does capital at home matter more than capital at school ......1999). in describing social capital,...

A

bafms©L

K

1

lFdi1oicg

f

0

Research in Social Stratification and Mobility 31 (2013) 1–21

Available online at www.sciencedirect.com

Does capital at home matter more than capital at school?Social capital effects on academic achievement

Mikaela J. Dufur a,∗, Toby L. Parcel b, Kelly P. Troutman ca Department of Sociology, 2008 JFSB, Brigham Young University, Provo, UT 84602, United States

b Department of Sociology and Anthropology, Campus Box 8107, North Carolina State University, Raleigh, NC 27603, United Statesc Department of Sociology, 3151 Social Science Plaza, University of California, Irvine, CA 92697, United States

Received 7 December 2011; received in revised form 17 August 2012; accepted 29 August 2012

bstract

A relatively neglected problem is how individuals derive social capital from more than one context and the extent to which theyenefit from the capital in each. We examine whether social capital created at home and at school has differing effects on childcademic achievement. We hypothesize that children derive social capital from both their families and their schools and that capitalrom each context promotes achievement. Using data from the National Longitudinal Education Study and structural equation

odeling, we show that capital from each context is helpful, with social capital in the family more influential than social capital at

chool. We discuss the implications of these findings for research on child achievement and for studies of inequality generally. 2012 International Sociological Association Research Committee 28 on Social Stratification and Mobility. Published by Elseviertd. All rights reserved.

Structu
eywords: Social capital; Academic achievement; Families; Schools;
. Introduction

It is well known that there is variation across ado-escents in tested levels of achievement (Farkas, 2003;ischer et al., 1996; Parcel & Menaghan, 1994) and thatifferences in achievement are associated with variationsn school success (Jencks et al., 1972; Sewell & Hauser,972). School success, in turn, is a critical predictor ofccupational and earnings attainment (Farkas, 1996), anmportant component of life in meritocratic society, and

onsequential to the transmission of inequality acrossenerations.

∗ Corresponding author. Tel.: +1 801 422 1720;ax: +1 801 422 0625.

E-mail address: mikaela [email protected] (M.J. Dufur).

276-5624/$ – see front matter © 2012 International Sociological Association Research Committee

http://dx.doi.org/10.1016/j.rssm.2012.08.002

ral equation modeling

It is also well known that resources from multiplesocial contexts influence academic attainment amongadolescents. After classic studies of occupational statusattainment demonstrated the importance of family back-ground and the intervening effects of education on adultattainment (Blau & Duncan, 1967), scholars developedincreasingly sophisticated models of why and how edu-cation was so consequential (Breen & Jonsson, 2005;Shavit & Blossfeld, 1993). Others studied why variousaspects of families, such as family structure (McLanahan& Sandefur, 1994) and family process (Conger, Conger,& Martin, 2010) were influential. Theorists such asColeman (1988, 1990) pointed to social capital as anunderlying construct influential in both families and

schools that influenced attainment, as well as to the gen-erality of the social capital concept in explaining othersocial outcomes. Despite these accomplishments, addi-tional questions remain.

28 on Social Stratification and Mobility. Published by Elsevier Ltd. All rights reserved.

dx.doi.org/10.1016/j.rssm.2012.08.002http://www.sciencedirect.com/science/journal/02765624mailto:[email protected]/10.1016/j.rssm.2012.08.002

l Stratifi
2 M.J. Dufur et al. / Research in Socia
Specifically, relatively little research explicitlyaddresses how family and school contexts affect theaccumulation and use of social capital as it is transmit-ted to educational outcomes. For example, does socialcapital pay off regardless of where it is generated? Oris the benefit of social capital dependent on the contextin which the investment is made? In addition, can socialcapital reflect resources in more than one context?

In this paper we address these issues with a study offamily and school social capital effects on adolescentacademic achievement. We hypothesize that childrenderive social capital from their families and from theirschools and that capital from each context is influential inpromoting academic achievement. We demonstrate thatsocial capital at home and social capital at school aredistinct constructs. We also show that capital from eachcontext is helpful in promoting child academic achieve-ment. Our empirical strategies differ from previous workin allowing us to evaluate the extent to which specificmeasures of capital actually reflect more than one con-text simultaneously, as well as the overall effectivenessof capital from each context.

We first review literature on social capital that iden-tifies which aspects of families and schools may beimportant to study. We then review literature pointingto limitations in prior studies of social capital effectson academic achievement, followed by details of ourempirical strategy and findings. We conclude by dis-cussing how our strategies can inform questions in thestudy of inequality more generally by identifying howresources in other contexts, such as neighborhoods andworkplaces, also differentially affect stratification out-comes.

2. Background

2.1. Social capital as a concept and a construct

By social capital we refer to resources that inherein the relationships among actors and that facilitate arange of social outcomes (Coleman, 1990). Social cap-ital is contained in relationships among individuals aswell as among corporate actors; it reflects the presenceof objective ties, as well as subjective relationships con-tingent upon trust, reciprocity, or other positive emotions(Paxton, 1999). In describing social capital, Coleman(1988) focuses specifically on the information, obliga-tions, and norms that are transmitted through social ties,

resources that help children acquire knowledge fromadults. Social capital can be used in the production ofindividual-level goods, such as when a worker recom-mends a friend for a job (Kirschenman & Neckerman,

cation and Mobility 31 (2013) 1–21

1991). Social capital can also be used in the produc-tion of group-level goods, as when immigrant groupsuse within-group ties to promote entire communities(Portes, 1998). We focus on individual-level creation ofand returns to social capital, with investments from thefamily and the school serving as key contexts from whichchildren may draw social resources.

Social capital theory implies purposeful investmenton the part of the actors. Considering children’s academicdevelopment, Coleman (1988) argued that the mere pres-ence of parental stores of knowledge is insufficientto ensure transmission of that knowledge to children.Instead, parents must make specific choices to invest intheir children’s development and engage in interactionswith a specific child to create the bonds along whichinformation can pass. Families who invest expect to seehigher levels of academic achievement in their children;they undertake these investments with the expectationthat strong academic achievement during childhood willtranslate into higher levels of education and occupationalattainment in later years, thus promoting upward mobil-ity in our stratification system. For this reason, familysocial capital investment has intergenerational conse-quences.

The creation of social capital is not limited to thefamily. Social capital associated with schools refersto investments between students and schools that canfacilitate educational outcomes. These bonds can reflectcommunity ties, but typically refer to the relationshipsthat parents and children form with school teachers andpersonnel. For example, arguments regarding the supe-riority of Catholic schools are based on the notion thatthe religious ties that link many attendees, their families,and their teachers promote common norms useful in sup-porting academic achievement (Coleman, 1988, 1990;Morgan, Sorensen, & Todd, 2009). In addition, attend-ing Catholic schools may promote the formation of socialcapital through social closure; parents are more likely tohave relationships with other parents whose children alsoattend the school. Adults in non-family settings, such ascoaches and church leaders, also incur costs of investingin children as they transmit knowledge and norms thatsupport academic achievement.

In addition, Putnam (2000) distinguished betweenbonding and bridging social capital. Bonding capitalincludes the intra-family connections we have describedabove, connections that we argue facilitate academicachievement for children and adolescents. In addition,

there is considerable evidence that children benefit fromthe social connections that parents have with otherssuch as neighbors, school personnel, and work col-leagues (Crosnoe, 2004; Dufur, Parcel, & McKune,

l Stratifi

22sgBisa

stcF(aotPctadaatsoPaccodihbc

ait

toHwTiacro

M.J. Dufur et al. / Research in Socia

008; Johnson, Crosnoe, & Elder, 2001; Parcel & Dufur,001a, 2001b). These connections illustrate bridgingocial capital; the stronger these connections are, thereater the resources to which children have access.ecause parents and schools share the responsibil-

ty of educating children, both bonding and bridgingocial capital may be important in promoting academicchievement.

The United States is not the only locale in whichcholars have studied these issues, often in the con-ext of analysis of cultural capital as well as socialapital.1 For example, using the 1992–1993 Netherlandsamily Survey, De Graaf, De Graaf, and Kraaykamp2000) demonstrated that parental reading behavior has

stronger effect than parental beaux arts participationn children’s educational attainment, particularly whenhe parents themselves have low levels of education.ensiero (2011) argued that it is child engagement inognitively stimulating activities, rather than participa-ion in organized activities, that predicts reading abilitynd locus of control. In addition, parental expectations,irect stimulation, parental interactions with the schoolnd children’s engagement in cognitively stimulatingctivities mediated more than half of the SES effect onhese two dependent variables, even in the presence oftrong controls. Notten and Kraaykamp (2010) studiedver 8300 respondents from the Family Survey Dutchopulation to investigate the role of parental readingnd television viewing, as well as parental guidance onhildren’s reading and television viewing, on child edu-ational attainment. They find that parental reading paidff for children, while parental television viewing wasisadvantageous. In addition, active parental investmentn children’s reading, such as giving children books foroliday gifts, reading to young children and discussingooks with them, had a positive effect on children’s edu-ational attainment.

Studies of Asian samples take a somewhat different

pproach. Park, Byun, and Kim (2011) found that parentsnvested considerable resources in locating and moni-oring private tutors for their children. Lee and Shouse

1 Some international research studied cultural and social capitalogether, often seeking to parse the effects of these respective formsf capital on academic outcomes of children and adolescents (Jager &olm, 2007; Notten & Kraaykamp, 2010). Our focus differs, in thate seek to untangle the effects of social capital at home and at school.hese lines of research are related, however, in that some studies used

ndicators of what they call cultural capital that others might identifys social capital (e.g., Yamamoto & Brinton, 2010). While addressingultural capital effects on child cognition is not our purpose here, theseelated approaches both offer important insights into the reproductionf social inequality across generations.

cation and Mobility 31 (2013) 1–21 3

(2011) showed that such “shadow education” was tiedto students’ desires to attend prestigious higher educa-tional institutions. Bassani (2006) studied the effects ofsocial capital at home and at school on math achieve-ment in Canada, Japan and the United States. She foundthat supportive family interaction had positive effects inall three societies, while the effects of social capital atschool were stronger than those at home in Japan, butweaker than those at home in the United States. Moregenerally, Ferguson’s (2006) review of studies from sev-eral countries highlighted the consistency across thesestudies, with investment in social capital tied to positivemeasures of children’s well-being.

Social capital theory, then, explains the mechanismsand processes by which bonds between children andother actors, such as their parents or their teachers,produce academic achievement. Adult investment inchildren is more than supervision; it creates the mech-anisms via which children are socialized and educated.Family and school environments characterized by lowlevels of social capital will be insufficient to transmit nec-essary information and knowledge to children, leadingto lower levels of achievement.

2.2. Promoting academic achievement: familiesversus schools

There are theoretical reasons to distinguish betweenthe capital created in the family and that created at school.The ties created between parents and children are strong,the result of repeated and frequent interactions; becausethe parent–child bond is one of the most intimate rela-tionships in early life, we would expect the social capitalcreated in families to exert a heavy influence on childacademic outcomes, even into adolescence.

The ways parents choose to invest in their connectionswith their children, then, can have powerful and long-lasting effects. Lareau (2011) argues that middle classand working class or poor children experience “unequalchildhoods” as a function of differing socializationstrategies at home. Middle class parents use concertedcultivation, creating a full schedule of activities for theirchildren to encourage academic development throughintense parent–child interaction and activities coordi-nated with schools. In contrast, working class and poorparents schedule far fewer activities and instead viewchild development as accomplishment of natural growth.They are less likely to actively play with their children,

leaving offspring more time to spend in free play. Theyalso talk with their children less (Hart & Risley, 1995).

Still, while parental influences remain strong through-out the teen years (Amato & Booth, 2000), relationships

l Stratifi
with people outside the kin circle become increasinglyimportant as children age. We expect that students whomake additional investments in social capital at schoolby participating in activities that introduce closer rela-tionships to peers, teachers, and other parents wouldsee returns to that investment in the form of greaterachievement. In addition, parents who become involvedin their child’s school or who get to know other par-ents are investing in a set of weak ties (Granovetter,1983) or in bridging social capital (Putnam, 2000)that provide broader access to social resources pro-moting academic achievement than parents who relyonly on the strong ties of close-knit kin and neigh-borhood circles. Lareau (2011) adds to this argumentwhen she notes that working-class parents are lesseffective in dealing with schools than middle-classparents because they are not as assertive in their chil-dren’s education in making requests of teachers andother school personnel. The discordance between theschools’ expectations for parental involvement andthe lower levels of this involvement among workingclass/poor families are a hindrance to child academicachievement.

However, debate continues as to the degree to whichschools actually influence children’s learning. Heckman(2008) argues that ability gaps across children developearly in their lives, and that the family environmentsof young children are major predictors of later suc-cess; he also argues that early interventions, includingthose within families, are more efficacious in promot-ing positive school outcomes, while later interventionssuch as improved pupil–teacher ratios have weakereconomic returns. Similarly, Grubb (2009) argues thatschool factors vary in their effects on student learning,and favors policies that bolster both families and schoolsin order to promote student academic outcomes. Oth-ers argue that the ways families shape students’ livesmay not be something that even very good schools canovercome (cf. Hanuschek, 1994; Morgan & Sorensen,1999). Some of Coleman’s evidence for the impor-tance of social capital, such as the effects of attendingCatholic schools or schools with low teacher–studentratios, diminish notably once background controls areintroduced (Alexander & Pallas, 1985; Hallinan &Kubitschek, 1999). At the same time, other studies linkschool social and financial resources to greater achieve-ment (cf. Elliott, 1998; Greenwald, Hedges, & Laine,1996; Parcel & Dufur, 2001a). The lack of consensus

on school effects underscores the need to more carefullyconsider the context from which students receive socialcapital when contemplating how to improve educationaloutcomes.


In addition, the debate about how to empiricallyhandle the potential differences between family andschool resources continues. For example, some scho-lars have argued that social capital as measured by thesocial ties between parents and children’s friends andbetween parents and other adults should be considered anindividual-level resource (Carbonaro, 1999; Hallinan &Kubitschek, 1999) while others view it as a school-levelresource (Morgan & Sorensen, 1999). While proponentsof both positions make compelling arguments, theirdivergent analytic choices lead to divergent findings,with individual-level social capital sometimes positivelyrelated to academic outcomes and group-level socialcapital potentially a negative predictor of academicoutcomes. Parcel, Dufur, and Zito’s (2010) review doc-uments these arguments.

This lack of consensus provides little guidance forresearchers trying to model the effects of capital on chil-dren’s achievement. We take this debate as a point ofdeparture. Our approach allows a more explicit test ofwhether social capital created at school can be distin-guished from social capital created in families. Further,we can examine whether family or school social capitalis stronger in its effects on academic achievement.

2.3. Social capital effects on academic achievement

Several studies deal obliquely with the notion thatcapital from more than one context may be influentialin predicting academic outcomes. Teachman, Paasch,and Carver (1996) posit that social capital protects stu-dents against dropping out of school. Their analysesshow strong protective effects of social capital, butthey consider attendance at a Catholic school and liv-ing in a two-parent family as two indicators of thesame idea—social capital—rather than investigatingwhether the site (in a school or at home) or type ofinvestment (bonding between teachers and students orbonding between parent(s) and children) makes a differ-ence. Similarly, McNeal (1999) explicitly looks at bothparent–child discussion and PTA participation as socialcapital in direct contrast to human and financial capital,but does not investigate whether these indicators mightbe substantially different from each other because oneis created at home and one is created at school. Pong(1998) demonstrates the influence of social capital on10th-grade mathematics and reading achievement as amacro-level aggregation of the effects of individual stu-

dents’ family capital, but this analysis focuses on theneed to control statistically for nested levels of data ratherthan distinguishing between contexts in which capi-tal might be created. Family-based indicators, such as

l Stratifi

pcwpaesc

itgndotsttcscrawotdtitatscsFtscstcm

dPatifia


ercent of students from a single-parent family or per-ent of students who say their parents know their friendsell, and potentially school-based indicators, such asrincipal-reported measures of parental involvement, arell modeled at the school level to try to capture systemicffects. In contrast, we view family-based indicators andchool-based indicators as potentially reflecting separateonstructs.

Other scholars speculate that social capital existsn more than one context but have not fully testedhis supposition. Swanson and Schneider (1999) sug-est that moving to a new school and moving to aew home have different consequences. Using NELSata and distinguishing among students who changenly houses, only schools, or both houses and schools,hey find that the type of mobility matters, withchool changes exerting important effects. However,hey frame these differences as mobility issues ratherhan focusing on the varying contexts in which socialapital might be accrued. The effects of changingchools may be due to the loss of school socialapital, while family social capital in their analysesemains largely intact. Furstenberg and Hughes (1995)ttempt to examine social capital in two separate spheres,ithin and outside the family, by examining 252 childrenf young mothers. They include both measures of whathey call “within-family” ties (such as parents and chil-ren sharing activities) and “family’s embeddedness inhe community” (such as school quality). They find pos-tive effects of social capital, but conclude in the endhat more research is required to distinguish adequatelymong potential types of social capital and how thoseypes affect different outcomes. In sum, these studieshow that social capital may be tied to many positivehild outcomes, but do not clearly estimate the relativetrength of capital effects across contexts. We agree withurstenberg (2005), who called upon researchers to more

horoughly study the embeddedness of actors in largerocial systems in order to measure more accurately socialapital and understand how actors use it. Our empiricaltrategies allow us to assess whether capital in one con-ext is more important than capital in another in affectinghild achievement, thus suggesting where interventionsight be more effective.Several analyses that have attempted to model chil-

ren’s social capital guide the current investigation.arcel and Dufur (2001a) argue that capital at home andt school may operate in parallel ways, demonstrating

hat both are helpful in promoting reading and mathemat-cs achievement among 5–13-year-old children. Theynd that capital acquired in families sometimes inter-cts with capital acquired at school to produce desired


outcomes; for example, more positive school social envi-ronments help blunt the negative effects of very highmaternal work hours. Similarly, Crosnoe (2004) usesdata from the National Survey of Adolescent Healthto argue that social capital may be housed in both thefamily and the school and that those institutions andthe capital they create may be intertwined to facilitateacademic achievement. His findings provide additionalevidence for the idea that family and school capitalwork both independently and together to affect childoutcomes. Finally, Hoffmann and Dufur (2008) findthat high-quality schools partly compensate for poorparental attachment and low parental involvement inschool in preventing juvenile delinquency, especially forlow-achieving youths.

What these studies could not demonstrate, however,is the extent to which a given indicator of capital mayactually reflect capital from more than one context. Forexample, when a parent participates in school organiza-tions, does this reflect family capital? Or does it reflectthe creation and strengthening of capital at school, giventhat the associations parents are building are derivedfrom the school context? Or does it reflect both forms ofcapital, with the potential that such involvement couldprovide two pathways of influence to promote childachievement?

Another limitation of these analyses is that their mod-els could not produce a summary measure of a complexconcept like social capital. Previous work, particularlystudies looking at the effects of family and school socialcapital on child well-being, has tended to use sets of sin-gle indicators to try to capture the effects of social capital.For example, Teachman et al.s’ (1996) excellent paperexplicitly distinguishes among family and school cap-ital, but uses sets of single indicators in a multivariatesetting to test that rather than measuring social capi-tal as a more global construct that might encompass anumber of investment approaches (see Parcel & Dufur,2001a, 2001b, and others for similar strengths and weak-nesses). Such an approach may introduce measurementerror and may not be sufficient to investigate the effects ofthe underlying concept of social capital (Paxton, 1999).In addition, using individual variables restricts the abil-ity to distinguish clearly between capital at home andat school, limiting the ways the findings may be usedto craft intervention programs. While previous researchreveals that more family indicators than school indica-tors were predictive of cognitive development (Parcel &

Dufur, 2001a), the findings do not allow us to concludedefinitively that family social capital is more importantthan school social capital in creating positive academicoutcomes. Through use of structural equation modeling,

l Stratifi
we address the issue of measurement error and are able toinclude more—and more accurate—information aboutthe form and function of children’s social capital (Paxton,1999). Our approach to testing whether social capital iscreated and used in different contexts also allows a com-parison of the strength of capital effects at home and atschool. This approach has proved useful in the study ofchild behavior problems (Dufur et al., 2008); these strate-gies may also illuminate analyses of child achievement.

2.4. Hypotheses

Our first hypothesis is that social capital is created inboth the family and at school. We expect that actors willcreate important ties and exchange information with bothfamily members and school personnel, but that socialcapital at school and social capital at home are dis-tinct constructs. Nonetheless, social capital created indisparate contexts may still be interrelated. Thus, oursecond hypothesis is that there are inter-relationshipsacross measures of each form of capital that reflectboth family and school influence on capital creation.Such inter-relationships would indicate that social capi-tal created in multiple contexts is closely connected andwould work together in some circumstances to promoteachievement. Third, we hypothesize that social capital inthe family and at school will each have positive effects onchild academic achievement. Fourth, given that severalinvestigations of child outcomes have found a greaternumber of family than school predictors significant, webelieve that the overall effect of family social capital willbe stronger than that of overall school social capital inpredicting child academic achievement.

3. Data

We use data from the second follow-up wave of theNational Education Longitudinal Study (NELS), con-ducted by the National Center for Educational Statistics.The NELS is a nationally representative study that gath-ered data from students, parents, teachers, and schooladministrators. The first wave of the study was conductedin 1988, drawing random samples of approximately25 eighth-grade students from each of 1000 randomlyselected schools. We employ data from the 12th gradesurveys, yielding a sample of 10,585 students. TheNELS has extensive data on schools and schooling fora large, nationally representative sample over time, and

includes many indicators of social capital. The NELS isan especially useful data set for this inquiry as it pro-vides potential indicators of social capital from multipletypes of actors. Using the NELS, we can incorporate


information in student, parent, and teacher behaviors,an especially important point in examining bonding andbridging social capital created across different actorsand in different settings (Coleman, 1988, 1990; Putnam,2000). In contrast, other data sets that include data onboth family and school contexts, such as the NLSYChild–Mother data set, contain data from fewer infor-mants, thus limiting the richness of the measures. Inaddition, the data attached to the NLSY Child–Motherdata set lack measures of child or parental attachmentto the school that are available in the NELS data. Giventhat our central focus is on theory testing, the age of thedata set need not be a major consideration; we addressthis issue further in Section 7.

4. Methods

4.1. Overall analytic strategy

To assess the structure of social capital, we firstperform confirmatory factor analyses using structuralequation models in AMOS. Confirmatory factor analysisis the most suitable method to use in this study becauseit allows us to determine (1) the extent to which eachindividual indicator does or does not reflect social capi-tal at home or social capital at school; (2) the frequencyand extent to which individual indicators reflect capitalin both contexts, the family and the school. In addition,this strategy allows us to use the best fitting models toestimate capital effects on academic achievement withina structural equation format. We also test for correlatederrors between similar items to provide unbiased models.After determining the structure of latent constructs tapp-ing family social capital and school social capital, weinvestigate the extent to which each of these influencesyouth academic achievement. We describe the modelsin greater detail below. We then add demographic con-trols to determine whether the effects of social capitalremain strong when additional factors that have beenshown to influence academic outcomes are considered.Finally, we subject these models to additional analysesthat demonstrate the robustness of the findings.

4.2. Measures

4.2.1. Dependent measureTable 1 describes the variables used in the analy-

ses. Our dependent variable, academic achievement, is acomposite standardized test score variable of knowledgein math, reading comprehension, and science. Studentscompleted the test during their senior year of high school.

M.J. Dufur et al. / Research in Social Stratification and Mobility 31 (2013) 1–21 7

Table 1Measures and variables.

Variable Description

Dependent variableComposite test scores Examines math, reading, science, and history knowledge. Scale = 0–100Capital indicatorsFamily social capitalParental trust in child Student report; 0 = only rarely to 6 = all the timeDiscuss issues with

parentsThree items measuring how often students report discussing (1) school programs, (2) school activities, and (3)school classes with parents (0 = only rarely to 6 = all the time)

Parent checks student’shomework

Parental report; 0 = only rarely to 6 = all the time

Parents attend schoolmeetings

Parental report on total number of times; range from 0 to 5 and above

Parents attend schoolevents

Parental report on total number of times; range from 0 to 5 and above

School social capitalStudent participation in

extracurricularactivities

Student report; four-item scale that includes athletics, academic clubs, and other school extracurricular activities(α = .74)

School contacts parent Parental report; 0 = never to 5 = five or more timesHigh teacher morale Teacher report; 0 = low to 5 = high. Higher scores = higher moraleLow conflict between

teachers andadministrators

Teacher report; 0 = frequent conflict to 5 = no conflict (reverse coded to reflect capital accumulation)

Teachers respond toindividual needs

Parental report; 0 = never to 5 = often

School environment 14-Item scale containing items asking students about the degree to which various school problems, such asdelinquency, violence, and absenteeism, are a problem in their schools (α = .88). Coded so that higherscores = more positive environment

ControlsIndividual-level controlsChild sex 1 = male; 0 = femaleChild race 1 = white; 0 = nonwhiteFamily socioeconomic

statusComposite of parental education and income, reported by NELS

Maternal marital status 1 = married; 0 = not marriedFamily size Parental report of number of children in sibshipSchool-level controls (Section 5.3)Percent free or

reduced-price lunchAdministrator report of the percent of students in the school receiving free or reduced-price lunch

Percent low-incomestudents

Administrator report of the percent of students in the school who qualify as low-income

Teacher education Aggregation of teacher reports of level of education received, recoded to years of post-secondary education priorto aggregation

Highest teacher salary Administrator report of the highest salary paid to any full-time teacher at the school. Responses range in dollarsfrom 0 to 24,999; 25,000 to 29,999; 30,000 to 34,999; 35,000 to 39,999; 40,000 to 44,999; 45,000 to 49,999; and50,000+

Teacher experience at Aggregation of teacher reports of how of how many years they have taught in their current school

4

ttwtn

school

.2.2. Independent measures: social capitalIn this paper, we focus on individual-level capital

hat individual students may accrue through their con-

act with key actors in their families and schools. Whilee agree that a global or collective form of social capital

hat students could draw on could be created in an orga-ization like a school, a thorough investigation of the

form and function of that collective good is beyond thescope of this paper. Instead, we utilize variables that indi-cate potential capital provided either by the adolescent or

an adult attached to that specific adolescent. We investi-gated multiple specifications of social capital using morethan 65 potential measures of family social capital andmore than 45 measures of school social capital derived

l Stratifi
from previous theoretical and empirical work. We usedfour criteria to select indicators for our models. First,the indicators had to reflect at least one aspect of eitherbonding or bridging social capital as outlined in previ-ous theoretical treatments. Second, they had to exhibitacceptable factor loadings on the respective constructs ofeither family or school social capital. Third, their load-ings had to be statistically significant. Finally, inclusionof the indicator had to improve the overall statistical fitof the model.

The indicators included in the models reported in Sec-tion 5.2 generally meet these four criteria. While we findthe exclusion of some variables surprising (for exam-ple, family attendance at cultural events), such variablessimply were not a good fit to models specifying socialcapital.2 In addition, while data sets examining youthoutcomes often include a myriad of school characteris-tics, a school characteristic does not necessarily indicateschool social capital. School characteristics such as aver-age standardized achievement scores are likely outcomesof interest that social capital might predict, but theywould not directly reflect the social ties, norm transmis-sion, or interconnectedness that would be characteristicof social capital. In the case of two variables (notedbelow), the indicators meet three of the four criteria; allothers meet all four criteria we establish above.

We measure family social capital with seven indi-cators. Previous theory-building in social capital hasargued that trust between individuals and institutionsis a key indicator of social capital (e.g. Paxton, 1999).We include a variable asking parents how much theytrust their children; higher scores indicate greater trust.Coleman (1988, 1990) argued that adult stores of humancapital would have little effect on children’s develop-ment without the availability of social capital to transmitthat knowledge to the children. We are able to tapparent–child interconnectedness, allowing the flow ofinformation from adult to child, in these data. Weinclude variables on how often students say they dis-

cuss (1) school programs, (2) school activities, and (3)school classes with their parents; higher scores indi-cate more discussion. We also use a variable tapping

2 Previous examinations of the effects of social capital on adolescentoutcomes using the NELS data have used multiple combinations ofsimilar measures (cf. Carbonaro, 1999; Hallinan & Kubitschek, 1999;Morgan & Sorensen, 1999; Teachman et al., 1996). Our contributionis not in identifying additional variables that could be used to measuresocial capital; rather, we use previously studied indicators to examinethe ways that capital may be created and used in different contextsand to examine whether the modeling of said indicators in differentcontexts is important in determining the effects of social capital.


how often the parent checks homework, allowing themto transmit information about specific academic sub-jects to children; higher scores indicate more contact.Parental contact with other adults or institutions createsthe kinds of closed networks Coleman argues are partic-ularly valuable in encouraging child development. Twoitems, parental attendance at parent–teacher meetingsand parental attendance at school events, tap parentalreports of interaction with schools. Although we usesome ordinal variables, all such variables have five ormore response categories, and none are notably skewed;they are therefore appropriate for use in structural equa-tion modeling.

We model school social capital in the NELS with sixitems. Looking to tap the networks that connect students,parents, and teachers to each other, building the socialcapital that will allow adults to transmit informationand norms to youth, we measure student participationin extracurricular activities with a four-item scale thatincludes athletics, academic clubs, and other schoolextracurricular activities (α = .74). To indicate the closednetworks Coleman describes, we also include a measureof how often parents say the school contacts the students’parents; although it is possible that schools contact par-ents in response to poor performance or inappropriatebehavior, this variable is positively correlated with ourmeasure of academic performance and so we use it asan indicator of positive contact. Higher scores indicatemore contact.

We also examine the connections among other actorsat school that could create a store of social capital indi-vidual students could access (Paxton, 1999; Putnam,2000). We note again that although the following vari-ables might indicate a more general store of socialcapital at school that connects students, teachers, andadministrators together, these variables are reported byindividual students and teachers and therefore differ foreach respondent, representing the individual’s percep-tion of the available social capital. Reports of moraleamong school employees and conflict between schoolemployees, reported by one of the target student’s tea-chers, provide an additional indicator of social capital;high morale and low conflict are linked to greater tiesamong adults in the school, providing greater overallschool social capital into which students can tap. Wereverse-code the original conflict measure so that higherscores on both variables would indicate greater socialcapital. An indicator of student–teacher contact asks

parents whether they feel teachers meet the needs ofindividual students. Higher scores indicate greater con-tact. We also utilize a 14-item scale containing variablesasking students about their school environment. Better

l Stratifi

stitwlai&dstaci

4

mdcfisaS

4

owtdiuwAas2cRdmimsctgf


chool environments reflect more general adoption onhe part of students of pro-social norms taught by adultsn the school. In addition, we argue that higher scores onhese scales indicate schools where adults and studentsork to create a positive environment and may be more

ikely to remain on school grounds and build social ties,n approach commonly used in the study of social capitaln neighborhoods (Putnam, 2000; Sampson, Morenoff,

Earls, 1999). These include items concerning theegree to which various issues are a problem in thechool: (1) school-related behaviors (such as absen-eeism), (2) delinquent behaviors (such as violence), (3)nd social issues (such as racism) (α = .88). We reverse-ode many of the included items so that higher scoresndicate more positive school environments.

.2.3. ControlsAfter the initial model predicting academic achieve-

ent, we estimate a second model that includes basicemographic controls to ensure the effects of the socialapital constructs are not spurious. These controls areor child sex and race (Parcel & Dufur, 2001a); fam-ly size (Downey, 1995; Parcel & Dufur, 2001a); familyocioeconomic status (Duncan & Brooks-Gunn, 1997);nd maternal marital status (Kim, 2011; McLanahan &andefur, 1994).

.3. Missing data

Initial analyses indicated that missing data on somef the school variables was of non-trivial proportions,ith as much as 15% missing on one school administra-

or variable. Because many commonly used methods ofealing with missing data produce substantial distortionn variables’ distributions and unrealistic variances, wese two strategies to deal with missing variables. First,e use the maximum likelihood estimation option inMOS, as maximum likelihood estimation improves the

ccuracy and the power of analyses relative to methodsuch as mean substitution or listwise deletion (Enders,010; Schafer & Graham, 2002). Second, to exert moreontrol over the utilized equations, we follow Little andubin (1987) in using interpolation and imputation toeal with cases that have missing data. If a variable isissing data on fewer than 10% of cases, we use linear

nterpolation to replace missing values. If a variable wasissing data on more than 10% of cases, we interpolate

cores on that variable through regression imputation,

reating a regression equation to predict the variablehat has missing data; we could then use the coefficientsenerated by that equation to create unique new valuesor each case based on other characteristics of that case.


Findings for both strategies are very similar; we reportthe latter here.

4.4. Models

An issue of interest in creating the measurement mod-els for social capital in the family and at school was howto deal with interaction between parents and their chil-dren’s schools. Is such interaction family social capital,school social capital, or something else? We model sev-eral possibilities (Fig. 1): (A) children’s social capitalis most appropriately modeled as one factor, regardlessof where the capital is created or used. If this modelbest fits the data, our findings will provide support forthe approaches most often taken in previous researchthat does not distinguish the site in which capital iscreated. That is, if this model fits best, we can con-clude that it does not matter where capital is created.(B) Children’s social capital is created and used in twodistinct contexts, families and schools. (C) Children’ssocial capital is most appropriately modeled in multi-ple contexts, with one latent construct tapping familysocial capital, one latent construct comprising schoolsocial capital, and a third latent construct reflecting socialcapital created jointly between families and schools.Indicators that tap substantial parent–school interaction,such as student participation in extracurricular activitiesor parental participation in PTA meetings, might reflectthis joint capital. D) Child social capital is created andused in two distinct contexts, families and schools, butsome parent–school interaction variables “cross-load”onto both latent variables, indicating both separate con-texts for capital and that capital is created and used jointlyacross two institutions. We test all four of these modelsand report the best-fitting solutions in Section 5.1.

We then use the social capital latent variables iden-tified in the confirmatory factor analyses to predictacademic achievement. We are able to consider familyand school social capital as single effects rather thanexamining individual coefficients associated with multi-ple variables, allowing us to use standardized coefficientsto compare the actual strength of capital created in thefamily to that created at school.

Although our models include some categorical vari-ables, most have enough response categories to assumeunbiased estimates; in addition, estimates we derivedusing polychoric correlations did not differ significantly(Joreskog & Sorbom, 1993). We therefore report the

findings from the non-polychoric analyses here as thisallows us to use typical standardized coefficients to indi-cate size of effect, and most models used in similarliterature employ the non-polychoric models, allowing

10 M.J. Dufur et al. / Research in Social Stratification and Mobility 31 (2013) 1–21

Model C

Model B

School Social Capital

Family Social Capital

Model A

Social Capital

Model D

: Tested Measurement Models

measure
Fig. 1. Tested
for comparison across findings. We report standardizedcoefficients, allowing us to compare the size of effectsacross variables.

5. Findings

5.1. Measurement models

Table 2 reports fit statistics for each of our fourpotential models. We use multiple fit indices to morefully evaluate the model (Browne & Cudeck, 1993;Tanaka, 1993). We report the comparative fit index(CFI), which measures the fit of the models as comparedto the null model; a CFI approaching one indicates agood fit. The Akaike Information Criterion (AIC), bycontrast, can make comparisons across nonnested mod-els. Smaller AICs indicate a better fit. Similarly, theBayesian Information Criterion (BIC) (Raftery, 1996)is an approximation to the log of a Bayes factor for theexperimental model compared to the saturated model,favors parsimonious models and is recommended with

large sample sizes, as we have here. Again, lower val-ues on the BIC indicate superior fit. Because the AICand BIC may be sensitive to large sample sizes, we alsoinclude the Browne–Cudeck Criterion (BCC), which

ment models.

imposes greater penalties for a lack of parsimony thandoes the AIC. Lower values on the BCC indicate superiorfit. We also report the root mean-squared error of approx-imation (RMSEA) to compare observed and predictedcovariance matrices and to take into account model com-plexity; for models such as ours, an RMSEA below .05indicates a good fit. We also report chi-square statisticsto test across models to determine whether a model fitsbetter than others.

Our findings indicate children’s social capital is bestrepresented as two separate latent factors, one reflectingsocial capital created in the family and one reflectingsocial capital created at school. Although this findingwould be consistent with either Model B or D, the bestfitting model is Model D, or the model in which indi-cators could “cross-load” onto both latent social capitalconstructs. In this case, one indicator—student participa-tion in extracurricular activities—cross-loads onto bothfamily social capital and school social capital. As Table 2indicates, this cross-loading model is a better fit to thedata, with a lower RMSEA (.045), a CFI closer to one

(.877), and lower values on AIC, BCC, and BIC scoresthan competing models. Although we note that each ofthe models has a significant chi-square, potentially indi-cating poor fit, because the chi-square is a comparison of


Table 2Comparison of social capital measurement models’ fit statistics for Model A (one latent variable), Model B (two latent variables; family socialcapital and school social capital), Model C (three latent variables; family social capital, school social capital, and joint family–school capital), andModel D (two latent variables—family social capital and school social capital—with shared indicators)—NELS.

Model X2 CFI RMSEA AIC BCC BIC

Model A 7329.666*** (df = 75) .476 .072 7417.67 7417.78 7609.82Model B 5483.571*** (df = 129) .847 .050 2281.99 2282.11 2474.14Model C 2419.817*** (df = 76) .583 .116 2380.00 2380.32 2690.63Model D 1783.16*** (df = 74) .877 .045 1873.16 1873.28 2072.65

N = 10,585.*p < .05 two-tailed tests.**

apphMn1satH

ttawcdmdpetaiiaf

asbpptbai

*p < .01 two-tailed tests.** p < .001 two-tailed tests.

model to a “perfect fit,” even a slight deviation from aerfect fit can induce a significant score, especially in theresence of a large sample size (cf. Paxton, 1999). Likeli-ood ratio tests between Model B and Model D, in whichodel B is nested, indicate that Model D fits the data sig-

ificantly better than does Model B (F = 410.83***; df). Thus, the model that best fits the data suggests twoeparate forms of social capital, one in the family andnother at school, with variables able to indicate capi-al in both settings, providing affirmative evidence forypotheses 1 and 2.In examining family social capital (see Fig. 2),

he standardized coefficients are all significant and inhe expected direction: more discussion with offspring,ttendance at school events, and supervision of home-ork can be interpreted as reflecting greater social

apital in the family. Parental trust that the student willo what is expected has a small but significant effect;odels without this indicator are not as good a fit to the

ata. Error terms are significantly correlated betweenarental attendance at school meetings and at schoolvents, a finding we would expect given that the ques-ions are structured in an almost identical fashion andny error associated with how the first item is answereds likely to be similar to the error associated with answer-ng the second. These findings provide support for ourrguments that these indicators are all good measures ofamily social capital.

Similarly, standardized coefficients are all significantnd in the expected direction when looking at schoolocial capital: more parental contact with the school;etter school environments; and better relations amongarents, teachers, and administrators all can be inter-reted to reflect greater social capital at school. Although

he effect of school contacting parent is small as indicatedy the standardized coefficients, the effect is significant,nd the model better fits the data with that indicatorncluded. Thus, this measure meets three of our four

criteria. One correlation between error terms is signif-icant: the errors terms associated with participation inschool activities and school contacting the parent aresignificantly correlated, possibly because participationin an extracurricular activity requires substantial coor-dination between parents and school personnel. Theseresults provide support that all of these measures reflectschool social capital.

The indicator of participation in extracurricularevents loads onto both the family social capital measureand the school social capital measure when both forms ofcapital are included simultaneously. Although the factorloading for extracurricular participation on school socialcapital is small, it is significant, the model fit is superiorwhen it is included, and the connection is theoreticallysupported. In other words, when modeled together, fam-ily social capital and school social capital as measuredin the NELS share an indicator, notably one that mightrequire active involvement of both parents and schools inorder for successful student participation. These findingsprovide some support for the idea that although the mea-surement of youths’ social capital must take into accountmultiple contexts, indicators of social capital may beshared across contexts.

5.2. Explanatory models predicting academicachievement

We now estimate the effects of these respective formsof capital on student academic achievement by usingthese constructs within path analytic models. The find-ings in Fig. 3 show that both family social capital andschool social capital are positive, significant predictorsof test scores. As predicted, the standardized effect of

family social capital (.371) is stronger than the effectof school social capital (.106). This difference providesevidence in favor of the hypothesis that capital in the fam-ily is more influential for children’s achievement than is

12 M.J. Dufur et al. / Research in Social Stratification and Mobility 31 (2013) 1–21

Figure 2: Measurement Model with Family Social Capital and School

Social Capital (NELS)

Parents check homework

Extracurricular activities

e

Teacher/ administrator conflict

Teachers meet individual needs

Teacher morale

0.051***

CFI 0.877 RMSEA 0.045

Schoolenvironment

Student/ teacher ratio

School contracts parent

School Social Capital

e e e e e

Discuss school programs

e

Parental Trust Parents attend school events

Parents attend meetings

Discuss classesDiscuss school activities

e e e e e

0.154***

e

Family Social Capital

social c
Fig. 2. Measurement model with family
capital at school; however, school capital is still a signifi-cant positive factor in promoting academic achievementeven in the presence of family social capital. Theremaining epistemic correlations are similar across themeasurement and explanatory models. Goodness of fitstatistics indicate that this model is a moderately goodfit to the data (CFI = .850; RMSEA = .049); we note,however, that our goal in this paper is testing whethercontextualized measures of social capital can be used toexplain outcomes such as academic achievement ratherthan explaining all variance in that outcome.

We test models that introduce basic demographiccontrols (race, family socioeconomic status, child sex,family size, and maternal marital status) (Fig. 4). Inthese models, the effects of family social capital andschool social capital diminish, in the case of schoolsocial capital by more than half, but remain positive,significant predictors of test scores, net of controls.These findings provide evidence that modeling socialcapital as reflecting separate, distinct constructs can beuseful in explaining academic achievement. They also

suggest that controlling for these background character-istics, while important for understanding both academicachievement and how social capital affects it, does notinterfere with our major conclusion. Specifically, family

apital and school social capital (NELS).

social capital is a more important predictor of child aca-demic achievement than is school social capital, net ofbackground controls.

5.3. Additional analyses

We also ran several robustness checks to strengthenour confidence in the findings showing that youth socialcapital should be measured in multiple contexts and thatsocial capital at home is more important than social cap-ital at school in predicting academic achievement. First,we consider the possibility that demographic statusessuch as family SES or race might affect the creation offamily social capital. While previous work has shownthat the effects of family and school social capital per-sist net of family and school human and financial capital,it is unclear to what extent other resources may haveindirect effects that operate through social capital (cf.Parcel & Dufur, 2001a). For example, if modeling anindirect effect of family socioeconomic status throughfamily social capital renders the direct effect of fam-

ily social capital on test scores nonsignificant, then ourconclusions should stress the indirect effect of familysocioeconomic status on test scores working throughfamily social capital. In this case, significant indirect


Expla natory Mo del wit h Family Social Cap ital and School

Social Capital Predictin g Test Scores (NELS )

Parents c heck homework

Extrac urricul ar acti vities

e

Standardi zed test scores

Teache r/ administrator conflict Teachers meet individ ual ne edsTeache r morale

0.050***

CFI 0.850 RMSEA 0.049

Schoo lenviron ment Studen t/ teacher ratioSchool con tracts parent

Schoo l Social Capit al

e e e e e

Disc uss sc hool pro gra ms

e


Parents attend meeti ngsDisc uss cl ass esDisc uss sc hool acti vities

e e e e e

0.031***

e

Family Social Cap ital

al and s

eiaif

tstoptoftfltc

Fig. 3. Explanatory model with family social capit

ffects might reveal that the relationship between fam-ly social capital and the dependent variable is largely

mechanism through which basic status characteristicsnfluence academic outcomes, or even that the effect ofamily social capital is spurious.

To test this possibility, we model indirect effectshrough family social capital of race, family SES, childex, maternal marital status, and family size. Each ofhese variables could potentially affect the creationf family social capital. For example, better educatedarents might have greater awareness of the impor-ance of helping their offspring with homework, onef our indicators of family social capital, and there-ore might be more willing or able to build capital inhat way. Greater income might be a proxy for more

exible work hours, allowing parents to interact with

he school more freely. Unfortunately, model identifi-ation issues made it impossible to test all of these

chool social capital predicting test scores (NELS).

indirect effects simultaneously, so each effect was testedindividually.

Table 3 shows the results of these analyses. Thefirst column lists the tested effects. The second column,Model 1, shows the effect of family social capital anddemographic controls on test scores without the inclu-sion of indirect effects (in other words, our originalmodel). Each subsequent pair of columns represents theeffects of a different test of indirect effects; the first col-umn in each pair displays the coefficients from a modelincluding one indirect effect of a demographic control,while the second shows the change in coefficients com-pared to the models without the indirect effect. Forexample, the third column, Model 2, shows coefficientsfrom the model where we test the indirect effects of race

on test scores through family social capital. The fourthcolumn shows the difference in the coefficients betweenModel 1 and Model 2, and so forth. The first row of

14

M.J.

Dufur

et al.

/ R

esearch in

Social Stratification

and M

obility 31

(2013) 1–21

Table 3Tests of indirect effects of individual background control variables on composite test scores through family social capital; standardized coefficients.

Effect of controlvariable

Model 1 Model 2 Model 3 Model 4 Model 5 Model 6

Modelwith noindirecteffects

White →familysocialcapital

Changefrom M1to M2 withindirecteffect

SES →familysocialcapital


Male →familysocialcapital


Mothermarried →familysocialcapital


Familysize →familysocialcapital


Control → familysocial capital

NA .167*** NA .395*** NA −.112*** NA .226*** NA −.198*** NA

Family socialcapital → testscores

.209*** .206*** −.003 .216*** .007 .206*** −.003 .208*** −.001 .207*** −.002

White → test scores .102*** .089*** −.013 .099*** −.003 .102*** .000 .115*** .013 .101*** −.001SES → test scores .407*** .403*** −.004 .361*** −.046 .405*** −.002 .382*** −.035 .391*** −.016Male → test scores −.081*** −.082*** .001 −.087*** .006 −.085*** .004 −.087*** .006 −.083 .002Mother

married → testscores

.255*** .240*** −.015 .250*** −.005 .256*** .001 .266*** .011 .249*** −.006

Family size → testscores

.312*** .311*** −.001 .319*** .007 .309*** −.003 .318*** .006 .304*** −.008

N = 10,585.*p < .05.**p < .01.*** p < .001.


Expla natory Model wit h Family Social Cap ital and School

Social Cap ital Predictin g Test Scores (wit h c ontrols) (NE LS)

Parents c heck homework

Extracurricular activities

e

Standa rdized test scores

Teache r/ administrator conflict

Teachers meet individ ual ne edsTeache r morale

0.388***

CFI .801 RMSEA 0.056

Schoo lenviron ment

Studen t/ teacher ratio

School con tracts parent

Schoo l So cial Capit al

e e e e e

Disc uss sc hool pro gra ms

e


Parents attend meeti ngsDisc uss cl ass esDisc uss sc hool activities

e e e e e

0.033***

e

Family Social Ca pital

Male

Parents married

White

Socio-economic

status

Family size

e

e

e

e

e

0.081***

chool so

croFcmosacmctTtc

Fig. 4. Explanatory model with family social capital and s

oefficients shows that all five demographic variables weeport here exercise statistically significant direct effectsn achievement, as expected and as noted previously inig. 4. The second row shows that the effect of familyapital on test scores remains strong in the presence ofodeled indirect effects, with only very small changes

nce these indirect effects are taken into account. Familyocial capital, then, exerts significant effects on academicchievement even in the presence any indirect effects ofhild race and sex, family SES or size, and maternalarital status. Subsequent rows display coefficients and

hanges in them across models for the demographic con-rols themselves; changes across models are minimal.

hus, inclusion of these indirect effects creates very lit-

le change in the strength of the effect of family socialapital on test scores when any of the indirect effects

cial capital predicting test scores (with controls) (NELS).

are included. The direct effects of the other controls ontest scores change only slightly in each model. Statisti-cally, then, inclusion of indirect effects does influencethe direct effect of family social capital on test scores;practically, however, these changes are negligible, sug-gesting that even after controlling for the ways familybackground influences family social capital, that formof capital still has a positive association with academicachievement.

We conducted a second robustness check concern-ing school characteristics to strengthen confidence inour findings. Recall that our focus has been on schoolsocial capital measured at the individual level. How-

ever, it remains a possibility that school-level factors mayinfluence academic achievement. For example, the per-centage of students receiving free or reduced-price lunch

l Stratification and Mobility 31 (2013) 1–21

on

com

posi

te

test

scor

es; s

tand

ardi

zed

coef

ficie

nts.

odel

1M

odel

2M

odel

3M

odel

4M

odel

5M

odel

6

ingl

e-ev

el odel

(no

choo

l-ev

elon

trol

s)

%

stud

ents

on

free

orre

duce

d-pr

ice

lunc

h

Cha

nge

from

M1

to

M2

with

scho

olef

fect

%

low

inco

me

stud

ents

Cha

nge

from

M1

to

M3

with

scho

olef

fect

Teac

her

educ

atio

nC

hang

efr

om

M1

to

M4

with

scho

olef

fect

Teac

her

sala

ry(h

ighe

st)

Cha

nge

from

M1

to

M5

with

scho

olef

fect

Teac

her

expe

rien

ceat

scho

ol

Cha

nge

from

M1

to

M6

with

scho

olef

fect

A.0

44*

NA

.038

*N

A

.063

**N

A

.067

***

NA

.081

***

NA

203*

**.2

08**

*.0

05

.207

***

.004

.208

***

.005

.212

***

.009

.205

***

.002

041*

**.0

43**

*.0

02

.045

***

.004

.046

**.0

05

.049

***

.008

.047

***

.006


and teacher qualifications/expectations have sometimesbeen found to influence test scores (Clotfelter, Ladd, &Vigdor, 2007; Roscigno & Crowley, 2001). It is possiblethat the effect of school social capital will no longer besignificant net of such contextual variables. Thus, we testthe possibility that the effect of individual-level schoolsocial capital is spurious by including school-level con-textual measures in the model.

Because school context variables were measured atthe school, rather than the individual, level, these testsrequire multilevel structural equation models, with thedependent variable, family social capital and schoolsocial capital as measured by individuals’ responses,and individual-level controls (race, sex, etc.) at Level1 and school characteristics at Level 2. We measureschool context with five variables: percent of studentsreceiving free or reduced-price lunch, percent of stu-dents from low-income families, percent of teachers atthe school with master’s degree or above, teacher pay(highest salary), and teacher experience at current school(see Table 1 for measurement details on school-levelvariables).3 We note, as above, that model identificationissues make it impossible to include all of these variablesin the model simultaneously. In addition, inclusion ofschool-level variables excludes a large number of casesfrom the sample; taken together, these limitations meanthat the following findings should be interpreted withcaution. We use the TWOLEVEL analysis protocol andmaximum likelihood estimation in MPlus to test thesemodels.

Results of these analyses are displayed in Table 4.The first column lists the effects being tested. The sec-ond column, Model 1, shows the effect of family andschool social capital on test scores without the inclu-sion of school-level controls (in other words, our originalmodel). Each subsequent pair of columns represents theeffects of a different school-level social capital mea-sure on achievement; the first column in each pairdisplays the coefficients, while the second shows thechange in coefficients compared to the models withoutthe school-level variable. For example, the third col-

umn, Model 2, shows coefficients from the model wherewe include the percent of students in the school whoreceive free- or reduced-price lunch. The fourth column

3 We ran the same tests using lowest teacher salary and teacher expe-rience in any secondary school setting and found similar results, so wedo not report them here for the sake of parsimony. We also testedurban school setting, type of school (public or private), and school sizeas measured by student/teacher ratio, but as none of these exerted asignificant effect on composite test scores and their effects on familyand school social capital were negligible, we do not report them here. Ta

ble

4Te

sts

of

scho

ol-l

evel

cont

rol v

aria

bles

M S l m s l c

Scho

ol-l

evel

cont

rol →

test

scor

esN

Fam

ily

soci

al

capi

tal →

test

scor

es.

Scho

ol

soci

al

capi

tal →

test

scor

es

.

N

=

5675

.*

p

<

.05.

**p

<

.01.

***

p

<

.001

.

l Stratifi

s1scmolemiwvfdcstwafa

6

eusBsiitmdb

tsathcacctoicts


hows the difference in the coefficients between Model and Model 2, and so forth. Row 1 shows that allix school-level capital variables we report here exer-ise statistically significant effects on achievement, asentioned above. Rows 2 and 3 show that the effects

f individual-level family social capital and individual-evel school social capital change only very slightly asach of the school-level controls is introduced to theodel. Changes in the effects of both family social cap-

tal and school social capital are also much smaller thanhen individual-level controls were introduced, as pre-ious research would suggest (see Figs. 3 and 4). Bothorms of capital, then, exert significant effects on aca-emic achievement even in the presence of school-levelharacteristics. Taken together, the robustness checksummarized in Tables 3 and 4 increase our confidencehat family and school social capital are built in differentays in different contexts, that both forms of capital are

ssociated with higher academic achievement, and thatamily social capital exerts stronger effects on studentchievement than does school social capital.

. Discussion

Our goal in this paper is to investigate whether consid-ring the context in which social capital is created andsed refines our understanding of how investment inocial resources affects child academic achievement.ecause test scores are associated with later school

uccess, inferences regarding where resources are mostnfluential are important to those concerned with hownequality is transmitted across generations. We arguehat social capital is not a monolith, but rather that invest-

ents in different social sites may potentially lead toifferent returns. Our analytic strategies help us moveeyond prior research in several ways.

Consistent with our first hypothesis, we demonstratehat children’s social capital is site-specific. In doingo, we improve upon prior studies that were unable tottend to the context from which social capital indica-ors are derived. There is less support for our secondypothesis regarding social capital created jointly acrossontexts. Only one item, participation in extracurricularctivities, acts as an indicator of social capital in bothontexts. While this may suggest that children’s socialapital is created and used in two very distinct con-exts, we note that this particular variable is indicativef shared effort between families and schools. Partic-

pation in extracurricular activities requires clear andonsiderable investments by school personnel, includingime spent coaching, supervising, or otherwise trainingtudents. Similarly, these activities require considerable


investments on the part of parents, who often have toarrange transportation, help raise funds, and chaperoneevents, bringing the parents into greater contact withtheir own children, their children’s peers, and other par-ents and teachers. It is possible that activities or situationsrequiring heavy investment from two parties are bestconsidered as being created across multiple contexts.

Consistent with our third hypothesis, we also demon-strate that capital from two distinct contexts, familiesand schools, positively affects academic achievement.Our confirmatory factor analysis reveals the extent towhich variables reflect the underlying concept of socialcapital, better enabling scholars to tie the specifics ofsocial capital theory to everyday social interactions bothat home and at school. As Coleman (1988) predicted,investment in the social ties that pass along informationand create obligations and shared norms helps transmithuman capital to the next generation.

In addition, this analytic strategy allows investigatorsto assess the relative strength of social capital in variouscontexts, an approach that has value for scholars and pol-icymakers alike. Consistent with our fourth hypothesis,we found that family social capital exerts stronger effectson academic achievement than does school social capital,a finding that clarifies ideas about sites of social invest-ment and may aid practitioners in crafting interventionpolicies that focus resources where they will have thegreatest benefit. These findings highlight the importanceof the social capital children experience in the home ascritical to promoting child development. In addition, wenote that a comparable analysis of child behavior prob-lems also demonstrated that families exerted strongereffects than schools; hence, the evidence appears to beaccumulating across these two dependent variables thatthe location in which social capital is created matters interms of promoting positive child development (Dufuret al., 2008).

7. Limitations, future research, and implicationsfor the study of inequality

A limitation to our analysis is that we have not studiedthe role of financial capital at home versus financial cap-ital in schools, nor have we studied the relative effectsof human capital at home compared to human capital atschool in their effects on child achievement. Althoughthese analyses were beyond the scope of this project,we have controlled for aspects of parental human and

financial capital in our models, thus reducing the risk ofspecification error. In addition, our models also provide aframework for studying the effects of these other formsof capital in future investigations. Such studies would

l Stratifi
provide additional information as to whether our con-clusions regarding the relative strength of social capitalat home and at school would be replicated for theseother forms of capital. If we were to find, for example,that parental financial capital (e.g., parental earnings)were more important than financial capital at school(e.g., per pupil expenditures), and that parental humancapital (e.g., parental education) were more importantthan schools’ human capital (e.g., teacher qualifica-tions), these findings would provide additional evidenceregarding the relative importance of families and schoolsin reproducing inequality. Most likely, however, the storyis much more complex (see Grubb, 2009). Finally, in ouradditional analyses, we were unable to simultaneouslycontrol for the measures of indirect and school-leveleffects included in Tables 3 and 4. It is possible that someof those specific findings could change if such analysescould be conducted.

Nonetheless, our theoretical model can be extendedto contexts beyond families and schools to more collec-tive stores of social capital. An obvious connection isto the excellent work by criminologists looking at thecapital available in neighborhoods (cf. Sampson et al.,1999). Although the data we use here do not allow for adetailed examination of neighborhood social capital, itis possible that youth may collect stores of social capitalin families, schools, and neighborhoods, as well as othersettings such as churches or sports teams. Schools, forexample, might be a site where collective stores of socialcapital are available for children to draw upon even ifthey or their own teachers or parents do not help to cre-ate that capital. While beyond the scope of this paper,future research could examine the way social capital iscreated and used collectively at the school level (such asthrough principal reports of global school capital) andwhether such capital exerts a positive influence on aca-demic achievement beyond the connections individualstudents make at their schools.

We also speculate that families influence academicachievement for both younger children and adolescents,but that other institutions such as schools exercise moreinfluence as children mature. In our study all childrenwere with a year or two of the same age, thus precludingour ability to investigate whether social capital at schoolwould matter more for older as compared with youngerchildren, a worthy topic for future research. In addition,peer groups likely become more important as childrenmature (Amato & Booth, 2000), another topic worthy

of investigation. Again, the models we have estimatedare flexible enough to use in testing such hypotheses.Studying children across age groups also brings up thepossibility that stores of capital may fluctuate over time,


both in terms of how much capital individuals mayaccess and in where individuals might create their cap-ital. While the investigation of how capital might shiftacross time is beyond the scope if this paper, the baselinewe provide here—that youth capital is created in multi-ple contexts—provides a stepping stone for examiningthose shifts in the future.

Our findings also have implications for scholars whoare particularly concerned with the role of schools inreproducing inequality. Schools in the United States havetraditionally been asked to do many things, from integrat-ing immigrants into society to ameliorating inequality(Cheadle, 2008; Cuban, 2010). Current debates aboutschool accountability dramatize the tendency for manyto argue that if student achievement is lacking, the solu-tion is to “fix” schools, despite decades of evidencefrom social science suggesting that family influences arevery consequential (Coleman et al., 1966; Cuban, 2010;Parcel et al., 2010). In addition, because of the primacyand privacy of family life, many may be reluctant toadvocate for strategies that would encourage the creationof greater social capital in the family, preferring insteadto encourage investment in capital at school. Our findingsin conjunction with previous research (Dufur et al., 2008)suggest that efforts to increase social capital at school,such as initiatives to reduce class size or attempts tocreate parent–school programs and ties, would probablyhave a beneficial effects on students. However, we alsofind that family social capital has a stronger influenceon child achievement than does school social capital,even after controlling for socioeconomic status and otherdemographic characteristics. It remains possible that therelative roles of the family and schools have changedover the years since the NELS data were produced. Thisis a possibility that future research should investigate.

Nonetheless, looking at these implications highlightsthe importance of acknowledging how social capital maybe constructed in different spheres and the consequencesof investment in each of those spheres. Thus, althoughinvestment in school social capital may be beneficial, ourresults suggest that investment in family social capitalwould reap greater returns. For example, social policiesand interventions focusing on increasing academic out-comes could be tailored to allow greater investment inboth family and school social capital. One possibilityis expansion of supportive workplace policies, such asflextime, that would allow parents to attend school meet-ings and participate in extracurricular activities with their

children without adversely affecting their jobs. Policiessuch as job-sharing, or encouraging part-time work, thatallow parents more time to discuss school with theirchildren or do homework together might build family

l Stratifi

scstsCca

ewsogNshtRtaocacidrt

A

tiFUuHAa

R

A

A

B


ocial capital and, therefore, have a greater effect onhild academic achievement than policies focusing onchool social capital. However, most families need full-ime work for one or two adults, thus rendering thesetrategies less practical than flextime in many cases.ommunicating to parents how efficacious family socialapital can be in promoting positive child outcomes maylso be worthwhile (Tough, 2009).

Finally, our findings suggest the need to study theffects of social capital in other contexts, particularlyhen studying inequality among adults. For example,

cholars studying labor markets note that weak ties,r those other than with family, can help explain whoets good jobs (Granovetter, 1983; Kirschenman &eckerman, 1991). Studies of immigrant experiences

how how social capital in the destination communityelps new immigrants become successful more quicklyhan those lacking ties and associations (Portes, 1998;aijman & Tienda, 1999). In both of these examples,

he context in which social capital is located and usedppears to be a critical component in gaining desiredutcomes. However, to date studies have not systemati-ally demonstrated which contexts are most important inffecting which aspects of inequality, and under whichonditions. Our study develops one model of how suchnvestigations might proceed. We look forward to theevelopment of a broad body of evidence regarding theelative importance of several contexts in the reproduc-ion of social inequality.

cknowledgments

An earlier version of this paper was presented athe 2005 Annual Meetings of the American Sociolog-cal Association, Philadelphia. We thank the College ofamily, Home and Social Sciences at Brigham Youngniversity for partial support of this research, Ben McK-ne for assistance with graphs, and Tim Heaton, Johnoffmann, and Pamela Paxton for helpful comments.dditional support was provided by the College of Social

nd Behavioral Sciences at the Ohio State University.

eferences

lexander, K. L., & Pallas, A. M. (1985). School sector and cognitiveperformance: When is a little a little? Sociology of Education, 58,115–128.

mato, P. R., & Booth, A. (2000). A generation at risk: Growing up

in an era of family upheaval. Cambridge, MA: Harvard UniversityPress.

assani, C. (2006). A test of social capital theory outside of the Amer-ican context: Family and school social capital and youths’ math


scores in Canada, Japan, and the United States. International Jour-nal of Educational Research, 45, 380–403.

Blau, P., & Duncan, O. D. (1967). The American occupational struc-ture. New York: Free Press.

Breen, R., & Jonsson, J. O. (2005). Inequality of opportunity in com-parative perspective: Recent research on educational attainmentand social mobility. In K. S. Cook, & D. S. Massey (Eds.), Annualreview of sociology (pp. 223–243). Palo Alto: Annual Reviews.

Browne, M. W., & Cudeck, R. (1993). Alternative ways of assessingmodel fit. In K. A. Bollen, & J. S. Long (Eds.), Testing structuralequation models (pp. 136–162). Newbury Park, CA: Sage.

Carbonaro, W. J. (1999). Opening the debate on closure and schoolingoutcomes: Comment on Morgan and Sorenson. American Socio-logical Review, 64, 682–686.

Cheadle, J. E. (2008). Educational investment, family context, andchildren’s math and reading growth from kindergarten through thethird grade. Sociology of Education, 81, 1–31.

Clotfelter, C. T., Ladd, H. F., & Vigdor, J. L. (2007). Teacher credentialsand student achievement: Longitudinal analysis with student fixedeffects. Economics of Education Review, 26, 673–682.

Coleman, J. S. (1988). Social capital in the creation of human capital.American Journal of Sociology, 94, S94–S120.

Coleman, J. S. (1990). Foundations of social theory. Cambridge, MA:Harvard University Press.

Coleman, J. S., Campbell, E. Q., Hobson, C. J., McPartland, J. M.,Mood, A. M., Weinfeld, F. D., et al. (1966). Equality of educationalopportunity. Washington, DC: Government Printing Office.

Conger, R. D., Conger, K. J., & Martin, M. J. (2010). Socioeconomicstatus, family processes, and individual development. Journal ofMarriage and Family, 72, 685–704.

Crosnoe, R. (2004). Social capital and the interplay of families andschools. Journal of Marriage and Family, 66, 267–280.

Cuban, L. (2010). As good as it gets: What school reform brought toAustin. Cambridge, MA: Harvard University Press.

De Graaf, N. D., De Graaf, P. M., & Kraaykamp, G. (2000). Parentalcultural capital and educational attainment in the Netherlands: Arefinement of the cultural capital perspective. Sociology of Educa-tion, 73, 92–111.

Downey, D. B. (1995). When bigger is not better: Family size,parental resources and children’s educational performance. Amer-ican Sociological Review, 60, 746–761.

Dufur, M. J., Parcel, T. L., & McKune, B. A. (2008). Capital andcontext: Using social capital at home and at school to predictchild social adjustment. Journal of Health and Social Behavior,49, 146–161.

Duncan, G. J., & Brooks-Gunn, J. (1997). Consequences of growingup poor. New York: Russell Sage Foundation.

Elliott, M. (1998). School finances and opportunities to learn: Doesmoney well spent enhance student achievement? Sociology of Edu-cation, 71, 223–245.

Enders, C. K. (2010). Applied missing data analysis. The GuilfordPress: New York.

Farkas, G. (2003). Cognitive skills and noncognitive traits and behav-iors in stratification processes. In J. Hagan, & K. Cook (Eds.),Annual review of sociology (pp. 541–562). Palo Alto, CA: AnnualReviews Inc.

Farkas, G. (1996). Human capital or cultural capital? Ethnicity andpoverty groups in an urban school district. New York: Aldine de

Gruyter.

Ferguson, K. M. (2006). Social capital and children’s well-being: Acritical synthesis of the international social capital literature. Inter-national Journal of Social Welfare, 15, 2–18.

l Stratifi
Fischer, C. S., Hout, M., Jankowski, M. S., Lucas, S. R., Swidler, A.,& Voss, K. (1996). Inequality by design: Cracking the bell curvemyth. Princeton, NJ: Princeton University Press.

Furstenberg, F. F. (2005). Banking on families: How families generateand distribute social capital. Journal of Marriage and Family, 67,809–821.

Furstenberg, F. F., & Hughes, M. E. (1995). Social capital and success-ful development among at-risk youth. Journal of Marriage and theFamily, 57, 580–592.

Granovetter, M. (1983). The strength of weak ties: A network theoryrevisited. Sociological Theory, 1, 201–233.

Greenwald, R., Hedges, L. V., & Laine, R. D. (1996). The effect ofschool resources on student achievement. Review of EducationalResearch, 66, 361–396.

Grubb, W. N. (2009). The money myth: School resources, outcomesand equity. New York: Russell Sage.

Hallinan, M. T., & Kubitschek, W. N. (1999). Conceptualizing andmeasuring school social networks: Comment on Morgan andSorensen. American Sociological Review, 64, 687–693.

Hanuschek, E. A. (1994). Money might matt

does capital at home matter more than capital at school ......1999). in describing social capital,...

Documents