The increasing violence perpetrated by far-right extremist groups in the United States has raised significant public concern regarding the spread of radicalization. Unlike previous assumptions that psychopathology is the primary driver, research points to a process involving destabilization from environmental factors, exposure to extremist ideology, and community reinforcement. This suggests a social contagion model, where extremist ideologies function as complex contagions, necessitating repeated exposure for adoption. This study aims to determine if far-right radicalization patterns align with this contagion process and to assess the influence of key endemic factors. The alarming rise in far-right extremist violence, accounting for a substantial majority of extremist murders in the US, underscores the urgency of this research. The study considers the far-right movement, encompassing white supremacists, neo-Nazis, and sovereign citizens, which is characterized by advocacy for violence to achieve an idealized future beneficial to a specific group, often based on racial, pseudo-national, or individualistic identities. The study's focus on this significant public health issue draws on previous research demonstrating the involvement of various environmental (endemic) factors, exposure to extremist ideologies, and community reinforcement in the radicalization process. The study frames radicalization as a social contagion, requiring multiple exposures for complete adoption, mirroring observed dynamics in broader political movements and other phenomena like extremist propaganda, hate crimes, intergroup conflict, and terrorism.

Existing research indicates that far-right radicalization is not primarily caused by psychopathology, but rather by a complex interaction of environmental factors, exposure to extremist ideology, and social reinforcement within communities. Even lone-wolf actors often interact with online extremist communities. This suggests a social contagion process where extremist ideologies act as complex contagions, requiring multiple exposures before adoption. Previous research has explored similar dynamics in phenomena such as extremist propaganda, hate crimes, intergroup conflict, and terrorism. While there is considerable research on how endemic factors predict radicalization and resulting violence, few studies have examined the interplay between these factors and contagion processes. Therefore, understanding the role of endemic factors like poverty rates, unemployment, income inequality, education levels, and the size of non-white populations, along with the influence of individual factors such as social media use, in the spread of radicalization is crucial. The potential for social media to enhance the spread is highlighted, given its role as a recruitment tool and a space for extremist communities to interact, potentially even augmenting physical organizing efforts.

This study employed a two-component spatio-temporal intensity (twinstim) model, an epidemiological method treating events in space and time as self-exciting point processes, to model the spread of far-right radicalization. Data from the Profiles of Individual Radicalization in the United States (PIRUS) database, including 416 far-right extremists exposed between 2005 and 2017, were used. The twinstim model separates the conditional intensity function into endemic and epidemic components, allowing the assessment of spatio-temporal covariates and epidemic predictors. The study considered endemic factors such as poverty rate, unemployment rate, income inequality, education levels, non-white population size, violent crime rate, gun ownership, hate groups per capita, and Republican voting. Individual-level variables like social media usage and group membership were included as epidemic predictors. To account for spatial clustering, a spatial step function with four 100km intervals and a maximum interaction radius of 400km was used. The temporal step function was similarly divided into four six-month intervals up to two years. The model also incorporated a centered time trend and population density as an offset endemic term. The best-fitting model was determined using Akaike's Information Criterion (AIC), and rate ratios were calculated to assess variable effects. A Monte Carlo permutation test was conducted to determine the statistical significance of the spatio-temporal interaction of the epidemic component. Simulations were also performed to assess the model's quality. Missing data was addressed through multiple imputation with chained equations and random forest machine learning. State-level gun ownership was estimated using a proxy measure based on suicide rates and hunting licenses.

The best-fitting model included seven endemic and two epidemic predictor variables. A significant time trend showed a 4.6% annual decrease in the endemic rate, indicating an increase in the epidemic component's strength over time. Poverty rates and the presence of hate groups had significant positive effects on radicalization probability, while Republican voting, the non-white population percentage, and unemployment rates had significant negative effects. Gun ownership, education level, and violent crime showed no significant effect. Crucially, group membership and social media radicalization had strong, significant positive effects on epidemic probability. The reproduction number (R₀) of 0.31 was significantly higher than simulated null models, indicating that the spatio-temporal interaction in the epidemic model is significant. Simulations showed that the model accurately captures the temporal and spatial dynamics of the data, though it is weighted towards high population density areas. The study notes a baseline increase in the endemic component between 2008-2012, possibly linked to the financial crisis, and a significant spike in the epidemic component around 2016, likely corresponding to the presidential election.

The findings strongly support the hypothesis that far-right radicalization spreads as a complex contagion, requiring reinforcement for transmission. The relatively low R₀ value (0.31) suggests that uncontrolled spread is unlikely, but outbreaks can still occur under conducive endemic and epidemic conditions. The significant positive effect of group membership and social media usage indicates that activism and organizing, rather than copycat effects, are the primary drivers of radicalization clusters. This highlights the enduring importance of local organizing by far-right groups and the augmented role of social media in this process. The negative correlation with certain factors, such as Republican voting and non-white population percentage, warrants further investigation, possibly indicating an interplay between political polarization and racial homogeneity. The lack of significance for violent crime is intriguing, suggesting different drivers for violent crime and extremist violence.

This study provides compelling evidence that far-right radicalization in the US spreads like a complex contagion. Social media and group membership significantly enhance this spread, emphasizing the roles of online and physical organizing. Poverty, low unemployment in regions with high poverty, a lack of racial diversity, and hate group activity increase radicalization risk. Given the persistent threat and the limitations in research funding, policymakers should reconsider their priorities and invest in initiatives addressing far-right extremism, such as developing effective online counter-narratives.

The study acknowledges several limitations, including the use of PIRUS data, which represents only a subset of radicalized individuals due to underreporting of hate crimes. The geographic precision of events is limited to the city-level, potentially affecting spatial clustering analysis. A significant amount of social media data was missing, impacting the interpretation of results despite robust significance. Finally, the limited spatial resolution of some endemic predictors (state-level data for gun ownership, for example) may have obscured local variations. Future studies should address these limitations through improved data collection and higher resolution data.