What does Big Data tell us about white supremacists?
- By J.M. BergerJ.M. Berger is co-author of ISIS: The State of Terror and is a nonresident fellow at the Brookings Institution.
A funny thing happened on the way to the data farm.
Last year, I started applying Big Data to extremist networks online, with an eye toward measuring influence and engagement among Twitter users following well-known white nationalists. The tools I used to parse the information proved to be powerful, and they offer new ways to test approaches to countering extremism online (see below). But it was the content I found in the network of Twitter users that turned out to be unexpectedly revealing.
The number 1 hashtag used by people following white nationalists online was #tcot — "top conservatives on Twitter." Number 3 was #teaparty, and number 5 was #gop. The content of links tweeted by the users also skewed sharply toward mainstream conservative and Republican content.
The group of followers included people who didn’t overtly tweet about white nationalism or white supremacy, but when the group was narrowed to openly extremist users, the percentages for those three hashtags didn’t go down, they increased.
A complete accounting of methods and results can be found in a paper I produced with software engineer Bill Strathearn, published today by the International Centre for the Study of Radicalization. Data for the paper was collected around the time of the U.S. election last November, which may have skewed the results, but as of last week, the most influential users from the group we studied were still heavily engaged with partisan Republican politics, based on an examination of the 50 most recent tweets from each of the 100 most influential users still active on Twitter.
The top hashtag remained #tcot. Coming in at number 3 was #cpac2013, and #gop was number 4. The tag #nra vaulted from 102nd place in last year’s data to number 6, no doubt due to recent discussions on gun control, while #teaparty dropped to number 10.
I didn’t set out to do a paper with a political dimension, and I wasn’t happy when I saw this data. Until now, I’ve been fortunate enough to work on extremism without getting snarled in contentious and potentially alienating political arguments. But there was no way to ignore the data, and it would have been wrong to try.
The issue of race in Republican politics has been roiling for some time, between the 2012 election — which President Obama won with only 39 percent of the white vote, sparking controversial comments from some mainstream conservatives, such as Bill O’Reilly — and racial extremists who saw an opportunity to push their agenda in the arena of mainstream politics. The issue came up again in force at a recent CPAC event on race relations, where one attendee suggested black slaves should have been grateful for the "shelter and food" provided to them by their oppressors.
The Twitter study shows that white nationalists feel an affinity for and seek engagement with mainstream conservatives, and the finding correlates to lots of other readily available evidence. So what do we do with it?
Based on the data, my chief recommendation is that conservatives need to be at the forefront of aggressively countering racist extremist narratives.
The majority of mainstream conservatives likely find overtures from white nationalists to be distasteful or offensive. But these contacts provide an opportunity to push back, and conservatives are far more likely to have a constructive impact on white nationalists than government-sponsored homilies or messaging by well-intentioned, left-leaning NGOs.
This should be a no-brainer for mainstream conservatives. There’s no downside to pushing back against neo-Nazis. It has positive potential for efforts to counter extremism, and it’s also potentially good for the Republican Party, which desperately needs to demonstrate that it’s tackling its internal race problem if it wants to remain relevant in an increasingly diverse country.
The problem is that the data also show that conservatives are hypocritical when it comes to addressing the relationship between the ideological center and its fringes. A significant number of prominent conservatives insist that mainstream Muslims bear meaningful responsibility for Islamist extremism. But those very same conservatives reject the idea that they themselves should be held responsible for the actions of terrorists like Anders Breivik, or the views of domestic extremists, such as neo-Nazis, Klansmen, and the Aryan Nations.
They can’t have it both ways.
Hypocrisy isn’t the sole province of conservatism, of course, and many conservatives are able to process these issues without falling into that trap. Some left-wingers won’t hesitate to attack right-wingers for fostering white nationalism, while defending Muslims against the very same kind of logic (and blithely ignoring the eco-terrorists and anarchists with whom they share certain values). Some Muslims will be happy to see the tables turned on their critics, even if it means relying on the same sloppy logic used to tar them in the first place.
Then there are the centrists, who take the easy road and simply declare that mainstream political and religious movements are not really relevant to extremism, an attitude that has obvious appeal to many. This might be a noble impulse, or it might be cowardly, but either way, the data show that it’s wrong. Extremism and mainstream politics and religion are linked, and it’s hard to talk about one while excluding all consideration of the other.
When we do talk about the overlap, it’s rarely constructive. If the subject comes up at all, it’s almost always in the context of attack and defense — X group attacks Y group as sharing responsibility for a particular extremist act or movement, then Y group angrily disavows the connection. Everyone gets overheated, each accuses the other of partisan motives, and almost no one tries to address the question thoughtfully. On the rare occasions that people do take a stab at a more careful consideration of the issue, it’s almost always in the context of the other team’s problems and not their own.
This is, at best, a zero-sum game. It further marginalizes people with radical, extremist, or even just nonconformist views, while never actually addressing the question of how the mainstream interacts with the fringes, a question that appears increasingly relevant to making progress on the problem of extremism.
It’s akin to analyzing tornadoes without studying wind, and it’s a crippling blind spot in our approach to the problem of extremism. It happens because people on all sides are more interested in attack-dog politics than in cleaning up their own backyards — or as someone once said, pointing out the speck of sawdust in another’s eye while ignoring the wooden plank in one’s own.
That’s no way to ru
n a political party, a religion, or a country. It’s time for a change.
How We Did It — and What It Means for Countering Extremism
Although the link between white nationalists and Republican politics is provocative, our research was always intended to focus on metrics for combating extremism, and its findings will arguably have a bigger impact on that effort.
My co-author, software engineer Bill Strathearn, implemented a method for collecting open-source data on 3,542 users who followed 12 well-known white nationalists on Twitter, and for scoring those users based on their influence (such as the tendency to inspire retweets) and exposure (a tendency to react to other users). We came up with a list of interaction types to evaluate, and Bill assigned numbers to each, then calculated the results.
We rated user scores using three different approaches. First, we measured just their interactions with the 12 "seed" accounts. Second, we measured just the followers’ interactions with each other. And finally, we measured interactions among the followers and the followers’ followers.
Door number 2 proved to be the most effective approach. When we limited our scoring criteria to only interactions among the 3,542 followers of seed accounts, the highest scoring users for both influence and exposure turned out to be very highly engaged with white nationalist ideology. When we added influence and exposure scores to create a measurement of interactivity, the trend became even more pronounced.
Of the top 50 most interactive accounts, 98 percent tweeted overtly and almost single-mindedly about white nationalism. For the top 100, the number was 95 percent, and for the top 200, the result was 83 percent. (If someone didn’t overtly tweet about white nationalism, that doesn’t mean they weren’t engaged with the ideology, it just means we didn’t see open evidence of such.)
In practical terms, what this means is the scoring system can be used to narrow a large social media dataset down to the most engaged and relevant users, as long as the seed accounts are clearly defined. We ran a comparison analysis on followers of anarchist Twitter accounts, and achieved similar results, although they were somewhat less precise due to the inherently nonconformist nature of anarchism.
Importantly, the scores were not based on any terminology or keywords specific to white nationalism. Based on these results, we believe this scoring system can be applied to almost any ideology where strongly defined seed accounts can be identified.
The scores can be used to track the total amount of engagement in a system, as well as the most influential messages and messengers. They can highlight which ideological thought leaders and ideas are gaining the most traction in a specific ideological group.
The best and clearest application of this tool is for efforts to counter violent extremism online. This approach, known as CVE, has become wildly popular in Washington policy circles as a possible alternative or supplement to counterterrorism efforts.
The idea is to head off violence before it begins by countering the narratives and ideologies of extremists at the thought level, before people turn to committing actual acts of terrorism and violence.
I’ve written extensively about my objections to CVE, both in principle and in practice. One of the biggest problems with the undertaking is that its practitioners are prone to effusive enthusiasm over approaches and assumptions that have never been quantitatively evaluated.
So far, most reported results have been anecdotal and subjective. There’s not much clear evidence as to whether CVE initiatives work at all, and if they do, which tactics work better than others.
If this paper accomplishes only one thing, it should be to end feel-good approaches to combating extremism online. If someone thinks broadcasting positive messages is an effective technique to reduce extremist engagement, let them field test it and we’ll measure. If someone believes disrupting online networks is a better approach, we can measure that too. If doing nothing at all is better than either style of CVE, we can put numbers on that as well.
This research will have other applications down the road, but the most immediate opportunity lies firmly in the realm of countering extremism online. Much ink has been spilled on this topic, and substantial amounts of public and private money have been spent on it. It’s time to figure out what works, if anything, and throw out everything that doesn’t.