Is Facebook causing anti-refugee attacks in Germany?

Here is the key result, as summarized by the NYT:

Wherever per-person Facebook use rose to one standard deviation above the national average, attacks on refugees increased by about 50 percent.

Here is the underlying Müller and Schwarz paper.  They consider 3,335 attacks over a two-year period in Germany.  But I say no, their conclusion has not been demonstrated.  Where to start?

Here is one picture showing a key correlation:

It is difficult to see if there is causation in the correlationThat looks pretty strong, doesn’t it?  Nein!  That is not how propaganda works, as an extensive literature in sociology and political psychology will indicate.  That is how it looks when you measure what is essentially the same variable — or its effects — two different ways.  For instance, that very big spike in the middle of the distribution?  As Ben Thompson has pointed out, it represents the New Year’s harassment attacks in Cologne.  Maybe that caused both Facebook activity and other attacks to spike at the same time?  Will you mock me if I resort to the “blog comment cliche” that correlation does not show causation?

To continue with the excellent Ben Thompson (he is worth paying for!), the identification method used in the paper is suspect, and he focuses on this quotation from the authors:

In our setting, the share of a municipality’s population that use the AfD Facebook page is an intuitive proxy for right-wing social media use; however, it is also correlated with differences in a host of observable municipality characteristics — most importantly the prevalence of right-wing ideology. We thus attempt to isolate the local component of social media usage that is uncorrelated with right-wing ideology by drawing on the number of users on the “Nutella Germany” page. With over 32 million likes, Nutella has one of the most popular Facebook pages in Germany and therefore provides a measure of general Facebook media use at the municipality level. While municipalities with high Nutella usage are more exposed to social media, they are not more likely to harbor right-wing attitudes.

The whole result rests on assumptions about Nutella?  What if you used likes for Zwetschgenkuchen?  Has a robustness test been done?  Was a simple correlation not good or not illustrative enough?  I’ll stick with the simple hypothesis that some municipalities have both more Facebook usage, due to high AfD membership, and also more attacks on refugees, and furthermore both of those variables rise in tense times.  AfD is the German party with the strongest presence on Facebook, I am sorry to say.

You will note by the way that within Germany the Nutella page has only verifiable 21,915 individual interactions, including likes (32 million is the global number of Nutella likes…die Deutschen are not that nutty), and that is distributed across 4,466 municipal areas.  (If you are confused, see p.12 in the paper, which I find difficult to follow and I suspect that represents the confusion of the authors.)  That should make you more worried yet about the Nutella identification strategy.  They never tell us what they would have without Nutella, a better tasting sandwich I would say.

I also would note the broader literature on propaganda once again.  Consider the research of Markus Prior: “…evidence for a causal link between more partisan messages and changing attitudes or behaviors is mixed at best.”  These Facebook results are simply far outside of what we normally suppose to be true about human responsiveness — so maybe the company is undercharging for its ads!

Ben adds:

I am bothered by the paper’s robustness section in two ways: first, every single robustness test confirmed the results. To me that does not suggest that the initial result must be correct; it suggests that the researchers didn’t push their data hard enough. There is always a test that fails, and that is a good thing: it shows the boundaries of what you have learned. Second, there were no robustness tests applied to one of the more compelling pieces of evidence, that Internet and Facebook outages were correlated with a reduction in violence against refugees. This is particularly unfortunate because in some ways this evidence works against the filter bubble narrative: after all, the idea is the filter bubbles change your reality over time, not that they suddenly inspire you to action out of the blue.

The authors do present natural experiments from Facebook and internet outages.  They find that “…for a given level of anti-refugee sentiment, there are fewer attacks in municipalities with high Facebook usage during an internet outage than in municipalities with low Facebook usage without an outage.” (p.28).  Again I find that confusing, but I note also that “internet outages themselves…do not have a consistent negative effect on the number of anti-refugee sentiments.”  That is the simple story, and it appears to exonerate Facebook.  pp.28-30 then present a number of interaction effects and variable multiplications, but I am not sure what to conclude from the whole mess.  I’m still expecting internet outages to lower the number of attacks, but they don’t.

Even if internet or Facebook outages do have a predictive effect on attacks in some manner, it likely shows that Facebook is a communications medium used to organize gatherings and attacks (as the telephone once might have been), not, as the authors repeatedly suggest, that Facebook is somehow generating and whipping up and controlling racist sentiment over time.  Again, compare such a possibility to the broader literature.  There is good evidence that anti-semitic violence across German regions is fairly persistent, with pogroms during the Black Death predicting synagogue attacks during the Nazi time.  And we are supposed to believe that racist feelings dwindle into passivity simply because the thugs cannot access Facebook for a few days or maybe a week?  By the way, in their approach if there is an internet outrage, mobile devices do not in Germany pick up the slack.

I’d also like to revisit the NYT sentence, cited above, and repeated many times on Twitter:

Wherever per-person Facebook use rose to one standard deviation above the national average, attacks on refugees increased by about 50 percent.

That sounds horrible, but it is actually a claim about variation across municipalities, not a claim about the absolute importance of the internet.  The authors also reported a very different and perhaps more relevant claim to the Times:

…this effect drove one-tenth of all anti-refugee violence.

I would have started the paper with that sentence, and then tried to estimate its robustness, without relying on Nutella.

As it stands right now, you shouldn’t be latching on to the reported results from this paper.


The goal is to get all media under control.

What Steve Sailer actually meant is that by blaming Facebook for extremist content, critics hope to regulate speech on Facebook.

Facebook being blamed for spreading extremist content is not the same as saying that the people actually spreading extremist content are the ones to be blamed for their spreading of extremist content. Such as the AfD, who would instantly regulate all speech on Facebook the second the AfD had the power to do so. The sort of people attracted to the AfD tend to be people who feel that a society that allows free speech is a weak society that needs a strong hand to correct its flaws - starting, of course, with forbidding any speech critical of the AfD and its goals.

Steve Sailer is longing for a world where his various ideological fixations concerning humans will not be challenged and mocked mercilessly, when not outright ignored. The strange thing is, that many of the people who share his fixations are generally explicit in their goals of in controlling the media, to keep us from living in a PC hellscape, where 'honest social scientists' need to fear any response except accolades for their 'honesty.'

I have never seen any comment by Steve Sailer that implies he would like to suppress the speech of those who disagree with him.

This is what I said about Sailer - 'Steve Sailer is longing for a world where his various ideological fixations concerning humans will not be challenged and mocked mercilessly, when not outright ignored.' One hopes that is an accurate summation, at least from his years of comments here.

Followed by an observation that is directed at those who share Sailer's opinions in one area, and what they tend to believe - 'The strange thing is, that many of the people who share his fixations....'

In all fairness, this may be more influenced by living in Germany, where the sort of people that share Sailer's beliefs concerning humans do not share American beliefs concerning the efficacy of free speech, apart from using it as a tool to destroy a society foolish enough to practice it. Which is their delusion, speaking as an American who feels the 1st Amendment is one of the greatest accomplishments in human history. But delusional or not, it remains a goal of theirs to remove free speech as a pillar of a free society - mainly because they do not believe in the idea of a free society.


Agreed. "Fake News" and "Hate Speech" are now just memes to cloak a desire for censorship.

Well...concern about the content of speech is certainly legitimate, and I think "Fake News" and "Hate Speech" are legitimate issues of concern.

The problem is what one does with that concern. Targeting the platform as the source of the problem is effectively a call for censorship.

"concern about the content of speech is certainly legitimate" But is it? Why? Would it be so bad if we let everything go?

Great ? I should definitely pronounce, impressed with
your web site. I had no trouble navigating through all the tabs
as well as related info ended up being truly easy to do to access.
I recently found what I hoped for before you know it
in the least. Quite unusual. Is likely to appreciate it for those who add forums or something,
website theme . a tones way for your customer to communicate.
Excellent task.

Would love to incessantly get updated outstanding web site!

Where do you draw the line between censorship and stopping foreign propaganda?

>> "Where do you draw the line between censorship and stopping foreign propaganda?"

I don't draw such a line at all. And even if I could, I wouldn't trust my "betters" with the power to enforce it over me.

Anyway, I find it useful to listen to foreign propaganda. Often it confirms they are knaves, but once in a while they have a point.

Foreign propaganda? Fake news? Nah, for my world news source of choice there is no issue:

I thought I would find some criticism of Tyler's post given the length of what you have written.... never mind...

Why bother with quoting Prof. Cowen's words when it seems to be so easily overlooked, starting with the observation that Facebook is a fairly new medium, and there is not a large body of literature detailing its effects, regardless of what Prof. Cowen may feel.

However, let me expand the response concerning ''that Facebook is somehow generating and whipping up and controlling racist sentiment over time.'

Facebook does not generate content, meaning that Facebook is 'generating' something is basically inaccurate. Further, though Facebook's basic algorithms allow for the amplification of a message, it is not really accurate to accuse Facebook of whipping them up - that reflects (in the main), the efforts of the users of Facebook. And of course, Facebook is vehement in denying that it 'controls' any sentiment on its content neutral service. And even if one accepts a certain kernel of truth within such a poorly formulated perspective (though the formulation is not that of the paper's authors), the reality is that Facebook can be generating and whipping up and controlling multiple sentiments over time, simultaneously.

So, the fact that the authors of the paper are summarized as presenting this perspective is clearly correct, and that it shows a lack of understanding on their part is obvious. Of course, that is not a criticism of Prof. Cowen's post, but of its source material.

Oddly enough, it is actually possible to criticize the same thing as Prof. Cowen appears to, and to demonstrate why it is worthy of being criticized.

Here are some ways that in future comments, if you agree with Tyler, you can make that clear, rather than making it seem like your comment is a rebuttal to the post and attack on its author:

"Tyler makes a good point because..."

"I agree, and also..."

"Yes, and..."

"Even though Tyler is a corrupt Koch-brothers-funded shill and this ignorantly Americo-centric blog is completely laughable, I begrudgingly admit that he is correct when he says..."

Great post. I would like to request that Tyler post angry more often.

+1. I thought it was Alex!

Tyler has fought long and honourably for civility, intellectual honesty, and the liberal tradition. I hold him in the highest regard.

But his cause, the old liberal centre, is failing. And he knows it. Sooner or later, like a Good German in 1938, he will have to choose before following his academic and political tribe into darkness, and following his conscience.

He is a good man. It would be nice if he survives this story.

This time the billionaires are backing the hard left. Cowen goes where billionaires lead.

once again Professor Cowen makes the long russian nights a little less dismal for the rest of us!

dare to dangle

bill maher no es un endocrinologist sexy

And what if the anti-refugee attacks simply correlate with just the number of refugees in the country? That seems like a more acceptable cause. What if refugees doesn't even get harassed more than any other Germans - maybe they should have investigated the attacks/1000 refugees and attacks/1000 native Germans: what if both rose in the same time?
By the way, the huge drop in the number of incidents recently can also be caused by the fact that those who migrated are no longer considered refugees, they earned a citizenship (or they simply assimilated or left the country).

Before anyone else says it, do the posts look like a trailing indicator to anyone?

Of course, they haven't done the stats for that, either.

If you ignore the second half of 2015, it's certainly a trailing indicator. To be fair, I assume that Tyler picked this graph to make a point that the paper isn't particularly convincing.

Also, I am now worried that Nutella is some kind of Neo-Nazi dogwhistle thing that I'm not getting. :-)

Did you just say the N-word?

I am ashamed at my prejudice towards other delicious confectionary spreads with their own stories and identities....

Another dog bites man story of a couple economists writing on a topic other social science disciplines have long worked on and completely ignoring said work.

But nonetheless, we must bow down to our new economist overlords, because they have an identification strategy!

Might I suggest the alternative title of, "Nutella, Making Rainfall Seem Plausible Since 2018"?



Perhaps, if this copy/paste is accurate from the original, I have found the source of Tyler's confusion;
“…for a given level of anti-refugee sentiment, there are fewer attacks in municipalities with high Facebook usage during an internet outrage than in municipalities with low Facebook usage without an outage.”
Read carefully my friends.

Sometimes an outage causes outrage, sometimes the other way.
Sometimes outrage is just the human condition.

"there are fewer attacks in municipalities with high Facebook usage during an internet outrage "

internet outrage? I'll cautiously assume that they meant internet outage. Which forces me to query, how common are lengthy internet outages in Germany?

"The Nutella identification strategy" sounds like the Twinkie defense. That should probably tell you something.

Also, in Fig 3, increases in anti-immigrant posts more often follow, rather than precede anti-immigrant incidents (when there is any evident relationship at all). Suggests authors of this paper may have cause and effect exactly backwards, if there is any relationship at all.

"for a given level of anti-refugee sentiment, there are fewer attacks in municipalities with high Facebook usage during an internet outrage than in municipalities with low Facebook usage without an outage"
If I'm reading this correctly, there is an ongoing effect from high usage, and a here-and-now effect from whatever you read in a particular day, and the here-and-now effect is bigger, but both are real. Reasonable?

In the absence of evidence to the contrary I will assume Nutella was used as one of many measures of Facebook usage. Then the authors used the one that resulted in the best model fit and did not report the other attempts.

And another relevant XKCD:

Two reasons to think you're wrong:

1. If you read the methodology, it required lots of manual work. Unlikely that they had the time and resources to do this for a bunch of different pages.

2. What are the odds that they tried this with a bunch of pages to cherry pick the best one, and then the best one just so happened to be the page with the most likes? The given explanation that they chose the page *because* it had the most likes is far more likely.

It's been embarrassing to see the tech community and libertarian/business types circle the wagons in response to this study. Obviously using proxies and clever data tricks to tease out causation is tricky. Obviously methodologies can always be critiqued and improved. And obviously no one should look at this paper and say that it is now *proven* that Facebook causes anti-refugee attacks. Obviously more research needed. None of that means this study is worthless.

I wonder what it would look like if you also put attacks by migrants on there. My priors are telling me it would look very peculiar indeed.

The garden of forking paths strikes again. Nothing could do the empiricists of the world more good than to read a few dozen random Andrew Gelman posts, grasp how DOA many supposedly empirical papers are, and then redouble their skepticism about the quality of peer review as a filter.

"While municipalities with high Nutella usage are more exposed to social media, they are not more likely to harbor right-wing attitudes."

Is that Germans having a go at being arch? Or just a statement of fact?

May be too late, but can someone take pity and explain what exactly visiting the Nutella social-media page was a proxy for? I don't follow this sort of thing very well, and usually give up quickly if the comments don't illuminate for me (that's right, the comments, TC!) - but food is my bailiwick and so, absurdly, I am feeling a bit stung.

Would Speculoos cookie butter have functioned just as well?

The idea was to measure Facebook usage. So engagement with the Nutella page was used as a proxy for Facebook use generally. The idea is that Nutella is sufficiently "neutral" in the sense that Nutella fans are unlikely to skew in any particular direction, politically speaking.

Does anyone else think it might be telling that Facebook hasn't disputed this study? They could replicate the study using *actual* Facebook usage data instead of the Nutella page engagement proxy and come to much firmer conclusions. But they haven't, so far anyway.

The methodology of this study is *necessarily* kind of a bank shot; measuring Facebook usage from outside the company is inherently difficult. It's easy to nitpick the choices they made. But the loud conspicuous silence from the one entity that actually has access to all of the data and a strong incentive to dispute the NYTimes story is disconcerting.

John cena is soon coming to india to meet all the bollywood stars and cricket fans to know more click here thanks !!!

Comments for this post are closed