Debate No. 2 – the battle

Even if neither Trump nor Clinton would make a good president, they certainly make good, or at least captivating television. This post is a little bit late, and perhaps many of you don’t want to think about this anymore, but here I am again, rambling to you about the debate from a linguistic point of view. Whereas the first debate wasn’t exactly rational discussion, this second one probably can’t even be described as discussions. It was Battle Royal – they wanted two people in, one person out.

So here is my rambling. For the first debate, I analysed non-linguistic features of Trump and Clinton, and found that the two had very different disfluency patterns, indicating that Clinton prepared and Trump didn’t; Clinton made a big effort on clarity, while Trump just dumped his (perhaps scattered) thoughts in real time without planning. We also saw that Clinton smiled much more and her smiles appeared much friendlier, even though those smiles served the same function as Trump’s bitter smirks. That was some fun analysis for me. The second debate was so charged, I thought I can’t do the same analysis. Let’s look at how emotional and biased people were, the candidates, the hosts and the people watching.

For those of you who didn’t watch it, the second debate had a town hall format, where the audience proposed and voted for questions they wanted to ask in advance. The two hosts – Martha Raddatz from ABC and Anderson Cooper from CNN – presented the most popular and representative audience questions, but they also asked follow up questions. Overall, Trump and Clinton spent the same amount of time talking (39.95 minutes for Trump and 39.3 minutes for Clinton), though like last time, Trump said more words (7442 words versus Clinton’s 6366 words). The general impression was that instead of talking about their own policies, they spent most the time attacking each other. Is that true? Let’s look at their most frequent words. Here are two word clouds, the bigger the word, the more times it was said. First, Trump. Some of his most frequent words were “country”, “Hillary”, “Russia”, “disaster”, “ISIS”, “Obama”, and “Clinton”.


Second, Clinton. Some of her most frequent words were “Donald”, “country”, “president”, “Supreme court”, “important”, “insurance”, “America”, “children”, and “women”.


We can see that apart from saying “country”, the most frequent thing they said was calling each other’s names. So it is true, they spent a lot of time attacking each other. When they talk about their own plans and views, they frequently made comparisons and pointed out how bad the other is in comparison. So they were doing similar things in the debate, but with noteworthily different styles.  I analysed the sentiment using a list of positive and negative words in English (I’ll talk about some technical details in another post). Clinton scored a strong positive value of 128, and Trump a negative value of -12. This means that Clinton’s speech contained many words that were associated with positive emotions (e.g. “respect”, “growth”, “care”, “important”), while Trump’s speech contained more words that are associated with negative emotions (e.g. “disaster”, “failed”). This difference is not surprising and is not new. But remember, even though their words had different sentiment, they were using them to do the same thing – attack. When Trump attacks, it was brutal and there was no hiding. However, Clinton was able to camouflage attacking by using sweet words and being indirect, just like her smiles in the first debate.

So the candidates were not rationally discussing their plans as future presidents, but rather staged sword fights with words. But did they choose to do that themselves, or did the hosts and the audience encourage them? You probably think that both the audience questions and what the hosts asked were impartial and concerned the country and people’s lives, or at least that’s what I thought. But was it the case? Let’s look at what the main topics were. The very first topic was whether the candidates are modelling appropriate behaviour for the youth, given what they did in the first debate. Clinton immediately attacked Trump with this question, and Anderson Cooper followed up on the question by bringing up Trump’s locker room banter tape, which became the most discussed topic in the entire second debate. Eventually Trump was able to redirect the attention to Clinton’s email controversy, which the hosts allowed and followed up on. So the first half an hour was entirely personal attacks. After that, they discussed health care,  Muslims (mainly directed at Trump), whether it is ok to be two-faced (directed at Clinton), taxes (mostly directed at Trump especially by Anderson Cooper), Syria, whether they can be devoted to all people (that’s where Anderson Cooper brought up Clinton’s “deplorables”), whether they have self-discipline (directly at Trump by Cooper), Supreme Court Justice and energy policy. By the end, perhaps the hosts felt that the debate has been so hostile, they had to restore a bit of harmony, so the last question was if they could say one positive thing about one other, to which Clinton said she respected Donald’s children (so she can’t say ONE positive thing about Trump himself then:)  ) , and Trump said Clinton is a fighter, she never gives up. From these topics you can see that at least half of them were questioning personal traits of the candidates. Trump and Clinton wanted to fight, but the audiences and the hosts clearly wanted to watch them fight.

Not only were the topics encouraging personal attacks, the hosts were also adding aiding the storm by time control. The time spent on different topics varied massively, plotted in the following graph (minutes spent on each topic). Three out of the top four topics were all Trump oriented! From this I don’t think we can say that the hosts were impartial, but whether Trump earned this treatment is a judgement I’ll leave to you.

duration topics.png

We can also see biases from the hosts by looking at the forms of the questions they posed. Cooper and Raddatz directly spoke to Clinton 35 times, including asking 8 follow-up questions. In comparison, they spoke to Trump 57 times, including asking 17 follow-up questions (five were repeats). Were the questions biased? Take a look at some of them:

Questions directed at Clinton:

1a.   Your husband called Obamacare, quote, “the craziest thing in the world,” saying that small-business owners are getting killed as premiums double, coverage is cut in half. Was he mistaken or was the mistake simply telling the truth?

1b.   Is it OK for politicians to be two-faced? Is it acceptable for a politician to have a private stance on issues?

Questions directed at Trump:

2a.  In the days after the first debate, you sent out a series of tweets from 3 a.m. to 5 a.m., including one that told people to check out a sex tape. Is that the discipline of a good leader?

2b.  Can you say how many years you have avoided paying personal federal income taxes?

In all four examples, the question had a closed yes-no form (polar question) rather than an open form (wh-question like “what”, “how”, “why”). Yes-no questions are often laden with biases, and if the answerer says anything other than “yes” or “no”, it can seem that they are deflecting the question. In addition to using closed question forms, there are also a lot of what linguists call “presuppositions”: information that was taken for granted, assumed to be true. In 1a, using the phrase “was the mistake” presupposes that it was a mistake (therefore making the first half of the question redundant). 1b used “is it ok to be two-faced”, which presupposed that Clinton WAS two-faced. 2a first presented strong evidence of lack of self-discipline, and then asked, “is that the discipline of a good leader” instead of “do you have the discipline of a good leader?”. 2b asked “how many years you have avoided tax”, presupposing that Trump definitely avoided federal income taxes.

So it looks like both the audience and the hosts came into the room with some judgements and emotions. What about the people watching? What emotions did the American people show? Did people from different political stances reacted differently? I scraped public comments to Facebook posts regarding the second debate from the evening of Oct 9th to Oct 10th  (I’ll probably talk about the technical details in another post). I chose one left-wing media: MSNBC, and one right-wing media: Fox News. Now, instead of looking at what people said, I looked at what emojis people used (only about 10 percent of comments contained emojis, which is lower than news posts about entertainment). I was expecting many negative ones, but it wasn’t the case. There were much less comments to MSNBC posts than to Fox News, but on both, people talked about Trump a lot more than about Clinton (72% about Trump in MSNBC Facebook comments, and 80% about Trump in Fox News). But, most of the emojis were positive! The most frequent emoji was the laughing with tears face (except for Fox News Trump comments, where the most frequent was the thumbs up sign). Laughing with tears emoji doesn’t indicate that people had positive sentiment towards the candidate, but rather they were amused by them. There were also many “neutral” emojis, and the vast majority were the US flag.

Media Candidate positive emojis negative emojis neutral emojis Total
MSNBC Clinton 41 10 10 61(28%)
Trump 95 37 25 157 (72%)
Fox News Clinton 148 42 21 211 (20%)
Trump 489 60 291 840 (80%)

Table 1: number of positive, negative and neutral emojis in comments to MSNBC and Fox News Facebook posts.

Here are the composition of positive, negative and neutral emojis for each media and each candidate, plus the most frequent emojis used.

Comments to MSNBC posts used a relatively small range of emojis (One may think that in this context, the US flag may very well carry a positive sentiment):

In comparison, comments to Fox News, especially those that mentioning Trump, used a wide range of emojis especially in the positive category.  This is the only place where the most frequent positive emoji is not a laughing with tears sign, but a thumbs up sign.

Surprisingly, the audiences seems to have enjoyed the debate, at last to some extent. There was indeed some differences between emojis to comments of MSNBC posts versus Fox News posts. Using a emoji sentiment “dictionary” compiled by (Kralj Novak et al., 2015, PLOS ONE, “Sentiment of Emojis”), I calculated the average emoji sentiment based. I excluded positive sentiments from laughing faces, because very likely they indicate that the user is amused, rather than being positive by the candidate. I found that the emojis were more positive about Trump than Clinton on Fox News (see sentiment scores in the following table, but not on NBC. Surprised? (Not!)

Media Mention both Mention Clinton Mention Trump
Fox News 1.10 0.54 1.79
MSNBC 0.24 0.38 0.35

Table 2: Sentiment score of emojis in comments to facebook posts: positive score indicates a positive sentiment

So overall, we see that the candidates spent much of their time attacking each other, though using different styles. Clinton was positive as usual and her attacks (at least in terms of words) were less brutal and more camouflaged. Trump used more negative words, and his attacks were brutal and direct. However, this kind of personal attacks were encouraged (intentionally or not) by the audience questions and especially by the hosts. They were not impartial and their questions reflected their own judgement. What was surprising was that the people who watched the video still found things amusing or positive, though unsurprisingly, how they felt was influenced by their political stance. It is not very clear who won this battle, but what a battle it was!

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s