Homophily and Contagion Are Generically Confounded in Observational Social Network Studies

“Or rather are they close in the network because of these similarities, through the processes known assortative mixing on traits, or more simply as homophily (McPherson et al., 2001)?”

“…inter-connected clusters of nodes, with minimal connection between clusters — has been thoroughly explored in the recent literature (explicitly in Girvan and Newman (2002); Newman and Girvan (2003); Bickel and Chen (2009); Porter et al. (2009); Fortunato (2010), implicitly in much smaller…”

“…of nodes, with minimal connection between clusters — has been thoroughly explored in the recent literature (explicitly in Girvan and Newman (2002); Newman and Girvan (2003); Bickel and Chen (2009); Porter et al. (2009); Fortunato (2010), implicitly in much smaller clusters in Elwert and…”

“The “easy” solution is to use expert knowledge of the system to identify all causally relevant variables, measure a sufficient set of them, and adjust for them appropriately (Morgan and Winship, 2007; Pearl, 2000; Spirtes et al., 2001).”

“…with minimal connection between clusters — has been thoroughly explored in the recent literature (explicitly in Girvan and Newman (2002); Newman and Girvan (2003); Bickel and Chen (2009); Porter et al. (2009); Fortunato (2010), implicitly in much smaller clusters in Elwert and Christakis (2008)).”

“…of actual processes of transmission and influence, which have been extensively studied in social psychology and allied fields since the 1920s (Bartlett, 1932; Sperber, 1996; Huckfeldt et al., 2004; Friedkin, 1998).13 However, not only is it adequate to demonstrate the existence of the…”

“19”

“Indeed, as noted by other authors (Lyons, 2011; Shalizi & Thomas, 2011; Thomas, 2013; VanderWeele et al., 2012), the apparent causality of the network effect could be influenced by the network structure or simple network autocorrelation through homophily (e.g., association with clinicians who are…”

“Homophilous selection of social ties complicates attribution of subsequent changes in attitudes to social influence (Kandel 1978; Shalizi and Thomas 2011).”

“In general these two factors are confounded when conducting observational studies [54].”

“Moreover, the impact of direct communication can be confounded by two structural mechanisms of communication networks in which the players are situated [29], namely, social influence mechanism [3] and triadic closure mechanism [28].”

“Influence effects in peer groups create dependency pathways between friendship formation and the use of psychological services, and thus, violate the independence assumptions of commonly used statistical methods (Shalizi & Thomas, 2011).”

“hand, information diffusion is better modeled with complex contagion due to the common presence of exogenous factors [25] and more complex forms of endogenous influence which include various social reinforcement mechanisms such as reciprocity [35], social feedback [36] and homophily [37].”

“However, identifying and estimating peer-influence under the presence of homophily have long remained a challenging problem [2, 8, 9].”

“As the heated debate regarding the series of papers using the Framingham Heart Study—an observational dataset—has demonstrated [7, 8, 9, 10, 11, 12, 13, 14], causal inference based on observational data is a major challenge, and the effort to extract as much as information from observational…”

“Similarly, to the extent that gang-related violence in other cities fosters retaliatory violence that diffuses through social networks rather than euclidian space, future work must separate non-random network clustering from network diffusion [41].”

“We begin by presenting the graphical causal model used in [43], which demonstrated that the phenomena of homophily (the tendency of people to form social ties with people similar to them) and of behaviour adoption due to social in-fluence from friends are confounded in observational social network…”

“1In [1] the following illustrative example is provided for homophily behavior in social networks: “If your friend jumped off a bridge, would you jump too?” A possible reasons for answering “yes” is that you are friends as a result of your fondness for jumping off bridges.”

“” However, this attention also brought scrutiny, including critiques that the peer influence (contagion) effects were not identifiable and therefore could not be distinguished from homophily (birds of a feather flocking together) and that the mechanisms for spread over networks remained obscure (2).”

“These findings have sparked substantial interest in network members’ attitudes and behaviors, and have led to debates about whether these findings reflect actual social influence or homophily – individuals’ tendencies to select into relationships with people who are similar to them [30,99,134,150].”

“Prior work by Shalizi and Thomas has already identified certain other assumptions this test may involve [16], and we hope that further exploration of the utility of this approach will be forthcoming.”

“Therefore, it is necessary to account for these alternative mechanisms when estimating the magnitude of the social-influence effect (Manski 1993; Aral et al. 2009; Shalizi and Thomas 2011; VanderWeel and An 2013; Ogburn 2018).”

“…peer effects estimation apply to the fore-going two studies (Angrist 2014; Bramoullé, Djebbari, and Fortin 2020; Elwert and Winship 2014; Shalizi and Thomas 2011), including the possibility of unobserved selection into deskmate allocations within classrooms (Keller and Takács 2019),…”

“…it is difficult even to estimate the magnitude and significance of associations (Dow et al. 1982), and second, inferring causality in networks (Shalizi and Thomas 2011) is often impossible because of how networks are both an independent variable, influencing future outcomes, and a dependent…”

“…distinguishing between social influence and social selection is not possible without longitudinal data 13 , and both may act in concert, therefore our analytical strategy necessarily requires some opening assumptions 21 , which in turn inform our modelling decisions and results interpretation.”

“Next to interpersonal processes, unobserved individual or contextual factors may contribute to the reinforcement of similar or opposed attitudes between individuals Shalizi and Thomas (2011).”

“Socialinfluence proceeding with nonnegative relations is based on the principle of “homophily,” where individuals are likely to imitate the behaviors from or make the same choices as their positive-relation neighbors [52], [53].”

“However, the current empirical design cannot determine the direction of this influence or identify strict causality [98].”

“Although homoph-ily and influence are not fully distinguishable in empirical data, they are conceptually distinct (Shalizi and Thomas 2011).”

“Shalizi and Thomas [54] show that in networks formed by latent homophily, contagion, and homophily can be confounded and the causal effect is not always identifiable.”

“Estimating causal peer influence has been an active topic of research with advancements in methodologies reported in Manski [1993], Shalizi and Thomas [2011], VanderWeele [2011], VanderWeele and An [2013], Bramoullé et al. [2009].”

“4 Empirically disentangling social influence and homophily is challenging (see Aral et al., 2009; Levitan & Visser, 2009; Shalizi & Thomas, 2011; Steglich et al., 2010).”

“Ma and Tresp (2021) explore violations of the SUTVA assumption (i.e., treatment “spillover” or “interference”) by creating node representations from a GNN that aggregate treatment information of local neighbors and the structural position of nodes separately.”

“In contrast, in homophilic networks we find that minorities are at a disadvantage regardless of the type of error present.”

“(2009) emphasized moreintrinsic network structure characteristics, including homophily (Miller and Smith-Lovin 1987; Shalizi and Thomas 2011), transitivity (Hoff, Raftery, and Handcock 2002), clustering (Frank and Strauss 1986; Wang and Wong 1987), and degree heterogeneity (Girvan and Newman 2002;…”

“This work is motivated by theories of homophily [53, 70] and aims to take advantage of the fact that similar users (across various dimensions) gather around specific interests.”

“Distinguishing these pathways is difficult (Shalizi & Thomas, 2011), but important for the development of effective intervention strategies.”

“…establishing causality in network influence studies has proven to be conceptually and methodologically challenging (e.g., Bramoulle et al., 2009; Shalizi & Thomas, 2011) and many wellknown examples of contagion and influence (e.g., Christakis & Fowler, 2007; Coleman et al., 1966; Fowler &…”

“For the design of recommender systems, there are four factors affecting recipients’ advice-taking decisions [2,3], namely cognitive homophily [25], tie strength [18], trustworthiness [27] and social capital [10].”

“) [1], [2], [19] questionning Manski’s group effect as the driving force explaining the observed phenomenon.”

“does a person start smoking cigarettes because their friends who smoke have influenced their behavior, or does the propensity to smoke help explain why they are all friends in the first place? Recent analyses of Shalizi and Thomas [23] show that homophily and contagion are generically confounded in observational socia…”

“A major difficulty in our study is that individual unobserved heterogeneity can be correlated with the formation of the social network and with the adoption of the iPhone 3G [18]–[20].”

“Identifying social influence in the presence of homophily using only observational data has been proven to be generally impossible, without the assumption that all homophily-related confounds are observed [29].”

“This is especially true in relational domains where capturing latent confounders in causal models is key to separating homophily-based correlations from contagion [Shalizi and Thomas, 2011, Lee and Ogburn, 2021].”

“While network formation theory distinguishes between these processes [Jackson and Wolinsky, 1996], empirically separating their effects in observational data is a well-known identification problem [Manski, 1993, Shalizi and Thomas, 2011, Tomasello et al., 2014].”

“It is generally understood that social selection and social influence mechanisms are typically confounded in observational studies of social and other networks (Shalizi & Thomas, 2011; Steglich, Snijders, & Pearson, 2010).”

“In addition, it is acknowledged that many studies have been conducted to test for contagion (Aral et al., 2009; Fowler et al., 2011; Shalizi & Thomas, 2011) and find it challenging to demonstrate it definitively.”

“A major strength of our findings is that they are not susceptible to homophily bias—bias that arises because of the tendency for people with similar unmeasured characteristics to form relationships based on those characteristics [45].”

“It is helpful here to build conceptual bridges between what Shalizi and Thomas (2011) refer to as “latent homophliy” and conceptual characterizations of the confounding of influence and homophily in political science, specifically by Franzese, Hays, andKachi (2012).”

“Of course, distinguishing between homophily and influence is difficult at best and, in some cases, might not even be possible (Shalizi & Thomas, 2011).”

“A key example with special relevance to efforts to describe social contagion is described by Shalizi and Thomas (2011) in “Homophily and Contagion Are Generically Confounded in Observational Social Network Studies”.”

“These two phenomena are also consistent with the sociological findings of social consistency [1] and social contagion [5].”

“…is an extensive literature on causal inference methods for so called “network settings” from a variety of perspectives (Hudgens and Halloran, 2008; Shalizi and Thomas, 2011; Ogburn and VanderWeele, 2014; van der Laan, 2014; Sofrygin and van der Laan, 2017; DeAmour, 2016; Aronow and Samii, 2017;…”

“For the specific case of the communication network, the source of peer effect is subject to discussion in the literature and it is probably a mixture of both explanations (Aral et al., 2009; Ma et al., 2014; Shalizi & Thomas, 2011).”

“Namely, such correlations may arise because of the confounding effects of unobserved covariates, telling us little about whether friends’ adoption causes one to adopt (Shalizi and Thomas, 2011).”

“the tendency for people to become more alike given a connection [79].”

“Due to the interconnectedness of actors, observational research on social networks presents myriad confounding problems that place considerable limits on the feasibility of causal identification (Shalizi and Thomas 2011).”

“, 2001), (Shalizi and Thomas, 2011).”

“In general, distinguishing social influence in decision making from homophily, which is defined as the tendency for individuals with similar characteristics and choice behaviors to form clusters in social networks, is currently a growing area of research and debate [3, 16].”

“Although these claims have been criticized [16], one could imagine that people transitively influence each other via social relations.”

“Our results further suggest that the GEE-based models are able to distinguish network influence from its absence even when there is a more subtle homophily—so-called latent homophily [14] on an unobservable intrinsic parameter for rate of change, rather than the observed characteristic itself.”

“Shalizi and Thomas [2011, § 4.3] did hint at a possible approach for identi-fication of social influence, even in an homophilous network.”

“(SHALIZI et altr.,2011).”

“Although our model is built on the notions of homophily and social influence described above, we point out that differ-entiating between the effects of these processes, particularly in observational settings, may be very difficult [35, 36].”

“Shalizi and Thomas (2011) used the example of two people jumping from a bridge to avoid an oncoming train.”

“Confounding factors, on the other hand, can simultaneously affect the formation of friendships and subsequent actions [31, 36].”