If you are covering “Twitter” as it if were a campaign rally, you are doing a disservice to democracy.

Bernie Bros, Obama Boys, TrumperTantrums, confirmations from a Tweet — before we ground our politics on arguments we need to be clear about what Twitter (and other social media) are and are not.

As a part of my research I have been collecting election discourse on Twitter, lots of it. Millions and millions of units of it. What is the executive summary? Twitter is awful. I have found in almost every dataset a propensity for the stream to be dominated by trolls and robots. I went out of my way in that last sentence to NOT use the word conversation. Why? Because Twitter is not a conversation. Other researchers have documented that during media events Twitter tended to become a transmit modeled medium — these dynamics play out in my samples as well. Retweets dominate, the only agents in my social networks with any meaningful centrality are celebrities (and some robots even). Here is a graphic of the interaction during the DemDebate mapped as a social network:

Image for post
Image for post
That mass of gray: tens of thousands of users with ZERO impact on anything.

Those pink core nodes are conservative celebrities. Tied in tweeters are green, still mostly celebrities, the vast mass of the rest of the network is a gray Borgish cube. Let’s be even clearer by using a cascade of eigenvector centralities from the DemDebate:

Image for post
Image for post
Dots mean ZERO impact.

Circles represent some control over flow in the network. Dots mean no impact on flow. Only a handful of users have any impact on the development of the network, these tend to include a short list of people:

  • Journalists
  • Celebrities
  • Various other alternative robots/celebrities

One might object to this way of modeling Twitter talk, perhaps you prefer Betweenness or Closeness or Borgatti DF for evaluating centrality. You would be right that Eigenvectors might not be the best math here. You would be wrong to believe that you would find any math that will say Twitter is not a disconnected mess of a network, that would have been your objection to my approach in the first place.

Using Twitter to find evidence of a BernieBro or a TrumperTantrum from a single data point is the same as going to a dive bar in near a college and waiting for townie to complain about the university messing up parking. The quote you get from that bar shouldn’t appear in an article in the New Republic, it doesn’t prove anything. Even Twitter agrees, this is why they are moving toward new timelines in the coming weeks: because the flow of garbage through the temporal network is just that repulsive.

If you go diving in the sewer, don’t be surprised if you end up in a pile of crap.

Here are some rules that might help people cover social media, based on my observations of my data:

  • If the user has no meaningful centrality number, their quote isn’t a story. Just like quotes from celebrities Tina Fey > Scott Adsit > some guy at an open mic night. If you cover the open mic, you are Homer Simpson:
  • Even if the user has a high centrality, be aware that you are basically covering celebrity gossip as news.
  • If you are observing a rapid retweet cascade, you are covering the work of a robot or troll, likely not individual users. Don’t fall for it.
  • Use data. Do not rely on individual Tweets or subjective impressions of your personal Twitter stream or hashtags you clicked on. Call a professor or a data scientist or someone to help you get something together aside from your personal Twitter view.

Many trolls on the internet are sadists. They work alone to get their kicks from hurting people. If you cover sadists and their violence as if it is joined to the identity of a political movement, you make deliberation more difficult. My datasets are full of vile garbage directed at basically every candidate. If we covered every candidate based on what trolls said about their adversaries, there would not be any candidate left.

Horse-race polling stories used to be the foundation of election coverage. Nate Silver ruined that for everyone in the business. Young people not watching This Week with George S. has ruined punditry as a story type too. Moving on from bad poll based articles and opinion-journalism to even worse readings of social network activity is not progress. This is not to say that there isn’t bad stuff out there or that people shouldn’t write cultural critiques. On the contrary, they should. What they should not be doing is drawing broad conclusions about social formations without data. The haphazard manufacture of social facts from columnist’s observations of their Twitter feeds contributes to trutherism (read: conspiracy nonsense) as it confirms that all facts are just vague impressions.

Reducing the scope of political journalism to what was on Twitter today weakens democracy. Social media is countable. The data is better quality than polling data, thus it actually reflects the awful mess of a public at scale. Can I confirm for you that Bernie Bros do or do not exist? No. Nor can you prove that does in anything except a very cursory way. The overwhelming story told by the data is of a dying transmit centric platform riddled with hateful trolls and celebrities selling references or giving out direct tweets as rewards to fans.

Take your time, write a better article.

Associate Professor of Social Media. Oregon State University. Read my book: Selling Social Media (Bloomsbury Academic), 2018.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store