Prediction markets have already proved to be effective and seem to have a huge potential in the future due to the wide pool of potential customers who are willing to pay for such kind of analysis. In this article, we will consider basic big data-based methods of the prediction of voting behavior.
According to the data provided by the Pew Research Center, in U.S. 86.8% of Americans are actually registered voters. On the other hand, statistics show that 55.7% of these people actually voted in the presidential election of 2016 when Donald Trump became the head of the country. Recent polling data shows that young people are less likely to vote than older people. YouGov has even called the age a new voting key predictor of voting intentions in Great Britain. Therefore, we will cover the main factors that can influence the election outcome and help make some predictions for political elections.
There are numerous aspects that influence voting behavior: the individual qualities of the candidate, who is running for a political seat. There are also other significant factors that define the voters’ choice:
Another factor is the political party that is represented by a specific candidate. These aspects can allow to simulate and predict the election outcomes using data science. So let’s get into the detail here.
The healthier country’s economy is, the more citizens will vote in favor of the incumbents. The incumbent is the person who is currently holding a political seat and has a desire to be re-elected for the next term. When the economy is in recession, the voters blame current governors for that. Conversely, if the country’s finances increase at the time of an election, voters will reward currently seated politicians with their thankful votes.
The 2015 study of the Cardiff University has shown that personal finances also influence the way people vote and which candidate they choose. The richer a person is at the time of the election, the more like he or she will vote in favor of an incumbent and vice versa.
Taking into account that people often lack information about the emerging candidates at the time of the election, incumbency helps voters make a decision. They are already familiar with those seated and know enough about the person who had a chance to improve the economic and political condition of the country. Between mediocrity and uncertainty, people tend to choose the first option. Therefore, incumbents often have obvious advantages in comparison with new candidates, especially when people evaluate current president positively. The more people are satisfied with the current president, the more chances that he will be re-elected there are.
If the current president can no longer be re-elected, the trend of the typical voter’s behavior associates the party, which the current president belongs to, with the incumbent. Therefore, if the new candidate belongs to the same party, he is automatically associated with both the party and current president. So, such a candidate has more chances to be elected.
Voter Fatigue is the factor that represents a level of voters’ tiredness of the frequent necessity to make choices and vote. With a high voter fatigue, people hardly get motivated to evaluate candidates in detailed and careful manner. In this case, voters rely on the incumbency and the closer they get to an election date, the lower participation percentage we get according to the research made by Sebastian Garmann, a chair of Public Economics of the Dortmund University.
To predict the election outcomes, various methods and techniques can be used. Big data market participants develop and implement new algorithms, mathematical models, and tools that allow to analyze data from thousands of sources and predict the winner of the upcoming election. Let’s consider the methodologies that help determine a future voting behavior.
A lexicon-based sentiment analysis is a predictive model that is aimed at improving outdated election prediction techniques that measure mostly the political attention towards to a particular party instead of evaluating its political support. The main problem in using a sentiment analysis is the differentiation of positive, neutral, and negative statements about a specific party on the Internet. This is where data science comes on stage. A supervised machine learning approach on a base of the hand-labeled training set allows calculating the percentage of positive references to a specific candidate. The advanced approach would be classifying sentiments that support or oppose a particular candidate instead of calculating positive or negative statements.
Data collection can be either manual or automatic. The main task is to wisely remove a “data noise” and collect the needed information as precisely as possible. To do so, it is important to choose the right keywords for text mining, such as names of the candidates and their current posts.
Existing tools and technologies for data collection and election predictions:
To ensure a correct predictive modeling, it is crucial to segment collected data in a convenient way (time, state, gender, etc). The dataset should be divided into two-three parts by specific periods corresponding to 2-3 months. It will help better understand how opinions are changed within each period and calculate a more precise result by comparing two datasets and generating an average number.
To segment opinions, various classifiers can be used: modified Huber, Logistic Regression, and Support Vector Machine (SVM) together with regularization methods (Elastic net, Lasso, and Ridge Regression). However, the calculation error can reach up to 15-20% according to the data of the New York based Levich Institute and Physics Department. That is why instead of the individual classification of each reference or mention, it is better to use the algorithms that directly estimate the aggregated distribution of opinions.
Despite the fact that U.S. election is decided by the electoral college, predictive modeling and calculating the chances of each candidate using data science is crucial for prediction of the election outcomes using the opinions of those who have already made a choice. Although, as the recent US presidential election showed, the opinions of undecided people had determined the final result.
Factors that usually influence a voting behavior can be used for the prediction of how undecided people will vote. And when it comes to estimation based on opinions of those who have already decided whom to vote for, big data analysis technologies and methodologies seem to be an effective and precise way to predict the outcome of the election.