Network structures, environmental technology and contagion

ABSTRACT

Diffusion of low-carbon technologies is critical to global decarbonization efforts aimed at achieving the Paris Agreement climate goals. In this paper, we represent a social system as a network of agents and model the process of technology diffusion as a contagion propagating in such a network. By setting the necessary conditions for an agent to switch (i.e. to adopt the technology), we address the question of how to maximize the contagion of a technology subject to learning effects (e.g. solar PV) in a network of agents. We focus the analysis on the influence of the network structure and technological learning on diffusion. Our numerical results show that clusters of agents are critical in the process of technology contagion although they generate high levels of variance in aggregate diffusion. Whatever the network structure, learning effects ease the technology contagion in social networks.

Key policy insights:

• Policy makers should take advantage of, or favour, clustered organizations to deploy renewables (e.g. cooperatives of farmers, online social networks)

• Social dimensions are critical in the design of environmental policies (e.g. peer effects)

• Technologies subject to learning effects spread further in the population

• When considering social connections, policy makers should design policies to cope with uncertainty in diffusion (clustered organizations exhibit high levels of uncertainty in adoption).

KEYWORDS:

• Networks
• complex contagion
• technology
• learning effects
• cascades

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 In this paper, we call ‘cascade’ the dynamics of diffusion.

2 See Farmer and Lafond (2016).

3 Remember that lattice networks exhibit high levels of clustering and (comparatively) very long path length; small-world structures demonstrate high level of clustering but with lower average path length; random networks are subject to low clustering and low average path length.

4 We relegate extreme scenarios $\alpha =\{0;1\}$ to the Appendix, Section A.1.

5 Although social networks are sparse, meaning they exhibit fewer links than the possible maximum number of links within that network, such framework is common in the literature on complex social networks (Cowan & Jonard, 2004; Snellman et al., 2019; Zhaoyang et al., 2018).

6 cf. Section 3.1 for description.

7 cf. Appendix, Section A.1 for $\alpha =\{0;1\}$.

8 cf. Appendix, Section A.2.

9 cf. Appendix, Section A.5.

10 cf. Appendix, Section A.4 for other scenarios.

11 Note: here we report associated seeds above which we observe no more positive values: when $\alpha =0.1$, negative values arise when $S_0=25$ (lattice), $S_0=29$ (random); when $\alpha =0.3$, negative values appear when $S_0=14$ (lattice), $S_0=25$ (random); when $\alpha =0.5$, negative values arise when $S_0=10$ (lattice), $S_0=15$ (random); when $\alpha =0.7$, negative values arise when $S_0=7$ (lattice), $S_0=14$ (random).