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”

“However, we cannot discount selection effects in observational studies like ours [25].”

“However, in the viral marketing applications, it is important to distinguish influence-based contagion from homophily-driven diffusion in dynamic networks [14], [15], and the existing influence maximizing or minimizing method without considering the causal nature of the influence may be biased.”

“More recently, homophily has been shown to be statistically confounded with influence (Shalizi & Thomas, 2011), meaning statistics cannot be used to differentiate between similar behavior due to homophilous linking or due to influence of Actor A over B.”

“Of course, any effects of network composition on political outcomes can be driven by both selection and social in fl uence, and unobserved covariates can drive both tie selection and attitude change (Shalizi & Thomas, 2011).”

“, the effect of social influence cannot be identified) [4, 73, 74].”

“This allows for unobserved homophily, a wellknown obstacle to identification in observational network data (Shalizi and Thomas, 2011).”

“Social influence is also observed in social settings, where people who interact can persuade others around them to do or not do something [39].”

“Such structural determination may be one of the keystones to discern between diverse features of influence in social networks such as homophily, social contagion and covariation (Shalizi and Thomas 2011).”

“…in which some unobserved and shared trait drives both the tendency for two people to form a friendship and for each to adopt (Aral et al., 2009; Shalizi and Thomas, 2011); either social property will increase the ability to predict a person’s adoption behavior beyond their personal…”

“Social characteristics not only define the relationship between devices but also illustrate the common interest and goals [21], [22].”

“) [1, 3, 37] questioning Manski’s group effect as the driving force explaining the observed phenomenon.”

“However, there are many scenarios whereby the interference between units matters [23, 44, 48].”

“…common exposure unobserved and correlated external events (confounding), and simultaneous influence of connected individuals (herding), or independent and simultaneous decision-making (the reflection problem) (Manski 2000, Hartmann et al. 2008, Shalizi & Thomas 2011, Aral & Walker 2012).”

“In a more formal treatment of the problem, students of social networks have highlighted that homophily, or the presence of similar individual traits, and social contagion are generically confounded in observational studies (Shalizi & Thomas, 2011).”

“As discussed in detail in “Appendix 7,” homophily (i.e., choice of social contacts based on similarity stemming from factors such as race, taste, and personality) seriously confounds studies of peer influence (Shalizi and Thomas 2011).”

“Studying social dependence is an important question in social network analysis (Man-ski 1993; Shalizi and Thomas 2011; Huang et al. 2016; Zhu et al. 2017).”

“This is a common issue for research where manipulations or random treatment are not possible (Shalizi and Thomas 2011).”

“Many studies document that both effects prevail simultaneously and distinguishing one from the other requires strong parametrical assumptions (Aral et al., 2009; La Fond and Neville, 2010; Manski, 1993; Noel and Nyhan, 2011; Shalizi and Thomas, 2010).”

“Equally, tendencies of networks to evolve endogenously, such as reciprocity or clustering/transitivity need to be taken into account (Shalizi and Thomas, 2011; Steglich et al., 2010).”

“In general, homophily is confounded with contagion [ ST11 ] .”

“To study the social contagion, we should notice that the behavior diffusion occurs mainly due to two reasons [23]–[25] of social contagion and homophily.”

“Intuitively, as a result of social influence or homophily [19, 3], the activeness of nodes are correlated with each other.”

“Shalizi and Thomas explore causal inference under the presence of homophily and contagion (Shalizi and Thomas 2011).”

“This type of interference is called interference by homophily (Shalizi and Thomas, 2011).”

“We also hope our results help demonstrate the value of randomized experiments in social network studies (Manski 1993; Shalizi and Thomas 2011).”

“As noted by Shalizi and Thomas (2010), separating social influence from social selection is not possible with cross-sectional data.”

“Abelson 1983) and contagion (Shalizi and Thomas 2011; Harrigan, Achananuparp, and Lim 2012), have been proved to be useful for”

“Distinguishing selection and influence requires strong assumptions on the absence of latent causal factors and on the parametrization of the underlying social process [Shalizi and Thomas (2011)].”

“Methodological problems stem from difficulties in ruling out confounding variables (Shalizi & Thomas, 2011).”

“Debate exists, however, over whether it is possible to demonstrate social learning with observational data without resorting to strong a priori assumptions (Shalizi and Thomas, 2011; Thomas, 2013; Hobaiter et al., 2014).”

“In this paper we propose techniques for quantification on networks that exploit the homophily effect observed in many social networks [16], [23], [22]: people tend mostly to relate with others whom they share some interests, ideas or beliefs.”

“Since observed and unobserved characteristics (i.e., covariates and fixed effects) are typically confounded, the presence of latent homophily significantly complicates identification of the homophily effects associated with the observables (e.g., Shalizi and Thomas, 2011).”

“This definition of interference only considers direct interference, leaving aside indirect mechanisms such as contagion (Ogburn 2018; Shalizi and Thomas 2011).”

“This absence of causal simultaneity has previously been assumed in the literature of causal peer effects (Egami, 2018; Liu & Tchetgen Tchetgen, 2020; McFowland & Shalizi, 2023; Ogburn & VanderWeele, 2014; Shalizi & Thomas, 2011).”

“Under these circumstances, refugee women’s or their partners’ gender-role values are not assigned at random, and effects will be biased (Shalizi and Thomas 2011).”

“(2) There are many challenges in identifying social influence effects, as they are often confounded with other unobserved variables (individual attributes, social-environmental factors, etc), especially homophily in the selection process (Aral et al., 2009; Frank & Xu, 2020; Shalizi & Thomas, 2011).”

“The new influence estimation approach proposed here aims to attribute impact by accounting for causal narrative propagation over the network, while addressing the challenge of discriminating between actual social influence and mere homophily [29, 32].”

“Using directed acyclic graphs, it has been shown that homophily results in a d-connected path between the ego’s prior obesity status (the exposure) and the alter’s subsequent obesity state (the outcome).(39) In other words, the population of alters connected to obese egos may not be exchangeable with the population of…”

“Modeling this more complex longer term process may be rather challenging: Shalizi and Thomas (2011) note teasing out the causality of events arising from homophily and social influence is quite difficult.”

“One reason for this may be the view that social structure is concomitant with quality; i.e., high-quality actors are assumed to be the ones with an early structural advantage (Barabási and Albert 1999, Mouw 2006, Shalizi and Thomas 2011).”

“In principle, the association could be explained by a preference for like-minded friends (homophily), friends influencing each other (diffusion), or other factors influencing both friendship formation and concern for global warming (e.g., Kossinets and Watts 2009; Wimmer and Lewis 2010; Shalizi and Thomas 2011).”

“…to control for latent traits X: Ŷit = b0 + b1Yi,t 1 + b2X̂i + b4 P j AijYj,t 1 P j Aij 3We tested a lagged variable in the data generating process with the same results 4Values from prior studies that use simulation studies to demonstrate bias in social influence studies (Shalizi & Thomas 2011).”

“…unexpected challenges of identifying causal peer effects when social relations (e.g., network ties or contextual exposures) are not randomly formed by outside intervention (e.g., Manski 1993; Sobel 2006; Shalizi and Thomas 2011; VanderWeele and An 2013; Angrist 2014; Ogburn and VanderWeele 2014).”

“Moreover the individual specific î‹ ó There is a clear complication if the ǫ i are allowed to be dependent as differentiating between f ( ) and the error term would not necessarily be possible (Shalizi & Thomas 2011).”

“Principle of homophily and social influence [21, 24] have pointed out that nodes with similar attributes tend to connect to each other in a social network, offering an intuitive way to take attribute information into account for graph clustering.”

“The combined causal effects from homophily and contagion are empirically indistinguishable without additional assumptions (Shalizi and Thomas, 2011).”

“This is a well-known confound of correlated behaviors (e.g., see Aral, Muchnik, and Sundararajan (2009); Shalizi and Thomas (2011); Jackson, Rogers, and Zenou (2017)).”

“…under interference, more recent work has shown how confounding social covariates may cause unidentifiability and biased estimates of causal effects (VanderWeele et al., 2014), especially when the social effects take place on a social network (Ogburn and VanderWeele, 2014; Shalizi and Thomas, 2011).”

“Many voters will not be able to construct congenial networks despite trying, since selectivity is constrained by the composition of the contexts out of which discussant can only be chosen.84 Although not 81 Shalizi and Thomas 2011.”

“Recent theoretical developments highlight the barriers to the identification of causal peer/contagion effects in networks with non-randomized, or observational, data (Lyons 2011; Shalizi and Thomas 2011).”

“, [25]).”

“While each of these pieces has an explicit mechanism of causation (see Shalizi and Thomas [1] for a prior discussion), it is the possibility that the lion’s share of this autocorrelation is due to contagious spread, yielding the concept of “three degrees of influence”.”

“Multiple social phenomena can confound the identification of social influence (see Hartmann et al. 2008; Shalizi and Thomas 2011, and discussion in the “Identification” section).”

“This identification problem is fundamentally rooted in unobserved confounding present in observational data (Hartmann et al. 2008; Phan and Airoldi 2015; Kossinets and Watts 2009; Aral et al. 2009; Shalizi and Thomas 2011; Kwon et al. 2017)6.”

“Hence, just as contagion and homophily may be confounded in observational data [74], it may be the case that our complementarity effects may be confounded with a corresponding phenomena like heterophily (the tendency to interact preferentially with people who exhibit opposing behaviour or traits).”