Table 1

Summary of included studies (n=10)

Author(s) (study name)CountryStudy designSample size and characteristicsData collection methodsMethod(s) for collecting and summarising social network dataMethod(s) for analysing social network dataEating behaviour and/or bodyweight measuresKey findings
Cohen-Cole and Fletcher (2008)22 (National Longitudinal Study of Adolescent Health (Add Health))USALongitudinalSchools: N/S; students: n=1988School-based and in-home surveys
  • Respondents asked to list up to 10 of their closest friends (five males, five females)

  • Dyads constructed based on ‘first friend named’ by each subject

Logistic and OLS regressionObesityOdds that an ego becomes obese if his or her ‘best friend’ is obese: unadjusted=0.8, after adjusting for school-specific trends=0.51, after adjusting for school-specific trends and individual fixed effects=NSE
de la Haye et al (2009)30AustraliaCross-sectionalSchools: n=2; students: n=385 (age 13–15 years; 202 males, 183 females)School-based surveyRespondents listed the first and last names of all their ‘close friends’ in their school grade (defined as friends they ‘hang around with’ most)Exponential random graph modelling, stratified by genderHigh-calorie food consumption (fast food, savoury snack foods, sweet snack foods, high-calorie drinks)
  • Girls: no significant within-group similarity

  • Boys: (1) fast food consumption: significant within-group similarity in 2/3 networks (MPEADE −0.08 to −0.12); (2) sweet snack food consumption: significant within-group differences in 1/3 networks (MPEADE 0.16); (3) savoury snack food consumption: no significant differences; (4) high-calorie drink consumption: no significant differences

Fowler and Christakis (2008)23 (Add Health)USALongitudinalSchools: N/S; students: n=1988School-based and in-home surveysRespondents asked to list up to 10 of their closest friends (5 male, 5 female)Logistic regressionObesity
  • Ego becomes obese if his or her ‘best friend’ is obese after adjusting for school-specific trends (coefficient 0.033, SE 0.014, p=0.020)

  • Evidence of ‘directional effects’ (that is, influence): ‘named friends’ not influenced by ‘namers’ (p=0.90)

Halliday and Kwak (2009)25 (Add Health)USALongitudinalSchools: n=16; students: wave 1, n=4617; wave 2, n=2970 (mean age 15 years; 48% male, 52% female; ethnicity: 55% Caucasian, 19% African–American, 12% Asian; SES: single-parent household ∼30%)School-based and in-home surveysRespondents asked to list up to 10 of their closest friends (5 male, 5 female)Logistic regression (school fixed effects)
  • BMI

  • Overweight: BMI at or above 85th percentile

  • Underweight: BMI below the 10th percentile

  • Correlations between own BMI and average peer BMI

  • BMI: girls: 10 point increase in friend BMI associated with 1.9 increase in own BMI; boys: 10 point increase in friend BMI associated with 1.8 increase in own BMI

  • Overweight: girls: 10% increase in friends' propensity to be overweight is associated with 1.1% increase in own propensity to be overweight; boys: 10% increase in friends' propensity to be overweight is associated with 0.8% increase in own propensity to be overweight

  • Underweight: no correlation

  • Determinants of changes in own BMI: adolescent who nominate overweight friends at wave 1 most likely to gain weight by wave 2

Hutchinson and Rapee (2007)29AustraliaCross-sectionalSchools: n=10; students: n=1094 (mean age 12.3 years; 100% female; SES: 84% born in Australia, 83% English first language, 82% two-parent family)School-based survey
  • Respondents asked to identify friends who they spend most time with at school from the Grade 7 roster

  • Friendship cliques (N=173) identified using UCINET software

MANCOVA within friendship cliques
  • BMI

  • Dieting

  • Extreme weight-loss behaviours

  • Binge eating

  • Body image concern

  • BMI: significant within-group similarity (effect size 1.69, p=0.000)

  • Dieting: significant within-group similarity (effect size 0.26, p=0.001)

  • Extreme weight loss behaviours: significant within-group similarity (effect size 0.27, p=0.020)

  • Binge eating: significant within-group similarity (effect size 0.24, p=0.026)

  • Body image concern: no significant differences

Paxton et al (1999)28AustraliaCross-sectionalSchools: n=6; students: n=523 (age 15–16 years, 100% female)School-based survey
  • Respondents asked to identify girls who they spend most time with at school from the Grade 10 roster

  • Friendship cliques (n=79) identified using UCINET software

  • MANCOVA within and between friendship cliques

  • Hierarchical regression models of friendship variables

  • BMI

  • Body image concerns

  • Dieting

  • Extreme weight loss

  • Binge-eating

  • Intraclique versus interclique variability:

  • (1) BMI: significantly higher between-group than within-group variance

  • (2) Body image concerns: significantly higher between-group than within-group variance (p=0.002)

  • (3) Dieting: significantly higher between-group than within-group variance (p=0.024)

  • (4) Extreme weight loss: significantly higher between-group than within-group variance (p=0.001)

  • (5) Binge-eating: no significant differences

  • Friendship variables predicted: 14% variance in body image concerns, 14% variance in dietary restraint, 12% variance in extreme weight-loss behaviours, 12% variance in binge-eating disorders

Pike (1995)26USACross-sectionalSchools: n=3; students: n=410 (mean age 16 years, range 14–19 years; 100% female)School-based survey.Respondents asked to identify up to 6 of their closest female friends
  • Girls were considered part of an individual's friendship network if they reciprocated the nomination of friendship

  • Hierarchical regression modelling, stratified by grade

Bulimic symptoms
  • Bulimic symptoms were predicted by: bulimic symptoms in the friendship network (OR 1.29, p<0.0001); rate of anorexia or bulimia among friends (OR 0.12, 0.0001)

  • Within-network similarities varied by grade: positive association at 9th grade; negative association at 12th grade

Strauss and Pollack (2003)21 (Add Health)USACross-sectionalStudents: n=17 557 (7th to 12th Grade; BME: 18% African–American, 12% Hispanic 12%; SES: 29% single-parent household)School-based and in-home surveys
  • Respondents asked to list up to 10 of their closest friends (5 male, 5 female)

  • Friendship networks were generated using Pajek and SAS/IML software

Multivariate regression analysisOverweight (BMI above 95th percentile)
  • In-degree (mean no. of nominations): girls: 3.41 overweight versus 5.01 normal-weight (p<0.001); boys: 3.38 overweight versus 4.55 normal weight (p<0.001)

  • Popularity (≤5 nominations):girls: 27% overweight versus 47% normal weight (p<0.001); boys: 28% overweight versus 40% normal weight (p<0.001)

  • Reciprocity (best male friend nominates as best friend): girls: 12% overweight versus 29% normal weight (p<0.002); boys: 31% overweight versus 34% normal weight (p=0.3)

  • Reciprocity (best female friend nominates as best friend): girls: 34% overweight versus 46% normal weight (p=0.002); boys: 18% overweight versus 24% normal weight (p=0.049)

  • Centrality (mean Bonacich score): girls: 0.53 overweight versus 0.78 normal weight (p<0.001); boys: 0.54 overweight versus 0.69 normal weight (p<0.001)

  • Social isolation (no nominations): overweight OR 1.71 (CI 1.39 to 2.20)

Trogdon et al (2008)24 (Add Health)USACross-sectionalSchools: n=16; students: n=3702 (mean age 16.1 years)School-based and in-home surveysRespondents asked to list up to 10 of their closest friends (5 male, 5 female)Multivariate regression analysis
  • BMI

  • Overweight (BMI at or above 85th percentile)

  • Friendship nominations: overweight adolescents less likely to be nominated as a friend

  • Influence of friends' weight: mean weight among friends positively associated with adolescent weight (OLS 0.30, p<0.001)

  • Influence of grade-level peers' weight: mean weight among grade-level peers positively associated with adolescent weight (OLS 0.23, p<0.005)

Valente et al (2009)27USACross-sectionalSchools: n=4; students: n=617 (age 11–15 years; 36% male, 64% female; ethnicity: 36% Asian, 30% Hispanic, 12% Caucasian 12%, 12% mixed, 6% other, 3% African–AmericanSchool-based surveyRespondents asked to nominate friends in their classRandom-effect logistic regression modelling, with individuals nested by school-classOverweight (BMI above 95th percentile)
  • Girls: (1) friends' average BMI: positive association with being overweight; (2) number of friends named: positive association with being overweight (OR 1.57, CI 1.01 to 2.46); (3) frequency named as a friend: negative association with being overweight for girls (OR 0.89, CI 0.75 to 1.07)

  • Boys: (1) friends' average BMI: positive association with being overweight; (2) number of friends named: no association with being overweight; (3) frequency named as a friend: no association with being overweight

  • BME, black and minority ethnicity; BMI, body mass index; MANCOVA, multivariate analyses of covariance; MPEADE, model parameter estimates for absolute difference effects; N/S, not stated; NSE, no significant effect; OLS, ordinary least squares; Pajek, large network analysis program (http://pajek.imfm.si/doku.php); SAS/IML, nInteractive matrix programming with integration to R (SAS Institute, Cary, North Carolina, USA); SES, socioeconomic status; UCINET, social network analysis program (Analytic Technologies, Lexington, Kentucky, USA).