Fake news has entered the political dictionary over the last year. Suddenly, politicians and commentators are worried that elections are being influenced by false stories being circulated that appear to be genuine. Social media platforms are under pressure to filter out such stories raising the old questions of censorship and “who guards the guards?” However, evidence on the extent and influence of fake news is thin on the ground.
Let’s start at the beginning and try and unpick the fake news issue. I think there are a few facts we can always agree on.
- Fake news have always provided a rich vein of comic material and general entertainment. Think of April Fool’s day, The Onion, The Daily Mash, National Enquirer, Sunday Sport, etc.
- Fake news have always existed since the printing press was invented. Journalists & editors are human after all so they are just as capable of lying like the rest of us.
- The internet makes it easier these days to distribute fake news.
- Fake news is not the same thing as slanting and twisting the facts which has existed since year dot.
- Fake news can be bracketed with conspiracy theories which again have existed since year dot.
So if we assume that there is more fake news around these days, what questions would we want answers to? My list includes –
- What % of the population have read fake news?
- What % of fake news readers are taken in by the fake news?
- What % of fake news readers who believe the fake news change their voting behaviour?
- What % of fake news readers who don’t believe the fake news change their voting behaviour?
- What are the demographics and political breakdowns for all 3 questions?
To my mind, question 3 is the critical one since if it can be proven that a significant proportion of people change their voting behaviour as result of believing fake news then clearly this is a problem for a democracy. If so, the natural follow up question is who are these people as per question 5? However, question 4 does need asking as well. A case can be made that supporters of a side who indulge in fake news might be disgusted enough by the behaviour to not vote or vote for the other side (think of Republicans put off by “Obama is a Muslim” conspiracy theories). If so, then, fake news might actually be self defeating if for every voter you attract to your side through fake news, you end up losing a voter who is turned off by your antics.
Let’s now apply my evidence hierarchy to these questions:
When fake news emerged last year, I think it is fair to say most people believed it was an issue for the right-wing in politics. Words like Alt-Right, comments made by Donald Trump, features on Fox News would have together left you with the impression that it was the right who were guilty of creating fake news. Furthermore, when it came to the readers and voters, the general assumption was that ignorant and uneducated people would be taken in by fake news and were more likely to vote for right wing parties as a result. This article in The Independent describing an FBI investigation into news sites like Breitbart quotes a lot of unnamed sources and is probably typical of articles describing the writer’s belief that fake news is an issue for the right.
Anecdotes about specific articles shown to be fake abound in the news. In December 2016, The Guardian published a list of articles shown to be fake from a number of countries around the world. Whilst what passes for left and right varies around the world, it is fair to say that most of the fake news listed here were intended to be detrimental to left wing or centrist candidates. This would support the common narrative we see about fake news.
However, listing examples is not sufficient to answer the questions I posed. All anecdotes do is verify that a problem exists but anecdotes are not capable of quantifying the extent and seriousness of the problem.
To get a handle on the extent of the problem as per my first question, we need someone to systematically track and verify the veracity of news stories over time. According the BBC article I quote at the beginning, the fact checking site Snopes has been doing this to some degree. This is not the same as counting all articles over time but by attempting to verify a wide variety of articles over time, they have observed that fake news favourable to the left has outnumbered fake news favourable to the right in the last week. Clearly this differs from the standard narrative that the right is more responsible for fake news.
It is clear though that answering question 1 is going to be hard. The only way to do this with any statistical rigour is to select a random sample of articles on a particular day and verify the accuracy of each article. The sample design would have to take into account the nature of the publisher of the article e.g. newspaper, TV channel, social media, personal blog, etc and would also need to account for the volume of readers. With web articles, Google Analytics would provide that data but print articles would be harder to verify the readership. This points to another issue with question 1. One might be able to estimate the % of articles read that constitute fake news but it is much harder to answer how many people actually read the fake news article and who those people were as per question 5.
What caught my eye about the BBC article was the description of a small experiment carried out by an American advertising company commissioned by CBS who in turn are producing a documentary to answer some of my 5 questions. The experiment involved creating two fake news articles, the first about a supposed police raid on some protestors (deemed to be favourable to the left), the second about a supposed congressional plot to unseat Trump (deemed to be favourable to the right). These articles were then published (where was not stated) and the company then tracked who was reading and who was sharing these articles.
Their conclusions completely contradict the standard narrative. Specifically,
- Left wing readers of the left wing favourable fake news were 34 times more likely to be graduates than the general population.
- Right wing readers of the right wing favourable fake news were 18 times more likely to come from the top quintile in terms of income.
- The more you consume, the more likely you are to vote.
I would have preferred to have seen some actual figures shown rather than relative risk factors (which I am not quite able to get my head around) but the general conclusion seems to be that fake news are much more likely to be read by wealthy, politically engaged graduates rather than poor, disinterested, uneducated people. So whilst the experiment can’t answer questions 1 to 4, it does give an answer to question 5 which surprised me.
Obviously this is only one small scale experiment but one of the signs of a good experiment (which I will post about in more detail at a later date) is whether or not it is reproducible. Reproducibility means an experiment that if it was repeated by a separate organisation, they would end up with a similar conclusion. Furthermore, this experiment can be scaled up to provide harder evidence. For example, I would have created probably 6 fake articles which covered a variety of political themes with perhaps two clearly favourable to left wing politics, two clearly favourable to right wing politics and two that were indeterminate but could be seen favourably by either side. After tracking how these are clicked on and shared, I would have tried to carry out a survey of the readers to find out how much the article was likely to change their voting behaviour which as I have said is the critical question to be answered. This wouldn’t have been easy but it would have been worth doing.
Further questions to consider
The BBC article included some theories as to why left wingers might be more susceptible to fake news. Specifically they raised a belief that confirmation bias was taking place which is the idea that we are more ready to believe and share news that confirms and strengthens our existing beliefs. No evidence was offered for this but one of the reasons why I would try and survey the fake news readers in my redesigned experiment was precisely to test this hypothesis.
If this hypothesis is true, then it suggests that the answer to my question 3 is that fake news does not change people’s voting behaviour, instead it reinforces existing voting behaviour. In other words, our votes won’t change but our willingness to consider and listen to opposing points of views will be reduced. A survey carried out in the aftermath of the 2016 US Presidential Election showed that Democrats were 3 times more likely to unfriend people on social media than Republicans. In my opinion, such an outcome is just as dangerous for democracy and the last thing I want to see is people segregating themselves into echo chambers which have been reinforced by fake news.