Social Interactions in Health Behaviors and Conditions
Summary and Keywords
Health behaviors are a major source of morbidity and mortality in the developed and much of the developing world. The social nature of many of these behaviors, such as eating or using alcohol, and the normative connotations that accompany others (i.e., sexual behavior, illegal drug use) make them quite susceptible to peer influence. This chapter assesses the role of social interactions in the determination of health behaviors. It highlights the methodological progress of the past two decades in addressing the multiple challenges inherent in the estimation of peer effects, and notes methodological issues that still need to be confronted. A comprehensive review of the economics empirical literature—mostly for developed countries—shows strong and robust peer effects across a wide set of health behaviors, including alcohol use, body weight, food intake, body fitness, teen pregnancy, and sexual behaviors. The evidence is mixed when assessing tobacco use, illicit drug use, and mental health. The article also explores the as yet incipient literature on the mechanisms behind peer influence and on new developments in the study of social networks that are shedding light on the dynamics of social influence. There is suggestive evidence that social norms and social conformism lie behind peer effects in substance use, obesity, and teen pregnancy, while social learning has been pointed out as a channel behind fertility decisions, mental health utilization, and uptake of medication. Future research needs to deepen the understanding of the mechanisms behind peer influence in health behaviors in order to design more targeted welfare-enhancing policies.
On the Importance of Social Interactions for Understanding Health Behaviors
Health behaviors are actions that affect health conditions by increasing or decreasing the odds of morbidity and mortality. Categories of health behaviors include diet, nutrition, and physical activity, tobacco, alcohol, and illicit drug use, sexual behavior, stress management, sleep patterns, healthcare use, and adherence to prescribed medication, among others. These behaviors may lead to health conditions such as obesity, diabetes, substance dependence, or depression, and may have significant impacts on a wide range of economic outcomes, including medical expenses, education, employment, wages, and crime (Cawley & Ruhm, 2012). Some health behaviors are more social in nature and as such are more likely to be subject to the influence of others through social interactions. In a comprehensive review of the economics of health behaviors, Cawley and Ruhm (2012) argue that “some of the most exciting recent economic research on health behaviors has incorporated factors such as peer groups, social capital and relative status.”
Social interactions, or peer effects, are any type of externalities in which the backgrounds, actions or outcomes of a reference group affect the decision process of an individual, either through changes in preferences, beliefs, or constraints. Market-based or price-based effects are excluded from this definition. Direct social interactions or externalities in preferences occur when the actions of others shape an individual’s preferences, either because consumption decisions by one or more members of the reference group change the individual’s payoff (payoff interactions or contagious effects), or because individuals do not wish to deviate from the social norm imposed by a reference group, perhaps due to fear of social sanctions (Hirshleifer, 1995). Externalities in expectations or beliefs occur when individuals look at behaviors among others in a reference group to guide them on the costs and benefits of different choices (Banerjee, 1992; Bikhchandani, Hirshleifer, & Welch, 1992). Finally, externalities in constraints may occur when fixed costs are high and supply does not take place until there is a critical mass of consumers. For example, health-food stores may not open in a neighborhood until there is a minimum critical mass willing to purchase that kind of food. Constraint externalities may also be a consequence of insufficient supply of shared resources, such as gyms (Cawley, Han, Kim, & Norton, 2017). In the presence of social spillovers, the first-order condition of the consumer’s maximization problem will show strategic complementarities in actions if the marginal utility to one person of undertaking an action is increasing in the average amount of the action taken by his peers (Glaeser & Scheinkeman, 2001). If marginal utility is decreasing in others’ actions, actions are strategic substitutes.
One of the main reasons why economists study peer effects is that social interactions may magnify the effects of interventions. In the presence of strategic complementarities in behaviors, social interactions will multiply the impact of policies and programs. The magnitude of this multiplier will depend on the strength of peer influence. The study of social interactions in economics has expanded in the past years because of the need to better understand the motivations and intentions behind individual behavior, in particular under the influence of behavioral economics and network analysis.
The Challenges of Estimating Peer Effects
In a seminal paper, Manski (1993) identified three types of associations in outcomes or behaviors that result from social interactions: correlated effects, contextual effects, and endogenous effects. Correlated effects are statistical associations between individuals’ behaviors that occur either because individuals tend to form ties to similar others (selection or homophily) or because they share the same environments with others. The other two categories involve particular cases of social influence. Contextual effects occur when the propensity of an individual to behave in certain way varies with the distribution of background characteristics in the group (Manski, 1993). Endogenous effects, in contrast, take place when the behavior of a peer or a group of peers directly influences the behavior of the individual. A key to endogenous effects is the simultaneity between the individual’s and peers’ choices.
While both contextual and endogenous effects are types of social interactions, they have very different policy implications. When endogenous effects are present, a policy such as a sin tax can have multiplier effects: the policy will impact the individual both directly and indirectly through the influence on his/her peers. If peer effects are just contextual, policy will not have multiplying effects. Contextual effects can only be exploited policy-wise by changing the composition of the group the individual belongs to and/or by focusing policy efforts on individuals strategically positioned in the social network.
Contextual effects are not easily distinguishable empirically from endogenous effects. A group of schoolmates may have unhealthy eating habits because their parents have preferences towards unhealthy food and the habits in these families spill over to the students in the same grade. If students’ food choices depend strongly on family habits and these habits are not easily observed by the econometrician, social interactions in food choices may be taken as endogenous when they are just contextual. In such case, a school policy that makes unhealthy food unavailable at school will mistakenly be assumed to have multiplier effects when it does not. Furthermore, even when knowing that the effects are contextual, the specific underlying trigger may not be well identified. In the example above, unhealthy eating habits may be correlated with peers’ family education, while it is family food preferences (correlated with family education) that really matters.
The main challenge in the estimation of endogenous effects stems from the simultaneity of influences between an individual’s behavior and that of the peer group. In the presence of endogenous social interactions, an individual’s behavior will reflect the behavior of the group, and the groups’ behavior will reflect that of the individual. This reflection problem, as defined by Manski (1993), is critical when the estimated effects are linear-in-means, that is, when the behavior of interest is influenced by the average behavior of the peer group. This is because the reflection problem relies strictly on perfect collinearity between an individual’s behavior and his/her peers’ average behavior. The problem is less critical in nonlinear settings, although multicollinearity may still be affecting the precision of the estimates in such cases.
Additional challenges in the estimation of peer effects include measurement error and specification error. One main source of measurement error arises from the difficulty in defining the relevant peer group. Compared to neighborhoods, schools have geographic and social boundaries of interaction that are more precise. However, students do not interact with everyone at school. The assumption that each neighbor or each student in a school, a grade or a class is a relevant influencer may introduce measurement error in the estimation (Ludwig et al., 2008). This problem is exacerbated when researchers do not have a full sample of all individuals that exert an influence on an ego. Using sampled students’ means or regional means in a regression is problematic because the sample means will be measured with sampling error, which attenuates estimates on the regional or school-level aggregates (Micklewright, Schnepf, & Silva, 2012).
Measurement error may also stem from misreports of health behaviors. Measures of health behaviors usually rely on self-reports from survey data. Misreporting may result from inaccuracies arising from comprehension, recall, and other cognitive operations or from factors related to the social desirability of the behavior and the interviewing conditions (Brener, Billy, & Grady, 2003). For some behaviors and populations, self-reports have shown small degrees of inaccuracies. For example, the validity of self-reported smoking is consistently high in population-based studies that compare self-reported data to biological markers of cotinine levels (Rebagliato, 2002; Vartiainen, Seppälä, Lillsunde, & Puska, 2002; Wagenknecht, Burke, Perkins, Haley, & Friedman, 1992). However, smoking status is more likely to be underreported by participants of smoking-cessation programs (Boyd, Windsor, Perkins, & Lowe, 1998) or by teenagers (Patrick et al., 1994; Rebagliato, 2002). Edoka (2017) shows that smoking misreporting among adolescents results in a downwards bias of marginal effect estimates of tobacco use. In the case of alcohol use, the evidence also points towards a high reliability of general population self-reports (Glovannucci et al., 1991; Polich, 1982), but to the underestimation of drinking behavior among specific populations, such as heavy drinkers (Townshend & Duka, 2002). Similarly, there is evidence that obese individuals, female endurance athletes and adolescents tend to understate their weight and food intake and to misreport weight control behaviors (Burkhauser & Cawley, 2008; Cawley & Burkhauser, 2006; Huybrechts, De Bacquer, Van Trimpont, De Backer, & De Henauw, 2006; Nawaz & Katz, 2001; Schoeller, 1995; Shields, Connor Gorber, Janssen, & Tremblay, 2011).
The above evidence suggests that misreporting tends to be higher among individuals more likely to participate in unhealthy behaviors such as substance use, intake of unhealthy food, or adoption of unhealthy weight-control methods. This negative covariance between the true explanatory variable of interest and measurement error does not have the classical errors in means specification. Most likely, it resembles the mean-reverting measurement error found by Bound and Krueger (1991) in earnings data. In such case, attenuation bias will decrease (relative to the case of classical measurement error) whenever the variance in measurement error is small relative to the variance in the underlying true variable. But it will increase, even to the point of reverting the sign of the coefficient, whenever the variance in measurement error is large relative to the variance of the underlying true variable. The problem introduced by measurement error is further complicated by the fact that the dependent variable in peer-effects models is also likely to show a negative correlation between the underlying true variable and measurement error.
Finally, misspecification error can also bias the estimates if model assumptions do not hold. These include assumptions about the choice of covariates, the interactions between social variables and individual decisions, or assumption derived from theory and functional form specifications.
Empirical Approaches in the Estimation of Peer Effects in Health Behaviors and Conditions
The typical social-interactions specification assumes that an individual’s health behavior depends linearly on personal characteristics, environmental characteristics, peers’ choices regarding that same health behavior, and (in some models) on peers’ background characteristics. Once a reference group is defined, most empirical models of peer influence in health behaviors assume that the association between peers’ actions and the individual’s action is linear in means.
Addressing the Reflection Problem
When the model is linear-in-means, the estimate of a coefficient of ego’s behavior on average-peer-behavior will be inconsistent due to the reflection problem. Approaches aimed at resolving reflection have included the use of peers’ background characteristics as instrumental variables, exploiting the incomplete structure of the social network, using lagged rather than contemporaneous measures of peer behavior, exploiting differences in the variance of behaviors at the aggregate level versus individual level, estimating structural discrete-choice models that solve for strategic equilibria, or running partial population models.
Identification Using Instrumental Variables: Peers’ Family Background, Exogenous Shocks and Genetic Phenotypes
A large number of studies have used instrumental variables to get past the reflection problem. The approach replaces the measure of peer behavior for some residual variation in behavior that is uncorrelated with the ego’s choices. Adequate instruments need to be sufficiently correlated with peer behavior, but excludable from the individual equation; the only channel through which they can influence the individual is through the alter.
Most studies dealing with peer effects in school settings have used peers’ average family-background characteristics to instrument for peers’ average behavior. Two assumptions must be made for these variables to be good instruments. The first one is that contextual effects are non-existent—that is, that peers’ parents and siblings do not directly affect the ego’s behaviors. Second, the model must adequately control for environmental characteristics that could be associated with peers’ background profile due to endogenous sorting. In papers dealing with substance use, the set of family background instruments has included peers’ parents’ sociodemographics (marital status, race, nationality, education), peers’ family structure, peers’ parents’ involvement with their children, availability of substances at peers’ households, and use of substances by peers’ parents and siblings (Ali, Amialchuk, & Dwyer, 2011; Fletcher, 2010, 2012; Gaviria & Raphael, 2001; Gwozdz et al., 2015; Halliday & Kwak, 2012; Lundborg, 2006; McVicar & Polanski, 2014; Powell, Tauras, & Ross, 2005; Sen, 2009). Studies analyzing peer effects in weight and obesity have used the body mass index (BMI) or obesity status of the peers’ parents to instrument for peer obesity in school settings (Renna, Grafova, & Thakur, 2008; Trogdon, Nonnemaker, & Pais, 2008) or regional means of sociodemographic characteristics to instrument for regional prevalence of obesity (Auld, 2011). Peers’ mothers’ teen childbearing, mother and fathers’ college education and peers’ average age of menarche have been used to instrument for peers’ likelihood of having a teen pregnancy.
The first problem with peer family-background instruments is that they rely on the assumption of no contextual effects. It is not hard to think of situations in which a student’s use of substances is affected directly by the availability of alcohol at friends’ homes, by friends’ older siblings, or by the parents of their friends, through conversations and example. Second, friendships selection may be associated to friends’ use of substances or weight status. If tendencies towards substance abuse or obesity are heritable, because of selection and heritability, an obese youth will be more likely to be surrounded by obese parents of peers. Finally, some of these instruments may violate the monotonicity condition required for the estimation of local average treatment effects. Monotonicity requires that the instruments shift the behavior of all individuals in the same way. For example, we would expect an individual who drinks when alcohol is not easily available at home to drink also when alcohol is available. Such assumption may not hold if the offspring of a heavy drinker reacts to parental substance abuse by being abstinent. In the presence of defiers, the instrumental variables estimates do not represent local average treatment effects.1
Other instrumental variables studies have used variation stemming from policy or other type of shocks that affects peers but do not affect the ego. In an analysis of spouses’ influence in smoking behavior, Cutler and Glaeser (2010) use smoking bans in the spouse’s workplace to instrument for the spouse’s smoking. Monstad, Propper, and Salvanes (2011) instrument a sibling’s teen pregnancy using variation in exposure to an educational reform that increased the minimum school-leaving age in Norway. Yakusheva and Fletcher (2015) use a close friend’s miscarriage as an exogenous shock to the friend’s likelihood of giving birth.
Finally, a very recent set of studies rely on peers’ genetic information to address the reflection problem. Sotoudeh, Conley, and Harris (2017) use peers’ distribution of polygenic scores associated with tobacco use to instrument for peer smoking. A polygenic score is a weighted average or composite score that takes into account information across an individual’s entire genome to measure his/her genetic predisposition or risk for a particular outcome. The authors argue that polygenic scores are good candidates for breaking the reflection problem because they are determined prior to birth. When running the instrumental variables regression, the authors control for a wide set of individual and peer average characteristics, including the ego’s own polygenic score and peers’ parents’ and siblings’ average smoking, to dismiss the effect of direct contextual influence. Cawley et al. (2017) criticize the use of genes as instrumental variables. While the use of genes may address the reflection problem, genes may not be excludable from the individual smoking equation. Peers’ genetic predisposition could trigger, for example, reactions by classroom teachers, who may devise strategies for students in that classroom to reduce their smoking. Or, genetic predisposition may correlate with other traits, such as discount rates, which may affect smoking through other channels (higher likelihood of being idle, using alcohol, and so forth).
Beyond the reflection problem, instrumental variables may contribute to address measurement error biases if the instruments are uncorrelated with the measured error. Because classical measurement error results in attenuation bias, we expect the peer estimates to be larger when running instrumental variables. On the other hand, in the absence of measurement error, and if actions are strategic complements, instrumental variables estimates should be smaller than naïve estimates. The sign and magnitude of the difference between instrumental variables estimates and naïve parameters will thus be an empirical matter.
A final and nontrivial issue with instrumental variables in general is that instruments may be weak after using fixed effects approaches that account for selection, and this weakness is likely to bias peer effects estimates. Halliday and Kwak (2012) show that many of the family background instruments used by prior researchers show little variability once school fixed effects have been accounted for.
Identification Using Network Structure
A more recent and growing source of identification comes from the structure of social networks (Bramoullé, Djebbari, & Fortin, 2009). The social network approach exploits the fact that not everyone interacts with everyone in a social group (i.e., networks are usually not complete). In network analysis, a peer group includes only those individuals that have connections with the ego. The social intransitivity in relationships that result from network incompleteness breaks the reflection problem, as the background characteristics of friends of friends, but not friends of ego, can be used to instrument for friends’ behavior. Studies using this identification strategy have used friends of peers’ parental marital status, parental educational attainment, presence or number of older sibling(s), family structure, alcohol availability at home, and financial, sociocultural, and physical environment characteristics, among others, to instrument for peers’ substance use, body weight, sport activities or food consumption (Ajilore, Amialchuk, & Egan, 2016; Fortin & Yazbeck, 2015; Leonard, McKillop, Carson, & Shuval, 2014; Liu, Patacchini, & Zenou, 2014; Mora & Gil, 2013; Moriarty, McVicar, & Higgins, 2016). Unlike prior literature, the social network approach gets around the reflection problem without having to assume the absence of contextual effects. It focuses, in addition, on peer groups that are arguably more relevant than those defined by analytical convenience (such as school peers, randomly assigned college roommates or neighbors). However, the network approach has not yet been able to resolve the problem of the endogenous formation of network links, an issue we come to in the next section.
Furthermore, the strength of identification in network analysis is likely to depend on the density of the network. As with other instrumental variables models, the instruments may end up being weak after using network fixed effects to account for network-related correlated effects.
Reduced-form Models that Combine Contextual and Endogenous Effects
Given the difficulties involved in finding good instruments and in getting rid of correlated effects in network analysis, some authors have turned to running reduced-form regressions of individual behavior on peers’ background characteristics, controlling for individual-level characteristics. The coefficients on these peer characteristics capture social interactions, but as reduced forms of more comprehensive social interaction models, they are unable to distinguish between endogenous and exogenous influences (Manski, 1993). Authors who have pursued this avenue in the substance use literature include Argys and Rees (2008), Bifulco, Fletcher, and Ross (2011), and Black, Devereux, and Salvanes (2013). Cawley et al. (2017) run reduced-form models of an individual’s weight and obesity on the genetic predisposition of his/her siblings’ towards having high BMI. They recognize that their estimated peer effect could capture not only endogenous, but also contextual effects running through different channels, such as the food that is prepared at home, the promotion of exercise in the household, or other traits of the individual different from BMI. Sotoudeh et al. (2017) take this argument further and question the adequacy of using siblings’ genes to explain social interactions at the family level. Because siblings’ genes will be correlated with parents’ genes, they may be a proxy for environmental common shocks that result from parents’ behavior (food- or-fitness related) and that influence both the ego and the siblings’ weight.
While statistically simpler, the approaches regressing ego’s behavior on peers’ background characteristics are unable to generate well-defined policy recommendations, as it is not clear whether these effects convey social multipliers that would enhance the impacts of policies or programs. Furthermore, reduced-form approaches cannot assert whether the source of social influence is the explanatory variable per se or if this variable is just a proxy for other unobservable triggers.
Some authors have suggested using perceptions of peer behavior, rather than measures of peer behavior, as a means for solving the reflection problem (Kawaguchi, 2004). Norton Lindrooth, and Ennett (2003) provide three reasons why such an approach leads to inconsistent estimates. First, individuals perceive their own behavioral choices as relatively common and alternative choices as relatively uncommon (false consensus effect), thereby projecting their own behavior onto the groups’ in subjective assessments. Second, there is evidence that certain populations, such as adolescents, tend to overestimate the prevalence of substance use, regardless of their own behavior (scaling effect). Finally, subjective estimates of peer health behaviors may incorporate simple random error (measurement error effect).
Identification Using Lagged Measures of Peer Use
An alternative road around the reflection problem has been to replace contemporary measures of peer behavior with measures that pre-date the appearance of a social link. This has been the usual approach in studies exploiting random assignment of college students (Duncan, Boisjoly, Kremer, Levy, & Eccles, 2005; Eisenberg, Golberstein, & Whitlock, 2014; Yakusheva, Kapinos, & Weiss, 2011; Yakusheva, Kapinos, & Eisenberg, 2014). Since current choices made by the individual cannot impact past peer choices, the use of past peer behavior addresses simultaneity. Some authors have used peer past behavior as the main explanatory variable in settings in which links were formed prior to collecting the data (Balsa, Gandelman, & Roldán, 2018; Clark & Lohéac, 2007). These studies may not get rid of simultaneity if current use by the ego (the dependent variable) is correlated with peers’ lagged use (the explanatory variable) through the ego’s own lagged use. One way to avoid this problem is to condition the regression on the ego’s past behavior. Still, these models must rely on certain assumptions about the lag structure of social influence and may run the risk of misspecification (Manski, 2005). Furthermore, lagged peer outcomes may overlook recent changes in peer behavior, which might influence an individual’s outcomes. They are therefore interpretable at best as lower-bounds estimates of the peer effect of interest (Hanushek, Kain, Markman, & Rivkin, 2003).
Identification Using Equilibrium Discrete-choice Models (Structural Estimation)
Structural models respond to the problem of reflection by endogenously modelling individual and peers’ choices and focusing on the equilibrium outcomes (Brock & Durlauf, 2001a). In discrete-choice models with endogenous social interactions, an ego’s net utility from an action depends on the observed or expected actions of other individuals. The model’s equilibria incorporate the feedback that each choice generates on the whole network and these equilibria define the conditional likelihood functions used to estimate the social interaction parameters.
Many of the discrete-choice structural studies of peer effects in health behaviors analyze interactions in small groups, and assume that individuals can fully observe the choices of others.2 In these models, individual payoffs depend on the actual choices of all others in the group (Card & Giuliano, 2013; Clark & Etilé, 2006; Harris & González López-Valcárcel, 2008; Krauth, 2006, 2007; Nakajima, 2007; Soetevent & Kooreman, 2007). Because the non-cooperative Nash equilibria of these models may generate multiple equilibria, it requires the choice of an equilibrium selection rule for estimation. The fundamental identification problem of these models relies on selecting the right equilibrium and on finding variation in the data that shifts an individual’s choice without shifting that of his/her peers.
Clark and Etilé (2006) estimate a structural game of peer effects in smoking across spouses. The authors consider a two-player non-cooperative binary game in which each individual’s smoking decision depends on own smoking and on the spouse’s past smoking. For certain combinations of private payoffs, the model allows for multiple equilibria, in which case a mixed-strategy equilibrium defines the appropriate threat-points. The model is estimated assuming a bivariate probit model. Nakajima (2007) assumes that youth smoking decisions occur sequentially rather than simultaneously. Peer interactions occur frequently and are reciprocal across students of the same school and cohort, making all outcomes endogenous. The author bases the model on a Markov dynamic interaction process, in which students respond to the observed smoking or non-smoking actions of others without trying to anticipate their actions. This process converges to a steady state distribution of the interaction process, which is used to construct the likelihood function for estimation. Identification is achieved through the parameterization of the utility function and by assuming interactions are uniform across individuals of similar types (gender, race). Soetevent and Kooreman (2007) also consider a discrete-choice model of smoking with a small number of players. They provide upper bounds on the number of equilibria that occur if interactions are strategic complements, which depend on the interaction strength and the number of agents. For identification, they assume that contextual effects are nonexistent. Harris and González López-Valcárcel (2008) explore asymmetric social influences in smoking behavior between siblings by considering a multiplayer non-cooperative binary game similar to the two-player game studied by Clark and Etilé (2006). The model allows for differential influence of smoking and non-smoking siblings. To avoid the indeterminacy due to multiple equilibria, they rely on a sequential-move solution, in which older players move first. Card and Giuliano (2013) model the behavioral choices (regarding sexual behavior, smoking, and marijuana initiation) of self-nominated friends’ pairs at school, allowing for social interaction effects as well as for correlated unobservable determinants of their joint behavior. They identify individual behavior from peers’ behavior by relying on individual traits (such as physical development) correlated with own behavior but not with that of the friends. Finally, Lin (2014) assumes that agents do not have complete information on peers’ behaviors. The author estimates a discrete-choice model in which agents interact only with a subset of the group, have partial information on peers’ choices, and form rational expectations about heterogeneous peers. Such model generates a unique equilibrium for a wide range of parameters.
While structural models have the advantage of embedding theoretical insights into the estimation, they also face a higher risk of misspecification. Identification usually relies on assumptions about functional forms, parameter distributions, and equilibrium selection rules, which are strong and not necessarily testable. Krauth (2005) shows that the bias due to an incorrectly specified equilibrium selection rule increases gradually in the strength of the peer effect. Furthermore, the computational cost and complexity of structural estimators is far greater than that of ordinary least squares or instrumental variables methods. As with much of the reduced-form literature, endogenous effects in many of these models are identified under the assumption that there are no contextual effects.
Individual- versus Aggregate-level Variation
Glaeser and Scheinkman (2001) and Glaeser, Sacerdote, and Scheinkman (2003) propose a different way of identifying social multipliers. Their approach infers the presence of social interactions from contrasts between regression coefficients at the individual and aggregate levels, as well as from comparisons between observed variance across regions and the variance that would be expected from sampling noise if people were randomly distributed across regions and there were no social interactions whatsoever. This technique is used by Auld (2011) to study regional peer effects on BMI and obesity, and by Cutler and Glaeser (2010) to explore peer effects on smoking. The main problem with this methodology is that aggregate and individual effects must be residualized from environmental correlates before being contrasted. This residualization usually relies on observable correlates but cannot get rid of unobservable correlated effects.
Partial Population Models
Partial population designs (Moffitt, 2001) originate from a sample of social groups between which there are no spillovers. Some of these groups are randomly assigned to a treatment category, and within these, a random subset of agents is treated. The difference between the mean outcomes of untreated subjects in treatment and control groups captures the spillover effects of group treatment on untreated subjects. Kremer and Miguel (2007) use this approach to assess spillovers in take-up of deworming medication in Kenya. Babcock and Hartman (2010) employ it to study the spillovers of a treatment aimed at incentivizing gym attendance.
Addressing Correlated Effects
Distinguishing true peer effects from correlated effects due to selection or homophily is not an easy task. The most credible approaches rely on the random or quasi random assignment of individuals to groups of peers. Studies using random assignment have analyzed social influences in health behaviors among individuals randomly assigned to roommates in college (Duncan et al., 2005; Eisenberg et al., 2014; Yakusheva et al., 2011, 2014), to squadrons in the air force (Carrell, Hoekstra, & West, 2011) or to neighborhoods (Ludwig et al., 2011).
Other studies have exploited the quasi-random assignment of students across cohorts in schools, defining the peer group at the school-grade level (Ajilore et al., 2016; Ali, Amialchuk, & Heiland, 2011; Bifulco et al., 2011; Clark & Lohéac, 2007; Fletcher, 2010, 2012; Halliday & Kwak, 2012; Moriarty et al., 2016; Sen, 2009: Sotoudeh et al., 2017; Trogdon et al., 2008), or across classes within a school grade if classes are randomly assigned within grades—in which case the peer group is defined at the class level (Balsa et al., 2018; Lim & Meer, 2018; Lundborg, 2006). These studies rely on parents’ ability to choose the school but not the cohort or class where their children will be placed in. In such settings, the use of school or school-year fixed effects captures time invariant unobserved features of the school correlated with parental preferences. Some studies with longitudinal data have also addressed unobserved heterogeneity by using individual fixed effects (Cohen-Cole & Fletcher, 2008; Gwozdz et al., 2015; Halliday & Kwak, 2009; Moriarty et al., 2016; Sen, 2009), although attenuation bias constitutes a non-trivial threat in settings with large number of individuals and few time periods (Deaton, 1995).
Studies based on random or quasi-random variation rely on peer groups that are opportunistic to the analyst, but that do not necessarily reflect the natural channels of influence. The influence that results from experimentally assigned groups may be quite different from that exerted in groups of peers that have formed by choice. Several studies have compared estimates from analyses defining the peers at the school-cohort level with estimates identifying peers from self-nominated friends (Ali, Amialchuk, & Heiland, 2011; Clark & Lohéac, 2007; Halliday & Kwak, 2012; McVicar & Polanski, 2014). Some of these authors interpret the higher magnitude and strength of estimates based on nominated friends as evidence of the stronger influence of close group of friends. Such comparisons, however, fail to account for the fact that links within networks are endogenous, and that the estimates based on nominated friends may suffer from selection bias.
Most analyses of peer influence in health behaviors exploiting social network structure adjust for school fixed effects or for network fixed effects. The assumption for identification is that, once the network has formed, within-network link formation takes place randomly or based on observable individual characteristics only. These studies ignore that link formation decisions may be given endogenously by unobservable characteristics that are also correlated with health behaviors (Fortin & Yazbeck, 2015).
In structural models, correlated effects have been taken care of through the inclusion of county, school, family or individual fixed effects and through the estimation of correlated random effects. Clark and Etilé (2006) exploit the variation in spouses’ smoking behavior over nine waves of data and allow for correlation between spouses’ individual constant traits and between their unobserved time-varying error terms. Nakajima (2007) controls for county-fixed effects to address unobserved heterogeneity at the county level, while Soetevent and Kooreman (2007) use school fixed effects to account for non-random selection into schools, and allow for within class correlation of error terms to account for common shocks. In Krauth (2006, 2007) selection and common shocks are addressed by imposing the assumption that the correlation in unobservable variables equals the correlation in observable variables. Harris and González López-Valcárcel (2008) assume a specific random-effects error structure to address the problem of common shocks. Card and Giuliano (2013) estimate correlated random effects across friends’ behavioral equations. A very recent and incipient literature is trying to address selection by embedding the mechanisms by which groups are formed in structural models of peer effects (Hsieh & Lee, 2015). This is a promising avenue of research.
Evidence of Peer Effects in Health Behaviors
In a recent review of the experimental and quasi experimental literature on peer effects, Sacerdote (2014) concludes that estimates of peer effects on social outcomes, such as playing sports or binge drinking, are more robust than those on test scores. The literature we present in the following section supports the claim that social interactions have a significant role in explaining a vast range of health behaviors, although not all of them. Our review focuses mostly on papers written since the 1990s and published in main economic outlets. We begin by reviewing the findings of the literature on peer effects in substance use (tobacco, alcohol, and illicit drugs), a summary of which is depicted in Table 1.
Peer Effects in Tobacco Use
Most analyses of peer influence in tobacco use aim at estimating the effects of peers’ tobacco prevalence on the individual’s likelihood of having smoked in the past 30 days. A number of studies in school settings that exploit idiosyncratic variation in cohorts within schools, and use instrumental variables (e.g., peer family background characteristics, peer genotype) to avoid reflection, tend to find statistically significant and relatively large smoking effects (Fletcher, 2010; Gaviria & Raphael, 2001; Lundborg, 2006; McVicar & Polanski, 2014; Powell et al., 2005; Sotoudeh et al., 2017). In the upper range, estimates amount to 0.6, implying that a 10 percentage point increase in peers’ prevalence of smoking increases the ego’s likelihood of smoking by 6 percentage points. Outside of the school setting, Cutler and Glaeser (2010) estimate an effect of 0.40 when using workplace smoking bans to instrument for a spouse’s smoking choice.
Among papers not relying on instrumental variables, Clark and Lohéac (2007) find that smoking by male peers increases ego smoking by 0.088. The authors use peers’ lagged smoking prevalence to address reflection and exploit idiosyncratic variation across cohorts within school to address selection. Also, in a school setting in Canada, Sen (2009) finds that smoking by same-gender peers has an effect of 0.298 on females’ smoking and of 0.387 on males’ smoking. They use individual fixed effects to control for unobserved heterogeneity but do not address reflection.
In terms of models exploring contextual effects, Argys and Rees (2008) show that for females (but not for males), being younger than the average student in the cohort increases the likelihood of smoking by 0.041 percentage points. However, Bifulco et al. (2011) do not find evidence of a correlation between the fraction of peers with college educated mothers or the fraction of minorities in the school-year and ego’s smoking, once controlling for school fixed effects and other covariates.
Unlike instrumental variables and other reduced-form studies, the only study relying on randomly assignment of college roommates (Eisenberg et al., 2014)—the cleanest approach to dismiss correlated effects—does not find statistically significant effects of past roommate’s smoking prevalence on the ego’s current use of tobacco. Neither do most structural models after addressing unobserved heterogeneity (Card & Giuliano, 2013; Clark & Etilé, 2006; Krauth, 2006; Soetevent & Kooremenan, 2007). One exception is Harris and González López-Valcárcel (2008), who find that an additional smoking sibling in the household increases the likelihood that ego smokes by 0.479, whereas an additional non/smoking sibling decreases his/her likelihood by 0.329. Nakajima (2007) also finds a coefficient of 0.12 on peer smoking when estimating a random utility model.
Peer Effects in Alcohol Use
Unlike the mixed case of tobacco, most empirical evidence points to positive and statistically significant peer interactions in alcohol use. Among studies using random assignment of college roommates, Duncan et al. (2005) find that, for male students that binge drank in high school, having a roommate who also binge drank in the past increases the frequency of binge drinking in college by 3.83 times per month (almost double the average frequency). No robust roommate effects are encountered for females. Eisenberg et al. (2014) estimate that being assigned a college roommate who binge drank before entering college increases the probability that the student binge drinks by 8.6 percentage points, even after conditioning on peers’ contextual characteristics.
As noted for the tobacco literature, papers that use peers’ family background characteristics to instrument for peer behavior (Case & Katz, 1991; Fletcher, 2012; Gaviria & Raphael, 2001; Lundborg, 2006; McVicar & Polanski, 2014) usually produce higher estimates of peer effects (from 0.2 to 0.7). Effects are quite large even in studies using friend of friends’ family background characteristics as instruments for peer behavior (Ajilore & Amialchuk, 2016).
Studies using reduced-form approaches also find evidence of positive peer effects in alcohol use. Clark and Lohéac (2007) find that a 10 percentage point increase in peers’ past prevalence, increases ego’s alcohol use by 1.8 percentage points, after controlling for school fixed effects and other covariates. Results in Argys and Rees (2008) show that, for females, being younger than the average student in a cohort increases the likelihood of using alcohol among teenagers in the United States by 3.5 percentage points. Findings from Balsa, French, and Regan (2014) reveal that a student’s relative socioeconomic status in the school year (relative deprivation) is positively associated with alcohol consumption and drinking to intoxication for adolescent males, but not for females. Bifulco et al. (2011), on the other hand, do not find evidence of peer effects when regressing a student’s likelihood of binge drinking on the fraction of peers with college-educated mothers or the fraction of minorities in the school year.
Lin (2014) uses a binary choice model with social interactions under heterogenous rational expectations. Both endogenous and contextual effects are identified in their model. Correlated effects caused by observed or unobserved characteristics shared by students within the same school are controlled for by school fixed effects. Using nominated friends as the relevant peer group, results show an endogenous peer effect of 0.499 for alcohol drinking.
Peer Effects in Illicit Drug Use
None of the studies exploiting random assignment of roommates in colleges in the United States finds evidence of statistically significant peer effects in illicit drug use (Duncan et al., 2005; Eisenberg et al., 2014). However, as with the other substances, peer estimates are statistically significant, and quite large, when employing instrumental variable methods that use family background characteristics to instrument for peer use (Ali, Amialchuk, & Heiland, 2011; Case & Katz, 1991; Gaviria & Raphael, 2001; Lundborg, 2006; and McVicar & Polanski, 2014). In these studies, peer effects range from 0.2 to 0.6. Using friends of friends’ family background characteristics as instruments (network intransitivity), Moriarty et al. (2016) are unable to reject the hypothesis that peer influence in marijuana use equals zero. Their estimate is positive but imprecise, probably due to weak instruments. When using individual fixed effects (rather than instrumental variables) with longitudinal data, they find a statistically significant effect on lagged friends’ prevalence of cannabis use of 0.076.
The findings from studies using reduced-form approaches deliver mixed results. Clark and Lohéac (2007) do not find evidence of peer effects in marijuana use when regressing ego’s use on male and female school-year-peers’ past prevalence of marijuana use. Argys and Rees (2008), on the other hand, find that females that are younger than the average student in the cohort are more likely to use marijuana. Bifulco et al. (2011) find that an increase of 10 percentage points in the fraction of peers with college-educated mothers decreases marijuana use by approximately 3 and 4–5 percentage points during and after high school, respectively.
Card and Giuliano (2013) provide structural estimates for peer effects in illicit drug use. They find that low-intensity use of marijuana by a best friend increases ego’s use by 0.45 percentage points, but do not find high-intensity use by peers to have a statistically significant influence. They, however, suggest precaution when interpreting this estimate, as they get a non-intuitive negative correlation between the parameters capturing unobservables of both friends.
Table 1. Social Interactions in Substance Use
Peer Effects in Weight and Weight Related Behaviors
The paper by Christakis and Fowler (2007) jumpstarted the economics literature on peer effects in weight and weight-related behavior (see Table 2). Christakis and Fowler studied peer effects in BMI and obesity for men and women in the offspring cohort of Framingham Heart Study, a densely interconnected social network of 12,067 people aged 21 or older, assessed repeatedly from 1971 to 2003. Running logistic regressions of ego’s obesity on each peer’s past and contemporaneous obesity, and conditioning on ego’s own past obesity, the authors find that a person’s chances of becoming obese increase by 57 percent if he or she has an obese friend (marginal effect of 0.62). They also show that effects are strongest for reciprocated friendships and nonsignificant when the friendship is perceived by the alter, but not by the ego.
Cohen Cole and Fletcher (2008) challenge Christakis and Fowler’s findings by claiming that their analysis does not sufficiently account for contextual effects, and that their method of controlling for selection is “much too narrow in scope.” To illustrate their claims, they replicate Christakis and Fowler’s analysis using Add Health data, and define the alter as the closest friend of the respondent (that with the highest friendship nomination in wave 1).3 They show that the peer effect becomes nonsignificant after the model accounts for school-specific trends and individual fixed effects. They conclude that the social correlations identified by Christakis and Fowler must be due to common environmental characteristics shared by the members of the network.
Fowler and Christakis (2008) reply by rerunning Cohen-Cole and Fletcher’s analysis on Add Health data with some variations: they track friendship nominations over time, use all available nominated friends rather than the first friend named, and impute missing data using expectation maximization. With these changes, they find a statistically significant effect of peers’ contemporaneous BMI on ego’s gain in BMI of 0.033, robust to the inclusion of school trends. They also run individual fixed effects and find a statistically significant coefficient of 0.053. Finally, they underscore the fact that peer effects are stronger when the alter is the named friend, and nonsignificant when the alter is the person who named the friend.
Other studies have also used Add Health data to estimate peer effects in weight and weight-related behaviors. Halliday and Kwak’s (2009) findings on peer effects in BMI and overweight lose statistical significance when individual fixed effects are included. Trogdon et al. (2008) and Renna et al. (2008) use friends’ background characteristics to instrument for close friends’ BMI and compare cohorts within school to account for unobserved heterogeneity at the school level.4 Trogdon et al. (2008) estimate an instrumental variable coefficient of 0.52, while Renna et al. (2008) find an instrumental variable coefficient of 0.252 for females, and an imprecise coefficient for males.5
Loh and Li (2013), Gwozdz et al. (2015) and Lim and Meer (2018) also use peers’ parental characteristics to instrument for peers’ average BMI. In Loh and Li (2013), the relevant peer group is children in the same age range and school, and residing in the same community in rural China. The authors find an instrumental variable estimate for the full sample of 0.273.6Gwozdz et al. (2015) study peer effects among children aged 2 to 9 in eight European countries. Findings reveal a peer effect of 0.368.7 Both the analyses in Loh and Li (2013) and Gwozdz et al. (2015) are limited by the fact that they use only observable community characteristics to control for correlated effects. Lim and Meer (2018) exploit the random assignment of students in 7th grade in South Korea across classes, ensuring their estimates are free from homophily. They employ peers’ number of older siblings as an instrument for peer BMI, finding an effect of peers’ BMI of 0.831. They do not find statistically significant effects of peers’ obesity on ego’s obesity (neither contemporaneous nor lagged).
Other studies exploit network structure and use friend of friends’ but not friend of ego’s covariates as instruments for peer BMI. These include Mora and Gil (2013) and Ajilore, Amialchuk, Xiong, and Ye (2014), both of whom study peer effects among adolescents in school settings, in Spain and the United States respectively. The instrumental variable marginal effects on nominated friends’ BMI are 0.371 in Mora and Gil (2013) and 0.416 in Ajilore et al. (2014). To address correlated effects, the former study adjusts for school and neighborhood fixed effects, while the latter controls for network fixed effects. None of the studies addresses the fact that network links are formed endogenously.
Yakusheva et al. (2011) and Yakusheva et al. (2014) estimate peer effects in weight gain taking advantage of the random assignment of first-year university students to roommates. The first paper analyzes a small sample of female students (N=144) at a private Midwestern university. Ego’s weight at the end of freshman year is regressed on the roommate’s weight at baseline (prior to exposure), on ego’s weight at baseline, and on dormitory fixed effects. Results show a negative and statistically significant peer effect: a one standard deviation increase in the roommate’s past weight decreases ego’s weight gain by 0.07 standard deviations. Together with evidence on positive peer influences in exercise frequency, choice of a meal plan, and use of weight loss supplements, the authors attribute the negative peer effect in weight gain to a higher engagement in weight-management behaviors by female students who weigh more, and to a transmission of these behaviors to their peers. Results need to be interpreted with caution due to the small sample size.
Yakusheva et al. (2014) estimate a similar model, but using a larger sample of male and female first year students at two universities. They regress ego’s weight at the end of the academic year on peers’ and own weight at baseline, ego’s and peer’s height at baseline, environmental preferences and university fixed effects. They find that peer’s weight at baseline increases ego’s weight gain by 0.034 for females, but has no influence on males. The influence of alter’s weight is stronger for female egos with higher weight than the alter, lower socioeconomic status, and less sexual experience. Influence is stronger also across pairs of roommates with similar academic achievement and similar political and religious views. The authors find little evidence of peer effects on exercise and eating-related behaviors and no alteration of the peer effect on weight estimates after controlling for these potential behavioral mechanisms.
Auld (2011) infers social interactions in BMI and obesity from excess variance across regions and from contrasts between regression coefficients at the individual and aggregate levels (Glaeser et al., 2003). The reference group is defined as individuals in the same county (or state) as the ego. Estimated multipliers from the regression analyses are less than 1 or even negative, suggesting biases due to unobserved time-varying influences on body weight at the county or state level. Overall, the results do not suggest the presence of large social multipliers on body weight.
Finally, Cawley et al. (2017) exploit genetic information to identify peer effects in BMI and obesity among siblings. The authors run pooled ordinary least squares regressions of ego’s BMI (or obesity) on a sibling’s genetic risk score for obesity, conditional on ego’s genetic risk score and other covariates. They find that a one-unit increase in the genetic risk score of the alter is associated with a 0.16 unit increase in ego’s BMI and with a 0.97 percentage point increase in the probability of obesity. The alter’s genetic risk score is significantly correlated with ego weight only when the ego is younger than the alter.8
Table 2. Social Interactions in Body Weight
A number of studies find significant peer effects in health behaviors that lead to overweight of obesity, such as exercise and food consumption (see Table 3). Carrell et al. (2011) study peer effects in a sample of 460 students entering the US Air Force Academy, who are randomly assigned to squadrons. Results show that a one standard deviation increase in the squadrons’ average fitness score during high school increases ego’s fitness gain by 0.129 standard deviations while in the academy, conditional on ego’s past high school fitness score and graduation class fixed effects. This effect is 30 percent as strong as the effect of own high school fitness in college fitness. Egos in squadrons with peers with higher fitness scores in high schools have lower probabilities of failing the Academy’s fitness requirements. These peer effects are caused primarily by friends who are the least fit. Babcock and Hartman (2010) use experimentally induced differences in the number of friends of ego exposed to a pay-to-exercise treatment in a university setting to study how peer usage of exercise facilities affect ego’s usage. For treated egos, they find that an additional treated friend increases the number of visits to an exercise facility by 0.13, whereas an additional non-treated friend decreases the number of visits by 0.065. No peer effects are found for untreated individuals.
Ali, Amialchuk, and Heiland (2011) study peer effects in weight-related behaviors using Add Health data from waves 1 and 2. They estimate ordinary least squares and probit regressions of ego’s behavior on nominated friends’ average behavior, controlling for contemporaneous and lagged measures of ego’s and peers’ BMI, parental location preferences, individual and peer covariates, and school fixed effects. They find peer effects of 0.079 in the likelihood of exercising at least three times a week, of 0.184 in the likelihood of practicing active sports, of 0.178 in the number of days per week eating in fast food restaurants, and of 0.066 in the likelihood of eating snacks. These estimates do not address the reflection problem.
Bramoullé et al. (2009), Leonard et al. (2014), and Fortin and Yazbeck (2015) exploit network intransitivity to estimate peer effects in the number of recreational facilities, food consumption, and frequency of fast food visits, respectively. Using Add Health data, and controlling for network fixed effects, Bramoullé et al. (2009) find that a one unit increase in the average number of recreational activities of self-nominated friends increases ego’s number of recreational activities by 0.467. Leonard et al. (2014) study a community of low-income, minority individuals in the United States. Controlling for network and environmental characteristics, they estimate a positive peer effect of 0.679 of geographically close family and friends on ego’s consumption of fruits and vegetables. Fortin and Yazbeck (2015) find that an additional average visit per week of nominated friends to fast food restaurants increases ego’s number of visits by 0.129 days. Their model introduces network fixed effects and peers’ contextual effects.
Finally, in a study outside economics, Cruwys, Bevelander, and Hermans (2015) analyze 69 experiments in which researchers independently manipulated the eating behavior of a social referent and measured either food choice or intake. They find evidence of peer influence (or “modelling effects”) in 64 of these studies, despite substantial diversity in methodology, food type, social context, and participant demographics.
Table 3. Social Interactions in Weight Related Behaviors
Peer Effects in Risky Sexual Behavior and Unwanted Pregnancy
Table 4 depicts a series of papers dealing with peer effects in risky sexual behavior and unwanted pregnancy. One of the first studies to discuss these issues was the paper by Case and Katz (1991), where the authors examine the influence of inner-city low-income neighbors in adolescent non-marital pregnancy. Using both probit and instrumental variable specifications, the authors find neighbors’ influence to be irrelevant for out-of-wedlock pregnancy.
Yakusheva and Fletcher (2015) and Fletcher and Yakusheva (2016) explore peer effects in fertility at the school level. In the former study, they use a friend’s miscarriage as a quasi-random fertility shock to peer fertility. They find that a close friend’s teen birth is associated with a 6 percentage point reduction in the probability of own teen pregnancy. The negative peer effect suggests a knowledge-sharing mechanism behind the peer influence, rather than a conformity effect. In the latter paper, Fletcher and Yakusheva (2016) use peer-level teen childbearing of mothers and peers’ average age of menarche to instrument for peers’ fertility. They find large peer effects: a 10 percentage point increase in peer pregnancies is associated with a 2 to 5 percentage point increase in the probability of own pregnancy. They argue that the influence of peers may occur through the construction of social norms rather than through information and knowledge sharing about pregnancy risks.
Two other papers explore schoolmates’ influence on teen fertility choices, but instead of focusing on endogenous effects, they run reduced-form specifications that associate students’ school or class composition with teen fertility decisions. After using instrumental variables techniques to address selection, Evans, Oates, and Schwab (1992) find that the fraction of economically disadvantaged students at school is not associated with the ego’s likelihood of having a teenage pregnancy. Black et al. (2013) exploit variation in cohort composition within schools over time in Norway. They find that a higher fraction of female classmates decreases teen pregnancy by 1.8 percentage points. They do not find statistically significant effects of peers’ age or socioeconomic status.
Kuziemko (2006) and Monstad et al. (2011) analyze peer effects in childbearing within the family. Kuziemko (2006) runs a model of ego’s childbearing on siblings’ prior fertility decisions conditioning on individual and age-in-years fixed effects. She finds that the probability of having a child increases by 15 percent during the 24 months following the birth of a niece or nephew. Monstad et al. (2011) exploit an educational reform in Norway that increased the minimum school leaving age to instrument for an older sister’s likelihood of having a teen pregnancy. They find evidence of strong peer effects in teen pregnancies across female siblings.
Several other papers analyze directly peer effects in sexual behavior at school. Ali and Dwyer (2011), Card and Giuliano (2013), Fletcher (2007), Halliday and Kwak (2012), and Richards-Shubik (2015) explore whether classmates or school friends can influence adolescent sexual activity initiation. Fletcher (2007) finds that a 10 percentage point increase in the rate of sexual activity at the school level increases the likelihood that a student initiates sex by 3 percentage points. Peer effects differ by age and race. Ali and Dwyer (2011) explore the role of close friends in influencing adolescent sexual initiation. The authors pursue an instrumental variable estimation strategy combined with school-level fixed effects and find that a 10 percentage point increase in the fraction of close friends initiating sex increases the individual’s probability of sexual initiation by 4 percentage points. Halliday and Kwak (2012) do not find evidence of peer effects in sexual initiation in a model that uses schoolfriends’ family background characteristic to instrument for friends’ sexual initiation. Richards-Shubik (2015) uses Add Health data to estimate an equilibrium model for the market of sexual partners in high school in the United States. The author finds evidence of large peer effects on the timing of sexual initiation, driven by within-gender peer norms. Richards-Shubik’s counterfactual simulations indicate that, when peer effects are removed, the number of boys and girls that initiate sexual activity during high school drops by 26 percent and 20 percent, respectively. Card and Giuliano (2013) analyze peer influence in sexual initiation by running a discrete-choice structural model of endogenous peer interactions. Their results show that the likelihood that one friend initiates intercourse within a year is increased by about 5 percentage points (on a base rate of 14 percent) if the other one also initiates intercourse.
Finally, two studies use randomly assigned peers to study peer influence in sexual behavior in college. Duncan et al. (2005) estimate the effect of having a college roommate who had sex in high school on the number of partners with whom the ego has sex during the first year of college. Eisenberg et al., 2014 analyze how the number of sexual partners that the roommate had in high school affects the ego’s number of sexual partners in freshman year. Both studies use data from U.S. universities and fail to reject the null of no peer effects.
Table 4. Social Interactions in Fertility and Sexual Behavior
Peer Effects in Mental Health
Mental illness is severely undertreated worldwide. The lack of knowledge regarding signs and symptoms, ignorance about how to access treatment, and stigma are among the main reasons for this treatment gap. Golberstein, Eisenberg, and Downs (2016) use college housing-assignment data to study peer effects in the use of mental health services. The study finds evidence of peer influence when groups are defined at the hall level, but not at the room level. The paper finds that exposure to a higher proportion of peers with a recent history of mental health treatment is associated with ego’s more positive beliefs about treatment effectiveness.
Some studies have found evidence suggesting that mental health may also be contagious among peers (see Table 5 for a review). Fowler and Christakis (2008) investigate whether happiness can spread from person to person in large social networks, reporting that friends who live within a mile and become happy increase the probability that ego is happy by 25 percent. This correlation increases to 42 percent when the alter who becomes happy lives less than half a mile away. Rosenquist, Fowler, and Christakis (2011) find also that individuals are 93 percent, 43 percent and 37 percent more likely to be depressed if the person they are connected to is depressed, considering one, two and three degrees of separation, respectively.
One of the main limitations of these studies is that they do not control sufficiently for correlated effects. In fact, when scholars address this problem, results indicate weaker links than suggested by previous studies. Taking advantage of the random assignment of first-year college students to roommates, Eisenberg, Golberstein, Whitlock, and Downs (2013) find no evidence of overall contagion of mental health, and evidence of small effects for specific mental health measures. They find no peer influence in happiness and modest influence in anxiety and depression. Pachucki, Ozer, Barrat, and Cattuto (2015) exploit longitudinal data on social interaction patterns to study peer effects in mental health among early adolescents. While they find evidence of cross-sectional correlations in mental health status across students in the same network, they find no evidence that connected students’ mental health status becomes more similar over time because of their network interactions. In line with these two studies, Ho (2017) finds no evidence of sibling spillovers in health symptoms that could have mental causes, such as loss of appetite or stomach pain. Taking advantage of sibling pairs who attend different schools, Ho instruments the health of a sibling by that of his or her classmates to identify a potential spillover effect. Among other findings, the study shows that “the spillover effect for a depressed sibling pair is significantly smaller than the spillover effect for a typical sibling pair, which implies that it is unlikely that spillovers in the associated symptoms are due to spillovers in depression” (Ho, 2017, p. 99).9
Table 5. Social Interactions in Mental Health
Mechanisms and Policy Implications
Externalities in Preferences, Beliefs, and Constraints
Despite the relatively large evidence on the existence of strategic complementarities in health behaviors across a wide range of behaviors, populations, reference group, and estimation techniques, there is not yet a good understanding of the underlying mechanisms. The literature has proposed three channels for peer effects in health behaviors (Rice & Sutton, 1998; Cutler & Glaeser, 2010; Nakamura et al., 2017): social spillovers in preferences (direct social interactions or payoff interactions); social spillovers in beliefs (social learning); and social spillovers in constraints.
The first channel, direct social interactions or payoff interactions, can arise for several reasons. An individual’s utility from an action may be enhanced when others take the same action. Eating and drinking are examples of health-related behaviors that usually satisfy this condition (Cutler & Glaeser, 2010). In general, people prefer to eat with others than by themselves. And an individual’s expected return from drinking is likely to rise if her friends go for a drink. This motive is referred to as social contagion. In addition, preferences may be shaped by the comparison of one’s actions or outcomes to those of a peer or reference group. Deviating from the social norm established by the reference group decreases utility, either because the group explicitly penalizes the individual or because deviating decreases individual implicit outcomes, such as social prestige or inclusion (for instance, an individual may choose to drink because he may risk not being invited next times his friends gather together). This motive is known as preference for conformism. Furthermore, spillovers in preferences could take place due to a purely imitative process if individuals have bounded rationality and for example, have preference uncertainty.
The second mechanism for peer influence, social learning, considers that the actions of others can be informative about the costs and benefits of one’s health-related choices (Bandura, 1977; Bikhchandani et al., 1992). In these models, the presence of friends who drink or smoke may provide evidence regarding the benefits of these activities. On the other hand, an environment in which nobody drinks or smokes could be considered an indication that these activities are not good choices.
The third mechanism operates when the costs of taking an action decrease as the number of individuals taking that same action increases (network externalities). These are market-mediated externalities caused by the existence of fixed costs. A healthy-food supermarket chain may require a critical mass of consumers to open a store in an area. Unless this critical mass is achieved, consumption of healthy food may be prohibitively expensive. Alternatively, a kid in school that only likes to play soccer may find it hard to exercise if only a few other kids in the school like soccer. Network externalities can also take place when joint childbearing decreases the costs of raising a child, because of shared resources, for example.
Social Interaction Mechanisms in Body Weight
Most studies have attributed peer effects in health-related behaviors to direct social interactions, either directly through influences in weight management behaviors or indirectly through social norms about the acceptability of body weight. This line of research gained especial attention due to the studies by Christakis and Fowler (2007, 2008a, 2008b, 2009), who claimed that different behaviors and conditions (including smoking, obesity and happiness) can spread by contagion through large social networks. Christakis and Fowler’s (2007) influential and controversial study on obesity concluded that peer effects are one of the culprits of the American obesity epidemic.
It is not clear whether peer influence in individual’s weight-related behaviors operates first through body weight norms and spills over to weight-related behaviors, or if it works the other way round. A few papers fail to find a mediating role for weight-related behaviors in the explanation of peer influence in body weight (Cawley et al., 2017; Fletcher & Yakusheva, 2016).
Many studies are in line with the idea that the willingness to conform to social norms is a key driver of peer effects in obesity. Graham and Felton (2005) suggest that changes in norms about appearance are key to explaining the variance in obesity incidence across socioeconomic cohorts in the United States. Eisenberg and Quinn (2006) show that being part of a social group where other members recently gained weight leads an individual to adopt obesity-related behaviors. Etilé (2007) analyzes whether social norms about ideal body weight (calculated by averaging individual perceptions of the ideal BMI over all observations within a reference group) predict individual’s perceptions of ideal body weight and individual’s food choices (e.g., alcohol, fat, sugar). He finds suggestive evidence that social norms and habitual BMI affect ideal BMI, which in turn influences food choices and, ultimately, actual BMI. His estimates imply that social norms of body shape are not resistant to changes in actual body weight. Gwozdz et al. (2015) find that parents are more likely to misperceive their children as being thinner than they actually are, as their peers’ BMI increases.
Interestingly, several papers suggest that social norms may affect preferences interactively rather than additively. In Pliner and Mann (2004), participants in an experiment eat more when led to believe that prior participants have eaten more. This effect is only present when individuals received palatable, rather than unpalatable, food. Babcock and Hartman (2010) show that study subjects who have been incentivized to increase gym usage, exercise more if they have more treated friends, whereas control subjects are not influenced by their treated or untreated peers. Yang and Huang (2014) find that an individual is likely to gain weight as his/her number of obese friends increases, but is unlikely to lose weight as the number of obese friends decreases.
Following the social-comparison theory (Festinger, 1954; Lockwood & Kunda, 1997; Miller, Turnbull, & McFarland, 1988), there is a growing literature (Bursztyn & Jensen, 2017) based on the idea that individuals evaluate themselves comparing their behavior and characteristics to those of others. Through this mechanism, individuals reduce uncertainty when they try to define their psychological self. For instance, Blanchflower, Van Landeghem, and Oswald (2009) provide evidence that perceptions of overweight among adults depend on peers’ BMI. In a study of young women in Europe, Costa-Font and Jofre-Bonet (2013) show that the larger peers’ body mass, the lower the likelihood of being anorexic. Nie Sousa-Poza, and He (2015) find that female Chinese teenagers whose peers have a higher BMI are, other things equal, less likely to consider themselves overweight. Arduini, Iorio, and Patacchini (2017) explore whether eating disorders are affected by social comparison. Taking advantage of Add Health data, they document significant negative effect of peers’ BMI on purging, on the onset age of eating disorders, and on the probability of developing eating disorder. The authors show that peers affect individual behavior through interpersonal comparisons as they find evidence that adolescents not only care about their absolute BMI, but also their relative one (i.e., the deviation of individual body size relative from peers’ body size).
Mechanisms Behind Social Interactions in Substance Use
Social norms can also be key in explaining social interactions in behaviors such as alcohol, tobacco, or drug use, although most of the literature on mechanisms does not come from economics (Chen, Chang, & He, 2003; Stormshak et al., 1999). Several authors posit that adolescents use substances to gain recognition and maintain their status among peers (Gaviria & Raphael, 2001; Mitchell & Amos, 1997; Plumridge, Fitzgerald, & Abel, 2002; Prinstein & Cillessen, 2003; Prinstein, Meade, & Cohen, 2003). Deviators (those who dare to say no) are likely to be punished through “ostracism or merciless bullying,” a mechanism that creates strong incentives to conform. Using Add Health data, Balsa, Homer, French, and Norton (2011) show that adolescents are socially rewarded (have more friendship nominations) when they conform to their peers’ alcohol use and penalized (to a lesser degree) when they consume above the level of their peers.
DeCicca et al. (2008) provide evidence in support of social norms as a key determinant of peer effects in smoking choices. The authors show that a social anti-smoking sentiment is more important than cigarette prices when it comes to explaining youth smoking participation. According to DeCicca et al. (2008, p. 745), “initiation decisions are typically made in adolescence and may be driven more by the desire for peer acceptance and other non-economic factors,” whereas “economic factors may play more of a role in decisions about cessation and daily demand.” Rodríguez-Planas and Sanz-de-Galdeano (2016) find that second-generation immigrant teenage girls with more gender equality in their country of ancestry are, ceteris paribus, relatively more likely to smoke, drink, and use drugs than boys.10
There is also evidence that social norms about substance use do not necessarily coincide with habitual prevalence. Adolescents are likely to perceive smoking or drinking behavior by their peers to be more salient than it really is (Reid et al., 2008; Steffian, 1999). Thus, teenage substance use may stem from a combination of inadequate beliefs and a taste for conformity.
Battaglini, Bénabou, and Tirole (2005) propose a theory of social influences that relates to substance use and, in particular, to how self-help groups (e.g. Alcoholic Anonymous, Narcotics Anonymous) leverage on the interaction with people with similar issues. As the authors note, these support groups promote informational interactions among members, rather than controlling their behavior by contracts, implicit agreements, or other commitment devices. Their theory explains how and when exposure to the choices made by a peer could be beneficial (or not) in the presence of self-control problems. First, observing how a peer deals with impulses and temptation improves self-control when the ego has at least a minimum level of expected self-control, but has the opposite effect when the ego has a low expected level of self-control. Second, the ideal friend is someone with a slightly worse self-control problem than the ego: this makes the ego’s successes more encouraging and her failures less discouraging. Intuitively, an individual can only get little benefit from having a friend who is too self-controlled or a friend who is too impulsive. Battaglini et al. (2017) find empirical evidence suggesting that interactions with peers that have similar impulses are more informative and, as a result, more likely to increase individual self-control.
Mechanisms Behind Social Interaction in Teen Sexual Activity and Pregnancy
As with other health behaviors, most studies attribute peer effects in teen pregnancy to social norms. Case and Katz (1991) and Fletcher and Yakusheva (2012) argue that exposure to high rates of adolescent pregnancy reduces the stigma cost of being a teenage mother, leading thus to higher rates of teenage pregnancy.
In some particular cases, however, the sign of the endogenous peer effect can be informative of social learning effects. If network externalities are not able to generate congestion effects, a negative correlation in peer behavior might indicate that peer effects are mainly responding to information mechanisms. One interesting example is the paper by Yakusheva and Fletcher (2015), where the probability of teen pregnancy drops after the realization of a close friend’s teen birth. As the authors explain, “we might imagine the possibility of a large amount of learning about the difficulties of being a teen mother if a high school friend has a child” (p. 29). The paper uses three different strategies to support the hypothesis of knowledge externalities. First, the negative peer effect is larger in schools with low fertility rates. Teenagers are expected to learn more from the teen childbearing experience of a peer in school as the rate of teen childbearing is smaller. Second, the peer effect increases when excluding women with a childbirth, who are expected to get less information from the experience of peers. Third, the treatment effect is smaller when the friend’s partner is ready to assume the parenting role: the harder the experience of peers, the larger should information externalities be.
Kuziemko (2006) also claims that social learning is likely to underlie her findings of peer influence across siblings in fertility decisions. She bases her claim on evidence that social influence is stronger for couples having their first baby, a result suggesting that new information is especially valued among non-experienced parents.
In addition, both Kuziemko (2006) and Monstad et al. (2011) suggest a role for network externalities behind peer influence in fertility across siblings, which would take place if siblings find it easier to share childrearing, information, and other resources when having a child within the same period.
Average vs. Aggregate Effects
When discussing direct social interactions, scholars usually think in terms of local-average models: peer behavior is viewed as the “social norm,” imposing a cost when individuals deviate from it. If that is the case, one should expect agents to conform to the average behavior of the reference group. But peer effects may also respond to the sum of peer efforts: the more friends involved in a particular activity, the higher the expected return from it. These are local-aggregate effects. The distinction between average and aggregate effects is far from being a minor issue, since these alternative models have substantially different policy implications. If we believe peer effects respond to the local-aggregate model, then the most effective policy would focus on one or more key agents. If we believe instead that the local-average model is the most appropriate to describe endogenous peer effects, adequate policies must tackle the norm of the group. In such case, policies must change the group perception regarding what should be considered a normal or appropriate behavior.
As discussed, the literature has put the spotlight on local-average effects when estimating and analyzing the mechanisms for health-related peer effects. It is worth asking whether that is the correct model. Liu et al. (2014) estimate a model allowing for two types of peer effects: average effects and aggregate effects. While average effects do not depend on the number of peers taking the action, aggregate effects do. Employing network intransitivity properties as well as measures of network centrality to identify both effects, the authors show that both social norms and aggregate effects explain peer effects in sports activities. Ajilore et al. (2014) also attempt to distinguish aggregate from average local effects in a model of peer effects in obesity and overweight. They find significant peer interactions in body mass index, which can be explained by both mechanisms of peer influence; the local average effect (social norm) is, however, much larger than the social aggregate effect. Peer influence for overweight status, on the other hand, appears to operate solely through social aggregate effects.
Some other studies suggest indirectly that the local-aggregate model might be suitable for peer effects in health-related behaviors. For example, Robalino (2016) uses Add Health data to analyze peer effects on cigarette smoking. He finds that the smoking propensity of the 20 percent most popular kids accounts for most of the aggregate peer effects in smoking, an effect that tends to persist over the years.
Mechanisms Dealing with Network Structure
Aiming for a better understanding of how health related behaviors can spread through social networks, scientists have developed theories on how network topology may affect this process. A notable example is the paper by Jackson, Rogers, and Zenou (2017), who suggest a taxonomy of networks characteristics, discussing the economic consequences of different social network structures. More precisely, the authors develop a framework that underscores four major network characteristics: degree distributions, homophily patterns, clustering and centrality of nodes. The degree of an agent is the number of connections to other agents. Homophily refers to the tendency of individuals to select similar individuals. Clustering measures the local dependence among the locations of the different links in a social network. And centrality usually refers to the degree of popularity of a particular agent in a given network. While Jackson and colleagues do not pay particular attention to health-related behaviors, scholars have found evidence of the importance of these characteristics when analyzing peer effects in health.
Homophily is a notable example. Peer similarity is not only a good predictor of connection between individuals, but also tells us about the potential relative strength of peer effects. Our review finds evidence of differential peer effects for individuals with different traits. For instance, Argys and Rees (2008) and Sen (2009) find different peer influence in smoking for males and females, whereas Yakusheva et al. (2014) find that peers’ weight increases ego’s weight gain for females, but has no influence on males.11
Centola (2011) studies the role of homophily in the process of adoption of health-related behaviors and innovations. He notices that the degree of similarity in social contacts can help promote diffusion, but can also exclude less healthy individuals from interactions with healthier peers. Conducting an internet-based social network experiment, Centola shows that those networks with more homophilous ties (i.e., where individual are clustered by gender, age and body mass index) exhibit more adoption than fully integrated networks (i.e., where participants are randomly mixed, regardless of individual traits). Moreover, the results show homophily does not restrict adoption of a health behavior only to healthier individuals, but instead increases the adoption by those who are most in need to adopt it. In line with Centola (2011), Yakusheva et al. (2014) find that peer effects in weight gain are stronger across peers with similar academic achievement and similar political and religious views.
Taking advantage of the same controlled experiment, Centola (2010) also shows the relevance of another characteristic highlighted by Jackson et al. (2017): clustering. According to his findings, the greater the clustering of the network topology, the more effective and faster these networks are in spreading a particular health-related behavior (i.e., to more people and in less time). This result supports Centola and Macy’s (2007) statement that people usually need multiple adopters to convince themselves of implementing a given health behavior.12 Interestingly, Centola (2010) also shows that social networks with more clustering are better for spreading a health behavior widely, even if the maximum distance across all pairs of nodes is larger than in other social structures (i.e., it has a larger diameter).
While most studies treat links between nodes as binary values (they can only be “on” or “off,” Jackson et al., 2017, pp. 64–65), empirical evidence suggests incorporating link heterogeneity and distinguishing weak ties from strong ones (Granovetter, 1973). Peer influence might be stronger between peers in reciprocated friendships (or undirected links), whereas the non-reciprocated (or directed) links are associated with weaker peer effects (e.g., Card & Giuliano, 2013; Christakis & Fowler, 2007; Fowler & Christakis, 2008).
Another way to incorporate heterogeneity in peer influence is by considering different hierarchical statuses. According to this approach, peer influence will depend on the relative standings of agents, where peers of higher status are expected to be the influencers, and those with lower statuses are expected to be the adopters. The studies by Yakusheva et al. (2014), Robalino (2016), and Cawley et al. (2017) find evidence of hierarchical peer effects in weight gain, smoking, and BMI, respectively. In Yakusheva et al. (2014), students are more likely to be influenced by peers that are slimmer, of higher socioeconomic status, and more sexually experienced relative to themselves. In Robalino (2016) only the 20 percent of the distribution with higher status exerts a significant influence on others. Cawley et al. (2017) shows that peer influences on body weight work from older to younger siblings.
Peer Effects in Economic Fundamentals
Finally, rather than focusing on direct influences in behavior, peer effects in health behaviors could result from peer influence in economic fundamentals. For example, peers may have similar attitudes towards substance use or sexual behavior because they exert an influence on each other’s risk attitudes or time preferences. Exploiting random assignment of MBA students to peer groups, Ahern et al. (2014) find positive peer effects in risk aversion, which they attribute to a desire for conformity. Balsa, Gandelman, and González (2015) estimate peer effects in risk attitudes in a sample of high-school students. They find a significant and quantitative large impact of peers’ risk attitudes on male individuals’ coefficient of risk aversion, but not on women’s.
What We Know and What We Do Not Know
In spite of the fact that no single econometric technique has solved for all the challenges involved in the estimation of peer effects, there is consistence evidence of social influence in alcohol use, body weight, weight-related behaviors, teen pregnancy, and teen sexual initiation across a wide range of populations, reference group definitions, and estimation techniques. The evidence is mixed, on the other hand, for tobacco, illicit drug use, and mental health conditions, although there is some evidence of positive peer effects in the utilization of mental health services.
Despite the widespread evidence on the existence of strategic complementarities in health behaviors, economists still know little about the mechanisms behind peer influence. Because a large body of studies rely on the use of peers’ family background characteristics for identification, scholars are unable to determine in these cases whether the effects reflect endogenous or contextual influence. Recent efforts have turned the spotlight on network structure and how it might impact the underling mechanisms of social interactions and peer influence. While these efforts are promising, the literature has yet to advance in addressing the endogenous formation of network links.
Understanding the mechanisms behind peer influence is key to the design of effective policies, as each mechanism has different implications in terms of how to promote healthy behaviors. Nakamura, Suhrcke, and Zizzo (2017) provide a good summary of the policy implications triggered by different sources of peer effects. If individuals learn from others about the potential health consequences of their behavior, the most appropriate policy would be to design educational policies that provide information about the consequences of others’ health-related behavior. As Nakamura and colleagues notice, empirical evidence indicates that over-estimation of peer smoking prevalence is a key determinant of smoking for adolescents (Reid, Manske, & Leatherdale, 2008). Providing information about real prevalence may contribute to adjust adolescents’ beliefs. If, on the other hand, peer effects stem primarily from social comparisons, providing information about the “right” behavior would not work. In this context, Nakamura and colleagues suggest manipulating the incentive system to increase the opportunity cost of conforming (e.g., punishing unhealthy behaviors or subsidizing healthy options). Finally, if individuals’ utility decreases when they are seen as deviating from a social norm, a possible solution would be to manipulate the perceived norm. In this case, Nakamura and colleagues’ suggestion is to take advantage of media power on individual’s perception regarding what others consider desirable and, for instance, campaigning to modify the normatively justified belief about the way people ought to behave. Leonard et al. (2014) propose geography-specific social marketing and education campaigns to address social norms related to diet patterns, as well as the type of food provided by non-profit agencies and peer behavior. Christakis and Fowler (2008b) suggest focusing on group-level interventions, in particular programs that create artificial social network ties (e.g. Alcoholics Anonymous, weight loss groups, symptom support groups), a prescription which is supported by the results in Battaglini, Díaz, and Patacchini (2017).
A deeper comprehension of the mechanisms behind peer influence and the use of this understanding to develop policy recommendations constitute the main challenges of the research agenda in peer effects for the years to come.
Akerlof, G. A., & Kranton, R. E. (2000). Economics and identity. Quarterly Journal of Economics, 115, 715–754.Find this resource:
Bramoullé, Y., Djebbari, H., & Fortin, B. (2009). Identiﬁcation of peer effects through social networks. Journal of Econometrics, 150, 41–55.Find this resource:
Brock, W. A., & Durlauf, S. N. (2001a). Interactions-based models. In J. J. Heckman & Edward Leamer (Eds.), Handbook of Econometrics (Vol. 5, pp. 3297–3380). Amsterdam, The Netherlands: Elsevier.Find this resource:
Card, D., & Giuliano, L. (2013). Peer effects and multiple equilibria in the risky behavior of friends. The Review of Economics and Statistics, 95, 1130–1149.Find this resource:
Cawley, J., & Ruhm, C. J. (2012). The economics of risky health behaviors. In Mark V. Pauly, Thomas G. Mcguire, & Pedro P. Barros (Eds.), Handbook of Health Economics (Vol. 2, pp. 95–199). Amsterdam, Netherlands: Elsevier.Find this resource:
Jackson, M. O. (2010). Social and Economic Networks. Princeton, NJ: Princeton University Press.Find this resource:
Jackson, M. O., Rogers, B. & Zenou, Y. (2016). Networks: An Economic Perspective. Oxford Handbook of Social Network Analysis. Oxford: Oxford University Press.Find this resource:
Jackson, M. O, Rogers, B. W., & Zenou, Y. (2017). The economic consequences of social-network structure. Journal of Economic Literature, 55, 49–95.Find this resource:
Liu, X., Patacchini, E., & Zenou, Y. (2014). Endogenous peer effects: Local aggregate or local average? Journal of Economic Behavior & Organization, 103, 39–59.Find this resource:
Manski, C. F. (2000). Economic analysis of social interactions. Journal of Economic Perspectives, 68(3), 115–136.Find this resource:
Sacerdote, B. (2014). Experimental and quasi-experimental analysis of peer effects: Two steps forward? Annual Review of Economics, 6, 253–272.Find this resource:
Sotoudeh, R., Conley, D., & Harris, K. M. (2017). The Influence of peer genotypes and behavior on smoking outcomes: Evidence from Add Health (Working Paper No. 24113). Washington, DC: National Bureau of Economic Research.Find this resource:
Yakusheva, O., Kapinos, K. A., & Eisenberg, D. (2014). Estimating heterogeneous and hierarchical peer effects on body weight using roommate assignments as a natural experiment. Journal of Human Resources, 49, 234–261.Find this resource:
Ahern, K. R., Duchin, R., & Shumway, T. (2014). Peer Effects in risk aversion and trust. Review of Financial Studies, 27, 3213–3240.Find this resource:
Ajilore, O., Amialchuk, A., Xiong, W., & Ye, X. (2014). Uncovering peer effects mechanisms with weight outcomes using spatial econometrics. Social Science Journal, 51, 645–651.Find this resource:
Ajilore, O., Amialchuk, A., & Egan, K. (2016). Alcohol consumption by youth: Peers, parents, or prices? Economics & Human Biology, 23, 76–83.Find this resource:
Akerlof, G. A., & Kranton, R. E. (2000). Economics and identity. Quarterly Journal of Economics, 115, 715–754.Find this resource:
Ali, M. M., Amialchuk, A., & Dwyer, D. S. (2011). The social contagion effect of marijuana use among adolescents. PLoS One, 6(1), e16183.Find this resource:
Ali, M. M., Amialchuk, A., & Heiland, F. W. (2011). Weight-related behavior among adolescents: The role of peer effects. PLoS One, 6(6), e21179.Find this resource:
Ali, M. M., & Dwyer, D. S. (2011). Estimating peer effects in sexual behavior among adolescents. Journal of Adolescence, 34, 183–190.Find this resource:
Aral, S., & Nicolaides, C. (2017). Exercise contagion in a global social network. Nature Communication, 8, 14753.Find this resource:
Aral, S., & Walker, D. (2012). Identifying influential and susceptible members of social networks. Science, 337, 337–341.Find this resource:
Arduini, T., Iorio, A., & Patacchini, E. (2017). Weight, reference points and the onset of eating disorders. IZA Discussion Paper, Institute of Labor Economics.Find this resource:
Argys, L. M., & Rees, D. I. (2008). Searching for peer group effects: A test of the contagion hypothesis. The Review of Economics and Statistics, 90, 442–458Find this resource:
Auld, M. C. (2011). Effect of large-scale social interactions on body weight. Journal of Health Economics, 30, 303–316.Find this resource:
Babcock, P. S., & Hartman, J. L. (2010). Networks and workouts: Treatment size and status specific peer effects in a randomized field experiment. Working Paper No. w16581. National Bureau of Economic Research.Find this resource:
Balsa, A. I., French, M. T., & Regan, T. L. (2014). Relative deprivation and risky behaviors. Journal of Human Resources, 49, 446–471.Find this resource:
Balsa, A. I., Homer, J. F., French, M. T., & Norton, E. C. (2011). Alcohol use and popularity: Social payoffs from conforming to peers’ behavior. Journal of Research on Adolescence, 21, 559–568.Find this resource:
Balsa, A. I., Gandelman, N., & González, N. (2015). Peer effects in risk aversion. Risk Analysis, 35, 27–43.Find this resource:
Balsa, A. I., Gandelman, N., & Roldán, F. (2018). Peer and parental influence in academic skills and risky behavior. Labour Economics, 55, 41–55.Find this resource:
Bandura, A. (1977). Social Learning Theory. New York, NY: General Learning Press.Find this resource:
Banerjee, A. V. (1992). A simple model of herd behavior. Quarterly Journal of Economics, 107, 797–817.Find this resource:
Battaglini, M., Bénabou, R., & Tirole, J. (2005). Self-control in peer groups. Journal of Economic Theory, 123, 105–134.Find this resource:
Battaglini, M., Díaz, C., & Patacchini, E. (2017). Self-control in peer groups: An empirical analysis. Journal of Economic Behavior & Organization, 134, 240–254.Find this resource:
Bifulco, R., Fletcher, J. M., & Ross, S. L. (2011). The effect of classmate characteristics on post-secondary outcomes: Evidence from the Add Health. American Economic Journal: Economic Policy, 3, 25–53.Find this resource:
Bikhchandani, S., Hirshleifer, D., & Welch, I. (1992). A theory of fads, fashion, custom, and cultural change as informational cascades. Journal of Political Economy, 100, 992–1026.Find this resource:
Black, S. E., Devereux, P. J., & Salvanes, K. G. (2013). Under pressure? The effect of peers on outcomes of young adults. Journal of Labor Economics, 31, 119–153.Find this resource:
Blanchflower, D. G., Van Landeghem, B., & Oswald, A. J. (2009). Imitative obesity and relative utility. Journal of the European Economic Association, 7, 528–538.Find this resource:
Bound, J., & Krueger, A. B. (1991). The extent of measurement error in longitudinal earnings data: Do two wrongs make a right? Journal of Labor Economics, 9, 1–24.Find this resource:
Boyd, N. R., Windsor, R. A., Perkins, L. L., & Lowe, J. B. (1998). Quality of measurement of smoking status by self-report and saliva cotinine among pregnant women. Maternal and Child Health Journal, 2, 77–83.Find this resource:
Bramoullé, Y., Djebbari, H., & Fortin, B. (2009). Identiﬁcation of peer effects through social networks. Journal of Econometrics, 150, 41–55.Find this resource:
Brener, N. D., Billy, J. O., & Grady, W. R. (2003). Assessment of factors affecting the validity of self-reported health-risk behavior among adolescents: Evidence from the scientific literature. Journal of Adolescent Health, 33, 436–457.Find this resource:
Brock, W. A., & Durlauf, S. N. (2001a). Interactions-based models. In J. J. Heckman & Edward Leamer (Eds.), Handbook of Econometrics (Vol. 5, pp. 3297–3380). Amsterdam, Netherlands: Elsevier.Find this resource:
Brock, W. A., & Durlauf, S. N. (2001b). Discrete choice with social interactions. Review of Economics Studies, 68, 235–260.Find this resource:
Burkhauser, R. V., & Cawley, J. (2008). Beyond BMI: The value of more accurate measures of fatness and obesity in social science research. Journal of Health Economics, 27, 519–529.Find this resource:
Bursztyn, L., & Jensen, R. (2017). Social image and economic behavior in the field: Identifying, understanding, and shaping social pressure. Annual Review of Economics, 9, 131–153.Find this resource:
Card, D., & Giuliano, L. (2013). Peer effects and multiple equilibria in the risky behavior of friends. Review of Economics and Statistics, 95, 1130–1149.Find this resource:
Carrell, S. E., Hoekstra, M., & West, J. E. (2011). Is poor fitness contagious? Evidence from randomly assigned friends. Journal of Public Economics, 95, 657–663.Find this resource:
Case, A. C., & Katz L. F. (1991). The company you keep: The effects of family and neighborhood on disadvantaged youth. Working Paper No. 3705, National Bureau of Economic Research. New York, NY: NBER.Find this resource:
Cawley, J., & Burkhauser, R. V. (2006). The importance of objective health measures in predicting early receipt of social security benefits: The case of fatness. Working Paper No. wp148, University of Michigan, Michigan Retirement Research Center.Find this resource:
Cawley, J., Han, E., Kim, J. J., & Norton, E. C. (2017). Testing for peer effects using genetic data (Working Paper No. w23719). National Bureau of Economic Research. New York, NY: NBER.Find this resource:
Cawley, J., & Ruhm, C. J. (2012). The economics of risky health behaviors. In Mark V. Pauly, Thomas G. Mcguire, & Pedro P. Barros (Eds.), Handbook of Health Economics (vol. 2, pp. 95–199). Amsterdam, The Netherlands: Elsevier.Find this resource:
Centola, D. (2010). The spread of behavior in an online social network experiment. Science, 329(5996),1194–1197.Find this resource:
Centola, D. (2011). An experimental study of homophily in the adoption of health behavior. Science, 334(6060), 1269–1272.Find this resource:
Centola, D., & Macy, M. (2007). Complex contagions and the weakness of long ties. American Journal of Sociology, 113, 702–734.Find this resource:
Chen, X., Chang, L., & He, Y. (2003). The peer group as a context: Mediating and moderating effects on relations between academic achievement and social functioning in Chinese children. Child Development, 74, 710–727.Find this resource:
Christakis, N. A., & Fowler, J. H. (2007). The spread of obesity in a large social network over 32 years. New England Journal of Medicine, 357, 370–379.Find this resource:
Christakis, N. A., & Fowler, J. H. (2008a). The collective dynamics of smoking in a large social network. New England Journal of Medicine, 358, 2249–2258.Find this resource:
Christakis, N. A., & Fowler, J. H. (2008b). Dynamic spread of happiness in a large social network: longitudinal analysis over 20 years in the Framingham Heart Study. British Medical Journal, 337, a2338.Find this resource:
Christakis, N. A., & Fowler, J. H. (2009). Connected: The surprising power of our social networks and how they shape our lives. New York, NY: Little, Brown.Find this resource:
Clark, A. E., & Etilé, F. (2006). Don’t give up on me baby: Spousal correlation in smoking behaviour. Journal of Health Economics, 25, 958–978.Find this resource:
Clark, A. E., & Lohéac, Y. (2007). “It wasn’t me, it was them!” Social influence in risky behavior by adolescents. Journal of Health Economics, 26(4), 763–784.Find this resource:
Cohen-Cole, E., & Fletcher, J. M. (2008). Is obesity contagious? Social networks vs. environmental factors in the obesity epidemic. Journal of Health Economics, 27, 1382–1387.Find this resource:
Costa-Font, J., & Jofre-Bonet, M. (2013). Anorexia, body image and peer effects: Evidence from a sample of European women. Economica, 80(317), 44–64.Find this resource:
Cruwys, T., Bevelander, K. E., & Hermans, R. C. (2015). Social modeling of eating: A review of when and why social influence affects food intake and choice. Appetite, 86, 3–18.Find this resource:
Cutler, D. M., & Glaeser, E. L. (2010). Social interactions and smoking. In D. A. Wise (ed.), Research Findings in the Economics of Aging (123–141). Chicago, IL: University of Chicago Press.Find this resource:
DeCicca, P., Kenkel, D., Mathios, A., Shin, Y., & Lim Y. (2008). Youth smoking, cigarette prices, and anti-smoking sentiment. Health Economics, 17, 733–749.Find this resource:
Deaton, A. (1995). Data and econometric tools for development analysis. In Jere Behrman & T. N. Srinivasan (Eds.), Handbook of Development Economics (vol. 3A, pp. 1785–1882). Amsterdam, Netherlands: North Holland.Find this resource:
Duncan, G. J., Boisjoly, J., Kremer, M., Levy, D. M., & Eccles, J. (2005). Peer effects in drug use and sex among college students. Journal of Abnormal Child Psychology, 33, 375–385.Find this resource:
Edoka, I. P. (2017). Implications of misclassification errors in empirical studies of adolescent smoking behaviours. Health Economics, 26, 486–499.Find this resource:
Eisenberg, D., Golberstein, E., Whitlock, J. L., & Downs, M. F. (2013). Social contagion of mental health: evidence from college roommates. Journal of Health Economics, 22, 965–986.Find this resource:
Eisenberg, D., Golberstein, E., & Whitlock, J. L. (2014). Peer effects on risky behaviors: New evidence from college roommate assignments. Journal of Health Economics, 33, 126–138.Find this resource:
Eisenberg, D., & Quinn, B. C. (2006). Estimating the effect of smoking cessation on weight gain: An instrumental variable approach. Health Service Research, 41, 2255–2266.Find this resource:
Etilé, F. (2007). Social norms, ideal body weight and food attitudes. Health Economics, 16, 945–966.Find this resource:
Evans, W. N., Oates, W. E., & Schwab, R. M. (1992). Measuring peer group effects: A study of teenage behavior. Journal of Political Economy, 100, 966–991.Find this resource:
Festinger, L. (1954). A theory of social comparison processes. Human Relations, 7, 117–140.Find this resource:
Fletcher, J. M. (2007). Social multiplier in sexual initiation decisions among US high-school students. Demography, 44, 373–388.Find this resource:
Fletcher, J. M. (2010). Social interactions and smoking: Evidence using multiple student cohorts, instrumental variables, and school fixed effects. Journal of Health Economics, 19, 466–484.Find this resource:
Fletcher, J. M. (2012). Peer influences on adolescent alcohol consumption: evidence using an instrumental variables/fixed effect approach. Journal of Population Economics, 25, 1265–1286.Find this resource:
Fletcher, J., & Yakusheva, O. (2016). Peer effects on teenage fertility: Social transmission mechanisms and policy recommendations. American Journal of Health Economics, 2, 300–317.Find this resource:
Fortin, B., & Yazbeck, M. (2015). Peer effects, fast food consumption and adolescent weight gain. Journal of Health Economics, 42, 125–138.Find this resource:
Fowler, J. H., & Christakis, N. A. (2008). Estimating peer effects on health in social networks: A response to Cohen-Cole and Fletcher; Trogdon, Nonnemaker, Pais. Journal of Health Economics, 27, 1400–1405.Find this resource:
Gaviria, A., & Raphael, S. (2001). School-based peer effects and juvenile behavior. Review of Economics and Statistics, 83, 257–268.Find this resource:
Glaeser, E. L., Sacerdote, B. I., & Scheinkman, J. A. (2003). The social multiplier. Journal of the European Economic Association, 1, 345–353.Find this resource:
Glaeser, E. L., & Scheinkman, J. A. (2001). Measuring social interactions. In S. Durlauf, & P. Young (Eds.), Social dynamics (pp. 83–133). Cambridge, MA: MIT Press.Find this resource:
Glovannucci, E., Colditz, G., Stampfer, M. J., Rimm, E. B., Litin, L., Sampson, L., & Willett, W. C. (1991). The assessment of alcohol consumption by a simple self-administered questionnaire. American Journal of Epidemiology, 133, 810–817.Find this resource:
Golberstein, E., Eisenberg, D., & Downs, M. F. (2016). Spillover effects in health service use: Evidence from mental health care using first-year college housing assignments. Journal of Health Economics, 25, 40–55.Find this resource:
Graham, G., & Felton, A. (2005). Variance in obesity across cohorts and countries: A norms-based explanation using happiness surveys. Washington, DC: Brookings Institution.Find this resource:
Granovetter, M. S. (1973). The strength of weak ties. American Journal of Sociology, 78, 1360–1380.Find this resource:
Gwozdz, W., Sousa-Poza, A., Reisch, L. A., Bammann, K., Eiben, G., Kourides, Y. . . Pigeot, I. (2015). Peer effects on obesity in a sample of European children. Economics & Human Biology, 18, 139–152.Find this resource:
Halliday, T. J., & Kwak, S. (2009). Weight gain in adolescents and their peers. Economics & Human Biology, 7(2), 181–190.Find this resource:
Halliday T. J., & Kwak, S. (2012). What is a peer? The role of network definitions in estimation of endogenous peer effects. Applied Economics, 44, 289–302.Find this resource:
Harris, J., & González López-Valcárcel, B. (2008). Asymmetric peer effects in the analysis of cigarette smoking among young people in the United States, 1992–1999. Journal of Health Economics, 27 (2), 249–264;.Find this resource:
Hanushek, E. A., Kain, J. F., Markman, J. M., & Rivkin, S. G. (2003). Does peer ability affect student achievement? Journal of Applied Econometrics, 18, 527–544.Find this resource:
Hirshleifer, D. (1995). The blind leading the blind. Social influence, fads, and informational cascades. In M. Tomassi, & K. Ierulli (Eds.), The New Economics of Human Behaviour. Cambridge, U.K.: Cambridge University Press.Find this resource:
Ho, C. Y. (2017). Estimating sibling spillovers in health: Evidence on symptoms. Economics & Human Biology, 27, 93–101.Find this resource:
Hsieh, C.-S., & Lee, L. F. (2015). A social interactions model with endogenous friendship formation and selectivity. Journal of Applied Econometrics, 31, 301–319.Find this resource:
Huybrechts, I., De Bacquer, D., Van Trimpont, I., De Backer, G., & De Henauw, S. (2006).Validity of parentally reported weight and height for pre-school-aged children in Belgium and its impact on classification into body mass index categories. Pediatrics, 118, 2018–2109.Find this resource:
Jackson, M. O, Rogers, B. W., & Zenou, Y. (2017). The economic consequences of social-network structure. Journal of Economic Literature, 55, 49–95.Find this resource:
Kapinos, K. A., Yakusheva, O., & Eisenberg, D. (2014). Obesogenic environmental influences on young adults: Evidence from college dormitory assignments. Economics & Human Biology, 12, 98–109.Find this resource:
Kawaguchi, D. (2004). Peer effects on substance use among American teenagers. Journal of Population Economics, 17, 351–367.Find this resource:
Kling, J. R., Liebman, J. B., & Katz, L. F. (2007). Experimental analysis of neighborhood effects. Econometrica, 75, 83–119.Find this resource:
Krauth, B. V. (2005). Peer effects and selection effects on smoking among Canadian youth. Canadian Journal of Economics, 38, 735–757.Find this resource:
Krauth, B. V. (2006). Simulation-based estimation of peer effects. Journal of Econometrics, 133, 243–271.Find this resource:
Krauth, B. V. (2007). Peer and selection effects on youth smoking in California. Journal of Business & Economic Statistics, 25, 288–298.Find this resource:
Kremer, M., & Miguel, E. (2007). The illusion of sustainability. Quarterly Journal of Economics, 122, 1007–1065.Find this resource:
Kuziemko, I. (2006). Is having babies contagious? Estimating fertility peer effects between siblings. Working paper, Harvard University.Find this resource:
Leonard, T., McKillop, C., Carson, J. A., & Shuval, K. (2014). Neighborhood effects on food consumption. Journal of Behavioral and Experimental Economics, 51, 99–113.Find this resource:
Lim, J., & Meer, J. (2018). How do peers influence BMI? Evidence from randomly assigned classrooms in South Korea. Social Science & Medicine, 197, 17–23.Find this resource:
Lin, Xu. (2014). Peer effects in adolescents’ delinquent behaviors: Evidence from a binary choice network model. Regional Science and Urban Economics, 46, 73–92.Find this resource:
Liu, X., Patacchini, E., & Zenou, Y. (2014). Endogenous peer effects: Local aggregate or local average? Journal of Economic Behavior & Organization, 103, 39–59.Find this resource:
Lockwood, P., & Kunda, Z. (1997). Superstars and me: Predicting the impact of role models on the self. Journal of Personality and Social Psychology, 73, 91–103.Find this resource:
Loh, C. A., & Li, Q. (2013). Peer effects in adolescent bodyweight: Evidence from rural China. Social Science & Medicine, 86, 35–44.Find this resource:
Ludwig J., Liebman J. B., Kling J. R., Duncan, G. J., Katz, L. F., Kessler, R. C., & Sanbonmatsu, L. (2008). What can we learn about neighborhood effects from the moving to opportunity experiment?American Journal of Sociology, 114, 144–188.Find this resource:
Ludwig, J., Sanbonmatsu, L., Gennetian, L., Adam, E., Duncan, G. J., Katz . . . McDade, T. W. (2011). Neighborhoods, obesity, and diabetes–A randomized social experiment. New England Journal of Medicine, 365(16), 1509–1519.Find this resource:
Lundborg, P. (2006). Having the wrong friends? Peer effects in adolescent substance use. Journal of Health Economics, 25, 214–233.Find this resource:
Manski, C. F. (1993). Identification of endogenous social effects: The reflection problem. Review of Economic Studies, 60, 531–542.Find this resource:
Manski, C. F. (2000). Economic analysis of social interactions. Journal of Economic Perspectives, 68(3), 115–136.Find this resource:
Manski, C. F. (2005). Social choice with partial knowledge of treatment response. Princeton, NJ: Princeton University Press.Find this resource:
McVicar, D., & Polanski, A. (2014). Peer effects in UK adolescent substance use: Never mind the classmates? Oxford Bulletin of Economics and Statistics, 76, 589–604.Find this resource:
Micklewright, J., Schnepf, S. V. & Silva, P. N. (2012). Peer effects and measurement error: the impact of sampling variation in school survey data (evidence from PISA). Economics of Education Review, 31, 1136–1142.Find this resource:
Mitchell, L., & Amos, A. (1997). Girls, pecking order and smoking. Social Science and Medicine, 44, 1861–1869.Find this resource:
Miller, D. T., Turnbull, W., & McFarland, C. (1988). Particularistic and universalistic evaluation in the social comparison process. Journal of Personality and Social Psychology, 55, 908–917.Find this resource:
Moffitt, R. A. (2001). Policy interventions, low-level equilibria and social interactions. In Durlauf, S., & Young, P. (Eds.), Social Dynamics (pp. 45–82). Cambridge, MA: MIT Press.Find this resource:
Monstad, K., Propper, C., & Salvanes, K. G. (2011). Is teenage motherhood contagious? evidence from a natural experiment. Discussion Paper Series in Economics 12/2011, Bergen: Norway: Department of Economics, Norwegian School of Economics.Find this resource:
Mora, T., & Gil, J. (2013). Peer effects in adolescent BMI: Evidence from Spain. Health Economics, 22, 501–516.Find this resource:
Moriarty, J., McVicar, D., & Higgins, K. (2016). Cross-section and panel estimates of peer effects in early adolescent cannabis use: With a little help from my “friends once removed.” Social Science & Medicine, 163, 37–44.Find this resource:
Nakajima, R. (2007). Measuring peer effects on youth smoking behaviour. Review of Economic Studies, 74, 897–935.Find this resource:
Nakamura, R., Suhrcke, M., & Zizzo, D. J. (2017). A triple test for behavioral economics models and public health policy. Theory and Decision, 83, 513–533.Find this resource:
Nawaz, H., & Katz, D. L. (2001). American College of Preventive Medicine Practice Policy statement. Weight management counseling of overweight adults. American Journal of Preventive Medicine, 21, 73–78.Find this resource:
Nie, P., Sousa-Poza, A., & He, X. (2015). Peer effects on childhood and adolescent obesity in China. China Economic Review, 35, 47–69.Find this resource:
Norton, E. C., Lindrooth, R. C., & Ennett, S. T. (2003). How measures of perception from survey data lead to inconsistent regression results: evidence from adolescent and peer substance use. Health Economics, 12, 139–148.Find this resource:
Pachucki, M. C., Ozer, E. J., Barrat, A., & Cattuto, C. (2015). Mental health and social networks in early adolescence: A dynamic study of objectively-measured social interaction behaviors. Social Science & Medicine, 125, 40–50.Find this resource:
Patrick, D. L., Cheadle, A., Thompson, D. C., Diehr, P., Koepsell, T., & Kinne, S. (1994). The validity of self-reported smoking: A review and meta-analysis. American Journal of Public Health, 84, 1086–1093.Find this resource:
Pliner, P., & Mann, N. (2004). Influence of social norms and palatability on amount consumed and food choice. Appetite, 42, 227–237.Find this resource:
Plumridge, E. W., Fitzgerald, L. J., & Abel, G. M. (2002). Performing coolness: Smoking refusal and adolescent identities. Health Education Research, 17, 167–179.Find this resource:
Polich, J. M. (1982). The validity of self-reports in alcoholism research. Addictive Behaviors, 7, 123–132.Find this resource:
Powell, L. M., Tauras, J. A., & Ross, H. (2005). The importance of peer effects, cigarette prices and tobacco control policies for youth smoking behavior. Journal of Health Economics, 24, 950–968.Find this resource:
Prinstein, M. J., & Cillessen, A. H. N. (2003). Forms and functions of adolescent peer aggression associated with high levels of peer status. Merrill-Palmer Quarterly, 49, 310–342.Find this resource:
Prinstein, M. J., Meade, C. S., & Cohen, G. L. (2003). Adolescent oral sex, peer popularity, and perceptions of best friends’ sexual behavior. Journal of Pediatric Psychology, 28, 243–249.Find this resource:
Rebagliato, M. (2002). Validation of self-reported smoking. Editorial. Journal of Epidemiology and Community Health, 56, 163–164.Find this resource:
Reid, J. L., Manske, S. R., & Leatherdale, S. T. (2008). Factors related to adolescents’ estimation of peer smoking prevalence. Health Education Research, 23, 81–93.Find this resource:
Renna, F., Grafova, I. B., & Thakur, N. (2008). The effect of friends on adolescent body weight. Economics & Human Biology, 6, 377–387.Find this resource:
Rice, N., & Sutton M. (1998). Drinking patterns within households: the estimation and interpretation of individual and group variables. Journal of Health Economics, 7, 689–699.Find this resource:
Richard-Shubik, S. (2015). Peer effects in sexual initiation: Separating demand and supply mechanisms. Quantitative Economics, 6, 663–702.Find this resource:
Robalino, J. D. (2016). Smoking peer effects among adolescents: Are popular teens more influential? IZA Discussion Paper No. 9714. Bonn, Germany: Institute of Labor Economics.Find this resource:
Rodríguez-Planas, N., & Sanz-de-Galdeano, A. (2016). Social norms and teenage smoking: The dark side of gender equality. IZA Discussion Paper No. 10134. Bonn, Germany: Institute of Labor Economics.Find this resource:
Rosenquist, J. N., Fowler, J. H., & Christakis, N. A. (2011). Social network determinants of depression. Molecular Psychiatry, 16, 273–281.Find this resource:
Sacerdote, B. (2011). Peer effects in education: How might they work, how big are they and how much do we know thus far? Handbook of the Economics of Education, 3, 249–277.Find this resource:
Sacerdote, B. (2014). Experimental and quasi-experimental analysis of peer effects: Two steps forward? Annual Review of Economics, 6, 253–272.Find this resource:
Schoeller, D. A. (1995). Limitations in the assessment of dietary energy intake by self-report. Metabolism, 44, 18–22.Find this resource:
Sen, A. (2009). Estimating the impacts of household behavior on youth smoking: Evidence from Ontario, Canada. Review of Economics of the Household, 7, 189–218.Find this resource:
Shields, M., Connor Gorber, S., Janssen, I., & Tremblay, M. S. (2011). Obesity estimates for children based on parent-reported versus direct measures. Health Reports, 22, 1–12. Ottawa: Statistics Canada.Find this resource:
Soetevent, A. R., & Kooreman, P. (2007). A discrete‐choice model with social interactions: with an application to high school teen behavior. Journal of Applied Econometrics, 22, 599–624.Find this resource:
Sotoudeh, R., Conley, D., & Harris, K. M. (2017). The Influence of peer genotypes and behavior on smoking outcomes: Evidence from Add Health. Working Paper No. 24113. Washington, DC: National Bureau of Economic Research.Find this resource:
Steffian G. (1999). Correction of normative misperceptions: An alcohol abuse prevention program. Journal of Drug Education, 29, 115–138.Find this resource:
Stormshak, E. A., Bierman, K. L., Bruschi, C. J., Dodge, K. A., Coie, J. D., & Conduct Problems Prevention Research Group. (1999). The relation between behavior problems and peer preference in different classroom contexts. Child Development, 70, 169–182.Find this resource:
Townshend, J. M., & Duka, T. (2002). Patterns of alcohol drinking in a population of young social drinkers: A comparison of questionnaire and diary measures. Alcohol and Alcoholism, 37, 187–192.Find this resource:
Trogdon, J. G., Nonnemaker, J., & Pais, J. (2008). Peer effects in adolescent overweight. Journal of Health Economics, 27, 1388–1399.Find this resource:
Ugander, J., Backstrom, L., Marlow, C., & Kleinberg, J. (2012). Structural Diversity in Social Contagion. Proceedings of the National Academy of Sciences, 109, 5962–5966.Find this resource:
Vartiainen, E., Seppälä, T., Lillsunde, P., & Puska, P. (2002). Validation of self-reported smoking by serum cotinine measurement in a community-based study. Journal of Epidemiology & Community Health, 56, 167–170.Find this resource:
Wagenknecht, L. E., Burke, G. L., Perkins, L. L., Haley, N. J., & Friedman, G. D. (1992). Misclassification of smoking status in the CARDIA study: A comparison of self-report with serum cotinine levels. American Journal of Public Health, 82, 33–36.Find this resource:
Yakusheva, O., & Fletcher, J. (2015). Learning from Teen Childbearing Experiences of Close Friends: Evidence Using Miscarriages as a Natural Experiment. Review of Economic Studies, 97, 29–43.Find this resource:
Yakusheva, O., Kapinos, K., & Weiss, M. (2011). Peer effects and the freshman 15: Evidence from a natural experiment. Economics & Human Biology, 9, 119–132.Find this resource:
Yakusheva, O., Kapinos, K. A., & Eisenberg, D. (2014). Estimating heterogeneous and hierarchical peer effects on body weight using roommate assignments as a natural experiment. Journal of Human Resources, 49, 234–261.Find this resource:
Yang, M., & Huang, R. (2014). Asymmetric association between exposure to obesity and weight gain among adolescents. Eastern Economic Journal, 40, 96–118.Find this resource:
(1.) Some have argued that using peers’ family malleable characteristics as instruments may even suffer from a “second-order case of the reflection problem: friend’s parents’ weight may be affected by friend’s weight which in turn may be affected by the respondent’s weight” (Cawley & Ruhm, 2012, p. 136).
(2.) This is unlike the original discrete-choice model proposed by Brock and Durlauf (2001a), who assumed that individuals interact in large groups, have incomplete information about others’ actions and form rational expectations about others’ behaviors (approximated by the population average of the observed choices).
(3.) Add Health is a nationally representative panel of US adolescents aged between 14 and 19 in 1994/1995
(5.) Trogdon et al. (2008) also analyze overweight and find that a 10 percentage point increase in peers’ prevalence of overweight increases ego’s overweight by 3.9 percentage point in the full sample and by 5.3 percentage points in the case of female subjects.
(6.) Their set of instruments include peers’ parents’ BMI and the fraction of peers’ parents born during the Great Famine.
(7.) Peer effects show substantial variation among the regions, with generally larger effects in the Spanish, Italian, and Cypriot regions than in the German, Swedish, Belgian, and Hungarian areas.
(8.) Ludwig et al. (2011) study the effects on overweight and obesity of participation in the Moving to Opportunity (MTO) project, a U.S. initiative that randomly assigned housing vouchers to poor families conditional on their moving to low-poverty areas. Families who moved were less likely to be obese and had lower BMIs. While these results could be indicative of social influence, they could also reflect environmental features, such as better access to healthy food or gyms.
(9.) Kling, Liebman, and Katz (2007) and Ludwig et al. (2011) find estimates of neighborhood effects in mental health that could result either from social interactions or from changes in the environment. Both studies analyze the Moving to Opportunity (MTO) project, a large-scale randomized social experiment that offered housing vouchers to poor families with children, conditional on their moving to low-poverty neighborhoods. Four to seven years after the random assignment, adults and female youths in families offered the vouchers showed substantial mental health benefits related to distress, depression, anxiety, calmness, and sleep. Ten to fifteen years after program implementation, these effects persisted among girls. The leading hypothesis is that moving away from dangerous neighborhoods substantially reduced the levels of stress faced by families.
(10.) Given that these and other risk behaviors are usually associated with boys, the authors suggest that moving towards gender equality may have the unintended spillover of making girls relatively more prone to engage in unhealthy behaviors.
(12.) According to Aral and Nicolaides (2017), there are three theories that associate network structure with behavioral contagions: Centola and Macy’s (2007) Complex Contagion theory indicates that, in order to induce a costly behavior (e.g., to exercise, to follow a diet pattern), individuals usually demand a signal of adoption by different peers and hence, clustered networks are more likely to spread these complex contagions; Ugander, Backstrom, Marlow, and Kleinberg (2012) suggest the Structural Diversity theory, arguing that structural diversity (i.e., the number of unconnected clusters that have at least one adopter) is key to explaining contagion; Aral and Walker’s (2012) Embeddedness theory indicates that the number of mutual connections is what really matters when we try to explain behavioral contagions.