Why and When "Flawed" Social Network Analyses Still Yield Valid Tests of no Contagion

Abstract

Lyons (2011) offered several critiques of the social network analyses of Christakis and Fowler, including issues of confounding, model inconsistency, and statistical dependence in networks. Here we show that in some settings, social network analyses of the type employed by Christakis and Fowler will still yield valid tests of the null of no social contagion, even though estimates and confidence intervals may not be valid. In particular, we show that if the alter’s state is lagged by an additional period, then under the null of no contagion, the problems of model inconsistency and statistical dependence effectively disappear which allow for testing for contagion. Our results clarify the setting in which even “flawed” social network analyses are still useful for assessing social contagion and social influence.

From social networks to health: Durkheim after the turn of the millennium

“…Unmeasured common cause variables, latent homophily (as well as other network formation processes that can masquerade as homophily, such as differential interaction opportunities (Feld, 1982; Kalmijn & Flap, 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). While some of the debates in this literature have taken an unnecessarily acerbic and “unrelentingly hostile” (Shalizi, 2012) (p.1) tone, fortunately this remains an active and, we hope, collegial area of ongoing research (Christakis & Fowler, 2013; Durlauf & Ioannides, 2010; VanderWeele, Ogburn, & Tchetgen Tchetgen, 2012). In the future, we expect the field will see more innovative and less expensive ways of collecting network-behavior panel data, as well as more sophisticated analyses of data from social media (Centola, 2013; Coviello et al, 2014).…”

Multiplex congruity: Friendship networks and perceived popularity as correlates of adolescent alcohol use

“…This problem becomes magnified when the relationship is mutual (VanderWeele et al, 2012). Although our exposure term represents a summary function of the outcome variable (as opposed to the outcome variable itself) based on outgoing ties (as opposed to mutually nominated ties), we are aware of the potential that this technique generates invalid standard errors for performing meaningful hypothesis testing.…”

Inference for influence over multiple degrees of separation on a social network

“…The original analyses of Christakis and Fowler [3, 4] came under critique by Lyons [5] because, under the alternative hypothesis of social influence, the models employed are in fact incompatible because of the use of contemporaneous ego-alter data so that certain variables appear as an outcome in one regression and as an independent variable in another regression in the same time period. Fortunately, in spite of these problems of model incompatibility, the analyses of Christakis and Fowler [3, 4] will still often constitute valid tests of the null of no social influence [6]. The original analyses of Christakis and Fowler [3, 4] allow one to do testing but not estimation.…”

“…The original analyses of Christakis and Fowler [3, 4] allow one to do testing but not estimation. Estimation would require lagging the alter’s state by an additional period [5, 6] so that e.g. the variables ( Y j,t , Y j,t −1 ) instead of ( Y j,t +1 , Y j,t ) would be included in the regression model (1).…”