Do Followers Count in Debate Tweets?


Q: Why was there such a large increase in the number of Hillary and Trump followers at 10:21 PM EDT during the second Presidential Debate?

A: I’m not sure, but maybe it has something to do with a woman’s right to choose and/or marriage equality?


During the last two presidential debates MITH has been collecting Twitter data using the words trump and hillary. One limitation of looking for patterns in collected Twitter data is that the presidential debates are high volume events. We know that the Twitter streaming API only gives us a portion of all the available tweets when there is a spike in traffic. For example our dataset for the first debate contains 1,303,084 tweets that mentioned hillary and the Twitter API let us know that at least 730,512 tweets were not delivered. Very little is known about what kind of sample Twitter provides, and without the full picture that the firehose provides it can be difficult to draw inferences from the numbers about Twitter users as a whole.

Followers Count

One interesting metric that is available is the number of followers a person has. This number can be particularly interesting to look at over time as it rises and falls. Every tweet that you get from the Twitter API includes a User Object which in turn contains the followers_count for that user at the time that the request was made of the Twitter API. So if you collect data in real time from the Twitter Streaming API you get a moving picture of a user’s followers. What’s also super handy is that in every retweet you get the user object for the retweeter, and also for the user who sent the original tweet.

During the debates you can expect lots of people to retweet the candidates. This means you can get smooth record of how many followers they have by the minute … or even second. The fact that you’re not getting all the tweets at a given point in time doesn’t really matter in this case.

New Followers

I thought it could be interesting to look at the number of new followers each candidate received during the first debate. In theory these would be “people” that had decided to follow Clinton or Trump on Twitter as they were watching the debate. I put people in scare quotes there because it is possible, and perhaps even likely, that there are bot armies creating these accounts. You’d have to track who these new accounts were and sample them to know for sure.

So with that caveat in mind here’s what the graph for the first debate looked like:

It was encouraging (for me) to see that Hillary was gaining more followers than Trump during the debate. There were a few interesting changes in the graph, but I was mostly distracted by the fact that Hillary was outpacing Trump so much in new followers.

We collected data for the second debate so I decided to look again with the same code.

I was pleased to see Clinton gained more supporters than Trump again…but this graph looks noticeably different. Do you see the bump in followers at 10:21 PM EDT (02:21 GMT). What happened there? After sharing this graph on Twitter I got this helpful tweet from Jane Wang:

Sure enough, when you look at the rate of change the bump is even more pronounced.


10:21 PM

So what happened at 10:21 PM EDT? Well, thanks to the Web you can see exactly what happened during the debate. It was the question from Beth Miller about the Supreme Court that begins at 10:19:54:

If you watch the time tick by you can see in the exact minute of 10:21 PM Hillary is saying this:

I want a Supreme Court that will stick with Roe v. Wade and a woman’s right to choose, and I want a Supreme Court that will stick with marriage equality. Now, Donald has put forth the names of some people that he would consider. And among the ones that he has suggested are people who would reverse Roe v. Wade and reverse marriage equality. I think that would be a terrible mistake and would take us backwards.

What is truly bizarre is that there is a bump at that point for both candidates. Maybe this points at a problem in my code, or my spreadsheet data (I’m happy to share the Tweet ID datasets if you would like to look for yourself). Or perhaps the bump is part of some kind of backlog in Twitter’s infrastructure? And there’s always the chance that there’s a bot army working for both candidates? It would be necessary to inspect the users to get a sense of that.

But maybe, just maybe, this data points at the fact that a woman’s right to choose and marriage equality are still polarizing and hot button issues, more than any other, for folks in this years election? Or at least for folks who use Twitter…which is definitely not all voters. That’s probably the biggest caveat there is.


I Don’t Know Putin

As a point of comparison here’s what the rate of change in new followers looked like during the third debate in Las Vegas.

9:26 PM EDT was the minute that Trump launched into his I don’t know Putin tirade, and Hillary crushes him with facts:

Note that the 300 new subscribers per minute at this moment is still dwarfed by the 1500/minute in the previous debate when Roe v Wade was mentioned.


Originally published at inkdroid.org on October 15, 2016.

A single golf clap? Or a long standing ovation?

By clapping more or less, you can signal to us which stories really stand out.