The Role of Emotions in Decision-Making
While people’s brains differ in how quickly they make decisions, or how confident they are about them, all modern science, from behavioural economics to affective neuroscience, agrees that emotions are essential in decision making. Humans are rational and emotional, and our moment-by-moment subjective sense of wellbeing influences our decisions.
If both thoughts and feelings COMBINED drive decision making, how do we measure this complex process?
In a changing world affected by realities such as climate emergency, automation, and an aging population it is increasingly important to understand and measure the collective emotions and feelings which heavily influence some of the most vital decisions in history — such as elections.
This case study illustrates the value of integrating emotion-based metrics in quantitative approaches. Excitingly, it demonstrates the improved predictive accuracy we can achieve when emotion-based measures are taken into account.
Adding the “secret sauce” of emotion text analytics improved the accuracy of a predictive model and uncovered the WHY behind voters’ decisions.
It shows how measuring emotions helps us to better understand why voters behave the way they do. Understanding the WHY could allow candidates to find better ways create and communicate their platforms, perhaps based on different measurements of success such as empathy, trust, and the wellbeing of their constituents. This would increase candidates’ likelihood of success and help create a stronger, more just society.
The Standard Code of Emotional Language
A study by Cornell University neuroscientist Adam Anderson shows that although feelings are personal and subjective, the human brain turns them into a standard code that objectively represents emotions across different senses, situations and even people. Anderson writes: “Despite how personal our feelings feel, the evidence suggests our brains use a standard code to speak the same emotional language.”
Emotion text analytics is a relatively new area of analytics and an exciting method for measuring emotions. Based on Anderson’s findings that all human brains generate a “standard code” for the entire valence spectrum from good to bad, and express it accurately using language, we can rely on emotion text analytics to improve the accuracy of predicting future events that are based on human decisions, such as elections, as well as consumer and shopper decisions.
Uncovering The Emotions Behind Voter Decision-Making
In the Spring of 2016, Heartbeat AI partnered with Lewers Research (Australia) to discover the emotions of Australians prior to the National elections in order to predict the election outcomes. The experiment worked better than we ever dreamed, so we repeated it using the same approach a few months later in the United States.
We used a combination of closed-ended and open-ended questions — an easy method to collect quality data related to people’s emotions.
Traditional or “rational” questions
Open-ended or “emotional” questions
Both studies were carried out in the same manner
The Google Survey Platform was used to collect data over a 3-month period leading up to election day.
An A/B testing approach of the two leading candidates was used to gather rational and emotional data about each candidate.
The open-ended responses were processed by the Heartbeat software to identify the primary and secondary emotions embedded within the context of the respondent’s answer. These binary variables were then weighted based on the combined factors of age, gender and region.
A Random Forest methodology was used to build and train the model.
We built three models: “Rational” (based on the three traditional questions), “Emotional” (based on the two open-ended questions with text converted into emotional variables), and a “Combined” model.
Three models were built in isolation to understand the predictive accuracy of the rational and emotional drivers on their own.
The “Combined” Model predicted the election outcomes at the National level
The 2016 Election was the closest race in Australian history since the 1940s, with a margin of just 0.72%. The “Combined” model was successful in predicting the results on both the National and Regional levels.
Australian Election: Actual and Predicted Results
The top 5 variables of the emotional model helped explain why Malcolm Turnbull won the election. While Joy and Anger played a similar role in voters’ preference, the differentiating factor was Trust. Being the incumbent, Turnbull had established a level of trust among Australians which provided sufficient advantage to win the election by the narrowest of margins.
The “Combined” Model predicted the election outcomes for each of the four swing states we identified
We used the same approach for the US Election model, but instead of focusing on predicting the results on the National level, we identified four swing states: Florida, Iowa, Ohio and Pennsylvania (because they were the most challenging states to predict). As with the Australian model, the strongest accuracy comes from using the “Combined” model.
Pollsters in this election were confident in their predictions of a Clinton win. And using only rational drivers, Clinton did appear as the stronger candidate based on experience, demeanour, political know-how, etc.
But once again, emotional drivers helped to explain how Trump won this election.
Predictably Predictable Emotions
The top five variables of Joy, Anger, Fear, and Trust within the emotional model clearly help to explain the role of voter emotions in selecting a candidate.
The Heartbeat algorithm identified voters’ secondary emotions and discovered their powerful influence on voters’ choice. Secondary emotions provided insight into why Trump was successful in the four key states polled.
For Trump, Anger was more aligned with the standard type of criticism directed at politicians.
For Clinton, the emotions were stronger and more negative, with resentment being the strongest.