Adoption as a social marker: Innovation diffusion with outgroup aversion

ABSTRACT

Social identities are among the key factors driving behavior in complex societies. Signals of social identity are known to influence individual behaviors in the adoption of innovations. Yet the population-level consequences of identity signaling on the diffusion of innovations are largely unknown. Here we use both analytical and agent-based modeling to consider the spread of a beneficial innovation in a structured population in which there exist two groups who are averse to being mistaken for each other. We investigate the dynamics of adoption and consider the role of structural factors such as demographic skew and communication scale on population-level outcomes. We find that outgroup aversion can lead to adoption being delayed or suppressed in one group, and that population-wide underadoption is common. Comparing the two models, we find that differential adoption can arise due to structural constraints on information flow even in the absence of intrinsic between-group differences in adoption rates. Further, we find that patterns of polarization in adoption at both local and global scales depend on the details of demographic organization and the scale of communication. This research has particular relevance to widely beneficial but identity-relevant products and behaviors, such as green technologies, where overall levels of adoption determine the positive benefits that accrue to society at large.

KEYWORDS:

• Agent-based model
• identity signaling
• innovation diffusion
• networks
• polarization
• social identity

Acknowledgments

We thank the members of NIMBioS working group, Evolutionary Approaches to Sustainability, for valuable discussion. For comments on an earlier version of this manuscript, we thank Jeremy Brooks, Fred Feinberg, Cristina Moya, Karthik Panchanathan, and Tim Waring. We are particularly grateful to Tim Waring for leadership in the NIMBioS working group.

Funding

The authors acknowledge the support of the National Institute for Mathematical and Biological Synthesis (NIMBioS), an Institute sponsored by the National Science Foundation through NSF Award #DBI-1300426, with additional support from The University of Tennessee, Knoxville.

Notes

¹ We use the terms “product” and “innovation” interchangeably, as our models apply equally to innovations, non-innovative products, and any other behavior that can be adopted via social influence.

² Bakshi et al. (2013) refer to these as “segments” of the population.

³ We provide, as a supplement, a Mathematica notebook for reproducing and altering the plots shown in Fig. 1 to enable a more thorough exploration of the analytical model by the curious reader.

⁴ Even when mathematical approximations are possible, they are imprecise and can miss important results only available to simulation approaches, as in de Aguiar, Rauch, and Bar-Yam (2004).

⁵ The line is the simplest organizational framework with which to study structured interactions. It also allowed for an easily implementable and interpretable spatial correlation between neighboring patches, which would be more complicated in network structures in which patches could share more than one neighbor in common, such as a square lattice or small world network. That said, our model is easily extendable to those and other network structures.

⁶ In assuming the spatial correlation between patches, we have in mind the smooth transitions that one experiences in moving from one media market to another and not the sharp distinctions in identities that one experiences when crossing the tracks or suburban boundaries, as in a Tiebout (1956) world.

⁷ In reality, even long-range observations will likely be skewed by homophily. Our formalization allows us to control the rate of homophilic interactions more precisely.

⁸ Product information flow in our model involves only direct social communication, and for simplicity excludes mass media influences. A modification in which agents received additional information from third-party sources, perhaps represented by “media agents,” could be added for future analyses.

⁹ Mathematically, it is of course possible for to be greater than one, further decreasing the probability of adoption when the product is rare. We restrict our analysis to the range (0, 1), because diffusion tends to fail for values of near or greater than one.

¹⁰ Model code is available at https://www.openabm.org/model/5237/version/1

Additional information

Funding

The authors acknowledge the support of the National Institute for Mathematical and Biological Synthesis (NIMBioS), an Institute sponsored by the National Science Foundation through NSF Award #DBI-1300426, with additional support from The University of Tennessee, Knoxville.