Friends and social contexts as unshared environments: a discordant sibling analysis of obesity- and health-related behaviors in young adolescents

Abstract

OBJECTIVE:

This study examines the contribution of best friends’ weight and the peer social context (time spent alone versus with friends) as sources of unshared environment associated with variability in weight and health behaviors among weight-discordant siblings.

METHODS:

Pairs of same-sex biologic siblings (N = 40 pairs; ages 13–17) were originally recruited as part of a study evaluating putative factors contributing to differences in adiposity among weight-discordant siblings. Siblings were asked to bring their best friends to the laboratory and siblings and friends’ height and weight were objectively measured. Siblings also completed multi-pass dietary recalls to assess energy intake and sugar sweetened beverage (SSB) consumption. Siblings’ physical activity was measured using accelerometry. Experience sampling methodology was used to assess sedentary behaviors/screen time and the number of occasions siblings spent alone and in the presence of friends. Multilevel models were used to estimate the relationships between predictors (best friends’ zBMI, time spent alone or with friends) and outcomes (siblings’ zBMI and obesity-related health behaviors).

RESULTS:

Best friends’ zBMI was the best predictor of participants’ zBMI, even when controlling for child’s birth weight. Best friends’ weight (zBMI) further predicted participants’ SSB intake and time engaged in sedentary behaviors. Being active with friends was positively associated with participants’ overall physical activity, whereas spending time alone was negatively associated with accelerometer counts regardless of siblings’ adiposity.

CONCLUSIONS:

A friends’ weight and the social context are unshared environmental factors associated with variability in adiposity among biologically-related weight-discordant siblings.

INTRODUCTION

In their 1987 seminal paper, Plomin and Daniels¹ asked: ‘Why are children in the same family so different from each other?’ The answer the authors proposed is that the unshared environment—a component of phenotypic variance—accounts for differential developmental pathways among siblings raised in the same household. Three decades later, the specifics of genetics and environmental influences on obesity, and how these influences interact, remain unclear (for example, Faith et al.²). From the possible sources of non-shared environmental effects on childhood obesity, differential parental treatment and more specifically maternal feeding practices, is the mechanism that has received the most attention from obesity researchers, including some studies that used a discordant sibling study design.^3–7 Studying same-sex biological siblings discordant for adiposity controls for some potential confounders such as family socioeconomic status and parental weight status. Testing differences between discordant siblings is a more robust design than comparing unrelated non-overweight and overweight youth. Other putative non-shared environmental mechanisms of youth obesity have been studied using a discordant sibling design. Roemmich and colleagues⁸ found no difference in responsivity to dietary variety between obese and non-overweight siblings. However, high food reinforcement and an inability to delay gratification contributed to siblings discordance in obesity.⁸

There is mounting evidence suggesting that youths’ friends and peer relationships are also likely contributors to unshared environmental effects on phenotype including eating,^9–12 physical activity^13,14 and sedentary behaviors.¹⁴ In comparison to the wealth of studies on parent-child relationships, however, relatively little attention has been devoted to the role of friends’ attributes and of the larger social context in explaining siblings’ discordance in adiposity.

The dearth of research on siblings’ unshared peer environment is surprising as there is now an increasing number of experimental and field studies showing that the behavior and characteristics of peers and friends can either promote or hinder children’s and adolescents’ energy balance behaviors.¹⁵ Specifically, spending time alone^14,16 and social exclusion (ostracism) decrease young adolescents’ motivation to be physically active and actual physical activity,¹⁷ and increase overweight youths’ energy intake.^18,19 Overweight youths are also more likely to have overweight friends than their normal weight peers,²⁰ and overweight friends, in turn, tend to share obesity risks^21–23 and co-engage in obesogenic behaviors.²⁴ These socially-mediated experiences likely provide important behavioral variability that contributes to differences in adiposity. Studies so far, however, could not clearly tease apart the effect of peers and friends from the influence of familial risk factors and exogenous contextual effects.²⁵

The present study moves the field forward by examining the associations between best friends’ standardized body mass index (zBMI) and siblings’ zBMI, and obesity-related health behaviors among biologically related weight-discordant siblings (one overweight or obese sibling ((BMI ⩾ 85th percentile) and one non-overweight sibling (BMI < 85th percentile)). The sample of same-sex biologic siblings was originally recruited as part of a larger study evaluating putative factors contributing to differences in energy balance behaviors and adiposity among weight-discordant siblings. The weight discordant sibling design allows for understanding the influence of such putative risk factors by controlling for on average 50% of the genetic variability between siblings and for some degree of the variance associated with shared aspects of the home and non-home environments.

On the basis of previous experimental work indicating that the social context influences youths’ eating and activity behaviors, this study further examines whether the frequency of time spent alone or in the presence of friends are associated with siblings’ health-related behaviors, and whether these relationships differ as a function of siblings’ weight status. It is important to assess whether the social context contributes to variability in obesogenic behaviors because these putative socio-environmental risk factors can be leveraged to promote change. Our focus on young adolescents further increases the significance of this study as peer relationships, or the lack thereof, may be especially influential during early adolescence as youths spend the majority of their waking hours in the company of peers and friends.²⁶ Furthermore, the influence of best friends’ attributes (as opposed to casual acquaintances) may be especially influential as closer and stronger connections or relationships provide broader and stronger possibilities for influence.^27–30

We hypothesize that best friends’ weight (zBMI) and the social context (spending time alone versus with friends) are sources of unshared environmental influence that may contribute to discordance in adiposity among siblings (or ‘egos’ to refer to each individual sibling or focal node in a pair).^1,2,31,32 It is important to note that a moderating effect of ego’s weight status (overweight/obese versus non-overweight) on the relationship between predictors and outcomes is not required for friends and social context to contribute to difference in adiposity among siblings. In other words, we do not hypothesize that best friend’s weight and social context operate differently in non-overweight and overweight or obese siblings. Rather, we hypothesize that the social context of overweight and obese adolescents and their best friends’ attributes are conducive to greater adiposity.

MATERIALS AND METHODS

Participants and procedure

Forty pairs of same-sex biologic siblings (ages 13–17, no more than 4 years apart) were originally recruited as part of a larger study evaluating putative factors contributing to differences in energy balance behaviors and adiposity among weight-discordant siblings. Families were recruited from newspaper advertisements and from a database of families who had inquired about previous studies. Parents were screened by phone for their children’s height, weight, a brief medical history and ethnic background. Children were excluded if they were below the 10th BMI percentile; had current psychopathology or developmental disability; and/or if they were on medications or had conditions that could influence their mobility or their activity level (for example, methylphenidate). If a sibling had a cold or upper respiratory distress they were rescheduled for testing. A total of 930 families contacted the study staff regarding the study (by email or phone). Typically the staff would leave three messages over the course of about 2–3 weeks in an attempt to reach the family and schedule a phone screen. From the original pool, 234 families did not return our call or were unable to complete the initial phone screen. A large number (n = 652) of contacted families were not eligible for the study (for example, did not meet BMI criteria, different parents, medical conditions, only had one child, siblings were twins, one sibling was adopted, siblings were different sex, not interested, did not meet age requirements, enrolled in another study). The remaining 44 families were enrolled in the study and four families dropped out before completion. Interested adolescents and their parents were scheduled to come to the laboratory for an information session. If they agreed to participate, siblings and parents were asked to provide written consent and siblings were trained to complete a series of measures and experimental tasks. As part of the study, siblings were also asked to bring their best friends to the laboratory and siblings and friends’ height and weight were objectively measured as described below. All procedures used in this study were approved by the Social and Behavioral Sciences Institutional Review Board of the University at Buffalo. Parents provided written informed consent for each sibling and the siblings provided assent.

Measures

Baseline participant characteristics.

The child participants were asked to report their demographic information including gender, age, grade-level, school and race/ethnicity. Parents provided parents’ education and household income.

Outcomes

BMI z-score.

Siblings’ height and weight were assessed using an electronic scale (Model BWB-800S, Tanita, Portage, MI, USA) and stadiometer affixed to the wall (Model PE-AIM-101, Perspective Enterprises, Portage, MI, USA). Body weight was measured to the nearest 0.01 kg and height to the nearest 0.1 cm. Participants were asked to remove their shoes, belts and heavy outerwear, and to empty their pockets. Height was measured in duplicate and if measurements were not within 0.5 cm, we obtained a third measurement. The weight data and mean of all height measurements were used to calculate BMI (kg m⁻²) percentiles and z-scores.³³ Child birth weight was assessed via parent’s report.

Dietary intake.

Multi-pass dietary recalls were completed by the youth to assess energy and macronutrient composition of the diet and the social context of eating. Dietary intake for each participant/sibling was based on 5 weekdays and 2 weekend days random 24-h multi-pass recalls collected over a 4-week period. Trained research staff conducted the 24-h diet recalls over the telephone using the Nutrition Data System for Research software version 2011 (June 2011), developed by the Nutrition Coordinating Center (NCC), University of Minnesota, Minneapolis, MN, USA.^34–36 If a child missed a scheduled recall, they were rescheduled for a day similar to the missed recall (that is, on another weekday or weekend day). The recalls were conducted on random days and children were not reminded of the dietary recalls before the phone calls. However, if the experimenter had to leave a message (for example, child not available at the time of the call), the child participant knew that the experimenter was trying to reach him/her to complete the food recall. Outcomes included total energy intake (Kcal) and consumption of sugar-sweetened beverages (SSB), solid fats and added sugars (SoFAS) items and fruits and vegetables.

Physical activity.

Physical activity was measured using the MTI Actigraph (Pensacola, FL, USA) activity monitor. The Actigraph is a small (5.1 × 3.8 × 1.5 cm), lightweight accelerometer worn around the waist that collects measures of raw acceleration, activity amount and intensity. The Actigraph has been validated in adolescents.^37–39 Activity was monitored and recorded for 7 consecutive days (5 weekdays, 2 weekend days). Youth received written instructions on use, including appropriate care and placement on the right iliac crest using a provided belt and had to wear the ActiGraph at least 10 h day⁻¹ for the day to meet the criterion for a full measurement day. The ActiGraph was initialized for 15-s epochs. Downloaded data were cleaned of spurious lines of > 16 000 counts and negative counts. Sequences of 20+ min of consecutive zero counts were scored as non-wear time. The main outcome variable is the average Actigraph counts/minute, an index of average total physical activity.

Sedentary behaviors/screen time.

The frequency of sedentary behaviors/screen time was captured using an experience sampling methodology (ESM) or ecological momentary assessment (EMA). EMA/ESM have been used to study a range of phenomena in psychology and behavioral medicine.^40–48 These methods make it possible to collect ecologically valid data, as they occur in participants’ natural environment. Although there are some variations in ESM/EMA methods, all involve (a) data collection that takes place in the participants’ environments; (b) assessment of participants’ current state or behavior; (c) assessment that may be event-based, time-based, or randomly prompted; and (d) completion of multiple assessments over a certain period of time.^49,50

In the present study, each sibling was given a cell phone to receive and send text messages related to the study. Text messages were sent to each sibling for seven consecutive days (5 weekdays and 2 weekends), approximately every 2 h between 15:00 hours and 21:00 hours on weekdays, and between 10:00 hours and 22:00 hours on weekend days. A total of seven texts were sent on weekend days and four texts were sent on weekdays. The text message alerted participants to indicate the activity they were doing (for example, screen time, eating, physical activity); the perceived difficulty of the activity (for example, sitting, walking, between a walk and a run, running) and the duration of activity (< 5 min, 6–10 min, 11–15 min, 16–20 min, 21+ min). We focused on the time engaged in screen-based activities (that is, sending/receiving email or text messages; visiting social networking sites; watching television; or playing video games), rather than school-related activities (for example, homework) to capture leisure-time screen usage.

Predictors

Best friend zBMI.

Siblings were asked to come to the laboratory with their best friends. Best friends’ height and weight were measured using the same procedures described above and zBMI was calculated based on these measurements.

Socio-environmental context.

The activity-related social context was captured using the EMA/ESM methodology described above. Participants were asked to report the social context in which sedentary activities/screen time and physical activity occurred. We have previously use this methodology to assess the relationship between social context and physical activity in young adolescents,¹³ and validated this approach using objective accelerometry.⁵¹

The eating-related social context (that is, eating alone or in the presence of friends) was captured during the food recall interviews. Research assistants who conducted the multiple-pass phone recalls asked participants to describe in detail the place in which each eating episode occurred (for example, home, school, restaurant), whether or not they engaged in screen activities (for example, TV, phone, video game), and the social context in which eating occurred (for example, alone, parents, friends).

Analytic models

This analysis focuses on the relationships between best friends’ weight and adolescents’ weight and obesity- and health-related behaviors (sedentary/screen behaviors, physical activity, energy intake and consumption of SSB). This study also examines the relationships between social context (being alone versus being in the presence of friends) and obesity-related health behaviors among weight-discordant siblings.

Siblings are clustered within families, so multilevel models were used to estimate parameters in the presence of clustering, with random intercepts at the family level,^52–54 using PROC MIXED models in the SAS software, Version 9.4.⁵⁵ Mixed models incorporate both random and fixed effects into the model, it assumes that the random effect (family) accounts for the correlation between measures from the same cluster. For each health behavior outcome, a main effects model is estimated (Model 1), followed by three models with interaction terms (with ‘ego’ referring to each individual focal node in a pair of siblings): Model 2a: ego’s zBMI × best friend’s zBMI; Model 2b: ego’s zBMI × time with friends; Model 2c: ego’s zBMI × time alone.⁵⁶ The tests of the hypotheses related to main effects of predictors on outcomes were evaluated in models that did not include interaction terms. Statistically significant interaction terms were interpreted using a graphical approach. To draw the graphs, we used the ‘pick a point’ approach.⁵⁷ This approach involves selecting representative high and low values (mean +/ − 1 s.d.) of the moderator variable and then estimating the effect of the focal predictor at those values.

RESULTS

Participant characteristics

The total analytic sample included 40 pairs of same-sex biologic siblings (Table 1). The sample was 57.5% male and 92.5% Non-Hispanic White, 5% African American, 2.5% Hispanic and 2.5% predominantly mixed race or other race (for example, Native American or Hispanic Black). Siblings’ mean age was 15.4 (s.d. = 1.4) years.

Table 1.

Participant characteristics (n = 80)

	Mean (s.d.; range or N%)
Outcomes
zBMI (kg m−2)	0.78 (0.89; −1.45–2.42)
Energy intake (Kcal)	1936 (508; 1086–3624)
Servings of SSB	9.4 (2.9; 0–15)
Sedentary time (instances reported via ESM/EMA)	23.3 (11; 6–61)
Accelerometer counts (counts/minute)	344 (125; 161–790)
Predictors
Best friend zBMI	0.59 (0.95; −1.61–2.55)
Instances with friends (reported via ESM/EMA)	8.9 (5.8; 0–25)
Instances alone (reported via ESM/EMA)	10.5 (6.2; 0–27)
Household incomea	5.4 (2.8; 1–10)
Mother’s educationb	6.4 (1.4; 4–8)
Father’s educationb	5.9 (1.4; 3–8)
Age (years)	15.4 (1.4; 13–17.8)
Male (n)	46 (57.5%)
Race/ethnicity
Black (n)	4 (5%)
Hispanic (n)	2 (2.5%)
Multiracial/other (n)	2 (2.5%)
White (n)	74 (92.5%)

Abbreviations: EMA, ecological momentary assessment; ESM, experience sampling methodology; SSB, sugar-sweetened beverages.

^aHousehold income: 1–10 scale, where 1 = Under $9,999, 10 = over 200 000.

^bParents education: 1–8 scale, where 1 ⩽ 7th grade, 8 = completed graduate degree.

Multilevel regression model results

All models controlled for siblings’ age, gender and household socioeconomic status (SES), which includes both parents’ income and education. Coefficients for these control covariates are not shown in Table 2. Models used to test zBMI hypotheses also included sibling’s birth weight as an additional control covariate.

Table 2.

Results of multilevel regression models for energy intake (Kcal), consumption of SSB, sedentary time (number of instances reported via ESM/EMA), and physical activity (accelerometer counts)

Predictors	Kcal per day	Servings of SSB	Sedentary time	Physical activity
Model 1: estimates (95% CIs)
Instances alonea	−12.84 (−30.42, 4.74)	−0.03 (−0.12, 0.07)	−0.06 (−0.23, 0.15)	−5.48 (−10.53, −0.43)*
Instances with friendsa	3.15 (−17.72, 24.03)	0.14 (0.03, 0.25)*	0.11 (−0.12, 0.35)	0.11 (0.07, 0.16)**
zBMI	−83.25 (−220.21, 53.71)	−0.35 (−1.10, 0.41)	−1.42 (−2.70, −0.14)*	5.33 (−26.34, 36.90)
Best friend zBMI	41.77 (−82.30, 165.83)	0.67 (0.02, 1.32)*	1.56 (0.20, 2.93)*	2.76 (−31.82, 37.34)
Model 2a: estimates (95% CIs)
Instances alone	−13.14 (−30.78, 4.51)	−0.03 (−0.12, 0.07)	−0.05 (−0.27, 0.16)	−5.11 (−10.29, 0.08)*
Instances with friends	5.61 (−15.86, 27.08)	0.15 (0.04, 0.26)*	0.11 (−0.13, 0.35)	0.11 (0.06, 0.16)*
zBMI	−106.42 (−249.51, 36.67)	−0.44 (−1.24, 0.35)	− 1.38 (−2.80, 0.03)	8.95 (−24.78, 42.68)
Best friend zBMI	−0.92 (−150.55, 148.71)	0.47 (−0.32, 1.26)	1.62 (−0.001, 3.23)	7.59 (−31.81, 46.99)
zBMI × best friend zBMI	61.61 (−57.92, 181.14)	0.29 (−0.36, 0.93)	−0.08 (−1.32, 1.16)	−9.02 (−39.20, 21.17)
Model 2b: estimates (95% CIs)
Instances alone	−12.7 (−30.42, 4.93)	−0.02 (−0.11, 0.06)	−0.05 (−0.27, 0.16)	−5.41 (−10.51, −0.31)*
Instances with friends	8.74 (−20.10, 37.59)	0.26 (0.12, 0.41)**	0.13 (−0.18, 0.44)	0.12 (0.07, 0.16)**
zBMI	−28.36 (−265.16, 208.43)	0.92 (−0.32, 2.15)	−1.26 (−3.67, 1.15)	11.44 (−30.82, 53.69)
Best friend zBMI	35.36 (−91.45, 162.16)	0.53 (−0.10, 1.16)	1.54 (0.14, 2.94)*	1.10 (−34.35, 36.55)
zBMI × friends	−6.65 (−30.24, 16.94)	−0.15 (−0.27, −0.03)*	−0.02 (−0.27, 0.23)	− 0.01 (−0.06, 0.04)
Model 2c: estimates (95% CIs)
Instances alone	−14.72 (−40.41, 10.97)	−0.08 (−0.21, 0.07)	−0.08 (−0.37, 0.20)	−5.27 (−12.42, 1.87)
Instances with friends	3.20 (−17.89, 24.29)	0.14 (0.03, 0.25)*	0.12 (−0.12, 0.35)	0.11 (0.07, 0.16)**
zBMI	−105.67 (−368.16, 156.83)	−0.87 (−2.28, 0.54)	−1.74 (−4.31, 0.82)	7.42 (−55.96, 70.80)
Best friend zBMI	41.82 (−83.47, 167.11)	0.67 (0.02, 1.32)*	1.55 (0.17, 2.93)*	2.77 (−32.72, 38.27)
zBMI × alone	2.11 (−19.23, 23.47)	0.05 (−0.06, 0.17)	0.03 (−0.19, 0.25)	− 0.20 (−5.64, 5.24)

Abbreviations: CI, Confidence interval; SSB, sugar-sweetened beverages.

^aNumber of instances alone or with friends via ESM/EMA. All models control for age, gender, and SES (parents’ income and education).

^*P ⩽ 0.05;

^**P < 0.01. For each outcome, a main effects model was estimated (Model 1), followed by models testing specific interaction terms (Model 2a: sibling zBMI × best friend’s zBMI; Model 2b: sibling zBMI × time with friends; Model 2c: sibling zBMI × time alone).

zBMI

Best friend’s zBMI was the only significant predictor of participants’ zBMI (estimate = 0.26, P < 0.05; 95% CIs = 0.003–0.42), even after controlling for child’s birth weight (P = 0.4). To assess whether this relationship distinguished overweight from non-overweight siblings, participants were dummy coded based on weight status; the only distinguishable feature shared across all sibling dyads. The interaction of siblings’ weight status (categorical) by best friend’ zBMI was not statistically significant (P = 0.62), indicating that the relationship between participants’ zBMI and best friend’s zBMI did not differ for overweight and for non-overweight siblings. In other words, siblings were similar to their best friends in zBMI, regardless of their weight status.

To further examine the specificity of the relationship between participants’ zBMI and their best friends’ zBMI, the siblings and their friends were crossed to test the relationship between siblings’ zBMI and their brother/sister’s best friends’ zBMI (that is, relationship between sibling A and best friend of sibling B; and relationship between sibling B and best friend of sibling A). Best friends’ zBMI was uniquely associated with ego’s zBMI (estimate = 0.27, P < 0.05) and not with the zBMI of their brother or sister (P = 0.43). Participants’ zBMI was not associated with time spent alone (P = 0.69) or with friends (P = 0.81), indicating that overweight siblings did not differ from their non-overweight siblings in terms of time spent alone (Overweight: M = 11.1 instances/units, s.d. = 6.3; non-overweight: M = 9.8 instances/units, s.d. = 6.1) or with friends (overweight: M = 9.2 instances/units, s.d. = 6.2; non-overweight: M = 9.6 instances/units, s.d. = 4.8).

Energy intake

Older participants reported consuming a greater (estimate = 118.3, s.d. = 9.4; P = 0.005; 95% CIs = 37.70, 199.07) number of kilocalories per day than younger participants. Boys (M = 2117, s.d. = 558) reported a greater daily energy intake than girls (M = 1721, s.d. = 337; estimate = − 407, P = 0.0006, 95% CIs = − 622, − 192). None of the other variables of interest were significant predictors of daily Kcal (Table 2), or of SoFAS or fruit and vegetable intake.

SSB consumption

As shown in Table 2, best friend’s zBMI was positively associated (estimate = 0.67, P = 0.04) with sibling SSB consumption. Spending time with friends was also associated (estimate = 0.14; P = 0.01) with greater SSB consumption. However, this relationship was qualified by siblings’ zBMI in Model 2b (estimate = − 0.15, P = 0.02). Differences of least square means indicated that for non-overweight siblings, spending more time with friends was associated with greater SSB consumption, whereas SSB consumption of overweight/obese siblings was not dependent on social context (Figure 1).

Figure 1.

Participants’ zBMI moderated the relationship between social interaction with friends and consumption of SSB. Non-overweight siblings who spent more time with friends, reported consuming more SSB than non-overweight siblings who reported spending less time with friends. Consumption of SSB of overweight siblings was not dependent on social interaction with friends.

Sedentary activities

Participant’s zBMI was inversely (estimate = − 1.42; P = 0.03) associated with the number of instances they reported being engaged in sedentary activities, whereas best friend’s zBMI was positively (estimate = 1.56, P = 0.03) associated with engagement in screen time/sedentary behaviors. The interaction of participants’ zBMI by best friend’s zBMI was not statistically significant (P = 0.90), indicating that the relationship between best friend’s zBMI and occasions engaged in sedentary activities did not differ as a function of siblings’ adiposity (Table 2).

Physical activity

Time spent time alone was inversely (estimate = − 5.48; P = 0.03; Table 2) correlated with average accelerometer counts. On the other hand, activity performed with friends (that is, accelerometer counts accumulated in the presence of friends) was the strongest predictor of participants’ overall activity (estimate = 0.11, P < 0.0001). These relationships were not moderated by participants’ zBMI (zBMI × alone: P = 0.94; zBMI × accelerometer counts with friends: P = 0.67).

DISCUSSION

This study contributes to the literature in several ways. First, to our knowledge, this is the first study to consider friends’ attributes (weight) and the peer social context as source of unshared environment associated with variability in adiposity among biologically related weight-discordant siblings. Second, this study included objective measurements of both participants and best friends’ height and weight, as well as a real-time assessment of social context using an ESM/EMA methodology.

Importantly, our findings indicated that best friend’s zBMI was the only significant predictor of ego’s weight, even after controlling for birth weight and some degree of shared environment and parental genetic variability. The relationship between siblings’ and best friends’ weight appears to be truly ‘unshared’ as best friends’ zBMI was uniquely associated with ego’s zBMI and not with their brother/sister’ weight. These findings add to the existing literature on best friends homophily indicating that overweight youths are more likely to have overweight friends than their normal weight peers.²⁰

Best friends’ zBMI was also strongly associated with SSB consumption, suggesting that best friends’ attributes may influence health behaviors related to obesity risks.^21,22 We have argued elsewhere²⁴ that close friends may act as ‘permission givers’ in early adolescence in either encouraging eating and activity-related behaviors and physical attributes (for example, weight) that are considered normative, or in sanctioning behaviors and traits that are not socially acceptable in the peer group. The literature on injunctive and descriptive norms in the area of youths’ alcohol and other drugs use further suggests that friends do not need to be present or co-engage in certain activities to influence daily behaviors.^58,59

Interestingly, best friend’s weight was positively associated with siblings’ time engaged in sedentary behaviors, but participants’ weight was inversely related to time engaged in sedentary activities. Although these findings are somewhat counterintuitive, they could indicate a reporting bias. Conceivably, overweight and obese youths may underreport engaging in screen time and other sedentary behaviors compared with non-overweight participants (for example, Simpkins et al.,⁶⁰ Jayawardene et al.,⁶¹ Wang et al.⁶²). By contrast, adolescents whose best friend is overweight or obese may not have the same concerns of incurring stigma related to obese individuals who engage in sedentary or screen time activities. It would be interesting to further explore whether best friends’ attributes and behaviors increase the accuracy of self-report information related to weight and energy balance.

Our results further indicated that the social context was associated with adolescents’ activity behaviors. Physical activity performed in the company of friends was associated with greater overall physical activity, whereas spending more time alone was associated with less overall activity. The positive influence of peers and friends and the detrimental effects of aloneness on adolescents’ physical activity are well documented.^{13,23,60,63–67} This study adds to this literature in using a rigorous assessment of physical activity (accelerometry) and a discordant sibling design. Overall, this work suggests that creating opportunities with peers and friends may be a promising approach to increase young adolescent’s physical activity or to complement comprehensive prevention and intervention efforts.

Finally, time spent with friends (versus alone) was associated with greater SSB consumption among non-overweight youths but not for overweight and obese youths. It is important to note that this moderating effect did not translate in overweight/obese youths consuming less SSB than their siblings. Rather, overweight and obese youths simply consumed similar levels of SSB whether they spent more time with friends or alone. One interesting hypothesis worth exploring revolves around the notions of cue-specificity or context-dependent behavior.^68–71 Non-overweight adolescents’ consumption of SSB may occur predominantly with friends and become associated with the social context; whereas overweight youths’ SSB intake may be associated and determined by other factors.

There were no significant predictors of ‘usual’ dietary intake including daily energy, SoFAS, and fruit and vegetable intake. Nor did dietary intake differ between adolescents of lower and greater zBMI. These null results are likely due to limitations in measuring dietary intake of children and adolescents. The 24-h dietary recall method used in the current study is widely used for national surveys of dietary intake of children and adolescents.⁷² However, even with the use of 24-h multiple-pass recalls, it is very difficult to obtain accurate measurements of dietary intake of youth.^73–75 In addition, social desirability may have biased self-reported dietary intake resulting in underreporting of energy intake and foods high in SoFAS.

Limitations

Given that uncontrolled, unshared environmental factors can bias the results of sibling studies, these aspects should be measured and controlled when possible.⁷⁶ This study focused on a limited number of unshared environmental factors, and we recognize that other (non)systematic aspects of the unshared environment likely influenced our findings. This said, evidence suggests that non-systematic factors tend to be events that do not relate in any general way to confounding factors.^77,78 Thus, the extent to which our results are biased by unshared factors remains unclear.

An important strength of this study is the inclusion of objective measurement of best friends’ zBMI; however, it did not include assessment of friends’ health-related behaviors. Therefore, significant associations between siblings and their best friends’ weight are not necessarily indicative of social influence. Although our experimental work (see refs Salvy et al.^79,80 for reviews) clearly shows that proximal social influences and conformity operate among friends and peers, it is unclear whether weight similarity between participants and their best friends in this study is due to social selection or homophily (tendency for friendships to form among youths who are similar),^3,23,51 or to a combination of social selection and social influence whereby similarities between friends increase over time as the relationship endure because of shared behaviors and opportunities. This concern has been raised previously, urging researchers to disentangle homophily from social influence.⁸¹ Until now, however, most designs (see de la Haye et al.²² for an exception) have not been able to distinguish association from causality and studies often overestimated the effects of social influence.^82–85 Future work should take advantage of recent advancements in statistical network models to tease apart social mechanisms⁸⁶ to have a clearer understanding of how friends operate on diet and choices of activity.

Another limitation of this study is that, despite collecting real-time data using an ESM/EMA methodology, we cannot make definitive conclusions regarding the directionality of the associations. For instance, the finding that best friend’s zBMI was positively associated with participants’ SSB consumption could be explained by reverse causality, whereby youths who drink more SSB have friends with greater adiposity. Further longitudinal work is needed in this area to replicate these findings and help disentangle causality in friends influence.

Finally, siblings involved in this study were asked to bring their best friends to the laboratory. However, the quality and/or duration of the identified ‘best friendship’ were not assessed in the present study. Relationship quality most likely moderates the influence of social contact on adolescents’ behaviors and future studies will need to account for this issue.

CONCLUSIONS

Despite some limitations, these results provide important insight on the relationship between friends’ attributes (weight) and young adolescents’ weight and obesity-related health behaviors. These findings add to previous work emphasizing the importance of friends when it comes to weight and health-related behaviors in young adolescents. Peer influence is increasingly the focus of intervention efforts targeting obesity and health promotion in early and late adolescence.⁸⁷ Recent simulations of social network dynamics and peer influence on adolescents’ weight and healthy behaviors suggest that interventions strengthening peer influence may further help reduce obesity incidence and increase healthy behaviors during adolescence.⁸⁷ This study is consistent with this literature and with our previous work (for example, Salvy et al.^63,80) in suggesting that afterschool and community-based programs and services that leverage existing peer groups and network dynamics are promising approaches to promote adolescents’ engagement in healthy behaviors.