The influence of Artificial Intelligence in electoral surveys

La influencia de la inteligencia artificial en los sondeos electorales

From C3 Digital we review practical cases that can inspire us to optimize our business models in the use of technology. In fact, we have been interested in reviewing the models used in the pre-election forecasts in the United States. We have evaluated two, the first is ‘The Economist’ (which combines surveys with advanced AI analytics). The second, the City University of New York (social listening, big data with AI).

According to The Economist (2020), pre-election forecasts indicated that Joe Biden will probably defeat Donald Trump in the elections. As in most predictions that are used, past patterns of voter behaviour are applied to new circumstances. For example, if you answer the question, “How often have previous candidates won similar positions?” If those historical relationships break down, the prediction will fail. But one of the paradoxes presented by Trump’s unprecedented presidency is that voters have mostly treated him as they would be any other Republican. This quantitative model is updated every day and combines state and national surveys with economic indicators to predict a variety of outcomes.

Precisely this model overestimated Biden ahead in so many states around the country that it was hard to imagine him losing positions in all of them. As the counting indicates, Biden has more than 305 votes, which is a 95% confidence interval. A level too high of certainty for the prediction. The following graph shows how The Economist failed in the pre-election surveys, underestimating the Republican vote.

University of the City of New York

On the other hand, the University of New York City used an artificial intelligence model in the Democratic candidate Joe Biden would be the winner over the Republican candidate Donald Trump in the US presidential elections. This prediction model was based on machine learning patterns trying to predict the future by extrapolating data from the past, combining big data technology with the ‘social listening and social understanding’ of more than one billion opinions on social networks. According to the experts of this university they affirm that, the investigation had already begun in the previous elections of 2016; and ending with the predictions for the 2020 US elections.

Taking into account that there is a lot of activity that occurs in each election, the researchers use this information to train their machine learning model by training a neural network using AI to predict the opinion of each user on the social media platform. The statistical analysis based on more than 800 million tweets in total gives weight to the results, thus making an approximation to the prediction.

However, the university researchers reveal that the method has certain limitations, because it is not yet fully ready to predict and call elections in an accurately way. Given that, one is the sampling bias that takes 20 million people across the United States, as long as they are the people who actually use Twitter.

Influence Artificial Intelligence in surveys

When only raw data is taken directly from social listening & social understanding tools’ without doing any of the weights or scales, Joe Biden has a huge advantage to win by high volume. However, in cases where there is a large voter turnout of people living in rural areas, Twitter penetration is very low. Another bias would be that you cannot predict how many people will vote. Therefore, controlling these two variables is important for the AI ​​tool to predict accurately. Therefore, the solution is to try to calibrate the current model, based on the previous choice, and including this choice.

The reflection of both models starts from the fact that the global context must be taken into account. These elections have been celebrated in the middle of a pandemic and a volatile economy; as well as the identification of variables that can help us improve the AI ​​training pattern to obtain optimal results.

C3 Digital has solutions and practical cases of machine learning, social listening & social understanding that can transform your business.

Follow our LinkedIn accounts to keep up to date with everything. MST / C3 / eAlicia.

Leave a Reply

%d bloggers like this: