ARTICLE IN PRESS

Health & Place 15 (2009) 1071–1077

Contents lists available at ScienceDirect

Health & Place

1353-82

doi:10.1

� Corr

Univers

3N6. Te

E-m

journal homepage: www.elsevier.com/locate/healthplace

Not all social capital is good capital

Spencer Moore a,b,�, Mark Daniel b,c,d, Lise Gauvin b,d, Laurette Dube e

a School of Kinesiology and Health Studies, Queen’s University, 69 Union St. PEC Rm. 215, Queen’s University, Kingston, ON, Canada K7L 3N6b CRCHUM, Centre de Recherche du Centre Hospitalier de l’Universite de Montreal, Canadac School of Health Sciences, University of South Australia, Australiad Departement de medecine sociale et preventive, Universite de Montreal, Canadae Desautels Faculty of Management, McGill University, Canada

a r t i c l e i n f o

Article history:

Received 27 May 2008

Received in revised form

9 March 2009

Accepted 5 May 2009

Keywords:

Social capital

Networks

Neighbourhoods

Well-being

92/$ - see front matter & 2009 Elsevier Ltd. A

016/j.healthplace.2009.05.005

esponding author at: School of Kinesiology

ity, 69 Union St. PEC Rm. 215, Queen’s Univers

l.: +1613 533 6000x78667; fax: +1613 533 20

ail address: mooresp@queensu.ca (S. Moore).

a b s t r a c t

Limited empirical research on social capital has investigated the potential downside of social capital on

health and well-being. We hypothesized that social capital and mastery might vary according to

education with lower-educated persons experiencing fewer advantages. This study used a stratified

cluster design to recruit a volunteer sample of 332 adult residents from 7 metropolitan census tracts.

The survey included a position generator to collect social capital network data. Generalized estimating

equations were used to account for the clustering of respondents in census tracts. Results indicated a

differential association between individual social capital and mastery according to educational

attainment. Among persons with a high school degree or more, higher social capital was associated

with a higher sense of mastery; among less-educated persons, higher individual social capital was

associated with lower mastery. Differences in the pathways by which lower- and upper-educated

groups access social capital may play a role in social capital’s negative association with psychological

well-being.

& 2009 Elsevier Ltd. All rights reserved.

Introduction

Social capital can be defined as the resources to whichindividuals or groups have access through their social relation-ships (Bourdieu, 1986; Portes and Sensenbrenner, 1993; Portes,1998). In early essays on social capital, Alejandro Portes cautionedagainst the portrayal of social capital as wholly beneficial with nosignificant downside (Portes and Sensenbrenner, 1993) ‘‘Socia-bility,’’ Portes (1998, p. 18) remarked, ‘‘cuts both ways’’. Socialconnections, which may lead to beneficial outcomes for someindividuals or groups, may lead to detrimental outcomes forothers. On the positive side, social capital can provide benefitssuch as familial support, social control, and privileged access toeconomic resources (Portes and Sensenbrenner, 1993). On thenegative side, social capital can restrict individual opportunitiesand freedoms or result in excessive claims and obligations beingplaced on a person. Who you know may tip the scales balancingthe positive and negative consequences of social integration(Mirowsky and Ross, 2003), but social factors, such as gender, age,or socioeconomic status (SES), can affect who you have an

ll rights reserved.

and Health Studies, Queen’s

ity, Kingston, ON, Canada K7L

09.

opportunity to know and the types of benefits that might accruethrough those connections.

Empirical research has been slow, however, to examine howthe potential advantages or disadvantages in health or well-beingdue to social capital might vary across social groups. Rojas andCarlson (2006), for example, found that the association of socialcapital with self-rated health in Russia was modified by education.Within impoverished Baltimore neighbourhoods, Caughy et al.(2003) showed that children whose parents had few neighbour-hood social connections had lower levels of behavioural problemsthan the children whose parents had more social connections.Carpiano (2007) found higher levels of neighbourhood socialsupport associated with higher likelihoods of smoking and bingedrinking. In other words, social connections do not in themselvesresult in health benefits; instead, the content of social relations,including the resources available in those connections, may play acritical role in how social capital impacts health and well-being.

Given this important yet under-examined issue, the aims ofthis study were threefold: (1) examine quantitatively whether ornot there were disparities in social capital according to educa-tional attainment among Montreal adults, (2) assess if highereducated groups used different network pathways in accessingsocial capital than those used by lower-educated groups, e.g,strong (family and friends) versus weak (acquaintances) ties, and(3) examine if educational attainment modified the association ofsocial capital with mastery. First, following the work of Bourdieu,

ARTICLE IN PRESS

S. Moore et al. / Health & Place 15 (2009) 1071–10771072

we hypothesized that like other forms of capital in society, socialcapital was unequally distributed across educational attainmentlevels. Second, following the work of sociologists such as Lin andGranovetter, we hypothesized that (1) a person’s weak, acquain-tance relationships were an invaluable source of social capital,particularly for those resources that could be used for instru-mental purposes, e.g., job attainment, and (2) that persons withhigher educational attainment had greater access to social capitalthrough their weak, acquaintance relationships. Third, althoughthe creation and maintenance of social connections can provideaccess to the resources of others, such actions can also be a drainon one’s own resources and time. We hypothesized that high SESgroups had more abundant personal resources to buffer againstthe greater number of obligations that come with more extensivesocial connections. In terms of mastery, high SES individuals withmore extensive social connections may view those connections ascontributing to a greater sense of control, whereas low SESindividuals with few resources but more extensive social connec-tions may view those connections as reducing their sense ofcontrol. We selected mastery as a particularly relevant outcomeon which to test these hypotheses since mastery has beenpostulated as lying along an indirect-cognitive path linking socialenvironmental characteristics to cardiometabolic outcomes (Da-niel et al., 2008). Mastery has been defined as the extent to whichone sees one’s life chances as being under one’s own control(Pearlin and Schooler, 1978), and is considered an importantdimension of psychological well-being and distress (Pearlin andSchooler, 1978; Mirowsky and Ross, 1986, 2003). Mastery is alsoassociated with positive health behaviour (Daniel et al., 2004,2006) and is an important health outcome itself (Pampalon et al.,2007).

In examining the association among social capital, education,and mastery, we sought to contribute to research on the potentialdownside of social capital for psychological well-being. In contrastto conventional indicators of social capital using trust andparticipation, we collected network data and constructed networkmeasures of individual social capital using a position generatorinstrument (Lin, 2001). The position generator measures anindividual’s social capital by assessing respondents’ ties topersons working in specific types of occupations and if thosepersons are family, friends, or acquaintances. Family and friendrelationships represent strong ties, whereas acquaintances repre-sent weak ties (Lin, 2001). There has been increased recognition ofthe importance of network data for measuring and assessing theimportance of social capital for well-being and health (Moore etal., 2005; Kawachi, 2006; Valente et al., 2007). One of theadvantages that formal network measures provide for the study ofsocial capital and health is greater attention to the types ofresources to which individuals or groups have access and theability to discriminate between the role of strong (e.g., family andfriends) and weak (e.g., acquaintances) ties in gaining access.

Methods

Sampling and recruitment procedures

Data for this study were drawn from the 2006 MontrealNeighbourhood Survey of Lifestyle and Health (MNSLH); theMNSLH was designed to collect self-reported and objectivebiological markers of health status. The MNSLH used a stratifiedcluster sampling design with clusters consisting of MontrealIsland census tracts (n ¼ 521); these clusters were stratified into(1) tertiles according to high, medium, and low socioeconomicstatus and (2) divisions of either predominantly French- orEnglish-speaking tracts based on the percentage of persons

reporting French or English spoken at home. French and Englishare the two official languages of Canada. Seven census tracts wererandomly selected from this stratified cluster design: one tractfrom each cell, with one additional medium-SES French tract lateradded to help augment sample size deficiencies. The individual-level sample was restricted to persons 18–55 years old. Recruiterscanvassed 100% of the census tract area distributing informationalmaterials on the study to households in those areas. Follow-upvisits by recruiters were used to enlist individual volunteers intothe study. Potential participants were given the opportunity tocomplete the socio-demographic, lifestyle, and self-reportedhealth questions either by telephone or online and schedule anurse’s visit at which time they would also return the self-administered social connectivity module of the questionnaire. Theunique recruitment protocol was implemented to facilitateobtaining as representative a sample of Montreal residents aspossible, and to reduce overall respondent burden often asso-ciated with population-based surveys and the collection ofbiological data using nurse home visits. Of 374 individualscontacted for a home visit, a total of 332 (89%) completed thesocial capital module. Compared to Canada census data of theselected census tracts, 2-sided exact binomial probability testsshowed that overall the MNS sample over-represented individualswho had a Bachelor’s degree, and those born outside of Canada;higher income and married individuals were over-represented in3 census tracts. There were no differences in gender or duration ofresidence in a given census tract. Four observations were excludedfrom analyses due to missing data on study covariates. Thepresent study is based on data from 328 respondents.

Measures

Dependent variable

We used Pearlin and Schooler’s (1978) original 7-item masteryscale with 5-point Likert categories ranging from stronglydisagree to strongly agree. Sample items included ‘‘Sometimes Ifeel that I’m being pushed around in life,’’ and ‘‘Sometimes I feelhelpless in dealing with the problems of life.’’ Five of the sevenitems, including the two previous examples, were reverse codedto ensure that higher scores on the scale reflected higher mastery.Respondents’ scores could range from 7 to 35. Mastery wasnormally distributed.

Independent variables

Individual social capital

We used a four-stage process to develop an individual socialcapital score for each respondent in the sample. First, wedeveloped a position generator instrument for the householdquestionnaire. Sixteen occupations were selected from a listing of90 occupations that had been ranked according to gender-neutraljob prestige scores within Canada (Goyder et al., 2003). To selectthe 16 occupations, we divided the Goyder et al. list into octilesranging from the highest to lowest prestige occupations. Fromeach octile in the list, two occupations were randomly selected.These 16 occupations were then randomly listed on the positiongenerator. The mean prestige value for the original list was 53.0with a range of values from 9.1 (squeegee kid) to 92.7 (physician).The mean prestige value for the study’s 16-item positiongenerator was 55.0 with a range from 33.6 (janitor) to 92.7(physician).

Second, rather than simply asking respondents if they knewsomeone on a first name basis with a given occupation, we asked

ARTICLE IN PRESS

1 Statistics Canada provides values for low income cutoffs only and does not

define middle- and high-income classifications. Therefore, we conducted analyses

to examine if the middle- and high-income distinction was sensitive to the original

cutoff values assigned. Low income was not significant when the cutoff between

middle and high was reduced to 3.00 from the original value of 4.00. We

triangulated this determination by examining the income-to-household size ratio

as a continuous variable. The association between income and mastery was

borderline significant at p ¼ 0.055. We therefore retained our original middle/high

cutoff value of 4.00. In no instance did changes in the income variable alter our

findings on social capital, education, and mastery.

S. Moore et al. / Health & Place 15 (2009) 1071–1077 1073

respondents to indicate the number of persons they knew whowere relatives, friends, or acquaintances. The distinction betweenfriends and acquaintances was left for respondents to decide. Toreduce potential response bias in respondent’s recall of the totalcount, we later dichotomized respondents’ answers to indicatesimply if they reported knowing a relative, friend, or acquaintancewith each occupation. Respondents were identified as havingaccess to an occupation if they indicated knowing a relative,friend, or acquaintance with the specific occupation. Cronbach’salpha was calculated to assess the internal consistency of the16-item instrument.

Third, after each respondent’s occupational connections wereidentified, we used the prestige value associated with eachoccupation to calculate three separate social capital indices: (1)upper reachability, (2) diversity or extensity, and (3) range. Upperreachability represents the highest prestige occupation accessed;diversity is the number of different occupations accessed; andrange is the difference between the highest and lowest occupationaccessed. The three indices were correlated: the highest correla-tions were between range and diversity (r ¼ 0.76, po0.001) andrange and upper reachability (r ¼ 0.76, po0.001), and lastlybetween upper reachability and diversity (r ¼ 0.59, po0.001).Fourth, given the objective to have a multidimensional measure ofsocial capital and the high correlation present among the separateindices, we created a linear component score using the optimallyweighted sum of the three. The weights were derived from aprincipal components analysis. This procedure has been con-ducted in other applications of social capital measures based onuse of the position generator instrument (Lin, 2001). In ancillaryanalyses, we also evaluated if the relations between mastery andthe three separate indices were similar to that of the compositesocial capital score.

Educational attainment

Respondents indicated their level of educational attainmentaccording to nine distinct categories ranging from ‘‘no degree,certificate or diploma’’ to ‘‘earned doctorate.’’ For our educationalattainment indicator, we reclassified the responses into threelevels: (1) less than high school, (2) high school or equivalent, and(3) more than high school, which included earning a tradecertificate on up to a doctorate.

Covariates

Covariates included income, age, gender, primary householdlanguage, marital or common-law relationship status, Canadian-born status, and residential duration. For income, respondentsidentified their income level from a list of 9 options, which rangedfrom ‘‘under $10,000 per year’’ to ‘‘over $200,000 per year.’’ Weconstructed an ordinal indicator of low, middle, and upper incomeadjusting for reported household size in the following manner.First, using the 9 income categories, we assigned each respondentan income value based on the maximum value for that category,e.g., ‘‘under 10,000 per year’’ was assigned a value of $9999. Theupper income category was assigned a value of $250,000 forcalculation purposes. Second, using reported household size, weassigned each respondent a low-income cutoff value based onhousehold size and the Statistics Canada based low-incomecutoffs for 2006. The cutoffs used are those assigned by StatisticsCanada for ‘‘before tax values for urban areas with more than500,000 persons.’’ (Statistics Canada, 2008). For example, a oneperson household has a value of $21,199 whereas a 3 personhousehold has a value of $32,446. Third, respondent’s assignedincome was divided by the low-income cutoff value. If the

resulting value was less than 1.00, the respondent was classifiedas ‘‘low income.’’ Values between 1.00 and 3.99 were classified asmiddle income. Respondents with a final score of 4 or more wereclassified as high income.1

Age was treated as a continuous variable. Primary householdlanguage was classified into French, English, or other. Respondentsindicated if they were male or female, and if they were in a maritalor common-law relationship, or not. Foreign-born status wasdetermined based on whether respondents reported being borninside or outside of Canada. Residential duration was based on theperiod of time that respondents reported living at their currentresidence.

Statistical analysis methods

Pairwise one-way ANOVA tests with Bonferroni correctionswere conducted to identify significant differences in upperreachability, diversity, and range scores among education groups.Second, generalized estimating equations (GEE) were used tomodel the association of individual social capital with mastery.This enabled accounting for the clustering of individuals within agiven census tract (i.e., ICC was estimated at 6.99% thuswarranting control for clustering). We used the ‘‘PROC GENMOD’’module in SAS, specifying a normal distribution with identity linkand an exchangeable correlation structure. GEE constructs a‘‘working correlation matrix’’ to account for the correlation amongrespondents within census tracts (Horton and Lipsitz, 1999); theexchangeable correlation structure is well suited to clusteredobservational data (Ying and Liu, 2006).

Continuous variables were standardized before analysis.Standardized coefficients represent the change in the standarddeviation units of mastery that occurs for each standard deviationunit change in predictors, holding all other variables constant.Four models were analysed. The first assessed the bivariateassociation of individual social capital with mastery. The secondassessed the association of individual social capital and educationwith mastery. The third examined the association of individualsocial capital and education with mastery after adjusting for studycovariates. The fourth and final model examined the interactionbetween individual social capital and education in their effects onmastery, adjusting for covariates.

Results

Table 1 provides descriptive information on the study sample,including educational attainment and covariates. The 7-itemmastery scale had a standardized Cronbach’s alpha of 0.70, andthe 16-item position generator instrument had a Cronbach alphaof 0.78. Table 2 provides information on occupations included inthe position generator, the occupational prestige score (Goyderet al., 2003), the percentage of respondents who reportedknowing someone in the occupation, and the percentage knownaccording to the type of relationship. The most commonlyaccessed occupation was registered nurse (50.9%); the leastcommonly accessed was funeral director (5.7%). The highest

ARTICLE IN PRESS

S. Moore et al. / Health & Place 15 (2009) 1071–10771074

prestige occupation, i.e., physician, was accessed by 42.5% of therespondents, mainly through weak ties, i.e., acquaintances; thelowest prestige occupation, i.e., janitor, was accessed by 20.5% ofthe respondents, again, mainly through weak, acquaintance ties.Table 3 provides average values of the three dimensions of socialcapital, i.e., upper reachability, diversity, and range, and the finalscore for (i) the overall sample, (ii) each level of educationalattainment, and (iii) each type of relationship. Pairwise one-wayANOVA tests with Bonferroni corrections showed significantdifferences in overall upper reachability scores across alleducation groups (F2,325 ¼ 28.72, po0.001); significantdifferences also existed between the ‘‘less than high school’’ and‘‘more than high school’’ groups in diversity (F2,325 ¼ 4.19,po0.001) and range (F2,325 ¼ 20.81, po0.001). There weresignificant differences between ‘‘less than high school’’ and‘‘more than high school’’ groups in the average upper

Table 1Characteristics of Montreal Neighbourhood Survey of Lifestyle and Health Sample

(social capital survey participants, n ¼ 328).

Variables % (n ¼ 328)

Education

Lower educated (less than high school) 7.3

Middle educated (high school or equivalent) 17.7

Upper educated (more than high school) 75.0

Income

Low income (o1.00 on income-to-household needs ratio) 36.0

Middle income (1.00–3.99 on income-household needs ratio) 54.0

High income (44.00 on income-household needs ratio) 10.0

Primary household language

French 72.9

English 12.5

Other 14.6

Married 46.1

Male 50.3

Canadian born 66.8

Mean (std. dev.)

Age 36.1 (8.8)

Residential tenure 5.2 (5.1)

OUTCOME

Mastery 23.4 (4.2)

Table 2Characteristics of position generator indicators of individual social capital (n ¼ 328).

Occupation Occupational prestige score

(Goyder et al., 2003)

% Total (std. de

High school teacher 68.4 47.8 (0.50)

Carpenter 52.4 25.3 (0.44)

Owner of retail store 60.8 32.6 (0.47)

Taxi driver 34.6 11.8 (0.32)

Physician 92.7 42.9 (0.50)

Janitor 33.6 20.4 (0.40)

Funeral director 58.8 5.8 (0.23)

Registered nurse 77.0 51.2 (0.50)

Sewing machine operator 35.9 13.7 (0.34)

Retail sales associate 39.1 19.5 (0.40)

Insurance Agent 48.8 27.4 (0.45)

Welder 55.2 14.0 (0.35)

Lawyer 74.4 39.3 (0.49)

Accountant 66.6 45.7 (0.50)

Trailer Truck Driver 42.0 16.4 (0.37)

Receptionist 40.4 32.6 (0.47)

reachability and diversity scores representing access through (i)family, (ii) friends, and (iii) acquaintances; there were significantdifferences between the two groups in the range score only inaccess through acquaintances. There was also a significantdifference between ‘‘less than high school’’ and ‘‘high schoolequivalent’’ groups in the upper reachability score in accessthrough acquaintances. For the final composite social capitalscore, the principal components analysis weighted the rangeindex at a value of 0.56 and upper reachability and diversity at avalue 0.22 each.

Table 4 presents the standardized coefficients and GEEstandard errors from the four models regressing mastery onindividual social capital, education, and covariates. Standardized bcoefficients are reported with confidence intervals (CI) forindependent variables. The bivariate regression of mastery onindividual social capital in model 1 indicated a significantassociation (b ¼ 0.21, CI: 0.13, 0.28, po0.001). Hence, increasingsocial capital by 1.00 (the standard deviation of social capital)results in a 0.21 standard deviation increase in mastery. In ourfinal model adjusting for socio-demographic covariates, educationmodified the association of social capital with mastery (b ¼ �0.41,CI: �0.54, �0.27, po0.001).

Fig. 1 illustrates the differential association of individual socialcapital with mastery as a function of level of educationalattainment. For the middle- and upper-educated strata therewas a positive association between individual social capital andmastery, i.e., as individuals’ degree of accessibility to the variousoccupations in the position generator increases, so, too, does senseof mastery. The interaction term indicated a negative associationbetween individual social capital and mastery for respondentswith less than a high school degree.

Social capital was moderately correlated with educationalattainment (r ¼ 0.31, po0.001) and income (r ¼ 0.21, po0.001),and educational attainment and income were similarly correlated(r ¼ 0.27, po0.001). In ancillary GEE analyses, we observedthat each particular dimension of social capital (e.g., upperreachability) interacted with education in association withmastery. Additional ancillary analyses with reclassification ofeducational attainment into four or five levels did not alterour results or change study conclusions. Our findings did notchange when analyses were repeated using imputed incomevalues for 24 non-respondents, with the exception that associa-tions with English-speaking household were no longer statisti-cally significant.

v.) % relative (std dev.) % Friend (std. dev.) % Acquaintance (std.

dev.)

14.3 (0.35) 28.9 (0.45) 20.7 (0.41)

7.6 (0.27) 10.7 (0.31) 10.7 (0.31)

4.6 (0.21) 12.5 (0.33) 19.2 (0.39)

1.5 (0.12) 4.9 (0.22) 6.1 (0.24)

12.2 (0.33) 21.3 (0.41) 23.8 (0.43)

1.8 (0.13) 6.4 (0.25) 13.4 (0.35)

0.9 (0.10) 1.2 (0.11) 3.7 (0.19)

17.0 (0.38) 24.7 (0.43) 24.4 (0.43)

5.2 (0.22) 7.6 (0.27) 4.3 (0.20)

3.7 (0.19) 10.9 (0.31) 11.6 (0.32)

3.7 (0.19) 11.6 (0.32) 16.2 (0.37)

4.9 (0.22) 5.5 (0.23) 5.2 (0.22)

11.6 (0.32) 18.3 (0.39) 21.0 (0.41)

11.6 (0.32) 24.1 (0.43) 21.9 (0.41)

3.7 (0.19) 7.3 (0.26) 7.9 (0.27)

5.8 (0.23) 14.0 (0.35) 16.5 (0.37)

ARTICLE IN PRESS

Table 3Individual social capital measures according to educational attainment and strength of tie (n ¼ 328).

Educational category Social capital dimension Overall (std.

dev.)

Relatives (std.

dev.)

Friends (std.

dev.)

Acquaintances (std.

dev.)

Total sample (n ¼ 328) Upper reachability (mean prestige value) 72.0 (28.3) 39.5 (37.5) 55.0 (35.0) 48.5 (38.7)

Diversity (mean number of ties) 5.4 (4.7) 1.1 (1.4) 2.1 (2.1) 2.3 (2.7)

Range (mean difference between high and low

prestige value)

31.9 (21.0) 7.1 (13.5) 16.6 (19.8) 17.4 (21.7)

Less than high school (23) Upper reachability 37.0��� (34.6) 15.14 (23.6) 27.24 (35.6) 12.04,� (24.3)

Diversity 1.94 (2.6) 0.44 (0.7) 1.04 (1.8) 0.44 (1.1)

Range 13.74,� (18.0) 2.1 (5.5) 10.6 (18.3) 2.44 (7.9)

High school degree or equivalent

(n ¼ 59)

Upper reachability 63.7��� (29.9) 32.5 (34.0) 43.5+ (36.9) 42.6� ((36.2)

Diversity 4.20+ (3.6) 1.0 (1.4) 1.6 (1.9) 1.6+ (1.9)

Range 28.0� (20.0) 4.5 (10.8) 11.9 (18.8) 14.2 (20.5)

More than high school (n ¼ 246) Upper reachability 77.3��� (24.2) 43.44 (38.4) 60.44,+ (32.7) 53.34 (38.4)

Diversity 6.14,+ (4.8) 1.24 (1.4) 2.34 (2.1) 2.64,+ (2.9)

Range 34.54 (20.6) 8.2 (14.4) 18.3 (20.0) 19.64 (22.2)

��� Significant mean differences among all education categories.+ Significant mean difference between ‘‘high school degree’’ and ‘‘more than high school’’ groups.4 Significant mean difference between ‘‘less than high school’’ and ‘‘more than high school’’ groups.� Significant mean difference between ‘‘less than high school’’ and ‘‘high school degree’’ groups.

Table 4Standardized coefficients estimating associations between mastery and individual social capital, education, and covariates; MNSLHS, 2006 (n ¼ 328).

Variable Model 1 coefficient (Std.err.a) Model 2 coefficient (Std.err.a) Model 3 coefficient (Std.err.a) Model 4 coefficient (Std.err.a)

Intercept 0.02 (0.09) 0.06 (0.09) 0.03 (0.12) 0.02 (0.12)

Individual social capital 0.23��� (0.03) 0.19�� (0.05) 0.15� (0.07) 0.20�� (0.07)

Educational attainment

Less than high school y �0.57�� (0.20) �0.49�� (0.17) �0.90��� (0.23)

High school or equivalent y �0.06 (0.15) �0.05 (0.10) �0.06 (0.10)

More than high school (referent) 0 (referent) 0 (referent) 0 (referent)

Social capital and education interaction

SC�low education y y y �0.46��� (0.08)

SC�med. education y y y �0.10 (0.05)

SC�high education (referent) y y y 0 (referent)

Income category

Low income y y �0.55��� (0.16) �0.54�� (0.16)

Middle income y y �0.38�� (0.12) �0.37�� (0.12)

High income (referent) y y 0 (referent) 0 (referent)

Primary household language

English y y �0.01 (0.09) 0.01 (0.09)

Other y y �0.15 (0.13) �0.16 (0.13)

French (referent) y y 0 (referent) 0 (referent)

Marital status

Married y y 0.35��� (0.10) 0.36��� (0.10)

Non-married (referent) y y 0 (referent) 0 (referent)

Gender

Male y y 0.13 (0.14) 0.13 (0.13)

Female (referent) 0 (referent) 0 (referent)

Age y y �0.25��� (0.04) �0.26��� (0.05)

Canadian-born status

Canadian born y y 0.31�� (0.11) 0.30�� (0.11)

Non-Canadian born 0 (referent) 0 (referent)

Residential duration y y 0.02 (0.05) 0.02 (0.05)

a GEE-empirical standard errors in parentheses.��� po.001.�� po.01.� po.05.

S. Moore et al. / Health & Place 15 (2009) 1071–1077 1075

Discussion

This study examined if social capital varies according toeducational attainment, and if education modifies the associationof individual social capital with mastery. In lower-educated

persons, greater individual social capital, i.e., more socialconnectivity and resource accessibility, was associated with alower sense of mastery. In higher educated persons, greaterindividual social capital was associated with higher sense ofmastery. Other covariates, such as marital status, age, and

ARTICLE IN PRESS

Fig. 1. Association of individual social capital with mastery across different levels of educational attainment. Note: Predicted values computed from coefficient estimates of

intercept, individual social capital, education, and the interaction terms reported in Table 4, Model 4.

S. Moore et al. / Health & Place 15 (2009) 1071–10771076

immigration status, were related to mastery in the expecteddirections. We focus here on the differential association ofindividual social capital with mastery according to educationalattainment, targeting two main issues: (1) disparities in indivi-dual social capital according to educational attainment andrelational type and (2) the negative association between socialcapital and mastery for lower-educated groups.

Social capital disparities

In our sample of Montreal residents, disparities were evidentin the distribution of social capital. These disparities took twoforms: (1) lower-educated individuals had a lower overall volumeof social capital than respondents with a more than a high schooldegree along each dimension of social capital and (2) lower- andupper-educated persons had different network pathways throughwhich they accessed social capital. First, in terms of the differentdimensions of individual social capital, lower-educated personshave lower average upper reachability, less diversity in theirnetwork (i.e., number of ties), and a smaller range of network ties.Second, lower- and upper-educated individuals had differentnetwork pathways through which they accessed social capital.Although individuals with a high school degree or more hadhigher levels of social capital in the upper reachability anddiversity dimensions along each pathway (i.e., relatives, friends,and acquaintances), these differences were significant across alldimensions, including range, for weak, acquaintance ties. In otherwords, there were no significant differences in the range of accessthat the lowest and highest educated groups had through relativesand friends; the difference in the range of the two groups lies inthe nature of their acquaintance ties. Low-educated individualshad more constricted weak ties and relied more on friends andfamily for access to social resources. As ties to close friends andfamily tend to be more homophilous, i.e., similar in socio-demographic background (Granovetter, 1973), the social networksof less-educated persons may have less socio-demographicdiversity than those of middle- and upper-educated persons.

Not all social capital is ‘‘good’’ capital

Understanding the differences between how upper- and lower-educated individuals access social capital is important for under-

standing the differential association of social capital with masteryaccording to education. In his work on the importance of weakties, Granovetter (1973, 1983) emphasized that individuals are lesslikely to be socially involved with acquaintances than they arewith family and friends (1973, 1983). In network terms, a person’snetwork of family and friend is likely more dense than theiracquaintance network. In other words, one’s family members andfriends more likely know each other than do one’s acquaintancesknow each other (Granovetter, 1983). The importance of weak,acquaintance ties resides in their potential bridging function, i.e.,in accessing information and resources beyond one’s immediatesocial circle. Without such ties, one is at a disadvantage inaccessing a range of resources compared to those who have suchties. Sociological research has shown that these disadvantagesdisproportionately affect marginalized groups through structuralbarriers that cap their access to a variety of more extensiveresources, including information, and affect their willingness orcapacity to develop weak, acquaintance relationships (Granovet-ter, 1983; Bourgois, 2003).

As a corollary to the relative absence of weak, bridging ties inone’s network is the fact that family and friends, i.e., strong ties,become even more important and necessary for one’s access toresources. This poses a double burden on marginalized groupswhose family and friends are also likely of low socioeconomicstatus with few personal resources. Moreover, access throughstrong ties may impose additional demands and pressures whichpotentially less likely develop when accessing resources throughweak ties. These pressures could lead to the development ofnegative consequences of social capital through the range ofmechanisms that Portes (1998) identifies: ‘‘restricted opportu-nities,’’ ‘‘excessive demands,’’ ‘‘limited freedoms,’’ and ‘‘down-leveling pressures’’ may all operate to reduce the sense of masteryof individuals with relatively higher social capital in a disadvan-taged network.

Limitations

Our findings should be considered in light of several limita-tions. First, the study is cross-sectional in design. Although ourstudy suggests that upstream social capital factors may influencemastery, it is also possible that mastery may influence a person’swillingness to establish relationships with others, particularly

ARTICLE IN PRESS

S. Moore et al. / Health & Place 15 (2009) 1071–1077 1077

acquaintance relationships. We have thus focused specifically onthe associational character of this relationship and not attemptedto draw any causal conclusions. Although reverse causality may bepresent in certain cases, sociological research on social integrationand psychological well-being has shown that social factors tend tobe more upstream in causal chains than psychological influences(Mirowsky and Ross, 2003). Longitudinal data would helpdelineate more clearly the direction of causal relationships.Second, the sample comprises volunteers, and lower-educatedpersons were limited in number. This is due in part to the fact thateducation is compulsory until a certain age but also that lower-educated individuals were less likely to volunteer for our study.The small sample size of lower-educated persons provided limitedstatistical power for conducting sub-sample analyses, particularlyon the specific association of weak ties with mastery. Third, therelatively few areas in the MNSLH prohibited use of multilevelmethods to test the associations between tract-level social capitaland individual mastery. Our analyses did however adjust for theclustering of observations. Fourth, the number of respondentsvaried across the seven study tracts. GEE can underestimatestandard errors with unbalanced cluster sizes or when thenumber of clusters is low (Feng et al., 1996).

Conclusion

Research on the negative consequences of social capital forhealth and well-being has remained a relatively unexamined areaof investigation, particularly compared to studies on its salubriousbenefits. Where research has found negative associations of socialcapital with health, there is little identification of the possiblemechanisms by which negative instead of positive outcomes maydevelop. The present study examines possible pathways by whichsocial capital results in negative consequences for mastery amonglower-educated compared to upper-educated individuals. Differ-ences between groups in the pathways that they use in accessingsocial capital may play an important role in understanding howsocial capital can negatively impact mastery. More generally, wesuggest that the identification of these different pathways hasimportant implications in designing programs that might leveragesocial capital in support of public health and health promotionpurposes.

Acknowledgements

We would like to acknowledge the contributions of CatherinePaquet in preparing the MNSLH data set for analysis and YanKestens in the development of the sampling plan. At the time ofthis research, SM held postdoctoral fellowships from the Fonds de

la Recherche en Sante du Quebec (FRSQ) and the Strategic TrainingProgram in Public and Population Health research of Quebec, apartnership of the Institute of Population and Public Health andthe Institute of Health Services and Policy Research of theCanadian Institutes of Health Research (CIHR) and the QuebecPopulation Health Research Network. M.D. holds a Research Chairfor Social Epidemiology from the University of South Australia andheld a Canada Research Chair for Biopsychosocial Pathways inPopulation Health from the Canadian Institutes of Health Researchduring the conduct of the research reported here. L.G. holds a

CIHR/CRPO (Centre de Recherche en Prevention de l’Obesite)Applied Public Health Chair in Neighbourhoods, Lifestyle, andHealthy Body Weight. L.D. holds a James McGill Chair inConsumer and Lifestyle Psychology and Marketing. Data collectionwas supported by a team grant from the FRSQ (Grant no. 8394,LD), the Canada Research Chairs programme (MD), CanadaFoundation for Innovation (Grant no. 201252, MD), and CIHR(Grant no. 200203 MOP 57805, LG).

References

Bourdieu, P., 1986. Forms of capital. In: Richardson, J.G. (Ed.), Handbook ofTheory for the Sociology of Education. Greenwood Press, Westport, CT,pp. 241–258.

Bourgois, P., 2003. In Search of Respect. Cambridge University Press, Cambridge.Carpiano, R.M., 2007. Neighborhood social capital and adult health: an empirical

test of a bourdieu-based model. Health & Place 13 (3), 639–655.Caughy, M.O., O’Campo, P.J., Muntaner, C., 2003. When being alone might be better:

neighborhood poverty, social capital, and child mental health. Social Science &Medicine 57 (2), 227–237.

Daniel, M., Moore, S., Kestens, Y., 2008. Framing the biosocial pathways underlyingassociations between place and cardiometabolic disease. Health & Place 14 (2),117–132.

Daniel, M., Brown, A., Garnggulkpuy, J., Cargo, M.D., O’Dea, K., 2006. Mastery,perceived stress and health-related behaviour in northeast Arnhem Land.International Journal for Equity in Health 5 (10), 1–10.

Daniel, M., Cargo, M.D., Lifshay, J., Green, L.W., 2004. Cigarette smoking,mental health and social support. Canadian Journal of Public Health 95 (1),45–49.

Feng, Z., McLerran, D., Grizzle, J., 1996. A comparison of statistical methods forclustered data analysis with Gaussian error. Statistics in Medicine 15,1793–1806.

Goyder, J., Guppy, N., Thompson, M., 2003. The allocation of male and femaleoccupational prestige in an Ontario urban area: a quarter-century replication.The Canadian Review of Sociology and Anthropology 40 (4), 417–440.

Granovetter, M., 1973. The strength of weak ties. American Journal of Sociology 78,1360–1380.

Granovetter, M., 1983. The strength of weak ties: a network theory revisited.Sociological Theory 1, 201–233.

Horton, N., Lipsitz, S., 1999. Review of software to fit generalized estimatingequation regression models. The American Statistician 53, 160–169.

Kawachi, I., 2006. Commentary: Social capital and health: making the connectionsone step at a time. International Journal of Epidemiology 35 (4), 989–993.

Lin, N., 2001. Social Capital: A Theory of Social Structure and Action. CambridgeUniversity Press, Cambridge.

Mirowsky, J., Ross, C., 1986. Social patterns of distress. Annual Review of Sociology12, 23–45.

Mirowsky, J., Ross, C., 2003. Social Causes of Psychological Distress. Aldine deGruyter, Hawthorne.

Moore, S., Shiell, A., Hawe, P., Haines, V.A., 2005. The privileging of communitarianideas: citation practices and the translation of social capital into public healthresearch. American Journal of Public Health 95 (8), 1330–1337.

Pampalon, R., Hamel, D., De Koninck, M., Disant, M.J., 2007. Perception of place andhealth: differences between neighbourhoods in the Quebec city region. SocialScience & Medicine 65 (1), 95–111.

Pearlin, L.I., Schooler, C., 1978. The structure of coping. Journal of Health and SocialBehavior 19 (1), 2–21.

Portes, A., 1998. Social capital: its origins and applications in modern sociology.Annual Review of Sociology 24, 1–24.

Portes, A., Sensenbrenner, J., 1993. Embeddedness and immigration—notes on thesocial determinants of economic-action. American Journal of Sociology 98 (6),1320–1350.

Rojas, Y., Carlson, P., 2006. The stratification of social capital and its consequencesfor self-rated health in Taganrog, Russia. Social Science & Medicine 62 (11),2732–2741.

Statistics Canada, 2008. Low income cut-offs for 2007 and low income measuresfor 2006. Income Research Paper Series. Ministry of Education. /http://www.statcan.ca/english/research/75F0002MIE/75F0002MIE2008004.pdfS.Accessed September 30, 2008.

Valente, T.W., Chou, C.P., Pentz, M.A., 2007. Community coalitions as a system:effects of network change on adoption of evidence-based substance abuseprevention program. American Journal of Public Health 97 (5), 880–886.

Ying, G., Liu, C., 2006. Statistical analysis of clustered data using SAS system. In:NorthEast SAS Users Group (NESUG) Proceedings, pp. 1–13.