Ayush Prasad
May 14: General Elections were held in 9 phases in the months of April and May 2014, to elect 543 members to the 16th Lok Sabha, the lower house of India’s Parliament. The data released by the Election Commission of India, the nodal Constitutional Body to conduct elections in India, states that 66.84% of the 814.5 million eligible voters exercised their franchise making this the largest ever democratic election in the history of humankind. The state does not fund election campaign in India, but approximately Rs 30,000 crore or US $ 500 million will be spent in the process of these elections. A further billion dollars is estimated to have been spent by various political parties.
The election is primarily between the ruling United Progressive Alliance led by the Indian National Congress Party and the National Democratic Alliance led by the Bhartiya Janta Party (BJP). The last election in 2009 was contested by over 178 national, regional and unrecognised parties. There is an unlikely challenger in the Aam Aadmi Party, who could be compared to the Tea Party Movement in the United States and has emerged from the Anti-Corruption Movement in 2011.
About half an hour after the last ballot had been cast, television channels across India began showing various “Exit Polls” or predictions on party positions that one could expect when the final tally came out from 8 am (IST-- 530GMT) on May 16.
Exit Polls
Most news channels have solicited the services of survey and analytics agencies. These agencies have offices and surveyors spread across the country. They use standard statistical techniques to analyse data. Their results are analysed and at times modified by political analysts and then presented on live television and on front pages of newspapers. The gap between polling and results of 86 hours is filled by analysing Exit Polls in television studios.
Exit Polls are serious business in India and there is a high premium on being accurate. One television channel (Times Now) even advertised that the sample size of its survey is 550,000 people, which would be approximately 80 times the sample size used for election surveys in the US, where the voters are just 8 times less than the number of voters in India.
The fundamental rule in surveys is that one ought to remove selection bias. In the US, surveyors use the Telephone Directory (Gallup’s 1936 onwards) as the universal set to randomize. The 2011 Census figures and subsequent projection shows that less than 60% of Indian Households have a phone and phone owners have multiple connections, telephone directories are not used. Instead, sample list is drawn from the publically declared Voters List in each constituency. At times, the polling station and the time duration for the survey are randomized, instead of the individual voters.
Predicting the Indian Election
I have picked up the data that was published on the front page of The Economic Times on May 12, 2014 and on the predictions made by major news channels in India on the same day. The data is represented on Table – 1.
Table 1 : Analysis of Exit Poll (1998-2004)
In Table -1, Prediction represents the data released by the Exit Pollsters while correlation shows the ratio of the prediction to the actual results. Standard Deviation is the variation in the predictions between different pollsters.
While declaring the predictions in Exit Polls, each news channel expresses a level of confidence in their figures of /- 3%, but it could be seen that in less than 8% of the predictions have been within that range. While over 40% of the predictions (shaded in red) have varied by more than /- 20% from the actual results. No single agency can be said to have phenomenally higher degree of accuracy.
The data also shows that since Exit Polls show a higher number of seats than actual seats for the National Democratic Alliance (NDA), than the United Progressive Alliance (UPA), traditionally supported by weaker sections of the society living in rural areas. This could be because of selection bias introduced due to convenience of reaching a surveyed sample. The Exit Polls in 1999, were the only time the Congress Party was given more seats than what they had actually achieved, their worst ever electoral performance. It shows moderation of statistical results.
The standard deviation between predictions of different agencies for the NDA has varied between 10-16 seats, while for the UPA it has varied from 7-12 seats. The pollsters have been more consistent with their performance for UPA than of the NDA. But in this election, the deviation for UPA has been higher than NDA, because of base-effect – UPA has been predicted to win a third of the seats of the NDA. There is no trend with regard to standard deviation, between exit poll predictions over the years – the pollsters remain as diverse as before.
The pollsters have been more accurate for the UPA at 13% than for the NDA at 16% deviation from actual results.
Predicting Election 2014
Exit Polls have been successful in predicting the winner in every election, apart from in 2004. In that year, it also failed to predict the falling trend of the NDA, but it did predict the rising trend in favour of the UPA. In 2004, both the political alliances were expected to increase the number of their seats, squeezing out other non-aligned parties.
In 2014 General Elections, the NDA is on a rising trend while UPA is on a falling trend. Hence, going by the past record of the Exit Polls, NDA could be declared the winner. The average number of seats that have been predicted in favour of the NDA is 285 seats. On an average, NDA has a positive correlation to results by 16% since 1998, therefore, it could be safe to say, that NDA would win 246 seats. Since, it is on an upswing, it could further be predicted to win 15 seats within the deviation, so my prediction is that the National Democratic Alliance (NDA) would secure between 246-261 seats. It will still fall short of 11 seats from majority. While the UPA is on a negative swing, so it could be said to win 104-120 seats only.
Elections in India are dependent on several factors beyond what can be understood in statistical numbers. Electoral victory is also a function of mobilising one's supporters in larger numbers than other candidates. At times it is about identities and positive vote towards someone, while at other times it is about negative vote against someone. The beauty lies in its unpredictability and mystery.