Thursday, August 23, 2018

What I Learned From An Analysis of 1.3 Million Articles About Brexit

The UK's vote to leave the European Union in 2016 led not only to intense debate and discussion, it also prompted over one million articles to be written about Brexit. In a fit of summer madness I decided to analyse these articles. These are my key findings:
  • nearly half of all social engagement with Brexit articles was with articles from just three websites
  • partisan content drives higher social engagements
  • Brexit articles that get tens of thousands of social engagements are outliers
  • most social engagements with Brexit articles take place on Facebook
  • the Facebook posts about Brexit that gain most engagement are humorous 

Full details of my analysis and findings are set out below.

It can be difficult to define when an article is about Brexit. I did consider applying a machine learning algorithm to the billions of articles published since 1st January 2016 in order to classify those about Brexit but this is probably something for academic researchers with more time and resources than I have. For this exercise I simply took all articles where Brexit appeared in the article title from January 2016 to 21 August 2018. This gave me 1.362m articles from the BuzzSumo database.

The chart below shows that there was a spike in articles about Brexit when the referendum took place in 2016. Since that time over 30,000 articles are published on Brexit most months and in July 2018 there were over 49,000 articles published.




The chart above also shows the average social engagements for each article, namely likes, shares and comments on Facebook, tweets and retweets on Twitter of the url, Reddit upvotes and engagements and pins on Pinterest. We can see that average engagement of articles has been slowly increasing since June 2016. If social engagements are an indicator of partisanship, as we explore below, then it might be the case that polarisation is slowly increasing or at least not declining. 

The blockbuster viral articles that gain tens of thousands of engagements on social networks are rare. Eighty five per cent of all Brexit articles received ten engagements or less. Only fifteen per cent (192,767) Brexit articles received more than 10 engagements. When we look at the distribution of shares for just these 192,767 articles, see the chart below, we can see the extent of the skewed distribution. Most articles receive little social engagement.


The table is cut off at 5,000 engagements but the line stretches right out to over 900,000 engagements for one article in the Independent. Only 6,225 articles gained more than 5,000 engagements out of the 1.3m articles published. Thus it is clear the big articles with tens of thousands of engagements are outliers. The median number of engagements for these 192,767 posts was 111.

Three Sites Dominate Brexit in Terms of Engagement


Some very interesting patterns emerge when we look at total engagements by domain or website. Articles from just three websites account for almost fifty per cent of all social engagement as we can see from the chart below. Only domains that published at least 5 articles about Brexit are included in this analysis.



The table below sets out the key data for the chart. These are total engagements for all articles with Brexit in the headline published from 1st January 2016 to 21st August 2018. 


The three dominant sites are all large and popular sites but it is surprising that the level of social engagements with Brexit articles published by just these three sites is so dominant. It is also surprising that sites such as the Express, Independent and The Guardian have more than twice as many social engagements with their Brexit articles than the BBC, which has far more website visitors. The BBC is ranked the seventh most visited website in the UK while the Express is ranked fiftieth and the Independent doesn't make it into the top 50. So why does their content get more social engagement?

One reason might be the tribal nature of content on these top three sites. For example, The Guardian and the Independent are staunchly pro-remain whereas the Express is pro-leave. This is reflected in some of their most shared Brexit articles over the last two years, for example:

The Guardian


The Independent


Express


By contrast the BBC content is less tribal, many of its most shared Brexit articles are news based rather than opinion based such as:


I think this provides some initial evidence that more partisan content gains more social engagement. I explore this further below.

Content Types and Partisan Content 


When I looked at the type or format of content that gets social engagement I found that 'why' posts and 'list' posts (any title starting with a number) had the highest levels of engagement. 


The higher levels of engagement with 'why' posts might suggest that people are actively looking for information and wanting to share useful information. Some 'why' posts do tend to be less polemical and partisan such as:


However, the majority of 'why' posts that get well shared are still very partisan. Here are some of the most shared 'why' posts:


List posts are typically more viral and get relatively high levels of shares, see my BuzzSumo research. This Buzzfeed post 27 Brexit Tweets Guaranteed To Make You Laugh, Cry, Or Probably Both gained over 30,000 shares. However, even list posts are used in a partisan way, for example:

10 reasons 'no deal' Brexit would be bonkers

From my analysis it seems that partisan content gains consistently higher levels of engagement relative to more neutral content.  In essence the content that your tribe will get behind and share.

Headline Phrases


I took the 192,000 posts with more than ten social engagements and analysed the top trigrams, three word phrases in headlines. The most used trigram was "after Brexit vote" which was used 1,221 times in headlines.  "No deal Brexit" was used 796 times.  "Trump and Brexit" was used 503 times.  "To stop Brexit" was used 574 times. This last phrase had a higher number of average shares than the other three phrases.

I also looked at the trigrams that gained the most engagements. For this exercise I only looked at trigrams that had been used at least ten times in headlines. Some of the notable top trigrams were as follows, average social engagements in brackets:

Tory MP Jacob (8,985)
UK should stay (7,659)
refuses to pay (6,196)
unite for Europe (5,400)
vote for independence (5,095)
to have final (4,837)
was a mistake (4,798)
stupid decision and (4,463)

Most Engagement Takes Place on Facebook


The majority of engagement with Brexit articles not surprisingly took place on Facebook, which is the largest social network by far. The average engagements with Brexit articles by network is below.

There is a correlation of 0.48 between Facebook Engagements and Twitter engagements. Whilst correlation is not causation this indicates that posts that do well on Facebook may also do relatively well on Twitter. By contrast there is a much smaller correlation between Facebook and Reddit of 0.18. This may reflect the difference nature of the audiences on these two social networks. 

I also looked separately at Facebook posts that mentioned Brexit. The posts with the most engagement tend to be amusing pictures and videos. Below is a selection from the top ten most engaged Facebook posts that mention Brexit.

The few serious Facebook posts that made it into the top 20 most engaging posts on Brexit included a Channel Four post on Angus Robertson refusing to be part of "a diminished little Britain" (223,000 engagements) and a BBC post on the House of Lords vote to guarantee EU nationals' right to stay in the UK (128,000 engagements). 

From this brief analysis it seems people widely share serious articles about Brexit on Facebook but the actual Facebook posts about Brexit that get most engagement are amusing posts with images and videos.

On YouTube the John Oliver video on Brexit was the most popular but a one hour Brexit The Movie video published in May 2016 and designed to inspire people to vote leave was shared over 200,000 times and viewed 2.8m times.