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”

“The temporal influence [11] is learned from direct causal interactions in cascades, which finely describes the temporal correlations of diffusion behaviors among users.”

“In this paper, we consider two widely acknowledged dependencies: temporal contagion and social homophily [28, 29].”

“For instance, Goldsmith-Pinkham and Imbens (2013) assumed a linear-in-means model and estimated peer effects under network endo-geneity (i.e., confounding by network, Shalizi and Thomas, 2011) using parametric Bayesian methods.”

“This approach however gave us experimental control which addressed confounding issues such as possible differential exposure to distorted content in depressed individuals’ social media timelines (52, 53).”

“…social signals, such as the number of reposts (Al-saedi, Burnap, and Rana 2016), topic tags (Dutta et al. 2018), the quantity of followers (Liu et al. 2012), Reddit’s ”karma” scores (Kano et al. 2018), and user influence (Li et al. 2015), along with social relationships (Shalizi and Thomas 2010).”

“There are often multiple forces driving associations that are unmeasured in these studies, and such latent homophily may bias inferences about the in fl uence of racial attitudes on friendship formation (Shalizi & Thomas, 2011).”

“While we have attempted to control for shared environment confounding through the use of spatial covariates (representing the assigned district), in general network contagion estimated through regression models is confounded by environment and/or homophily [24].”

“The presence of homophily contributes to network formation, and hence, indirectly to the spread of behavior through these newly formed network ties [14].”

“Although the underlying mechanisms are complex, 9 behavioral similarities may be due to shared environments ( i.e. , common causes), to the inclination to have friends with common interests ( i.e. , homophily), or to the influence that those in our PSN can have on our lifestyle behaviors ( i.e. ,…”

“…the observed units (Lyons, 2010), and possibly biased effect estimates that could have resulted from: model misspecification (Lyons, 2010; VanderWeele and An, 2013), network-induced homophily (Shalizi and Thomas, 2011), and unmeasured confounding by environmental factors (Shalizi and Thomas, 2011).”

“Aral, Muchnik, and Sundararajan (2009), Aral and Walker (2011), and Shalizi and Thomas (2011) discuss the implications of homophily in detail. sampling of network nodes.”

“In detail, when someone receives information from others who hold similar opinions with him/her, he/she will be more likely to perform the dissemination behavior [10–12].”

“, 2009], [Aral, 2010], [Aral and Walker, 2011], [Shalizi and Thomas, 2011], and [Bapna and Umyarov, 2014]).”

“1In [2] 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.”

“Attributes are essential as they can both influence and be influenced by the structure of links [24].”

“These phenomena pose significant challenges to the identification of social influence in observational data [5].”

“…network data, Abrutyn and Mueller (2014a) argue that suicidal behavior spreads between friends, but the analysis is likely affected by unmeasured confounding as well as network-specific problems like homophily bias and partial censoring of the peer group (Shalizi and Thomas 2011; Griffith 2022).”

“This similarity can occur through a variety of forces, and (especially with cross-sectional data), it can be very difficult to determine which is driving the autocorrelation (Shalizi and Thomas 2011).”

“In network theory, we observe homophily when nodes tend to connect preferentially with nodes with similar characteristics (New-man 2002; Kossinets and Watts 2009; Shalizi and Thomas 2011; Hasheminezhad and Brandes 2023).”

“The central problem, as Shalizi & Thomas (2011) brilliantly expose, is that homophily (the tendency to form ties with similar others) and contagion (influence or diffusion through those ties) are “generically confounded” in observational studies.”

“Regarding voting behavior and related contagious social outcomes, different layers contain predictive information due to homophily, contagion, and shared environments [43].”

“However, the extent to which these relationships drive political change remains uncertain, as most data struggle to disentangle social influence from selection effects (Bello and Rolfe 2014; Shalizi and Thomas 2011).”

“One key methodological challenge in identifying behavioral diffusion is accounting for self-selection (Manski 1993; Shalizi and Thomas 2011), whereby individuals with similar tendencies naturally form geographic clusters: residential self-selection.”

“Shalizi and Thomas (2011) discussed that it is nearly impossible to reliably estimate causal peer effects in observational studies using direct confounding adjustment methods (e.g., regression-based adjustment) due to pervasive concerns about unobserved network confounding.”

“By adding the corresponding nodes information from X ′ and Y ′ to the inherent relation learning Pfeiffer III et al. (2014); Shalizi & Thomas (2011), the revised generation model is In general, SGDD concurrently optimizes the parameters X ′ and A ′ .”

“…evidence regarding behavioral associations between members of a network has resulted in a growing methodological literature (e.g., Fowler et al. 2011; Ogburn et al. 2014, ), with some suggesting that it is nearly impossible to isolate peer effects in observational data (Shalizi and Thomas 2011).”

“In the full interference setting, Tchetgen Tchetgen et al. (2021) propose a chain graph representation for interference along with an estimation strategy, known as auto-g-computation , for”

“In fact, with cross-sectional data, they are confounded and cannot be separated (Shalizi and Thomas, 2011).”

“Homophily and social influence are generally confounded with each other, and their causal effects are best distinguished through controlled experiments (Shalizi & Thomas, 2011).”

“An individual ’ s obesity or weight status becoming more similar to their social ties over time suggests an interesting phenomenon of “ obesity contagion ” (Cohen-Cole & Fletcher, 2008; Shalizi & Thomas, 2011; Smith et al., 2020).”

“Distinguishing homophily from contagion is generally difficult (Shalizi and Thomas 2011).”

“9 As a further test of robustness, and to specifically investigate endogeneity (or selection or homo-phily; Shalizi and Thomas 2011), we estimated social network models of networks and protected lands (Snijders 2001).”

“To further investigate the causal relationship, Shalizi & Thomas (2011) provided three factors underlying such interaction: homophily, or the formation of social ties due to matching individual traits; social contagion; the causal effect of an individual’s covariates on his or her measurable…”

“Some social science research has shown that the correlation of data can well reflect the interaction between individual attributes [1].”

“In particular, we follow Shalizi and Thomas [61] (translated to the neighborhood context) and argue that the likelihood of following previous out-movers is confounded by factors that account for the high degree of similarity between neighbors within a particular area, which can also cause moving-out behavior.”

“In fact, their differences are often very vague and undetectable in empirical social network studies [19].”

“The model does not assume influence effects outside households, but this is a very open controversy in science (Aral et al, 2009; Shalizi and Thomas, 2011).”

“Social comparison theory and social influence network theory suggest that entity similarity also has an impact on behavior spreading. found that the behaviors or opinions of social network users were similar simply because their characteristics were similar [46, 47].”

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

“Real social networks present three potential confounders to social influence [49, 50].”

“It has also been recognized that other external factors, such as geographic proximity, may explain similarity in purchase patterns among friends on social medial [19], [20].”

“Community membership: Community membership is another candidate proxy for latent homophily [8].”

“Shalizi and Thomas [32] show that homophily and social influence are generally confounded with each other; thus, distinguishing between them requires strong parametric assumptions.”

“Users in homophilic relationships will share common characteristics [15].”

“Homophily is thus consistent with the view that triangle closing is fundamental (Shalizi and Thomas, 2011).”

“One is the issue of the separation of causal effects from selection effects regarding similarity between subjects and their significant others (Yamaguchi and Kandel 1997; Shalizi and Thomas 2011).”

“Drumwright defined the term Corporate Societal Marketing as “encompass marketing initiatives that have at least one non-economic objective related to social welfare and use the resources of the company and/or one of its partners” [26].”

“Shalizi and Thomas (2011) have shown that when there is an unobserved trait that co-determines both behavior and network choice, social influence effects are generally unidentifiable as social influence and homophily (selection) are generically confounded through this unobserved trait.”

“, 2017), future work will likely need to confront the equivalent problem–separating non-random network clustering from non-random network diffusion (Shalizi and Thomas, 2011).”

“Multiple previous works has studied causal inference in the presence of social network [20, 24, 25, 66, 4, 57, 41].”

“Importantly, under our assumption that the relevant covariates are observed, homophily does not by itself invalidate identification: strong ignorability conditional on the local covariate environment remains sufficient (Shalizi & Thomas, 2011).”

“The comparison concerns pathways through reached derivatives, but observed agreement or ancestry alone cannot establish population-wide influence [Shalizi and Thomas 2011].”

“Shared exposure is the classic alternative to influence in network data (Aral et al., 2009; Shalizi and Thomas, 2011); Section 5.6 tests its most direct form, contact with the attacker.”

“Correlated behaviour alone does not identify a peer effect [14, 15], and an agent’s own explanation is not a complete account of its decision [21].”

“The imitation model among the traders is not causally identified [2,25].”

“The homophily principle states that nodes with similar attributes tend to connect to each other, which provides an intuitive theoretical basis for exploiting attribute information [9].”

“Individuals experiencing similar concerns may also be more likely to interact, making interpersonal influence difficult to distinguish from pre-existing similarities and shared circum-stances (Shalizi and Thomas, 2011).”

“Similar work demonstrates that high-dimensional adjustment can reduce bias relative to experimental benchmarks [24], [25].”