Spatial clustering in vaccination hesitancy: The role of social influence and social selection

Abstract

The phenomenon of vaccine hesitancy behavior has gained ground over the last three decades, jeopardizing the maintenance of herd immunity. This behavior tends to cluster spatially, creating pockets of unprotected sub-populations that can be hotspots for outbreak emergence. What remains less understood are the social mechanisms that can give rise to spatial clustering in vaccination behavior, particularly at the landscape scale. We focus on the presence of spatial clustering, and aim to mechanistically understand how different social processes can give rise to this phenomenon. In particular, we propose two hypotheses to explain the presence of spatial clustering: (i) social selection, in which vaccine-hesitant individuals share socio-demographic traits, and clustering of these traits generates spatial clustering in vaccine hesitancy; and (ii) social influence, in which hesitant behavior is contagious and spreads through neighboring societies, leading to hesitant clusters. Adopting a theoretical spatial network approach, we explore the role of these two processes in generating patterns of spatial clustering in vaccination behaviors under a range of spatial structures. We find that both processes are independently capable of generating spatial clustering, and the more spatially structured the social dynamics in a society are, the higher spatial clustering in vaccine-hesitant behavior it realizes. Together, we demonstrate that these processes result in unique spatial configurations of hesitant clusters, and we validate our models of both processes with fine-grain empirical data on vaccine hesitancy, social determinants, and social connectivity in the US. Finally, we propose, and evaluate the effectiveness of two novel intervention strategies to diminish hesitant behavior. Our generative modeling approach informed by unique empirical data provides insights on the role of complex social processes in driving spatial heterogeneity in vaccine hesitancy.

Fig 1

(a) A schematic of the landscape-level spatial network in the United States. Counties, which contain populations of individuals, are represented by nodes, and interactions between the county populations determine the edges of the network. (b) A schematic of the change in vulnerability due to the social influence and social selection processes. Under social influence (top arrow), each community (node) starts with a level of hesitancy (denoted in blue in the top network), and hesitancy behavior spreads through the network across social interactions (edges) between communities from hesitant (dark blue) communities to non-hesitant ones (light blue). Under social selection (bottom arrow), communities with similar traits (denoted in pink in the lower left network) tend to be connected, and some traits independently lead to hesitancy behavior (denoted in blue in the bottom left network). Each process can give rise to the same distribution of vulnerability to outbreaks (dark orange = vulnerable community, light orange = protected community in the right network).

Fig 2. Spatial clustering generated by social selection (left) and social influence (right).

For each process we show the spatial clustering in vaccine hesitancy as a function of the network structures, from highly spatial (ring) to aspatial (random) networks. Each curve corresponds to a different intensity of the given social process, from high (top, darker shades) to low (bottom, lighter shades).

Fig 3

(a) An affectable society with α = 0.2, which has a low social influence parameter and (b) A determined society with α = 0.6 which as a high social influence parameter. Each curve corresponds to a different level of initial social selection; from homophilous (dark) to random (light) networks.

Fig 4

(a) We show estimated social selection β, social influence α, the spatial clustering values based on observed hesitancy data, and that from stochastic simulations of our theoretical model. b) Pink colored counties ($●$) are vulnerable (high hesitancy) and green colored counties ($●$) are protected (low hesitancy). We illustrate the observed data for 2015, alongside the observed and social influence modeled estimates for 2018. The star marks the county of San Diego, California immediately to its north.

Fig 5

Reduction in spatial clustering via interventions, relative to no intervention: (a) We consider the impact on relative spatial clustering under the social selection strategy, which targets nodes to intervene on (target randomly, high hesitancy, high social traits) for societies with increasing levels of social selection (b) We consider the impact on relative spatial clustering under the social influence strategy, which reroutes edges with different probabilities of edge rerouting in societies of increasing levels of social influence.