I audited Kai Ryssdal and Molly Wood’s Twitter Accounts
By Jon Keegan
A few weeks ago I got an email from the folks at the “Make Me Smart” podcast with an intriguing challenge — “Could you audit the Twitter accounts of our hosts?” Specifically I was asked to look for any telltale signs of an “echo chamber.” Did hosts Kai Ryssdal or Molly Wood’s Twitter experience feature more voices from one side of the political spectrum? Or more from one group of citizens? In my work as a visual journalist I have examined what polarization looks like on Facebook, and studied how Twitter was used by Hillary Clinton and Donald Trump in the 2016 presidential campaign. This “Make Me Smart” exercise sounded interesting, so I accepted.
You can listen to the episode where I talk to Kai and Molly here:
91: How do you get out of your echo chamber?
It's tough to admit you have a filter bubble, and even harder to break out of it. But we all live in one: Social…
So to figure this all out, Kai and Molly (good sports as they are) agreed to provide me with first-class data. Through their Twitter account dashboard, they requested a download of all of the data from their accounts (You can find instructions on how to get your own data from this page.) I only needed two things: the full archive of their tweets, and the list of accounts that they are following. This was an important part of the process, as Twitter’s Developer API (the interface for interacting with their platform via code) only lets users download the 3,200 most recent tweets from a public account. Since Kai and Molly requested their own data, I got everything they have ever tweeted.
Data in hand, I needed to do a few things before I could interpret it. If you open the text file named “tweet.js” that contains your tweets, you may be overwhelmed at first. On their public-facing side, Tweets seem really small, just up to 240 characters and a link or picture, maybe the occasional hashtag. But Twitter is actually a pretty complicated platform, and each tweet contains a huge amount of metadata — or data about the tweet that you may never see, but stays attached to the Tweet forever. Take, for example, this tweet from POTUS:
When you look at the raw data behind that tweet, there are 5,198 characters and 162 lines of data under the hood!
We didn’t need all of this to do our examination, so I converted these verbose JSON files to the more compact CSV (comma separated values) format. I then loaded this into a database server (I use MySQL). Once the data was in a database, I could start to answer some questions.
So, How Are Kai and Molly Using Twitter?
There are so many different ways to use Twitter. Before speaking with our hosts, I wanted to see what I could tell about their use of the platform.
So here are some top-line stats (as of Nov. 7, 2018) that I pulled together:
OK, so now we’re getting a 10,000-foot view of the basic stats. Twitter first launched on March 21, 2006 (with this very first tweet) so Molly and Kai are long time users of the platform, having joined in 2007 and 2009 respectively.
And they tweet a lot; Kai produces an average of 12.8 tweets per day, which is double Molly’s number of 5.7.
Kai does not ❤️ many tweets. While Molly has hit that “like” heart 7,825 times Kai has only done so 124 times.
Another big difference that jumps out at me looking at both of their stats is this: Kai’s “Following” number — the number of accounts that he is following — sits at 373. That’s far below the 1,188 that Molly follows. Molly follows about three times as many accounts as Kai does.
Patterns of usage: When
I always find it interesting to see what people’s patterns are on Twitter. Looking at the timestamp on all of the tweets for Molly and Kai, we can group them by hour (assumes Pacific Time) and get a sense for their most active times on the platform. In general it seems their schedules are offset, with Kai up and tweeting earlier, and Molly starting later, and staying up longer.
We can do the same for day of the week. Molly’s schedule looks fairly steady during weekdays, while Kai builds up a lot of momentum to peak on Fridays.
It is always interesting to see which days they tweeted the most in the history of their time on the platform. The top three days for Kai were October 29, September 17 and August 7, 2015. All were days when he was live tweeting the GOP Presidential Debates.
Molly’s top days were covering the Google IO event on May 28, 2015, covering the Apple reveal event on September 9, 2015 for the iPhone 6S, and April 11, 2018 when Mark Zuckerberg testified before Congress.
Patterns of usage: How
For every tweet, one piece of metadata that Twitter serves up is which app sent the tweet. You can see Kai’s Twitter client of choice is TweetDeck, which is a desktop-friendly but information-dense multi-column view of Twitter favored by social media managers and power users (like Kai). Molly appears to prefer using the web client for Twitter, which is a good way to consume all that information, without the addictive push notifications of a native mobile app.
Patterns of usage: What
Finally we should look at what our hosts are tweeting about. Here’s Molly’s most popular tweet:
And here’s Kai’s most popular tweet:
Another thing that Twitter breaks out in the data for each tweet is what hashtags are used. Molly has used hashtags to associate her tweets with events, such as Apple’s World-wide Developer Conference (#WWDC) or Austin’s South by Southwest Festival (#SXSW). Kai uses them in a similar way, but also to promote #RedPantsFriday.
Patterns of usage: Who (they follow)
To answer the question about echo chambers, the best data to look at are who Kai and Molly follow. This is one the more difficult tasks, as we need to be able to categorize the accounts, and there is no data field that says, “This account is a musician, or journalist or president of a large country.” As Kai was following a relatively small number of accounts, I manually classified as many accounts as I could, which ended up being around 66%. Did the same for Molly, and was able to get a larger chunk classified with around 83%. Machine learning could be used for this task if we had a much larger set of accounts to classify, but this is tricky, as a lot of accounts straddle different worlds. I encountered a lot of bio texts that featured lines like “writer/podcaster/founder.” Hard for a human to classify these, much harder for an algorithm. So not a perfect measure, but definitely enough to give us a sense of the makeup of these voices which appear in their feeds.
As you can see for both Kai and Molly’s accounts, the largest category of account by far was “journalist” followed by “news outlet.”
News outlets are definitely one of the places where polarization can manifest itself most clearly, so it was worth drilling down into these news-related categories to see if we could find out more about the news outlets represented.
Drilling down into journalists / news outlets…
To understand what the makeup of news voices followed by Molly and Kai’s accounts, I did use a form of machine learning to help with this tricky task. One of the most powerful things machine learning tools can do is extract “entities” from a body of text. Basically the program reads in the full text you want to analyze, then it returns a list of different known entities, along with a category. For example it might find “NBC” and classify it as “ORGANIZATION,” or “Baghdad, Iraq” as “LOCATION.” So first I took a list of all the accounts that I categorized as “News outlet ” or “journalist”, then took those accounts’ “bio” texts, and ran them through Amazon’s Comprehend (a cloud based machine learning service) and grabbed all of the entities that were organizations. Sorting this list gives us a snapshot of the news outlets reflected in their feeds.
So looking through the makeup of these news sources, we mainly see large, established mainstream news outlets – a good foundation for any healthy news diet. But there was a notable lack of conservative voices, so that is one area where both Kai and Molly could benefit to help round out their feeds.
The other suggestion I would make would be to follow more of the official accounts from elected officials. More and more politicians and government agencies are embracing President Trump’s platform of choice to get their message out.
Yes, it can be a lot of work to go find these accounts, figure out which is the actual official account (plenty of fakes out there), and then you may just not be happy with the mix you have after changing the makeup of your feed. You may see a lot of stuff that annoys you, or doesn’t feel relevant to you, or it just overtaking your feed to an extent you don’t like. Thankfully, Twitter has a great tool to help solve this problem and combat filter bubbles at the same time, which is Twitter Lists.
In one click you can subscribe to a curated collection of accounts, which lets you browse a custom timeline with voices from a community or field of interest (Update: see my correction below). If you haven’t used Lists, you should definitely give them a try. For example, C-SPAN maintains a great collection of lists related to politics and government news. list for all of the members of Congress, as well as the official cabinet level agencies for the U.S. Federal Government.
I’ve put together a short “Red feed” list that includes some of the bigger voices that were favored by conservative Facebook users, according to a large 2015 study. Slate has also produced a solid lists of established conservative and liberal voices.
Twitter can be a powerful tool for keeping up with breaking news or tapping into lively communities online. That has definitely been the case for me with the journalists I follow on Twitter. But unless you take advantage of all that the platform has to offer, you will only get so much out of it. Try an experiment and add some more voices to your Twitter experience by browsing some of these some of these Twitter lists for a week. Perhaps it will help get you outside of an echo chamber you aren’t even aware you’re in.
Nov. 26, 2018 — I incorrectly described the functionality of Twitter lists in an earlier version of this story. I wrote “In one click you can follow a curated collection of accounts, which can instantly fill your feed with voices from a community or field of interest”. That’s not exactly true it turns out (much to my chagrin). You can “subscribe’” to a list in one click, but those accounts tweets will not show up in your main timeline. You need to view that list’s timeline. So if you want to see what the voices of the members of Congress are tweeting about, you need to view their timeline here: https://twitter.com/cspan/lists/members-of-congress?lang=en
Apparently Tweetbot (iOS) offers easy access to switch between list timelines, which seems like a good way to use these lists. I apologize for the error, and I’ve updated the text above to reflect a more accurate description of the feature.
Suggested Twitter lists:
“Right leaning tweets” by Slate: https://twitter.com/Slate/lists/right-leaning-tweets/
“Left leaning tweets” by Slate: https://twitter.com/Slate/lists/left-leaning-tweets/members
All US senators: https://twitter.com/cspan/lists/senators/
The Cabinet: https://twitter.com/cspan/lists/the-cabinet
Foreign Leaders: https://twitter.com/cspan/lists/foreign-leaders
Read more about polarization and echo chambers on social media:
Hidden Tribes of America
The Hidden Tribes of America is a year-long project launched by More in Common to better understand the forces that…
Is the Internet Causing Political Polarization? Evidence from Demographics
We combine nine previously proposed measures to construct an index of political polarization among US adults. We find…
How do Americans use Twitter for news?
A look at how researchers analyzed news habits on Twitter using a small but representative sample of users drawn from a…
Replication Data for: Exposure to Ideologically Diverse News and Opinion on Facebook - Facebook…