Can Systems Engineering find a way to counter Violent Radicalization?

Systemic innovation for countering violent radicalization: Systems engineering in a policy context - Clancy - Systems Engineering - Wiley Online Library

by Timothy Clancy, Bland Addison, Oleg Pavlov, Erika Palmer, Khalid Saeed 2w

Systems Engineering

Abstract

This paper brings a systems engineering approach to policymaking in the context of violent radicalization. We test strategies to combat terrorism under the premise that violent radicalization is a complex system of social contagion resulting in terrorism. 

We built a simulation using DIME-PMESII military standards to replicate a terror contagion occurring over 10 years in both physical and online environments under optimal, realistic, and worst-case scenarios. We then tested antiterrorism, counterterrorism, and counter radicalization strategies as policy experiments in this simulation. 

These experiments identified four key dynamics relevant for developing policies to reduce terrorism. 

• First, most well-known policies are ineffective in containing terrorism driven by social contagion. 
• Second, strategies generating backlash can become worse than doing nothing at all. 
• Third, perceived grievance determines the carrying capacity of terrorism in a system, allowing disrupted networks to regenerate. 
• Fourth, variable public support may result in sharp secondary waves of violence under certain contingencies. 

Experimenting with our model, we explore effective ways to address the violent radicalization problem.

Conclusion

Our goal in this paper was to show how systems engineering can be used in policymaking by conducting simulated experiments into antiterrorism, counterterrorism, and counter radicalization policies for combating terrorism under the premise that violent radicalization operates as a social contagion known as a terror contagion. An important part of this is to show how the analysis of policy life cycles from established methods not commonly applied in systems engineering environments (i.e., system dynamics to evaluate sociotechnical systems and policy life cycles) could be brought into a systems engineering context– using a domain that is ubiquitous in systems engineering literature, defense. 

1. The first key finding was that terror contagions are driven by positive feedback loops, like a viral infection. The more high-risk population activated to commit acts of terrorism, the larger the number of cultural scripts operating to radicalize others. A system memory, like a viral reservoir, retains cultural scripts conveying violent ideology and methods of mass violence. The reservoir can then reinfect high-risk populations years later. These two factors combine so that once a terror contagion passes a threshold, it gains a high fault tolerance, and years can pass between subsequently completed attacks.
2. The second finding was that even when policies are effective in the best case, if they spark backlash effects by being too broadly targeted, they can generate perceived grievance within the high-risk population or the larger population within which the high-risk population exists. This backlash effect can cause a steady increase in the high-risk population, which undermines the policy's effectiveness over time. 
3. The third finding shows how this perceived grievance is the carrying capacity of the system of a terror contagion serving as both the source of a terror contagion and renewing or restoring violent radicalization networks disrupted by periodic interventions. 
4. Our fourth finding shows how even a policy that overcomes the previous three challenges can struggle over the long run because as terrorist violence reduces, public perception of risk reduces, and public support for successful intervention declines.

Though early-stage, these experiments in policy research help understand the opportunities and challenges of combatting terror contagions. Future research would advance the simulation from its current form by creating simulations calibrated to specific violent ideologies and corresponding high-risk populations in different geographic regions. Simulation improvements could build upon our existing profiles and data set of 4500 incidents leveraging GTD data.⁵⁵ Incorporating clustering, hyperdimensional, or machine learning techniques could improve profiles and support calibration.⁷⁵ Additional calibrations could incorporate natural language processing of social media to identify specific cultural scripts used in a profile.

Finally, within our stated limitations, we note that focused deterrence and terror contagion containment have significant empirical evidentiary support in adjacent domains. Numerous ongoing pilots of focused deterrence show its benefit against criminal gang violence and domestic partner violence. There are several decades of research on countering the Werther effect of celebrity suicide contagion. Although applying these methods to combating terror contagions will require work, it is not starting from scratch. These results should encourage additional research under the premise that violent radicalization is a form of social contagion, either through the terror contagion hypothesis or other alternatives. Developing strategies along this line to counter violent radicalization may open new effective opportunities for reducing terrorist violence.

Summary

Here are the key points from the document:

• Violent radicalization leading to terrorism is a major global security threat. Mass shootings, far-right extremism, and terrorist groups like ISIS are manifestations of this issue.
• Policies to counter violent radicalization fall into three main categories: antiterrorism, counterterrorism, and counter-radicalization. Examples include professional training, threat assessments, gun control laws, community advocacy, and counternarratives.
• Systems engineering and system dynamics modeling can provide a unique perspective on evaluating and improving these policies over their lifecycle. The models examine how policies perform under best case, realistic constraints, and worst case scenarios.
• Key findings from the simulations:
  • Popular policies often struggle due to positive feedback loops and cultural scripts acting as a "viral reservoir" to reinfect high-risk populations.
  • Policies can backfire if they create grievances that increase the high-risk population over time.
  • Perceived grievance is the "carrying capacity" that allows radicalization to renew itself after interventions.
  • Successful policies can undermine themselves as reduced terrorism leads to reduced public support.
• Combined policy portfolios like focused deterrence and "terror contagion containment" show promise, but require careful implementation to avoid backlash effects.
• Overall, systems engineering provides a useful approach to evaluating and improving policies to counter violent radicalization and terrorism over time. But more research is needed, especially on regional differences.

Human Irrationality

Based on the information provided in the documents, it does not appear that the simulation model directly accounted for factors like human irrationality, religious beliefs, clan/family ties when modeling violent radicalization. The model takes a high-level perspective focused on population-level dynamics.

However, the model does try to capture some relevant macro-level influences that could reflect these micro-level factors, in an abstract way:

• Perceived grievance - This could encompass irrational yet strongly held grievances rooted in religious, clan or tribal identities.
• Ideological similarity - The tendency for radicals to identify with perpetrators of their same ideology may relate to innate group biases.
• Normal abandonment - Even among radicals, there are innate moral inhibitions against violence that lead some to abandon plans.
• Backlash effects - Aggressive policies may fuel irrational desires for revenge tied to family or group honor.

So while not directly modeled, the structure does try to represent some of the emergent effects of irrational biases, religiosity, family/clan ties on radicalization.

To increase confidence in the model, the researchers could try to incorporate more low-level agents representing varied cognitive, cultural, and social dynamics. Adding and testing these against individual case studies could improve validity. This increased resolution may reveal new insights into how the balance of top-down ideological factors and bottom-up identity factors influence radicalization and policy effectiveness over time.

Verification and Validation

Here is how the researchers verified proper implementation and validated the model against historical data:

Model Verification:

• Used standard system dynamics methods like structure verification and parameter verification
• Checked model structure followed documented causal loop diagrams
• Confirmed parameters were consistent with data sources and literature
• Tested boundary adequacy by evaluating behavior at extremes
• Ran sensitivity analyses to check robustness to parameter changes
• Used unit checking in equations and converted units where necessary
• Peer reviewed model with system dynamics experts

Model Validation:

• Recreated historical growth patterns of terrorist incidents using parameter profiles
• Generated behavior modes (equilibrium, struggle to grow, contagion etc.) that matched real world reference modes
• Checked model could reproduce contagion patterns like Columbine school shootings
• Quantitatively compared simulation output to 4600 real incidents from Global Terrorism Database
• Conducted docking by simulating reduced subsets of model and testing against overall behavior
• Confirmed simulated policies like focused deterrence produced results in line with literature

In summary, they used standard modeling best practices like verification, validation, docking, and sensitivity analysis to ensure the model was properly implemented and consistent with real-world data on terrorism and violence. This provides confidence in using the model to evaluate policies. But the researchers acknowledge the need for continued refinement and validation as new data becomes available.

Policies

  the key policies that were tested in the system dynamics simulations include:

Antiterrorism:

• Enhanced law enforcement and bystander response training
• Hardened facilities
• Gun control laws

Counterterrorism:

• Counterterrorism investigations
• Threat assessments
• Red flag laws

Counter-radicalization:

• Community awareness and advocacy
• Efforts to reduce root causes and grievances

Combined policies:

• Focused deterrence (antiterrorism and counterterrorism)
• Terror contagion containment (media guidelines, counter-narratives, highlighting failed attacks)

Some other interventions analyzed:

• Disrupting physical or online radicalization "safe havens"
• Reducing grievances in radicalized populations
• Serial raids/strikes against extremist groups

The simulations tested these policies in isolation and in combined portfolios to evaluate their effectiveness across different scenarios from best case to worst case. The goal was to understand how policies perform over their lifecycle when faced with real-world constraints.

Evaluation Metrics

the key metrics used to evaluate the effectiveness of policies in combating violent radicalization and terrorism were:

• Behavior mode of terrorist incidents over time - Policies were assessed on their ability to shift the behavior mode down to lower levels of terrorism. Going from contagion to struggle to grow to failure to grow and ideally down to equilibrium or below was seen as success.
• Number of contagion incidents - The mean and range of suspected terror contagion incidents were numerically tracked over 1000 simulation permutations. Effective policies reduced the mean number of contagion incidents compared to the baseline.
• Adjusting baseline conditions - Policies were evaluated on their ability to weaken or strengthen the baseline condition they were tested against. For example, shifting from a strong contagion baseline to just a contagion baseline was seen as an improvement.
• Percentage change in contagion incidents - The percent change between the mean number of contagion incidents before and after a policy intervention was calculated. Larger reductions signaled a more effective policy.
• Ability to sustain improvement - Policies were assessed on their ability to sustain improvements over time under realistic constraints like backlash and declining public support. Sudden reversals or subsequent waves of incidents suggested limited sustainability.
• Falsification testing - Policies thought to be effective were falsified by removing components to ensure benefits were arising from hypothesized mechanisms.

The combination of visual behavior modes, numerical metrics, comparisons to baseline conditions, and falsification testing allowed the simulations to rigorously evaluate policies over time under different scenarios.

Simulation and Plans for Future Research:

Simulation Overview:

• System dynamics simulation model representing the lifecycle of violent radicalization and terror contagions
• Models the interactions between incidents, agents, networks, systems of spaces (governed vs ungoverned), and overall system behavior
• Uses stochastic modeling of discrete terror incidents along with underlying differential equations

Key Controllable Parameters:

• Success rates of terror incidents (planning, execution)
• Number of fatalities per incident
• Size of high-risk population
• Perceived grievance levels
• Influence of extremist groups and safe havens
• Strength of policy interventions like deterrence, grievance reduction, contagion containment

Artifacts Generated:

• Timeseries graphs of key variables like number of incidents, fatalities, radicalized population size
• Tables summarizing number of contagion incidents under different scenarios
• Behavior mode diagrams showing changes to overall system behavior over time under policies

Future Research Planned:

• Create simulations calibrated to specific violent ideologies and geographical regions
• Incorporate clustering, machine learning, and natural language processing to improve profiles
• Advance simulation structure based on empirical validation and feedback
• Test combined policy portfolios like focused deterrence and terror contagion containment in field pilots
• Evaluate simulation approach against other methods like agent-based modeling

Overall, the simulation provides a flexible virtual testbed to evaluate policies over their lifecycle under different scenarios. There are many opportunities to advance the artifacts and approach through additional research.

Further Reading and References

In some ways, spread of violent extreme radicalization could be viewed as a contagious disease. Education and absorption of immigrants into the culture might be regarded as vaccination. Curing radicalized individuals doesn't seem to be practical, and should be looked into. Communications serve to moderate the contact rate. I've also assembled a few papers from psych, sociology, and criminology community on the subject, which seem to address other aspects of the problem.

SIR Modeling

---

---

Introduction

"A brief introduction to the SIR model and what is tells us about the spread of COVID-19.

SIR Model Simply Explained by “Micheal Porter”

The SIR model is one of the most basic models for describing the temporal dynamics of an infectious disease in a population. It compartmentalizes people into one of three categories: those who are Susceptible to the disease, those who are currently Infectious, and those who have Recovered (with immunity). At its most basic level, the SIR model is a set of equations that describes the number (or proportion) of people in each compartment at every point in time. The SIR model is often represented with the following flow diagram that shows the three states (S, I, and R) and arrows depicting the direction of flow between the states.

At this point in the pandemic, you may have seen the usual graphical output from a SIR model that shows the number (or proportion) of people in each state over time

These curves come from the (continuous time) SIR model which specifies a set of three ordinary differential equations:

$$\frac{d{S}_t}{dt}=-\frac{\beta I_t{S}_t}{N},\frac{d{I}_t}{dt}=\frac{\beta I_t{S}_t}{N}-\gamma I_t,\frac{d{R}_t}{dt}=\gamma I_t$$

“As the saying goes,”An equation is worth a thousand words".

Unpacking SIR

Every day a contagious person will randomly move around the population infecting any susceptible person they come into close enough contact with. Let’s say a contagious person will come into close contact with (and cough on, sneeze on, kiss, etc.) an average of $\beta$ (beta) people each day; if those people are susceptible (i.e., not immune) they will become infected otherwise they will just benefit from the close contact. If there are $S_t$ susceptible people on day $t$, the expected number infected by our carrier is $\beta S_t/N$, where $S_t/N$ is the probability a close contact is with a susceptible person.

Our infectious carrier will continue this process of randomly contacting people each day until they are no longer contagious (either by recovery or death). Suppose that every day a contagious person has probability $\gamma$ (gamma) of becoming non-contagious. Some readers will recognize that this specifies a geometric distribution for the length of time an infected person is contagious. One property of this model is that the average number of days that someone stays contagious is $1/\gamma$.

Notice that we have specified two model parameters: $\beta$ and $\gamma$. Because these are greek letters you know they are important! Combined, they make another parameter you may have heard of, the $R_0$ (“R nought”). The $R_0$, known as the basic reproductive number, is defined as the expected number of people infected from a contagious person over the length of their contagiousness (in a fully susceptible population). If you were paying attention and keeping notes, you may be able to work out that $R_0=\beta /\gamma$ which is the expected number of close contacts per day ($\beta$) multiplied by the average number of days contagious ($1/\gamma$). If $R_0>1$ (or equivalently, $\beta >\gamma$), then we will likely have an epidemic on our hands as each infected person will infect more than one other on average, who will infect more than one other, and so on (i.e., exponential growth) until we reach herd immunity (but more on that later).

“But what about those equations?”

Refer back to the SIR ODE equations, they describe the change in counts at an instance of time. We can consider a discrete time version of these equations that describe the changes each day using the concepts described above.

Noten that this will give an approximation of the actual SIR output (and there are better approximations then the one described below). However, don’t let this disrupt your sleep; remember that the SIR model is only an approximation of reality and a discrete time perspective is, in my opinion, a better representation than continuous time (e.g., humans don’t randomly mix for 24 hours each day, data is reported daily, etc). Also keep in mind that the model parameters must be estimated, so the fitted curve for the discrete time version should be similar to the continuous time version (even if the estimated parameter values are slightly different).

For notation, take $S_t,I_t,R_t$ as the number of Susceptible, Infectious, and Removed people in the population on day $t$. The total population size $N$ is assumed to stay constant over the observation period and is equal to the sum of all counts (i.e., $N=S_t+I_t+R_t$). This discrete time version of the SIR model specifies the equations:

$$S_{t+1}-S_t=-\frac{\beta I_t{S}_t}{N},I_{t+1}-I_t=\frac{\beta I_t{S}_t}{N}-\gamma I_t,R_{t+1}-R_t=\gamma I_t$$

This shows that the three SIR equations describe how the counts in each category change in one day. Consider the change in the number of infectious; it is the sum of two components:

$$\begin{array}{rl}{I}_{t+1}-I_t& =\text{newly infectious}-\text{removed}\\ & =\frac{\beta I_t{S}_t}{N}-\gamma I_t\\ & =I_t\left(\beta S_t/N-\gamma \right)\end{array}$$

The last line makes clear that this term just multiplies what we described above for a single infectious person by $I_t$, the total number of infectious people at time $t$. In other words, if one infectious person is expected to infect $\beta S_t/N$ other people on day $t$, then $I_t$ infectious people are expected to infect $I_t\cdot (\beta S_t/N)$ on day $t$. Likewise, if each infectious person has probability $\gamma$ of recovery or death on day $t$, then the expected total number of infected people that recover (or die) on day $t$ is $I_t\gamma$.

We can also examine the first term, corresponding to the number of newly infectious people, with a slightly different formulation

$$\begin{array}{rl}\text{newly infectious}& =\left(\frac{\beta I_t}{N}\right)S_t\\ & =f_t{S}_t\end{array}$$

As a reminder, $\beta$ is the transmission parameter and it controls the rate that infectious people can spread the disease (larger values of $\beta$ imply faster spread). The second line makes it explicit that the number of newly infected people at time $t+1$ is proportional to the number of susceptible people at time $t$. Specifically, $f_t=\beta I_t/N$ is referred to as the force of infection and is the fraction of susceptible people who become infectious during time t. Notice that $f_t$ depends on both the transmission parameter $\beta$ and the proportion of people who are infectious at time $t$.

The second term in the equation for the change in infectious people models the number of infectious people that are removed (i.e., recover with immunity or die):

$$\begin{array}{rl}\text{removed}& =\gamma I_t\end{array}$$

The $\gamma \in [0,1]$ parameter is the fraction of infectious people that are removed each day. When $\gamma$ is large, people will recover (or die) quickly and won’t have as much time to infect the susceptible population. However when $\gamma$ is small, people stay infectious longer and have more opportunity to infect others. Because this model is based on expected values, we can consider the separation of $\gamma$ into two sub-components to handle deaths and recoveries separately (e.g., $\gamma =p_{\text{death}}{\gamma }_{\text{death}}+(1-p_{\text{death}}){\gamma }_{\text{recovery}}$).

If you look back at the SIR equations you’ll notice that the change in susceptibles and removed are based on the two components we just described. That is, the change in susceptibles is minus the number of newly infected and the change in removed is the number of newly recovered. This ensures that total number of people in all bins is always equal to $N$.

For the visually inclined, here is my attempt to represent the flow of counts between times $t$ and $t+1$. The equations above the arrows show the fraction of the counts in the left side bin that get transferred to the right side bin with color indicating if the count is positive (black) or negative (red).

Notice that the flow out of each bin must sum to 1 to preserve the required property that total counts must equal the population size. Look now at the bins on the right side. Because $S_{t+1}$ only gets a negative flow it will only decrease (or stay the same) over time. The number of removed at time $t+1$, $R_{t+1}$ only gets positive flow and will only increase (or stay the same) over time. The situation for the number of infectious, $I_{t+1}$ is different in that it can increase or decrease over time.

Hopefully this provides you with a little better understanding of the SIR model. It also reveals that the quote is wrong - an equation must be worth more than 1000 words since I have probably reached that limit and still have a few more things to say about herd immunity and $R_0$.

Make it stop: Herd Immunity and $R_0$

Let’s concentrate on the number of infectious individuals and write out explicitly the estimate for time $t+1$:

$$\begin{array}{rl}{I}_{t+1}& =I_t+\left(\frac{\beta I_t}{N}\right)S_t-\gamma I_t\\ & =I_t\left(1+\beta \frac{S_t}{N}-\gamma \right)\end{array}$$

This shows that $I_{t+1}$, the number of infectious tomorrow, can be expressed as a multiple of the number of infectious today. This leads to the following conditions that determine if the number of infections is increasing, decreasing, or at its peak based on the two model parameters $\beta$ and $\gamma$ as well as the proportion of the population that is susceptible $S_t/N$.

$$\begin{array}{rlrl}\beta \frac{S_t}{N}& >\gamma & & \text{Infectious Count Increasing}\\ \beta \frac{S_t}{N}& =\gamma & & \text{Peak Infectious Count}\\ \beta \frac{S_t}{N}& <\gamma & & \text{Infectious Count Decreasing}\end{array}$$

The peak of the epidemic will occur when the proportion of susceptible people is equal to the ratio $\gamma /\beta$

$$\begin{array}{rl}\frac{S_t}{N}& =\frac{\gamma }{\beta }=\frac{1}{R_0}\text{Peak Conditions}\end{array}$$

which shows one reason $R_0=\beta /\gamma$, the “R nought” value mentioned above, is a prominent part of the discussion.

Let’s examine this in more detail. According to the SIR model, the peak of the infection, that is the day when the number of infected people is greatest, is the first day when the proportion of susceptible peoples fall below $1/R_0$. Equivalently, this suggests the epidemic will peak when the cumulative proportion of the population that has been infected exceeds $1-1/R_0$. This is what is meant by the term herd immunity - when the fraction of the population that is immune reaches a large enough level that the number infected starts to decline.

A recent post by epidemiologists at Johns Hopkins University suggests that the US is not close to herd immunity. Below is a plot that shows, according to the SIR model, the percentage of a population that needs to be immune before it reaches herd immunity for a given $R_0$ value.

The Johns Hopkins article suggests about 70% of the US needs to be immune to reach herd immunity; this would put $R_0=3.33$. There are all sorts of estimates on $R_0$ for COVID-19, but most that I have seen range between 1.5 and 6.5 (which will of course vary region to region and over time).

Changing the disease dynamics

Now that you know something of the SIR model, it can be insightful to consider how the dynamics change under different strategies. Here are three things a population may consider doing to limit the number of peopled infected with COVID-19

1. Reduce the transmission rate (decrease $\beta$)
2. Reduce the length of time that someone is contagious (increase $\gamma$)
3. Reduce the number of susceptible people (decrease $S_1$)

Below is a plot that shows how the infection dynamics are impacted under each strategy. The left facet shows the percentage of the population that is infectious ($I_t/N$) and the right facet gives the percentage of population that has been infected ($(I_t+R_t)/N$). The baseline model uses $\beta =0.30,\gamma =1/10$ ($R_0=3.0$) in a population of 100,000 people. This implies that each infectious individual will come into contagiously close contact with an average of 0.30 people per day and will stay contagious for an average of 10 days. On day 1, there is a single infectious person leaving $S_1=\mathrm{99,999}$ susceptible, and $R_1=0$ recovered. Strategy 1 reduces $\beta =0.25$, Strategy 2 increases $\gamma =1/8$ (contagious for 8 days on average), and Strategy 3 considers that 10% of the population is immune on day 1 ($S_1=\mathrm{90,000}$).

Social distancing policies, limitations on group sizes, and quarantines are used to reduce the transmission rate (decrease $\beta$). This action has been taken by most governments around the world to limit the number of people who are infected and results in a “flattening of the curve” as shown by Strategy 1 in the plot. This strategy forces the peak down (and therefore reduces hospital overcrowding), but extends the total outbreak duration. As can be seen from the cumulative plot on the right, it also reduces the total percentage of the population that will be infected.

The second strategy is to reduce the time someone is infectious (increase $\gamma$). This could be accomplished by medications, supplements, nutrition, exercise, and other treatments. Remdesivir, for example, may reduce the infectious period for up to 4 days on average. In this hypothetical scenario, the SIR model suggests that even a 2 day reduction in the time someone is infectious can have a large impact in the peak number of infections and total number of people infected.

The third strategy is to reduce the susceptible population. This refers to a vaccine, which doesn’t yet exist for COVID-19. According to this SIR model a 10% vaccination rate would not only lead to a reduction in the peak infectious count but also provide the largest reduction in total number of infections (out the three scenarios).

Hopefully, this sheds some light on the different public health responses you have been hearing about in the headlines. There are more variations on the types of mitigation efforts available and certainly better models in which to evaluate how such efforts will impact the disease, but these three strategies and the basic SIR model will take you far in understanding how it all works.

Where to go from here

The purpose of this article is to provide a brief introduction to the SIR model and explain a few of its properties. Hopefully this has cleared up some of the confusion surrounding terms you have been hearing about in the news and also given you a bit more insight into how a compartmental epidemic model works. It may have also left you with other questions and concerns; I’ll outline a few thoughts on where to go next.

The first thing worth mentioning is that the SIR model is undoubtedly wrong in the sense that it doesn’t explain all the complexities and dynamics in the COVID-19 pandemic. But as George Box famously said “All models are wrong, but some are useful”. And the SIR model can be useful in many ways. First, it is a good starting place for understanding the more complex models. Also, never forget that the parameters of these models must be estimated. A simple model can provide better predictions than a more realistic (and complex) model when there is not enough data or the data quality is poor. And we are dealing with both sparsity and quality issues in the COVID-19 data. This is an often under-appreciated aspect of modeling and data analysis.

Nevertheless, I’d say the basic SIR model is too simplistic to be used for accurate modeling of the COVID pandemic. Fortunately, there are several relatively simple ways to improve it. First, we can consider that the transmission parameter $\beta$ (and hence $R_0$) can change over time, especially due to social distancing and stay at home regulations. I would be skeptical of any COVID-19 model that didn’t account for changes in the transmission rate, especially with the unprecedented movement restrictions we have seen around the world. Here is a simple analysis that shows how the infectious counts respond if the transmission parameter $\beta$ is abruptly changed during the course of the epidemic.

The baseline scenario is the same as what was considered in Changing the Disease Dynamics section, $\beta =0.30,\gamma =1/10$ ($R_0=3.0$). The other scenarios consider that $\beta$ is reduced to $0.20$ on a certain day (e.g., due to a shelter in place order). The top left panel is the baseline scenario. The other panels show how the infectious rate varies if the intervention occurs at different time points in the course of the epidemic. The black line is provided as a reference to the baseline scenario. These plots reveal that early changes in the transmission rate have the strongest effect. A change close to the herd immunity induced peak (bottom right panel) has very little effect.

The basic SIR model is deterministic. This means that once the model parameters ($\beta$ and $\gamma$) are specified there is no randomness in the model output. This is apparent by considering that all transitions between compartments are specified by expected values and not probability distributions. One consequence of this is that no uncertainty is represented in the model output - something that can lead to poor decision making. Uncertainty in the model output can be estimated with Monte Carlo simulation of a stochastic SIR model. In a stochastic SIR, instead of modeling exactly $f_t{S}_t$ newly infectious on day $t+1$, the number of newly infectious is randomly determined according to a Binomial distribution ($\text{Bino}(n=S_t,p=f_t)$). Likewise, the number of infectious people that recover is no longer exactly $\gamma I_t$, but randomly determined according to the Binomial distribution ($\text{Bino}(n=I_t,p=\gamma )$). The stochastic SIR will produce a different set of curves every time it is simulated. To get a sense of the uncertainty, the stochastic SIR can be simulated 100’s or 1000’s of times. For example, below is the result of running a stochastic version of the SIR model 100 times.

Notice that while the peak number of infectious is very consistent (around 31.2K), the timing of the peak has much greater variability, starting as early as day 58 in one simulation and including one simulation that doesn’t reach the peak until day 93. A word of caution: the stochastic SIR only addresses the uncertainty in the model output, but doesn’t address the uncertainty in estimating the model parameters (i.e., it assumes $\beta$ and $\gamma$ are known exactly). Parameter uncertainty, especially early in the outbreak when data is limited, will have an even greater influence on how precisely we can make predictions. Making decisions under uncertainty is the rule rather than the exception for this pandemic!

There are many other compartmental models that extend SIR. The closest is the SEIR model that splits out the infected population into two sub-groups, those who are infected but not yet contagious and those who are infectious. The SEIR model is very similar to the SIR model, but Susceptible people who become infected move first move into the Exposed group. There is one additional parameter that controls how long a person stays in the Exposed group before they move into the Infectious state.

The basic framework can be extended to include more detailed compartments and transitions, like having separate parameters for different age groups or how likely people from different geographic locations will interact, etc. For very complex model specifications, Agent-Based Models (ABM) can be employed to understand the corresponding disease dynamics. You can think of an ABM as a type of stochastic compartmental model on steroids; it can capture very complex population behavior using agents that follow relatively simple rules (e.g., leave home between 8-9am, go to work, stop for groceries once per week, go out to eat 2 times per week) and can have individual disease progression (e.g., an older agent has higher probability of becoming hospitalized).

I’ll close this article with mention of an all together different way to estimate epi-curves is with a pattern-based or curve fitting approach. This is the approach taken by Institute for Health Metrics and Evaluation and is based on the observation that over a wide range of parameter values, the SIR/SEIR models maintain a similar shape. For example, notice from the plots above that the curves representing the number/percentage of infectious people look rather bell-shaped. For those of you who have taken a Statistics course, the mention of “bell-shaped” will evoke the wonderful memories of the Central Limit Theorem and the Gaussian or Normal distribution. The IHME modelers exploit this observation and directly estimate the parameters of a Gaussian shaped epi-curve. This approach offers a different way to estimate model parameters (e.g., using fatality data) with a focus on forecasting.

 

Giovanni Valentini (2024). SIR Epidemic Spread Model (https://www.mathworks.com/matlabcentral/fileexchange/75100-sir-epidemic-spread-model), MATLAB Central File Exchange. Retrieved February 1, 2024.  

The SIR Model for Spread of Disease - The Differential Equation Model

Author(s): David Smith and Lang Moore

maa.org

---

As the first step in the modeling process, we identify the independent and dependent variables. The independent variable is time  t,  measured in days. We consider two related sets of dependent variables.

The first set of dependent variables counts people in each of the groups, each as a function of time:

S = S(t)	is the number of susceptible individuals,
I = I(t)	is the number of infected individuals, and
R = R(t)	is the number of recovered individuals.

The second set of dependent variables represents the fraction of the total population in each of the three categories. So, if  N  is the total population (7,900,000 in our example), we have

s(t) = S(t)/N,	the susceptible fraction of the population,
i(t) = I(t)/N,	the infected fraction of the population, and
r(t) = R(t)/N,	the recovered fraction of the population.

It may seem more natural to work with population counts, but some of our calculations will be simpler if we use the fractions instead. The two sets of dependent variables are proportional to each other, so either set will give us the same information about the progress of the epidemic.

1. Under the assumptions we have made, how do you think  s(t)  should vary with time? How should  r(t)  vary with time? How should  i(t)  vary with time?
2. Sketch on a piece of paper what you think the graph of each of these functions looks like.
3. Explain why, at each time  t,  s(t) + i(t) + r(t) = 1.

Next we make some assumptions about the rates of change of our dependent variables:

• No one is added to the susceptible group, since we are ignoring births and immigration. The only way an individual leaves the susceptible group is by becoming infected. We assume that the time-rate of change of  S(t),  the number of susceptibles,¹ depends on the number already susceptible, the number of individuals already infected, and the amount of contact between susceptibles and infecteds. In particular, suppose that each infected individual has a fixed number  b  of contacts per day that are sufficient to spread the disease. Not all these contacts are with susceptible individuals. If we assume a homogeneous mixing of the population, the fraction of these contacts that are with susceptibles is  s(t).  Thus, on average, each infected individual generates  b s(t)  new infected individuals per day. [With a large susceptible population and a relatively small infected population, we can ignore tricky counting situations such as a single susceptible encountering more than one infected in a given day.]

• We also assume that a fixed fraction  k  of the infected group will recover during any given day. For example, if the average duration of infection is three days, then, on average, one-third of the currently infected population recovers each day. (Strictly speaking, what we mean by "infected" is really "infectious," that is, capable of spreading the disease to a susceptible person. A "recovered" person can still feel miserable, and might even die later from pneumonia.)

Let's see what these assumptions tell us about derivatives of our dependent variables.

4. The Susceptible Equation. Explain carefully how each component of the differential equation

   (1)

   
   follows from the text preceding this step. In particular,
   • Why is the factor of  I(t)  present?
   • Where did the negative sign come from?
   Now explain how this equation leads to the following differential equation for  s(t).

   (2)

5. The Recovered Equation. Explain how the corresponding differential equation for  r(t),

   (3)

   
   follows from one of the assumptions preceding Step 4.
6. The Infected Equation. Explain why

   (4)

   
   What assumption about the model does this reflect? Now explain carefully how each component of the equation

   (5)

   
   follows from what you have done thus far. In particular,
   • Why are there two terms?
   • Why is it reasonable that the rate of flow from the infected population to the recovered population should depend only on  i(t)   ?
   • Where did the minus sign come from?

Finally, we complete our model by giving each differential equation an initial condition. For this particular virus -- Hong Kong flu in New York City in the late 1960's -- hardly anyone was immune at the beginning of the epidemic, so almost everyone was susceptible. We will assume that there was a trace level of infection in the population, say, 10 people.² Thus, our initial values for the population variables are

S(0) = 7,900,000

I(0) = 10

R(0) = 0

In terms of the scaled variables, these initial conditions are

s(0) = 1

i(0) = 1.27 x 10- 6

r(0) = 0

(Note: The sum of our starting populations is not exactly  N,  nor is the sum of our fractions exactly  1.  The trace level of infection is so small that this won't make any difference.) Our complete model is

(6)

We don't know values for the parameters  b  and  k   yet, but we can estimate them, and then adjust them as necessary to fit the excess death data. We have already estimated the average period of infectiousness at three days, so that would suggest  k = 1/3.  If we guess that each infected would make a possibly infecting contact every two days, then  b  would be  1/2.  We emphasize that this is just a guess. The following plot shows the solution curves for these choices of  b  and  k. 

7. In steps 1 and 2, you recorded your ideas about what the solution functions should look like. How do those ideas compare with the figure above? In particular,
   • What do you think about the relatively low level of infection at the peak of the epidemic?
   • Can you see how a low peak level of infection can nevertheless lead to more than half the population getting sick? Explain.

In Part 3, we will see how solution curves can be computed even without formulas for the solution functions.

---

¹ Note that we have turned the adjective "susceptible" into a noun. It is common usage in epidemiology to refer to "susceptibles," "infecteds," and "recovereds" rather than always use longer phrases such as "population of susceptible people" or even "the susceptible group."

² While I(0) is normally small relative to N, we must have I(0) > 0 for an epidemic to develop. Equation (5) says, quite reasonably, that if I = 0 at time 0 (or any time), then dI/dt = 0 as well, and there can never be any increase from the 0 level of infection.

David Smith and Lang Moore, "The SIR Model for Spread of Disease - The Differential Equation Model," Convergence (December 2004)

 

Counter-Terrorism Module 2 Key Issues:
Radicalization & Violent Extremism

KKIENERM

unodc.org

---

  This module is a resource for lecturers  

As with the concept of 'terrorism', there is no universally agreed definition of the term 'violent extremism'; indeed, somewhat confusingly, the terms can sometimes be employed interchangeably. There are, however, a number of definitions which have been developed at the national, regional and international levels. A recent United Nations High Commissioner for Human Rights (UNHCHR) Report on good practices and lessons learned on how protecting and promoting human rights contribute to preventing and countering violent extremism examined existing State practice on policies and measures governing 'violent extremism' (General Assembly, Human Rights Council report A/HRC/33/29). This revealed very diverse national approaches (para. 17), a number of which are included in the 'interest box' below. The challenges associated with defining the phenomenon are also revealed in the Report's finding that "[i]n other cases, definitions employed do not make fully clear whether 'violent extremism' presupposes violent action or inciting violent action, or whether lesser forms of conduct that do not normally trigger criminal law sanctions would also be included." (Para. 17). Generally, the diversity of definitional approaches reveals some consistency in that the phenomenon of 'violent extremism' is regarded as being broader than that of terrorism. This is also reflected in the VE Action Plan in which the Secretary-General observed that "violent extremism encompasses a wider category of manifestations" than terrorism since it includes forms of ideologically motivated violence that falls short of constituting terrorist acts (General Assembly report A/70/674, para. 4).

The diversity of what may constitute 'violent extremism' has, to some extent, been shaped by the activities of terrorist groups such as Islamic State of Iraq and the Levant (ISIL), Al Qaeda and Boko Haram, which spread messages of hate and violence as well as religious, cultural and social intolerance. In doing so, groups engaged in violent extremism often distort and exploit religious beliefs, ethnic differences and political ideologies to legitimize their actions as well as to recruit and retain their followers.

Potential pitfalls, that some PVE/CVE efforts have become casualties of, include oversimplification of the phenomenon inter alia with respect to its association with any specific religion, nationality, civilization or ethnic group which can have the effect of furthering rather than hindering violent extremism agendas. For instance, the United States of America's Department of State, in its 2016 Strategy on Countering Violent Extremism, recognized this diversity where itnoted that "the drivers of violent extremism vary across individuals, communities, and regions" (US Department of State and USAID, 2016, p. 3). Despite the focus being on returning foreign terrorist fighters, the understanding of the United States resonates with the contents of The Hague-Marrakech Memorandum, where it was recognized that there is a need for a move towards an individual approach to PVE/CVE efforts. For instance, at Good Practices 16 and 19, the Memorandum suggested that States should deploy individual risk assessment tools, that consider a variety of factors, with the assessments being overseen by trained professionals (GCTF, (A), p. 8). This was reinforced in the addendum to the Memorandum, which recommended for the individual risk assessment tool to be implemented by an expert "proficient in understanding the many facets of radicalisation and the local and cultural context" (GCTF, (B), p. 4).

One aspect that States as well as commentators have sometimes oversimplified is the notion of 'radicalization', a concept which has attracted much attention (and related controversies) including in relation to counter-terrorism prevention discourse and efforts. The UNHCHR Report observed that:

The notion of 'radicalization' is generally used [by some States] to convey the idea of a process through which an individual adopts an increasingly extremist set of beliefs and aspirations. This may include, but is not defined by, the willingness to condone, support, facilitate or use violence to further political, ideological, religious or other goals. (Report A/HRC/33/29, para. 19).

Some commentators have suggested that 'radicalization' can be understood as the process by which individuals adopt violent extremist ideologies that may lead them to commit terrorist acts, or which are likely to render them more vulnerable to recruitment by terrorist organizations (Romaniuk, 2015, pp. 7-8).

Recruitment patterns

Recruitment can take many different approaches. Here is one set of proposed models as to how these might be categorized and critiqued:*

• 'The Net': violent extremist and terrorist groups disseminate undifferentiated propaganda, such as video clips or messages, to a target population deemed homogeneous and receptive to the propaganda.
• 'The funnel': entails an incremental approach, to target specific individuals considered ready for recruitment, using psychological techniques to increase commitment and dedication. Even targeted children who resist complete recruitment may develop positive outlooks on the group's activities.
• 'Infection': when the target population is difficult to reach, an 'agent' can be inserted to pursue recruitment from within, employing direct and personal appeals. The social bonds between the recruiter and the targets may be strengthened by appealing to grievances, such as marginalization or social frustration.

* Gerwehr, Scott, and Sara A. Daly (2006). Al-Qaida: terrorist selection and recruitment. Santa Monica, California, RAND Corporation.  Pp. 76-80. Cited in United Nations, UNODC (2017). Handbook on Children Recruited and Exploited by Terrorist and Violent Extremist Groups: Role of the Justice System . Vienna. P. 13.

As many commentators and governmental/intergovernmental entities now recognize, historically too much emphasis was given to religio-centric ideology as a driver of terrorism (Kundnani, 2015, pp. 10-11), often at the expense of other critical underlying factors being overlooked or given inadequate attention. At the heart of the movement critical of this limited approach is the work of Botha, who drew attention to the significance of individual psychology as being an essential component in the turn to extremism (including terrorism), with Botha concluding that in order to further prevent terrorism it is essential that improved understanding is developed as to what motivates an individual to turn to terrorism (Botha, 2015, p. 3). For example, one of the key findings of the United Nations Development Programme (UNDP) report, Journey to Extremism in Africa (UNDP Report) (2017), was that while 51% of people interviewed cited religious grounds as a reason for joining violent extremist groups, as many as 57 percent of the respondents also admitted to limited or no understanding of religious texts.

The previously mentioned study conducted by Botha supplements this, as it was determined that far from religion being a key component in radicalization, one of the strongest influences was that of an individual losing faith in politicians and political systems. Critically, Botha's study revealed that anger was commonly targeted at agents of the State, due to their role in protecting the incumbent; the impact of this can be seen when the following is considered: "instead of preventing and combating terrorism [the repressive approaches of agents of the State] ensure that young people affected by them - and even other family members - [were] radicalised" (Botha, 2015, p. 13). Certainly, there has been increased recognition that an over emphasis on radicalization may lead to overly simplistic conclusions regarding the causal links between radicalization (resulting in extremist thoughts) and acts of violent extremism. This may invite a 'deprogramming' approach as a/the solution in response without adequately examining other pathways to violence such as socio-economic factors, discussed in the next section. Certainly, the benefit of increased understanding and lessons learnt are reflected in the approach of the VE Action Plan around which the analysis in this Module has been framed.

It is interesting to note that, at times, academic scholarship has been ahead of governmental and intergovernmental institutions in terms of its understanding and thinking on PVE/CVE related issues. For example, Martha Crenshaw, writing back in 1988, noted that the "actions of terrorist organisations are based on a subjective interpretation of the world rather than objective reality", with Crenshaw arguing that the perception of the political and social environment is filtered through their own beliefs and attitudes (Crenshaw, 1988, p. 2). Nowadays, there is increased understanding that the process of radicalization is highly individualized, with no single pathway and often taking many different forms (General Assembly, Human Rights Council report A/HRC/31/65, para. 15). Scholars have drawn on the social-psychological distinctions within beliefs, feelings, and behaviours to disaggregate the radicalization process. Those who turn to terrorist action only form the apex of a pyramid of a larger group of sympathizers who share their beliefs and feelings (McCauley and Moskalenko, 2008; General Assembly report A/70/674, para. 32).

In examining drivers of violent extremism, great caution must be exercised in the terminology used in order to avoid being misinformed by incorrect and/or unchallenged related assumptions. Consequently, some entities have reviewed their definitional and conceptual approaches, such as the European Police Office (Europol) which recently proposed a move away from the term "radicalization" to "violent extremist social trend" (EUROPOL, 2016).

One significant issue, which must be clarified from the outset, is that what is critical to counter-terrorism discourse and efforts is not per se whether individuals hold 'radical' or 'extremist' views (terms which can be relatively subjective in nature and therefore susceptible to misunderstanding) but whether such views are translated into violent acts (which is the exception rather than the norm). Potentially, millions of people drawn from different social, ethnic, cultural, religious, or geographical backgrounds have what some others might regard as 'radical' or 'extremist' views, especially when compared with their own ones, yet do not commit violent or terrorist acts. Indeed, even how the terms 'violent' and 'violent extremism' are defined can vary contextually depending on the method and methodologies used. For instance, a positivist understanding of 'violent extremism' would differ from one derived from the application of a method of 'micro-narratives' or collecting life stories. Micro-narratives are undoubtedly important for better comprehending or addressing more local drivers of violent extremism.

Definitional approaches to 'violent extremism'

There are many different governmental and intergovernmental definitional approaches to the concept of violent extremism, some examples of which are given here.

Governmental

Australia (1*): "Violent extremism is the beliefs and actions of people who support or use violence to achieve ideological, religious or political goals. This includes terrorism and other forms of politically motivated and communal violence."

Canada (2**): "[V]iolent extremism" is where an offence is "primarily motivated by extreme political, religious or ideological views". Some definitions explicitly note that radical views are by no means a problem in themselves, but that they become a threat to national security when such views are put into violent action

USA (3*): The FBI defines violent extremism as the "encouraging, condoning, justifying, or supporting the commission of a violent act to achieve political, ideological, religious, social, or economic goals", whilst USAID defines violent extremist activities as the "advocating, engaging in, preparing, or otherwise supporting ideologically motivated or justified violence to further social, economic or political objectives".

Norway (4*): Violent extremism constitutes activities of persons and groups that are willing to use violence in order to achieve political, ideological or religious goals.

Sweden (5*): A violent extremist is someone "deemed repeatedly to have displayed behaviour that does not just accept the use of violence but also supports or exercises ideologically motivated violence to promote something".

UK (6*): Extremism is defined as the vocal or active opposition to fundamental values, including democracy, the rule of law, individual liberty and the mutual respect and tolerance of different faiths and beliefs, as well as calls for the death of United Kingdom armed forces at home or abroad.

Intergovernmental

Organization for Economic Cooperation and Development (OECD) (7*): "[P]romoting views which foment and incite violence in furtherance of particular beliefs, and foster hatred which might lead to inter-community violence".

United Nations Educational, Scientific and Cultural Organisation (UNESCO) (8*): Whilst recognizing that there is no internationally agreed-upon definition, UNESCO, within the Preventing violent extremism through education: a guide for policy-makers document, suggested that the most common understanding of the term, and the one which it follows within the guide, is one that "refers to the beliefs and actions of people who support or use violence to achieve ideological, religious or political goals". This can include "terrorism and other forms of politically motivated violence".

1* Parliament of Australia (2015). " Australian Government measures to counter violent extremism: a quick guide." February.

2* Public Safety Canada (2009). " Assessing the Risk of Violent Extremists." Research Summary, vol. 14, no. 4.

3* USAID (2011). " The Development Response to Violent Extremism and Insurgency: Putting Principles Into Practice." USAID Policy, September 2011. P. 2.

4* Norwegian Ministry of Justice and Public Security (2014). " Action Plan Against Radicalization and Violent Extremism." P.7.

5* Government Offices of Sweden (2011). " Sweden Action Plan to Safeguard Democracy Against Violence Promoting Extremism." Government Communication 2011/12:44, Point 3.2.

6* HM Government (UK) (2015). Counter-Extremism Strategy . London, Counter-Extremism Directorate, Home Office. Para. 1. See too HM Government (2011). Prevent Strategy . The Stationery Office, Norwich. Annex A. Note that the 2013 UK Task Force on Tackling Radicalisation and Extremism defined "Islamist extremism".

7* Organization for Economic Cooperation and Development (OECD), Development Assistance Committee (2016). DAC High Level Meeting, Communiqué of 19 February 2016 .

8* United Nations, UNESCO (2017). Preventing violent extremism through education: A guide for policy-makers . Paris, France.

Next page

Back to top

 

Radicalization and Violent Extremism:
Risk Factors, Causes and Threat Assessment Workshop at UCSB

Ben Smith

orfaleacenter.ucsb.edu

---

The focus of the workshop is on identifying risk factors, causes and drivers of radicalization, including the role of self-uncertainty, small group dynamics, personality profiles, criminogenic factors, mental health issues as well as social-ecological characteristics of the surrounding environment. At the same time the workshop seeks to address the question of how empirical evidence on risk factors and causes of radicalization should/are informing threat assessment tools and policies of prevention and interdiction in the area of countering violent extremism. The workshop explores these questions in the context of all forms of violent extremism; lone actors and group-based; religious and politically motivated; domestic and foreign fighting. Core issues to be addressed include:

• Conversion, group new-comers and status seeking
• Interaction between personality and self-uncertainty in radicalization
• Appropriateness of existing risk assessment tools of violent extremism
• Overlap between criminal and extremist milieus – the uniqueness of risk factors of violent extremism
• Neighborhood effects on the emergence of radical groups in some localities rather than others
• Enclave deliberation, group polarization and self-uncertainty reduction
• Effectiveness of civic engagement encouragement and other principles in counter-radicalization work
• Systematic differences/similarities between various forms of radicalization and violent extremism

The purpose of the workshop is to bring together leading scholars in the area of terrorism and security studies, the psychology and social psychology of extremist violence and political aggression, criminology and sociology of violence for an inter-disciplinary exchange of views on empirical research findings, their implications for threat assessment and policies and future research projects and ideas. The format of the workshop will be a mix of presentations and plenary discussions in a relatively small setting of invited participants (about 15).

Presenter Biographies and Abstracts

---

Martha Crenshaw “What is new about “radicalization”?”

Interest in the individual level of terrorism or “violent extremism” is longstanding. Many studies have investigated the question of why a small number of people who have had similar experiences and been exposed to the same ideologies turn to violence in the service of a political cause. What has changed since researchers began analyzing the subject in the 1970s and 1980s? Human psychology at the individual and group levels remains constant. 

• What has changed is likely to be the environment. The first critical transformation concerns what can be called permissive conditions. Here the rise of social media in particular as well as advances in communications technology in general have increased the velocity and the worldwide scope of “radicalization” processes.
•  More problematic is the second question of possible change in the instigating or motivating factors for an individual’s turn to violence. Is there a fundamental difference between contemporary radical Islamist/jihadist ideologies and historical predecessors such as revolutionary socialism or nationalism?

---

Lasse Lindekilde “British Muslims Mobilization against Islamist Extremism: The Importance of Action Appeals and Trust”

Research on social movements and collective action has most often been silent on the issue of the effect of characteristics of the mobilizer on the success or failure of mobilization. Most studies in these fields investigate either successful cases of mobilization or collect data on intended behaviours without specifying to respondents, who it is that makes the appeal for action. This paper takes a first step in providing answers to the question of whether and how characteristics of the mobilizer matter to the success of mobilization. It does so by testing hypotheses derived from the literature on persuasive communication and related research, on the case of British Muslims‘ collective action against Islamist extremism.

The data come from a survey experiment of British Muslims (n=825). It features a short scenario about a terrorist attack perpetrated by British Muslim attackers and a subsequent call to action by three different mobilizers (conditions): the British Government, the Muslim Council of Britain (MCB) or nobody specific. We investigate to what degree it matters for British Muslims‘ willingness to engage in collective action, who calls upon them to stand up against Islamist extremism. In doing so we also take into account other relevant variables, including trust in Government/MCB, group efficacy, protest experience, issue saliance, ideological attitudes, identity strength, and emotions.

---

Lella Nouri "‘Fit into our community or piss off back to the prehistoric shithole you came from’: Threat-based boundaries and social-media political community building by the Far Right.”

On 14th March 2018, Facebook banned Britain First from its platform, reasoning that they had “repeatedly posted content designed to incite animosity and hatred against minority groups” (Guardian, 2018). This came just a week after Britain First Leader Paul Golding and Deputy Leader Jayda Fransen were collectively convicted of four counts of hate crime and jailed for a combined period of 54 weeks. The removal of Britain First from Facebook has sparked debate about the content of the group’s Facebook page and its posts, with high-profile figures such as the Mayor of London, Sadiq Khan, describing the group as “a vile and hate-fuelled group whose sole purpose is to sow division” (The New York Times, 2018). By this time last year, Britain First had succeeded in receiving over 1.9 million likes and attracting 1.6 million followers on Facebook (Nouri, Lorenzo-Dus and Di-Cristofaro, 2017), thus making it by far the most popular political party on the platform. Since then these numbers increased still further, with over 2 million likes and 1.8 million followers, meaning that until its removal Britain First had the “second most liked Facebook page in the politics and society category in the UK – after the royal family” (Hope not Hate, 2018).

Against this backdrop, an important question emerges, why does use of social media by groups like Britain First result in such unprecedented popularity? By examining comparative data collected from Facebook and Twitter on two groups that can be classified under the far-right umbrella: Britain First and Reclaim Australia. 

The main aim of our paper is to examine how they use social media. This is crucial for reaching a nuanced understanding of causation in radicalisation processes since ‘how things happen is why they happen’ (Tilly 2006, 410). Drawing upon a +4million word corpus comprising all the Facebook and Twitter posts by Britain First and Reclaim Australia between 21st January and 11th April 2017, our paper examines the linguistic strategies used by both groups. 

Our analysis reveals that the overriding goal of these far-right groups’ use of social media is to establish an ‘imagined political community’ (Anderson,1991). Crucially, the relative salience and inter-relations between the constitutive features of these groups’ communities differ significantly from those proposed by Anderson. Most noticeably, establishing threats against boundaries – rather than developing national narratives – emerges as the main pillar on which these groups’ imagined communities rest. Our analysis also reveals that threat-based boundary establishment is primarily realised by these groups through discursive practices of othering (Coupland 2010), which are known to have an influential effect in digital terrorist propaganda (Lorenzo-Dus & Macdonald 2018).

By providing, to our knowledge, the first systematic study of Britain First and Reclaim Australia’s social media discourse, our paper advances understanding of how the far right uses social media and why their use proves so popular. Additionally, and by empirically testing theories of political group formation (specifically, Anderson’s seminal work) on these groups’ use of social media, our paper puts forward a new model of imagined political community for the far right – one that we hope may be further tested and applied to other extremist groups and media platforms.

---

Mikkel Hjelt  “The ’Where’ of Radicalization: The Social Ecology of Radicalization in Denmark”

My presentation focuses on the connection between individuals and their immediate surroundings in producing radicalization. Therefore, I investigate the immediate socio-physical or virtual environment by characterizing the functions of the radical settings, understanding the individual’s exposure to radical settings, and explaining what characterizes neighborhoods in which radical settings emerge. The data collection has been carried out in Aarhus, Denmark, wherefrom several foreign fighters have left for Syria and Iraq. The data collection applies a mixed methods approach, combining one-on-one interviews and focus groups.

Criminology constitutes the core theoretical framework of the project, more specifically, Situational Action Theory. Furthermore, the project draws on more specific theories of radicalization and extremism. My empirical findings underline the importance of the contextual level in understanding radicalization. 

1. First, exposure to radical settings begins early through a combination of social selection, networks, and recruitment since these young people live close to each other, attend the same schools, or participate in the same leisure activities. 
2. Second, these radical milieus are settings where you connect with other likeminded, socialize and develop a certain common identity through various activities, and discuss religious and political relations and actions, for example, leaving for Syria and Iraq. These radical settings intertwine a physical and virtual companionship. 
3. Third, the radical settings center on locations such as mosques or apartments in certain neighborhoods in Aarhus. 
4. Fourth, the broader socio-physical context seems to affect the emergence of radical settings. The neighborhood of my field research is, on the one hand, a community with a strong sense of mutual trust, helping, looking out for, and knowing each other. 

On the other hand, the same positive neighborhood characteristics also have a downside. These aspects of social organization, collective efficacy, and social capital seem to be limited internally. In this sense, the area is partly detached from the broader society. Furthermore, this neighborhood encapsulates a certain moral context that promotes certain norms, values, and feelings of identity, some of which may be conducive to radicalization.

---

Michael Hogg  “Identity Uncertainty and the Social Psychology of Radicalization”

According to uncertainty-identity theory (e.g., Hogg, 2012), a key motivation associated with group identification and group and intergroup behaviors is reduction of self/identity related uncertainty. Because this motivation is more effectively satisfied by highly entitative groups that provide a distinctive and clear identity, people can develop a social identity preference under uncertainty for xenophobic groups that are intolerant of internal dissent and have autocratic leaders (e.g., Hogg, 2014). Uncertainty identity theory research has shown: 

• (1) self-related uncertainty strengthens group identification, 
• (2) particularly with distinctive groups that have a clearly defined and unambiguous identity; 
• (3) in the absence of an available socially valued identity uncertainty can motivate identification with societally devalued groups; 
• (4) uncertainty motivates people to accentuate the distinctiveness of a group they belong to and of its identity, and 
• (5) can cause people to dis-identify with groups that do not have a distinctive identity; 
• (6) uncertainty motives a need for leadership, 
• (7) particularly leadership that delivers an unambiguous identity message in an affirmational and directive/autocratic style that is uncertainty reducing; 
• (8) uncertainty can also create a preference for leaders who possess the Dark Triad traits of Machiavellianism, narcissism and psychopathy, and can facilitate such people’s successfully attainment of leadership; and 
• (9) people who are uncertain about their membership status in a group that matters to them are the most likely to become zealots who engage in extreme intergroup behavior on behalf of the group. 

In this talk I overview uncertainty identity theory and its findings in order to underscore the potentially powerful role that identity uncertainty plays in radicalization and violent extremism.

---

Oluf Gøtzsche-Astrup “Resorting to violence: A generalized dark mindset mediates the effect of uncertainty on intentions to engage in political violence”

I present and test a theoretical framework that explains how fundamental uncertainty can lead to acts of political violence at the individual level. Existing theories claim that fundamental uncertainty is central to predicting political violence, but there is debate concerning the mechanism through which this happens. 

Drawing on social identity theory and cognitive and behavioral neuroscience, I argue that the negative emotionality caused by fundamental uncertainty is attributed to broader causes in the world through cognitive biases, which causes the individual to hold a dark mindset of the world. This dark mindset in turn makes acts of political violence, but not legal activism, seem necessary and attractive as an ingroup defense. 

In an observational pilot study and a population representative concurrent double randomization design study in the United States and Denmark, the mechanism linking fundamental uncertainty to political violence is tested. The results support the theoretical model and indicate a new target for policy interventions that attempt to lessen political violence caused by economic and political instability in society.

---

Clark McCauley “Evolution of the Psychology of Terrorism: Lessons for Analysts and Security Officials”

Looking back over developments in the psychology of radicalization and terrorism, several trends emerge. First was a turn away from seeing terrorists as crazy to seeing them in a rational choice framework in which at least terrorist leaders try to maximize the effectiveness of strategies and tactics. A more recent development gives greater attention to emotions in explaining terrorist behavior. Second, and related to the first, there was initially an emphasis on individual-level explanations of terrorism, then a recognition of the power of group, and most recently increased attention to the social movement and public opinion contexts in which terrorism occurs. Third, initial efforts to understand terrorism focused on them—the terrorists. Building now, if slowly, is attention to the action-and-reaction dynamic of the conflict between terrorists and the government they oppose. Several implications of recent trends are briefly identified.

---

David Parker  “Encouraging Public Reporting of Radicalisation: The Impact of Framing”

Anecdotal and qualitative research suggests that when encouraging the public to report concerns of radicalisation, framing this action in terms of safeguarding the individual in question (i.e. a care framing) is the most effective. 

This approach is employed in a range of countries, including the UK and Denmark. Despite this, data available indicates that of all referrals received, the proportion from the public is extremely low. Furthermore, communication efforts to date have prioritised specific communities rather than the general population. However, anecdotal information indicates the potential for future radicalisation awareness communications to fall under broader CT branding that prioritises a risk framing. Such an approach would be in line with research that suggests risk framing can be an effective approach in influencing public opinion / behaviour. 

This presentation discusses the use of an experimental survey with an N-set of 4,000 nationally representative members of the public (2,000 in the UK and 2,000 in Denmark). The survey experiment seeks to identify causal inference on behavioural intentions when communications prioritise a risk or care framing (using intimacy as a further treatment) and attitudes towards the behaviour that could act as barriers of drivers of reporting. This data will add to the academic literature on both framing and counter radicalisation as well as having significant policy relevance for a range of states delivering radicalisation prevention strategies.

---

Noemié Bouhana “Nothing More Practical Than a Good Theory: Development of a Causal Meta-Framework to Support the Assessment of Radicalisation and Terrorism Risk”

Changes in terrorist patterns over time and the unreliability of risk factors and indicators reflected in the absence of empirically-supported, stable terrorist profiles have hampered efforts to develop structured tools in support of frontline risk assessment of individual radicalisation and terrorism. As low base rates remain an insurmountable obstacle to the adoption of actuarial methods, structured professional judgement has emerged as the approach of choice for use in preventative, investigative and prison settings. This begs the question, however, of how such a judgement should be structured. 

This paper presents work carried out to develop a risk analysis, causally-informed meta-framework grounded in criminological theory (specifically, Situational Action Theory and opportunity theories) to support risk assessment practice above and beyond the use of risk factor-based tools. Findings from the EU-funded FP7 PRIME project on lone actor extremism, which draw from a dataset of 125 offenders, and from a qualitative study of radicalisation carried out in UK high-security prisons, are used to illustrate the conceptual necessity and the practical, diagnostic value of the proposed framework.

---

Amy-Louise Watkin and Seán Looney “‘The Lions of Tomorrow’ A News Value Analysis of Child Images in Jihadi Magazines”

This paper reports and discusses the results of a study that investigated photographic images of children in five online terrorist magazines to understand the roles of children in these groups. The analysis encompasses issues from 2009-2016 of 

• Inspire, 
• Dabiq, 
• Jihad Recollections (JR), 
• Azan and 
• Gaidi Mtanni (GM) . 

The total number of images was 94. A news value framework was applied which systematically investigated what values the images held that resulted in them being ‘newsworthy’ enough to be published. 

This paper discusses the key findings which were that Dabiq distinguished different roles for boys and girls, portrayed fierce and prestigious boy child perpetrators, and children flourishing under the caliphate. It is thought that these images intended to inspire sympathisers to join and were a form of psychological warfare. Inspire and Azan focused on portraying children as victims of Western-backed warfare which is thought to have been an attempt to create feelings of guilt and anger towards the West. Finally, GM portrayed children supporting the cause peacefully, and JM contained no re-occurring findings. This is thought to be because the groups believed that using children would portray them negatively.

Terrorism and the internet: How dangerous is online radicalization?

ncbi.nlm.nih.gov

Jens F. Binder 1 , * , † and Jonathan Kenyon 2 , †

---

• Journal List
• Front Psychol
• PMC9606324

Abstract

This work is concerned with the extent and magnitude of threat related to online radicalization in the context of terrorist acts and related offending. Online influences have been depicted as major drivers for the propagation and adoption of extremist ideologies, which often contain an element of collective grievance, and subsequent acts of violence. This is most pronounced in the discussion of so-called lone actor terrorism, but extends to all forms of extremist offending, and beyond. The present work situates online radicalization leading to terrorist acts within the wider context of grievance-based beliefs and attitudes. Further, it addresses current positions and debates surrounding the relevance and mechanisms of online radicalization in terrorist offending. Recent evidence from quantitative studies is reviewed to estimate prevalence of online radicalization and the level of threat that results from it. This is followed by a discussion of plausible, but opposing, interpretations of the estimates presented. While online radicalization does occur, with and without reference to offline processes, the resulting threat is not overly high. This assessment, however, refers to the present only and is unlikely to hold for the future, given the general growth and acceleration of online activity among terrorist actors.

Keywords: online radicalization, terrorism, risk assessment, extremism, terrorist offending

Online radicalization as a cause for common concern

This work is concerned with the extent and magnitude of threat related to online radicalization in the context of terrorist acts and related offending. Online radicalization is here understood as a process during which individuals get exposed to, imitate and internalize extremist beliefs and attitudes, by means of the Internet, in particular social media, and other forms of online communication. This definition is adopted for entirely pragmatic reasons and should not mask the fact that almost none of its terms has gone uncontested (Neumann, 2013; Gill et al., 2015; Macdonald and Whittaker, 2019; Evans and Williams, 2022; Rothut et al., 2022). From a forensic perspective, such radicalized individuals are seen as at an increased risk of committing offenses which may take the form of violence, causing harm and death to many, as in violent acts of terrorism (Kenyon et al., 2021a; Evans and Williams, 2022; Hamid and Ariza, 2022).

The present work will, first, situate online radicalization leading to terrorist acts within the wider context of grievance-based beliefs and attitudes. This will allow for an outline of the extent to which the specifics of terrorism studies are generalizable and can contribute to a wider integrative perspective on grievance and violence. Second, we will address current positions and debates surrounding the relevance and mechanisms of online radicalization in terrorist offending. Third, we will review recent evidence that is available on the prevalence of online radicalization and the resulting level of threat. For this, our emphasis is on recent, quantitative studies, less so on qualitative and theory-driven work, although we acknowledge the wealth of important contributions from such work in the wider thematic area. This allows us to arrive, fourth, at a quantification of threat levels, which, we believe, is crucial to current debate.

Online radicalization processes have been of major concern, not only in the area of terrorism, but in the wider field of grievance-based violence. In fact, recent work has introduced a comparative approach that builds on the commonalities between perpetrators of, for example, high school shootings, hate crimes, and terrorist attacks (Brooks and Shaw, 2022; Clemmow et al., 2022; Ebbrecht, 2022). In particular for offenders deemed to be lone actors, the boundaries between terrorism and other forms of offending are blurred (Capellan, 2015; Kenyon et al., 2021a; Clemmow et al., 2022). Capellan (2015) sees both ideological and non-ideological mass shootings as belonging to one broader type of homicide defined as lone actor grievance-fueled violence. Similarly, Clemmow et al. (2022) propose a general Lone Actor Grievance-Based Violence framework that accommodates both lone actor terrorists and mass murderers, based on a detailed cluster analysis on several dimensions (propensity, situation, preparatory, leakage, and network indicators).

Over the past decade, the way in which the Internet presents, selects, connects and curates information, by virtue of its architecture as much as through user activity, has been identified as particularly concerning in the context of extremist ideologies. Broad concepts that have emerged address the dangerous normalization and acceptance of extremist messages that result from such information management. For example, Von Behr et al. (2013) reviewed evidence that pointed to the formation of echo chambers online, structures in which individuals can surround themselves with likeminded others and help reinforce each other’s views, thus contributing to an amplification of opinions. Related to echo chambers, filter bubbles (Pariser, 2011) have received sustained attention. For these, automated algorithmic selection of content is the main driver. Individuals are exposed to more and more content of the same type, at the expense of alternative viewpoints. Although a solid understanding of the actual effects of such broad mechanisms on radicalization has not been reached yet (Reed et al., 2019), their potentially sweeping generality and relevance is without question. Studies have documented problematic Internet uses for specific platforms across a spectrum of different forms of extremism, ranging from clearly political ideologies (e.g., right wing; O’Callaghan et al., 2014) to those that can be associated with religion (e.g., jihadi-inspired; Clifford and Powell, 2019; Macdonald et al., 2019) and those that are more difficult to categorize such as entrenched misogynistic world views (Speckhard et al., 2021). In the following, we will focus on a more detailed review of the role and specific mechanisms of online radicalization in the context of terrorism.

Online radicalization and terrorism

The specific context of terrorism

Although there is some indication of a common basis for grievance-based forms of offending, there are a number of specific factors that surround acts of terrorism. These are important to highlight for a further investigation of online radicalization. All definitions of extremism and terrorism are contentious, but there is general agreement that a frame is provided by some ideology supportive of violent changes to societal and/or political order (see, for example the perspective adopted by the United Kingdom government in its most recent counter-terrorism strategy: UK Government, 2018). This also means that there are pre-existing structures and organizations that represent, shape and use such ideologies and exert influence on individuals as members and followers. These organizations are concerned with recruitment or member management and the channeling of activity. At their most powerful stage, they assume para-military and quasi-governmental forms (as in the example of Al Qaeda; see Gunaratna and Oreg, 2010).

But next to these organizational forms, a wider gray area can be identified, in which individuals are inspired to commit acts of violence. This is captured by the label of the lone actor terrorist (Gill, 2015; Kenyon et al., 2021a). Online radicalization in the context of terrorism can therefore occur in direct exchange with networks and groups with a high interest in recruitment and a readiness to invest resources in communication and outreach activities; online radicalization can also occur in a less systematic way, driven by the individual. This duality is further reflected in theoretical explanations of radicalization that focus either on bottom-up (i.e., through emerging group dynamics; Sageman, 2004, 2008) or top-down dynamics (i.e., through hierarchies that channel influence from an organization to those to be radicalized; Hoffman, 2008), or, indeed, a synthesis of both (Conway and McInerney, 2008).

It should also be considered that a wide range of content generated by organizations classed as terrorist or extremist is deemed illegal in many countries, as is the formal or informal organizational membership. This poses a dilemma for terrorist organizations operating online: high levels of secrecy can be achieved through encryption, thereby minimizing the risk of detection, but this limits outreach to recruits and sympathizers severely. Further, the accessibility of extremist materials online can be very high, but digital files leave traces on individuals’ devices, and the mere downloading of certain materials can lead to detection and prosecution. As a result, terrorist groups have shown substantial adaptability and flexibility in their use of online services and platforms (UK Home Office, 2019). A common strategy established over the last few years consists of using entry points on mainstream sites that can be used to guide those interested to other digital locations such as encrypted services or dedicated web sites (Clifford and Powell, 2019; Macdonald et al., 2019).

Recent work on online influences in terrorist offending has provided evidence both for and against a perspective on Internet activities as a specific risk factor. Separate lines of research suggest that offenders radicalized online pose the least threat to society when compared with those who have more, and more face-to-face, social exchanges (Kenyon et al., 2021b, 2022a; Hamid and Ariza, 2022). In addition, online radicalization has been criticized as an overly simplistic, artificial construct that neglects the realities of today’s seamless transitioning between online and offline spheres (Gill et al., 2015; see also Conway, 2016). At the same time, the evidence base also indicates that online radicalization can and does occur, with potentially violent consequences, as in the case of some lone actor cases (Kenyon et al., 2021a).

There is no doubt that online activities play an important role in most forms of terrorism. Research has documented how terrorist organizations and terrorist actors have kept pace with technological development. To the extent that the Internet permeates all aspects of our daily lives, it is also an integral part of the propagation of extremist ideologies and resulting actions and operations. From the start of more wide-spread Internet use, research has documented how novel forms of online engagement have led to novel aspects of terrorist activity (e.g., Weimann, 2006, 2014). A recent overview by Evans and Williams (2022), based on a synthesis of earlier studies, groups online activity into five broad domains: Financing, networking and coordination, recruitment and radicalization, knowledge transfer, and mobilization to action. The main conclusions by Evans and Williams (2022) indicate that all extremist movements engage in online activities, in ways and with platforms that are no different from normal everyday uses of the Internet. In relation to radicalization, the Internet plays an integral role in the generation, consumption and spread of extremist propaganda.

Other work has focused on the facilitating role of the Internet during the radicalization process itself, often emphasizing that online and offline influences are intertwined and reinforce each other (Von Behr et al., 2013; Gill, 2015; Valentini et al., 2020). Jensen et al. (2018) assign an accelerating role to social media, in particular for the radicalization of foreign fighters, but see radicalization as a process which is not exclusively online or offline. Likewise, Herath and Whittaker (2021), similar to Gill’s (2015) earlier work, take issue with a clear-cut dichotomy of either online or offline and provide evidence for several radicalization pathways that combine both types of influence in different measure. Indeed, Whittaker (2022) argues that any separation of online and offline radicalization is meaningless since both domains are part of the same information environment and cannot give rise to different processes of radicalization.

Next, to any conceptual debate, however, it remains a fact that any individual on a pathway toward increasing radicalization may obtain relevant information from the online world, in large quantities and at comparatively low levels of environmental restrictions and control. This poses a challenge to policy makers and regulators. Regulating and monitoring the online world requires measures, resources and, often, legislation different from those needed in an offline public sphere. Simply declaring the Internet to be an integral and inseparable part of our lives will not resolve this challenge and does not offer nuanced responses. Where authors have considered radicalization to happen (nearly) exclusively through online means, opinions on resulting threat are mixed. Hamid and Ariza (2022) concede that online radicalization exists and poses a problem, however, they conclude that it constitutes a lower threat than other forms of radicalization and is of lesser pertinence to security. Other work also indicates that threat levels differ depending on the online and offline means of radicalization (Gill et al., 2017; Jensen et al., 2018; Kenyon et al., 2021b, 2022a).

Mechanisms of online radicalization

Before the estimation of prevalence and threat are addressed in the next section, the frame of the debate is best shaped by addressing an a priori question, namely whether Internet technologies are suitable and have the actual power to lead to radicalization “on their own.” In this section, the focus is on studies that have outlined how Internet technologies can support and facilitate radicalization processes, potentially independent of any offline exchanges. Core questions that emerge from these studies concern the role of active and passive uses of the Internet and how these can further extremist attitudes and beliefs. Such mechanisms provide a more solid basis to consider online radicalization as a persistent problem.

Echo chambers and filter bubbles, as outlined above, have been identified as possible mechanisms almost a decade ago (Von Behr et al., 2013; Reed et al., 2019). Some research findings tentatively affirm that such mechanisms have also been effective in recent years when it comes to radicalization. Further, there are now studies that have followed individuals and their online activities much more closely and allow for a more detailed understanding of the mechanisms at play. It should also be noted that the general consensus sees the role of the Internet as that of a facilitator or catalyst, far less as a driving causal factor (see Meleagrou-Hitchens and Kaderbhai, 2017). As such, the question here is not so much how the Internet would cause radicalization, but how precisely it can support such a process in those individuals who are particularly vulnerable.

Gill (2015) focused on a behavioral analysis of lone actor terrorism, i.e., cases characterized by an absence or scarcity of social interaction. The Internet main roles concerned the reinforcement of the individual’s radical mind set, the dissemination of propaganda and information leakage prior to an attack. Of those functions, reinforcement is most likely to be of relevance during the radicalization process. The comprehensive analyses by Hamm and Spaaij (2017), covering more than 60 years of lone actor terrorism in the U.S., may contribute to a wider understanding of the reinforcement that can be obtained online. The authors found that lone actors were more likely to maintain some affinity with an extremist organization in the time period before the 9/11 attacks compare to after. Hamm and Spaaij explained this shift with increased online activity and a change in audience and social influence. Lone actors are thought to obtain ideological direction not through organizations, but networks of anonymous online activists, a crucial transformation that has made lone actor terrorism more decentralized and leaderless.

There is some suggestion that exposure on its own has some substantial effects. Hassan et al. (2018) conducted a systematic review on the link between exposure to extremist online content and violent radicalization. Having identified a set of 11 empirical studies, using a range of methods and focusing on several extremist ideologies, the review concludes that there is tentative evidence that exposure leads to radicalization, although it is not clear which level of involvement is needed on the user’s side to become more radicalized. Similarly, Wolfowicz et al. (2022) reviewed and integrated experimental and observational evidence in a comprehensive meta-analysis. Based on four experimental studies, the authors obtained a small effect for mere exposure to media content, i.e., with passive study participants, on radicalization outcomes (Hedges’ g = 0.08), which was slightly increased in case of high trait aggression (g = 0.13).

Focusing in detail on the Twitter activity by 110 self-proclaimed Daesh supporters, Smith et al. (2020) were able to show how conformity to the linguistic and stylistic aspects of an extremist group environment increased over time and was positively related to engaging in group mobilizing interaction. While these findings demonstrate how radicalization processes can be detected online, and are expressed in social media activity, the focus is clearly on users who are neither passive nor in social isolation. A similar level of activity is described in the study by Speckhard et al. (2021) on self-defined involuntary celibates (“incels”) online. The authors provided an account of how a subset of those identifying as “incels” are further radicalized in online forums that support the immersion in a grievance-based perspective and lead to an increased endorsement of violence. Within active and extended online networks, there is also the possibility that radicalizing messages are controlled by feeder accounts, thus channeling influence in a more organized manner, as in the study on the Twitter networks surrounding foreign fighters in Syria by Klausen (2015).

Using a large (44 k) sample of Twitter users, Magdy et al. (2016) compared interactions online (use of hashtags, retweets, replies, and mentions) prior to and after the 2015 terrorist attacks in Paris. Negative attitudes toward Muslims and endorsement of extremist hashtags after the event could be predicted to a substantial extent from previous Twitter activity (e.g., consumption of anti-Muslim tweets), even in the absence of prior references to Islam by the user. These findings point to the possible effects of more passive social media consumption, or social media activity that is not focused on a particular target, which increases the readiness for developing more specific extremist views.

Next, to the question of how much activity or engagement is necessary online to support radicalization, other work has focused on the type of format and content that is most effective. Wolfowicz et al. (2022), in their meta-analysis, attempted to separate online exposure from other forms of media consumption. Pooling outcomes from 49 observational studies, they conclude that TV consumption carries no effect while active and passive online exposure to radical content are related to risk of radicalization (r = 0.22 for active, and r = 0.24 for passive online consumption). Among active information seeking online, accessing jihadist magazines showed the strongest association with radicalization (up to r = 0.29), in contrast to beheading videos (r = 0.16), possibly because these are more indicative of violence and aggression more generally. This finding coincides with the study by Frissen (2021) on a large sample (>1,800) of Belgian young adults: self-reported cognitive radicalization was most pronounced.

The empirical evidence to date has been integrated in several theoretical analyses and frameworks. For example, Mølmen and Ravndal (2021) derive a total if six factors of theoretical importance to the process of radicalization from a review of the literature, three of which carry particular relevance in an online context. These are facilitation, acceleration, and echoing. Facilitation encompasses any intensification in the exposure to extremist content, acceleration refers to the shorter timeframe that is assumed for online radicalization as compared to offline processes, and echoing implies a further reinforcement, and normalization, of an extremist mind set due to the like-mindedness of the sources of influence encountered online. Likewise, Neo (2019) has proposed a model of internet-mediated radicalization that outlines the supportive functions of Internet technologies during five phases of the radicalization process: reflection, exploration, connection, resolution, operation. It is worth noting that while the connection seems to suggest actual communication with others, this phase can also be dominated by unidirectional online influences, without interaction.

In sum, the Internet provides several functions and mechanisms that allow for online radicalization, and likely so in the absence of actual social interaction. It seems that development of a grievance-based perspective, and the deeper immersion therein, are most effectively achieved by combining both asocial and social engagement online. It should be added, however, that there is general agreement that a combination of online and offline processes is seen as most effective in the furthering of the radicalization process.

Evidence on threat levels

In this section, recent studies are reviewed to, firstly, establish our understanding of the prevalence of online radicalization and, secondly, to arrive at some informed estimate of the actual threat level that results from such radicalization. To this end, the focus is on quantitative studies that are based on some clearly defined population of terrorist actors and allow for statistical interpretation and generalization, to a certain extent. As will become clear, all such studies differ from each other in terms of the underlying data sources, the type of terrorist actor under investigation and the precise set of variables and operationalizations used. Following a review of prevalence and threat, a wider discussion is initiated of the divergent interpretations that can be derived from the current state of knowledge. By alternating between conflicting critical narratives, the aim is to get closer to an answer of a core question of the present work: How dangerous is online radicalization?

Evidence on prevalence of online radicalization and associated threat

In their landmark study from 2015, Gill et al. provided a detailed account of online activities of 227 United Kingdom-based terrorist actors, covering the period from 1998 to 2013. Overall, there was evidence for some online activity related to an attack or relevant terrorist offense in 61% of all cases. Looking at specific activities, 54% of all cases used the Internet for learning, 44% for the spread of extremist online media, 32% for attack preparation. Some of these figures, unsurprisingly, were markedly increased toward the end of the time period covered. In a follow-up study, using a modified data set with 223 entries, Gill et al. (2017) further differentiated online activity based on several offender characteristics. So-called lone actors were substantially more likely (2.64 times) than group-based terrorists to learn online. The type of attack was likewise correlated with online activity, with those concerned with using Improvised Explosive Devices (IEDs) being more likely to engage in online learning compared to other attackers.

The figures from Gill et al. (2015) are roughly confirmed by Whittaker (2021) who used a data set on 231 U.S. based Daesh (IS) terrorists, all that were recorded during the period 2010 to 2020, and their online activities. Some online activity was found to be present in 92% of all cases; more than 80% interacted online with co-ideologues, 80% used social media platforms for at least some of their activities, 36% had disseminated propaganda online. The somewhat increased percentages are not surprising given the extended time period up to 2020 and the fact that the peak activity of Daesh/IS falls into the years 2015 and 2016, after what the study by Gill et al. (2015) was able to consider.

These findings indicate the overall importance of Internet technologies for terrorist actors, and they provide important detail on the type of activity, for the United States and the United Kingdom. They stop short, however, of assigning a specific role of such activities to the radicalization process proper. While Gill et al. (2017) link lone actor terrorism to both online activities and to the severity of the chosen plot and attack method, it would be premature to conclude that lone actors define all relevant cases of online radicalization. It can be assumed that radicalization is an ongoing development and continues while actors are fully operational. Under this assumption, all online activity would also be relevant to radicalization, by definition. Other studies, in contrast, have placed a direct emphasis on the role of online activity within the radicalization history of individuals, as far as this can be reconstructed from sources. The focus here is, again, on quantitative studies that allow for some estimate of overall prevalence and threat level.

Bastug et al. (2020) investigated the role of social media for 51 Canadian Islamist extremists from 2012 onwards. Information on radicalization was available for 32 individuals. Of these, online activities were underpinning the radicalization process in 21 cases. This puts the prevalence rate at in between 41% and 66%, for an overall group size of 51 or 32, respectively. In this study, however, online activities could occur alongside other radicalization mechanisms. The prevalence rate therefore refers to mixed modes of radicalization as much as to more exclusive online influences.

Similarly, Jensen et al. (2018), using the comprehensive PRIUS data base of U.S.-based extremists, noted that radicalization involving social media rose substantially over time. In the period from 2011 to 2016, social media were assigned a primary role in radicalization for 17% of all cases (n = 295), across a spectrum of causes including jihadist, far-left, far-right and single issue ideologies. A primary role of social media was assumed if exposure to extremist ideologies and more than half of the socialization took place online. This provides a more restrictive criterion for online influences, but again assumes a mixed-model of radicalization. The study also provided an opportunity to discuss the acceleration potential of the Internet for the process of radicalization. By focusing on a sub-set of jihadist foreign fighters, Jensen and colleagues were able to define a meaningful start and end point to radicalization (i.e., from the first time contact with extremist ideologies to the first attempt to take up the role of foreign fighter), and they found that as social media engagement increased the duration of the process decreased.

Returning to the challenge of interpreting reported prevalence, using mutually exclusive categories for online and offline radicalization pathways allows for more insightful estimates. In a study on individuals arrested in Spain for activities related to jihadi terrorism, Reinares et al. (2017) collected information on 178 cases. The time period covered reaches from 2013 to 2016. For 119 cases in the sample, information on the radicalization environment was available and a classification according to Internet activity could be established. An environment that was exclusively online was found in 35% of cases, offline only was the case for 24%, and for 40%, a mix of online and offline influences was found. Of note, radicalization was defined here as development prior to involvement in terrorist activities.

The comprehensive study by Hamid and Ariza (2022) provides a rare opportunity to relate the radicalization pathway to the severity of the terrorist act. This allows for a direct, quantified estimate of the level of threat that follows from different radicalization modes. Focusing on attack behaviors, the authors created a database containing 439 jihadist attackers active in eight Western countries in between 2014 and 2020. Of these, 54% were radicalized mostly offline, 18% online, and for 9%, a mix of online and offline influences could be established. Online radicalization typically came with social interaction. Only 2% of the sample conformed to a pathway labeled asocial online radicalization. A radicalization pathway could not be established for 17% of all cases.

When it comes to threat levels, those radicalized offline showed a three times higher likelihood of successful attack completion when compared to those radicalized online (Hamid and Ariza, 2022). The only exception were the few cases of asocial online radicalization; for these, successful attack completion was 2.5 times more likely than for the offline group. The severity of outcomes was likewise related to radicalization and showed more severe outcomes when offline processes were involved. Online radicalization, both social and asocial, did not play a role for attacks with more than 20 people injured or more than 5 attack casualties, in contrast to offline or hybrid radicalization.

So far, the findings discussed are based on publicly available data, often involving carefully maintained open data bases, but also integrating media reports, court proceedings, and other documentation. In contrast, our own work on extremist offenders in the United Kingdom, England and Wales specifically, is based on closed-source data generated within the Prison and Probation Service (HMPPS; Kenyon et al., 2021b, 2022a,b). A data set was generated by coding Extremism Risk Guidance Reports (ERG22+; Lloyd and Dean, 2015; National Offender Management Service, 2017), together with two Structured Risk Guidance Reports, an earlier version of the ERG22+ report, covering cases across a range of causes and ideologies. These reports constitute detailed accounts of an offender’s background and radicalization journey prior to the offense. In the majority of cases, offender interviews form part of the basis for the ERG22+ reports although a range of other restricted and more freely accessible sources get consulted, e.g., court reports, police reports, sentencing remarks, prison intelligence reports, among others.

Importantly, the detailed accounts are supplemented by formalized risk assessments on a total of 22 variables. These are aggregated to represent three different dimensions of the risk, or threat, that an offender poses: engagement, intent and capability. Reports and assessments are generated by HMPPS professionals who have undergone a specialized training program. Thus, the ERG22+ reports constitute one of the few standardized sources that allow for a triangulation of radicalization pathway, offense characteristics and current levels of risk and threat.

Within a total of 269 case reports, all related to the Terrorism Act, 235 cases of radicalized extremists could be identified. These conformed to the definition by Silke (2014): there was evidence they had held extremist views prior to coming into custody and that they had engaged in extremist activity outside prison. Of those, 12% had been radicalized primarily online, 40% primarily offline, and 48% through a mix of influences. Online radicalization coincided with a greater likelihood of mental illness or personality disorder being present as well as a lower degree of social connection with other extremists offline (63% were classed as lone) when compared to the other two categories. Further, online radicalization was characterized by a lower likelihood of being in an attacker role compared to radicalization offline.

In terms of risk assessments, online radicalization came with the lowest level of risk on all three dimensions. Engagement, as defined in the ERG22+, refers to a growing interest or identification with an extremist ideology or any group in support of such an ideology. Only 32% of the online group were classed as highly engaged, in comparison to 67% in the mixed group and 50% in the offline group. Intent refers to future readiness to overcome inhibitions and take action by committing offenses on behalf of the group or cause. Here, 15% of the online group were classed as high, with 48% in the mixed group and 36% in the offline group. Finally, capability encompasses knowledge, skills, networks and the general training necessary for carrying out terrorist acts. The online group showed significant (i.e., highest) levels in only 5% of all cases while for the offline group this figure was 41% and for the mixed group 22%.

In sum, the prevalence rate of online radicalization, in particular in the decade from 2010 and 2020, stands roughly at 12%–35% within a wider population of terrorists. This range is derived by looking across Western countries and somewhat differently defined populations. While Reinares et al. (2017) and Hamid and Ariza (2022) focus on jihadist terrorists, the former with a focus on individuals actually apprehended, the latter with a focus on terrorist attacks on record, Kenyon et al. (2022a,b) work with information on incarcerated offenders covering a wider spectrum of ideological backgrounds. It should be noted that the label “online” here refers to instances where the clear dominance or near exclusivity of online processes could be established with sufficient confidence. If mixed forms of radicalization were included, prevalence figures would be higher (as, for example, in Bastug et al., 2020) although it would then no longer be warranted to assign a driving force to Internet technologies.

Individuals radicalized online do not typically pose the highest level of threat. Considering the few successful attacks identified by Hamid and Ariza (2022) and the few individuals attributed high levels of threat, in particular any significant levels of capability, in Kenyon et al. (2022a,b), it seems that substantially dangerous individuals constitute no more than 2% in populations of Western-based terrorist actors. These figures need to be taken with great caution, given the scarcity of quantifiable findings. They do resonate, however, with the low threat levels found previously for lone actors (e.g., Gill et al., 2017), although findings also show clearly that the overlap between lone actors and those radicalized online is far from complete.

Opposing narratives compatible with the evidence

The review in the preceding section exemplifies, first and foremost, the substantial challenges that come with any attempt at quantifying the extent and outcomes of online radicalization. Equally challenging, however, is an appraisal and interpretation of the outcomes of quantification. In this section, this point will be developed through the presentation of two opposing narratives. In contrast to section Evidence on prevalence of online radicalization and associated threat, the purpose here is to outline a more holistic perspective rather than to reference again relevant literature.

In the first narrative, online radicalization is seen as posing a low threat by itself. While Internet activity has increased over time in all studies reviewed, this often seems to be attributable to the wider spread of Internet technologies and Internet use in society. In particular, the rise of social media among terrorist actors is closely mirrored by the global development these platforms have seen. Mixed modes of online and offline radicalization emerge as a standard model, and most studies have assigned a reinforcing, facilitating, possibly accelerating role to the Internet in there (Gill et al., 2015; Jensen et al., 2018). When considering mutually exclusive radicalization pathways, captured as solely online, solely offline, or combined, the Internet emerges still much more as an enabler, rather than a driver on its own.

When viewing online radicalization as a specific radicalization pathway, threat levels appear even lower. In our work to date (Kenyon et al., 2022a), online radicalization does coincide with an offender type that is socially isolated, more prone to mental illness and associated conditions, and less likely to commit acts of violence. This type is assessed as low in engagement with extremist ideologies, or groups representing such ideologies, and further shows lowest levels of intent and low capability compared to other radicalization pathways. Other work has also highlighted a tendency for information leakage online, and where this is prior to an offense, it can help to thwart attacks (Gill et al., 2017; Kenyon et al., 2021a; Hamid and Ariza, 2022). This has led to the view that those radicalized online are comparatively powerless, in particular when it comes to translating online activity into offline violence (Conway, 2016). As far as the evidence goes, successful attackers who have radicalized primarily online are very rare when compared to any wider extremist offender population. As far as convicted individuals within the Western world are concerned, online radicalization and violence do not share a strong or direct link.

The second narrative assigns a much higher threat level to online radicalization. A few individuals on this radicalization pathway manage to commit acts of violence, and these may be particularly difficult to detect when leakage does not occur, fails to trigger a security response or is intentional and helps to enhance the effectiveness of an attack [see Hamm and Spaaij (2017) on the distinction between intentional and unintentional information disclosure online]. There is also the possibility that many online-only offenders are simply at an earlier stage in their pathway toward violence-endorsing extremism (in comparison to those who have already forged stronger social connections). In our work, 32% were highly engaged with extremist causes and groups while 15% showed high levels of intent regarding future offending. It is therefore hard to conclude that online radicalization results in, more or less, harmless offenders.

In terms of overall prevalence, the studies that offer a breakdown over time (Jensen et al., 2018; Hamid and Ariza, 2022; Kenyon et al., 2022b) indicate that exclusive or predominant online radicalization has been on the rise until recently, and most likely still is. Although the percentages are comparatively low, they are markedly above zero. Indeed, most of the discussion surrounding online radicalization follows a particular logic whereby risks and threats are described as comparatively low, not negligible. In this context, it also needs to be considered that mixed forms of radicalization, involving some form of online activity, are becoming the norm. For these offenders, our previous work (Kenyon et al., 2022a) has shown some of the highest levels of risk: 67% on this radicalization pathway showed high levels of engagement and 48% high levels of intent.

Finally, online radicalization and non-violent (online) offending are still likely to encourage and endorse violence and contribute to the perpetuation of an online culture of extremist beliefs, stabilizing a grievance-based climate that carries the ongoing potential of encouraging acts of violence in others. Given the inherently global outreach of the Internet, this may be one of the strongest arguments to take online influences seriously. The prevalence of hate speech and related materials online can be deemed high, and recent studies in this area have shown that occasional encounters with such online content are experienced by 40% to 50% of younger individuals (Saha et al., 2019; Costello et al., 2020). As such there are constant opportunities for the initiation of further radicalization processes within large populations, and no suggestion of any downward trend.

When considering again the empirical evidence for these opposing narratives, it is noteworthy that the few studies that allow for some quantification differ markedly on a range of important dimensions, yet converge, by and large, in their findings. The underlying data sources are mostly openly accessible, with the exception of our work on reports held by HMPPS, the United Kingdom penal system. Looking at a prison-based population comes with several restrictions. Only terrorist actors that are both apprehended and sentenced will undergo this risk assessment. This is in contrast to comprehensive databases of all known terrorist acts. These are incident-based and therefore likely to register more actors. Another variation concerns the way studies can address radicalization as a process and estimate threat. This ranges from assessing the effectiveness and severity of the offense to prospective risk assessments of the individual. Some studies have focused exclusively on jihadi-inspired actors whereas others have covered the whole spectrum of what falls under some definition of extremist ideology. All studies, however, focus on actors based in the Western World, i.e., the United States, Western Europe, and Australia.

The two opposing narratives outlined so far can also be linked to different courses of action regarding prevention and counter-terrorism measures. When focusing on relative risk, any allocation of resources for prevention needs to consider that online radicalization does, at present, not constitute a main source of threat. When focusing on absolute risk, it is crucial to note that exclusive online radicalization does occur, for a non-trivial proportion of all terrorist actors and incidents, and constitutes one established and growing pathway for terrorist activities.

Current limitations and outlook

Online radicalization seems to pose a manageable risk. This evidence is based, however, chiefly on data that falls in the decade from 2010 to 2020. When assessing the present state of affairs and attempting to forecast future developments, a number of unknowns need to be taken into account. The first concerns any effects due to the global COVID-19 pandemic, which since 2020 has altered the modes of work and socializing for large parts of the world population and has increased online activities in many domains of life (Feldmann et al., 2020). Concerns over the effects of the pandemic on extremism are high, and first evidence shows that online extremism may have increased in particular for grievance-based ideologies (Davies et al., 2021). It remains to be seen whether the pandemic has changed and steepened the growth trajectory for online radicalization, but there is at the very least a substantial risk of acceleration.

Another unknown factor concerns, by necessity, the ongoing evolution of the Internet. This concerns both the functionality and accessibility of technologies and their relevance to terrorist activity. For example, adding a virtual reality component to training units could increase the capability of online actors once the technology has become more of a standard for wider populations of users. Further, developments such as the Internet of Things could provide novel forms of both radicalization and attacks (Henschke, 2021; Sullivan and Montasari, 2022). These developments also extend to easily accessible attack equipment such as weaponry generated through 3D-printing and promoted on social media (The Guardian, June, 2022).

A final unknown noted here concerns the uncertainty over effect sizes for any indirect harm caused by the perpetuation of extremist online networks and extremist online culture. As noted above, the Internet provides the mechanisms for radicalization and the opportunities for encountering relevant content (Magdy et al., 2016; Hassan et al., 2018; Costello et al., 2020; Smith et al., 2020; Saha et al., 2019). Those radicalized online can therefore have effects on others by the endorsement and spreading of propaganda and similar content. A quantification of such indirect harm, however, seems exceedingly difficult at present.

Lastly, the question of generalizability of terrorism-related findings to other forms of grievance-based violence needs revisiting. Many of the considerations in the present work are not confined to terrorism, but can be extended to other forms of grievance-based offending. Evidence on the online radicalization process does not, generally speaking, presuppose any specifics in the domain of terrorism, and findings on information leakage and general offender characteristics are, as pointed out in the beginning, very similar across different forms of grievance-fueled violence (Capellan, 2015; Kenyon et al., 2021a; Clemmow et al., 2022). There is still, however, careful scrutiny required since terrorist offending can comprise much more than immediate acts of violence.

The data bases used to establish prevalence and threat related to online radicalization contain not only violent attackers, but a multitude of roles including supporters, facilitators, recruiters, propagandists and so forth. While upper threat estimates derived from the data are related, by definition, to violent attackers, other roles can still pose a substantial danger to society. In addition, the links to grievance may not be as strong in the context of terrorism as they are for related forms of violence. The most pertinent example concerns the engagement dimension in the ERG22+ reports analyzed in our research (Kenyon et al., 2022a). Engagement, within this framework of risk assessment, can refer to immersion in an extremist ideology as much as it can refer to group identification and attachment. While the first type of engagement would imply a strongly held grievance-based belief set, the second may be more a matter of social influence, peer pressure and a need for belonging. Again, this means that terrorism is, on a practical level, treated as a more broadly defined category and shows only partial overlap to the category of grievance-fueled violence.

To return to our initial question, within the domain of terrorism, online radicalization, as a process dominated or entirely guided by Internet-related activity, does occur and poses a discernible threat, although both prevalence and threat level have so far been lower in comparison to other forms of radicalization.

Author contributions

All authors listed have made a substantial, direct, and intellectual contribution to the work and approved it for publication.

Conflict of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

References

• Bastug M. F., Douai A., Akca D. (2020). Exploring the “demand side” of online radicalization: evidence from the Canadian context. Stud. Conflict Terror. 43, 616–637. doi: 10.1080/1057610X.2018.1494409 [CrossRef] [Google Scholar]
• Brooks N., Shaw R. (2022). Fixated and grievance-fuelled persons: Considerations on the dangers of gaps, silos and disconnects. Psychiatry Psychol. Law, 1–17. doi: 10.1080/13218719.2021.1995518 [PMC free article] [PubMed] [CrossRef] [Google Scholar]
• Capellan J. A. (2015). Lone wolf terrorist or deranged shooter? A study of ideological active shooter events in the United States, 1970–2014. Stud. Conflict Terror. 38, 395–413. doi: 10.1080/1057610X.2015.1008341 [CrossRef] [Google Scholar]
• Clemmow C., Gill P., Bouhana N., Silver J., Horgan J. (2022). Disaggregating lone-actor grievance-fuelled violence: comparing lone-actor terrorists and mass murderers. Terror. Polit. Viol. 34, 558–584. doi: 10.1080/09546553.2020.1718661 [CrossRef] [Google Scholar]
• Clifford B., Powell H. (2019). Encrypted extremism: Inside the English-speaking Islamic state ecosystem on telegram. Washington DC: George Washington University. [Google Scholar]
• Conway M. (2016). Determining the role of the internet in violent extremism and terrorism: six suggestions for progressing research. Stud. Conflict Terror. 40, 77–98. doi: 10.1080/1057610X.2016.1157408 [CrossRef] [Google Scholar]
• Conway M., McInerney L. (2008). Jihadi video and auto-radicalization: Evidence from an exploratory YouTube study. In European Conference on Intelligence and Security Informatics (pp. 108–118). Heidelberg: Springer.
• Costello M., Barrett-Fox R., Bernatzky C., Hawdon J., Mendes K. (2020). Predictors of viewing online extremism among America’s youth. Youth Soc. 52, 710–727. doi: 10.1177/0044118X18768115 [CrossRef] [Google Scholar]
• Davies G., Wu E., Frank R. (2021). A witch’s brew of grievances: the potential effects of COVID-19 on radicalization to violent extremism. Stud. Conflict Terror., 1–24. doi: 10.1080/1057610X.2021.1923188 [CrossRef] [Google Scholar]
• Ebbrecht C. K. (2022). Systematic review: risk factors and mechanisms of radicalization in lone-actor grievance-fueled violence. Nordic Psychol., 1–35. doi: 10.1080/19012276.2022.2074524 [CrossRef] [Google Scholar]
• Evans A. T., Williams H. J. (2022). How extremism operates online. RAND Corporation. Available at: https://www.rand.org/pubs/perspectives/PEA1458-2.html
• Feldmann A., Gasser O., Lichtblau F., Pujol E., Poese I., Dietzel C., et al.. (2020). The lockdown effect: Implications of the COVID-19 pandemic on internet traffic. In Proceedings of the ACM Internet Measurement Conference (pp. 1–18). ACM.
• Frissen T. (2021). Internet, the great radicalizer? Exploring relationships between seeking for online extremist materials and cognitive radicalization in young adults. Comput. Hum. Behav. 114:106549. doi: 10.1016/j.chb.2020.106549 [CrossRef] [Google Scholar]
• Gill P. (2015). Lone actor terrorists: A behavioural analysis. Oxon: Routledge. [Google Scholar]
• Gill P., Corner E., Conway M., Thornton A., Bloom M., Horgan J. (2017). Terrorist use of the internet by the numbers: quantifying behaviors, patterns, and processes. Criminol. Public Policy 16, 99–117. doi: 10.1111/1745-9133.12249 [CrossRef] [Google Scholar]
• Gill P., Corner E., Thornton A., Conway M. (2015). What are the roles of the internet in terrorism? Measuring online behaviours of convicted UK terrorists. EU FP7 VOX-Pol report. Available at: http://voxpol.eu/what-are-the-roles-of-the-internet-interrorism.
• Gunaratna R., Oreg A. (2010). Al Qaeda’s organizational structure and its evolution. Stud. Conflict Terror. 33, 1043–1078. doi: 10.1080/1057610X.2010.523860 [CrossRef] [Google Scholar]
• Hamid N., Ariza C. (2022). Offline versus online radicalization: which is the bigger threat? Tracing outcomes of 439 jihadist terrorists between 2014–2021 in 8 Western countries. GNET report. Available at: https://gnet-research.org/wp-content/uploads/2022/02/GNET-Report-Offline-Versus-Online-Radicalization.pdf
• Hamm M. S., Spaaij R. (2017). The age of lone wolf terrorism. New York: Columbia University Press. [Google Scholar]
• Hassan G., Brouillette-Alarie S., Alava S., Frau-Meigs D., Lavoie L., Fetiu A., et al.. (2018). Exposure to extremist online content could lead to violent radicalization: a systematic review of empirical evidence. Int. J. Dev. Sci. 12, 71–88. doi: 10.3233/DEV-170233 [CrossRef] [Google Scholar]
• Henschke A. (2021). “Terrorism and the internet of things: cyber-terrorism as an emergent threat,” in Counter-terrorism, ethics and technology. eds. Henschke A., Reed A., Robbins S., Miller S. (Cham: Springer; ), 71–87. [Google Scholar]
• Herath C., Whittaker J. (2021). Online radicalization: moving beyond a simple dichotomy. Terror. Polit. Viol., 1–22. doi: 10.1080/09546553.2021.1998008 [CrossRef] [Google Scholar]
• Hoffman B. (2008). The myth of grass-roots terrorism-why Osama bin Laden still matters. Foreign Aff. 87, 133–138. [Google Scholar]
• Jensen M., James P., LaFree G., Safer-Lichtenstein A., Yates E. (2018). The use of social media by United States extremists. National Consortium for the Study of Terrorism and Responses to Terrorism (START).
• Kenyon J., Baker-Beall C., Binder J. (2021a). Lone-actor terrorism: a systematic literature review. Stud. Conflict Terror. 1–24. doi: 10.1080/1057610X.2021.1892635 [CrossRef] [Google Scholar]
• Kenyon J., Binder J., Baker-Beall C. (2021b). Exploring the role of the internet in radicalization and offending of convicted extremists. Available at: https://www.gov.uk/government/publications/exploring-the-role-of-the-internet-in-radicalization-and-offending-of-convicted-extremists
• Kenyon J., Binder J. F., Baker-Beall C. (2022a). Online radicalization: Profile and risk analysis of individuals convicted of extremist offences. Legal Criminol. Psychol. 1–17. doi: 10.1111/lcrp.12218 [CrossRef] [Google Scholar]
• Kenyon J., Binder J. F., Baker-Beall C. (2022b). Understanding the role of the internet in the process of radicalization: an analysis of convicted extremists in England and Wales. Stud. Conflict Terror., 1–25. doi: 10.1080/1057610X.2022.2065902 [CrossRef] [Google Scholar]
• Klausen J. (2015). Tweeting the Jihad: social media networks of Western foreign fighters in Syria and Iraq. Stud. Conflict Terror. 38, 1–22. doi: 10.1080/1057610X.2014.974948 [CrossRef] [Google Scholar]
• Lloyd M., Dean C. (2015). The development of structured guidelines for assessing risk in extremist offenders. J. Threat Assess. Manag. 2, 40–52. doi: 10.1037/tam0000035 [CrossRef] [Google Scholar]
• Macdonald S., Grinnell D., Kinzel A., Lorenzo-Dus N. (2019). Daesh, twitter and the social media ecosystem: a study of outlinks contained in tweets mentioning Rumiyah. RUSI J. 164, 60–72. doi: 10.1080/03071847.2019.1644775 [CrossRef] [Google Scholar]
• Macdonald S., Whittaker J. (2019). “Online Radicalization: Contested Terms and Conceptual Clarity”, in Online Terrorist Propaganda, Recruitment, and Radicalization. ed. J. R. Vacca (Boca Raton: CRC Press; ). [Google Scholar]
• Magdy W., Darwish K., Abokhodair N., Rahimi A., Baldwin T. (2016). #isisisnotislam or #deportallmuslims? Predicting unspoken views. In Proceedings of the 8th ACM Conference on Web Science (pp. 95–106). ACM.
• Meleagrou-Hitchens A., Kaderbhai N. (2017). Research perspectives on online radicalization: a literature review, 2006–2016, vol. 19. Dublin: International Centre for the Study of Radicalization, 1–98. [Google Scholar]
• Mølmen G. N., Ravndal J. A. (2021). Mechanisms of online radicalization: how the internet affects the radicalization of extreme-right lone actor terrorists. Behav. Sci. Terror. Polit. Aggress., 1–25. doi: 10.1080/19434472.2021.1993302 [CrossRef] [Google Scholar]
• National Offender Management Service (2017). ERG22+ & extremism risk screen: Summary and overview. London: Ministry of Justice. [Google Scholar]
• Neo L. S. (2019). “An internet-mediated pathway for online radicalization: RECRO,” in Combating violent extremism and radicalization in the digital era. eds. Khader M., Neo L. S., Ong G., Tan E., Chin J. (Hershey, PA: IGI Global; ), 197–224. [Google Scholar]
• Neumann P. R. (2013). Options and strategies for countering online radicalization in the United States. Stud. Confl. Terror. 36, 431–459. doi: 10.1080/1057610X.2013.784568 [CrossRef] [Google Scholar]
• O’Callaghan D., Greene D., Conway M., Carthy J., Cunningham P. (2014). Down the (white) rabbit hole: the extreme right and online recommender systems. Soc. Sci. Comput. Rev. 33, 459–478. doi: 10.1177/0894439314555329 [CrossRef] [Google Scholar]
• Pariser E. (2011). The filter bubble: What the internet is hiding from you. London: Penguin Books. [Google Scholar]
• Reed A., Whittaker J., Votta F., Looney S. (2019). Radical filter bubbles: Social media personalization algorithms and extremist content. London: Global Research Network on Terrorism and Technology. [Google Scholar]
• Reinares F., Garcia-Calvo C., Vicente Á. (2017). Differential association explaining jihadi radicalization in Spain: a quantitative study. CTC Sentinel 10, 29–34. [Google Scholar]
• Rothut S., Schulze H., Hohner J., Greipl S., Rieger D. (2022). Radikalisierung im Internet: Ein systematischer Überblick über Forschungsstand, Wirkungsebenen sowie Implikationen für Wissenschaft und Praxis. Bonn: Core-NRW. [Google Scholar]
• Sageman M. (2004). Understanding terror networks. Philadelphia, PA: Pennsylvania University Press. [Google Scholar]
• Sageman M. (2008). Leaderless jihad: Terror networks in the 21st century. Philadelphia, PA: University of Pennsylvania Press. [Google Scholar]
• Saha K., Chandrasekharan E., De Choudhury M. (2019). Prevalence and psychological effects of hateful speech in online college communities. In Proceedings of the 10th ACM conference on web science (pp. 255–264). [PMC free article] [PubMed]
• Silke A. (2014). “Risk assessment of terrorist and extremist prisoners,” in Prisons, terrorism and extremism: Critical issues in management, radicalization and reform. ed. Silke A. (London: Routledge; ), 108–121. [Google Scholar]
• Smith L. G., Wakeford L., Cribbin T. F., Barnett J., Hou W. K. (2020). Detecting psychological change through mobilizing interactions and changes in extremist linguistic style. Comput. Hum. Behav. 108:106298. doi: 10.1016/j.chb.2020.106298 [CrossRef] [Google Scholar]
• Speckhard A., Ellenberg M., Morton J., Ash A. (2021). Involuntary celibates’ experiences of and grievance over sexual exclusion and the potential threat of violence among those active in an online incel forum. J. Strat. Secur. 14, 89–121. doi: 10.5038/1944-0472.14.2.1910 [CrossRef] [Google Scholar]
• Sullivan A., Montasari R. (2022). “The use of the internet and the internet of things in modern terrorism and violent extremism,” in Privacy, security and forensics in the internet of things (IoT). eds. Montasari R., Carroll F., Mitchell I., Hara S., Bolton-King R. (Cham: Springer; ), 151–165. [Google Scholar]
• The Guardian (2022). Meta banned firearms sales. Why are they still available on Facebook and Instagram? Available at: https://www.theguardian.com/technology/2022/jun/15/meta-firearms-sales-facebook-instagram-ghost-guns
• UK Government (2018). CONTEST: The United Kingdom’s strategy for countering terrorism (Revised June 2018). Available at: https://www.gov.uk/government/publications/counter-terrorism-strategy-contest-2018
• UK Home Office (2019). Online harms white paper. Available at: https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/793360/Online_Harms_White_Paper.pdf
• Valentini D., Lorusso A. M., Stephan A. (2020). Onlife extremism: dynamic integration of digital and physical spaces in radicalization. Front. Psychol. 11:524. doi: 10.3389/fpsyg.2020.00524, PMID: [PMC free article] [PubMed] [CrossRef] [Google Scholar]
• Von Behr I., Reding A., Edwards C., Gribbon L. (2013). Radicalization in the digital era: the use of the internet in 15 cases of terrorism and extremism. RAND Corporation. Available at: https://www.rand.org/pubs/research_reports/RR453.html
• Weimann G. (2006). Terror on the internet: The new arena, the new challenges. Washington, D.C.: US Institute of Peace Press. [Google Scholar]
• Weimann G. (2014). New terrorism and new media (Vol. 2). Washington, D.C.: Commons Lab of the Woodrow Wilson International Center for Scholars. [Google Scholar]
• Whittaker J. (2021). The online behaviors of Islamic state terrorists in the United States. Criminol. Public Policy 20, 177–203. doi: 10.1111/1745-9133.12537 [CrossRef] [Google Scholar]
• Whittaker J. (2022). Rethinking online radicalization. Perspectives on. Terrorism 16, 27–40. [Google Scholar]
• Wolfowicz M., Hasisi B., Weisburd D. (2022). What are the effects of different elements of media on radicalization outcomes? A systematic review. Campbell Syst. Rev. 18:e1244. doi: 10.1002/cl2.1244 [CrossRef] [Google Scholar]

---

Articles from Frontiers in Psychology are provided here courtesy of Frontiers Media SA

---

Extremist ideology as a complex contagion: the spread of far-right radicalization in the United States between 2005 and 2017

Youngblood, Mason

nature.com

---

Introduction

The far-right movement, which includes white supremacists, neo-Nazis, and sovereign citizens, is the oldest and most deadly form of domestic extremism in the United States (Piazza, 2017; Simi and Bubolz, 2017). Despite some ideological diversity, members of the far-right often advocate for the use of violence to bring about an “idealized future favoring a particular group, whether this group identity is racial, pseudo-national, or characterized by individualistic traits” (National Consortium for the Study of Terrorism and Responses to Terrorism (START), University of Maryland, 2017). Over the last decade, the far-right movement was responsible for 73.3% of all extremist murders in the United States. In 2018, this statistic rose to 98% (The Anti-Defamation League Center on Extremism, 2019). The increasing severity of far-right extremist violence, as well as the associated rhetoric on social media (Davey and Ebner, 2019; Winter, 2019), has generated public concern about the spread of radicalization in the United States. Former extremists have referred to it as a public health issue (Allam, 2019; Bonn, 2019), an idea advocated for by some policy experts as well (Sanir et al., 2017; Weine and Eisenman, 2016).

There is little evidence that radicalization is primarily driven by psychopathology (Misiak et al., 2019; Post, 2015; Webber and Kruglanski, 2017). Rather, radicalization appears to be a process in which individuals are destabilized by various environmental factors, exposed to extremist ideology, and subsequently reinforced by members of their community (Becker, 2019; Jasko et al., 2017; Jensen et al., 2018; Mills et al., 2019; Webber and Kruglanski, 2017). Even “lone wolves”, or solo actors, often interact with extremist communities online (Holt et al., 2019; Kaplan et al., 2014; Post, 2015). As such, radicalization may spread through a social contagion process, in which extremist ideologies behave like complex contagions that require multiple exposures for adoption (Guilbeault et al., 2018), which has been observed for political movements more broadly (González-Bailón et al., 2011). Previous research suggests that extremist propaganda (Ferrara, 2017), hate crimes (Braun, 2011; Braun and Koopmans, 2010), intergroup conflict (Buhaug and Gleditsch, 2008; Gelfand et al., 2012), and terrorism (Cherif et al., 2009; LaFree et al., 2012; Midlarsky et al., 1980; White et al., 2016) exhibit similar dynamics.

The environmental factors favoring radicalization, referred to here as endemic factors, include variables like poverty rate that may influence an individual’s risk of adoption in particular regions. As such, endemic factors have the potential to enhance or constrain the spread of contagions through populations across geographic space. Although significant research has been done on how endemic factors predict radicalization (and resulting violence) (Goetz et al., 2012; LaFree and Bersani, 2014; McVeigh, 2004; Medina et al., 2018; Piazza, 2017), few studies have investigated how these factors interact with contagion processes. The aim of this study is to determine whether patterns of far-right radicalization in the United States are consistent with a contagion process, and to assess the influence of critical endemic factors. After controlling for population density, I assessed the following endemic factors that have been implicated in previous research on radicalization, extremism, and mass shootings: poverty rate (Durso and Jacobs, 2013; Gale et al., 2002; Kwon and Cabrera, 2019b; Lin et al., 2018; Medina et al., 2018; Piazza, 2017; Suttmoeller et al., 2015, 2016, 2018), unemployment rate (Espiritu, 2004; Gale et al., 2002; Goetz et al., 2012; Green et al., 1998; Jefferson and Pryor, 1999; Majumder, 2017; Pah et al., 2017; Piazza, 2017), income inequality (Goetz et al., 2012; Kwon and Cabrera, 2017, 2019a, b; Majumder, 2017; McVeigh, 2004; McVeigh and Cunningham, 2012), education levels (Durso and Jacobs, 2013; Espiritu, 2004; Florida, 2011; Gladfelter et al., 2017; Kwon and Cabrera, 2017, 2019a; McVeigh et al., 2014), non-white population size (Gladfelter et al., 2017; LaFree and Bersani, 2014; McVeigh, 2004; Medina et al., 2018), violent crime rate (Gladfelter et al., 2017; McVeigh and Cunningham, 2012; Sweeney and Perliger, 2018), gun ownership (Anisin, 2018; Lin et al., 2018; Pah et al., 2017; Reeping et al., 2019), hate groups per capita (Adamczyk et al., 2014), and Republican voting (McVeigh et al., 2014; Medina et al., 2018). Furthermore, I aim to determine whether individual-level variables, such as social media use, enhance the spread of far-right radicalization over space and time. Social media platforms increasingly appear to play a role in radicalization, both as formal recruitment tools (Aly et al., 2017; Awan, 2017, Bertram, 2016; Wu, 2015) and spaces for extremist communities to interact (Amble, 2012; Dean et al., 2012; Pauwels et al., 2014; Winter, 2019). If social media platforms augment physical organizing (Bowman-Grieve, 2009; Holt et al., 2016), then they may also enhance the spread of radicalization.

Although social media platforms relax geographic constraints on communication, evidence suggests that social media networks still exhibit spatial clustering. For example, the majority of an individual’s Facebook friends live within 100 miles of them (Bailey et al., 2018), the probability of information diffusion on social media decays with increasing distance (Liu et al., 2018), and online echo chambers map onto particular locations (Bastos et al., 2018). Since complex contagions require reinforcement, and the majority of online friendship ties are within a close radius, the diffusion of extremist ideologies online should still exhibit some level of geographic bias. This idea is supported by evidence that social media enhances physical organizing among extremists (Bastug et al., 2018; Gill et al., 2017; von Behr et al., 2013), and anecdotes of “self-radicalized” individuals using social media to contact other extremists in their area (Holt et al., 2019).

In order to model the spread of far-right radicalization I used a two-component spatio-temporal intensity (twinstim) model (Meyer et al., 2017), an epidemiological method that treats events in space and time as resulting from self-exciting point processes (Reinhart, 2018). In this framework, future events depend on the history of past events within a certain geographic range. Event probabilities are determined by a conditional intensity function, which is separated into endemic and epidemic components. This allows researchers to assess the combined effects of both spatio-temporal covariates and epidemic predictors. Epidemic, in this framework, refers to any level of contagion effect and does not necessarily imply uncontrollable spread. With a couple of notable exceptions (Clark and Dixon, 2018; Zammit-Mangion et al., 2012), previous applications of self-exciting point process models in terrorism and mass shooting research have not simultaneously modeled diffusion over both time and space (Collins et al., 2020; Garcia-Bernardo et al., 2015; Johnson and Braithwaite, 2017; Lewis et al., 2012; Porter and White, 2010; Tench et al., 2016; Towers et al., 2015; White et al., 2013).

The radicalization events in this study, which correspond to where and when a radicalized individual’s extremist activity or plot was exposed, came from the Profiles of Individual Radicalization in the United States (PIRUS), an anonymized database compiled by the National Consortium for the Study of Terrorism and Responses to Terrorism (START) (National Consortium for the Study of Terrorism and Responses to Terrorism (START), University of Maryland, 2017). PIRUS is compiled from sources in the public record, and only includes individuals radicalized in the United States who were either arrested, indicted, or killed as a result of ideologically-motivated crimes, or were directly associated with a violent extremist organization. I chose to use PIRUS instead of the Terrorism and Extremist Violence in the United States (TEVUS) database because events in PIRUS are disambiguated by individual and include social variables that may influence the diffusion process.

A contagion effect in this modeling framework could result from one of two forces. The first is a copy-cat effect, in which individuals copy behaviors observed directly or in the media. Although this effect has been proposed in terrorism and mass shooting research in the past (Nacos, 2009; Towers et al., 2015), it seems to be a more plausible contagion mechanism for specific methods of violence (Helfgott, 2015) (e.g., suicide bombings (Tominaga, 2018)) rather than radicalization more broadly. The second is linkage triggered by activism and organizing, or ideologically-charged events (e.g., elections, demonstrations, policies), in that region. To differentiate between these two forces, I included two sets of epidemic predictors in the modeling. The first two event-level variables, plot success and anticipated fatalities, might be expected to increase epidemic probability if a copy-cat effect is present. This is because successful large-scale events are probably more contagious due to increased media coverage (Towers et al., 2015). Alternatively, the two individual-level variables, group membership and social media use, might be expected to increase epidemic probability if activism and organizing drive the linkage between events.

Methods

Data collection

All individual-level data came from PIRUS. Only individuals with far-right ideology who were exposed during or after 2005 (the earliest year with social media data) with location data at the city-level or lower (n = 416; F: 6.0%, M: 94.0%) were included (see Figs. 1 and 2). For each individual, the date and location of their exposure (usually when their activity/plot occurred), whether their plot was successful (34.9%), the anticipated fatalities of their plot (0: 69.5%, 1–20: 26.0%, >20:2.6%, >100: 1.9%), whether they were a member of a formal or informal group of extremists (58.4%), and whether social media played a role in their radicalization (31.2%), were included. Unknown or missing values for each predictor (plot success: 0.5%, anticipated fatalities: 13.5%, group membership: 0%, social media: 54.8%) were coded as 0. To ensure that the coding procedure for missing predictor values did not introduce bias, I checked whether the results of the full model were consistent after multiple imputation with chained equations and random forest machine learning (see Table S1). The location of each exposure was geocoded from the nearest city or town using the R package ggmap (Kahle and Wickham, 2013). Since domestic terrorists tend to commit acts in their local area (Klein et al., 2017; Marchment et al., 2018; Smith et al., 2008), I assumed that exposure locations reflect where individuals were radicalized.

Fig. 1: The number of far-right extremists exposed in the PIRUS database between 2005 and 2017.

Number of cases is on the left, and cumulative number of cases is on the right.

Full size image

Fig. 2: The untied locations of far-right extremists exposed in the PIRUS database between 2005 and 2017.

The color of each county corresponds to its log-transformed population density.

Full size image

State-level gun ownership was estimated using a proxy measure based on suicide rates and hunting licenses (Seigel et al., 2014). Using data from 2001, 2002, and 2004 (the only three years for which state-level gun ownership data is available), Seigel et al. found that the following proxy correlates with gun ownership with an R² of 0.95:

$$\left(0.62\cdot \frac{{\mathrm{FS}}}{{\mathrm{S}}}\right)+(0.88\cdot {\mathrm{HL}})-0.0448$$

(1)

where FS/S is the proportion of suicides that involve firearms (from the Centers for Disease Control and Prevention, or CDC), and HL is hunting licenses per capita (from the United States Fish and Wildlife Service) (Seigel et al., 2014). Missing suicide rates (five years for DC, two years for Rhode Island) were replaced with the mean values for that state. State-level hate group data was collected from the Southern Poverty Law Center, while violent crime data was collected from the Federal Bureau of Investigation’s Uniform Crime Reporting Program.

County-level demographic data was collected from the US Census using the R package censusapi (Recht, 2019). This included population density, poverty rate, Gini index of income inequality, percentage of the population that is non-white, percentage of the population that has at least a high-school diploma, and unemployment rate. County-level income, race, education, and unemployment data is only available after 2009, so the 2010 data was used for 2005–2009. County-level presidential election voting records were collected from the Massachusetts Institute of Technology Election Lab, and non-election years were assigned the data from the most recent election year.

Geographic data was collected from the US Census using the R package tigris (Walker, 2019).

Model specification

Twinstim modeling was conducted using the R package surveillance (Meyer et al., 2017). To convert the data to a continuous spatio-temporal point process, all tied locations and dates were shifted in a random direction up to half of the minimum spatial and temporal distance between events (1.52 km and 0.5 days, respectively) (Meyer and Held, 2014).

Step functions were used to model both spatial and temporal interactions. Visual inspection of the pair correlation function for the point pattern indicates that the data is significantly clustered up to 400 km (see Supplementary Fig. 1). As such, the spatial step function was split into four 100 km intervals with 400 km as the maximum interaction radius (Nightingale et al., 2015). The temporal step function was split into four six-month intervals up to two years (based on the the high degree of variation in radicalization and attack planning times among domestic extremists (Bouhana et al., 2018; Silkoset, 2016; Smith and Damphousse, 2009)). I attempted the analysis with different combinations of power-law, Gaussian, and Student spatial functions, and exponential temporal functions, but these variations converged to unrealistically steep spatial and temporal interaction functions that approached zero around two km and two days, and appeared to be significantly influenced by the tie-breaking procedure (Meyer and Held, 2014).

Population density (county-level) was log-transformed and used as an offset endemic term. A centered time trend was also included to determine whether the strength of the endemic component has shifted over time. Poverty rate (county-level), Gini index of income inequality (county-level), gun ownership (state-level), percentage of the population that is non-white (county-level), percentage of the population that has at least a high-school diploma (county-level), unemployment rate (county-level), percentage of voters that vote Republican in presidential elections (county-level), violent crime rate per thousand residents (state-level), and number of hate groups per million residents (state-level) were included as dynamic endemic predictors that change annually. Plot success, anticipated fatalities, group membership, and social media radicalization were included as epidemic predictors.

All possible models with all possible combinations of predictors were run and ranked by Akaike’s Information Criterion (AIC) (Burnham and Anderson, 2002). The best fitting model with the lowest AIC was used to assess the effects of each variable on event probability. Rate ratios were calculated by applying exponential transformation to the model estimates.

Permutation test

To determine whether the spatio-temporal interaction of the epidemic component was statistically significant, I used the Monte Carlo permutation approach developed by Meyer et al. (Meyer et al., 2016). Using this approach, a twinstim model with all endemic predictors from the best fitting model and no epidemic predictors was compared to 1000 permuted null models with randomly shuffled event times. For each permutation I estimated the reproduction number (R₀), or the expected number of future events that an event triggers on average, which represents “infectivity”. A p-value was calculated by comparing the observed R₀ with the null distribution of the subset of permutations that converged.

For additional support, I also ran a likelihood ratio test and a standard Knox test of spatio-temporal clustering. The Knox test was conducted with spatial and temporal radii of 100 km and six months (the upper bounds of the first steps in the step functions), respectively (Knox and Bartlett, 1964).

Simulations

To further assess the quality of the model, I conducted simulations from the cumulative intensity function using Ogata’s modified thinning algorithm according to Meyer et al. (2012). Using the parameters of the best fitting model, I conducted 1,000 simulations of the last six months of the study period and compared the results to the observed data.

Results

The results of the best fitting model (ΔAIC < 2), which included seven endemic and two epidemic predictor variables, are shown in Table 1. Firstly, there is a statistically significant time trend whereby the endemic rate decreases by 4.6% each year, indicating that the strength of the epidemic component has increased over time. There appears to be a baseline increase in the endemic component between 2008-2012 which likely corresponds to the financial crisis (Funke et al., 2016), as well as a significant spike in the epidemic component around 2016 which likely corresponds to the presidential election (Giani and Meón, 2019; Rushin and Edwards, 2018) (Fig. 3). There are also significant positive effects of poverty rates (p < 0.01) and the presence of hate groups (p < 0.0001) on radicalization probability. Interestingly, the percentage of voters that vote Republican in presidential elections (p < 0.0001), the percentage of the population that is non-white (p < 0.05), and unemployment rates (p < 0.0001) appear to have significant negative effects on radicalization probability. Gun ownership, education level, and violent crime all have no significant effect on radicalization probability. When Republican voting was replaced with the absolute percent difference between Republican and Democratic voting, a proxy measure for the competitiveness of elections, it was no longer significant. A variance inflation factor test identified no multicollinearity problems among the time-averaged endemic predictors (VIF < 3) (Zuur et al., 2010).

Table 1 The results of the twinstim modeling, with estimated rate ratios (RR), Wald confidence intervals, and p-values.

Full size table

Fig. 3: The total intensity (in black), as well as the isolated endemic component (grey), over time.

Total intensity can be interpreted as the proportion of radicalization probability that is explained by the endemic and epidemic components.

Full size image

Both group membership and radicalization via social media have strong and significant positive effects on epidemic probability. Exposures of individuals who belong to formal or informal extremist groups are over four times more likely to be followed by future exposures in close spatial or temporal proximity (p < 0.01). Similarly, exposures of individuals radicalized on social media are almost three times as likely to be followed by future exposures (p < 0.01). Anticipated fatalities and plot success did not appear in the best fitting model. Estimates of the decaying spatial and temporal interaction functions, as well as model diagnostics, can be seen in Supplementary Figs. 2 and 3, respectively. A variance inflation factor test identified no multicollinearity problems among the epidemic predictors (VIF < 3) (Zuur et al., 2010).

Based on the permutation test, the observed R₀ (0.31) is significantly higher than the null distribution of the converged permutations (N_conv= 739, p < 0.01) (Fig. 4). This indicates that the spatio-temporal interaction in the epidemic model is significant. Both the likelihood ratio test of the epidemic against the endemic model (p < 0.0001) and the Knox test (p < 0.0001) support this result.

Fig. 4: The results of the Monte Carlo permutation test.

The grey bars show the null distribution of R₀ from the 739 permutations that converged, whereas the red dashed line shows the observed R₀ (0.31) calculated from the twinstim model.

Full size image

The results of the simulations can be seen in Figs. 5 and 6. On average the simulations neatly match the observed cumulative number of exposures between June 2017 and January 2018 (Fig. 5), indicating that the model accurately captures the temporal dynamics in the data. Similarly, the model appears to do a good job of capturing the spatial dynamics in the data, although it is clearly weighted towards high population density areas (Fig. 6).

Fig. 5: The cumulative number of exposures during the last six months of the study period (red).

The results of 1000 simulations are shown in grey.

Full size image

Fig. 6: The exposure locations during the last six months of the study period.

The color gradient, ranging from blue (low) to yellow (high), represents a Gaussian kernel density for the results of 1000 simulations with a bandwidth of 200 km. The contour lines segment the kernel density into 10 levels.

Full size image

Discussion

By applying novel epidemiological methods to data on 416 extremists exposed between 2005 and 2017, this study provides evidence that patterns of far-right radicalization in the United States are consistent with a contagion process. Firstly, the estimated reproduction number is significantly higher than those from simulated null models, indicating that endemic causes alone are not sufficient to explain the spatio-temporal clustering observed in the data. The reproduction number for radicalization (R₀ = 0.31) is also lower than one, suggesting that extremist ideologies behave like complex contagions that require reinforcement for transmission. Fortunately, this means that extremist ideologies are unlikely to spread uncontrollably through populations like seasonal influenza (R₀ = 1.28) (Biggerstaff et al., 2014), but outbreaks can occur under the right endemic and epidemic conditions. For example, regions with higher rates of poverty and hate group activity are more likely to experience far-right extremism, whereas regions with a larger non-white population, more Republican voting, and higher rates of unemployment are less likely to experience far-right extremism. Most importantly, radicalizations involving extremist groups or social media significantly increase the epidemic probability of future radicalizations in the same location. This suggests that clusters of radicalizations in space and time are driven by activism and organizing rather than a copy-cat effect.

The fact that group membership significantly increases the epidemic strength of events, and the presence of hate groups significantly increases radicalization probability, suggests that local organizing remains a potent recruitment tool of the far-right movement. This idea is reflected in recent increases in rallies across the country, such as “Unite the Right” in Charlottesville, VA in August of 2017, that have been attended by regional chapters of white nationalist and militia organizations. It also suggests that concerns about typological “lone wolves” radicalized over social media should not overshadow the persistent and expanding far-right movement in the United States. Only 10.8% of people in this study were radicalized on social media independently of an extremist group, indicating that solo actors are still the minority in the far-right movement. That being said, solo actors radicalized on social media, such as Omar Mateen (Pulse nightclub shooting in 2016) and Dylann Roof (Charleston church shooting in 2015) (Holt et al., 2019), are typically deadlier than group members in the United States (Phillips, 2017), and should thus be the subject of much future research.

Radicalization on social media also significantly increases the epidemic strength of events, indicating that social media platforms augment physical organizing and that the diffusion of extremist ideologies online is likely geographically biased. The increasing role of social media in far-right extremism and radicalization is well established (Costello and Hawdon, 2018; Holt et al., 2019; Lowe, 2019; Ottoni et al., 2018; Winter, 2019). Social media platforms like Twitter provide extremist communities with low cost access to large audiences that might not otherwise engage with far-right content (Bertram, 2016; Wu, 2015). For example, one report found that only 44% of people who follow high-profile white nationalists on Twitter overtly express similar ideologies (Berger and Strathearn, 2013). As mainstream platforms clamp down on hate speech, extremist users have just shifted their traffic to alternative sites such as 8chan and Gab (Blackbourn et al., 2019; Hodge and Hallgrimsdottir, 2019). Given the centrality of social media in far-right organizing, future research should explore how counter-narratives (van Eerten et al., 2017; Voogt, 2017) and other strategies could be used to fight the spread of extremist ideologies online.

The results indicate that county-level poverty rates increase the probability of far-right radicalization. While there is little to no evidence that poverty predicts extremism at the state-level (Durso and Jacobs, 2013; Gale et al., 2002; Lin et al., 2018; Piazza, 2017), studies at the county-level have found that poverty predicts both mass shooting rate (Kwon and Cabrera, 2019b) and hate groups (presence (Medina et al., 2018) not longevity (Suttmoeller et al., 2015, 2016, 2018)). This discrepancy between geographic resolutions indicates that using state-level poverty data obscures local variation. The results of this study also reveal a negative effect of unemployment rate on radicalization, adding to the remarkably contradictory evidence for links between unemployment and extremism in the United States (Espiritu, 2004; Gale et al., 2002; Goetz et al., 2012; Green et al., 1998, Jefferson and Pryor, 1999; Majumder, 2017; Piazza, 2017). Although this result appears to be counter-intuitive, I hypothesize that poverty and unemployment may interact in driving radicalization. For example, individuals from regions where jobs are plentiful but poverty remains high may be the most disillusioned and susceptible to extremist ideologies. Interestingly, income inequality did not appear in the best fitting model, and had no significant effect when included. This suggests that overall deprivation, as measured by poverty rate, is more important in driving radicalization than inequality. Previous studies that have found a positive impact of income inequality on hate groups or crime either used state-level data (Majumder, 2017), did not account for poverty rate (Goetz et al., 2012; McVeigh, 2004), or combined income inequality with poverty rate into a single index (McVeigh and Cunningham, 2012). Interestingly, both unemployment (Pah et al., 2017) and income inequality (Kwon and Cabrera, 2017, 2019a, b) appear to be strong predictors of mass shootings. Although this seems paradoxical, the majority of mass shootings are not ideologically driven (Capellan, 2015), so the socioeconomic drivers may be different than for far-right radicalization.

Violent crime appears to have no influence on radicalization. Although one study of the Ku Klux Klan found that high levels of far-right activity can increase homicide rates in the long-term (McVeigh and Cunningham, 2012), there is little evidence that violent crime rates drive increases in extremist violence or radicalization (Sweeney and Perliger, 2018). Hate crime is only very weakly correlated with violent crime (Gladfelter et al., 2017), and extremist violence is even more rare (LaFree and Dugan, 2009), so they are likely driven by different factors.

Previous studies have found strong evidence for a negative relationship between education and hate crime rates (Espiritu, 2004; Gladfelter et al., 2017), a positive relationship between education and mass shooting rates (Kwon and Cabrera, 2017, 2019a), and no evidence for a relationship between education and hate group organizing (Durso and Jacobs, 2013; Florida, 2011; McVeigh et al., 2014). The results of this study are consistent with the latter category, which makes sense given that the majority of the plots in the dataset were non-violent.

The negative effect of Republican voting on event probability could be because individuals on the far-right of the political spectrum who live in counties with more Democratic voters may feel more partisan hostility (Miller and Conover, 2015). Interestingly, this effect does not appear to be the result of more competitive elections (Suttmoeller et al., 2015), as the absolute difference between Republican and Democratic voting did not significantly influence event probability. Alternatively, the negative effect of Republican voting may be due to the fact that many of the rural counties that lean heavily Republican have low population densities and no recent history of extremist violence. A previous study that found mixed evidence for a positive influence of Republican voting on the presence of hate groups excluded counties without hate groups from the modeling, which may have eliminated this skew effect (Medina et al., 2018).

The fact that the percentage of the population that is non-white negatively predicts far-right extremist violence is consistent with the intergroup contact hypothesis, which suggests that prolonged contact between racial groups reduces conflict under certain conditions (Allport, 1954). Although other researchers have suggested that population heterogeneity increases far-right radicalization (McVeigh, 2004), the only study to find evidence of this in the United States did not explicitly account for population density (LaFree and Bersani, 2014). Other studies controlling for population density have found that both anti-black hate crimes and hate groups appear to be more common in white dominated, racially homogeneous areas (Gladfelter et al., 2017; Medina et al., 2018). Despite mixed evidence for the intergroup contact hypothesis, it is widely accepted that community diversity and tolerance is key to fighting radicalization and extremist violence globally (Ellis and Abdi, 2017; Ercan, 2017; Gunaratna et al., 2013; Hoffman et al., 2018; Southern Poverty Law Center, 2017; United Nations Development Programme, 2016).

Gun ownership does not predict radicalization in this model, which is unsurprising since only 30.5% of people in this study planned on committing fatal attacks but interesting given the centrality of gun control in debates following mass shootings in the United States (Joslyn and Haider-Markel, 2017; Luca et al., 2020; Pierre, 2019). Despite strong evidence that gun ownership is linked to mass shooting rates at the national-level (Reeping et al., 2019), evidence for same pattern at the state-level remains mixed. Previous studies have found that it either positively predicts mass shootings overall (Reeping et al., 2019), when combined with particular gun control laws (Anisin, 2018), or not at all (Lin et al., 2018; Pah et al., 2017). Unfortunately, CDC funding for research on gun ownership was restricted by Congress in 1996 after lobbying by the National Rifle Association, so potential links between extremist violence and gun ownership remain understudied (DeFoster and Swalve, 2018; Lemieux, 2014; Morall, 2018; Winker et al., 2016).

Several limitations of this study should be highlighted. Firstly, the PIRUS database only represents a subset of radicalized individuals in the United States. The creators of the database used random sampling to maximize its representativeness over different time periods, but there remains a possibility of spatial or temporal bias in the original data due to underreporting by victims and law enforcement effort (DiIulio, 1996). Instances of hate crime are notoriously underreported relative to other forms of crime (Pezzella et al., 2019), because victims often fear retaliation or mistrust the police (Pezzella, 2017; Weiss et al., 2016; Wong and Christmann, 2016). There is also a great deal of variation in hate crime training among police departments, and the personal beliefs of individual officers can influence whether or not instances are reported (Boyd et al., 1996; Pezzella, 2017). Both of these factors are likely to be more pronounced in areas with legacies of far-right extremist violence, and historical crossover between far-right groups and the police (Barnes, 1996; Johnson, 2019; Rowe, 1976). In addition, the geographic locations of events are only geocoded to the city-level, potentially enhancing the spatial clustering of the data. Furthermore, social media data were missing for a significant number of individuals (54.8%). The significance level of the estimate for social media usage is extremely low and robust to imputation, indicating that it likely reflects a real effect, but researchers should exercise caution when interpreting this result (Safer-Lichtenstein et al., 2017). Lastly, the spatial resolution of three of the endemic predictors was limited to the state-level, which may have flattened some important local variation. One of these variables, gun ownership, was also a proxy measure. Policymakers should release historical restrictions on research funding for gun violence and hate crime research to improve data resolution for future studies.

In conclusion, far-right radicalization in the United States appears to spread through populations like a complex contagion. Both social media usage and group membership enhance the contagion process, indicating that online and physical organizing remain primary recruitment tools of the far-right movement. In addition, far-right radicalization is more likely in Democrat-majority regions with high poverty and low unemployment, fewer non-white people, and more hate group activity. While the federal government has acknowledged the threat of far-right extremism (The Department of Homeland Security, 2019), funding for organizations researching or fighting the movement has decreased in recent years (O’Toole, 2019). Based on the results of this study, I recommend that policymakers reconsider their funding priorities to address the expanding far-right extremist movement in the United States. Future research should investigate how specific interventions, such as online counter-narratives to battle propaganda, may be effectively implemented to mitigate the spread of extremist ideology.

Similar content being viewed by others

N. Velásquez, P. Manrique, … N. F. Johnson

Marta Martins

Koen van der Zwet, Ana I. Barros, … Peter M. A. Sloot

Data availability

All data used in the study are available online either publicly or upon request from PIRUS. The R code used in the study is available on Harvard Dataverse: https://doi.org/10.7910/DVN/WPYCKJ

References

• Adamczyk A, Gruenewald J, Chermak SM, Freilich JD (2014) The relationship between hate groups and far-right ideological violence. J Contemp Crim Justice 30(3):310–332

  Article  Google Scholar 

• Allam H (2019) ‘We were blindsided’: families of extremists form group to fight hate. National Public Radio (12)

• Allport G (1954) The nature of prejudice. Addison-Wesley, Cambridge

  Google Scholar 

• Aly A, Macdonald S, Jarvis L, TM Chen (2017) Introduction to the special issue: terrorist online propaganda and radicalization. Stud Confl Terror 40(1):1–9

  Article  Google Scholar 

• Amble JC (2012) Combating terrorism in the new media environment. Stud Confl Terror 35(5):339–353

  Article  Google Scholar 

• Anisin A (2018) A configurational analysis of 44 US mass shootings: 1975-2015. Int J Comparat Appl Criminal Justice 42(1):55–73

  Article  Google Scholar 

• Awan I (2017) Cyber-extremism: Isis and the power of social media. Society 54(2):138–149

  Article  Google Scholar 

• Bailey M, Cao R, Kuchler T, Stroebel J, Wong A (2018) Social connectedness: measurement, determinants, and effects. J Econ Perspect 32(3):259–280

  Article  PubMed  Google Scholar 

• Barnes RD (1996) Blue by day and white by (k)night: regulating the political affiliations of law enforcement and military personnel. Iowa Law Rev 81:1079

  Google Scholar 

• Bastos M, Mercea D, Baronchelli A (2018) The geographic embedding of online echo chambers: evidence from the Brexit campaign. PLoS ONE 13(11):1–16

  Article  CAS  Google Scholar 

• Bastug M, Douai A, Akca D (2018) Exploring the "demand side” of online radicalization: evidence from the Canadian context. Stud Confl Terror. https://doi.org/10.1080/1057610X.2018.1494409

• Becker MH (2019) When extremists become violent: examining the association between social control, social learning, and engagement in violent extremism. Stud Confl Terror. https://doi.org/10.1080/1057610X.2019.1626093

• Berger JM, Strathearn B (2013) Who matters online: measuring influence, evaluating content and countering violent extremism in online social networks. The International Centre for the Study of Radicalisation and Political Violence. https://bit.ly/35lvqby

• Bertram L (2016) Terrorism, the Internet, and the Social Media Advantage: Exploring how terrorist organizations exploit aspects of the internet, social media and how these same platforms could be used to counter-violent extremism. J Deradical 7:225–252

  Google Scholar 

• Biggerstaff M, Cauchemez S, Reed C, Gambhir M, Finelli L (2014) Estimates of the reproduction number for seasonal, pandemic, and zoonotic influenza: a systematic review of the literature BMC Infect Dis 14:480

  Article  PubMed  PubMed Central  Google Scholar 

• Blackbourn J, McGarrity N, Roach K (2019) Understanding and responding to right wing terrorism. J Polic Intell Count Terror 14(3):183–190

  Article  Google Scholar 

• Bonn T (2019) Former extremists call for violent extremism to be treated as public health issue. The Hill (9)

• Bouhana N, Corner E, Gill P, Schuurman B (2018) Background and preparatory behaviours of right-wing extremist lone actors: a comparative study. Perspect Terror 12(6):150–163

  Google Scholar 

• Bowman-Grieve L (2009) Exploring stormfront: a virtual community of the radical right. Stud Confl Terror 32(11):989–1007

  Article  Google Scholar 

• Boyd EA, Berk RA, Hamner KM (1996) "Motivated by hatred or prejudice”: categorization of hate-motivated crimes in two police divisions. Law Soc Rev 30(4):819–850

  Article  Google Scholar 

• Braun R (2011) The diffusion of racist violence in the Netherlands: discourse and distance. J Peace Res 48(6):753–766

  Article  ADS  Google Scholar 

• Braun R, Koopmans R (2010) The diffusion of ethnic violence in Germany: the role of social similarity. Eur Sociol Rev 26(1):111–123

  Article  Google Scholar 

• Buhaug H, Gleditsch KS (2008) Contagion or confusion? why conflicts cluster in space. Int Stud Q 52(2):215–233

  Article  Google Scholar 

• Burnham KP, Anderson DR (2002) Model selection and multimodel inference: a practical information-theoretic approach. Springer, New York

  MATH  Google Scholar 

• Capellan JA (2015) Lone wolf terrorist or deranged shooter? a study of ideological active shooter events in the United States, 1970-2014. Stud Confl Terror 38(6):395–413

  Article  Google Scholar 

• Cherif A, Yoshioka H, Ni W, Bose P (2009) Terrorism: mechanisms of radicalization processes, control of contagion and counter-terrorist measures. Preprint at arXiv: 0910.5272

• Clark NJ, Dixon PM (2018) Modeling and estimation for self-exciting spatio-temporal models of terrorist activity. Ann Appl Stat 12(1):633–653

  Article  MathSciNet  MATH  Google Scholar 

• Collins B, Hoang DT, Yoon HJ, Nguyen NT, Hwang D (2020) A survey on forecasting models for preventing terrorism. In: Advanced computational methods for knowledge engineering. Springer, Cham, pp 323–334

• Costello M, J Hawdon (2018) Who are the online extremists among us? sociodemographic characteristics, social networking, and online experiences of those who produce online hate materials. Violence Gender 5(1):55–60

  Article  Google Scholar 

• Davey J, Ebner J (2019) ‘The great replacement’: the violent consequences of mainstreamed extremism. Institute for Strategic Dialogue. https://bit.ly/2yXdl7h

• Dean G, Bell P, Newman J (2012) The dark side of social media: review of online terrorism. Pak J Criminol 3(3):107–126

  Google Scholar 

• DeFoster R, Swalve N (2018) Guns, culture or mental health? framing mass shootings as a public health crisis. Health Commun 33(10):1211–1222

  Article  PubMed  Google Scholar 

• DiIulio JJ (1996) Help wanted: economists, crime and public policy. J Econ Perspect 10(1):3–24

  Article  Google Scholar 

• Durso RM, Jacobs D (2013) The determinants of the number of white supremacist groups: a pooled time-series analysis. Soc Probl 60(1):128–144

  Article  Google Scholar 

• Ellis BH, Abdi S (2017) Building community resilience to violent extremism through genuine partnerships. Am Psychol 72(3):289–300

  Article  PubMed  Google Scholar 

• Ercan SA (2017) Engaging with extremism in a multicultural society: a deliberative democratic approach. J Peacebuild Dev 12(2):9–21

  Article  MathSciNet  Google Scholar 

• Espiritu A (2004) Racial diversity and hate crime incidents. Soc Sci J 41(2):197–208

  Article  Google Scholar 

• Ferrara E (2017) Contagion dynamics of extremist propaganda in social networks. Inf Sci 418-419:1–12

  Article  Google Scholar 

• Florida R (2011) The geography of hate. The Atlantic (5)

• Funke M, Schularick M, Trebesch C (2016) Going to extremes: politics after financial crises, 1870-2014. Eur Econ Rev 88(2011):227–260

  Article  Google Scholar 

• Gale LR, Heath WC, Ressler RW (2002) An economic analysis of hate crime. East Econ J 28(2):203–216

  Google Scholar 

• Garcia-Bernardo J, Qi H, Shultz JM, Cohen AM, Johnson NF, Dodds PS (2015) Social media affects the timing, location, and severity of school shootings. Preprint at arXiv: 1506.06305

• Gelfand M, Shteynberg G, Lee T, Lun J, Lyons S, Bell C, Chiao JY, Bruss CB, Dabbagh MA, Aycan Z, Abdel-Latif AH, Dagher M, Khashan H, Soomro N (2012) The cultural contagion of conflict. Philos Trans R Soc B 367(1589):692–703

  Article  Google Scholar 

• Giani M, Meón PG (2019) Global racist contagion following Donald Trump’s election. Br J Polit Sci. https://doi.org/10.1017/S0007123419000449

• Gill P, Corner E, Conway M, Thornton A, Bloom M, Horgan J (2017) Terrorist use of the internet by the numbers: quantifying behaviors, patterns, and processes. Criminol Public Policy 16(1):99–117

  Article  Google Scholar 

• Gladfelter AS, Lantz B, Ruback RB (2017) The complexity of hate crime and bias activity: variation across contexts and types of bias. Justice Q 34(1):55–83

  Article  Google Scholar 

• Goetz SJ, Rupasingha A, Loveridge S (2012) Social capital, religion, Wal-Mart, and hate groups in America. Soc Sci Q 93(2):379–393

  Google Scholar 

• González-Bailón S, Borge-Holthoefer J, Rivero A, Moreno Y (2011) The dynamics of protest recruitment through an online network. Sci Rep 1(197):1–7

  Google Scholar 

• Green DP, Glaser J, Rich A (1998) From lynching to gay bashing: the elusive connection between economic conditions and hate crime. J Pers Soc Psychol 75(1):82–92

  Article  CAS  PubMed  Google Scholar 

• Guilbeault D, Becker J, Centola D (2018) Complex contagions: a decade in review. In: Complex spreading phenomena in social systems. Springer, Cham, pp 3–25

• Gunaratna R, Jerard J, Nasir SM (2013) Countering extremism: building social resilience through community engagement. Imperial College Press, London

  Book  Google Scholar 

• Helfgott JG (2015) Criminal behavior and the copycat effect: literature review and theoretical framework for empirical investigation. Aggress Violent Behav 22:46–64

  Article  Google Scholar 

• Hodge E, Hallgrimsdottir H (2019) Networks of hate: the alt-right, “troll culture”, and the cultural geography of social movement spaces online. J Borderl Stud https://doi.org/10.1080/08865655.2019.1571935

• Hoffman AJ, Alamilla S, Liang B (2018) The role of community development in reducing extremism and ethnic conflict. Palgrave MacMillan, Cham

  Book  Google Scholar 

• Holt TJ, Freilich JD, Chermak SM (2016) Internet-based radicalization as enculturation to violent deviant subcultures. Deviant Behav 38(8):855–869

  Article  Google Scholar 

• Holt TJ, Freilich JD, Chermak SM, Mills C, Silva J (2019) Loners, colleagues, or peers? assessing the social organization of radicalization. Am J Crim Justice 44(1):83–105

  Article  Google Scholar 

• Jasko K, LaFree G, Kruglanski A (2017) Quest for significance and violent extremism: the case of domestic radicalization. Polit Psychol 38(5):815–831

  Article  Google Scholar 

• Jefferson PN, Pryor FL (1999) On the geography of hate. Econ Lett 65(3):389–395

  Article  MATH  Google Scholar 

• Jensen MA, Seate AA, James PA (2018) Radicalization to violence: a pathway approach to studying extremism. Terror Polit Violence. https://doi.org/10.1080/09546553.2018.1442330

• Johnson VB (2019) KKK in the PD: white supremacist police and what to do about it. Lewis Clark Law Rev 23(1):205–261

  Google Scholar 

• Johnson SD, Braithwaite A (2017) Spatial and temporal analysis of terrorism and insurgency. In: The handbook of the criminology of terrorism. Wiley, Hoboken, pp 232–243

• Joslyn MR, Haider-Markel DP (2017) Gun ownership and self-serving attributions for mass shooting tragedies. Soc Sci Q 98(2):429–442

  Article  Google Scholar 

• Kahle D, Wickham H (2013) ggmap: spatial visualization with ggplot2. R J 5(1):144–161

  Article  Google Scholar 

• Kaplan J, Lööw H, L Malkki (2014) Introduction to the special issue on lone wolf and autonomous cell terrorism. Terror Polit Violence 26(1):1–12

  Article  Google Scholar 

• Klein BR, Gruenewald J, Smith BL (2017) Opportunity, group structure, temporal patterns, and successful outcomes of far-right terrorism incidents in the United States. Crime Delinq 63(10):1224–1249

  Article  Google Scholar 

• Knox EG, Bartlett MS (1964) The detection of space-time interactions. J Royal Stat Soc C 13(1):25–30

  Google Scholar 

• Kwon R, Cabrera JF (2017) Socioeconomic factors and mass shootings in the United States. Crit Public Health 28(2):138–145

  Article  Google Scholar 

• Kwon R, Cabrera JF (2019a) Income inequality and mass shootings in the United States. BMC Public Health 19:1147

  Article  PubMed  PubMed Central  Google Scholar 

• Kwon R, Cabrera JF (2019b) Social integration and mass shootings in U.S. counties. J Crime Justice 42(2):121–139

  Article  Google Scholar 

• LaFree G, Bersani BE (2014) County-level correlates of terrorist attacks in the United States. Criminol Public Policy 13(3):455–481

  Article  Google Scholar 

• LaFree G, Dugan L (2009) Tracking global terrorism trends, 1970-2004. In: To protect and to serve: policing in an age of terrorism. Springer, New York, pp 43–80

• LaFree G, Dugan L, Xie M, P Singh (2012) Spatial and temporal patterns of terrorist attacks by ETA 1970 to 2007. J Quant Criminol 28(1):7–29

  Article  Google Scholar 

• Lemieux F (2014) Effect of gun culture and firearm laws on gun violence and mass shootings in the United States: a multi-level quantitative analysis. Int J Crim Justice Sci 9(1):74–93

  Google Scholar 

• Lewis E, Mohler G, Brantingham PJ, Bertozzi AL (2012) Self-exciting point process models of civilian deaths in Iraq. Security J 25(3):244–264

  Article  Google Scholar 

• Lin PI, Fei L, Barzman D, Hossain M (2018) What have we learned from the time trend of mass shootings in the U.S.? PLoS ONE 13(10):1–13

  Google Scholar 

• Liu L, Chen B, Ai C, He L, Wang Y, Qiu X, Lu X (2018) The influence of geographic factors on information dissemination in mobile social networks in China: evidence from WeChat. ISPRS Int J Geoinf 7(5):1–16

  Google Scholar 

• Lowe D (2019) The Christchurch terrorist attack, the far-right, and social media: what can we Learn? The New Jurist (4)

• Luca M, Malhotra D, Poliquin C (2020) The impact of mass shootings on gun policy. J Public Econ 181:104083

  Article  Google Scholar 

• Majumder M (2017) Higher rates of hate crimes are tied to income inequality. FiveThirtyEight (1)

• Marchment Z, Bouhana N, Gill P (2018) Lone actor terrorists: a residence-to-crime approach. Terror Political Violence. https://doi.org/10.1080/09546553.2018.1481050

• McVeigh R (2004) Structured ignorance and organized racism in the United States. Soc Forces 82(3):895–936

  Article  Google Scholar 

• McVeigh R, Cunningham D (2012) Enduring consequences of right-wing extremism: Klan mobilization and homicides in southern counties. Soc Forces 90(3):843–862

  Article  Google Scholar 

• McVeigh R, Cunningham D, Farrell J (2014) Political polarization as a social movement outcome: 1960s Klan activism and its enduring impact on political realignment in southern counties, 1960 to 2000. Am Sociol Rev 79(6):1144–1171

  Article  Google Scholar 

• Medina RM, Nicolosi E, Brewer S, Linke AM (2018) Geographies of organized hate in America: a regional analysis. Ann Am Assoc Geogr 108(4):1006–1021

  Google Scholar 

• Meyer S, Elias J, Höhle M (2012) A space-time conditional intensity model for invasive meningococcal disease occurrence. Biometrics 68(2):607–616

  Article  MathSciNet  PubMed  MATH  Google Scholar 

• Meyer S, Held L (2014) Power-law models for infectious disease spread. Ann Appl Stat 8(3):1612–1639

  Article  MathSciNet  MATH  Google Scholar 

• Meyer S, Held L, Höhle M (2017) Spatio-temporal analysis of epidemic phenomena using the R package surveillance. J Stat Softw 77(11):1–55

  Article  CAS  Google Scholar 

• Meyer S, Warnke I, Rössler W, Held L (2016) Model-based testing for space-time interaction using point processes: an application to psychiatric hospital admissions in an urban area. Spat Spatio-Temporal Epidemiol 17:15–25

  Article  Google Scholar 

• Midlarsky MI, Crenshaw M, Yoshida F (1980) Why violence spreads: the contagion of international terrorism. Int Stud Q 24(2):262–298

  Article  Google Scholar 

• Miller PR, Conover PJ (2015) Red and blue states of mind: partisan hostility and voting in the United States. Polit Res Q 68(2):225–239

  Article  Google Scholar 

• Mills CE, Freilich JD, Chermak SM, Holt TJ, LaFree G (2019) Social learning and social control in the off- and online pathways to hate crime and terrorist violence. Stud Confl Terror. https://doi.org/10.1080/1057610X.2019.1585628

• Misiak B, Samochowiec J, Bhui K, Schouler-Ocak M, Demunter H, Kuey L, Raballo A, Gorwood P, Frydecka D, Dom G (2019) A systematic review on the relationship between mental health, radicalization and mass violence. Eur Psychiatry 56:51–59

  Article  PubMed  Google Scholar 

• Morall A (2018) The science of gun policy: a critical synthesis of research evidence on the effects of gun policies in the united states. RAND Health Quart 8:1

  Google Scholar 

• Nacos BL (2009) Revisiting the contagion hypothesis: terrorism, news coverage, and copycat attacks. Perspect Terror 3(3):3–13

  Google Scholar 

• National Consortium for the Study of Terrorism and Responses to Terrorism (START), University of Maryland (2017). Profiles of Individual Radicalization in the United States (PIRUS). https://bit.ly/2yWtmu8

• Nightingale GF, Laland KN, Hoppitt W, Nightingale P (2015) Bayesian spatial NBDA for diffusion data with home-base coordinates. PLoS ONE 10(7):e0130326

  Article  PubMed  PubMed Central  CAS  Google Scholar 

• O’Toole M (2019) Trump officials have redirected resources from countering far-right, racism-fueled domestic terrorism. Los Angeles Times (8)

• Ottoni R, Bernardina P, Cunha E, Meira W, Magno G, Almeida V (2018) Analyzing right-wing YouTube channels: hate, violence and discrimination. In: WebSci 2018 Proceedings of the 10th ACM Conference On Web Science, Amsterdam, pp 323–332

• Pah AR, Hagan J, Jennings AL, Jain A, Albrecht K, Hockenberry AJ, Amaral LA (2017) Economic insecurity and the rise in gun violence at US schools. Nat Hum Behav 1(2):2–7

  Article  Google Scholar 

• Pauwels L, Brion F, Schils N, Laffineur J, Verhage A, de Ruyver B, Easton M (2014) Explaining and understanding the role of exposure to new social media in violent extremism: an integrative quantitative and qualitative approach. Academia Press, Gent

  Google Scholar 

• Pezzella FS (2017) Hate crime statutes: a public policy and law enforcement dilemma. Springer, New York

  Book  Google Scholar 

• Pezzella FS, Fetzer MD, Keller T (2019) The dark figure of hate crime underreporting. Am Behav Sci. https://doi.org/10.1177/0002764218823844

• Phillips BJ (2017) Deadlier in the U.S.? on lone wolves, terrorist groups, and attack lethality. Terror Polit Violence 29(3):533–549

  Article  Google Scholar 

• Piazza JA (2017) The determinants of domestic right-wing terrorism in the USA: economic grievance, societal change and political resentment. Confl Manag Peace Sci 34(1):52–80

  Article  Google Scholar 

• Pierre JM (2019) The psychology of guns: risk, fear, and motivated reasoning. Pal Commun 5(1):1–7

  Article  Google Scholar 

• Porter MD, White G (2010) Self-exciting hurdle models for terrorist activity. Ann Appl Stat 4(1):106–124

  Article  MathSciNet  MATH  Google Scholar 

• Post JM (2015) Terrorism and right-wing extremism: the changing face of terrorism and political violence in the 21st century: the virtual community of hatred. Int J Group Psychoth 65(2):242–271

  Article  Google Scholar 

• Recht H (2019) censusapi: retrieve data from the census APIs. R package version 0.6.0

• Reeping PM, Cerdá M, Kalesan B, Wiebe DJ, Galea S, Branas CC (2019) State gun laws, gun ownership, and mass shootings in the US: cross sectional time series. BMJ 364:1542

  Google Scholar 

• Reinhart A (2018) A review of self-exciting spatio-temporal point processes and their applications. Stat Sci 33(3):299–318

  Article  MathSciNet  MATH  Google Scholar 

• Rowe G (1976) My undercover years with the Ku Klux Klan. Bantam Books, New York

  Google Scholar 

• Rushin S, Edwards GS (2018) The effect of President Trump’s election on hate crimes. SSRN Electronic Journal. https://doi.org/10.2139/ssrn.3102652

• Safer-Lichtenstein A, LaFree G, Loughran T (2017) Studying terrorism empirically: what we know about what we donat know. J Contemp Crim Justice 33(3):273–291

  Article  Google Scholar 

• Sanir J, Nicholson A, Giammaria C (2017) Countering violent extremism through public health practice: proceedings of a workshop. In: Forum on medical and public health preparedness for disasters and emergencies. The National Academies Press, Washington, DC

• Seigel M, Ross C, King C (2014) A new proxy measure for state-level gun ownership in studies of firearm injury prevention. Inj Prev 20(3):204–207

  Article  Google Scholar 

• Silkoset E (2016) Lone-actor terrorists: how long does it take to plan a terrorist attack? Dissertation, University College of London

• Simi P, Bubolz BF (2017) Far right terrorism in the United States. In: The handbook of the criminology of terrorism, Wiley, Hoboken. pp 297–309

• Smith BL, Cothren J, Roberts P, Damphousse KR (2008) Geospatial analysis of terrorist activities: the identification of spatial and temporal patterns of preparatory behavior of international and environmental terrorists. Terrorism Research Center in Fulbright College, University of Arkansas. https://bit.ly/2SibIIc

• Smith BL, Damphousse KR (2009) Patterns of precursor behaviors in the life span of a U.S. environmental terrorist group. Criminol Public Policy 8(3):475–496

  Article  Google Scholar 

• Southern Poverty Law Center (2017) Ten ways to fight hate: a community resource guide. https://bit.ly/35j4TLN

• Suttmoeller M, Chermak S, Freilich JD (2015) The influence of external and internal correlates on the organizational death of domestic far-right extremist groups. Stud Confl Terror 38(9):734–758

  Article  Google Scholar 

• Suttmoeller MJ, Chermak SM, Freilich JD (2016) Only the bad die young: the correlates of organizational death for far-right extremist groups. Stud Confl Terror 39(6):477–499

  Article  Google Scholar 

• Suttmoeller MJ, Chermak SM, Freilich JD (2018) Is more violent better? the impact of group participation in violence on group longevity for far-right extremist groups. Stud Confl Terror 41(5):365–387

  Article  Google Scholar 

• Sweeney MM, Perliger A (2018) Explaining the spontaneous nature of far-right violence in the United States. Perspect Terror 12(6):52–71

  Google Scholar 

• Tench S, Fry H, Gill P (2016) Spatio-temporal patterns of IED usage by the Provisional Irish Republican Army. Eur J Appl Math 27(3):377–402

  Article  MathSciNet  MATH  Google Scholar 

• The Anti-Defamation League Center on Extremism (2019) Murder and extremism in the United States in 2018. https://bit.ly/2VLy5If

• The Department of Homeland Security (2019) Strategic framework for countering terrorism and target violence. https://bit.ly/2xhCCsB

• Tominaga Y (2018) Thereas no place like home! examining the diffusion of suicide attacks through terrorist group locations. Appl Spat Anal Policy 11(2):355–379

  Article  MathSciNet  Google Scholar 

• Towers S, Gomez-Lievano A, Khan M, Mubayi A, Castillo-Chavez C (2015) Contagion in mass killings and school shootings. PLoS ONE 10(7):e0117259

  Article  PubMed  PubMed Central  CAS  Google Scholar 

• United Nations Development Programme (2016) Preventing violent extremism through promoting inclusive development, tolerance and respect for diversity. https://bit.ly/2KJzxVa

• van Eerten JJ, Doosje B, Konijn E, de Graaf B, de Goede M (2017) Developing a social media response to radicalization: the role of counter-narratives in prevention of radicalization and de-radicalization. Wetenschappelijk Onderzoek-en Documentatiecentrum. https://bit.ly/2VMRkBc

• von Behr I, Reding A, Edwards C, Gribbon L (2013) Radicalisation in the digital era: the use of the internet in 15 cases of terrorism and extremism. RAND Europe. https://bit.ly/37AOnIn

• Voogt S (2017) Countering far-right recruitment online: CAPEas practitioner experience. J Polic Intelligence Counter Terrorism 12(1):34–46

  Article  Google Scholar 

• Walker K (2019) tigris: load vensus TIGER/line shapefiles. R package version 0.8.2

• Webber D, Kruglanski AW (2017) Psychological factors in radicalization: a “3 N” approach. In: The handbook of the criminology of terrorism. Wiley, Hoboken, pp 33–46

• Weine S, Eisenman, D (2016) How public health can improve initiatives to counter violent extremism. National Consortium for the Study of Terrorism and Responses to Terrorism. https://bit.ly/3cVtvNm

• Weiss JC, McDevitt J, Iwama JA (2016) Group work with victims of hate crimes. In: Greif GL, Knight C (ed) Group work with populations at risk, 4th edn. Oxford, New York, pp 291–311

• White G, Porter MD, Mazerolle L (2013) Terrorism risk, resilience and volatility: a comparison of terrorism patterns in three southeast Asian countries. J Quant Criminol 29(2):295–320

  Article  Google Scholar 

• White G, Ruggeri F, Porter MD (2016) Modelling the proliferation of terrorism via diffusion and contagion. Preprint at arXiv 1612.02527

• Winker MA, Abbasi K, Rivara FP (2016) Unsafe and understudied: the US gun problem. BMJ 352:i578

• Winter A (2019) Online hate: from the far-right to the ‘alt-right’ and from the margins to the mainstream. In: Online Othering. Springer, Cham, pp 39–63

• Wong K, Christmann K (2016) Increasing hate crime reporting: narrowing the gap between policy aspiration, victim inclination and agency capability. Br J Commun Justice 14(3):5–23

  Google Scholar 

• Wu P (2015) Impossible to regulate? social media, terrorists, and the role for the U.N. Chic J Int Law 16(1):281–311

  Google Scholar 

• Zammit-Mangion A, Dewar M, Kadirkamanathan V, Sanguinetti G (2012) Point process modelling of the Afghan War Diary. PNAS 109(31):12414–12419

  Article  ADS  CAS  PubMed  Google Scholar 

• Zuur AF, Ieno EN, Elphick CS (2010) A protocol for data exploration to avoid common statistical problems. Methods Ecol Evol 1(1):3–14

  Article  Google Scholar 

Download references

Acknowledgements

I would like to thank David Lahti, Bobby Habig, and the rest of the Lahti lab for their valuable conceptual feedback.

Author information

Authors and Affiliations

1. The Graduate Center, City University of New York, New York, NY, 10017, USA

   Mason Youngblood

2. Queens College, City University of New York, New York, NY, 10017, USA

   Mason Youngblood

Authors

1. Mason Youngblood

   You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Mason Youngblood.

Ethics declarations

Competing interests

The author declares no competing interests.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/.

Reprints and permissions

Applying Public Health Models and Approaches to Countering Violent Extremism

ncbi.nlm.nih.gov

---

Highlights of Main Points Made by Individual Speakers^a

• Public health prevention models for CVE could address the broad range of political, social, economic, and historical forces or grievances that can create and reinforce the conditions for violent extremism (Eisenman).

• Instead of only looking at what individuals are doing at a specific instance, recognize the policies and circumstances that lead to the behavior and address those (Wen).

• Even the most well-intentioned and best thought-out efforts and interventions can fail if they are not supported by local community champions who can leverage their own credibility and encourage others to understand that CVE is a health issue worth engaging (Arthur).

• Because a small number of people do pose a threat that warrants reporting, health professionals are tasked with striking a delicate balance between their social responsibilities and their duties of care (Wynia).

• Health professionals' poor ability to accurately detect credible threats is a critical concern particularly when the determination is used as justification for breaching civil liberties (Wynia).

a

This list is the rapporteurs' summary of the main points made by individual speakers and participants, and it does not reflect any consensus among workshop participants.

Chapter 5 focuses on how elements of the existing health and public health frameworks can be applied to strengthen and support CVE programs. By incorporating lessons learned in analogous programming that combat other types of violence, health and public health practitioners could promote community resilience, encourage community engagement, and help to gather evidence about the risk and protective factors that underpin radicalization toward violence. The evaluation of these programs, an essential component of public health models, is also discussed, as are the set of practical and ethical challenges health professionals face when working in CVE roles.

THE PUBLIC HEALTH APPROACH TO ADDRESSING THREATS

Eisenman noted that more than 2,500 federal, state, and local agencies constitute the backbone of the U.S. public health system and bear the legal responsibility for assuring the delivery of essential public health functions. The public health profession incorporates both health care delivery systems and academic public health sciences; however, he pointed out that there are multiple other sectors that contribute to the public health system and to population health (see Figure 5-1).

FIGURE 5-1

The intersectoral public health system. SOURCE: Eisenman presentation, September 8, 2016.

Steps in Implementing a Public Health Approach to Violence Prevention

In their respective presentations, Georges Benjamin and Leesa Lin detailed CDC's four key steps for implementing a public health approach to address health threats, including violence prevention¹:

1.

Define the problem using reliable data: “who,” “what,” “when,” and “how.”

2.

Identify risk and protective factors using scientific research methods: what factors protect people or put them at risk?

3.

Develop and test prevention strategies using evaluation.

4.

Assure widespread adoption using additional evaluation, training, and/or technical assistance.

Step 1: Define the Problem Using Reliable Data: “Who,” “What,” “When,” and “How”

Benjamin explained that the first step is to use data to define and monitor all facets of the problem at hand. This includes understanding who is at risk, what they are at risk of, when the problem is more likely to occur, and where it is likely to occur. Gaining clarity about the problem and how to address it also provides the necessary groundwork for putting measures into place for monitoring interventions, he explained.

Step 2: Identify Risk and Protective Factors Using Scientific Research Methods

The second step, according to Benjamin, is to identify the risk factors and protective factors at the individual, community, and broad societal levels that contribute to making certain communities more susceptible than others to experience a health threat. Such factors include issues such as the availability of firearms, religiosity, and level of wealth. He explained that the public health approach focuses heavily on exploring how these factors coalesce around particular problems.

Step 3: Develop and Test Prevention Strategies: Four-Tiered Model of Public Health Prevention

Benjamin stated that the third step is to apply knowledge gained about the problem, as well as its risk and protective factors, toward developing hypotheses about potentially effective interventions and putting programs into place. He noted that actually measuring and evaluating the efficacy of public health programs tends to be erratic. In many cases, a program is implemented and associated with a desirable health outcome; the program is then championed as effective without the benefit of a formal evaluation to determine whether or not the outcome was actually caused by chance. This highlights the importance of incorporating evaluation measures to assess success in preventing or mitigating health threats, he argued. He explained, however, that regardless of whether the program's success has been empirically validated, if it is ostensibly successful and contributing to positive momentum, then the next goal is to work toward prevention of the problem. Eisenman contended that it would be fair to construe all public health activities as preventative. Furthermore, he suggested that it would be more apt to frame the entire CVE enterprise as the prevention of violent extremism, not countering it.

To explain how the public health sector implements the task of prevention, both Eisenman and Benjamin explored the multitiered public health model, which encompasses multiple levels of prevention (see Figure 5-2).

FIGURE 5-2

Tiered model of public health prevention. SOURCE: Eisenman presentation, September 8, 2016.

Primordial prevention strategies are situated furthest upstream, noted Benjamin. Eisenman explained that the primordial prevention level is not always included in the public health model (some of which have only three tiers of prevention), but suggested that it warrants more attention. For example, by blocking American tobacco companies from introducing tobacco products into developing countries, public health advocates were successfully able to prevent the introduction of new risk factors for disease into those countries. He explained that in subsequent years, primordial prevention has come to be associated with changes in the country or changes in macrolevel social and political policies to prevent disease. Lin described primordial prevention as addressing fundamental health determinants at the system level.

Benjamin explained that the goal of the next tier, primary prevention, is to avert a health threat by addressing its root causes before they result in a disease or injury. Primary prevention efforts are upstream strategies aimed at reducing exposures to hazards, altering risky behaviors, and increasing resistance to disease or injury should exposure occur. He cited several examples of primary prevention strategies, including preventing exposure to secondhand smoke, building sidewalks to encourage and enable walking to promote cardiovascular health, promoting the use of seat belts or bicycle helmets, and vaccinating against pneumococcal pneumonia to protect individuals and strengthen herd immunity. He suggested that another primary prevention strategy with particular relevance to the CVE space—and violence prevention more broadly—is geared toward preventing early childhood trauma, which he described as a predictor of impulse control problems and future violent tendencies.

The goal of secondary prevention is to reduce the effect of a disease or an injury that has already occurred, said Benjamin. This process may involve strategies for detecting and treating a disease or injury as soon as possible to halt or slow its progression, encouraging personal strategies to prevent reinjury or recurrence, and implementing programs to return people to their original health and ability to function. Secondary strategies include, for example, tobacco cessation support for a person with coronary artery disease, mammograms and colonoscopies for detecting and treating early-stage malignancies, and treatment with antibiotics for pneumococcal pneumonia infection. In the context of violence prevention, he suggested that secondary prevention strategies might involve engaging with youth who are gang members without a history of violence. Intervening at that point may reduce their elevated risk of violent behavior as they enter adulthood and possibly become more involved with gang activities.

Benjamin described tertiary prevention strategies as concentrated on reducing the effect of an ongoing illness or injury that has lasting effects, as well as preventing its recurrence. He explained that tertiary prevention is accomplished by helping a person to manage a chronic disease and disability, such as by using medications to control hypertension in a stroke patient or providing rehabilitative therapy to a person who has had a myocardial infarction. Eisenman contributed the example of vocational rehabilitation to help retrain workers after they recover from an injury. According to Benjamin, the goal of tertiary prevention is to improve the person's ability to function, quality of life, and life expectancy. Intervening with an individual who has been a victim of violence to prevent retaliation is another form of tertiary prevention, because engaging with the victim can help to break the cycle of violence that can escalate from relatively minor injury to retaliatory serious injury or death, Benjamin noted.

Step 4: Assure Widespread Adoption Using Additional Evaluation, Training and/or Technical Assistance

After effective strategies have been identified, the fourth step is to assure their widespread adoption by disseminating and implementing them broadly. Benjamin highlighted the four Es of prevention: education, environment, enforcement, and evaluation. He suggested that education efforts should cast a wide net and cover individuals at risk, policy makers, and the public at large. Environmental efforts should aim to make the environment as safe as possible for individuals (e.g., providing home safety inspections for older people who have fallen and broken their hips). Benjamin emphasized that solutions must “go viral” with improved efforts to disseminate best practices that can be customized for specific environments.

Applying the Tiered Model of Public Health Prevention to CVE

Benjamin emphasized that the relationships between the four categories of prevention activities are complex, context dependent, and not always easily delineated. However, he suggested that understanding where a given program fits into the spectrum of prevention strategies is an important step in building a system for CVE. Both Benjamin and Eisenman explored how public health's four-tiered model of prevention can help to map out strategies for preventing extremist violence. Eisenman noted that most activities within existing CVE programs would fall under some level of the public health prevention model (see Figure 5-3).

FIGURE 5-3

Applying the four-tiered model of public health prevention to prevention of extremist violence. SOURCE: Eisenman presentation, September 8, 2016.

Universal prevention strategies, according to Eisenman, frame violent extremism as a problem for all communities, and such strategies are predicated on the assumption that all communities have opportunities for violence prevention. In the CVE space, he noted, primary prevention is concerned with targeting the majority of the population who have not engaged in problematic behaviors associated with violent extremism. Eisenman explained that such activities can include community programs and coun-ternarrative media campaigns to reduce risk factors and strengthen protective factors for individuals, families, and communities. These activities may either be CVE relevant or CVE specific, he noted. Benjamin suggested that programs in the primary prevention category could be broadly construed as community-level strategies that mitigate modifiable risks, such as limiting the availability of extremist media and reducing social isolation and exclusion. He contended that this can be done sensitively without stigmatizing or profiling individuals but by using data-driven methods to identify individuals at risk, a process that will require ongoing scientific research.

Eisenman explained that secondary prevention targets a subset of the population who are considered to be at risk for violent behaviors. He likened secondary strategies for detecting the behavioral changes in individuals that precede violent acts to strategies for detecting and addressing the preclinical changes that occur before a disease manifests.^2,3 Examples of CVE-specific and CVE-relevant program activities include targeted violence threat assessment programs, “off-ramps,” and interventions for people at risk before they commit violence. Benjamin similarly described secondary prevention strategies as directed at individuals who have some characteristics that put them at an elevated risk for violent extremism (e.g., exposure to extremist ideologies or proximity to a radical social network). On the contrary, Lin pointed out that exposure to extremist ideologies cannot be considered as a precursor of the “disease” and thus screening at the population level to detect and treat a disease at an early stage to prevent it from progressing (aka secondary prevention) cannot be done for CVE. She argued that given there is not sufficient evidence of the risk factors for violent extremism—science has yet to determine what puts individuals at a higher risk of committing an act of violence—secondary intervention for violent extremism is not feasible.

Tertiary prevention strategies, according to Benjamin, are directed at individuals who have already adopted extremist ideologies or are in contact with violent extremists, but are not engaged in planning or carrying out acts of violence, in order to divert them away from the path to violence. Eisenman explained that tertiary prevention activities, such as deradicalization programs, can help to manage and rehabilitate people who have already manifested criminal intent and violent extremist behaviors; these activities are generally housed in CVE-specific programs.

Eisenman suggested that primordial prevention for CVE could address the broad range of political, social, economic, and historical forces or grievances that can create and reinforce the conditions for violent extremism (e.g., Western foreign policy, wars in the Middle East, and the global distribution of wealth). He noted that while primordial prevention strategies are not generally on the forefront of the public health approach, there are advocates in the field who are active in attempting to address the social determinants of risk by promoting policies that improve human rights and reduce inequalities on both national and global levels. He noted that many such activities may not be CVE specific, but are CVE relevant.⁴ Lin added that primordial prevention addresses broad health determinants at the system level to minimize future hazards to health and inhibit the establishment factors.

Lin suggested adopting a fifth tier—quaternary prevention—that would include actions taken to oversee the civil liberty and ethical issues intrinsic in all levels of prevention, in order to ensure that people are not overtreated for behaviors that are not a risk (see Figure 5-4).

FIGURE 5-4

Levels of prevention in public health. SOURCE: Lin presentation, September 8, 2016.

Woodie Kessel, senior scholar at the Koop Institute, cautioned against assuming there is one most efficient strategy for prevention, arguing that all levels of prevention will require promotion, attention, investment, and consideration.

Applying CDC's “10 Essential Functions of Public Health” to CVE

Eisenman introduced CDC's 10 Essential Functions of Public Health in order to provide guidance about how the public health approach, system, and workforce can contribute to preventing extremist violence. He explained that these 10 essential functions are widely accepted as forming the foundation for all public health activities, and presented a figure (see Table 5-1) detailing how each of the 10 essential functions can be applied to prevent violent extremism (Weine et al., 2016).

TABLE 5-1

CDC 10 Essential Public Health Functions and CVE.

Eisenman explained that the first two functions fall under the auspices of public health assessments; in the CVE realm, this involves applying public health data to help understand CVE needs. He reiterated Benjamin's contention that data is the foundation of the public health enterprise, pointing to the usefulness of using relevant information from population-level public health surveillance surveys to inform CVE activities, particularly as research continues to improve understanding of the relevant risk and protective factors for violent extremism. For example, he suggested that many of the protective factors that are posited to mitigate against violent extremism—such as social cohesion and access to health care—could be the same factors that allow communities to withstand stresses and sustain healthy behaviors in the face of adversity. He noted that perceived discrimination and trust in government are often measured in public health surveillance surveys. Eisenman suggested that public health could also contribute by codifying definitions of violent extremism that lend themselves to prevention programs with a clear connection to health and well-being.

Functions 3, 4, and 5 in Table 5-1 pertain to the core public health function of developing policies and plans for health, according to Eisenman. Activities include informing and empowering communities, mobilizing partnerships, and developing policies and plans. He proposed that CVE could be addressed within the wider context of violence prevention generally, with the public health sector serving as convener and helping to assemble community sectors and agencies around CVE, as well as providing technical assistance on program planning and grant funding. Furthermore, he suggested that public health could contribute to policy making and program development, thereby helping to shift CVE away from its dependence on law enforcement and closer to the mental health education, youth development, and other human services sectors.

Eisenman explained that functions 6 through 10 represent the public health core functions of ensuring the populations' universal access to a culturally competent health system, for both general health and prevention. These functions include enforcing health and safety laws and regulations, linking people to needed health services, assuring a competent health workforce exists, researching and evaluating programs, and advocating for CVE-related laws and policies that guard against civil liberties violations and stigmatization.

Protective Factors and Interventions: Scope and Subjectivity

Wynia asked whether the types of interventions under consideration play an important role. If someone is inaccurately predicted to be on the path to violent behavior, he questioned whether it matters which interventions are available in terms of making that judgment. Steele replied that it depends on the scope of the program and the nature of the intervention. If the scope of the program is violence prevention writ large, for example, and the type of intervention is social services, then those services may help an individual to better adjust and could potentially prevent that individual from going on to commit violent extremism. Even if the individual goes on to join a gang, she remarked, it is still positive value for that individual if violent behavior is averted. She contended that when it is a positive intervention in terms of social services, the risk of a false positive may be mitigated, and suggested that the key is to implement an iterative process to determine the right scope of programs, find positive opportunities for intervention, and improve the specificity of intervention tools.

Interventions that practitioners may view as positive and free of risk may not be perceived or experienced as such by the person who receives it, Wynia pointed out. For instance, a practitioner might think that there is no harm in engaging with a person who seems to be on the path to violence and referring the person to counseling. However, if the consequences of that counseling referral are loss of employment and social standing in the community, then that was neither a risk-free nor an inevitably beneficial intervention. Wynia explained that as a practitioner, he tends to categorize public health interventions as binary: either positive or negative. Positive interventions are those that promote health; negative interventions are ones that prevent the spread of disease. He suggested that there may be a gray area in between where, depending on the person's perspective, an intervention may be positive or negative. He noted that this type of subjectivity seems particularly common in the CVE arena. Steele suggested that integrating health approaches into CVE efforts could bring valuable lessons about how to mitigate that type of stigma, as well as offer guidance about tailoring interventions to reduce the potential for stigma.

In discussing protective factors and the scope of interventions, Wynia noted that much discussion has centered on moving the bell curve by looking at populations and the social determinants of peoples' choices and behaviors. He asked about the appropriate balance between a focus on shifting the bell curve, and a parallel focus on identifying and intervening with people on the path to violence. Steele remarked that CVE is a challenge faced by the whole of society, which thus requires “whole of society” approaches. She suggested that the partners involved in improving and amplifying protective factors may be different than the partners involved in the more narrowly focused interventions. She reported that discussions with partners in the education sector have focused on what type of work can be done to improve protective factors, to deliver positive messages, and to foster a sense of inclusion in schools. She suggested that these are the types of interventions that are more likely to have broad public support, because they are ones in which a range of partners can add value. As a caveat, she noted that those types of interventions should not be implemented to the exclusion of other types of efforts that may have more difficulty in garnering public support.

Public Health Approach to Building Community Resilience

Eisenman called for the public health sector to help promote the concept of community resilience in the context of violent extremism.⁵ He defined community resilience from the perspective of public health as “the capacity of the community as a whole to prepare for, respond to, and recover from adverse events and unanticipated crises that threaten the health of all,” and suggested that it could be reframed in a way that is pinned toward achieving CVE-specific outcomes. He introduced a public health theory about how to build community resilience after a disaster, and suggested that a similar theory could be effective in the CVE arena (see Figure 5-5).

FIGURE 5-5

Theory of community resilience. NOTE: NGO = nongovernment organization. SOURCE: Eisenman presentation, September 8, 2016.

According to Eisenman, the theory posits that factors contributing to community resilience include wellness, access, education, engagement, self-sufficiency, partnership, quality, and efficiency. These factors serve as levers to bolster the core components of community resilience within society.

Hick observed, from a practical standpoint, that there is ample room for improvement. He offered the example of overwhelming meningococce-mic disease: by the time the disease is eventually recognized and diagnosed in the emergency department, the patient will have a high risk of death even with the best treatment possible. But in the early stages, there is nothing a physician can do to recognize that it is different from any other viral illness, so the only strategy is to vaccinate. Applying the disease analogy to CVE, he suggested there are circumstances in which primordial and primary strategies must provide thrust and “inoculate” the community with culturally competent and specific values. Hick said,

I can tell my kids hate is not a Christian value, and you can tell your kids, if you're Muslim, that hate is not a Muslim value. How do we continue to build that into the system? Movement of small shifts in the population may make a difference.

Wynia asked whether there may be any potential drawbacks or disadvantages linked to efforts to promote resilience. He went on to say that the only place he had encountered negative feedback about such efforts was in clinical care settings; the perception among the doctors and nurses was that it was an effort to prolong their ability to work. He questioned whether securitizing or medicalizing these types of community-based interventions would run the risk of a similar response from the public that resilience is being used as a tactic to control political beliefs under the banner of a healthy community. Steele emphasized that resilience is not intended to be about belief system control, but about channeling resilience to nonviolent action. She suggested that there are ways to reduce the risk of violence while still giving a voice to those grievances.

Working on Violence as a Public Health Issue: Reflections from a Practitioner in the Field

Wen introduced the concept “hurt people hurt people” as framing her work in addressing violence from a public health perspective. For example, she reported that among the people currently incarcerated in Baltimore City, 40 percent have mental illness and 80 percent use illegal substances; nationwide, only 11 percent of addicts can get the help they need. She remarked:

We continue to treat addiction as a moral failing when in fact it is a disease. . . . Whose fault is it when individuals with addiction end up leaving jail without having their addiction treated in the first place, and then end up in the cycle once more because they do not have employment or because they do not have housing? Whose responsibility is that to break that cycle? We see that as our responsibility in public health to start that conversation, but with the help of everybody else. . . . We must dig deeper and not just see that person in front of us as the perpetrator of violence.

Wen related a formative experience that informed her understanding of violence as a public health issue. While she was working as an emergency room physician, a teenage patient was admitted with gunshot wounds to the abdomen and chest; health care providers in the room recognized him as having been admitted multiple times with various violence-related injuries over the course of the previous year: a gunshot wound to the arm, a hand fracture from punching someone, and a head injury received in a fight. She characterized him as a powerful exemplar of being caught in a cycle of violence, a cycle in which the perpetrators of violence are often the victim of violence as well. Wen cautioned against looking at what individuals are doing in a single instance without recognizing the policies and circumstances that led to the behavior, such as discriminatory policies in policing, mass incarceration, incarceration of people who have addictions, and treating addiction as a crime and not as a disease.

Wen offered three messages based on her experience working on violence through a public health lens. First, she urged the group to dig deeper and to look beyond individuals as the perpetrators of violence, and to recognize that they are the victims of deep trauma. She explained that in Baltimore City, an ambitious program is in place to train every front-line worker in the city (including social workers, doctors, police officers, and teachers) on trauma-informed care in order to strengthen their abilities to understand, respond to, and treat the impacts of trauma. More than 1,200 people have been trained so far, and the next step is to train every individual in the city who will somehow be in touch with a child to recognize and understand the effects of trauma.

Her second message highlighted the need to invest earlier. In Baltimore, the City Health Department houses the Office of Youth Violence Prevention, predicated on the idea that investing in early childhood is the best way to prevent violence. She described a program in Baltimore City called “Vision for Baltimore,” which was founded in 2016 to achieve one basic goal: to provide glasses to every child in the city who needs them by screening all students in Baltimore City public schools between 2016 and 2019.⁶ Less than 20 percent of children who are screened as needing glasses actually have them, leaving up to 10,000 children in the city with uncorrected vision. She emphasized that the simple investment in a child's glasses can potentially be a violence prevention strategy: it may prevent a child from being labeled disruptive and losing traction in school to the point that he or she engages in a cycle that ultimately leads to violence. She stated that investing in children as early as possible is crucial. A program called Baltimore for Healthy Babies, founded in 2009, has thus far resulted in a 28 percent reduction in infant mortality and a 36 percent reduction in teen birth rates. Nurses, social workers, and case workers visiting homes to teach about safe sleep practices and other child rearing practices not only improves a child's educational outcomes, but also reduces the likelihood of the child being involved in violent activities later on in life.⁷

Wen listed her third point as being at the core of public health: looking at the cost of doing nothing. In a sense, she noted, the face of public health is one that does not exist because it saves lives without recognition—from the prevention of food-borne illnesses, to interventions that help mediate conflicts or improve a child's educational outcomes, to intervening in the pathway to violence. She urged everyone involved in public health to talk about the cost of inaction, particularly in high-risk communities.

She explained that the Baltimore Safe Streets program,⁸ based on the national Cure Violence program, continues to face funding problems on a yearly basis, despite former gang members and criminals making efforts to stop conflict (mediating nearly 700 conflicts in the previous year) and to give back to their communities. She cited a forthcoming study from Johns Hopkins University that found the Baltimore Safe Streets program to be more effective than any other public health intervention in the previous 10 years, at a cost of just $1.7 million per year, which is dwarfed by the millions of dollars it saves in Medicaid, disability, unemployment, and incarceration. The program is so effective because it is implemented by individuals who have literally walked in the shoes of the people they are helping. Similarly, needle exchange programs have been in place in the city for more than 20 years, reducing the percentage of individuals living with HIV from IV drug use from 64 percent in 2000 to 8 percent in 2014. She posited that its success is driven by the fact that people working on the program are people who are themselves in recovery or living with HIV or hepatitis.

Cure Violence Program: Promoting Local Credible Messengers

Jalon Arthur, director of Innovation and Development for the Cure Violence program, School of Public Health, University of Illinois at Chicago, explained that the national Cure Violence program⁹ is an evidence-informed health approach that views and treats violence in all its forms as an infectious health issue, because violence is the leading cause of death in the communities and populations the program serves. He pointed out that like other infectious diseases, violence is not distributed equally: there are epicenters in communities, places like Chicago, Honduras, and Syria. He remarked that violence spreads through exposure, with violence begetting violence: whether it is acts of violence in the home, in the street, in war, or acts of violent extremism.

Many people in communities do not want to see their loved ones or their peers become radicalized, but at the same time, they do not want to see their loved ones or their peers arrested or imprisoned, explained Arthur. Thus, there may be a host of missed opportunities to effectively interrupt the process of radicalization because many people will simply not engage with law enforcement for fear of imprisonment of themselves, loved ones, or peers. In many cases, there is information that is under the radar of the CVE space but known by someone at the community level who does not know how to respond to it.

Credible messengers at the community level can take advantage of their own powerful networks to facilitate a system for efficient detection and response, Arthur emphasized. Empowering community members to serve as a bridge providing critical services for vulnerable people is a critical piece of what Cure Violence and other health approaches can offer, he said. In a community that has a network of credible messengers who are promoting the process of radicalization, it is crucial to empower another smaller credible group within that community to push in the opposite direction away from radicalization.

Training individuals and community groups to recognize people at risk is critical, he explained. A common thread among those who become radicalized is that somebody close to them notices that something is “off” or wrong. By educating a community through the Cure Violence program, or other health-based violence prevention approaches, and by using a pool of credible messengers, a community is able to detect and interrupt this radicalization process. This can provide a trusted and respected alternative to contacting law enforcement. This has been effective in Cure Violence programs within communities where daily shootings used to take place, but have now gone 1–2 years without a single shooting or a homicide. This is an average of a 40–70 percent decrease in violence, according to five independent evaluations, Arthur explained.

Another component of the Cure Violence program is the convening of trusted stakeholders in community forums, town hall meetings, and conferences to strategize and put forth solutions on how to address an epidemic of violence. Arthur explained that these venues offer opportunities to air grievances, share and challenge ideas, exchange perspectives, and present peaceful alternatives. Trusted community insiders can meet without judgment (in the absence of law enforcement), serving as a powerful vehicle for encouraging positive recruitment and allowing people who feel alienated to become part of a bigger group dynamic. He suggested that in the CVE space, given the right trusted messengers, similar approaches could provide platforms for deradicalization.

In regard to CVE, Arthur observed some recurring narratives and concerns that warrant consideration. One is that with a low incidence of violent extremism, many communities are reluctant to buy into CVE efforts at the expense of other issues pervading their communities. Another common concern is that the government has a vested interest in vulnerable communities only to the extent to which the threats—either real or imagined—pose to the larger society. The takeaway for some people at the community level is that whole communities are permitted to suffer through tragedies and more traditional forms of violence, as long as the problems are contained within that community and do not escalate to what would be considered violent extremism.

He urged that these sentiments should be taken into consideration and warned that programs perceived to be government led can often be counterproductive and lack community-level buy-in, without which the program will not work. He emphasized that even the most well-intentioned and best thought-out efforts and interventions can fail if they are not supported by local community champions who can leverage their own credibility and encourage others to understand that it is a health issue worth engaging.

PUBLIC HEALTH MODELS FOR EVALUATING CVE PROGRAMS

Evaluating Community-Led Interventions

Ramchand remarked that there is a tradition in public health to begin working toward prevention immediately when a crisis emerges, even before the mechanisms that underlie the outcomes are fully understood.¹⁰ Ramchand explained that this tradition holds in the CVE domain: while experts continue to do the important work of investigating the factors that lead certain individuals to violent extremism, community-led organizations and programs are already on the front line implementing interventions to prevent it. However, he emphasized that because little is known about the actual efficacy of the CVE interventions, it is now critically important to scientifically assess and evaluate those programs. He contended that given the resource-constrained environments for funders, assessment and evaluation can be used to guide decisions about which programs should be sustained with funding. Evaluations of how different interventions and activities are working can also inform the science of understanding the potential trajectories of violent extremism, he suggested. A further benefit he highlighted is that instilling a culture of evaluation within those community-led organizations allows them to take ownership of identifying and initiating their own program changes and improvements as needed.

Developing a Tool Kit to Aid Community-Based Organizations: Getting to Outcomes

RAND was commissioned by DHS in 2011 to develop a tool kit to aid community-based programs in evaluating their own CVE programs, explained Ramchand. As an initial step, the researchers turned to an existing resource called Getting to Outcomes, an evidence-based approach to evaluation that has been used to aid many community-based organizations in the public health sphere in assessing their own programs. RAND adapted the model to create a similar tool kit for community-based programs seeking to prevent suicide. He noted that there are parallels between suicide prevention programs and CVE programs. For example, the outcomes the programs are seeking to prevent are significant but also very rare, low-base-rate events with high levels of false positives. He observed that this can deter organizations from evaluating their programs, which hinders their ability to adjust and improve. To create a model for evidence-based program evaluation that would be applicable to a diversity of program types, RAND reviewed a corpus of relevant research, drawing on 166 peer-reviewed evaluation studies of suicide prevention programs.

Ramchand explained that the next step was to apply this work toward building a five-step tool kit for CVE program evaluation:

• Step 1: Identify your program's components, and build a logic model.

• Step 2: Design an evaluation.

• Step 3: Select evaluation measures.

• Step 4: Analyze evaluation data.

• Step 5: Use evaluation data to improve the program.

He focused on the formative research that was carried out to understand the activities and outcomes common to U.S.-based CVE programs in order to develop the CVE-focused content for Steps 1 and 3. The first step involves programs identifying core competencies and organizing those components in a logic model; he noted that effective evaluations require all of the components of a program to be well specified.

Ramchand detailed how after programs create their own logic model, they are linked to appropriate evaluation measures and available validated metrics that program managers can use to assess and measure the programs' outcomes (Step 3 in RAND's five-step tool kit).

To identify the activities, outcomes, and metrics relevant to the CVE space, Ramchand reported that the researchers performed a literature review of prior assessments of CVE programs' effects and limitations. A systematic search of 200 CVE assessment publications produced only seven studies that assessed CVE outcomes and demonstrated the potential effectiveness of those programs. However, he noted that the scientific rigor of those studies was extremely limited compared to other academic disciplines (e.g., lack of control group, lack of empirically validated metrics). A survey was also carried out to identify activities and outcomes common to U.S.based CVE programs. He reported that researchers were able to carry out successful interviews with 28 of 94 CVE programs contacted (33 percent response rate; program focus: 46 percent Islamic extremism, 54 percent other forms of extremism).

Reasons for Lack of CVE Program Evaluations

To tailor the tool kit for CVE, Ramchand explained, program administrators were asked specifically about evaluation activities and why such activities were not pursued. He reported that the most common reasons cited for the lack of evaluations were lack of resources and administrator uncertainty about data. Regarding the former, Ramchand explained that administrators responded that they did not have enough resources to follow up with participants or conduct surveys, that they had few (if any) staff who are familiar with evaluation, and that they prefer to devote scarce resources to program administration.

Ramchand noted that administrators' uncertainty and confusion about data revealed multiple concerns: they were worried about appearing intrusive in asking for structured feedback from people who participated in the program; they perceived available data options as uninformative (especially with respect to the Likert scale); and they felt uncertain about the most appropriate metrics for their programs. Furthermore, administrators reported struggling with how to measure and demonstrate the counterfactual—the number of extremist acts the program had prevented. Ramchand suggested that employing proximal outcomes can resolve this issue, allowing administrators to focus on their programs' near-term objectives.

Shaping the CVE Tool Kit

Based on the information gleaned from the literature review and survey, Ramchand noted that the researchers returned to the logic model framework to create a tool kit that was designed to effectively address the barriers identified. He explained that to identify relevant outcomes, programs must clearly specify the audience their activities are targeting. That is, different activities may be geared toward targeting the subset of individuals at risk of committing violent extremism, versus those targeting the families, communities, and institutions that influence or interact with individuals at risk of committing violent extremism.

Once a target audience is specified, the tool kit helps guide programs into considering their likely program objectives. He observed that programs targeting individuals are generally focused on the following outcomes:

• Countering violent extremist opinions and ideology

• Improving psychological well-being and addressing moral concerns

• Enhancing positive social networks

• Reducing political grievances

• Improving social and economic integration

In contrast, he observed that programs targeting the communities surrounding individuals at risk of extremism are generally focused on the following objectives:

• Helping community members understand and identify violent extremism and its risk factors.

• Building community capacity to identify and engage with those at risk.

• Building the capacity of positive, influential members to credibly counter violent extremist ideologies.

• Creating environments that accept minority groups.

• Promoting policies that address political grievances.

• Strengthening capacity to curtail violent extremism.

He explained that identifying their objectives helps guide programs as they customize their logic model. They are then consequently provided with the appropriate available metrics that they can use to assess whether the program is stimulating the intended types of change. Furthermore, he noted that the tool kit provides comprehensive guidance for all five steps of the tool kit, including how to design an evaluation, analyze evaluation data, and use evaluation data to improve programs. The aim, according to Ramchand, is to create an ethos of continuous and ongoing evaluation within programs. Feedback from administrators who piloted the tool kit suggested that cultivating this ethos is crucial, because community organizations will not use the tool kit without active efforts to educate them about the value of evaluation.

Applying a Public Health Approach to the Evaluation of CVE

Lin traced the development of a pilot evaluation program for preventing violent extremism in the greater Boston area. She noted that the project's timeline allowed for independent external evaluators to become involved in a formative evaluation capacity before the pilot was implemented, which informed the program's design.

Lin reported that more than 50 stakeholders from 45 organizations took part in the initial interview process.¹¹ Two interviewers visited each interviewee's workplace, which were primarily community-based organizations, to gain contextual information about the communities they serve, the programs in place, and the interpersonal relationships at play. Ultimately, the researchers conducted more than 24 hours of interviews and collected approximately 2,000 statements. She emphasized how important it is for evaluators to ensure transparency and timely communication in order to foster trust among interviewees.

The next step, according to Lin, was to identify challenges faced in the design and implementation of violent extremism prevention programs based on the stakeholders' feedback and a thorough review of available literature. The principal challenge they identified concerned the name and scope of the programs; she explained that the controversial history associated with the term CVE was discouraging stakeholders' participation. She estimated that 98 percent of interviewees stated bluntly that they would not take part in a program with the CVE label because it would risk undermining the trust and relationships they had worked to build with the communities they serve. Furthermore, she noted a lack of support for any violence prevention program that is too narrow in scope, as in focusing only on a particular subset of the population.

The second challenge Lin's team identified is the lack of a clear, agreed-upon definition for violent extremism and of clear outcomes across CVE programming, according to Lin. To illustrate, she cited the National Strategy on Empowering Local Partners to Prevent Violent Extremism (The White House, 2011), which defines violent extremists as “individuals who support or commit violent extremism to further political goals.” She reported that in trying to use that definition as guidance for defining the outcomes of the pilot evaluation program, the researchers realized that the definition fails to specify what type(s) of ideologies are being referred to, what type of violence is being targeted (for example, whether it should include bullying or verbal discrimination), and what is meant by “support” (for instance, the types of social media interactions that constitute support).

Limited evidence regarding the risk factors for violent extremism emerged as the third challenge, she noted. She explained that this challenge in particular limits the application of secondary prevention strategies that are typically used in the public health sector—namely, screening people in order to detect and treat conditions at an early stage. Thus, the limited evidence available about risk factors obstructs secondary efforts to detect and intervene with people who have the propensity for violent extremist behavior. Essentially, this involves screening people in the precriminal space, because population-level screening for violent extremists is not currently feasible.

Lin explained that without the ability to screen as a secondary prevention strategy, the researchers developed a theory of change that spans the primary, tertiary, and quaternary domains. She detailed how in Boston the theory is informed by feedback from stakeholders about how to define the program's success, which falls into three broad categories. The first category, building trust and earning social support, addresses the stakeholders' need to be listened to and have their opinions validated, their desire to prevent profiling, and their suggestion to expand community policing. She explained that the second component is designed to address the second category, fostering civic engagement and cultural awareness. Such activities include the provision of safe spaces for encouraging civic conversation and open forums for nonmainstream discussions, increasing diversity in the system (that is, government, social service providers, and schools) and improving cultural sensitivity, as well as developing counternarratives against violence and discrimination. The third category, according to Lin, comprises activities to treat the root causes of violence and violent extremists by improving conditions to the extent that people can reach their full potentials; this includes investing in school system and education initiatives, expanding youth programs and services, addressing housing issues, and nurturing healthy and safe neighborhoods.

Looking forward, Lin reported that in August 2016, the Massachusetts Executive Office of Health and Human Services issued a request for proposals (RFP) for the Promoting Engagement, Acceptance, and Community Empowerment (PEACE) Project to be implemented in Boston. Its goals are to prevent violence (and prevent people from joining organizations that promote, plan, or engage in violence) as well as to promote resilience by strengthening protective factors. She commended the organizers for incorporating stakeholder feedback that was shared with them by the researchers of the evaluation pilot program. Specifically, the most recent RFP for the PEACE project focuses on primary (rather than secondary) prevention and completely excludes concepts such as CVE, radicalization, ideology, and risk factors. She explained that instead, the RFP pinpoints specific results and provides clear definitions of violence and violent extremism as “an act that violates state or federal law and causes physical harm to a person, or property and is motivated by hate and/or the intention of domestic terrorism.” She noted that the pilot program has offered its resources for technical support should the PEACE project choose to have an evaluation component of its program.

Evaluating the Capacity of the Mental Health Sector to Support the CVE Enterprise

Stevan Weine, a professor of psychiatry, the director of the International Center on Responses to Catastrophes, and the director of Global Health Research Training at the Center for Global Health, University of Illinois at Chicago, remarked that many new community-based initiatives to address violent extremism are seeking to engage the mental health sector. He described efforts to incorporate the mental health sector into activities to address both violent extremism and violence writ large within the Los Angeles CVE framework. Because a very large proportion of lone actors who commit violence have mental health and psychosocial problems that necessitate mental health intervention, Weine noted, it is a challenge to find the sufficient organizational capacity and professional expertise to conduct mental health interventions of the scope needed for CVE. According to Weine, there is not yet a credible consensus model to implement and support this work in a way that protects civil rights and civil liberties.

Project Eval LA, a collaboration with UCLA, is a partner of the Los Angeles Region Intervention Steering Committee,¹² and it has prepared a forthcoming formative evaluation report assessing intervention efforts in Los Angeles.¹³ Weine shared some of the preliminary findings of the report.

Proposed Los Angeles CVE Intervention Services Flow Chart

Weine presented the services flow chart created by the steering committee, which is subdivided into the three domains of community, assessment, and treatment (see Figure 5-6).

FIGURE 5-6

Proposed Los Angeles CVE intervention services flow chart. NOTES: CRT = Community Response Team; CST = Community Support Team; MH = mental health. This proposed flowchart was not ultimately adopted by Los Angeles. SOURCE: Weine presentation, September (more...)

Weine explained that the services flowchart draws heavily on a threat assessment framework used by the School Threat Assessment Response Team program¹⁴; it is not based on radicalization theory but on a pattern of communications and behaviors that people exhibit prior to committing violent extremism. He explained that within the community domain, a community support team composed of multidisciplinary professionals accepts referrals from a wide range of organizations, conducts threat-assessment training, and forges links with a broad network of professionals or paraprofessionals from various organizations. In the assessment domain, a community-based professional (for example, a school or workplace psychologist) conducts the first level of assessment to determine if there is a credible problem that warrants referral to the community support team for further threat assessment and behavioral assessment, or if other types of care or support are appropriate. In the treatment domain, Weine explained, people who have been screened as having a medium to high level of threat, based on established criteria,¹⁵ receive a thorough threat assessment, a behavioral assessment, and a safety plan, all of which are implemented by the community support team. People referred for care receive ongoing case management, law enforcement referral (if needed), social support services, mental health services, a follow-up plan, and a plan for eventual exit from the program as appropriate.

Tabletop Exercise to Assess Mental Health Capacity

According to Weine, the steering committee has grappled with whether the program should be a targeted violence program that incorporates school violence, workplace violence, and hate crimes, rather than focusing exclusively on violent extremism. To explore this possibility, he reported that the steering committee has plans under way to work with LA County DMH to expand the school violence program into a broader violence prevention program that includes expertise on violent extremism.¹⁶

As part of that process, a pair of tabletop exercises were conducted in July 2016 to assess the performance of the DMH in a CVE situation by using two simulated scenarios of violent extremism. During the exercise, teams of multidisciplinary partners discussed decisions such as determining the veracity of the threat, which types of services to recommend, whether law enforcement should be involved, and how to deal with the community.

Weine reported that the tabletop exercise revealed an instructive panoply of existing capacities and gaps in capacity. It identified 23 existing capacities: for instance, multidisciplinary teams were well established and proficient in working with community members and professionals from multiple disciplines, and the evaluation teams were able to conduct thorough threat and behavioral assessments and cooperate effectively to plan assessments and treatments. However, Weine noted that twice as many gaps (46) were identified as capacities. These included the cultural competency of care regarding Muslim Americans, use of measures and tools, how to make referrals and dispositions, activation and coordination with community leaders and organizations, and monitoring and responding to media.

According to Weine, the exercise revealed that mental health and public health services can make key contributions to targeted violence reduction. Instead of focusing on a very small number of people who may potentially commit acts of violent extremism, the focus with targeted violence reduction is on a larger number of people, organizations, and providers involved in violence prevention.

Finally, Weine suggested that tabletop exercises can be an effective way to jumpstart the implementation of targeted violence reduction programs through engagement with the mental health sector, by fostering trust among disparate stakeholders and community partners,¹⁷ and by exposing capacities and gaps to be addressed.

THE ROLE OF HEALTH PROFESSIONALS IN CVE: LEGAL AND ETHICAL ISSUES

Participants discussed the landscape of health professional ethics within the CVE space, focusing on the roles and responsibilities of health professionals with respect to threat assessment and obligatory reporting. Runnels observed that there seems to be a tension between what it means to be a health practitioner involved with CVE and what it means to be someone who does research in these areas, in terms of ethical issues, legal concerns, and the pressures to build an evidence base and actually take action by implementing programs and measures.

Legal Reporting Obligations for Health Professionals

Wynia noted that incidents of extremist violence often spawn legislative attempts to mandate health providers to report on individuals they believe to be dangerous, citing the wave of bills passed in California, New York, and Tennessee in the wake of the 2012 Sandy Hook massacre.¹⁸ Hick observed that there are difficult questions to answer with respect to the specific role of health professionals in reporting suspicious issues, such as whether and when to engage law enforcement. He explained that legal requirements regarding whether the duty to warn and the duty to protect are obligations for providers vary by state (see Figure 5-7). In states where both types of reporting are obligatory, if a health provider witnesses any specific threat or thinks there is the probability of a violent occurrence, then it must be reported to the relevant responsible public enforcement agency and the person at risk must be informed.

FIGURE 5-7

Reporting obligations nationwide. NOTE: Green: permitted to report; blue: obligated to report; yellow: no obligation to report. * Arizona, Delaware, and Illinois have different duties for different professions.

Hick explained that the 1976 case Tarasoff v. Regents of the University of California served as the catalyst for many of the laws concerning the duty to warn and the duty to protect. The case involved a student in California who reported to his psychologist that he was going to kill his female peer. The psychologist reported the claim to law enforcement, but the psychologist did not inform the student that was at risk and her family. He killed the woman when she returned to campus after a study abroad program; her family sued and won a judgment in the case. In the years since, Hick noted that most states have implemented Tarasoff-like laws that oblige health professionals to report specific and probable threats made by their patients. He continued that in many states, medical providers must report any gunshot wound unless law enforcement is already involved, as well as any cases of suspected domestic violence, child abuse, or child neglect. Wynia noted that there are further legal criteria that can be used to breach personal liberties in certain circumstances. For example, HIPAA allows for disclosure to law enforcement with a warrant, including provisions for “intelligence and national security activities” to assure “proper execution of a military mission,” and to “provide protective services to the President.”

Potential Consequences of Obligatory Reporting for Individuals and Communities

Wynia highlighted the tension between the dual loyalties both inherent and pervasive in health care. Because a small number of people do pose a threat that warrants reporting, health professionals are tasked with striking a delicate balance between their social responsibilities and their duties of care. He emphasized that legislatively mandated reporting inevitably leads to overreporting and false positives, because providers are concerned about the severe consequences of failure to report a patient who goes on to commit violence. However, he pointed out that a false positive report can have a similarly devastating effect on the lives, families, and careers of patients who are reported and then found not to be a threat. Obligatory reporting can also compromise the therapeutic process, according to Wynia, which often requires verbalizing hostile impulses in presumed confidence.

Taking a step back, Wynia observed that there is historical precedence for the practice of requesting (or requiring) health professionals to serve as agents of the state in pursuit of social policy aims under the pretext of protecting the community. He provided several examples to illustrate. The 19th-century diagnosis of drapetomania was applied to slaves who made “unreasonable” repetitious attempts to escape. California's Prop 187, which was never actually implemented, required physicians to report patients seeking care without appropriate documentation of legal immigration status. In Pakistan, a fake hepatitis B vaccination campaign was designed to gather intelligence on Osama bin Laden; as a result, Pakistan is now home to 60 percent of the world's polio cases after trust in the vaccination enterprise was decimated. In 2013 in New Mexico, at least three people were brought into the hospital under suspicion after routine traffic stops and subjected to rectal exams, colonoscopies, and CAT scans—none of which resulted in any evidence of criminal activity.

Wynia reiterated that missteps and overreach in this arena can be catastrophic for patients, referring to the U.S. Supreme Court case of Buck v. Bell (1927), which justified the forced sterilization of Carrie Buck. In his decision, Justice Oliver Wendell Holmes stated:

We have seen more than once that the law may call upon the best citizens for their lives. It would be strange if it could not call upon those who already sap the strength of the State for these lesser sacrifices . . . in order to prevent our being swamped with incompetence. . . . The principle that sustains compulsory vaccination is broad enough to cover cutting the Fallopian tubes. Three generations of imbeciles are enough.

The issue persists today, noted Wynia, who is member of a panel for the American Psychological Association (APA) that is considering how to respond to the Hoffman Report, which showed that key leadership and staff at the APA colluded in the coercive interrogations program during the Bush administration, which included aspects of research on how best to torture people.

Hick remarked that making a report that turns out to be false can have severe and unintended consequences not only for the individual, but for the cultural community at large. Hick cited the example of an emergency medical services (EMS) crew being called to transport a woman who was ill from an apartment complex in Minneapolis in which the majority of residents were Somali. The crew noticed large amounts of electronic supplies, metal cylinders, and other initially suspicious items; they reported it to their supervisor and then to the police. Ultimately it was determined that a man living in the apartment was a radio control vehicle enthusiast. The situation was resolved, but according to Hick, it undermined relations between the EMS and other residents of the apartment complex (e.g., instead of allowing crews to enter to their apartments, residents would bring the family member needing care down to the lobby). He remarked that although the incident was well intentioned, it created a major trust problem. He suggested that while failing to report a threat that comes to fruition in a terrorist attack can have severe consequences, there needs to be appropriate care taken in reporting suspicious activity.

Hick further noted that suspected child neglect can sometimes require a nuanced understanding of cultural context because it can expose the discrepancies between Western medicine and medical traditions of immigrant communities. He explained that providers are protected from legal repercussions if the report is in good faith, but that does not mitigate potential cultural backlash. For example, reporting can result in the state taking protective custody of children who are considered to be at risk based on Western medical criteria but whose parents decline further diagnostics or interventions. He commented that this can severely destabilize the immigrant community's trust in providers (which has happened in both the Hmong and Somali communities in Minneapolis).

Wynia remarked that despite the tenets of the Hippocratic Oath, in which protecting patients' privacy and confidentiality is a key principle, upholding confidentiality is not an absolute value for health professionals. He emphasized that there are no absolute values in biomedical ethics. He explained that the most common means of analyzing ethical problems in medicine is called principalism, which involves balancing all the relevant principles to optimize the overall outcome while conceding that not every principle will be optimized. In the realm of ethics, remarked Wynia, the key criteria in determining when it is ethically justifiable to breach confidentiality include credible threat, significant harm, the presence of an identifiable third party, and the likelihood that the warning will be effective.

Practical and Ethical Challenges in Individual Threat Assessment: Health Approach Perspective

Can Health Professionals Predict Violent Behavior

Health professionals' poor ability to accurately detect a credible threat is a critical concern, according to Wynia, particularly when the determination is used as justification for breaching civil liberties. He cited data from the 1970s suggesting that physicians are no better than chance at predicting whether an individual will become violent. That said, he conceded that the best possible study design to examine that capability would be completely unethical because it would require tracking individuals who are predicted to be dangerous while letting them roam free. However, he noted that a 1967 U.S. Supreme Court decision (Baxtrom v. Herold) released 967 “criminally insane” patients from prison because they were not receiving therapy for their mental illnesses, despite being evaluated annually for dangerousness.¹⁹ The patients were tracked for 4 years after their release, and only 2.2 percent (n = 21) were returned to prison for any criminal act (error rate ~97 percent). A subsequent study (Lidz et al., 1993) revealed these types of predictions of violence have poor differential ability, reported Wynia. Of 357 patients who were tracked in the study, 53 percent of the individuals whose psychiatrists predicted they might commit violence did so within 6 months; however, 36 percent of patients predicted not to be violent actually did commit violence.

Wynia reflected that although practitioners and studies do have a very difficult time accurately predicting individuals who are going to commit violence, it is not impossible. He pointed to the human instinct that makes us feel that it is possible to tell when a person you observe is potentially dangerous. Benjamin commented that individual threat assessment is contingent on a clinician's ability to be astute in the difficult task of “connecting the dots” to recognize that an individual is in the spectrum of threat. He observed that further upstream, public health focuses on risk assessment by astute citizens who “see something and say something.” Both approaches need to be further refined. Miller raised the issue of how exposure to vicarious trauma may lower the threshold for health care providers' perception of individual-level and community-level threats. Stewart suggested that the poor ability to predict violence may be attributable to the wrong questions being asked or, perhaps, because it is not predictable at all, despite how counterintuitive that may seem.

Hick explained that the types of credible or probable direct threats for which providers are routinely vigilant are different than the suspicious activities they are asked to report with respect to the threat of violent extremism. CVE threats also have a slightly different threshold. Hick noted that in other contexts, providers are given anticipatory guidance (e.g., questions about suicide risk or domestic violence) that is based on known risks and relatively well-established community-wide interventions with known efficacy. In the CVE space, he explained, it is very difficult to select the appropriate interventions and to select the population at risk. He suggested that by the time an individual's behavior raises red flags that are sufficient for a health care provider to act from legal and other obligatory standpoints, there are often concerns over whether the evidence is sufficient to mandate an assessment or to place restrictions on that person. He expressed doubt that there is much scope for health care providers (from a recognition and treatment standpoint) to do more than perform the types of mental health threat assessments that are already standard practice, without giving rise to significant ethical concerns.

Hick noted that a common criticism leveled against threat assessment is that certain factors can be more heavily weighed and thus prone to subjective bias, but he remarked that this is also a problem with most psychological assessments. He contended that many nonmental health providers have an instinctive feeling about patients that factors into risk assessment; even though there are some criteria they consider when performing assessments, it is by no means a rigid proscriptive set of criteria.

Threat Assessment Methodology and Expertise

Weine observed that the threat assessment field grapples with the issue of whether the people being targeted in CVE interventions are the same people who go on to commit violence, and if efforts should be targeted on a specific subset of people. He advised that there must be a way to differentiate between a person with an overt and potentially modifiable emotional disturbance who is also considering violent behavior, and a person who is a committed criminal who has just not acted violently yet; only the former person would be responsive to CVE-type case management or treatment.

Runnels noted that public health tends to focus its efforts on the broader community, and questioned what the implications of an individual risk assessment strategy would be when dealing with larger populations. Lin explored the issue of population-level screening for violent extremism in her presentation. She explained that a screening program is only as good as the predictive ability of the risk factors on which the population being screened is based. It is also heavily dependent on the prevalence of what is being screened for, so screening is not usually feasible for rare outcomes. The key question in considering a screening test for violent extremism, she explained, is to determine how likely it is that a person with a positive test result is actually a violent extremist. She described the following thought experiment to illustrate the answer.

As of 2016, the United States population was 323 million, including a presumed 10,000 violent extremists. If there were an extremely powerful hypothetical screening test that could pick up 99 percent of violent extremists when screened, then screening the population would identify 9,900 violent extremists (true positives) and miss 100 of them (false negatives). If the hypothetical test also had a 99 percent specificity, then the test would correctly identify more than 320 million people as innocent (true negatives). However, it also means that more than 3.2 million people will be incorrectly labeled as being violent extremists despite being innocent, which is an extremely high number of false positives. In statistical terms, the positive predictive value of this very powerful hypothetical screening test is just 0.3056 percent.²⁰

Thus, Lin concluded that population-level screening for violent extremism is not possible at the current stage. She clarified that no screening test with 99 percent sensitivity or specificity actually exists; generally, these types of tests have specificity of between 70 percent and 80 percent.

Wynia noted that individual threat assessment does not typically employ population-level strategies because they do yield such high numbers of false positives. Instead, they target a very rich sample of individuals in clinics and emergency departments who have a high probability of violence prior to screening, which makes testing that sample worthwhile. Wynia stated that a specificity level of 70–80 percent is acceptable in a sample where the probability of disease is high: “If we start with 9 out of 10 people who are going to be doing something bad anyways, then having a 70 percent test is good enough.”

Hick commented that Bayes' Theorem is also relevant: when testing a high-probability population for the presence of a disease, the performance of that test is likely to be much better and much more predictive; if the test is applied to everyone in the population regardless of risk, then the performance of the test becomes significantly worse and results in a tremendously higher number of false positives. He emphasized that deciding who should be screened is incredibly important, regardless of the performance characteristics of the test. Lin remarked that this type of testing is only one indicator that factors into clinical judgment and not its sole basis. She highlighted the importance of differentiating between assessment tools and diagnostic tests: assessment tools attempt to predict a future outcome (as in suicide assessment). Diagnostic tests confirm or determine the presence of a condition. Beyond clinical decisions, she argued, the focus should be on ethical issues as well as understanding the limitations of these tests in order to improve them.

Wynia remarked that typical types of nondichotomous clinical testing (e.g., predictions about the likelihood of a heart attack or of losing a limb due to diabetes) operate on a receiver operating characteristic (ROC) curve, which requires determining what counts as positive and what does not. Moreover, the same will be true of threat assessments; what “score” on a threat assessment should count as a positive or negative result? He explained that the task of threat assessment in the CVE arena becomes even more complex because unlike most medical testing, the intervention of doing the assessment itself can end up affecting the person's subsequent behaviors (he likened this to a kind of Hawthorne effect). He noted that there are analogues in medicine: it is possible, for example, that after telling a patient that he or she is on a path toward diabetes based on the hemoglobin A1C test, the patient suddenly changes his or her behavior. Even though the hemoglobin A1C test actually is predictive of progression to diabetes, some people get that information and change their behavior as a result. For any test that is a predictor of future activities over which the individual has a choice, he explained, it is more challenging to calculate things like an ROC curve, positive predictive value, or sensitivity/specificity (he noted that these are commonly used terms in public health and medicine, but they are difficult to apply to behavioral issues).

Eisenman clarified that there are other tests outside of the domain of the threat assessment approach, which itself is not supported by a huge body of evidence. However, the school-targeted threat assessment approach is considered the best practice based on the models built out at Virginia Tech. He explained that the approach is not dichotomous in the “no violence” versus “yes violence” sense; rather, at the barest minimum, it categorizes individuals as low, medium, and high risk. Furthermore, he noted that medium-risk and high-risk individuals generally are not lost to follow up because almost all of them continue to receive ongoing follow-up services. In that sense, he argued, the school-targeted threat assessment approach is not strictly a test but actually a complex intervention.

Weine cautioned against lumping together the psychiatric and psychological assessment of suicide and homicide with the practice of threat assessment. He suggested that while the average clinical mental health professional does have a poor ability to predict violent behavior, the emerging discipline of threat assessment is an entirely different, highly specialized discipline beyond the realm of the average emergency room physician's or outpatient clinician's expertise. He remarked: “Nobody is trying to take health professionals and turn them into proxy for law enforcement to do investigations. This would be a serious error, ethical and otherwise.” He noted that that there is a body of specialized knowledge and practice that is being applied to the task of training to assess threats. However, he emphasized that threat assessment is not a “witch hunt” or designed to encourage people to report on each other to “drum up cases.” Rather, he explained that the aim is to create resources within institutions so when people do make threats, somebody with the appropriate expertise is assessing them.

Ramchand remarked that although discussions of screening protocols and their predictive power are important, there is also a need for more discussion about how to intervene with people who screen positive. In suicide prevention, he noted, an effective and low-cost intervention (that has been replicated) is to send a letter to people after they have been discharged from outpatient clinics that says “You may do fine, that's great, but if you're not, we're here for you, and you should come back.” He argued that if such low-cost interventions might help even a small percentage of individuals they reach, then that should supersede concerns about the false positive rate.

Cultural Relativism: Implications for Mental Health and Threat Assessment

Hick described how the Minneapolis, Minnesota, area is home to more than 25,000 Somali immigrants (representing one-third of the U.S. Somali population), many of whom maintain very close ties to their homeland. Assimilation-related issues are very common in the community,²¹ and they are complicated by longstanding and targeted efforts by Al-Shabaab and other organizations to recruit community members to enter conflicts abroad.²² Hick noted the widespread belief that Somali youth in Minneapolis are susceptible to radical messages and are at potential risk of committing violent extremism. He further stated that the vulnerabilities that drive certain individuals into such ideologies are not necessarily predictable on the basis of sociodemographics or various types of anticipatory guidance for health providers.

To illustrate, he described how the perception of mental health and disease is very different among the Somali population than among other populations. While stigma surrounding mental health issues still persists in the United States, he explained that mental health is immensely more stigmatizing in Somalia, and persons are usually categorized in strictly binary ways: crazy or sane. A person deemed crazy will often suffer social indignities, and their family will also be ostracized from fear of a genetic component of the mental condition. Yet, because of the amount of psychological trauma, physical trauma, and abuse that they have endured, he reported that more than one-third of Somalis have diagnosable mental health conditions. Many of their mental health conditions manifest in physical ways, such as chronic abdominal pain and headaches and make them more difficult to conceal from others. He explained that to find out what is actually going on with a Somali patient requires setting aside at least 30 minutes and saying, “Tell me the news.” For Somali patients, the “news” covers everything that is happening in the community and its impact on their personal physical and emotional health. Because providers' time constraints make such lengthy visits infeasible, Hick explained that providers have established a specific Somali clinic staffed by Somali liaisons that are available to devote adequate time and then liaise about what they learned with a team that includes medical professionals, allowing for more sensitive detection of mental health issues, as well as assessment for potential risk.

Ongoing trust issues and concerns about profiling have proven challenging. Hick noted that progress is very slow, and a single high-profile incident can be a huge setback. However, he suggested that there may be an opportunity to use funding directed to CVE—regardless of ideological or procedural concerns—to promote community-wide equity of access to chronically underresourced mental health services.

1 

Source: U.S. Centers for Disease Control and Prevention (CDC), National Center for Injury Prevention and Control, Division of Violence Prevention. For more information on the Public Health approach see http://www​.cdc.gov/violenceprevention​/overview​/publichealthapproach.html (accessed November 8, 2016).

2

He noted such behaviors are often referred to as precriminal, but clarified that it is not a public health term and its use may be counterproductive and perpetuate stigmas.

3

For example, the Los Angeles Targeted Violence Threat Assessment Program is a secondary prevention program for people who have been identified with precursor behaviors but who have not yet committed a violent extremist act.

4

5

He noted that several national policies now incorporate community resilience: Federal Emergency Management Agency's (FEMA's) National Disaster Recovery Framework; National Health Security Strategy of the United States (2009); Public Health Preparedness Capabilities; and FEMA ESF #14: Long-term community recovery.

6

7

8

9

10

For example, the Gay Men's Health Crisis was formed in New York City before HIV was identified.

11

Types of agencies (n = 45): academia (20 percent) and health care and mental/behavioral health (16 percent), which she noted are not mutually exclusive; community-based organizations (40 percent); and government agencies (including law enforcement and schools) (24 percent).

12

As discussed by Joumana Silyan-Saba earlier in the proceedings, the Los Angeles Region Intervention Steering Committee is convened and coordinated by the Los Angeles Mayor's Office in partnership with the DHS LA Regional Office. It has used a joint public health and mental health approach to help augment the Los Angeles CVE framework by building logic models, a services flow chart, evaluation tools, training plans, and other types of supporting materials needed to implement a program.

13

He explained that such reports are commissioned to ensure that a given program is well formed, is well developed, has clear goals and objectives, has clear outcomes and outputs, has a hypothesized process based on sound theory, and is evaluable.

14

Run by the LA County Department of Mental Health Emergency Operations Bureau.

15

In a given year, he reported that the department receives between 3,000 and 4,000 referrals for assessment; around 100 people considered to be medium or high risk are screened.

16

He noted that the DMH is the only organization in Los Angeles with the capacity and professional expertise to perform this type of intervention work on the large scale required (10 million people in 188 cities within Los Angeles County).

17

Weine reported that during the exercise, the strongest critics of CVE were impressed at the sensitive way in which law enforcement, mental health professionals, and others were able to parse the evidence, make determinations, and share (or not share) information.

18

New York requires mental health professionals to report anyone who “is likely to engage in conduct that would result in serious harm to self or others” to the state's Division of Criminal Justice Services, which then alerts the local authorities to revoke the person's firearms license and confiscate his or her weapons. California mandates a 5-year firearms ban for anyone who communicates a violent threat against a “reasonably identifiable victim” to a licensed psychotherapist. Tennessee requires in-state mental health professionals to report “threatening patients” to local law enforcement, which was passed “in response to mass shootings.”

19

In addition, they had spent more time in prison than they would have had they been convicted of their original crime rather than being judged criminally insane.

20

Positive predictive value (PPV) = 9,900 true positives/(true positives [9,900] + false positives [3,229,900]) = 0.003056 (0.3056 percent).

21

He noted the important benefits to be derived from immigrants maintaining their cultural identities, but that it should be balanced with efforts to ensure that they are accepted (and feel like) members of the community.

22

He reported that in 2015, six people were arrested and indicted for planning to go and join ISIL, and were in various stages of prosecution as of September 2016. Another Somali who left Minneapolis in 2007 has been linked to continuing ongoing aggressive recruitment efforts on the Internet.