solution.
Correlations for the Ghana data are all significant at the .0001 level with
the exception of the question about helping another in the previous six
months, which is not statistically correlated to the question pertaining to
criticism.
We conjecture that the first four items in Table 11 measure the same con-
struct, namely volunteerism, and that the fifth measures a dimension only
partly related to volunteerism. Hence, we suggest that the first two items
listed in Table 11 be retained as measures of the volunteerism dimension of
social capital; the third and fourth be entertained given reliability consider-
ations or the particulars of the community under investigation, and that the
last item be eliminated.
Confirmatory Factor Analysis
While exploratory factor analysis is largely directed at identifying a relatively
small set of underlying clusters or dimensions that subsume a larger number
of intercorrelated variables, confirmatory factor analysis statistically tests
these clusters as well as the predictive validity of the factor structure. It is a
theory testing vs a theory-generating method (Stapleton, 1997).
An SEM approach was used to perform the confirmatory analyses. In
particular, AMOS 3.6 (Arbuckle, 1997), a graphics and text-based SEM
program similar to LISREL and EQS, was used. Confirmatory factor analy-
sis uses a maximum likelihood approach to extract prespecified dimensions
4
and test if the residual covariance matrix still contains significant variation
(Gorsuch, 1983).
It is appropriate in confirmatory factor analysis to test the relationship
between various theoretical models. The approach used here is consistent
with commonly advocated procedures (see, for example, Bentler and
Bonnett, 1980; Breckler, 1990; Gorsuch, 1983), in which several, increasingly
more detailed models are evaluated. We began with a one-factor model in
which all variables were loaded onto one general factor and covariances left
unspecified.
Subsequent manipulation of the model included eliminating three vari-
ables included in the exploratory factor analysis, and specifying a number of
Narayan and Cassidy: Measuring Social Capital 79
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 79
covariances among error terms. The final model is significant (
2
= 171.53;
d.f. = 148; p = .09), permitting acceptance of the null hypothesis that there is
no significant difference between the measurement model and the observed
data.
The final model provides an excellent fit for the data across several key
indices. The RMR index was equal to .049, below the acceptable threshold of
.05 for a good fit, as is RMSEA (.013) (Browne and Cudeck, 1993). The CFI
(.99) exceeds the conventionally accepted threshold of .95. In essence, the
confirmatory analysis resulted in a model that, with minor exceptions, is vir-
tually identical to the model that emerged in the exploratory work.
In examining the relationship between each variable and its associated
dimension, none of the critical ratios (t values) is less than 2.0. Hence, it can
be concluded that each of the dimensions has a significant effect on its associ-
ated variable (p .05). In addition, with few exceptions, all standardized
regression weights are relatively strong, i.e. > .70.
Determinants of Social Capital
Pride and Identity Given that only two questions were used to measure this
dimension, it is not viable to perform a factor analysis on the data. The inter-
correlation, however, between the two variables is significant for both Ghana
(r = .42; p < .0001) and Uganda (r = .67; p < .0001).
Political Engagement Results of analyses conducted on both the Ghana and
Uganda data sets are presented in Table 12. While we include this dimension
in this section on determinants, we argue later in the article for treating politi-
cal engagement as an outcome of social capital.
The factor structures for both countries are remarkably similar. The first
two variables, for example, load on factor 2 in both Ghana and Uganda; two
of the last three variables load on factor 1 in both solutions. While the variable
80 Current Sociology Vol. 49 No. 2
Table 12 Political Engagement – Ghana and Uganda
Ghana Ghana Uganda Uganda
Variable Factor 1 Factor 2 Factor 1 Factor 2
*Number of times attended town meeting .458 .996
*Number of times contacted politician .811 .388
Number of times joined protest
*Voted in last local election .769 .794
**Voted in last state election .760 .891
Would you vote for a candidate from a
different ethnic group? .438
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 80
related to frequency of contacting a politician loads relatively weakly on factor
2 in the Uganda solution, its high loading in Ghana suggest its inclusion in
subsequent research. Protest activity fails to load at a value greater than .30,
and thus is not recommended for future survey administrations. The last vari-
able listed, while loading at .438 in the Uganda solution, should also be elim-
inated principally because it fails to discriminate well. For example, in Ghana
only 15 percent of those queried responded that they would not vote for a
candidate from another ethnic group/caste/tribe/race/religion or linguistic
group. In Uganda, those responding similarly were approximately 10 percent.
It is possible, however, that the lack of variability in response is a function of
the binary scale used (yes/no). It might, therefore, be worthwhile to rephrase
the question using a scale with multiple response options.
Communication Table 13 shows the factor structure for communication
variables in Ghana. The first factor subsumes variables related to media; the
second largely addresses proximity variables, while the third is related exclu-
sively to roads. All variables are significantly intercorrelated (p < .0001). The
amount of variance accounted for by the first two variables is, respectively,
19.8 percent and 18.1 percent. The third accounts for 11.7 percent of the vari-
ance in the solution.
Outcome Variables
In this section, we consider several outcome dimensions of social capital: for
instance, quality of government; honesty; and peace, crime and safety. Other
Narayan and Cassidy: Measuring Social Capital 81
Table 13 Communication Variables (Rotated Factor Matrix – Ghana)
Factor
———————————
123
*How often do you listen to the radio? .946
**Do you have a radio in your household or have
access to a radio somewhere else nearby? .701
**How often do you read a daily newspaper or have
one read to you? .471
*How close is the nearest telephone? .770
**How close is the nearest working post office? .668
*How far is your village/neighborhood from a big
town center or city? .539
Do you have or have access to a television? .323 .365
*How would you rate the general condition of the
roads in your area? .796
**Is your household easily accessible by roads? .561
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 81
outcomes, such as self-reported happiness, are not addressed as dimensions
but are examined separately in the next section on the predictive validity of
the questionnaire variables.
Quality of Government Of the variables included in Table 14, only the first
two are recommended for inclusion for subsequent research; both load with
high values on the first factor for both data sets. The third variable listed is
subsumed by the second factor in Ghana, which accounts for approximately
9 percent vs 31 percent associated with the first factor.
The fourth variable (time spent dealing with regulations and bureau-
cracy) was not asked in Uganda, and the strength and location of its loading
in the Ghana solution do not justify its sustained inclusion. The last variable
listed, while not loading above .30 in Ghana, does, however, load strongly in
Uganda and hence is suggested for inclusion. In addition, it is weakly and
insignificantly correlated with the other variables.
Honesty/Corruption Both the wording and the scale were changed between
the initial administration of the instrument in Ghana and its subsequent
administration in Uganda. Moreover, several of the authorities inquired
about (for instance property/land registration officials; traditional authori-
ties) were added or deleted in the two administrations.
Table 15 shows the results from the factor analysis performed on Ghana.
Drawing from the Uganda data as well (Table 16), the additional variables
marked with asterisks are also recommended or suggested for inclusion.
The pattern in the Uganda data in fact is intriguing, suggesting a
bifurcation in the factor structure potentially related to familiarity with or
proximity to the officials. Note in Table 16, for example, how the authorities
82 Current Sociology Vol. 49 No. 2
Table 14 Quality of Government – Ghana and Uganda (Rotated Factor
Matrix)
Factor
—————————
12
*If a household pays additional money is service delivered? .990;
.713
*Do you generally have to pay additional money to .727;
government to get things done? .821
How often do the rules, laws, etc. change without warning? .406 .490
Time spent dealing with government regulations and bureaucracy N/A .337
*Extent government takes your concerns into account when
making changes .956
Note: Ghana values are in plain type; Uganda values are in bold type.
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 82
with whom one might expect respondents to have greatest familiarity gener-
ally cluster in factor 1. The notable exception is the post office, which is also,
interestingly, negatively, albeit weakly, correlated with the other variables.
Competency Competency was measured in the GSCS by asking respon-
dents about their level of perceived competency of various officials. The
results of the two factor structures are shown in Table 17.
All variables recommended or suggested for future research are signifi-
cantly intercorrelated with the other measures so identified in both the Ghana
Narayan and Cassidy: Measuring Social Capital 83
Table 15 Honesty/Corruption – Ghana (Rotated Factor Matrix)
Factor
—————————
12
*Add. payment – judges/courts .792
**Add. payment – police .688
*Add. payment government medical systems .527 .418
Add. payment electric/power companies .524 .333
Add. payment – property/land registration .504 .312
Add. payment – license officials .477 .357
Add. payment – banks .414 .327
Add. payment – local government .325 .551
Add. payment post office .533
Add. payment – village/ward officials .503
*Add. payment – government school teachers .378 .477
Add. payment – tax authorities .422 .458
Add. payment – housing authorities .337 .418
Add. payment – local NGOs .341
Note: ‘Add. payment’ refers to moneys paid to government agencies to get things done.
Table 16 Honesty/Corruption – Uganda
Honesty of: Factor 1 Factor 2
Government medical system .780
Government school teachers .555
Local government officials .357
Village officials
Judges .652
License officials .508
Police .318 .478
Post office –.396
Traditional authorities .347
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 83
and Uganda data sets. While the judges/courts variable cross-loads in the
Ghana analysis, it is the highest loading variable in Uganda, loading solely on
factor 1. As such, it seems prudent to retain it in future research.
Peace, Crime and Safety The final outcome dimension we examine is peace,
crime and safety. The results of the analyses for both countries are shown in
Table 18. What can be derived from Table 18? First, the patterns are remark-
ably consistent across the two samples. The first three items are significantly
intercorrelated and obviously measure the same construct. Hence, the second
and third items are tagged as suggested variables. Second, the fourth variable
(confidence in the government for protection) is negatively and significantly
correlated with the first three variables in the Ghana data. Its recommendation
84 Current Sociology Vol. 49 No. 2
Table 17 Competency – Ghana and Uganda
Competency of: Factor 1 Factor 2
*Police .605
.605
**Local government .560
.467 .363
*Government school teachers .396 .724
**License officials .386
.613
Village/community officials .348 .381
Post office .378 .585
*Judges/courts .370 .489
.741
Note: Ghana values are in plain type; Uganda values are in bold type.
Table 18 Peace, Crime and Safety
Variable Factor 1 Factor 2
*Safety of neighborhood .935
.821
**Safety of household .773
.800
**Safety walking down street .549
.757
*Confidence in government for protection –.308 .951
.588
*Victim of non-violent crimes .900
**Victim of violent crimes .843
Note: Ghana values are in plain type; Uganda values are in bold type.
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 84
for inclusion is predicated on the strength of its loading on the second factor
in Ghana. Finally, while neither of the last two variables listed load meaning-
fully in the Ghana analysis, they do in the Uganda analysis, likely a function
of the substantial differences between the two samples. Since, based on the
Uganda results, they both measure the same construct and since both are sig-
nificantly interrelated in Uganda (r= .763; p.0001), either may be employed.
The variable related to non-violent crimes is recommended because of its
slightly higher loading and because it is anticipated that in general more
persons will be affected by non-violent rather than violent crimes.
Predictive Validity
If the variables recommended as measures of social capital are of intrinsic
worth, they should be able to predict a substantial amount of the variance in
key outcome variables. In this section, we attend to this topic.
The recommended GSCS social capital variables perform admirably in
explaining variance in a number of outcome variables at the community level.
They do less well predicting at an individual household level. For example,
in Ghana, approximately 13 percent of the variance in the measure ‘How
happy are you?’ was explained in a stepwise regression that eliminated all but
a few social capital variables. In Uganda, only 6.7 percent of the variance was
explained – and by a different set of predictors. The story, however, is quite
different when the households are aggregated at the EA level. Unfortunately,
we were unable to perform this level of analysis on the Uganda data because
of the small number of EAs represented. We limit our results, then, to the
Ghana data set, with 70 EAs represented in the analysis. Table 19 depicts the
variance associated with each factor. Note that the factor labels reflect some
overlapping.
The EA level, eight-factor solution, is, despite some minor differences,
Narayan and Cassidy: Measuring Social Capital 85
Table 19 Ghana EA Factor Structure Variance
Cumulative
Factor Variance (%) Variance (%)
1. Trust 16.1 16.1
2. Group characteristics 15.2 31.3
3. Volunteerism and group characteristics 10.4 41.7
4. Everyday sociability – crafts and recreation 10.0 51.7
5. Everyday sociability – eating meals and chores 6.8 58.5
6. Togetherness 5.6 64.1
7. Everyday sociability and trust 4.1 68.2
8. Everyday sociability – frequency of recreational activities 3.5 71.6
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 85
86 Current Sociology Vol. 49 No. 2
Table 20 Regression Results, Ghana – EA Level
Adjusted
Dependent Variable R
2
Significant Predictors
Subjective well-being
How satisfied are you with your life? .348 Are people who don’t volunteer
criticized?
Expectation of people volunteering
Trust in politicians
Optimism about future .597 Criticism of people who don’t
volunteer
Trust – same club members
Decision-making index
Frequency people visit home
Frequency – doing chores with others
Same gender heterogeneity index
Frequency of volunteerism
Frequency – arts/crafts
Perceived ability to survive a crisis .329 Expected that people will
(financial, health, etc.) volunteer?
Who visits you at home?
Same religion heterogeneity index
Frequency – people visit you at home
Honesty and corruption
Honesty of judges .532 Trust – family members
Frequency eating outside home
How well people in community get
along
Same neighborhood heterogeneity
index
Same tribe heterogeneity index
Honesty of police .259 Ask neighbors for help if sick
Participation index
Frequency – eating meals outside
home
Same neighborhood heterogeneity
index
Peace, crime and safety
Safety of household from crime and .499 Feeling of togetherness
violence Frequency – play cards/games with
others
Trust – family members
Competency
Competency of police .389 Trust – politicians
Feelings of togetherness
With whom do you eat outside
home?
Competency of government medical .357 Trust – politicians
system Feelings of togetherness
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 86
strongly similar to the household level analysis. Both, for example, have as
their first two factors group characteristics and trust. Both cluster similar
variables in similar ways.
Table 20 summarizes selected results from a number of multiple regres-
sion analyses.
Narayan and Cassidy: Measuring Social Capital 87
Table 20 continued
Adjusted
Dependent Variable R
2
Significant Predictors
Quality of government
Do you have to pay money to receive .558 Criticism for not volunteering
services How well people get along
Trust – people in village
Same neighborhood heterogeneity
index
Ask for help if child sick
Frequency – doing arts/crafts, etc.
If a household pays additional money, .659 Criticism for not volunteering
are services delivered? Expectation to volunteer
Togetherness
Ask for help if child sick
Have you volunteered?
Frequency – doing chores
Frequency – people visiting you at
home
Trust in politicians
Extent government takes concerns .436 Trust in politicians
into account Trust in business owners
Ask for help if child sick
Same tribe heterogeneity index
Frequency – chores outside house
Political engagement
Vote in last local election .869 Participation index
Volunteered?
With whom do you spend time doing
chores?
Trust – fellow club members
Same education/income
heterogeneity index
Same neighborhood heterogeneity
index
EA .719 Trust – politicians
Number of memberships
Expectation of volunteerism
Time spent doing chores
Trust– same tribe
How well people get along
Togetherness
With whom do you eat meals?
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 87
The results are impressive across virtually all variables representing
major outcome dimensions. What predicts political involvement, perceived
competency and honesty of government institutions, feelings of safety and
so forth are the fundamental components of social capital: trust, social
interaction, group involvement and affiliation, volunteerism, for instance.
The results also have face validity in the best sense of the term – that is, they
are sensible. For example, the two variables concerning competency of
government institutions reflected on p. 87 are predicted in part by trust in
politicians. The variable related to perceived safety of the household is pre-
dicted by variables related to the cohesion of the community, trust in family
and social engagement with others in the community. Another noteworthy
finding is the high variance explained in voting behavior (R
2
= .869). And
what predicts the behavior: participation, volunteerism, trust and group
characteristics. The key element across all outcome variables is basically
involvement with others in a meaningful way: namely, social capital. The last
row in Table 20 predicts the EA, based on several key predictors. The vari-
ance accounted for is high (R
2
.719), and the predictors are not surprising.
What happens, however, when tangible assets are added to the list of
independent variables? In truth, there is some positive influence, but gener-
ally not great. For example, perceived honesty of judges and courts increases
from the .532 noted in Table 20 to .625, and cattle and motorcycle owner-
ship enter as significant predictors. The adjusted R
2
value for satisfaction
with life increases from .348 to .512 when respondents own sewing machines
and cattle, but these changes are not surprising. Unsurprising as well is the
change in R
2
when assets are included in the regression conducted to predict
whether policies change without involvement of those affected. Without
assets in independent variables, the adjusted R
2
is .436. Including assets
increases the R
2
to .594 and reflects ownership of a sewing machine and
refrigerator. One might reasonably speculate that ownership of tangible
assets might increase one’s sense of self-efficacy. Conversely, there is no
change in the R
2
or the significant predictors for the variable related to per-
ceived power to change one’s life (R
2
= .245). So also, there is no change in
the predictors for safety or in the adjusted R
2
. In other words, tangible assets
may have some salutary influence on the predicted variable, but the most
important predictors, consistently, are those related to the social capital a
community possesses.
Including Determinants
We noted earlier that some variables, such as pride and identity, may be con-
sidered determinants or outcomes of social capital. When treated as out-
comes, the results are quite interesting, as shown in Table 21.
The variable contributing most to a sense of pride is a social interaction
variable. Cattle and bicycle ownership come into play, but six of the eight
88 Current Sociology Vol. 49 No. 2
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 88
significant predictors are measures of social capital, not economic assets. A
similar pattern emerges in regard to identity. The data support that respon-
dents’ identity is a function principally of social capital variables: everyday
sociability and trust.
Determinants as Predictors We also created nine indices of social capital
(one for each of the eight factors and one cumulative index), based on the EA
level Ghana data. Each index was weighted in relation to the percentage of
variance accounted for by the associated factor. Factor 1, which accounted
for the largest percentage of the variance (16 percent), was set at unity. Factor
2, accounting for 15 percent of the variance, was given a weight of .93, and
so on. A regression was then conducted with each of the eight weighted
indices. When regressed against EA, the variance accounted for was 66.5
percent, somewhat less than the 71.9 percent documented above when the
individual variables were used. This is not surprising: there is less predictive
precision in an index than there is in the individual variables of which it is
comprised. All indices were, however, included as significant.
We also examined determinants as predictors of social capital. An index
of total social capital, comprising the weighted subindices discussed previ-
ously, was generated. Included in the stepwise regression were all communi-
cation variables, pride and identity variables and political engagement
variables. An adjusted R
2
of .671 was calculated with the following variables
all significant predictors (p .05): voting in last state election; proximity to
nearest telephone; sense of identity; and willingness to vote for a candidate
outside the respondent’s ethnic group, race and so on. These results suggest
that political engagement may indeed be a meaningful determinant of social
capital, as well as the individual’s sense of identity. That only one of the
Narayan and Cassidy: Measuring Social Capital 89
Table 21 Pride and Identity as Outcome Variables
Pride and Identity Adjusted R
2
Significant Predictors
Pride .648 With whom do you play games?
Cattle ownership
With whom do you do arts/crafts?
Trust – own tribe
Bicycle ownership
Same gender heterogeneity index
Same family heterogeneity index
Trust – family
Identity .528 Who visits you at home?
Same tribe heterogeneity index
Bicycle ownership
Trust – family members
Frequency with which you do arts/crafts
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 89
communication variables emerged as a significant predictor is somewhat sur-
prising. However, the results do provide some empirical justification for
defining communication as a determinant of social capital. In addition, it is
conceptually appealing to posit that communication nourishes social capital.
Internal Homogeneity Another measure of the adequacy of the items is the
overall reliability of the item set. As such, additional analyses were performed
to assess the internal consistency of the questions used in the factor analysis.
Cronbach’s alpha values for those variables included in the final Ghana
models indicate high reliability (household = .89; aggregate = .81), exceeding
the customary .70 threshold (Nunnally and Bernstein, 1994).
Summary and Conclusions
The reader may recall that Figure 1 represented our point of departure.
Figure 3 reflects our current thinking, influenced largely by the analyses con-
ducted in this investigation. What are the principal changes? First, we now
conjecture that political engagement is a consequence rather than a dimen-
sion of social capital. Second, we now believe that empowerment is better
90 Current Sociology Vol. 49 No. 2
Illustrative Proximate
Determinants
of Social Capital
Dimensions of
Social Capital
Illustrative Social, Political and
Economic Outcomes
of Social Capital
Group characteristics,
including memberships
in informal groups
and networks with
particular characteristics
Generalized norms
Togetherness
Everyday sociability
Neighborhood connections
Volunteerism
Trust
Empowerment
Communication
Government
competence
Government
honesty and
corruption
Quality of
government
Peace and
safety
Political
engagement
Figure 3 A Revised Measurement Framework
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 90
defined as a determinant of social capital than as an outcome. Third, com-
munity solidarity (togetherness), initially perceived as a determinant, we now
believe is better defined as a dimension of social capital. Finally, communi-
cation has been added as a determinant.
A reliable, valid and consistent set of measures advances theory develop-
ment and assists empirical research. Our principal goal was to develop such
a set of measures for use in subsequent investigations. While some measure-
ment issues persist (for example, if certain variables are better defined as
determinants, outcomes or dimensions of social capital), we believe that
some, at least partially, have been resolved. We have demonstrated, for
example, the relative stability of the dimensions and the measures at differ-
ent levels of aggregation and across substantially divergent populations.
Additional work in the measurement of social capital is obviously needed.
For example, it would be worthwhile to use the dimensions identified for the
prediction of direct rather than self-reported measures. It would be worth-
while to assess the robustness of the measures in other societies and environ-
ments. We acknowledge, as well, the inherent limitations of a cross-sectional
study and recommend focused, longitudinal research using the suggested and
recommended measures.
A secondary goal of the article was to present findings on social capital
in two African republics. From the data in these two different samples we
find evidence that supports the importance of social capital in societal well-
being. Optimism, satisfaction with life, perceptions of government insti-
tutions and political involvement all stem in large degree from the
fundamental dimensions of social capital. Trust, community involvement,
social engagement, volunteerism and so forth appear to influence, positively
or negatively, attitudes and behaviors. Going beyond the data, we speculate
that varying amounts of social capital in Ghana and Uganda might partially
explain the economic disparities between the communities examined.
Empirical verification of this, however, awaits future research.
Narayan and Cassidy: Measuring Social Capital 91
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 91
Appendix: Recommended and Suggested Items for Measuring
Social Capital, Determinants and Outcomes
Variables preceded by a * are recommended.
Variables preceded by ** are suggested.
Social Capital Measures – Group Characteristics
Variable/Comments Scale
* How many groups or organizations do you belong to? (Absolute frequency)
These could be religious groups, sports teams, or just
groups of people who get together regularly to do an
activity or tasks.
[If the specific types of groups are not of interest, a simple
question that asks about total number of groups is
sufficient. The question should, however, clarify for the
respondent the types of groups the respondent is asked to
consider when answering.]
* On the average, how much money, if any, do you (Absolute frequency)
contribute to the groups to which you belong in a
month?
[If the investigator is interested in distinguishing among
groups, (e.g. most important group, second most
important group) the question should be rephrased
accordingly. In this article, indices related to money
contributed, participation, etc. were additive.]
*On average, how often do you participate in the (Absolute frequency)
activities of the groups to which you belong in a month?
*To what extent do you participate in the group(s)’(s) 1. To a very small extent
decision-making? 2. To a small extent
3. Neither a small nor
large extent
4. To a large extent
5. To a very large extent
Thinking about the members of this group, would you Record response:
say that most are from the same: Yes
1. ** Neighborhood/village/community? No
2. * Family or kin group?
3. ** Tribe/caste/ethnic/linguistic group?
4. ** Religious group?
5. * Educational background and income level?
6. * Gender?
92 Current Sociology Vol. 49 No. 2
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 92
[We assume that the respondent is being asked about the
most important group. Should the researcher be
interested in differentiating among two or more groups,
response options would be provided for each. Indices
referred to in this article, (e.g. same gender heterogeneity
index) were additive across the three most important
groups identified by respondents. Questions with
slashes(/) indicate that the interviewer should use a single
term most relevant to the interview situation.]
Social Capital Measures – Generalized Norms
Variable/Comments Scale
* Generally speaking, would you say that you can’t be 1. You can’t be too
too careful in dealing with people, or that most people careful
can be trusted? 2.
[Only the anchors of the scale are labeled.] 3.
4. Most people can be
trusted
** Would you say that most of the time people are just 1. Are just looking out
looking out for themselves, or they are trying to be for themselves
helpful? 2.
[Only the anchors of the scale are labeled.] 3.
4. Try to be helpful
** Do you think that most people would try to take 1. Would take advantage
advantage of you if they got the chance, or would they of you
try to be fair? 2.
[Only the anchors of the scale are labeled.] 3.
4. Would try to be fair
Social Capital Measures – Togetherness
Variable/Comments Scale
* How well do people in your community/village/ 1. Not getting along at
neighborhood get along these days? Using a five-point all
scale where 1 means not getting along at all and 5 means 2. Not getting along
getting along very well, how well are people in your very well
community/village/neighborhood getting along? 3. So-so
[For this and other questions with standardized, multiple 4. Getting along quite
response options, respondents were shown cards with well
the response options when asked the questions.] 5. Getting along very
well
Narayan and Cassidy: Measuring Social Capital 93
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 93
* How would you rate the togetherness or feeling of 1. Not close at all
belonging in your neighborhood/village/community? 2. Not very close
[Again, use a five-point scale where 1 means people do not 3. So-so
feel close at all and 5 means that people feel very close 4. Somewhat close
to each other.] 5. Very close
Social Capital Measures – Everyday Sociability
Variable/Comments Scale
In addition to participating in group activities or
associations, people also do many activities informally
with others. How often do you do each of the following?
[Each set of questions (frequency and with whom) in this
section are preceded with a general branch question. For
example, ‘Do you get together with a usual group of
people to play cards, games, board games?’ Where
appropriate, these are documented below.]
* On the average, how often in a month do you get (Absolute frequency)
together with a group of people to do arts, crafts, or other
recreational activities?
[Branch question: ‘Do you get together with a group of
people to do arts, crafts, or other recreational activities?’]
** Who are these people with whom you do arts, crafts, 1. Family members or
or other recreational activities? friends
[Questions such as these were asked in an open-ended 2. Friends from the
fashion. The interviewer then recorded the response same caste/religion/
using a precoded response set.] ethnic/education/
wealth/gender group
3. Friends from different
caste/religion/ethnic/
education/ wealth/
gender groups
* Who are the people with whom you get together to (Same as above)
play cards, games, or board games?
** On average, how often do you get together with (Absolute frequency)
others to play cards, games, or board games?
* Who are the people with whom you spend time 1. Family members or
outside the household in other ways, such as doing friends
94 Current Sociology Vol. 49 No. 2
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 94
chores, shopping, talking, drinking, or just spending 2. Friends from the same
time together? caste/religion/ethnic/
[Branch question: ‘Do you spend time with people education/ wealth/
outside your household in other ways, such as doing gender group
chores, shopping, talking, drinking, or just spending 3. Friends from different
tome together?’] caste/religion/ethnic/
education/wealth/
gender groups
**Who are the people who visit you at home? (Same as above)
[Branch question: ‘Do people visit you at your home?’]
** Who are the people with whom you eat meals (Same as above)
outside the home?
[Branch question: ‘Do you eat meals with people
outside the home?’]
Social Capital Measures – Neighborhood Connections
Variable/Comments Scale
* On a scale from 1 to 5, where 1 is very unlikely and 5 1. Very unlikely
is very likely, how likely is it that you would ask your 2. Unlikely
neighbors to take care of your children for a few hours 3. Neither unlikely nor
if you were sick? likely
4. Likely
5. Very likely
* On a scale from 1 to 5, where 1 is very unlikely and 5 1. Very unlikely
is very likely, how likely is it that you would ask your 2. Unlikely
neighbors for help if you were sick? 3. Neither unlikely nor
likely
4. Likely
5. Very likely
Social Capital Measures – Volunteerism
Variable/Comments Scale
* In your community/neighborhood, it is generally 1. Strongly disagree
expected that people will volunteer or help in 2. Disagree
community activities. 3. Neither disagree nor
agree
4. Agree
5. Strongly agree
Narayan and Cassidy: Measuring Social Capital 95
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 95
* People who do not volunteer or participate in 1. Strongly disagree
community activities are likely to be criticized or fined. 2. Disagree
3. Neither disagree nor
agree
4. Agree
5. Strongly agree
** Most people in your community/neighborhood make 1. Strongly disagree
a fair contribution to community/neighborhood 2. Disagree
activities. 3. Neither disagree nor
agree
4. Agree
5. Strongly agree
** On average, how many times per month do you (Absolute frequency)
volunteer in community activities?
Social Capital Measures – Trust
Variable/Comments Scale
Now I want to ask you how much you trust different 1. To a very small extent
groups of people. On a scale from 1 to 5, where 1 means or not at all
‘to a very small extent’ and 5 means ‘to a very large 2. To a small extent
extent’, how much do you feel you can trust the people 3. Neither small nor
in each of the following groups? great extent
[Cards with response options were shown to respondents.] 4. To a great extent
5. To a very great extent
6. N/A – No such group
* People in your tribe/caste/race/religion/ or ethnic group? (Same scale as above)
** People in other tribes/castes/race/religion/or ethnic (Same scale as above)
groups?
**People in your village/neighborhood? (Same scale as above)
** People who belong to the same clubs, organizations, (Same scale as above)
or groups as you?
** The business owners and traders you buy things from (Same scale as above)
or do business with?
* Politicians? (Same scale as above)
96 Current Sociology Vol. 49 No. 2
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 96
* People in your family? (Same scale as above)
* Government service providers (education, health, (Same scale as above)
electricity, water, etc.)?
* Local/municipal government? (Same scale as above)
* Judges/courts/police? (Same scale as above)
Determinant Measures – Pride and Identity
Variable/Comments Scale
* We have talked to many people about their sense of 1. Very confused about
identity – that is, who they are, where they come from, who I am
and their sense of belonging. Using a five-point scale, 2. Somewhat confused
where 1 means having a very weak sense of identity and about who I am
5 means having a very strong sense of identity, how 3. Neither clear nor
would you rate your own sense of identity? unclear about who
[As discussed in the text, the limit of two variables for this I am
dimension precluded conducting a data reduction analysis. 4. Somewhat clear about
While these two items are highly intercorrelated, the who I am
researcher may wish to consider adding one or more items 5. Very clear about
to measure this dimension and improve overall reliability who I am
of the subscale.]
We are also interested in how proud you are of who you 1. Very ashamed
are and the larger group to which you belong. Using a 2. Ashamed
scale where 1 means very ashamed and 5 means very 3. Neither proud nor
proud, how would you rate your own sense of pride? ashamed
4. Proud
5. Very proud
Determinant Measures – Communication
Variable/Comments Scale
* How often, if at all, do you listen to the radio? 1. Never
2. Monthly
3. Weekly
4. Daily
** Do you have a radio in your household, or have access Yes
to a radio somewhere else nearby? No
Narayan and Cassidy: Measuring Social Capital 97
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 97
** On average, how many times per week, if at all, do (Absolute frequency)
you read a daily newspaper or have one read to you?
* How long does it take you to get to the nearest 1. In or near the
telephone? household
2. 1–10 minutes
3. 11–20 minutes
4. 21–30 minutes
5. 31–40 minutes
6. 41–50 minutes
7. 51–60 minutes
8. More than an hour
**How long does it generally take you to reach the 1. 1–10 minutes
nearest working post office? 2. 11–20 minutes
3. 21–30 minutes
4. 31–40 minutes
5. 41–50 minutes
6. 51–60 minutes
7. More than an hour
* About how far is your village/neighborhood from a 1. Household is in the
big town center or city? biggest town
center/city in area
2. Within 20 km
3. Within 50 km
4. Within 100 km
5. Within 200 km
6. More than 200 km
away
* How would you rate the general condition of roads in 1. Very bad
your area? 2. Somewhat bad
3. About average
4. Somewhat good
5. Very good
** Is your household easily accessible by roads all year 1. Yes, throughout the
long, only during certain seasons, or not at all? year
2. Yes, only during
certain seasons
3. No, not easily
accessible at all
98 Current Sociology Vol. 49 No. 2
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 98
Outcome Measures – Quality of Government
Variable/Comments Scale
* Households like yours generally have to pay some 1. Strongly disagree
additional money to the government from time to time 2. Disagree
to get things done. 3. Neither disagree nor
[An overall introduction is given to respondents indicating agree
that they will be read a series of statements and asked 4. Agree
how much they agree or disagree with each statement.] 5. Strongly agree
* If a household pays the required additional money to (Same as above)
the government officials, the service is delivered as agreed
or the problem is solved.
* The government takes into account concerns voiced (Same as above)
by you, your household, or people like you when making
changes in rules, laws, and policies that affect your lives.
Outcome Measures – Honesty and Corruption
Variable/Comments Scale
Sometimes it is common for households to pay additional
money to government agencies to get things done. As I
read the following list of agencies, please tell me, on a
scale from 1 to 5, where 1 is very dishonest and 5 is very
honest, how you would rate each of the following
government services:
* Judges/court system 1. Very dishonest
2. Dishonest
3. Neither dishonest nor
honest
4. Honest
5. Very honest
** Police (Same as above)
** Government medical system, clinics, hospitals (Same as above)
* Post office (Same as above)
* Government school teachers (Same as above)
Narayan and Cassidy: Measuring Social Capital 99
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 99
Outcome Measures – Competency
Variable/Comments Scale
Our lives are affected by government institutions. I am
now going to read a list of institutions. For each, please
tell me how competent you think each of the following is.
Use a scale where 1 is very incompetent, and 5 is very
competent.
* Police 1. Very incompetent/
inefficient
2. Somewhat
incompetent/inefficient
3. Average
competency/efficiency
4. Somewhat
competent/efficient
5. Very
competent/efficient
6. No contact
**Local government/municipality officials (Same as above)
* Government school teachers (Same as above)
** License officials (Same as above)
* Judges/court system (Same as above)
Outcome Measures – Peace, Crime and Safety
Variable/Comments Scale
* In general, how safe would you say your neighborhood 1. Very unsafe
is from crime and violence? 2. Unsafe
3. Neither unsafe nor
safe
4. Safe
5. Very safe
** In general, how safe would you say you and your (Same as above)
household are from crime and violence at home?
** How safe do you feel walking down your street after (Same as above)
dark?
100 Current Sociology Vol. 49 No. 2
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 100
* How much confidence do you have that government 1. Not confident at all
authorities can protect you, your household and your 2. Not confident
property from crime and violence? 3. So-so
4. Somewhat confident
5. Extremely confident
* In the past 12 months, how many times have you or (Absolute frequency)
anyone else in your household been the victim of
non-violent crime such as theft, robbery, or destruction
of property?
** In the past 12 months, how many times have you or (Absolute frequency)
anyone else in your household been the victim of a violent
crime such as physical assault or mugging?
Outcome Measures – Political Engagement
Variable/Comments Scale
In the past year, how many times, if ever, have you done
any of the following:
* Attended a town meeting, public hearing, or public (Absolute frequency)
affairs discussion group?
* Met, called, or sent a letter to a local politician? (Absolute frequency)
* Did you vote in the last local government/community Yes
election? No
N/A (no elections/
can’t vote)
** Did you vote in the last state/presidential election? (Same as above)
Notes
1 We are grateful to participants of the workshop; they included: Hans Blomkvist,
Uppsala University; Ashutosh Varshney, Columbia University; Stephen Knack,
American University; Thierry van Bastelaer and Satu Kahkone, IRIS, University
of Maryland; Dawn Crossland, Princeton Survey Research Associates; and Paul
Collier, Michael Walton, Brian Levy, Christiaan Grootaert and Ravi Kanbur,
World Bank.
2 The variables included in the final factor structures are not identical across the two
data sets. The primary reason for this reflects the iterative nature of factor analysis.
Narayan and Cassidy: Measuring Social Capital 101
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 101
For example, items with low communalities were eliminated, as were items with
relatively low factor loadings (i.e. < .30), and/or items cross-loading on two or
more factors. As such, some items eliminated from Ghana may have been retained
in the Uganda set and vice versa.
3 The indices are additive across the questions inquiring about key aspects
(frequency of participation, participation in decision-making and money
contributed) of the three groups most important to the respondents. Some
rescaling was performed on the Uganda data to permit comparisons with Ghana.
4 We use the terms ‘construct’, ‘factor’ and ‘dimension’ interchangeably.
102 Current Sociology Vol. 49 No. 2
07 Narayan (to/d/k) 30/7/01 3:21 pm Page 102
... The most general insights that show the background of trust as a concept are its relation to values, norms, or attitudes (Claridge, 2004;Bodor et al., 2020;UNDP, 2021). Previous studies (Markowitz & Shariff, 2012;Hornsey et al., 2016;Weber, 2016;Marshall et al., 2019) have shown that the forms and methods of analyzing trust depend greatly on the goals of researchers. Trust is assumed to be an important social construct (Cologna & Siegrist, 2020) that is applicable to various topics but is especially relevant when analyzing public attitudes toward issues of concern to society (e.g., climate change). ...
... Trust as an element of social capital (supplemented by the authors followingNarayan & Cassidy, 2001) ...
Article
Although there is a growing trend in scientific and political discussions about how people feel about climate change in the context of trust, adequate research in this area is lacking in Central Eastern Europe (CEE). This article aims to determine whether trust is an important factor for the perception of climate change by the inhabitants of selected CEE countries. European Social Survey(ESS) Round 10 data were used, and a multivariable binary logistic regression method was applied. This paper considers three different dimensions of climate change, concern, personal norms, and attribution, as well as a composite indicator reflecting climate attitudes (skeptic or proponent). Four different models are presented that differ only in terms of their dependent variables, while the independent variable (trust) and the control variables (socio-demographic variables) are the same in all the cases. The results show that both personal and institutional trust are very low in the selected countries. Institutional trust was significant in all four models, while personal trust was significant in the environmental norm and skepticism models. Greater trust in institutions decreases the odds of being worried about climate change and increases the odds of having feelings of personal responsibility for climate change reduction. An increased level of personal or institutional trust increases one’s chances of being in the climate proponent group. Of the socio-demographic variables examined only political left-right scale did not affect attitudes significantly.
... Social Capital Theory (SCT) posits that robust social networks, including bonding, bridging, and linking, significantly benefit sustainable development, especially in developing countries like Sri Lanka. Social Capital, encompassing relationships among family, friends, colleagues, and communities, alongside shared values and norms, promotes knowledge sharing and collective learning (Katungi, et al., 2008;Narayan & Cassidy, 2001;Coleman, 1988). Trust, reciprocity, and shared understanding within these networks are crucial for improving livelihoods. ...
... We did this by asking participants if they were in favor of or against paying "IDR 16,500" a month for better garbage management. Prior studies [27][28][29][30][31] showed that children's relationships with people outside the village are examples of characteristics of social capital. These characteristics are deeply embedded in Indonesian culture. ...
Article
Full-text available
Sustainability in environmental practices depend on better household waste management, and environmental policy planning needs to take into account the variables affecting residents’ willingness to pay (WTP) for waste management. This research aims to measure the amount of WTP to improve waste management at Tasikmalaya, Indonesia. With the goal of improving home waste management, this study explores how residents’ WTP is shaped by social capital, a multifaceted notion that includes social networks and trust as well as the socio-demographic aspects. To calculate residents’ WTP for waste management, the study used the contingent valuation technique (CVM). We investigated the effects of social capital on residents’ willingness to engage using a logistic regression mode. According to the findings, 64% of residents were prepared to make a monetary contribution toward the waste management process. The research results show that the WTP value is “IDR 16,500”. The important role that social capital plays in influencing WTP in the context of better household waste management is highlighted by this study’s result. By taking social capital’s effects and dimensions into account, it offers policymakers important insights for creating waste management policies that work better. This research contributes to the literature by highlighting the environmental and health risks of pollution caused by landfill.
... We utilized a logistic regression analysis [26] to determine the association between social capital and the willingness of households to enhance their waste management practices. The household's willingness, with 1 denoting agreement and 0 denoting disagreement, is the dependent variable in the model. ...
Article
This study looks at the connection between people’s willingness to pay (WTP) for waste management services anf their social capital. The research utilitizes a linier regression model to examine the relationship between social capital and people’ desire to interact, making use of contigent valiation, a technique useful for assessing financial prices and preferences for non-market environmental assets. With an average WTP of IDR 17,000 (USD 1,09), the result show that 58.2% of the community is willing to pay for garbage management. The study finds a positive correlation between social capital and WTP, suggesting that strong community bonds, high levels of trust, and external family connections enhance the likelihood of supporting sustainable waste management initiatives. In addition to sociodemographic traits, the WTP's ability to manage waste is also impacted. Legislators and waste management experts should take note of these findings, which emphasize the need of fostering social capital to boost public involvement and support for sustainable waste management projects. Building trust, promoting community involvement, and establishing shared environmental principles can all contribute to the development of an environment that is conducive to effective waste management techniques. This study contributes to the expanding body of information about the socioeconomic drivers of waste management practices by highlighting the crucial role that social capital plays in creating WTP for waste management solutions.
Article
Understanding the roles of tourism development in poverty reduction and livelihood capital through a microlens is essential to rural ethnic areas’ sustainable development. However, few assessment methods are available to unravel the complex relationships between these three facets, and gaps remain in measuring residents’ livelihood. Based on the sustainable livelihood approach, this paper first selected core components of the sustainable livelihood framework to construct an evaluation index system and then provided specific indicators and evaluative criteria. Taking villages in China’s Longji Rice Terraces as a sample case, this study employed objective approaches (e.g., the entropy weight method and gray relational analysis) to assess residents’ livelihood capital. The aim was to showcase the sustainability of this area’s livelihood through a microlens. Findings offer several important contributions. First, this study presents a reliable measurement development method and a set of evaluation criteria for sustainable livelihood in tourism contexts, including five forms of capital and 21 indicators. Second, local inhabitants were found to benefit from tourism development; their livelihood options were diversified, especially in terms of growing financial capital. Yet these benefits were not evenly distributed, with notable differences observed across communities. Third, farmers’ livelihood capability did not improve uniformly over time. These findings offer valuable insights for tourism and hospitality stakeholders and can help steer rural regions toward a path of prosperity and growth. Strategies for inspiring entrepreneurship should especially be considered. In closing, theoretical implications are discussed, and suggestions for future work are provided.
Article
Full-text available
Research Paper Social capital plays a vital role in strengthening community organizations by enhancing collective action, promoting resilience, and fostering sustainable development. This study systematically reviews existing literature to assess the role of social capital-comprising trust, networks, and civic engagement-in bolstering community organizations. The primary objective of the study is to consolidate knowledge on how bonding and bridging social capital interact to influence community organizations' structure, resilience, and operational effectiveness. Through a comprehensive analysis of secondary data from multiple databases, the study identifies the significant contributions of social capital to community organization outcomes, including improved decision-making, resource mobilization, and community cohesion. The findings suggest that while bonding social capital strengthens internal ties, bridging social capital enhances the organization's ability to access diverse resources and foster collaboration across different social groups. Additionally, the study highlights gaps in the literature, particularly in terms of measuring the impact of social capital on specific community organization outcomes and the underrepresentation of rural and marginalized communities.
Article
Full-text available
Pakistan’s evolution into a research-centric nation, recognized by Clarivate Analytics, suggests a transition towards a knowledge-based economy. However, the prevalence of academic dishonesty within Pakistan’s academic institutions raises concerns, potentially impeding the realization of its intellectual prowess. This research therefore examines the drivers of academic dishonesty in Pakistani universities, drawing on data collected from 457 students in four prominent public sector universities. Regression analysis of the dataset reveals several noteworthy findings. The study identifies that male gender, larger family size, risk preference, anxiety, and susceptibility to peer influence correlate positively with academic dishonesty. Conversely, students’ socio-economic status and their perception of diminished respect for dishonest behavior act as deterrents. Furthermore, the impact of detection on academic dishonesty exhibits a threshold effect, while personal integrity/morality and punitive measures appear to have limited impact on academic dishonesty. Consequently, this study underscores the importance of addressing socio-cultural and environmental factors as pivotal elements in combating unethical academic conduct within Pakistan.
Book
Full-text available
Baluran National Park as a conservation area has a wide variety of flora and fauna. But this time Baluran National Park to experience a variety of threats, interruptions and damage to ecosystems. Based on information from the Baluran office (2014) damage incurred includes: (i) Forest fires reached ± 946.75 ha, or about 4% of the total land area is due in large part caused by the irresponsible and not because of natural factors; (Ii) Activities ± 400 ha of land clearing for agriculture plants business; (Iii) timber theft, especially in resort Labuhan Peacock; (Iv) Pastoral wild is a problem that is quite prominent, especially in the area Karangtekok, Labuhan Merak, and Balanan with an area of 3,450 ha. Cattle grazing types, namely cows and goats with an average of 1,447 head of cattle per day; (Iv) Local Trasmigrasi Settlements (Translok) since 1976, covering an area of 57 ha in Pandean area, Wonorejo village; (V) illegal encroachment and the tilling of the soil; and (vi) hunting of wildlife by people with firearms, snares, poison and sap that often occur during the dry season. This study is a non-laboratory scale with the purpose of: (i) to analyze and explain the effect of each variable system of institutions, incentives, social capital ,, the socio-demographic, and social learning to the level of community participation; (Ii) to analyze and explain whether the public welfare mediates the effect of variable incentives on the level of community participation; (Iii) to analyze and explain whether the public welfare mediates the influence of social capital on the level of community participation. The study population includes. By using cluster sampling based group / region relating to the location of the existing research in Baluran National Park 60 respondents. It is based on a number of variables used in this study were 6 to 39 indicator variables by using Equation Model Sructural analysis.
Article
Sosyal sermaye, ekonomik gelişme ve kalkınmanın ön koşulu olarak kabul edilen ayrıca vergi uyumunun sağlanması için de üretilmesi gereken bir sermaye türüdür. Sosyal sermayenin unsurları arasında yer alan güven, sosyal ağlar ve sosyal normlar toplumsal işbirliğinin oluşturulmasında önemli argümanlardır. Toplumsal işbirliğinin en önemli göstergelerinden biri de vergi uyumudur. Ayrıca vergi uyumu unsurları arasında yer alan vergi adaleti, vergi bilinci ve vergi harcamaları vergi uyumunun sağlanmasında etkin rol üstlenen diğer bileşenlerdir. Çalışma ile amaçlanan sosyal sermaye ve vergi uyumu unsurları arasında yer alan vergi adaleti, vergi bilinci ve vergi harcamalarının vergi uyumu üzerindeki etkilerini ortaya koymaktır. Çalışmada vergi yükünü en fazla hisseden mükellef grubuna ulaşmak amaçlanmış, bu nedenle gelir vergisi kapsamında yer alan ticari kazanç ve serbest meslek kazancı elde eden mükellefler örneklem olarak belirlenmiştir. Çanakkale merkez ilçede ikamet eden 446 mükellef ile anket yapılarak analizler için gereken veriler elde edilmiş, Faktör Analizlerinin ardından Yapısal Eşitlik Modeli kurulmuştur. Veri analizlerinde IBM SPSS 27 ile SmartPLS hazır yazılımları kullanılmıştır. Model sonucunda genelleştirilmiş güvenin vergi uyumu üzerinde etkili olmadığı ancak stratejik güven, kurumsal güven ve sosyal normların vergi uyumunu artırdığı belirlenmiştir. Vergi adaleti ve vergi bilinci, vergi uyumunu artırırken vergi harcamalarının vergi uyumunu olumsuz etkilediği elde edilen diğer bulgulardır.
ResearchGate has not been able to resolve any references for this publication.