Hardly Anyone Is Publishing Anti-Semitic Tweets

One of the most widely disseminated images created by the brilliant pro-Trump minds who have been harassing Jewish journalists.Photo: spectrereturns/Twitter

Anti-Semitic hate has been a big, unfortunate theme of this presidential campaign: Journalists in particular will tell you that there has been what feels like a marked uptick in online anti-Semitic rhetoric, particularly targeting those who are critical of Donald Trump. Yesterday, the Anti-Defamation League published an analysis of Twitter’s anti-Semitism problem, the purpose of which was twofold: to take a big, zoomed-out look at overall anti-Semitic activity on Twitter between August 2015 and July 2016, and to look more closely at the subset of explicitly anti-Semitic tweets aimed at journalists.

This is an awkward thing to type, particularly as a Jewish journalist who has been targeted by anti-Semitic idiots online, but there’s a strong case to be made that the ADL’s report — and some of the subsequent coverage of it — overstates the extent of this problem on Twitter. That doesn’t mean anti-Semitism online isn’t painful and harmful, or that anti-Semitism isn’t a terrible force in the world, but it does mean that we may be systematically overestimating its Twitter severity in a way that lends significant benefit to Twitter’s small, loud subset of dedicated anti-Semitic abusers.

Let’s go directly to the report:

ADL conducted a search of tweets using a broad set of keywords (and keyword combinations) designed by ADL to capture anti-Semitic language. These keywords did not include any terms associated directly with the 2016 presidential campaign. This yielded 2.6 million results.

These 2.6 million tweets, which were posted by 1.7 million Twitter users, appeared an estimated 10 billion times — which means that this language was potentially seen 10 billion times. That’s roughly the equivalent social media exposure advertisers could expect from a $20 million Super Bowl ad — a juggernaut of bigotry we believe reinforces and normalizes anti-Semitic language and tropes on a massive scale.

Our next step, a manual review of tweets containing anti-Semitic language, yielded 19,253 overtly anti-Semitic tweets mentioning 800 journalists. The 19,253 Tweets were seen approximately 45 million times, and 60 percent of these tweets were replies with anti-Semitic content sent directly to journalists or other users.

Sixty-eight percent of the 19,253 Tweets were sent by 1,600 Twitter users, confirming that these were persistent attacks on journalists by a relatively small cohort of Twitter users. [emphasis mine]

A few things jump out. The first is that, if this method is at all correct in categorizing anti-Semitic tweets, the average anti-Semitic Twitter user tweets just 1.5 anti-Semitic tweets in a year, and that many such tweets are one-off events. Obsessively prolific anti-Semites, then, are just a tiny minority of the total population of Twitter users spreading such hate.

More important, the ADL’s methodology for measuring how often anti-Semitic tweets are seen is pretty broken. It’s unclear exactly what the organization is saying when it writes that those 2.6 million tweets “appeared an estimated 10 billion times — which means that this language was potentially seen 10 billion times.” What is clear is that potential impressions (how many times a given tweet is delivered into users’ feeds) and actual impressions (how many times a given tweet is actually seen) should never be conflated on Twitter. In fact, the service is fairly notorious for the extent to which tweets get launched into the ether, never to be seen, let alone engaged with.

When you look at the math, there’s just no reason to think 10 billion “appearances,” whatever that metric means (I can’t say I’ve seen it before in other discussions of Twitter engagement), translates to anywhere near 10 billion actual impressions. For it to be the case that those 2.6 million tweets were viewed 10 billion times, that would mean each tweet got more than 3,800 impressions. That’s a very high number. By way of comparison, this tweet of mine from last night’s presidential debate — a time when Twitter’s audience, particularly the people I follow and am followed by, was glued to the service — got about 2,400 impressions. This one got about 3,350.

Because I work for a media organization, I have about 11,000 followers, which is a lot more than the average user, but even for those of us lucky enough to have a decent number of followers, low impression percentages are the norm. Here, for example, is a user with 390,000 followers showing that his tweets regularly land in the approximately 5,000 to 7,000 impressions range. It’s safe to say, then, that the average tweet gets nowhere near 3,800 impression. For what it’s worth, a Twitter spokesperson expressed skepticism about the ADL’s 10 billion number to the Washington Post’s Julie Zauzmer: “Impressions are non-public data,” they said. “We don’t believe these numbers are accurate, but we take the issue very seriously. We have focused the past number of months specifically on this type of behavior and have policy and products aimed squarely at this to be shared in the coming weeks.” Obviously a Twitter spokesperson is not an unbiased source, but there are solid reasons to be skeptical about this Super Bowl figure. (There is a good explanation of the basics of Twitter-engagement statistics here.)

As for the journalists being attacked, it’s striking that such a high percentage of those attacks come from 1,600 accounts. This isn’t a “relatively small” cohort of Twitter users — it’s a tiny, tiny cohort. It is almost zero. And as anyone who has interacted with this crowd knows, many of them seem to run multiple accounts — the actual number of Twitter users dedicated to tormenting Jewish journalists with Auschwitz imagery is probably significantly lower than 1,600. Whatever the actual figure is, it’s a particularly hyperactive crowd skilled at projecting the illusion of greater numbers than they actually have. “Bethany Mandel, a freelance reporter who wrote critically about Trump, was also viciously harassed on Twitter,” notes the ADL report. “One user tweeted about her for 19 hours straight, and she received messages containing incendiary language about her family, and images with her face superimposed on photos of Nazi concentration camps.” Anecdotally, this rings true: The sorts of people who would tweet at someone for 19 hours straight are overrepresented in the anti-Semitic Twitter crowd.

Especially when you combine this 1,600 number with the overall low number of average anti-Semitic tweets per anti-Semitic tweeter, it becomes clear that a huge chunk of this problem is emanating from a tiny, tiny slice of the service’s users. And yet they’ve been able to launch a national conversation about online anti-Semitism. This is similar to what happened with GamerGate: Because the movement consisted of obsessive types who would tweet endlessly at their targets (including a smaller subset that would inundate these targets with rape and death threats) — and because it can be difficult to determine scale and significance on a radically flat platform like Twitter — the movement felt bigger than it really was. There were never that many GamerGaters, and there are (I would wager) far fewer openly anti-Semitic Twitter users. In both cases, we’re talking about people who are extremely good at making a lot of noise online, in spite of rather small numbers.

An important caveat: To the victims of the recent uptick in anti-Semitic tweets — and there does seem to have been an uptick that is connected to Trump, both anecdotally and per the ADL’s figures — there’s thin comfort in any sort of rational, analytically inclined attempt to understand how prevalent this stuff is. Because of the dynamics of Twitter, even a small crew of dedicated harassers can feel like an army, and in this case that fake army is spewing some of the most emotionally charged hate imaginable. Plus, it’s undeniably the case that Twitter is often terrible at responding to instances of extreme harassment and threats — I have firsthand experience with this — which is a separate question from just how many of these asshats there are.

But while it’s always a tough sell to claim that a really awful-seeming problem is less dire than people are claiming it is, and while it’s understandable that there’s such a potent stigma against actually-ing complaints about anti-Semitism and other forms of hate speech, it’s nonetheless important to understand how big the problem really is, as opposed to how crappy it can feel to be on the receiving end of a horde of Holocaust Pepes. It does not appear, based on the ADL’s numbers, that there are all that many people interested in engaging in anti-Semitic harassment. And yet today, people are reading a Washington Post headline relaying the false claim that “in 2016, people have read anti-Semitic tweets 10 billion times, many from Trump supporters.” This is absolutely an overreaction in light of the ADL’s data.

Twitter isn’t the world, of course. There’s certainly a strong case to be made that anti-Semitism is a growing problem at the moment, in the European Union if not in the U.S. as well. It is a vicious, murderous belief system that needs to be monitored vigilantly, forever. But when it comes to our assessment of the online world, if we panic in the wake of a small group of obsessive idiots, we give them power they haven’t earned, and which they will surely abuse. They know that they have our attention, and we keep giving them more and more of it.