Social influences on obesity: current knowledge, emerging methods, and directions for future research and practice

Abstract

Purpose of Review:

Individuals tend to be socially connected with those of similar weight and obesity status. To inform future research and intervention development, we reviewed recent literature examining social influences on weight with a focus on mechanisms of social influence, populations studied, and emerging analytical methods.

Recent Findings:

Social networks appear to influence weight gain and weight loss. It remains unclear what underlying mechanisms (e.g. social norms, social comparison, behavioral modeling) drive this relationship. Stochastic actor-oriented modeling is an important method in the field, but other work has leveraged natural experiment or randomized designs to study social influence.

Summary:

Future networks and obesity research should examine social influence mechanisms, focus on diverse populations across the life course, and carefully consider how to adequately control for competing factors of social selection and physical environments.

Introduction

Nearly four in ten US adults have obesity, a figure that has been steadily increasing since 2000 (1) and is substantially higher than the Healthy People 2020 goal of 30.5% (2). In turn, obesity is associated with a higher risk for various costly and preventable chronic diseases including cardiovascular disease, type 2 diabetes, and multiple cancers (3, 4). One factor that is associated with obesity development is an individual’s social network: individuals who are socially connected tend to have a similar weight status (5). This social patterning of weight can also be called assortativity – the tendency for individuals to share connections with individuals who share similar traits (6). Other terms for this phenomenon include homogeneity bias or network autocorrelation (7).

While assortativity can be discussed without emphasizing an underlying social process, these processes are often of interest to researchers. In the field of social networks and health, three commonly accepted processes that may lead to assortativity by weight are shared environments, social selection, and social influence (Figure 1) (7–9). Aspects of the built environment that are shared between individuals, such as grocery stores and green spaces, may exert effects on both individuals and their social ties who live in the same area. Social selection, often referred to as homophily, reflects the possibility that individuals selectively choose social ties based on shared characteristics, such as obesity or weight status (10). Social influence, on the other hand, reflects the tendency of an individual to become more like their social ties over time. The contributions of shared environments, social selection, and social influence became a matter of considerable debate after Christakis and Fowler’s seminal work evaluating person-to-person spread of obesity through a social network over the span of 32 years (8, 9). The debate spurred critical consideration and ongoing discussion about new research approaches and analytic methods that could isolate the effects of these three processes that are challenging to independently assess (9, 11, 12).

Figure 1:

Social processes that may lead to assortativity by weight

Note: Arrows between individuals indicate social ties; arrows from boxes indicate shared physical environment. Color changes symbolize potential behavioral changes across the two time points (i.e., no color change indicates no behavioral change, color change indicates change in potential behavior).

Understanding the contributions of these three processes to assortativity by weight is critical to develop effective interventions, programs, and policies (hereafter, ‘interventions’) to combat the increasing prevalence of obesity (13). If shared physical environments are the main explanation, then we may want to pursue interventions that promote changes to physical environments, such as increasing the walkability of neighborhoods or the availability of fresh fruits and vegetables in stores. If social selection is at play, then we may want to take advantage of social clustering and target public health or clinical interventions at high risk groups within a social network. Conversely, if social influence is the predominant reason for assortativity by weight, the ramifications for interventions are substantively different. Specifically, leveraging social influence processes within a network could potentially lead to more effective interventions by changing the behaviors of specific target individuals who would subsequently spread positive behaviors via social ties (14). Further, social influence is particularly important to study in detail because there are many candidate mechanisms that could drive social influence. For example, do individuals change their behavior mainly because of shared norms within social networks, social comparison to others, or by observing the behaviors of social ties? These mechanisms are important to understand because they would each suggest different intervention targets and strategies.

To date, the literature is supportive of assortativity by weight (5, 11, 15–25). Furthermore, prior reviews have largely concluded that net of shared environments and social selection, individual’s obesity status or weight tends to become more like their social ties over time (5, 17, 21). However, these reviews have also suggested gaps in the study of social influence that warrant attention. First, the ongoing debate about appropriate methods to study social influence suggest a need to assess the state of emerging research designs and analytic methods. Second, there appears to be an increasing interest in developing interventions that leverage social influence (26–31); yet reviews have not synthesized our current understanding of specific social influence mechanisms, which are a critical piece of effective intervention development. Third, much of our understanding has come from studies making use of data from the National Longitudinal Study of Adolescent to Adult Health (Add Health), a nationally representative dataset on adolescents which, while still collecting follow-up data, was instantiated over 20 years ago (5, 20, 21). As such, current understanding of social influence across the life course and across different socio-demographics groups is limited. Therefore, in this review, we discuss if recent research continues to support social network influences on body weight and explore specific topics found to be major knowledge gaps by previous reviews (5, 20, 21). Specifically, we review recent literature to identify 1) commonly used methods to study social influence, 2) if specific mechanisms of social influence have been evaluated, and 3) what populations have been included in social influence studies. We conclude by discussing limitations of the current literature and directions for future research and practice.

Search strategy

We searched three online databases (PubMed, Sociological Abstracts, and PsychInfo) on July 2, 2019 for original research published within the last five years (2015-present). Our searches included terms relevant to social influence and obesity such as “social influence”, “peer influence”, “contagion”, “obesity”, “body mass index”, and “body weight.” Full search terms are available from the authors upon request. We required that articles collected some form of social network data and assessed a weight or diet outcome. Network data could be egocentric or sociocentric. Sociocentric data is centered on a whole, yet bounded, group and collects information on individuals and their social ties (e.g., all students in a school, all employees of a workplace) (32). Egocentric data, on the other hand, is centered on an individual and collects information from an individual about their social ties, typically without collecting information from those ties directly (32). We included articles analyzing egocentric data if their data collection involved asking about specific characteristics of social ties, rather than general questions about their social environments. We also excluded articles that focus solely on laboratory studies of behavior modeling which are outside the scope of this review. Reference lists of included full texts were searched for additional literature.

Results

Our initial search returned 1,486 articles after removing duplicates between the databases. Of those, 114 full-text studies were assessed for inclusion. We reviewed these full-text articles and their reference lists for relevance to our specific aims and selected 19 for inclusion. Articles were mainly excluded because they 1) measured a social factor but did not measure any form of social networks and/or 2) examined an outcome that was not weight or diet (e.g., health status, cardiovascular disease, body dissatisfaction or norms). Included studies are summarized in Table 1.

Table 1:

Selected articles included in review

Citation	Methods for Influence?	Mechanisms?	Noteworthy population?	Main research question	Main findings	Notable Limitations and Unique Contributions
(Bruening et al. 2018)	✓			What social processes (social selection, social influence, both) impact BMI changes among a social network of diverse college students?	Social influence present: students’ BMI changed to become more like the average of their friends’ BMIs. Social selection also present: students appeared to avoid friends with BMIs <22 or >26.	Emerging adult population, but predominantly white. Found that young adults may exclude individuals with extreme BMIs.
(Datar and Nicosia 2018)**	✓		✓	Are military families assigned to counties with higher obesity rates more likely to be obese compared to families assigned to counties with lower obesity rates?	Social influence present: parents and children in families assigned to counties with higher rates of obesity were more likely to be obese, and this association was stronger with time. No evidence of self-selection.	Creative use of a natural experiment to answer a question about social processes. Unique population. Clear discussion of policy and practice implications of each process. County-level data allowed for control of shared physical environments.
(Fortin and Yazbeck 2015)	✓			Within Add Health data, is fast food consumption one pathway by which social networks influence weight?	Social influence present: adolescents change fast food habits in response to their social ties’ fast food consumption. That in turn leads to increases in adolescents’ weight.	Evaluates how social networks may indirectly influence weight through fast food consumption.
(Meng 2016)	✓			What social processes (social selection, social influence, both) impact weight among individuals within an online health social network?	Social influence present: Individuals had a greater odds of losing weight if their social network contacts were also losing weight. Social selection also present: individuals formed connections with those with similar weight status and who were at a similar stage of weight loss.	Focuses on weight loss rather than weight gain. Social network from online interactions, rather than inperson interactions.
(Simone, Long, and Lockhart 2018)*	✓			What social processes (social selection, social influence, both) impact unhealthy weight control behaviors among Add Health adolescents? And do those relationships vary by gender?	Social influence not present: unhealthy weight control behaviors (e.g., vomiting) do not appear to be socially influenced. Social selection present: adolescents do appear to select their friends based on similarity in unhealthy weight control behaviors. Relationships did not appear to vary by gender.	Focuses on weight control behaviors, rather than weight gain. Attempted to evaluate whether the magnitude of effect of social selection and social influenced varied by gender.
(Lim and Meer 2018)	✓			Is BMI socially influenced among South Korean students randomly assigned to classrooms?	Social influence present: a student’s BMI is positively associated with the BMI of their peers	Leverages econometric methods to take advantage of a natural experiment to control for social selection and evaluate social influence.
(Smit et al. 2016)	✓			Does a social network-based peer influence intervention increase water consumption and decrease sugar-sweetened beverage consumption among primary school children in the Netherlands?	Social influence present: water consumption increased and sugar-sweetened beverage consumption decreased in the intervention group.	Randomized control trial that was informed by social network measures and used social influence as a process to change behavior. However, analyses did not control for dependency within the social network.
(Franken, Smit, and Buijzen 2018)	✓	✓		Does a social network-based peer influence intervention increase water consumption and decrease sugar-sweetened beverage consumption among primary school children in Aruba?	Social influence present: sugar-sweetened beverage consumption decreased overall in the intervention condition. Water consumption increased only for those children in the intervention group who reported high injunctive norms.	Utilized an experimental design as above. However, analyses did not control for dependency within the social network. Evaluated how a specific social mechanism, injunctive social norms, moderated social influence.
(Shakya, Christakis, and Fowler 2015)*		✓		Is an individual’s self-comparison with their social ties associated with weight loss behaviors (dieting)?	If individuals felt they were thinner or fitter than their friends, then they were less likely to diet.	This article tries to specifically evaluate social comparison as a mechanism of social influence. Uses a national longitudinal survey.
(Perry and Ciciurkaite 2019)		✓	✓	Are there social network associations in BMI, and are those associations moderated by an individual’s level of self-monitoring?	The magnitude of social network associations in BMI was moderated by self-monitoring: individuals with higher self-monitoring were more susceptible to social influence.	Links individual psychosocial characteristics (self-monitoring) to social network processes, but not able to independently assess social influence.
(Perry et al. 2016)		✓	✓	Is there an association between partners who live together, and are normative body size, social control, and behavior modeling potential mechanisms for social influences on weight-relating eating behaviors?	Behavior modeling was significantly associated with weight-related eating behaviors (fast food consumption, produce consumption). Normative body size and social control were not significantly associated with fast food consumption. Partner body size was positively associated with fast food consumption.	Study is cross-sectional and egocentric so did not have the ability to truly disentangle selection from influence. However, this study is one of the first to examine multiple potential mechanisms of social influence at once.
(Winston et al. 2015)		✓	✓	Is the support of social network members associated with weight loss in overweight/obese Black and Hispanic adults during a weight loss study?	Weight loss was greater among participants with helpful social network members, and individuals with harmful social networks gained weight.	Study is egocentric so cannot disentangle selection from influence. Evaluates social support as a mechanism for social influence on weight.
(Leahey et al. 2015)**		✓		Are normative influences for obesity associated with weight loss outcomes during an obesity intervention?	Having fewer casual friends who were overweight was associated with better weight loss outcomes. More network acceptance of unhealthy eating habits was associated with worse weight loss outcomes.	Study is egocentric so cannot disentangle selection from influence. Attempts to evaluate social norms as a mechanism for weight change - both weight and unhealthy eating norms. Evaluated associations of mechanisms according to various types of social ties (i.e., emotionally close vs. casual friends).
(Rancourt et al. 2015) **		✓		What is the impact of weight-related social comparisons on dieting behaviors in overweight young adult women?	Comparing oneself to a thinner social tie resulted in increases in calorie restriction, and that effect was amplified if the social tie was a friend.	Study is egocentric so cannot disentangle selection from influence. Evaluates social comparison as a potential mechanism. Data collected using Ecological Momentary Assessment to improve external validity.
(Gudzune et al. 2018)			✓	Is there a relationship between the dietary habits of public housing residents and their social network?	An individual’s sugar intake was significantly associated with their reported network’s sugar intake.	Study is cross-sectional and egocentric so did not have the ability to truly disentangle selection from influence. Unique population and focus on diet outcome.
(Gudzune et al. 2019)			✓	Is there a relationship between the BMI of public housing residents and their social network using egocentric data?	The weight of an individual was inversely related to that of their reported social network.	Study is cross-sectional and egocentric so did not have the ability to truly disentangle selection from influence. Unique population.
(Marquez et al. 2018)			✓	Among Latinas participating in a weight loss intervention, are the weight and weight control behaviors of Latinas and their top three social ties related?	The weight change and weight control behaviors of Latinas and their social ties were significantly associated.	Study is cross-sectional and egocentric so did not have the ability to truly disentangle selection from influence. Unique population.
(Jackson, Steptoe, and Wardle 2015)			✓	Is an individual’s weight change associated with their partners weight change among couples in the English Longitudinal Study of Aging?	If an individual had a partner who was overweight and subsequently lost weight, their own odds of losing weight increased.	Focus on weight loss, rather than weight gain. Expands knowledge to older adults.
(Cobb et al. 2015)			✓	Is an individual’s weight change associated with their partners weight change among couples in the Atherosclerosis Risk in Communities Study?	An individual was more likely to gain weight if their partner also gained weight. If their partner become obese, then an individual had a greater risk of becoming obese themselves.	Expands knowledge to older adults.

Notes: ‘Methods for Influence’ = Did the paper use methods that could possibly distinguish between social selection, social influence, and shared physical environments? ‘Mechanisms’ = Did the paper try to assess a mechanism of influence? ‘Noteworthy population’ = Did the article study a population that was not Add Health and was not adolescent, high-income, or primarily white?

^*Of Importance

^**Of Outstanding Importance

We found that the literature continues to be predominantly supportive of social influence as a major factor in the social patterning of obesity and weight. Individuals’ weight outcomes tend to become similar to those of their social ties over time (13, 33–37). The articles also provided evidence of a relationship between social influence and dietary behaviors (38). One article did not find evidence of social influence among adolescents’ unhealthy weight loss behaviors but did find that students tended to select friends with similar behaviors (39). We discuss methods, mechanisms, and populations studied in further detail below. Briefly, it continues to be difficult to discern among specific social processes and many articles fail to account for the potential influence of shared physical environments. Articles that were able to disentangle social processes did not examine specific mechanisms of social influence. Articles that did examine specific mechanisms of social influence used egocentric data and mainly focused on one specific mechanism, as opposed to comparing multiple possible mechanisms. Lastly, work has begun to expand research to include participant populations outside of the Add Health data.

Methods Used

We found three articles that used stochastic actor-oriented modeling (SAOM) to study weight or weight-related behaviors (33, 36, 39). Briefly, SAOM is one popular way to evaluate the relative contributions of various social processes, but requires longitudinal data on both social networks and health (40). This modeling approach uses a simulation-based estimation procedure to develop parameter estimates (40). By incorporating both longitudinal social network and health data, SAOMs investigate how social networks and health evolve simultaneously over time. The three articles used SAOM to leverage longitudinal data and investigated the co-evolution of social networks and health, which allowed them to disentangle social selection and social influence (40). While controlling for shared environments is possible with SAOM procedures, none of these studies incorporated control variables for physical environment confounders.

Of the three SAOM articles, two found evidence of social influences on both weight gain and weight loss (33, 36). The third study using SAOM examined unhealthy dieting behaviors and did not find any evidence of social influence (39). More specifically, Bruening et al. used SAOMs with social network and obesity data collected from emerging adults on a college campus in the 2015–16 academic year (33). Their results supported a social influence mechanism: individuals had a higher odds of gaining weight if their friends had a higher BMI (33). Their work also suggested that emerging adults may exclude individuals with extreme BMIs (BMI less than 22 or greater than 26) from their social group via social selection (33). Another study modeled data on a social network of individuals derived from a health-related app to examine social selection and influence among a network of individuals focused on losing weight (36). This work also showed evidence of social influence: individuals with social ties who lost weight had a higher odds of losing weight themselves (36). Finally, one study used SAOM to examine social network data from Add Health (39). Rather than assessing weight or obesity, the authors examined unhealthy dieting behaviors such as inducing vomiting or taking diet pills (39). Their modeling did not show evidence of social influence – it did not appear that adolescent’s unhealthy diet behaviors converged over time (39). However, there was evidence of social selection – adolescents may choose to be friends with individuals who have similar unhealthy weight control behaviors (39).

In addition to SAOM, we found several studies that used other methods and analytical approaches to study social influence processes. We found one study that used econometric methods to investigate a specific pathway by which social networks may influence weight and obesity using Add Health data (38). Specifically, the study incorporated fast food consumption as an intermediate effect between social networks and weight. They found that an adolescent was more likely to increase their fast food consumption in response to an increase in their friends’ fast food consumption, which in turn translated to increases in weight.

We also found four articles that used natural or random experiments to examine social influence (13, 28, 31, 34). Within experimental designs, individuals typically do not have control over certain aspects of their social network. The two natural experiments we reviewed examined social influence by leveraging random military installation assignment and classroom assignment to control for social selection. For example, the ability to select certain social ties is altered – friendships may change, but family ties will not. Datar and Nicosia took advantage of the military’s assignment of families to specific installations to assess whether families living in counties with higher rates of obesity were more likely to be obese compared to families in counties with lower rates of obesity (13). County level data was also used to control for shared physical environments. Both parents and children in families assigned to counties with higher rates of obesity had higher odds of being overweight or obese, and those odds were larger the longer the family had resided in that county. Findings were robust to adjustment for shared physical environments and tests for self-selection. Results supporting the social influence hypothesis have also been found using classroom assignments in South Korea (34).

In addition to leveraging natural experiments, we found research that has also used investigator-randomized assignments. We found two recent articles that used social network data to identify specific individuals to deliver an intervention to their peers (28, 31). In brief, this intervention used ‘peer influencers’ in Dutch and Aruban schools who were instructed to promote water consumption among their peers. Peer influencers were chosen by asking all children to name up to five individuals whom they thought were good leaders, respected, or would go to for advice. These peers were then trained to drink more water and promote water consumption. Both trials found that the intervention decreased children’s actual consumption of sugar-sweetened beverages (28, 31).

Social Influence Mechanisms

We found seven research articles that examined specific mechanisms of social influence such as social control, behavioral modeling, social norms, social support, and social comparison (31, 41–46). We found one randomized study that explored mechanisms (31). Specifically, the intervention study to promote water consumption among Aruban schoolchildren collected additional data on children’s self-reported social norms (31). Descriptive norms were assessed by asking children how often their friends consumed water, and injunctive norms were assessed by asking how often their friends approved of them drinking water (i.e., perception of how supportive friends were of water drinking). The intervention was effective at increasing water consumption only for children who reported high injunctive norms (i.e., perceived friends as more supportive) (31).

We also found several studies that examined mechanisms with egocentric data. One study examined the relationship between social support and weight loss among participants in a weight loss intervention (41). The results showed that reporting helpful, supportive social ties was associated with greater weight loss, and having harmful, unsupportive social ties was associated with weight gain during the intervention (41). Another study examined social norms and weight loss, again among participants in an obesity intervention (42). Weight norms were measured using the weight status of social ties, and unhealthy eating norms were measured by self-reported acceptability (e.g., acceptability among their friends of eating unhealthy foods, being encouraged by friends to eat unhealthy foods, etc.). Having more casual friends who were overweight and greater social network acceptability of unhealthy eating were both associated with less weight loss (42). Notably, the weight norms set by ‘emotionally close’ social ties (i.e., partners and best friends) were not associated with treatment outcomes, while the weight norms set by casual friends were (42).

Two articles focused on self-comparison as a mechanism (43, 44). One used self-reported egocentric data from a nationally representative sample collected at two timepoints (43) while the other used self-reported ecological momentary assessment (EMA) data in overweight young adult women collected six random times per day over five days (44). EMA data is sampled in real time, which limits the potential for recall bias and so is has the potential to improve the extent to which the research can be generalized to real-life, natural settings (47). In both studies, self-comparison was measured as downwards or upwards. Upwards social comparison meant that an individual saw their weight as worse off than those around them. Both studies found that upwards social comparison was associated with dieting behaviors (43, 44). Conversely, downwards social comparison meant that an individual saw their weight as better than those around them. The two studies’ findings on downward social comparison differed. Within the nationally-representative sample, downwards social comparison of weight was associated with individuals being less likely to diet (43). On the other hand, data from the EMA sample showed that downward social comparison was associated with subsequent dieting behaviors (44).

Finally, one study simultaneously examined three possible mechanisms of social influence on fruit and vegetable as well as fast food consumption (45). The authors examined the relationship of three mechanisms set by a partner or spouse – social norms, social control, and behavior modeling – with eating behaviors of a respondent. Social norms were assessed using the partners’ body weight. Social control was assessed by asking “To what extent does your partner or spouse attempt to manage or control what or how much you eat?”. Behavior modeling was assessed by asking the respondent to report how often their partner or spouse ate produce and fast food. Their analysis found significant associations between the behavior modeling mechanism and eating behaviors, and social norms (i.e., partners body size) were significantly associated with a respondents’ fast food consumption (45). In contrast, the social control mechanism was not significantly associated with either eating behavior.

Populations Studied

We also assessed whether studies have expanded from the Add Health dataset to focus on racial/ethnic minority populations, economically diverse populations, or age groups other than adolescents. We found two studies that used Add Health data when reviewing the literature since 2015 (38, 39). All other work used alternative data sources. A few articles focused on particularly noteworthy populations. One article focused on Black and Hispanic adults in New York City (41). Two articles used data from the Networks and Obesity: Relationships and Mechanisms (NORMs) study (45, 46), which used deliberate sampling for socioeconomic and racial heterogeneity in Lexington, Kentucky. We also found studies of egocentric social networks within public housing residents in Baltimore and Montgomery County, Maryland (48, 49) and among Latina women (50). Non-US research articles that we reviewed were conducted in high-income countries; namely Netherlands (28), Aruba (31), and South Korea (34).

In terms of age groups, we found two recent studies that focused on social influence among an aging population, specifically using adult and older-adult married couples using data from the Atherosclerosis Risk in Communities Study (ARIC) (35) and English Longitudinal Study of Aging (ELSA) (37). These two studies examined whether an individual’s weight change was associated with a partner’s weight change. Among ARIC participants, individuals had higher odds of becoming obese if their partner became obese (35), and among ELSA participants, individuals had higher odds of weight loss if they had an overweight partner at baseline who lost weight (37). These results generally support the conclusion that partners’ weight changes are related, and it is possible that intervention effects may spill over from one partner to another. However, neither of these articles considered the shared physical environment of married couples and did not account for factors that could impact both partner selection and later life weight change.

Future Directions

We found continuing support for the influential role of social networks in shaping weight. The studies we reviewed used several approaches to isolate social influence effects, primarily SAOM models, econometric methods, and experimental designs. Our review suggests that research on specific social influence mechanisms remains limited and there remains a need to expand longitudinal social network studies to more diverse populations.

We found that SAOM was a popular approach. SAOM is one of the state-of-the-art methods for disentangling social influence from social selection, but this type of network statistical model still has limitations (51). One recent simulation study indicated that the social selection parameters in SAOM may be quite sensitive to misspecification (51); thus, a variety of approaches should be considered. Natural experiments offer another approach that do not require use of network statistical models. Only two studies used this approach to leverage natural variation in social ties to examine social processes in the context of obesity or weight loss (13). A common natural experiment that could be leveraged further is the random assignment of university dormitory roommates (52). As an alternative method, latent space modeling is a statistical approach that models the probability of relationships between actors as dependent on an individual’s unobserved position in “social space” (53, 54). We did not find any studies that used latent space modeling.

Understanding the specific mechanisms that lead to social influences on obesity can aid in development of high quality and targeted interventions (45). Studies that used the most advanced analytic methodologies to distinguish social influence from social selection did not examine social mechanisms, which is a similar finding to a previous systematic review (21). Studies that attempted to examine social mechanisms showed evidence for a variety of mechanisms including social support (41), social norms (31, 42), social comparison (43, 44), and behavior modeling (45). However, only one study examined more than one mechanism (45), which limits our ability to clearly distinguish which mechanism might play a predominant role in social influence. Further, the studies that examined social mechanisms used egocentric data for which there are no methods currently available to isolate social influence from other processes (55). It may be possible to combine econometric methods (such as fixed effects and instrumental variables) with egocentric data to study mechanisms of social influence while controlling for social selection and shared physical environments. This is an area that is ripe for future methodological research.

We found two articles that detailed the effect of one social influence intervention on children’s beverage consumption (28, 31). Randomized interventions could have potential utility to evaluate mechanisms of social influence. For example, one article examined if social norms moderated the effect of a randomized intervention (31). However, the intervention targeted two mechanisms of social influence simultaneously: the peer influencers were given a reusable water bottle (behavioral modeling) and were tasked with encouraging their peers to drink more water (social norms) (28). Thus, the extent that each specific mechanism influenced change is not distinguishable. Future randomized intervention studies should consider comparing various social network interventions that capitalize on different potential mechanisms. Factorial designs that allow researchers to assess how multiple factors influence an outcome, both independently and together, may be appropriate. Alternatively, researchers could also consider a target trial framework, which uses observational social network data to mimic a randomized trial for a specific intervention plan (56–58). In general, a target trial framework would not directly answer which mechanism is at play but could allow clear assessment of the effects of potential programs structured around specific mechanisms.

In addition to studying mechanisms of influence, research should also evaluate whether demographic, psychosocial, or network factors moderate social influence. Social influence appears to be a salient concept in most social networks, but it is unclear if the magnitude and specific mechanisms of social influence vary by individual characteristics (e.g., race, personality characteristics) or types of social ties (e.g., romantic partners, coworkers). This knowledge could help identify individuals and groups who would be most responsive to social influence interventions. For example, Perry & Ciciurkaite found that self-monitoring, an individual personality characteristic, moderated network associations in weight (46). The association between an individual’s weight and the weight of their network members was greater among those individuals with high self-monitoring (i.e., those who were more likely to modify their behaviors based on people around them) (46).

One of the largest limitations of research on assortativity by weight is the relative homogeneity of populations studied. Much of our knowledge on social influence and obesity has come from the Add Health study (5, 20, 21). This singular source is problematic because the data were collected before the peak of the obesity epidemic and was designed to be representative of adolescents in high school in 1994–1995 (59). We found work has expanded into more diverse sociodemographic populations, but these studies typically had more substantial limitations in terms of design. For example, two studies used longitudinal data from older-adult married couples, which adds knowledge of social influences over the life course, a noted limitation of social networks research (20). However, limiting the focus to the married dyad provided a narrow view of an individual’s social network and did not allow for disentangling different social processes. Therefore, we echo other researchers in suggesting that more longitudinal data on social networks and health is needed, and that data should include diverse sociodemographic groups (5, 20, 21).

Conclusions

Research indicates that the obesity, weight, and some dietary behaviors of an individual are influenced by their social ties. It is unclear which specific mechanisms underlie social influence; candidate mechanisms include social support, social norms, social comparison, or behavior modeling. Currently, it is not clear if influence effects differ for individuals at different stages of the life course and in different demographic groups. Furthering our knowledge in those two domains while using the appropriate analytic methods and research designs is critical to developing high-quality weight-loss and obesity prevention interventions that leverage social networks.