You may recall last week a spate of stories and tweets claiming that fake news spreads further and faster on Twitter. For instance, there is Steve Lohr at the NYT, who doesn’t quite get it right:
And people, the study’s authors also say, prefer false news.
As a result, false news travels faster, farther and deeper through the social network than true news.
That struck me as off-base, and you can find other offenders, so I went back and read the original paper by Vosoughi, Roy, and Aral. And what did I find?:
1. The data focus solely on “rumor cascades.” The paper does not establish the relative ratio of fake news to real news, for instance. The main questions take the form of “within the data set of rumor cascades, what can we say about those cascades?”
2. It still may (or may not) be the case that real news has its major effects through non-cascade mechanisms. Most people are convinced of 2 + 2 = 4, but probably not because they heard it through a rumor cascade.
3. Within the universe of rumor cascades, this paper measures average effects. It does not mean that at the margin fake news is more powerful. For instance, the rumor “Hillary Clinton is Satan” may have been quite powerful, but that does not mean a particular new rumor can achieve the same force. 2 + 2 = 5 won’t get you nearly as far in terms of retweets, I suspect.
4. Overall the results of this paper remind me of another problem/data issue. At least in the old days, children’s movies used to earn more than films for adults, as stressed by Michael Medved. That doesn’t mean you have a quick money-making formula by simply making more movies for kids. It could be a few major kid’s movies, driven perhaps by peer effects, suck up most of the oxygen in the room and dominate the market. And then, within the universe of cascade-driven movies, kid’s movies will look really strong and indeed be really strong. That also doesn’t have to mean the kid’s movies have more cultural influence overall, even if they look dominant in the cascade-driven category. In this analogy, the kid’s movies are like the fake news.
5. I am not sure how much the authors of the paper themselves are at fault for the misunderstandings. They can defend themselves on the grounds of not being literally incorrect in their statements in the paper. Still, they do not seem to be going out of their way to correct possible and indeed fairly likely misinterpretations.
6. The strongest argument for the coverage of the authors’ paper is perhaps the coverage of the authors’ paper itself. The incorrect interpretations of the result did indeed spread faster and further than the correct interpretations. I even delayed the publication of this post by a few days, if only to make its content less likely to be true.