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”

“This inclination implies that information rooted in distinct professional disciplines tends to naturally group into network homophily (Shalizi & Thomas, 2011; A. Yang & Saffer, 2019).”

“Unfortunately, determining whether homophily or contagion is responsible for similar behavior among friends is notoriously difficult (and often impossible [51]).”

“Additionally, existing theories of social science, like homophily and social influence [21, 55], do not apply directly to signed networks.”

“…is notoriously difficult, due to intrinsic confounding of contagion — a network effect marketers and policy-makers seek to leverage — and homophily (Shalizi and Thomas 2011), as well as endogenous group formation, correlated unobservables, and simultaneity (Hartmann et al. 2008), the last often…”

“based diffusions are driven by influence of neighbors whereas homophily-based diffusions are driven by properties of nodes (which are correlated among neighbors) [3, 62, 83].”

“There are methodological issues when examining influence processes within such contexts— differentiating between selection or influence is generally difficult—limiting the types of inferences one can make (Shalizi & Thomas, 2012; VanderWeele, 2011).”

“(That is why we used the time-delayed model in Shalizi and Thomas (2011).)”

“This is not unique to offline settings; a study of social influence in Wikipedia shows that users become rapidly similar after their first online communications, and this keeps increasing over time, making it difficult to tease social influence and homophily apart [17].”

“However, the correlation between feature and graph information is well known in literature [56, 57], overlooking the impact will lead to bad performance.”

“The setting is informed by standard assumptions made to identify the causal effect of social influence under the presence of homophily in a pairwise setup, when examining the causes behind why a person manifested a behavior at a given time [50, 77].”

“Second, in using egocentric network data, we do not address some issues specific to causal inference in network analysis, such as homophily and social contagion in sociocentric network data (Shalizi and Thomas 2011), the emphasis of much research by Christakis and Fowler (2007, 2008, 2013).”

“In the presence of confounding effects such as homophily and shared local or global influences, contagion can be hard to identify [63].”

“Rather, secondary forms of homophily can reinforce network homogeneity if different forms of homophily are intertwined (Shalizi and Thomas 2011).”

“In fact, homophily effect and social influence are frequently confounded and difficult to separate in non-experimental studies (Shalizi and Thomas 2011).”

“However, PSM still fails to provide full causal interpretations because it does not control for unobserved effects [29].”

“This relates to the dilemma of discerning homophily from influence [20]; better hitters might already have coalesced onto certain teams, and the contagious hitting is due to this structural grouping as opposed to a diffusive effect.”

“Since a community is a set of users who share similar topics [29], users’ interests in multiple topics result in multiple topic distributions in communities.”

“Disentangling homophily from social contagion is difficult because similar people select into shared networks and encounter common environments and shocks, which align adoption times and produce clustering that is observationally equivalent to peer influence 8,9,44,45 .”

“Determining the balance of these two forces toward similarity is difficult because the behaviors are generally endogenous (Shalizi and Thomas 2011), with each causing the other.”

“The possible confounding of peer effects with social homophily has been noted in the literature (Shalizi & Thomas, 2011).”

“We then examined the social network theories of sociality (some people are more social than others, Table 1) (17), health-behavior homophily, and social contagion that could explain the increase in network participants who discussed living donation with other patients and requested a living donation from family and fr…”

“As long as all of the variables affecting the behavioral outcome or both the behavior and selection are observed and controlled using this relatively simple model, contagion effects can be correctly estimated [16,20].”

“As shown in Shalizi and Thomas (2011), W is always, often implicitly, adjusted for in any causal diffusion analysis because researchers need to compare observations with similar spatial/network pre-treatment characteristics.”

“Thus, it would be particularly difficult to determine and infer from observational data how personal moods spread or diffuse.(12) Even without claiming the causal effects that are essential for explaining spread or diffusion, however, it would be revealing to examine whether and how similar patterns of concurrent mood…”

“This allows us to gain better understanding of human behaviour related to diffusion and contrasts with studies driven by observational data, in which confounding effects can always be present [36].”

“That is a much harder task [4, 61] because the benefit of randomization is lost, as the set of individuals who received the treatment is often pre-determined.”

“Noting that it is difficult to distinguish selection and causal effects [20] clustering seems to occur together with homophily.”

“Although influence is the predominate focus of health studies, inferences of network influence are frequently overstated (Shalizi & Thomas, 2011).”

“Because these possibilities are well studied and occur routinely in network analysis, not just mediation analysis (An 2015a; An et al. 2022; Centola 2010; Shalizi and Thomas 2011; Steglich et al. 2010), I do not belabor the point here.”

“To this end, observational techniques [16], [23] and randomized controlled trial (RCT) methods [24], [25] have been proposed in the literature.”

“Two mechanisms, identi fi ed in earlier research albeit using different terminology (see Shalizi and Thomas 2011) are suggested to be driving this pattern.”

“In order to deal with this situation, as Shalizi and Thomas (2011) and VanderWeele (2011) show, it is possible to use a sensitivity testing approach to assess the robustness 7 The term 'spatial' is applied to all variables which have been weighted by W, regardless of whether this is a contiguity…”

“Thus, a causal effect of peer influence should be inferred with caution, although the SAOM has been used to estimate peer influence and friend selection simultaneously (Shalizi & Thomas, 2011).”

“Those with a higher weight often experience more pressure to be slim and have been identified as a unique cluster, based on homophily and contagion assumptions, in social network studies of obesity (79).”

“However, Shalizi and Thomas (2011) caution that is usually difficult to observe all confounders; unobserved factors that drive connections and purchases can still bias our estimates of social influence.”

“Due to group homophily in networks [8], individuals in a community share similar topic interests, and they generate intracommunity links.”

“It is well known that empirical identification on whether the behavior of friends affects one’s behavior is hard to obtain with just observational data because homophily and reverse causality hinder our ability to do so [41, 54, 24].”

“, assortative mixing [39] or “homophily”), unobserved influence, and transitivity of influence [46].”

“The Homophily Confound: Shalizi and Thomas (2011) identify three factors that affect the activities in a social network: ‘Homophily’, ‘Social Influence’ and ‘Co-Variate Causation’.”

“We illustrate this approach on case studies of two Apache Software Foundation (ASF) projects: Ant and Axis2/Java, on which we model three individual behaviors: one’s commit count, high file ownership, and minor contributor count (or low file ownership), all over time.1 We found that: 1.”

“…this puzzle amounts to separating the effect of social influence (i.e. when network influences attitude/behavior) from homophilous interactions (i.e. when attitude/behavior influences network formation) and context effects (Borgatti and Foster 2003; McPherson et al. 2001; Shalizi and Thomas 2011).”

“…significant rho in the neighborhood effects literature – in health behaviors is small (e.g. Christakis and Fowler, 2007; Eisenberg and Quinn, 2006) and largely debated (e.g. Lyons, 2011; Cohen-Cole and Fletcher, 2008a, 2008b; Noel and Nyhan, 2011; Sainsbury, 2008; Shalizi and Thomas, 2011).”

“More recently, a debate has flared up on the identifiability of influence effects in networks (Shalizi and Thomas, 2011; VanderWeele, 2011), triggered in part by the remarkable findings of Christakis and Fowler (2007) and Fowler and Christakis (2008) on the contagion of obesity, happiness, and other traits.”

“Whereas we used a straightforward behavioral genetic design, they performed social network research, critiques of which are now emerging in the literature (Lyons, 2011; Shalizi & Thomas, 2011).”

“Despite their central importance to social science, causal peer effects are notoriously difficult to estimate reliably in empirical data (Aral 2011; Manski 2000; Shalizi and Thomas 2011).”

“Furthermore, regression analysis confounds homophily-based diffusion with social-influence based diffusion [6,92].”

“In addition to primary selection effects (friendships develop among students with similar perceptions of teaching quality), so-called secondary selection effects may occur as students ’ homophily is not one-dimensional and also operates based on other attributes (see also e.g., [7,36]).”

“Homophily must also be considered because it is often conflated with, and empirically very difficult to separate from, contagion [51, 52].”

“Specifically, “homophily,” which is the tendency of individual actors with similar attributes to form ties with one another (Kossinets & Watts, 2009; Shalizi & Thomas, 2011), has been regarded as an important mechanism of network formation.”

“It is thought to be an important aspect of many social behaviours and phenomena, including group in fl uence, contagion, and information di ff usion (Aral et al ., 2009; Bakshy et al ., 2012; Shalizi & Thomas, 2011).”

“The use of a potential outcomes framework provides an approach to deal with well-understood biases, such as confounding and selection bias, and address complex dependencies in outcomes due to the network structure [56–58], as well as biases specific to network designs, including homophily [60] and network dependence […”

“To this extent, we introduce into the diffusion framework, the fact that not only network structure matters, but also the way in which agents influence each other [33].”

“However, Spirtes (1995); Neal (2000) show that without any specific assumption regarding the nature of dependence (i.”