Last updated: Mar 1, 2016

Trump's Primaries

This plot shows dots representing 27.1 million tweets about Donald Trump during the month of February while Trump battled rival Republicans in the primaries in Iowa, New Hampshire, South Carolina and Nevada.

Click on the links in the text below to follow along or explore the data yourself as you would a web-map: double-click, mouse-wheel, pinch or use the +/- buttons top left to zoom in/out; click and drag to pan around.

Climbing up the Mountain Plot

What is a mountain plot?

Every tweet is a dot. As each tweet comes in, it will be plotted based on the time of the tweet and the author id.

Time. The horizontal axis represents time. The far left side is Feb 1, the far right is Feb 29.

Author. The vertical axis represents unique authors. The first author that we process starts in the very first row at the bottom. The next tweet processed from a never-before-seen user starts another row. Rows keep adding for each user - in this dataset we have 3.7 million unique authors.

Slope. Because each tweet is added as it comes in, the leading edge of the mountain grows more quickly when more new authors start tweeting about Trump. At the beginning of February it grows very quickly, because there are lots of people tweeting about Trump all the time. About Feb 9, there's a steep bump that corresponds to the New Hampshire primaries and lots of new people talking about Trump.

Dots, Stripes and Zoom

Bright dots indicate more tweets.

Bright vertical bands occur at times when most of the people are talking about Trump. These correspond to events, such as the GOP debate on Feb. 6 and Feb. 13. Hit the home button (top left corner) to zoom out again.

Dark vertical bands correspond to times when very few people are talking about Trump. The darkest vertical bands are night-time in the US when most people are asleep.

Dark horizontal bands correspond to a group of people who all joined the conversation at the same time, but really don't talk much about Trump again beyond their original tweet. Here are 12,000 users who pile in on Feb 17 to retweet this tweet from Trump himself:

.@FoxNews is so biased it is disgusting. They do not want Trump to win. All negative!

— Donald J. Trump (@realDonaldTrump) February 17, 2016

This cohort isn't generally engaged in the discourse on Twitter (hence the dark band) but interestingly they jump in again en masse. This time it's all within a span of just a few minutes to retweet the following tweet, shown as a strong vertical stripe:

Who agrees that Trump should step down from his campaign for president? #NeverTrump

— optrump2016 (@optrump2016) February 27, 2016

Anomalous patterns stand out because they don't follow the patterns around them. These can be interesting to zoom in and investigate. For example, bots can show up as bright smudges or sequences of bright dots that don't match the periodicity of their neighbors. Back in early February there's some bright coordinated activity among many new posts which we identified as suspect Twitter accounts in our previous post ( Two Days in Iowa).

Top 10 Topics via Hashtags

Over top each square region are the top 10 hashtags (excluding #trump, #donaldtrump, and #trump2016). We're collecting tweets (using the Twitter public streaming API) that mention keywords Trump (e.g. DonaldTrump, Trump4President), WakeUpAmerica, and MakeAmericaGreatAgain. Hashtags are useful to see what topic is being discussed.

Pope Francis became a popular topic starting about Feb 18th. This occurs after the pope makes comments regarding Trump's call for deportation of undocumented immigrants.

Ben Carson is a popular topic via the hashtag #TheOnlyOutsider. Zooming in for closer inspection shows that there isn't really broad support for this hashtag - it never spreads out vertically. This may indicate that very few accounts are using the hashtag. A quick check in the data shows only a single twitter account - @wethepeople2017 - using this hashtag, but generating tweets every few minutes:

Tweets by @wethepeople2017

Opportunists can be similarly isolated. Both #nazi and #ebay seem to occur together at various places when zoomed in on the plot. Checking twitter seems to show many messages of people selling war memorabilia presumably using popular hashtags to simply increase visibility of their wares. Similarly, the tag #fitn (aka fitness) is rampant throughout the data in early February: likely lots of robots at work but we haven't investigated that subset in any detail.

Competition can be identified. For example, which other politicians are co-mentioned with Trump? Presumably those mentioned most frequently, with the broadest support across all the tweets, are the biggest competitors. We've made a few shortcuts so you can take a quick look at the macro-level: #Cruz, #Bush, #Hillary, and #Bernie. Didn't see Jeb anywhere when you clicked on Bush? Unfortunately, Jeb Bush didn't bubble up to the top anywhere in Trump conversations and he withdrew shortly before the Nevada vote.

Issues can also be seen within cohorts. For example, the topic #woman occurs repeatedly across a row. Twitter search shows many interrelated topics about Trump's tumultuous relationship with women ranging back to his fiery exchanges with Megyn Kelly, other accusations, references to Hillary, and so on.

Comedy of course, can be uncovered - for example, #greetthealiensin5words during the New Hampshire primaries. The hashtag #fodtrumpmovie comes up in a few places and is a link to Johnny Depp playing Donald Trump in a mockumentary.

So what?

The Twitter conversation is filled with great patterns representing real topics, events, debates; but also robots and interlopers. Regardless of the debates, there's more discussion around Trump and democrats than there is about Trump and the other Republican candidates: Cruz is the only candidate making a presence in the Trump conversation, other than a single Carson supporter.

How'd we do this?

The Trump Mountain Plot was built using the open source software Uncharted Salt. Salt is a free software library for developers to assemble massive amounts of data, organize it, mine it, and create interactive displays. We use the library, for example, to analyze GPS data, massive networks, bitcoin, stock markets, product sales, and other big data. Here's a couple other examples:

Check out our blog or contact Uncharted Software if you have any questions about Salt.