• Sean Farrell

Twitter Votes: How Social Media Scored the First Presidential Debate

‘Who won?’

This is the question journalist Graeme Dobell asked us when writing his recent article 'Reporting on a President who shouts and shifts'. Probably the best thing we can say about the first Presidential Debate of the 2020 US election season is “it wasn’t boring”. Infuriating perhaps, but definitely not dull. We saw lots of emotional responses on Twitter to the debate (particularly state by state), but can we declare a victor? Does ‘winner’ vs ‘loser’ even make sense in this scenario? How did people respond to each subject, or react to particular statements made? For example, how did people react to Biden’s “will you shut up man?” comment or Trump’s pronouncement to “stand back and stand by” when asked to denounce a particular far-right extremist organisation?

To dig deeper into this question, we grouped the emotional responses on Twitter from tweets aimed at Trump or Biden and posted throughout the debate (a much bigger sample this time, allowing us to slice the data down to 1 minute time scales) from Twitter users based in the US. We then compared the 21 emotional measures we extracted using Receptiviti’s proprietary models (stay tuned for the public release) with the timing of each discussion subject and a number of arbitrarily selected contentious statements made by each candidate and threw tweet volume into the mix for fun and games.

Some really interesting patterns scream through the data. First of all, in our sample (possibly biased, we admit) there are far more tweets aimed at Biden throughout the debate than at Trump (Fig 1).

Figure 1 – Volume of tweets targeting Trump (red) vs Biden (blue) during the first debate, which started around 9pm ET on the 29th of September 2020.

Secondly, the volume of tweets aimed at Trump are incredibly strongly correlated with our emotionality metrics, particularly ‘good feel’ and ‘bad feel’ (in equal measures no less). In contrast, the tweets aimed at Biden showed almost no correlation at all between volume and emotionality, though the sharp spike in tweet volume does correlate with a peak in emotionality and bad feelings (see Fig 2). Strangely, after the first hour of the debate, the emotionality and volume of tweets aimed at Biden both drop significantly. Could this be evidence of debate fatigue? Our Chief Scientist Prof. Jamie Pennebaker believes it’s a strong indicator that the people tweeting at Biden in the first hour of the debate are a different population than the people tweeting at him after the first hour. Very interesting signal regardless.

Figure 2 – Emotionality vs time from tweets aimed at Trump (left) and Biden (right). The grey bars indicate tweet volume targeted at each candidate, and the coloured lines indicate emotional intensity. Note: the tweet volume is scaled for visualisation clarity, and Biden’s tweet volume is significantly higher than Trump’s in our sample.

When tweet volume peaked for Biden, we saw a huge jump in anger (Fig 3), along with a boost in bad feeling and emotionality but no subsequent jump in good feeling. This corresponded with the period where Biden was being pushed on whether he would pack the Supreme Court, and also when he lost his cool and exclaimed “will you shut up man?”. It seems pretty clear that Twitter users weren’t impressed either with his evasion around the Supreme Court packing question or finally cracking over Trump’s interruptions.

Figure 3 – Anger aimed at Trump (left) and Biden (right). Key statements made by each candidate are indicated by the dashed lines. Tweet volume is indicated by the grey bars.

Another really interesting emotion that jumped out was disgust. Overall, far more disgust was aimed at Trump than at Biden, and it trended steadily upwards throughout the debate, spiking often when he was attacking his opponent (and particularly following his “stand back and stand by” statement when asked to disavow white supremacist organisations). In contrast, disgust aimed at Biden initially trended slightly downwards at the start of the debate before reversing direction after – you guessed it – he lost his cool and told Trump to shut up. There aren’t many obvious correlations between statements and disgust spikes for Biden, other than when Biden accused Trump of being a racist. It seems clear though that the Twitter public weren’t impressed with either candidates’ snarky attacks on each other.

Figure 4 – Disgust aimed at Trump (left) and Biden (right). Key statements made by each candidate are indicated by the dashed lines. Tweet volume is indicated by the grey bars.

Figure 5 – Good feeling aimed at Trump (left) and Biden (right). Key statements made by each candidate are indicated by the dashed lines. Tweet volume is indicated by the grey bars.


Figure 6 – Bad feeling aimed at Trump (left) and Biden (right). Key statements made by each candidate are indicated by the dashed lines. Tweet volume is indicated by the grey bars.

But does this tell us who ‘won’ the debate? Not really. We can, however, look at the relative balance of good feeling vs bad feeling for each candidate from before and after the debate and use that as a proxy for a score card. Across the entire US, pre-debate tweets aimed at Biden had a good-versus-bad metric of 42.6%, while after the debate that rose to 50.0%. As for Trump, his good-versus-bad pre-debate metric was 52.8%, while afterwards it was 51.1%. So, from this I guess you could say that good feeling for Trump went down slightly by 1.7%, while good feeling for Biden went up by 7.4%. A decisive win? Not exactly, but in alignment with at least some of the post-debate commentary.

As we well know, the US has a rather unique voting system known as the Electoral College which means that winning the popular vote (if Twitter qualifies as a proxy… no comment from us) is insufficient to win the Presidency. There are a number of ‘battleground states’ that are expected to be most competitive and deterministic over who wins the election. As we have limited our sample to tweets that have been geo-tagged as being posted by people based in the US (sorry, should probably have mentioned that earlier) we can generate a state-by-state summary of emotion directed towards each candidate from a period prior to the debate (taken from tweets posted 30 mins or earlier before it kicked off) and those posted after it finished (30 mins or more after the climax). So, what does that tell us?

For starters, Biden looked pretty solid before the debate kicked off, with around 80% of the states in the Twitterverse having a higher positive feeling for him than Trump (lets ignore bias and sample sizes for this exercise). After the debate sentiment shifted, with only around 60% having a better opinion of Biden than Trump after the fact. But this is still meaningless to a large extent, as it’s more important which states he wins in than overall numbers.

The New York Times has a really cool interactive tool to let you mix-and-match which battleground states each candidate might win, to see how the outcome will affect the overall election (assuming the non-battleground states will vote as expected). Using our good-to-bad metric as a proxy for how each state will vote here’s what it predicts: Biden will win Texas, New Hampshire, Ohio, Michigan, Maine, Nevada, Arizona and Wisconsin, while Trump will win Minnesota, Florida, Georgia, Iowa, North Carolina, Nebraska and Pennsylvania. If Biden wins all the battleground states that still had more positivity towards him than towards Trump, he'd win 316 electoral college votes to Trump's 222, giving him a decisive win according to the New York Times. Of course, this is just emotions of people on Twitter who say they are based in the US so highly uncertain plus immensely biased, so take this with a huge grain of salt.

Figure 7 – Continental US map of good-vs-bad feeling for Trump vs Biden prior to the first debate. Blue = more favourable towards Biden, red = more favourable towards Trump.

Figure 8 – Continental US map of good-vs-bad feeling for Trump vs Biden after the first debate. Blue = more favourable towards Biden, red = more favourable towards Trump.

So, what can we conclude from this? There’s still another debate later this week (at least at the time of writing, and if we’ve learned anything from 2020 it’s that the future is uncertain) and none of this is solid, but it’s fun to speculate. All we can say is this… it’s pretty clear that people in the Twittersphere were not impressed with the bickering and insults. It will be interesting to see whether either candidate has learned from this experience.

  • LinkedIn
  • Facebook
  • Twitter

© RECEPTIVITI INC. 2020