Reply to “Comment on Geodesic Cycle Length Distributions in Delusional and Other Social Networks”

Abstract

Martin (2020) describes a misinterpretation of exponential random graph (ERGM) parameters in my contribution (Stivala 2020), with the use of this parametric model obscuring, rather than illuminating, the data. He suggests that this is symptomatic of a trend in the social networks community towards a methodological monoculture focussed on the use of ERGMs. In this Reply I try to clarify how this situation arose in this specific case, and address some more general issues Martin raises, including the use of nodal covariates, what we can learn from ERGMs, and methodological monoculturalism in social network research.

Geodesic Cycle Length Distributions in Delusional and Other Social Networks

“…An advantage of the 𝑑𝑘-series approach is that we do not need to contend with the potential difficulties in estimating ERGM parameters, where for some networks, such as most of the fictional character networks considered here, it can be difficult or perhaps even impossible (without adding a lot of extra constraints, such as fixing the presence or absence of particular ties) to find a converged model. Further discussion of the relative merits and shortcomings of these approaches can be found in Martin (2017); Stivala (2020a); Martin (2020); Stivala (2020b).…”

“…Commentary on Stivala (2020a) mostly concerned the [mis-]interpretation of ERGM parameters and the trend (or otherwise) in the social networks field to be moving towards a monoculture of using parametric models, and in particular ERGMs, for everything (Martin, 2020;Stivala, 2020b). What has not yet been addressed, is to test the hypothesis suggested in Stivala (2020a), that fictional character networks (see Labatut and Bost (2019) for an overview), and specifically those with only a single author (unlike the one fictional character network examined in that work), might have anomalous geodesic cycle length distributions, like the "Patricia" networks.…”