It's 2019, and the general elections are in full swing across India as the world's largest democracy soaks in some fierce poll banter. Like many other years, old records tumble, and new ones are created to enthuse the 900 million to exercise their mandate. The consequence of this massive exercise is that accurately predicting the result becomes a challenge. Although opinion and exit polls spell out their prophecies, there is an element of surprise that shakes up the masses and the markets. With the advent of social media, political parties have redesigned their election campaigns and devised strategies to reach their intended audience better. So, could social networks then hold clues to the impending results of elections?
A collaborative study by researchers at the Indian Institutes of Technology at Kharagpur and Kanpur, and the Princeton University, USA, says yes. They have designed a model, based on social networks of voters, online and offline, to make reliable predictions on the ‘surprise’ element of election results and suggest possibilities of decreasing it. The study has been published in the Proceedings of the ACM India Joint International Conference on Data Science and Management of Data.
"Surprise is a phenomenon only of closely contested elections", says Dr Swaprava Nath from IIT Kanpur, who is one of the principal investigators of this study. The 'element of surprise', defined from a voter's perspective, is the scenario when the voter's expected candidate does not win. “Our focus is on predicting the surprise of a given voter, who is embedded in a social network and 'perceives' the winner through her interaction with her connections in the social network,” he explains.
The sources of information for the voters about the candidates' positions are through their social connections, both online and offline. The impact of media on surprise is also important, but that is posed as a future plan of investigation in this early version of the research. Based on the connection information and their preferences (which is easy to guess since they are socially close and a voter can follow their social neighbours’ activity on the social media), they make ‘intelligent’ guesses on which candidates/parties win. If the outcome differs, the voter is surprised and could be unprepared for it.
The mathematical model proposed by the researchers of the current study is inspired by the ‘homophily effect’ of social networks. It states that people tend to form connections with others who have similar mindset. The model shows that an election outcome will lead to a ‘surprise’ when the difference between the ‘estimated bias’ and the ‘true bias’ crosses a threshold.
Estimated bias is the probability of connection to another voter (either of similar or dissimilar political view), as estimated by a voter, while the true bias is the actual probability of the connection. The values of these two biases may not be equal, thus a voter may wrongly perceive some candidate to be the winner, which may be far from reality -- resulting in a surprise/shock after the results are declared.
There are different voting rules according to which a mandate can be expressed. Some elections (this is what we use in the general elections in India) use plurality, where the voters just vote for their favourite candidate and a candidate getting the maximum number of votes win. However, there is another voting rule named Borda scoring, where voters rank the candidates and each position is given a weight – the candidate getting the maximum total weight wins. Similarly, there is a rule called veto, where every voter disapproves one single candidate. This study explores why and how a 'surprise' occurs in elections. The model and its empirical testing predict that no single voting rule can reduce the surprise element for all sections of the voters, implying that it is less likely that a voting rule can be put forward that will yield unsurprising outcomes for all voters, particularly when the election is closely contested.
The researchers have tested their model using the data of the UK-EU referendum, popularly known as the Brexit. It was a referendum held in 2016 where the UK voted to leave the European Union. It was a closely contested election with two outcomes—leave or remain. The mandate to leave came as a surprise/shock to the entire world. The researchers found that the results obtained from this model hold good for this data.
However, if the margin of victory of one party/candidate winning over another is massive, there will be less surprise at the outcome, because the voters would have expected this result despite small errors in their estimated bias. “So, if the global conditions for an unsurprising election outcome actually hold in an election, and you are a good estimator of your own political biases, then your local observations (say your friends' voting patterns) say a lot about who will be the winner in the election.”, says Dr Nath.
The study shows that surprise may come if (a) the voters are not good estimators of their biases, and (b) election is a close-contest. “While it is tempting to predict the results of an election beforehand, it is (almost) impossible to predict such a thing without extensive sampling and normalizing them statistically, which itself is a much more expensive process”, concludes Dr Nath.
This article has been run past the researchers, whose work is covered, and the institution to ensure accuracy.