School Connectedness in the Health Behavior in School-AgedChildren Study: The Role of Student, School, and SchoolNeighborhood CharacteristicsDouglas R. Thompson, Ronaldo Iachan, Mary Overpeck, James G. Ross, Lori A. Gross

ABSTRACT: School connectedness includes liking school and positive relations with teachers and peers. School connectedness isassociated with a variety of positive health outcomes. The goal of this study was to identify characteristics of students, schools, andschool neighborhoods that are related to school connectedness. In the Health Behavior in School-Aged Children (HBSC) Study, schoolconnectedness was reported by 13,207 students (grades 6-10) in 340 schools. HBSC measured a variety of student characteristics.Characteristics of schools were culled from data maintained by Quality Education Data, and school neighborhood characteristicswere derived from the 2000 decennial census. Associations between connectedness and student, school, and school neighborhoodcharacteristics were estimated using hierarchical linear models. Characteristics of students, schools, and school neighborhoods wereassociated with school connectedness. Connectedness was greater among younger students, females, students with better academicperformance and greater extracurricular involvement, students with greater self-rated physical attractiveness, students with morefriends, students from 2-parent families, and students whose parents were more involved with school. Connectedness was greater insmaller schools, more racially homogeneous schools, and schools with more students from relatively wealthy households. Schoolconnectedness was higher in neighborhoods with a greater percentage of non-US citizens. As the percent of renters in theneighborhood increased beyond 20%, school connectedness tended to decrease. The findings point to possible strategies forfostering school connectedness. (J Sch Health. 2006;76(7):379-386)

School connectedness has been defined in differentways,1 but common indicators include liking school,

a sense of belonging at school, positive relations withteachers and friends at school, and an active engagement inschool activities. A growing literature shows that schoolconnectedness predicts a variety of health outcomes. Stu-dents who feel connected to school report higher levels ofemotional well-being, less substance abuse, better health,decreased levels of suicidal ideation, decreased depressivesymptoms, and decreased risk of violent or deviant behav-ior and teen pregnancy.2-7 Given the variety of positiveoutcomes associated with school connectedness, it isworthwhile to gain a better understanding of factors thatcould lead to higher levels of school connectedness and,conversely, risk factors for low levels of connectedness.This was the goal of the present research.

Past studies have identified predictors of school con-nectedness. Higher levels of connectedness have beenfound in younger students, those from 2-parent families,males, students with a high grade point average, studentswith more educated parents, and students with a largenumber of friends, while lower levels of connectednesshave been found among black students, students whofailed to participate in extracurricular activities, and thosewho skipped school.4,5,8 At the school level, connectednesswas higher in schools with well-managed classrooms,a greater percentage of Hispanic students, a greater per-centage of 2-parent families, and a greater percentage ofstudents participating in extracurricular activities, whileconnectedness was lower in schools with harsh disciplinarypractices, urban schools, and larger schools.4,5,8

The goal of the present study was to examine factorsassociated with school connectedness, specifically charac-teristics of students, schools, and school neighborhoods.The primary source of data was the 2001-2002 HealthBehavior in School-Aged Children (HBSC) Study, with across-sectional, school-based sample of students in grades6-10 in the United States. The HBSC was first conductedin a handful of European nations in 1982, and by the2001-2002 school year there were 35 member nations offi-cially involved in the HBSC.9,10 The HBSC was conductedin collaboration with the World Health Organization.

The analysis included variables that were related toschool connectedness in past studies. At the student level,these were gender, age, urbanicity, student academic per-formance, 2-parent household, parental education, race/ethnicity, and participation in extracurricular activities. Atthe school level, school size and percent of students indifferent racial/ethnic groups were examined.

This study also tested hypotheses that were not exam-ined in past work on school connectedness. Specifically,at the student level, the analysis included immigrant sta-tus, parental involvement in the school, self-rated physicalattractiveness, and friendships with other students of thesame sex and the opposite sex; at the school level, it

Douglas R. Thompson, PhD, Senior Research Scientist (dthompson@mmri.org), Maryland Medical Research Institute, 600 Wyndhurst Ave,Baltimore, MD 21210; Ronaldo Iachan, PhD, Senior Statistician (ronaldo.iachan@orcmacro.com); James G. Ross, MS, Senior Vice President(james.g.ross@orcmacro.com); and Lori A. Gross, MHS, Project Manager(lori.gross@orcmacro.com), ORC Macro, 11785 Beltsville Drive,Calverton, MD 20705; and Mary Overpeck, DrPH, Epidemiologist(overpecm@hrsa.gov), Health Resources and Services Administration,HRSA/MCHB Office of Data and Program Development, 5600 FishersLane room 18-41, Rockville, MD 20857. The US component of the HBSCStudy was funded under contract from the Maternal and Child HealthBureau, Health Resources and Services Administration, Department ofHealth and Human Services to ORC Macro (Macro International Inc), Con-tract No. FY01-PAJ-003. The National Institution for Child Health andHuman Development at the National Institutes of Health contributed partof the funds to support this contract. The authors wish to thank the 340schools who served as hosts for the study and members of the HBSC Coor-dinating Center in Edinburgh, Scotland; the International Data Coordinat-ing Center in Bergen, Norway; and the World Health Organizationin Copenhagen, Denmark. For additional information about the HBSC,contact Dr. Erio Ziglio, Head, WHO European Office for Investmentfor Health and Development (ezi@ihd.euro.who.int); Dr. Candace Currie,Director, Child and Adolescent Health Research Unit, University ofEdinburgh (candace.currie@ed.ac.uk); and Vivian Barnekow Rasmussen,Technical Advisor, Promotion of Young People’s Health, WHO RegionalOffice for Europe (vbr@euro.who.int).

Journal ofSchool Health d September 2006, Vol. 76, No.7 d � 2006, American School Health Association d 379

included household wealth of the student population. Thehypothesized relations between each of these variablesincluded in the study and school connectedness aredescribed in Table 1.

No past work on school connectedness considered theinfluence of the neighborhood around the school. Theschool neighborhood may be important because students’feelings of safety and impressions of school as a stable,reliable environment may be affected by the school neigh-borhood. In this study, school neighborhood characteris-tics measured in the decennial census were used asindicators of wealth and poverty in the neighborhood andtransiency of the local population.

METHODData

This research drew upon linked data files constructedfor continued use of the funding agency, the Health Re-sources and Services Administration. The files includeddata compiled at the student level, school level, and dis-trict level. Data sources included the HBSC Study; theQuality Education Data (QED) school database, whichdescribes 67,345 schools in the United States; and theCommon Core of Data and the Private School Survey,compiled by the National Center for Education Statistics.In addition, census data were linked to schools and dis-tricts based on geographic location.

The target population for HBSC was public and privateschool students in grades 6-10 in the United States. Amultistage stratified sampling strategy was developed tomeet precision requirements. The sample had 2 compo-nents. First, there was a main US sample with no over-sampling of minorities. This sample selected primarysampling units consisting of school districts (some verylarge districts were split), with probabilities proportionalto enrollment in the target grades, and schools wereselected by grade with probabilities proportional to enroll-ment in each grade. Second, a supplemental sample wasdesigned to oversample minorities. The supplementalsample included with greater probabilities areas andschools with greater concentrations of African Americanand Hispanic students.

Data collection began in October 2001, and ended inMay 2002. Of 548 schools contacted, 340 (62%) chose toparticipate. Within the participating schools, 18,620 stu-dents were selected and 15,245 (82%) participated. TheQED data were available for 339 of 340 schools (99.7%)in the US HBSC sample. The Census 2000 short- andlong-form data (available for download from www.cen-sus.gov) were available for 330 of the 340 schools(97.1%) in the HBSC sample (for 10 schools, school geo-graphic location data in the frame based on the QED fileswere not sufficient to match to census tracts). Theresearch protocol was approved by the InstitutionalReview Boards of ORC Macro (Federalwide AssuranceNumber FWA00000845) and the National Institute ofChild Health and Human Development. All schools weregiven the option of using active or passive parental per-mission forms depending on school district policy. Anactive form must be returned with a parent’s signatureand the ‘‘yes’’ box checked for a student to be allowed

to participate. A passive form must be returned with aparent’s signature and the ‘‘no’’ box checked to prevent astudent’s participation. Both variations meet the require-ments of the No Child Left Behind Act, which for studiesnot funded by the US Department of Education merelyrequires that parents be offered a means of ‘‘opting out’’on behalf of their child. Emancipated minors may signtheir own parental permission form. Before participatingin the survey, students with parental permission were fullyinformed and invited to assent to the survey without pen-alty for opting against participation.

MeasuresThe study examined associations between school con-

nectedness (the dependent variable) and student, school,and school neighborhood characteristics.

The HBSC student questionnaire consisted of 59 itemson nominal or ordinal scales (2-10 response categoriesper item), primarily about health, social relationships, andexperiences in school. Several past studies of school con-nectedness2,5-8 used data from the National LongitudinalStudy of Adolescent Health (AddHealth11). In the AddHealthstudies, school connectedness was defined as the sum ofresponses indicating agreement with statements including‘‘I feel close to people at this school,’’ ‘‘I feel like I ampart of this school,’’ ‘‘I am happy to be at this school,’’‘‘The teachers at this school treat students fairly,’’ and‘‘I feel safe in my school’’ (studies differed in the subsetof items used). The HBSC did not have the same measuresas AddHealth, but the measures were similar enough todefine school connectedness in a comparable fashion. Spe-cifically, school connectedness in HBSC was defined as thesum agreement with the statements ‘‘I feel safe at thisschool,’’ ‘‘When a student in my class(es) is feeling down,someone else in class tries to help,’’ ‘‘The students in myclass(es) enjoy being together,’’ ‘‘Most of the students inmy class(es) are kind and helpful,’’ ‘‘Other students acceptme as I am,’’ (each item scored 1 ¼ ‘‘strongly disagree’’ to5 ¼ ‘‘strongly agree’’), plus the response to ‘‘How do youfeel about school at present?’’ (collected on a scale of1 ¼ ‘‘I don‘t like it at all’’ to 4 ¼ ‘‘I like it a lot,’’ multipliedby 1.25 to make the scale roughly comparable to measurescollected on a 1-5 scale). Higher ratings indicated a greaterdegree of school connectedness. A student with missingdata for any item of the scale received a code of missingfor the entire scale. Cronbach’s alpha (0.77) indicatedadequate scale reliability. The validity of these items wasnot specifically investigated in HBSC, but past studiesusing very similar or identical items found that they weresignificantly associated with a variety of student healthoutcomes,2,5,6 supporting criterion validity.

Variables that were examined as potential predictors ofschool connectedness are listed in Table 1. Student char-acteristics were coded based on students’ responses to theHBSC survey. School characteristics were culled fromQED, and school neighborhood characteristics werederived from census data.

AnalysisSchool connectedness was modeled as a function of

student, school, and school neighborhood characteristics.School connectedness was treated as a continuous

380 d Journal of School Health d September 2006, Vol. 76, No.7 d � 2006, American School Health Association

Table1

VariablesIncludedin

theAnalysis,HypothesizedRelationWithSchoolConnectedness,Measure(s)andCoding,andUnivariate

Statistics*

(Continuedonnextpage)

Hypothesis

(BasedonPastResearchWhere

Available)

Measure(s)andCoding

Meanor%

(SE)

Studentcharacteristics

Youngerstudents

willexhibitmore

connectedness

Age13oryounger(¼

1)vs14orolder(¼

0)

52.4%

(1.8)

AfricanAmericanstudents

willexhibitlessandHispanic

students

more

schoolconnectedness

AfricanAmerican(¼

1)vsother(¼

0)

17.3%

(1.8)

Hispanic(¼

1)vsother(¼

0)

12.6%

(1.1)

Students

whoare

born

outsidetheUnitedStatesmay

feellesspartoftheschool,dueto

languageor

culturald

ifferences

Born

inUnitedStates(¼

1)vsother(¼

0)

94.3%

(0.5)

Femaleswillexhibitlessschoolconnectedness

Female

(¼1)vsother(¼

0)

51.9%

(0.6)

Students

livingin

urbanareaswillexhibitless

schoolconnectedness

Urbanresidence(¼

1)vsother(¼

0)

36.6%

(2.6)

Ruralresidence(¼

1)vsother(¼

0)

33.7%

(2.9)

Students

from

2-parenthouseholdswillexhibitmore

schoolconnectedness

2parents

inthestudent’sprimary

residence

(¼1)vsfewerthan2parents

inprimary

residence(¼

0)

74.8%

(0.9)

Students

withmore

educatedparents

willexhibitmore

schoolconnectedness

College-educatedparent(1

orboth

parents;¼1)vsother(¼

0)

43.8%

(1.2)

Parentalinvolvementwithschoolm

ayencourage

students

tovalueschoolm

ore

highly

Parents

talkwithteachers:‘‘stronglyagree’’

thatparents

are

willingto

cometo

school

totalkwithteachers

(¼1)vsother(¼

0)

43.2%

(0.6)

Parents

help

withhomework:‘‘stronglyagree’’

thatparents

are

willingto

help

withhomework

(¼1)vsother(¼

0)

53.0%

(0.7)

Diligentstudents

withgoodacademicperform

ancewill

exhibitmore

connectedness

Self-ratedacademicperform

ancerelativeto

otherstudents

intheschoolis‘‘very

good’’

(¼1)vsother(¼

0)

25.9%

(0.7)

Homework:studentspends1ormore

hours

eachday

(includingweekends)onhomework

(¼1)vsother(¼

0)

35.0%

(1.0)

Lackofextracurricularactivitywillberelatedto

less

schoolconnectedness

Frequentextracurriculars:studentisinvolvedin

clubsor

organizations5ormore

days/week(¼

1)vsother(¼

0)

25.8%

(0.8)

Noextracurriculars:studentisinvolvedin

clubsor

organizations0dayseachweek(¼

1)vsother(¼

0)

29.3%

(0.8)

Students’feelingsabouttheirphysicala

ppearancemay

indicate

socialp

opularity

andconsequently

schoolconnectedness

Goodlooking:self-ratedappearanceis‘‘very

good

looking’’or‘‘quitegoodlooking’’(¼

1)vsother(¼

0)

45.7%

(0.9)

Havingmore

friendsofthemore

‘‘connected’’gender

(males,accordingto

pastwork)mayhaveanespecially

greatim

pactonthelessconnectedgender(females)

31

closemale

friends(¼

1)vsother(¼

0)

74.9%

(0.5)

31

closefemale

friends(¼

1)vsother(¼

0)

76.0%

(0.4)

Journal of School Health d September 2006, Vol. 76, No.7 d � 2006, American School Health Association d 381

Table

1VariablesIncludedin

theAnalysis,HypothesizedRelationWithSchoolConnectedness,Measure(s)andCoding,andUnivariate

Statistics*

(Continuedfrom

previouspage)

Hypothesis

(BasedonPastResearchWhere

Available)

Measure(s)andCoding

Meanor%

(SE)

Schoolcharacteristics

Schoolswithwealthierstudents

mayhavelowerlevelsof

socialp

roblemsthatare

associatedwithlowlevelsof

schoolconnectedness,forexample,deviantbehavior

andteenpregnancy

Wealthystudents:RWIabove84(the75th

percentile

forschoolsin

theQEDdatabase)

(¼1)vsother(¼

0)

Poorstudents:RWIbelow49

(the25th

percentile

forschoolsin

the

QEDdatabase)(¼

1)vsother(¼

0)

6.8%

(1.5)

49.6%

(5.4)

Students

inlargerschoolswillexhibitlessconnectedness

Schoolsize:naturallogofthenumberofstudents

intheschool

Raw:1041(41);

log:6.7

(0.04)

Schoolswithmore

Hispanicstudents

willexhibitmore

Percentblackstudents

intheschool

19.8

(1.43)

connectedness,while

thosewithmore

African

Americanstudents

willexhibitlessconnectedness

PercentHispanicstudents

intheschool

15.3

(1.27)

Schooln

eighborhoodcharacteristics

Poorerneighborhoodsmayhavehigherlevelsof

Medianhousehold

incomein

theneighborhood

40696.0

(1030.38)

socialp

roblems(eg,crime)thatcontribute

tostudents

feelinglesssafe,thuslessconnected

totheschool

Thepercentageofindividuals

intheneighborhoodbelowthe

povertylevel

15.7

(0.64)

Inneighborhoodswhere

residents

haveatendency

Percentrenters

intheneighborhood

35.1

(1.19)

tobetransient,students

aswellasschoolstaff

mayfeellessattachedto

theschool,dueto

their

recentarrivala

nd/ortheperceptionthattheymay

moveawaysoon

Percentnon-U

Scitizensin

theneighborhood

6.6

(0.50)

*Statisticsforstudentcharacteristicswere

weightedandthesurveydesignwastakeninto

accountin

thecomputationofvarianceestimates.Theestimatesforschoolswere

unweightedsample

estimates.

RWI:RelativeWealthIndication.

382 d Journal ofSchool Health d September 2006, Vol. 76, No.7 d � 2006, American School Health Association

variable. All student characteristics were collected onordinal or nominal scales in HBSC; therefore, they wererepresented as binary or as sets of indicator variables. Atthe school level, wealth was coded as 2 binary variablesto indicate schools with relatively wealthy students (abovethe 75th percentile of the Relative Wealth Indicator(RWI) of all schools in the QED database) and schoolswith relatively poor students (below the 25th percentile).The RWI is calculated as 100 3 (1 � [the number of stu-dents falling below the federal poverty guidelines/the totalnumber of students in a district’s boundaries]). Schoolsize and percent African American were coded, respec-tively, as the natural logarithm of number of students anda quadratic term for percent African American, to main-tain comparability with past work.8 A quadratic term forpercent Hispanic students was also tested. All communitycharacteristics were represented as continuous. Becausepreliminary graphs indicated nonlinearity for percentrenters and percent non-US citizens, these were representedwith splines to allow the slope to change at differentpoints on the scale.12 For percent renters, there were 2 splineknots (25% and 40%) and there was a single knot forpercent non-US citizens (10%). The model included1 hypothesized interaction, gender by friendships, as ex-plained in Table 1.

Hierarchical linear modeling (HLM13,14) was used toaccount for the nonindependence of students within schoolsand school districts. All models included a random inter-cept. It was of particular interest to explain why schoolsvaried in mean ratings of connectedness. Therefore, a seriesof nested models was constructed to determine what per-centage of the variance in connectedness among schoolsand districts was explained by student, school, and schoolneighborhood characteristics. The relative fit of each modelwas assessed using the likelihood ratio chi-square. PROCMIXED in SAS version 9.1 (SAS Institute, Cary, NC) wasused to conduct the analysis.15,16

RESULTSOf the 15,245 HBSC respondents, 427 (3%) were

excluded from the present analysis due to excessive miss-ing or unreliable data and an additional 1601 (11%) wereexcluded due to missing data on items comprising theschool connectedness scale. Among those for whomschool connectedness could be computed (n ¼ 13,207),the mean was 21.66 (standard error ¼ 0.10; 25th percen-tile ¼ 18.75, median ¼ 22.00, 75th percentile ¼ 24.75).Univariate statistics for school, student, and school neigh-borhood characteristics are provided in Table 1.

Four models were constructed: (1) A model with a ran-dom intercept but no other predictors, (2) a model withstudent characteristics added, (3) a model adding schoolcharacteristics, and (4) a model adding school neighbor-hood characteristics. Below, results from the final model(model 4) are described. The primary purpose of the othermodels (models 1-3) was to estimate the variance inschool connectedness explained by the set of characteristicsadded at each level (student, school, school neighborhood).

The results of models 1-4 are shown in Table 2. Thelikelihood ratio chi-square tests (ps , .0001) indicate thateach model was a significant improvement over the previ-ous one.

Random Effects (School-Level Variance Estimates)Initial models included separate random effects for

schools and districts. However, variance at the districtlevel was nonsignificant and including it did not improvethe model fit; therefore, the final version of all modelsonly included random effects for schools. The school-level variance in model 1 (random intercept but no otherpredictors) was 1.77. This is an estimate of the unadjustedvariance among schools in mean connectedness. Addingthe student characteristics (model 2) explained 38.4% ofthe variance among schools, adding school characteristics(model 3) explained 21.1% of the remaining variance,and adding school neighborhood characteristics (model 4)explained 15.1% of the remaining variance. The entire setof predictors together explained 59.3% of the varianceamong schools in connectedness (compare the varianceestimates of model 1 and model 4).

Fixed EffectsThe fixed-effect estimates are interpreted as regression

coefficients. The estimate indicates the predicted increasein school connectedness (or decrease, if the estimate isnegative), given a 1-unit increase in the correspondingindependent variable (predictor). Only significant effects(estimates significantly different from 0, p , .05) in thefull model (model 4) are described below.

Student Characteristics. School connectedness washigher among younger students, Hispanic students, fe-males, and students from 2-parent households. Connected-ness was also higher among students who strongly agreedthat their parents were willing to talk with teachers andamong students whose parents helped them with home-work. ‘‘Very good’’ academic performance, diligent home-work, and high levels of extracurricular participation werepredictive of greater school connectedness, while partici-pating in no extracurriculars was predictive of less con-nectedness. Being very/quite good looking was predictiveof greater connectedness. Gender interacted with friend-ships. Among males, close friendships with �3 males and�3 females predicted approximately equal increases inschool connectedness (b ¼ .92 and .86 for close friend-ships with �3 males and females, respectively), whileamong females, close friendships with �3 males wereless important (predicted increase in school connected-ness ¼ .92 � .51 ¼ .41), but close friendships with otherfemales appeared much more important (predicted in-crease in school connectedness ¼ .86 1 .57 ¼ 1.43).

School Characteristics. In schools with more studentsfrom relatively wealthy households, connectedness wasgreater, while in larger schools there was less connec-tedness. For percent of black students, the significantquadratic term indicates a U-shaped trend where schoolswith either a large or small percentage of black studentshad the highest levels of connectedness and schoolswith a more diverse racial mix tended to exhibit lessconnectedness.

School Neighborhood Characteristics. Two schoolneighborhood measures (percent renters and percent non-US citizens) were predictive of connectedness. In neigh-borhoods with few renters (0-25% renters), percentage ofrenters bore little relation with school connectedness(slope ¼ .02, nonsignificant). However, in communities

Journal of School Health d September 2006, Vol. 76, No.7 d � 2006, American School Health Association d 383

with between 26% and 40% renters, there was a sharp drop inconnectedness (slope ¼ .02 � .08 ¼ �.06), which remaineduniformly low above 40% renters (Figure 1). For non-UScitizens, there was an increase in connectedness between 0%and 10%, which leveled off above 10% (Figure 2).

Post hoc power analysis using Monte Carlo simula-tions17 was done to quantify power to detect associationswith school connectedness in HBSC. The power analysesassumed HLM models with a random intercept for eachschool, controlling for student demographics and RWI at

Table 2Parameter Estimates (SE) and Significance of Estimates in 4 Sequential Models of School Connectedness

Parameter or Fit Indicator Model 1 Model 2 Model 3 Model 4

Model fit�2 log likelihood 77,618 68,098 68,045 68,011Likelihood ratio v2 N/a 9520***** (df ¼ 20) 53***** (df ¼ 7) 34***** (df ¼ 7)

Random effectsVariance 1.77 (0.191)***** 1.09 (0.135)***** 0.86 (0.114)***** 0.73 (0.101)*****

Fixed effectsIntercept 21.61 (0.089)***** 17.19 (0.268)***** 18.95 (0.862)***** 19.03 (0.970)*****

Student characteristicsAge , 14 0.63 (0.111)***** 0.58 (0.114)***** 0.58 (0.112)*****African American �0.24 (0.132)* �0.05 (0.142) �0.03 (0.142)Hispanic 0.48 (0.136)**** 0.47 (0.144)*** 0.44 (0.144)***Born in United States 0.23 (0.175) 0.18 (0.175) 0.19 (0.176)Female 0.45 (0.207)** 0.48 (0.207)** 0.47 (0.207)**Urban �0.03 (0.112) 0.03 (0.112) 0.07 (0.113)Rural �0.01 (0.121) �0.05 (0.123) �0.03 (0.125)2-parent household 0.40 (0.095)***** 0.38 (0.095)***** 0.39 (0.095)*****College-educated parent �0.05 (0.084) �0.09 (0.084) �0.09 (0.084)Parents willing to talk to teachers 0.76 (0.089)***** 0.76 (0.088)***** 0.76 (0.088)*****Parents willing to help with homework 1.81 (0.089)***** 1.81 (0.089)***** 1.82 (0.089)*****School performance ‘‘very good’’ 1.02 (0.092)***** 1.03 (0.092)***** 1.03 (0.092)*****Homework 11 hours/day 0.47 (0.085)***** 0.45 (0.086)****** 0.44 (0.086)*****Extracurricular activities: 51 days/week 0.20 (0.097)** 0.20 (0.097)** 0.20 (0.097)**Extracurricular activities: never �0.53 (0.097)***** �0.50 (0.098)***** �0.50 (0.097)*****Very/quite good looking 0.77 (0.082)***** 0.78 (0.082)***** 0.78 (0.082)*****31 close male friends 0.95 (0.162)***** 0.92 (0.162)***** 0.92 (0.162)*****31 close female friends 0.85 (0.130)***** 0.86 (0.130)***** 0.86 (0.130)*****Female 3 31 close male friends �0.54 (0.202)*** �0.52 (0.202)** �0.51 (0.202)**Female 3 31 close female friends 0.61 (0.210)*** 0.57 (0.210)*** 0.57 (0.210)***

School characteristicsWealthy students 0.76 (0.277)*** 0.61 (0.283)**Poor students �0.11 (0.191) �0.05 (0.189)Natural logarithm of number of students �0.22 (0.120)* �0.25 (0.120)**% black �0.04 (0.010)**** �0.03 (0.010)***% blacky 0.00 (0.000)*** 0.00 (0.000)***% Hispanic 0.02 (0.011) 0.01 (0.012)% Hispanicy �0.00 (0.000) �0.00 (0.000)

School neighborhood characteristicsMedian income (thousands) �0.00 (0.007)% below poverty level �0.01 (0.011)% renters 0.02 (0.017)25, % renters < 40 �0.08 (0.027)***% renters. 40 0.06 (0.020)***% non-US citizens 0.07 (0.029)**% non-US citizens . 10 �0.05 (0.037)

***** p, .0001; **** p , .001; *** p , .01; ** p , .05; * p, .10.y Percentage squared (quadratic term).N/a: not applicable; SE: standard error.

384 d Journal of School Health d September 2006, Vol. 76, No.7 d � 2006, American School Health Association

the school level, with the same sample size as HBSC. Thesimulations indicated that for student-level variables,school connectedness differences of 0.3 and 0.4 could bedetected with power ¼ 0.80 and 0.90, respectively, whilefor school-level variables, school connectedness differencesof 0.6 and 0.7 could be detected with power ¼ 0.80/0.90.

DISCUSSIONThe objective of this study was to extend the findings

of past studies of school connectedness, using a large,nationally representative school-based sample providinginformation that was not examined in past work on schoolconnectedness. Like past studies,4,8 school connectednesswas greater among younger students, students in 2-parenthouseholds, students who performed well in school, stu-dents who participated in many extracurricular activities,and students with a large number of friends. Also consistentwith past work,4,5 at the school level, there was less con-nectedness in larger schools, and there was a U-shapedrelation between percent African American students andconnectedness.

In contrast to other studies,4,8 the present study showedthat connectedness was higher among females (past workfound more connectedness among males) and school con-nectedness was not significantly associated with being anAfrican American (defined at the student level), living inan urban area, or having highly educated parents. Failure toreplicate some findings may be due to the fact that themodels in the present study included variables that werenot examined in past work. Possibly, the inclusion of addi-tional predictors in this study accounted for some of the as-sociations that were found in past work. However, thisstudy differed from past work in several ways (the sample,data collection methods, specific wording of survey items,and analytic methodology); thus, the source of differencesin results cannot be known with certainty.

In addition to examining predictors of school connect-edness that were examined in past work, the present studyexamined other potential predictors. The results supportedthe hypothesis that parental involvement in school pre-dicts connectedness. Further, good-looking students (pos-sibly indicating social popularity) felt more connected. Ashypothesized, the association of friendships with connect-edness depended on gender—close friendships with malesand females were about equally important for male stu-dents, whereas friendships with other females predictedmuch greater levels of connectedness for female students.As expected, connectedness was higher in schools withrelatively wealthy students.

Above and beyond student and school characteristics,school neighborhood characteristics contributed to theprediction of school connectedness. As expected, con-nectedness was lower in neighborhoods with a large per-centage of renters (.25%). The models includedindicators of wealth and urbanicity; thus, this associationcannot simply be due to the fact that neighborhoods withmore renters are more likely to be urban or they havemore (or less) wealthy residents. Possibly, students inneighborhoods with more renters are more likely toperceive the school as a temporary, transient environ-ment—they may be moving away soon, or their friends,staff members, and/or teachers may be likely to moveaway. Contrary to the hypothesis that percent of non-UScitizens would be an indicator of transiency and thuspredictive of less connectedness, neighborhoods withmore non-US citizens exhibited higher levels of schoolconnectedness.

A general theme emerging from the results is thatpopulations traditionally considered ‘‘minorities’’ often

Figure 1Predicted School Connectedness by Percent

Renters in the Community

21.2

21.4

21.6

21.8

22

22.2

22.4

22.6

22.8

0Percent renters in community

Pre

dic

ted

co

nn

ecte

dn

ess

10 20 30 40 50 60 70 80 90 100

Note: Predicted connectedness was estimated byvarying the percent of renters in the communitybetween 0 and 100, holding all other variablesconstant at their respective modes (categoricalvariables) or means (continuous variables).

Figure 2Predicted School Connectedness by Percent

Non-US Citizens in the Community

18.5

19

19.5

20

20.5

21

21.5

22

Percent non-US citizens

Pred

icted

co

nn

ected

ness

0 10 20 30 40 50 60 70 80 90 100

Note: Predicted connectedness was estimated byvarying the percent of non-US citizens in the

community between 0 and 100, holding all othervariables constant at their respective modes(categorical variables) or means (continuous

variables).

Journal ofSchool Health d September 2006, Vol. 76, No.7 d � 2006, American School Health Association d 385

exhibited relatively high levels of school connectedness.Controlling for economic and family factors (number ofparents in the household and parental involvement withschool), connectedness was higher among Hispanic stu-dents and students in schools with predominantly AfricanAmerican students. Although at the student level, beingborn in another country was unrelated to connectedness,schools in neighborhoods with a large percentage of non-US citizens exhibited relatively high levels of connected-ness. Some past studies found that connectedness waslower among minorities; in part, this may have been dueto failure to control for some important economic or fam-ily factors in past work.

Several limitations of the study should be noted. First,the study was based on a nonexperimental design. Conse-quently, there is uncertainty about the mechanisms under-lying the reported associations. Second, relatively littleinformation on school neighborhoods was available; thedecennial census lacked some potentially useful measuressuch as neighborhood crime and neighborhood residents’attitudes about education. Third, the study had excellentpower to detect statistical associations with school con-nectedness, but some of the effect sizes were small. Forexample, the estimated coefficient for frequent extracur-ricular activities (b ¼ 0.20) indicates that, other thingsbeing equal, students who often engage in extracurricularsexhibit school connectedness about one fifth of a pointhigher than students who seldom or never do such activi-ties. Future work is needed to determine the practical sig-nificance of these associations, that is, whether theypredict meaningful differences in student health andwell-being.

The results of this study suggest possible strategies topromote school connectedness. Encouraging greaterparental participation with school activities may be 1 wayto boost connectedness. Encouraging participation inextracurricular activities may also help to foster connect-edness. Targeted efforts may be needed to increase con-nectedness in certain kinds of schools, particularly thosethat are heterogeneous with respect to African Americanstudents (eg, those where around 50% of the students areblack), those that do not have wealthy student popula-tions, and large schools. The findings on neighborhoodcharacteristics may have policy implications: if the neigh-borhood in which a school is located has a significantimpact on connectedness, this may indicate that certain

kinds of neighborhoods should receive major consider-ation in deciding where to build new schools. j

References1. Libbey HP. Measuring student relationships to school: attachment,

bonding, connectedness and engagement. J Sch Health. 2004;74:274-283.

2. Resnick MD, Bearman PS, Blum RW, et al. Protecting adolescentsfrom harm: findings from the National Longitudinal Study of AdolescentHealth. JAMA. 1997;278:823-832.

3. Eccles JS, Early D, Frasier K, et al. The relation of connection, reg-ulation, and support for autonomy to adolescents’ functioning. J AdolescRes. 1997;12:263-286.

4. Bonny AE, Britto MT, Klosterman BK, et al. School disconnected-ness: identifying adolescents at risk. Pediatrics. 2002;106:1017-1021.

5. Blum RW, McNeely CA, Rinehart PM. Improving the Odds: TheUntapped Power of Schools to Improve the Health of Teens. TechnicalReport. Minneapolis: Center for Adolescent Health and Development,University of Minnesota; 2002.

6. McNeely C, Falci C. School connectedness and transitions into andout of health-risk behavior among adolescents: a comparison of socialbelonging and teacher support. J Sch Health. 2004;74:284-292.

7. Jacobson KC, Rowe DC. Genetic and environmental influences onthe relationships between family connectedness, school connectedness, andadolescent depressed mood: sex differences. Dev Psychol. 1999;35:926-939.

8. McNeely CA, Nonnemaker JM, Blum RW. Promoting school con-nectedness: evidence from the National Longitudinal Study of AdolescentHealth. J Sch Health. 2002;72:138-146.

9. Currie C, Roberts C, Morgan A, Smith R, Settertobulte W, Samdal O,Rasmussen VB, eds. Young People’s Health in Context: International Reportfrom the HBSC 2001/02 survey. WHO Policy Series: Health Policy forChildren and Adolescents Issue 4. Copenhagen, Denmark: WHO RegionalOffice for Europe; 2004.

10. World Health Organization. The WHO cross-national study ofhealth behavior in school-aged children from 35 countries: findings from2001-2002. J Sch Health. 2004;74:204-206.

11. Bearman PS, Jones J, Udry JR. The National Longitudinal Studyof Adolescent Health research design. Available at: http://www.cpc.unc.edu/projects/addhealth/design.html. Accessed September 15, 2004.

12. Marsh LC, Cormier DR. Spline Regression Models. ThousandOaks, Calif: Sage; 2002.

13. Bryk AS, Raudenbush SW. Hierarchical Linear Models. NewburyPark, Calif: Sage; 1992.

14. Snijders T, Bosker R. Multilevel Analysis. Thousand Oaks, Calif:Sage; 1999.

15. Singer JD. Using SAS PROC MIXED to fit multilevel models,hierarchical models, and individual growth models. J Educ Behav Stat.1998;24:323-355.

16. Littell RC, Milliken GA, Stroup WW, et al. SAS System for MixedModels. Cary, NC: SAS Institute; 1996.

17. Muthen LK, Muthen BO. How to use a Monte Carlo study todecide on sample size and determine power. Struct Equation Model.2002;4:599-620.

386 d Journal of School Health d September 2006, Vol. 76, No.7 d � 2006, American School Health Association