The burning question on everyone’s lips is “Will Trump win the 2020 US election?”
Over the last 6 weeks, I’ve been analysing petabytes of the world’s online content with the power of AI to uncover this story of who is going to play the pivotal role of leader of the free world.
This one hasn’t been easy, unlike my previous predictions, but I’ve got the story.
Here is my prediction for whether Trump will win a second term.
Everyone has an opinion. We’re all entitled to one. We engage with content and place our personal bias on interpreting information to form our world view. Most people in my circle say Trump will win but are quietly hopeful he won’t.
We live in our bubbles, but the world is bigger than our bubble.
So as I watch this closely, I seek to go outside my bubble and explore the universe of data outside to understand what ‘the market’ view is, through analysing engagement and emotion.
An EMOTION is an intense feeling that is short-term and typically directed at a source (an event, a person, a word). Emotions come and go, like the tides in the ocean.
We measure over 400 emotions through natural language processing, which is a branch of machine learning, which is a subset of Artificial Intelligence (AI). We process petabytes of online content to measure emotion of markets to uncover what people deeply care about, in an unbiased way. This is how I predict human behaviour.
Emotions drive behaviour. Strong emotions change behaviour.
This methodology has been proven to predict trends in the market for brands and industry. By measuring how people deeply feel about a topic using the largest sample on earth (ie. everything online), it gives insights into what will happen in the future, through the change in people’s behaviour. If you’re curious, see it in action when I predicted the winners of The Voice 2019 and 2020, and MasterChef Australia 2020.
This works. Exciting stuff.
Okay, are you ready? Let’s take a look at the US Presidential Election.
STEP INTO THE MIND OF A DATA MAVEN
In this analysis, I looked at the data from many angles and was mindful of my personal bias when analysing and translating what I see.
I started with calibrating historical data using the 2016 election. I needed to understand how this market works concerning US elections.
Trump vs Clinton (US Election 2016)
At the time, all the major polls were saying that Clinton was going to win, easily. The infamous Nate Silver of fivethirtyeight.com predicted Clinton with a 71.4% chance of winning. The whole world believed she was going to win. All the experts and opinions around the world were aligned. Reminds me of our Australian Federal Election in May 2019.
Then what happened… the polls got it wrong. So wrong!
So I unpicked the data…and then I uncovered why.
When we measured the emotions around Trump and Clinton just before polling day in November 2016, this told a very clear story.
What you see below is an emotion wheel which depicts the emotions of the market for a narrative. Green is positive. Red is negative. The intensity of colour is the strength of emotion. Width of emotion is the contribution of that emotion to the overall emotion. Broader emotions are in the inner ring and graduating to finer emotions in outer rings. This measure can be taken over time, as emotions change.
What do I see? Very clearly, Trump was liked, actually intensely loved, more than we all thought (according to the polls). Clinton was disliked more than we thought.
What does that say about polling? Biased. Then can we please stop using polls!
This story goes deeper. It was worth taking a look at this from another angle…
The view above was measuring emotion from the English speaking market. What about the non-English speaking market – Hispanic and Latino are the largest ethnic minority comprising 18% of the US population?
The narrative was that Clinton was strong with the Hispanic population. Traditionally this ethnic group has voted for Democrats.
So what did the Spanish speaking market feel about these candidates?
Wow! Can you see what I see?
- Clinton was intensely negative (deep red) – very strong anxiety, anger, fear, distress – Latinos were not her amigos.
- Trump still dominantly negative (pale red) but not as intense
- Trump had stronger expectation (blue) – despite promising to build the wall, there were high hopes for Trump from the Hispanic community
So looking at the data, it is clear to me how the polls got it wrong.
Okay, so have I won you over with my method?
Are you ready to dive into 2020?
Trump vs Biden (US Election 2020)
No doubt the hottest question on earth right now is ‘Will Trump win 2020?’. Notice how I framed it – not ‘Who will win’ but will Trump win or lose. Subtle, but important, not just semantics.
One of the campaign slogans for the Democrats this year was “Doesn’t matter who, just vote Blue”.
That just says it all.
Trump knows how to engage people, even if he is polarising. In my world, polarising means intense emotions on either side – not a bad thing, polarising changes the world. People intensely love him (and hence he can get away with anything) or intensely hate him (and will break social distancing protocol to protest). People are driven by emotion. Strong emotions change behaviour.
In predicting this election, this is not an easy task. It will depend on which side of the polarised camp has more people who vote and where they sit across the 50 states. He knows how to market himself and create not just attention, but engagement.
Over the last few weeks, I ran analysis at key checkpoints. This has been a very volatile market to read, like trying to hoist a sail in a stormy sea, in territory I’m not familiar with. We saw lots of volatility (similar to Trump’s temperament). He has demonstrated power to shift the narrative that serves him, which has worked to get him this far.
So has it served him with the loyal base? Has he won over more people amid this global pandemic? Has his miraculous recovery from covid given people hope that America will be great again?
Let’s see what the market says…
Here is a snapshot of key events and my commentary on what I observed in the market using AI.
First checkpoint: Lead up to 1st Presidential Debate
- September: Postal voting opened
- 23 September: US covid death toll passes 200K
- 24 September: The week before 1st presidential debate
The polls were saying Biden had a very strong likelihood to win. According to Nate Silver, the forecast was a 88% chance of winning. Didn’t we hear this song before?
What does the market and AI say?
You should be good at reading these emotion charts by now…
Early signs are as expected:
- Trump is very negative and overall more disliked than liked.
- Biden is not as liked as we all presumed or told by the polls. Same lyrics to that same song again.
- Biden is not engaging. Pale tones indicate he’s ‘meh’ (that’s my technical term for ‘people don’t care’). This is a danger zone because Biden should have people feeling something more than ‘meh’.
Remember, this is a race of relativities. It’s not about who we like best, it’s about who we dislike least. Again, semantics but important.
At this point, this is reiterating my theory that Biden is not even in the picture, hence the question “Will Trump win or lose” is the key question.
Next checkpoint: 1st Presidential Debate September 29
The 1st presidential debate was televised globally and 73 million people tuned in to watch two privileged grown white men bicker in the schoolyard and call each other names. I could only watch a few minutes. My 6-year-old twins argue more gracefully than that. It was clear that the Trump strategy was to bully and dominate and expose Biden as weak. It worked, to me at least.
But stepping into a voter’s shoes for a second, if I was living in fear, loss of income, death by covid, the impending war with China, distrust in the media, #BLM, then my world is scary. I would be living in a mental state of anguish and despair. So I can understand how turning to the biggest baddest bully in the schoolyard enters my decision process. It’s survival. No time for idealism. I don’t have to like Trump as a person, but I’m looking for strength, certainty, leadership and safety.
And if someone was to ask me in a poll, I may not tell them what I really believe.
So what did the market feel, 24 hours after the 1st debate?
What I see:
- Blink and you’ll miss it. Biden didn’t move the needle. The sentiment around him did not change at all. Not good. He made absolutely no impact, positive or negative.
- Trump moved the market. Despite the overall negative emotion, it reduced in width and lightened in tone. Notice the shift towards expectation (blue). Now that is a healthy sign.
Trump won that debate. So at this point, my prediction was that Trump will win.
However, this signal wasn’t strong enough. Like a splinter in my heel, I had to dig deeper and work it out. Get it right or don’t bother. Okay *tiger mother settles down.
So I decided to adjust tack. Measuring the temperature of something so volatile required more than 2 thermometer measurements on 2 dates. This angle was already making me cross-eyed. I can’t rely on this view only. There’s gotta be a better way to analyse this.
So my question became, how has affect changed over time?
ANALYSIS PIVOT: AFFECT OVER TIME
Affect is the psychological description or underlying experience of feeling, emotion or mood. It encompasses a broad range of feelings that people can experience. Often it refers to the behavioural expression of mood (which is observable).
Another angle to look at this changing tide of emotions is to consider the overall affect of Trump and Biden, side by side, over time.
This view is very interesting when measured over time. The affect timeline below shows positive (above the line) and negative (below the line). Colours depict particular affect and emotional states.
We measured this daily from early September up to the week after the 1st debate.
Aaaah, what a delight to see this story unfolding – art and science in motion. This story is still, mostly about Trump.
This is what I see:
- Overall majority negative affect towards both candidates over time
- Overall, deep and growing fear, anger, fury panic in early September
- Deep distress and anxiety projected at Trump in line with US covid death toll passing 200k (all focussed on Trump)
- Despite the escalating circumstances surrounding the pandemic, Trump has been able to drive positive affect – joy, optimism and belonging – in parallel to the distress and anxiety (he has an unbelievable influence to shift the narrative)
- Trump has consistently driven positive affect over time (more than Biden)
- Biden has lacklustre positive affect to date – again, who’s Biden, oh, the other guy
- Biden does not show up as positive at all (strange given it shouldn’t be too hard to be likeable against Trump)
- Biden is again, just ‘meh’, the other old white guy
From this view and at this point (now early October) it looks consistent with my prediction that Trump will win.
Okay, just when I was putting the coding gloves back into their satin case, this happened…
Trump tests covid positive 1 October.
Was it real? Was it a strategy?
Nine days after Trump’s coronavirus diagnosis, he stood on a balcony at the White House and delivered an address to a crowd of a few hundred black supporters.
Trump refused to participate in the second US presidential election debate against Biden after it was announced the event would be held virtually because of his coronavirus diagnosis.
Ahhh, meanwhile, the polls were still saying Biden was going to win. Their view hadn’t changed – kinda like the old sailor asleep while this storm has been blowing up and down.
Did the market change?
(I knew this wasn’t going to be easy.)
Last checkpoint: Affect response post covid test
Now, look at that!
I’m seeing the tide turning positively towards the other guy, Biden. Quietly rejoicing here. Biden is building positive affect momentum.
We are 2 weeks out from the election.
My prediction now has been revised to Biden.
HOWEVER, don’t take out the champagne just yet.
We are still 2 weeks out and we just saw how the tides changed quickly.
If polling day was tomorrow, my call would be Biden.
HOWEVER, we know how this US voting system works. This will impact the overall outcome:
- Not compulsory voting
- Not based on direct popular vote (ie. most votes do not win, remember the other lady, Clinton? The largest democracy on earth designed the electoral college system back in 1787 to avoid populous states from exercising their full democratic right. Oh the irony.)
- All 50 US states and Washington DC have a set number of “electors” in the electoral college – roughly proportionate to the size of each state. Each has its own system of voting and counting votes.
Murky? It’s designed to be.
And lastly, we know that a record number of people are expected to vote before November 3 by opting for postal votes. This began in early September (remember, when the tide had swung towards Trump and I would have predicted Trump to win?).
So the recent warming towards Biden does not guarantee that he will win. This will be decided on the key states and whether this emotional tide towards Biden is representative of the sentiments in these states – North Carolina, Florida, Michigan, Wisconsin, and Arizona are the ones to watch.
Unfortunately, the powers of Aunty AI here are not quite accurate yet to narrow down affect measure by state. We also did not explore the Hispanic view. The assessment above has been purely based on the English view in the US.
So my prediction of who will win, may not necessarily turn out to be a reality because of these factors above.
If I had more data, I could take a look at it. But I don’t so I can’t.
Sometimes, you just have to put pens down and make a decision.
So for now, it’s looking like Biden is no longer just the other guy.
I predict Biden can win the US Election 2020. Finally, I’m in line with the polls (note: which haven’t changed at all during this tidal drama. Interesting…)
Election day Tuesday, 3 November 2020. I’ll get the coding gloves out again in 2 weeks.