Brexit on Twitter Week Ending June 3rd
Overview
The following analysis uses social network analysis methodology created by Right Relevance — more information in this post. Scroll to the bottom to see tables indicating which users were the most influential during the week.
What follows is a personal analysis of some of the interesting elements of the conversation on Twitter during the week.
The network map of the Twitter conversation during the week shows the familiar pattern from previous analysis with a noisy leave side and more integrated Remain side.
In a previous commentaries I illustrated that the leave side appears to be a noisy echo chamber and the remain side much more diverse and that the main PR accounts of both parties were behaving differently and re-enforcing that pattern of behaviour.
I also posed the question of whether their was a media bias towards remain based on the position of the main media accounts in the conversation map.
A study by the Guardian newspaper, however, suggested that there was actually a bias towards leave in the main newsprint media.
Our twitter conversation analysis groups users into communities which we refer to Flocks and Tribes.
Tribes are groups of people who share a long term common interest. In this case member of the same political party are example of Tribes. These groups are slowly changing and reflect a fundamental quality about the members. Tribes are detected by analysing the followers and following of users along with the communication (by retweets, mentions and replies )between users. Here is an example of the UK Labour Party topic on Right Relevance which illustrates this kind of tribe.
Flocks are groups of people who simply communicate on Twitter within a period of time. The existence of a Flock is detected purely by the communication during that period of time. The Brexit campaign is a good example of this kind of flocking behaviour as there are people campaigning together who belong to very different tribes.
The twitter conversation maps above illustrate the formation of flocks in two ways. Firstly the, layout, with users who communicate being closer together and forming into distinct groups. Secondly, the colours. Both the colours and layout are created by algorithms which detect the presence of these groups. In a network of this complexity and size the algorithms detect many thousands of communities the diagrams are greatly simplified to illustrate only the main flocks.
Often the colours and the layout align as in this map which clearly identifies a leave component and remain component.
These larger communities often break down into sub communities which can also be seen in a map.
However, representing this kind of information in a 2D picture means that in some cases a more complex set of communities is much harder to represent in a map.
In the map below shows one community coloured in pink to illustrate that the grouping does not always neatly align with the colouring.
The Centre Ground
Here is the conversation map for the week ending 3rd June as shown above but coloured to just indicate the main campaigns. Highlighted is an area on the map which consists of many smaller users on Twitter who are supporting the leave campaign. More on these later.
There is no equivalent large group of ‘popular support’ on the remain side. The position of the media within the map is exclusively outside this area which reflects a more central position and which the media would be expected to occupy.
If the centre of the map is shifted to exclude this group the more central position of the media is more clear.
Filtering the map to show only the large leave community in green and highlighting the same media reveals the fact that many of the media are positioned within the remain campaign community but coloured to indicated they are part of the leave community.
The two types of algorithmic community detection detect different types of communities.
As mentioned above, in a previous post I examined the role of the two campaigns media relations accounts. Examining how this effects the Guardian account helps explain what is happening.
The maps above show the ego networks for the Guardian and for the StrongerIn Press. An ego network is all the accounts directly connected to a central account.
As mentioned above, the Guardian is part of the larger leave community as indicated by the colour green. The Guardian ego network illustrates that the Guardian twitter account is involved in a lot of communication with members of the leave campaign and the remain campaign.
Here is an example of a Tweet that attracted communication with both sides.
On the right hand side is the ego network for the StrongerInPress account. As pointed out in the previous article this StrongerInPress PR account is very well connected on both sides, but more strongly on the remain side. and creates a community on the remain side. The Guardian also belongs to this community which is reflected in it’s position in the overall map. The Guardian therefore, shows up in flocks on both sides of the argument. Visually it is pulled towards the remain side by its presence in the StrongerInPress community.
That is not the case with another major media account, the Telegraph.
Here are the same kind of ego maps for the Telegraph and for the Vote Leave Media account.
It is immediately obvious that these ego maps contain a much less diverse group of users. The Telegraph ego network does not contain any major users from the remain side.
Here is an example tweet from the Telegraph. The replies to this tweet illustrate the strong connection with the leave campaign which are the connections in the green part of the ego network map above.
In the Vote Leave Media ego network on the right there is also a distinct lack of diversity. Only David Cameron (who occupies a unique position) is present from the remain side.
These four examples illustrate what appears to be a difference in approach to campaigning on Twitter by both sides. Returning to the overall map (below) the structure provides an illustration of this difference which could be broadly categorised as:
Leave — High volume of tweets and echoing tweets of others from within the leave community.
Remain - Lower volume of messages and engaging in dialogue with those with a wider range of views.
These descriptions are about the structure of the conversation on Twitter. This may provide some insight into the way the campaigns are run in the wider context as Twitter contains a lot of messages about topics in the wider world. Further analysis would be required to confirm how closely the conversation on Twitter mirrors the overall public conversation and sentiment.
Organised Structure?
I said above that I would return to the ‘popular support’ part of the map.
Here is another view of that part of the map.
The reason that I have put ‘popular support’ in inverted commas is that I suspect a high degree of organisation of these accounts.
The visualisation reveals a structure which looks unlikely to be the result of a natural process.
I have posted in the past about detecting structured and organised tweet campaigns.
Looking at these accounts which are prevalent in this part of the map there appears to be a lot of very similar types of accounts. Accounts with a relatively low number of followers who are tweeting and retweeting on the single issue of the Brexit campaign. I have highlighted four I picked at random.
If you look a these four accounts you will see that they are similar in nature.
There are some similar type accounts on the remain side which I have highlighted below. However, they are much less prevalent on the remain side.
To put this in context I should point out that all the maps used above are filtered to remove lots of ‘noise’ to make them readable and meaningful. If more of the noise is left in they look more like this which illustrates the shear volume of messages coming from the leave side.
The ebola and data science examples above illustrate malevolent uses of this kind of organised Twitter activity. I should point out that so far I have not detected any evidence of that kind of activity and there are several plausible and reasonable explanations of what is happening.
This could be a genuine grass roots movement that is well coordinated and organised to use twitter to promote the leave cause.
Whatever the reason behind this phenomenon, it has two significant potential effects.
One, creating an echo chamber which, whilst it appears successful at face value, is not as effective at getting the message out to the wider population.
Two, insulating the important influencers in the campaign from some of the more extreme views which are expressed on Twitter.
Extreme Views
Here is an example of a Tweet tying the Brexit leave campaign to extreme views about islam.
The account from which this was posted is quite prominent in the network.
In the table of top influencers overall, this account shows up as being highly influential.
These kind of extreme views could do a lot of damage to the leave campaign which has been accused of being racist. Twitter is particularly susceptible to the spreading of extreme views due to the fact that accounts can be anonymous and that tweets are public. The large volume of activity on the leave side allows the main protagonists to maintain a very focused campaign on Twitter without being closely associated with extreme views.
Conclusion
Today has brought a set of opinion polls which indicate a gain in support for the leave campaign. First indications from recent data suggests an upswing in tweets supporting the belief that the leave campaign is performing well.
Echo chambers have an unfortunate effect of convincing those within them who strongly believe in their cause that they are doing better than they are in reality.
Given the analysis of the Twitter conversation there are clear warning signs of this for the leave campaign. The remain campaign looks more robust in terms of a wider range of views and connections with others which are more likely to provide negative feedback which is useful in assessing the real nature of public sentiment.
Example of Echo Chamber
Above is a tweet which was posted using one of my images to support the fact that Vote Leave were winning the Twitter battle. This was retweeted immediately by quite a high number of people using this to support the leave campaign.
The tweet had no link to my article and no explanation of what this kind of conversation map means. Therefore, those users who retweeted it were clearly guilty of confirmation bias which is the cornerstone of an echo chamber.
This may be a small example but does demonstrate neatly how echo chambers work.
Tables of Twitter Performance
Overall
Leave
Remain
