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”

“In observational studies of friends, one of the fundamental problem is that the friendship bond is endogenous to the characteristics of the individuals in the friendship, making it challenging to disentangle the effects of friends from effects of characteristics contributing to friendship formation (63).”

“…of the treatments assigned to other units that are related to unit i through the network; and network-correlated outcomes (McPherson et al., 2001; Shalizi & Thomas, 2011; Manski, 2013), where the network informs the correlation among the potential outcomes because the potential outcomes of unit…”

“Hence, the network autocorrelation models only examine similarity between respondents’ and community members’ logics given their prior interactions but do not fully address respondents’ possible selfselection into communities (i.e., Shalizi & Thomas, 2011).”

“While status-based homophily, also referred to as network selection , may certainly shape the observed patterns of similarity between connected individuals, a distinct mechanism, network influence (or ‘ contagion ’ ), can also cause this observed similarity [201].”

“Without longitudinal data, the network autocorrelation models can only identify an association between peer self-efficacy and teacher self-efficacy and is therefore incapable of ruling out social selection, rather than peer influence, as the reason behind the association (Shalizi & Thomas, 2010).”

“In the context of causal inference, methods for dealing with data dependence are developed under the heading of interference [32, 9, 27, 10, 36, 29, 20, 35].”

“In reality, all these effects are interconnected [9, 3] and are difficult to disentangle from observational data [35].”

“It has been recently shown that these mechanisms are generically confounded [11], but it is important to remember that this does not make observational data worthless.”

“One lesson here is that policies aimed at encouraging or discouraging the spread of behaviors in networks must be backed by rigorous empirical studies that convincingly separate homophily effects from effects due to social contagion (Aral et al., 2009; Shalizi and Thomas, 2011).”

“Although we refer to this independent variable as homophily (see Methods), the effect of this variable is in fact interpreted as either homophily or contagion [20,21].”

“However the relative effect of seeing one’s friends is only partly measured through our observational study [32].”

“Distinguishing these two phenomena in an observational study without additional strong assumptions is not possible [51].”

“A major challenge in this area is that homophily—the tendency of similar people to cluster in a network—is generally confounded with contagion—the influence people have on their neighbors [ST11].”

“A COMPUTATIONAL TEST im ite d Gianluca Manzo L (C ) E me ral d G rou p P ub lis hin g”

“Likewise, any research involving observational data, such as ours, will have difficulty distinguishing social contagion from homophily [16, 38].”

“contraceptives users [26,27]).”

“Besides, the conventional graph convolutional network encoders [2], [152] are based on the homophily assumption [153], i.e., the connected nodes contain similar semantics.”

“are either flawed (Cohen-Cole and Fletcher, 2008; Lyons, 2011; Shalizi and Thomas, 2011) or limited by strong assumptions and burdensome data requirements.”

“Thus, even when issues of simultaneity (including “reflection”, Manski, 1993) can be avoided, it is thus generally not possible to identify peer effects using observational data without the often suspect assumption that adjusting for available covariates is sufficient to make peer behavior unconfounded (i.e., as-if ra…”

“…differ-ence between the well known problem of homophily (the phenomenon by which individuals with similar traits are more likely to form social ties with one another) and peer influence (Cohen-Cole and Fletcher, 2008; Lyons, 2011; Noel and Nyhan, 2011; Shalizi and Thomas, 2011; VanderWeele, 2011).”

“…is not feasible without assumptions on the interference mechanism has been suggested in the literature for observational studies (e.g., see Shalizi and Thomas, 2011), and also in the context of randomized experiments; for instance Aronow and Samii (2013) state that under arbitrary…”

“We cannot rule out that emotional well-being homophily was the result of shared social foci that individuals with certain levels of emotional well-being engage in (i.e., secondary homophily, Shalizi & Thomas, 2011).”

“To avoid overlooking of the inherent relation [30, 71] between A′ and the corresponding node information (i.”

“…2001)), multiple peer effects, and the boundary specification problem all pose significant problems for researchers attempting to understand the phenomena of contagion and social influence (Berndt & Keefe, 1995; Laumann, Marsden, & Prensky, 1983; Manski, 1993; Shalizi & Thomas, 2011; Thomas, 2013).”

“This could be because of a “contagious” effect, where people imitate the behavior of their peers, or a feeling of homophily, in which some unobserved traits drive people’s tendency to form a friendship and to exhibit a similar behavior (e.g., Aral et al. 2009, Shalizi and Thomas 2011).”

“Because each of these phenomena may lead to correlations between the phenotypes of connected individuals (Christakis and Fowler, 2007; Shalizi and Thomas, 2011), methods to parse these associations apart are required.”

“Because friendship group position represents a snapshot of a continually evolving and highly endogenous social network, proving causality would be a tall order (Shalizi & Thomas, 2011).”

“Similarly, Shalizi and Thomas [32] famously proved that contagion (causal inheritance of traits due to peer influence) cannot be distinguished from homophily given a single network observation.”

“, [2, 53, 57, 69, 70]) have shown increased interest in distinguishing the effects of social influence from those”

“This formulation means that the homophily mechanisms are assumed to be dependent only on the process Y.t/ and any measured covariates; latent homophily is not considered here (Shalizi and Thomas, 2011).”

“While much of the early statistical work on networks focussed on models for understanding network formation [8, 10], there has been a recent surge in drawing causal inference from experiments on networks [6, 19, 20, 21, 22].”

“Age, median (IQR), y 22 (11-45) 21 (11-45)”

“der linear outcome models. Recent works in causal inference show that confounding social covariates lead to unidentiability and biased estimates of causal e ects ( 63 ), especially on social networks (64,65), and how longitudinal studies (66,67) and design of experiment for specic peer e ects (68) provide a way forwar…”

“These works address applications such as the study of infectious diseases [61, 14] or behavior and interactions in social networks [55, 30, 57, 11, 61, 24, 18].”

“A careful, self-contained characterization of the challenges inherent in this problem can be found in Shalizi and Thomas [177].”

“Also, in social networks, features and links are not independently presented but are strongly correlated, and the root cause of the correlations can be explained by some social science theories such as social influence and the principle of homophily [7, 30].”

“People are connected, and so their health is connected (Smith and Christakis, 2008).”

“Another advantage of in vitro studies is the possibility to differentiate social influence from other confounding factors such as homophily thanks to the anonymity of influencing individuals (see also [63, 64]).”

“In a large graph with few clusters, nodes in each cluster may generate different response than other clusters, obfuscate GATE estimation if we assign treatment/control at cluster level: the difference in the responses of different clusters might be unrelated to the treatment effect, but instead due to endogenous node…”

“However, ascertaining the direction of a causal relationship between network structure and link or nodal attributes, i.e., whether it pertains to homophily or contagion, is difficult (e.g., see Doreian, 2001 and Shalizi and Thomas, 2011 and references therein).”

“Recent theoretical research in statistics and elsewhere has noted the difficulty – if not impossibility – of distinguishing between these two mechanisms, even with experimental data (Aral et al., 2009; Aral & Walker, 2011; Shalizi & Thomas, 2011).”

“There has been much discussion of the challenge that confounding due to homophily poses for learning about peer effects (Shalizi and Thomas 2011).”

“…dynamics relate to health, above and beyond baseline network features, focuses on how dynamic social processes can exert unique influences on individuals (Cerhan & Wallace, 1997; Eng et al., 2002; Friedman & Aral, 2001; Giordano & Lindstrom, 2010; Seeman et al., 2011; Shalizi & Thomas, 2011).”

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

“However, the occurrence of additional pledges by contacts of the original donors is hard to interpret as evidence of the existence of a network effect when relying on observational data (Shalizi and Thomas 2011).”

“Specifically, it is difficult to distinguish between homophily and contagion by simply observing the network properties and without strong assumptions about the social process [4, 26].”

“Diffusionpredictionreliesheavilyontemporal influence[20],which describes the dynamic changes in user transitions and cascade content with the evolution of diffusion.”

“The effect of the level two variables can be due to homophily mechanisms that, despite the variables included in our model, can create latency (Shalizi and Thomas, 2011).”

“Several studies on online social networks provide valuable insights into social influence, information spread, behavioral diffusion, and the structural aspects (who has ties to whom) [16-20].”

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

“As noted by Lyons (2011) and by Shalizi and Thomas (2011), this is, of course, a very strong assumption.”

“Much of the research on walkability and wellbeing in both the health and transportationplanning disciplines has relied on more classical transportation approaches that examine the correlation between environmental and contextual characteristics (often of the home neighborhood) to the health and wellbeing of residents…”

“If the characteristics are unmeasured, their effects cannot be blocked and so will be indistinguishable from social influence (peer effects) [62].”

“Health outcomes for chronic conditions are modulated by health behavior, which in turn might be expected to show covariation sensitive to underlying social network structure (Centola, 2010; Shalizi and Thomas, 2011; Christakis and Fowler, 2013).”

“However, we are not able to make causal inferences regarding the transmission of health behaviours between peers in this study as this causal relationship is not identifiable in the absence of full information concerning both the choice of friends and the proclivity to engage in a given health behaviour [20-22].”

“Causality has been studied in many fields, such as biology [1-2], medicine [3-7], economics [8-12], epidemiology [13-15], and sociology [16-20].”

“Many different processes may produce similarity between peers (McPherson et al., 2001; Shalizi & Thomas, 2011), but two in particular are often at the forefront of understanding peer delinquency: influence and selection.”

“A further drawback is that—in many cases—user feedback can be conceived as a combination of social and semantic signals whose separation is hard, and they can be easily misclassified [26].”

“Although contagion and homophily are often presented as confounding phenomena, both may occur simultaneously in real-world networks(50) and may vary by study context.”

“we found an increased likelihood of connection between authors with the same role or health condition, it is impossible to differentiate homophily effects from contagion effects in observational data [95]. Future experimental work would enable exploration of these differing causes. In examining the full network, we al…”

“However, few studies examine two-mode network event sequences (De Nooy, 2011; Stadtfeld and Geyer-Schulz, 2011; Quintane et al., 2014; Malang, Brandenberger and Leifeld, 2017).”

“ShaliziandThomas[29]provedakeynegativeresult—homophily and influence are fundamentally confounded in observational studies.”