Election metrics | Analysing exit polls, Election , 2014 lok sabha election updates, Lok sabha election 2014, Exit poll, 2014 election , latest news.
- Details
- Parent Category: Political News
- Last Updated: Thursday, 22 January 2015 20:48
Election metrics | Analysing exit polls, Election , 2014 lok sabha election updates, Lok sabha election 2014, Exit poll, 2014 election , latest news.
The polls that have been published so far are trying to make judgements based on people’s opinions almost a year ahead of the general election.
Polling in India is a hard business for several reasons. First, the country is politically fragmented. We have a large number of regional parties, whose influence is restricted to certain states or even parts of states. Then, even if two regions see battles between the same parties, elections there are more often than not likely to be fought on different issues. Finally, and most importantly, we have a first-past-the-post system combined with multi-cornered contests, which makes predictions difficult.
The standard practice for pollsters is to divide the country into regions and conduct surveys in randomly chosen constituencies in each region. The results of these surveys are used to arrive at a possible vote-share distribution in each region, which is then put through what is likely to be a complicated formula to be turned into seats. While each step of this process is prone to errors (what if one didn’t segment the regions properly? what if the samples weren’t random? what if there were errors in aggregation?); the trickiest step is easily the last one—where the total partywise vote share in a region is to be converted into seats. For example, in the 2009 parliamentary elections, the Bahujan Samaj Party (BSP) was the largest party in terms of vote share in Uttar Pradesh, garnering 27%. However, the party secured only 20 seats out of 80 in the state, less than that secured by the Congress party, which had a vote share of only 18% (figure 1). The same figure shows that despite performing similarly in terms of vote share, the Bharatiya Janata Party (BJP) got half the seats that the Congress got.
So, assuming that Uttar Pradesh is a region for my opinion poll (it is unlikely to be and, given its vastness, I would poll separately in each sub-region), and that my polls throw up the same vote shares as were reflected in the elections in 2009, how should I predict?
Or take, for example, the southern state of Karnataka (figure 2). The BJP had a small lead over the Congress in terms of vote share but got three times the number of seats. Again, given vote share figures of 41%, 37% and 13% for the three major parties in the region, what are the odds that you would predict that they would get 19, six and three seats, respectively?
Then there is the time factor. The next general election is scheduled for May, unless it is brought forwardSo the polls that have been published so far are trying to make judgements based on people’s opinions almost a year ahead of the elections. Considering that factors such as pre-poll alliances and choice of candidates are key in determining the results of elections, it is extremely likely that people’s preferences will change as we get closer to the polls.
Rajeeva Karandikar, director of the Chennai Mathematical Institute and one of India’s top election statisticians, says in The Hindunewspaper that the churn (of voting intentions) that happens in the last few days before the elections is much more than that in the preceding weeks and, thus, the predictive power of any opinion poll conducted even weeks before the elections is weak (http://goo.gl/9dH5I2). For this reason, we will not assume any predictive power for the recent spate of opinion polls. Instead, we will look at them as barometers of how the country would have voted around the time when these opinion polls were held.
As we write this, five agencies have conducted and published the results of their opinion polls. The results are summarized in table 1. However, before we dive into analysing these polls, a caveat is in order. Elections are still some time away and we don’t know what the National Democratic Alliance (NDA) or the United Progressive Alliance (UPA) or the Third Front will actually mean. We will assume, however, that the pollsters have defined these alliances as they existed on the day when they conducted their polls.
The first thing that should strike you when you look at the table is the degree of confidence pollsters (apart from the CNN IBN-The Hindu-CSDS group) have in their predictions. It is a reflection on the state of the Indian opinion poll industry that in polls conducted almost a year before the elections, only one agency thought it worthy to give a prediction range. Given that these are the results of opinion polls, it is unlikely that any statistical model would throw up a precise number as prediction.
Still, let us assume that the pollsters have been accurate and honest in their analysis, and see what the polls show.
A simplistic reading of the polls shows that the BJP-led NDA is slipping. While both the polls that came out in May gave the alliance over 200 seats, the alliance’s numbers in subsequent polls have been dwindling.
As far as the NDA is concerned, two important events took place in early June, which is likely to affect its fortunes in next year’s elections.
First, at a party convention in Goa in early June, Gujarat chief minister Narendra Modi was declared as the leader of the BJP’s campaign for next year’s elections.
Less than a week later, the Janata Dal-United, or JD(U), which had been part of the NDA since its inception, decided to break away from the alliance.
In the 2009 elections, the NDA got 32 out of the 40 seats in the state of Bihar, with the JD(U) contributing 20 of them.
With the alliance breaking down, this number is under threat, and might be the reason for the NDA’s lower numbers in July vis-à-vis May.
As for the incumbent UPA, the polls don’t show any particular trend. Nevertheless, the polls are in agreement that this alliance is likely to get far fewer seats than in 2009.
In 2009, the Congress, which leads the UPA, alone got over 200 seats. As per the above polls, it is unlikely that the entire alliance will come to that number next year.
Back in May, when the exit poll results for the Karnataka assembly elections came out, while there was vast disparity in the seat predictions by the various agencies, there was a near consensus in terms of vote share (http://www.mintasia.com/karnatakapolls). Again, we find better consensus when it comes to vote share (interestingly, only three of the five polls have published vote shares).
The NDA is predicted to get around 27-29% of the total votes in 2014 (up from about 23% in 2009), while the UPA is likely to drop to the same range (significantly down from the 37% it got in 2009). Again, the fact that near agreement on vote shares has led to vastly different seat projections shows the difficulties in converting votes to seats.
Apart from polling on possible seat distributions among parties in the next Parliament, one other question most polls tend to ask is the respondent’s preferred choice as the country’s Prime Minister.
I’m of the opinion that given our Westminster structure, such polls make no sense.
For example, I might want Modi as my Prime Minister (PM) but the BJP, or any of its allies, fails to put up a candidate in my constituency. In this case, my desire for Modi becoming PM achieves nothing. Similarly, if the BJP were to put up a crook as its candidate in my constituency, it is unlikely I will vote for him even if I strongly desire that Modi become PM. Preference of one prime ministerial candidate plays only a small part in the overall analysis when a voter decides who to vote for. Hence, surveys on preferred prime ministerial candidates are best ignored.
The biggest problem in analysing opinion and exit polls in India is the lack of disclosure on methodology. Apart from the CNN-IBN-The Hindu-CSDS poll, which put out a detailed note on the methodology(http://goo.gl/xQMw8B), none of the polls even mention the number of respondents for their polls (at least, not in any of their public disclosures in newspapers).
While this is perhaps consistent with a culture that sees it fit to give out a single number as a prediction a year before polls, the lack of disclosure of methodology significantly impacts the credibility of a poll.
Hopefully, as we get closer to the elections, we will get more disclosures, which will help us make better predictions on what India’s 16th Lok Sabha is going to look like.
source:http://www.livemint.com/