They’re Still There, He’s All Gone: American Fatalities in Foreign Wars and Right-Wing Radicalization at Home*

What explains right-wing radicalization in the US? Research shows that demographic changes and economic decline both drive support for the far-right. We contribute to this research agenda by 1) studying the elusive early stages in the process of radicalization and 2) highlighting an additional factor that contributes to right-wing radicalization in the US: the impact of foreign wars on society at home. We argue that the communities that bear the greatest costs of foreign wars are most prone to high rates of right-wing radicalization. To support this claim, we present robust correlations between participation in the far-right social media website Parler and fatalities among residents who served in the US wars in Iraq and Afghanistan. This correlation holds at both the county and census tract level, and persists after controlling for the level of military service in an area. The costs of the US’s foreign wars have important effects on domestic US politics. On January 6, 2021, supporters of President Donald Trump illegally breached the U.S. Capitol in an attempt to prevent Congress from certifying then President-elect Biden’s election victory. FBI director Christopher Wray, and later a number of Capitol police officers in Congressional testimony on July 27, described the events as domestic terrorism. Since January 6, nearly all Federal agencies responsible for U.S. national security have declared far-right extremism a grave threat to the country. Though extraordinary, the insurrection of January 6 can be understood as part of a broader trend in far-right radicalization in the U.S.1 Even before the insurrection, the 2020 Homeland Threat Assessment highlighted far-right extremism as an urgent, long-overlooked security concern. Far-right violence is most visibly associated with the “militia” movement (e.g. Oath Keepers, Three Percenters) and white supremacist and racist political organizations (e.g. KKK, Proud Boys, Attomwaffen Division). The most lethal incident of far-right terrorism in American history was the 1995 Oklahoma City bombing, which killed 168 people and injured hundreds more. More recently, individuals steeped in far-right politics have perpetrated mass shootings in Charleston, SC in 2015, Pittsburgh, PA in 2018, El Paso, TX and Poway, CA in 2019, among too many other similar incidents. *We thank Gregoire Phillips for substantial comments on an earlier draft. We also thank Christopher Blair, Anna Meier, and James Fearon for helpful comments. †University of Pennsylvania, Perry World House. ‡University of Connecticut, Human Rights Institute. §Washington University in St. Louis. 1See, among others, Perliger (2020); Jardina (2019); Belew (2019)

An individual's escalation to extremism and violence is rarely sudden or immediate, but rather the culmination of a gradual radicalization process that begins with nonviolent and less provocative activities such as disseminating radical ideas or participating in far-right organizations (McCauley and Moskalenko 2008;Bartlett and Miller 2012;Miller-Idriss 2020).Analogously, elite polarization, in the form of increasingly radical Republican Party leaders and the discourse dominating Fox News, drives macro-level trends in far-right violence and participation in far-right organizations (Kydd 2021).While only a fraction of radicalized individuals engage in extremism and violence, their nonviolent activities have a variety of other negative consequences; from distorting policy and its distributive benefits to eroding confidence in government and democratic institutions.
This research note investigates the early stages of right-wing radicalization in the contemporary United States.
Following McCauley and Moskalenko (2008) and Bartlett and Miller (2012), we define radicalization as the process of increasing extremity of beliefs and behaviors challenging the status quo political order and/or the bounds of mainstream political debate.Extremism refers to the beliefs on this continuum that deviate from the status quo to the greatest extent and behaviors that deploy or justify violence.Far-right radicalization and extremism, specifically, refer to objectives to "preserve or restore traditional politics," or protect the power of those "who have traditionally enjoyed authority, privilege, or wealth" (Jackson 2020, 9-10.).
Existing literature highlights two main causes of right-wing radicalization in the United States: economic anxiety and racial resentment toward minorities, potentially rooted in demographic shifts in the population.In the wake of major structural changes in the economy (e.g.globalization and deindustrialization) or financial crises, some who lose careers or financial security may adopt radical populist, nativist, and anti-globalist ideas. 2 Following a similar logic, erosion of social capital or increased social isolation may be associated with political expression such as voting for right-wing parties (Bolet 2021).Demographic shifts in the racial or ethnic make-up, and cultural shifts away from conservative or traditional values, may trigger a "cultural backlash" against social and political changes (Norris and Inglehart 2019), especially when the majority or privileged group perceives these trends as threatening to erode their political power in the future (Kydd 2021).The politicization of these demographic and cultural changes leads some to develop out-group antipathies (Kinder, Sanders, and Sanders 1996;Kinder and Sears 1981;McConahay 1983) or deeper attachment to group identity (Jardina 2019), in extreme cases leading to privilege violence against groups perceived as enemies. 3e agree that both economic anxiety and racial resentment explanations are important for understanding right-wing radicalization, but highlight a third factor missing from this debate: the impact of U.S. foreign military engagements on politics and society at home.We argue that communities that bear the costs of these wars, specifically in terms of fatalities among community members, may be more prone to high rates of radicalization.Foreign military engagements swell the ranks of the armed services, and sending young adults to fight means that some will return with physical or 2 See, among others, Carreras, Irepoglu Carreras, and Bowler (2019), Baccini and Weymouth (2021).mental health issues and others will not come back at all.Family and friends that survive those killed may acutely oppose the war effort, assess its achievements unfavorably relative to the costs borne, and distrust and vilify the institutions and politicians that support(ed) its continuation.They may seek out narratives and belief systems that channel their grievances, including from the radical right.In addition to connecting with one another and with veterans sharing their disillusionment with government, those directly affected also socialize other community members into radical belief systems and social/political organizations.Therefore, we expect communities that suffer greater fatalities will generate more individuals participating in the early-stage activities of far-right radicalization, such as sharing (mis)information or expressing support for radical ideas on social media.
Experts have long noted a connection between military or law enforcement and far-right politics in the United States.Membership in far-right organizations in the U.S. expands following major war efforts rather than surges in poverty, nativism, populism, or other grievances in the population; and war veterans frequently "bring the war home," joining and leading these organizations (Belew 2019, 20-21).Many American far-right organizations recruit individuals with military or law enforcement backgrounds for their training and capabilities (Ware 2019) and/or encourage their members to enlist in order to acquire relevant experience (Donnelly 2021).Some veterans may be particularly susceptible to recruitment and radicalization for a variety of reasons.In today's volunteer force, individuals from the socio-economic, demographic, and cultural backgrounds that the economic anxiety and racial resentment arguments suggest are more prone radicalization are disproportionately represented in the military.Paralleling the Vietnam War story (Belew 2019), those who served in Iraq and Afghanistan may confront a sense of futility as their counterinsurgency and state-building efforts repeatedly failed to change the security situation due to political realities on the ground, while political backlash against the wars among the population at home may contribute to a sense that their service is unappreciated.Our argument complements this work by emphasizing the effects of military engagements on society more broadly, examining the individuals and communities disproportionately affected by military engagements beyond (exclusively) those who serve.
We provide evidence for this claim by examining publicly available data from Parler, the online social media platform now infamous as a forum for right-wing radicalization, which includes geo-located information on users' uploaded videos.Efforts to measure early stages of radicalization systematically typically face major barriers: extremist organizations remain clandestine, individuals may be wary of revealing their participation, and, unlike violent incidents, the mundane forms of social and political action that mark the early stages of radicalization are not widely investigated or reported on by law enforcement, government agencies, or the media.The Parler data provide a unique opportunity to capture the early stages of far-right radicalization for systematic empirical investigation.
To measure the costs of US wars, we map war fatalities in Iraq and Afghanistan since 2003 to the deceased's hometown.We aggregate both sets of data to the county and census tract levels.We find that the number of military fatalities in a county or tract is strongly related to the number of videos uploaded to Parler.This result is robust to covariate adjustment for a variety of potential confounders, including partisan, income, and ethnic composition.
Crucially, this effect persists even when controlling for participation in the military.This shows that it is not the rates of military service in a community itself that is driving its rate of right-wing radicalization, but its costs in the form of war fatalities.

Research design
To provide empirical support for our claims, we investigate the spatial correlation between right-wing radicalization, measured by videos uploaded to Parler, and the number of war fatalities.Recording and uploading a video to a social media platform requires a higher level of investment than making text posts to the same platform , and video content presents a higher risk of identification and deanonymization.For these reasons, we argue that users uploading videos to Parler allow us to measure a more engaged and mobilized subset of users from the total user base.

Parler videos
Our corpus consists of nearly all public videos uploaded to Parler from its launch in August, 2018 to its removal from its Amazon Web Services hosting on January 10, 2021. 4The data consist of two parts: an archive of metadata files (hereafter metadata), each one representing a video uploaded to Parler, and an archive of the corresponding video files. 5  The metadata contain 1,032,523 records in JSON format.64,520 contain geolocation information on where the video was (presumably shot and) uploaded.Of these, 58,680 videos were uploaded from the United States.We limit our sample to videos uploaded from the start of 2020 until Parler was shut down on January 10, 2021: 57,159 videos.We conduct a human review of a random sample of one percent of the videos to assess their originality, which we discuss in detail in the appendix.The most relevant finding of this investigation is that 27.62% of the reviewed videos contained footage of users filming video on televisions or computer screens.This provides further support for our empirical strategy of using attributes measured in local geographic areas to explain video counts, as many of these videos are uploaded from users' homes.

Location
As our empirical strategy relies on matching counts of videos with other data measured at the same geographic units, we must ensure that videos were uploaded from the user's area of residence.One key threat to this assumption consists of videos recorded and uploaded at protests and rallies, where users by have traveled great distances to participate.To address this problem, we omit videos uploaded within 25 kilometers of a protest or riot recorded on the same day in the Armed Conflict Location and Event Dataset (ACLED) (Raleigh et al. 2010). 6While this may eliminate some uploads from users who live in the vicinity of the event, slightly reducing the sample size is preferable to introducing bias by including videos uploaded outside of a user's home geographic unit.

Ethical considerations
Many of these videos were uploaded from Parler users' homes, and as such the geolocation information constitutes potentially personally identifying information (PII).We take multiple steps to address this concern.First, our statistical results are in no way based off of individual videos; the smallest unit of analysis in our dataset is a count of videos aggregated up to the census tract level.Census information is publicly available at the lower levels than the census tract level from the United States Census Bureau.The privacy concerns that led to changes in the 2020 census were about the block group, not the tract level, since the tract level does not contain enough information to re-identify any individuals (Long 2020).Second, our replication materials do not contain the metadata and thus do not provide geographic coordinates for individual videos.

Independent variable and controls
Our main measure for the costs of foreign wars borne by a community is the number of fatalities for each hometown in the United States.This data was collected in from the iCasualties.orgwebsite, which tracks fatalities in the US wars in Iraq and Afghanistan using information from a variety of official US military and civilian agencies.Their data includes the name, date, service branch, location, rank, age, and hometown and state.To geolocate the war fatalities, we match the hometown to corresponding location in the U.S. Board on Geographic Names for entities.8 We explore data at two different levels of aggregation: the county and census tracts.Census tracts nest within counties and counties nest within states.This allows use to used fixed effects for each higher level of aggregation in all our models (for example, county fixed effects when the unit of analysis is the census tract).
As many scholars have argued that ethnic diversity leads to support for the far-right, we include various demographic controls.All models include measures of the population as a control, split into the racial categories in the US census, since our independent and dependent variables are measured as a sum.We also include the number of refugees settled in a county and the difference in the nonwhite population between 2019 and 2010.We also control for income, which is Finally, to control for military participation, we use official Army Enlistment data that Dean (2018) acquired from FOIA requests.We collapse annual figures to the county level.We also use US Census and American Community Values data on the total number of people in each census unit (tract or county) that have served in the military.We split both the variables into military service before or after 2001 to account for the potential heterogeneous effects to control for an older generation's participation in the military, as military participation in the US is highly clustered within families across generations.Moreover, we control for whether a military base is located in or near the relevant unit of analysis, as we suspect the relationship we theorized could be driven by areas with extremely high military participation.

Results
Figure 2 plots the bivariate relationship between war fatalities and the log of Parler videos uploaded at the county level, normalized by population.While a number of counties that experience no war fatalities do have a large number of video uploads, there is a strong positive correlation overall, as indicated by the regression line.predictor of right-wing radicalization: doubling the number of war fatalities in a county leads to about a 12% increase in the number of videos uploaded to Parler.This is about one fifth the effect that doubling the population has.Moreover, the coefficient for the log of war fatalities (0.126) is precisely estimated with a p-value close to 0.  In addition to the war fatalities variable, this model shows that, as predicted, an increase in the Republican vote share in the 2016 election is correlated with more right-wing radicalization.Interestingly, we find that counties experiencing an increase in the nonwhite population from 2010 to 2019 have less participation in Parler, although the effect of refugees is positive.Income is positive and precisely estimated.We interpret these coefficients to mean that republican leaning counties with a higher than average income are counties that are likely to have higher right-wing radicalization.
The model at the census tract level yield similar results.County fixed effects are included and standard errors are clustered at the county.The scale of each coefficient in the tract models is smaller than in the county level models since the mean logged number of video uploads is much smaller at 0.24 (representative of the more granular unit of analysis).
Nevertheless, the relative magnitudes are the same, with the effect of war fatalities being similar in magnitude as the effect for the size of the white population.To measure socioeconomic variables, we include the share of households with income over $100,000, and with under $50,000, per year.Consistent with the county-level model, higher income census tracts have a higher share of Parler videos, while lower income census tracts have a lower share of Parler videos.
We include whether a military base is located near the tract.Unfortunately, adequate voting and partisanship data at the census tract level does not exist.Many of these controls capture outcomes that occur after the deaths of servicemembers and thus can introduce post-treatment bias.Theoretically, our war fatalities measure is a proxy for the costs of US wars, which is not a one-time event but has persistent effects.To probe the robustness of our results to various combinations of control variables, we conduct an Extreme Bounds Analysis (presented in the Appendix) and find that the effect of the effect of war fatalities on Parler video uploads is consistent across over eight thousand model specifications, and the effect size is very often larger than the model specification we present here.

Conclusion
This research note complements existing explanations of right wing radicalization, a phenomenon of growing of importance in the United States, with attention to the impact of foreign military engagements on radicalization at home.
Our statistical results show a strong correlation between areas in the US whose residents have died in overseas wars and the level of participation in a far-right social media website.This is a remarkably robust result that holds even when controlling for military participation.It is not participation in the military that leads to far-right radicalization, it is specifically the harmful domestic repercussions of foreign military interventions that lead to far-right radicalization in the United States.Our explanation for far-right radicalization helps to provide an additional explanation for the rise of the right in the US that is not located directly in racial animus or economic anxiety.Future research should probe how demographics, economics, and the costs of wars interact to fuel the far-right.
Many studies have focused on the determinants of voting for the far-right or participating in far-right terrorism.
Within these two types of activities lies a process of radicalization.The Parler data allows us to examine early stages of radicalization, typically difficult to observe systematically.The findings show that phenomena that are largely thought of as driven by purely domestic factors may have important international components.
One implication of our argument is that the rise of the far-right in the United States may have different causes than the rise of the far-right in Europe, where overseas military participation is much lower.Future research can explore if any relevant analogs that explain far-right radicalization in Europe.Left unexplored in this research note is the presence of any mitigating effects: are areas that are economically more prosperous less likely to see a correlation between war fatalities and radicalization?Is there any analogous form of radicalization occurring for other ideologies or is this limited to the far-right?We leave these questions for future research.

Figure 1 :
Figure 1: Choropleth map of Parler videos and war fatalities in contiguous 48 US states correlated with right-wing support and military enlistment.Many have argued that globalization has led to right-wing radicalization, so we also control for data on trade shocks.Republican Party vote-share is, almost by definition, correlated with far-right radicalization, and also correlates with military service.We control for broadband access, since it likely drives social media usage.Ensuring that our measures are on similar scales is important.Nonwhite population change, median income, level of education, trade shocks, and Republican vote share are all normalized by subtracting the mean and dividing by the standard deviation.Population variables (including demographic variables and military participation) are logged sums of the number of individuals in each category.

Figure 2 :
Figure 2: This plot shows the bivariate correlation between war fatalities and Parler participation at the county level.Both variables are normalized by population, and the number of videos is logged.The blue line represents the least squares line.

Figure 3 :
Figure 3: These plots show the results of two linear models where the number of Parler video uploads (logged) is the dependent variable.The main independent variable of interest is the number of Iraq and Afghanistan war fatalities in each unit of analysis.The unit of analysis for the model on the left is the county level and the tract level for the model on the right.The county model has state fixed effects with standard errors clustered at the state, while the tract model has county fixed effects with standard errors clustered at the county.Horizontal bands indicate 95 percent confidence intervals.