Post Mortem of @realDonaldTrump: An Introduction
Our research team at the University of Colorado Boulder is launching a series of posts that reveal how influential the @realDonaldTrump account was leading up to and during the 45th US Presidency. Like no other political person or even celebrity online, this account had tremendous impact on not just its 88M followers but also on the online information environment in the large.
This post is the first in a series we are running in winter/spring 2021.
It was clear in April 2020 that online disinformation was a significant problem, but it still caught us by surprise when we saw the evidence in an utterly innocuous place. As long-time crisis informatics researchers, we are accustomed to seeing all kinds of behavior online, some supporting safety and some not—but this digital behavior suggested something different was afoot.
We tuned into Twitter early in the coronavirus pandemic as we do in all the natural hazards events we investigate, with an initial goal of studying risk communication between officials and the public. We started by collecting everything the Center for Disease Control (CDC) and the World Health Organization (WHO) posted — as well as who was sharing these posts.
Accidental Disinformation Researchers
Then it happened: @realDonaldTrump broke our analytical workflow. On the morning of April 25, @realDonaldTrump retweeted 8 tweets from the CDC within 6 minutes. Our data visualizations of who was retweeting the CDC spiked in unexpected ways, and was a suspicious deviation from forms of large-scale “collective behavior” that we would expect to see under such circumstances. We eventually realized that we were witnessing an otherwise hard-to-detect amplification machine springing into action and retweeting @realDonaldTrump in lockstep (and is a mini-study this blog will present in future posts).
This was the moment we became accidental disinformation researchers. We had stumbled upon a signal in the noise of the Twittersphere that demonstrated just how influential the @realDonaldTrump Twitter account really was beyond its immediately observable scope.
What followed were months of data collection and analysis in an attempt to appreciate the magnitude and mechanisms of its range. This made it hard to be surprised on January 6, 2021 when rioters stormed the US Capitol in a violent, deadly attack. After weeks of flagging @realDonaldTrump re/tweets for distributing disinformation, Twitter indefinitely suspended the account on January 8.
In light of these and other remarkable events of 2020 and 2021, we are performing an analytical post mortem inspection of the behaviors surrounding the @realDonaldTrump Twitter account. So significant was the @realDonaldTrump account that we are reviewing the highlights so that others can come to appreciate its influence as well.
Macro View of Re/Tweeting Patterns
We begin this series with consideration of just the volume and frequency of tweets. In this first data visualization (using data from the TheTrumpArchive.com), we display @realDonaldTrump’s activity from January 1, 2015, the year Trump initially declared his candidacy for President. All times are displayed in Eastern Time. Each tweet @realDonaldTrump posted is displayed as a blue dot, with retweets as triangles. If @realDonaldTrump deleted a re/tweet, that is displayed as a red dot. It was not until late 2020 that Twitter began to flag @realDonaldTrump tweets as potentially containing disinformation: those are depicted as orange dots.
Each row of dots represents a single day, with the horizontal position corresponding to the time of day the tweet was posted. A count of all re/tweets posted by @realDonaldTrump each year is displayed between the year labels on the left.
In the weeks to come, we will highlight what was happening at the time these re/tweets were posted. For now, consideration of the whole dataset is important to understanding the overall behavior of the @realDonaldTrump Twitter account.
Increasing Volume over the Years
The first point to note is the high (and ever-increasing) volume of @realDonaldTrump’s re/tweets. From January 1, 2015 until Twitter suspended the account on January 8, 2021, @realDonaldTrump posted 28,262 tweets and 9,852 retweets (for a total of 38,114; all of which are represented here). Of these, 653 tweets and 439 retweets were deleted after posting (the red dots). We will return to a closer analysis of deletion behavior below.
Additionally, though there are multiple authoring platforms used for @realDonaldTrump tweets, we found that in 2018, 2019, and 2020, nearly 99% of @realDonaldTrump tweets were posted from an iPhone. Others have speculated on ways to determine who is actually authoring these tweets, but the consistency of this platform use is a feature of note.
Daily Patterns of Re/Tweeting
The second behavioral feature to note is the high frequency of tweeting across the day and into the night. In 2015, a campaign year, we see even more tweeting during a typical nighttime sleep cycle on the US east coast. This activity slows down as 2016 approaches.
Tweeting in the morning increases after the 2016 election and becomes more notably dense in mid-2019 through 2020, as does the overall volume of tweeting. Also in 2020, tweeting in the earliest hour of the morning (12 midnight to 1am) increases, but tweeting from 2am-5am nearly ceases in comparison to prior behavior.
When considering this macro view, other than the dense tweeting of the morning, tweeting happens throughout the workday and into the night. The bar chart in Figure 2 collapses all tweets since January 1, 2015 until the date of account suspension, and depicts the distribution across the 24-hour day, confirming the morning peak as well as the remarkable consistency throughout the day, rising again after 9pm.
Patterns Of Deleted Re/Tweets
An account holder can delete tweets or re/tweets after they are posted. For @realDonaldTrump, we note that tweets in the middle of the night have a high frequency of being self-deleted (the red dots in Figure 1). If we again aggregate all deleted tweets into one day, we see that about 1 in 10 tweets posted between 2 and 5am were later deleted, as Figure 3 illustrates:
Considering now deletion over the course of years (Figure 4), with just a few exceptions, all the President’s deleted re/tweets occurred before about 9am and then after about 9pm eastern time until 2020. (Deletions on March 17, 2019 were retweets of Jack Posobiec). However, the 45th President begins a marked new habit in early 2020 of regularly deleting his re/tweets during the workday—and continues this practice throughout the year.
In our next post, we will examine specific historical moments to consider @realdonaldtrump tweet volume and content. Across this series, we will continue to offer quantitative assessment, while also boring in on select tweet content.
Overview of Gaining Follower Counts
Finally, we return to the full display of re/tweeting behavior but in this visualization, we depict @realDonaldTrump’s follower count (Figure 5). Upon the mid-year approach of the 2016 election, the account broke 10M followers. Within the short span to the 2017 inauguration, the account gained another ~10M followers. The account gained another ~25M followers in the first year of Trump’s presidency, ending 2017 with roughly 45M followers. Going into 2020, the account had about 70M followers and, at the point of suspension on January 8 2021, had reached 88M. As an immediate comparison, @barackobama tweeted 3,150 times over the same period, with 128M followers today. But more comparative analysis is to come: In future posts, we will analyze the difference between current political leaders and celebrity accounts in comparison to @realDonaldTrump, including the behavior of followers.
In our next post, we will draw out the historical highlights that appear in this social media record. In later posts, we will turn to a closer accounting of the events of 2020 as reflected in this account, show the effects of the “amplification machine” in digital trace data of its followers, compare this account to other influential accounts, and more.