Divergent exit polls under question
Real story lies in sampling methods, vote-seat conversion method
With most exit polls forecasting the National Democratic Alliance (NDA) would either get a simple majority or miss the majority mark by a whisker, the possibilities of obtaining divergent results need to be dispassionately dissected.
The real story lies in how seats are chosen for sampling, how diligently swings are calculated and, importantly, what model has been adopted for vote-seat conversion. In close contests, the margin of error could be larger than the difference between the two principal contestants. And, that is where the real challenge to most projections is. To a great extent, the vote-to-seat conversion depends on the outcome of not just alliances but how the first-post-the-past system pans out. It was the first-past-the-post system that gifted 206 seats to the Congress, while a set of alliances gifted the 2004 election to the Congress, both outcomes of the Westminster model.
Today, in more than half the Lok Sabha seats, we see multi-cornered contests; in close to a third of these, we have quadrangular contests. That explains why some of the traditional psephological models could be contested, as well as the failure of many a large sample to project the right number of seats.
The challenge to the exit polls this time comes from Uttar Pradesh, Bihar, Jharkhand and, to an extent, West Bengal, Odisha and Madhya Pradesh. In UP, Bihar and Jharkhand, it is the multi-polarity of the contest that can completely alter the results. With 28 per cent of the votes, the Bharatiya Janata Party (BJP) won eight of 14 seats in Jharkhand in 2009. A mere reorganisation of non-BJP votes, even without explicit alliances, could drastically reduce BJP's tally. Many a time, such a reorganisation might not be captured in the exit poll. Now, extend this scenario to the fluid situation in UP, Bihar and a large number of multi-cornered contests; we know the probability of a surprising tally on May 16. Importantly, NDA's upsurge is heavily predicated on these two states.
Since 1999, almost all exit polls have overestimated BJP in UP. In 2004 and 2009, the party barely touched the two-digit tally in the state, though most exit polls had given it around 30 seats. Assam is another state where most polls have since 1996 anticipated BJP's rise to prominence. But the party has, so far, had only a brief moment of glory: In 2009, when its alliance with the Asom Gana Parishad helped it win four seats. It's clearly a case of failing to project seats in multi-cornered contests, and our psephological models are yet to establish they have arrived there conclusively.
Take 2004, for instance: If exit polls' projection of 35-odd seats for BJP in UP and a superior performance in Bihar (where it actually won 11 seats) had come true, NDA would have ended up with 225 seats (and not 185) and BJP would have got an outside possibility of forming the government. An error of projection in these two large states, and many others with multi-cornered contests, can completely alter the trajectory on May 16 and India could be in for some serious surprises.
The writer is a political analyst and psephologist. Views expressed are personal
The real story lies in how seats are chosen for sampling, how diligently swings are calculated and, importantly, what model has been adopted for vote-seat conversion. In close contests, the margin of error could be larger than the difference between the two principal contestants. And, that is where the real challenge to most projections is. To a great extent, the vote-to-seat conversion depends on the outcome of not just alliances but how the first-post-the-past system pans out. It was the first-past-the-post system that gifted 206 seats to the Congress, while a set of alliances gifted the 2004 election to the Congress, both outcomes of the Westminster model.
Today, in more than half the Lok Sabha seats, we see multi-cornered contests; in close to a third of these, we have quadrangular contests. That explains why some of the traditional psephological models could be contested, as well as the failure of many a large sample to project the right number of seats.
The challenge to the exit polls this time comes from Uttar Pradesh, Bihar, Jharkhand and, to an extent, West Bengal, Odisha and Madhya Pradesh. In UP, Bihar and Jharkhand, it is the multi-polarity of the contest that can completely alter the results. With 28 per cent of the votes, the Bharatiya Janata Party (BJP) won eight of 14 seats in Jharkhand in 2009. A mere reorganisation of non-BJP votes, even without explicit alliances, could drastically reduce BJP's tally. Many a time, such a reorganisation might not be captured in the exit poll. Now, extend this scenario to the fluid situation in UP, Bihar and a large number of multi-cornered contests; we know the probability of a surprising tally on May 16. Importantly, NDA's upsurge is heavily predicated on these two states.
Since 1999, almost all exit polls have overestimated BJP in UP. In 2004 and 2009, the party barely touched the two-digit tally in the state, though most exit polls had given it around 30 seats. Assam is another state where most polls have since 1996 anticipated BJP's rise to prominence. But the party has, so far, had only a brief moment of glory: In 2009, when its alliance with the Asom Gana Parishad helped it win four seats. It's clearly a case of failing to project seats in multi-cornered contests, and our psephological models are yet to establish they have arrived there conclusively.
Take 2004, for instance: If exit polls' projection of 35-odd seats for BJP in UP and a superior performance in Bihar (where it actually won 11 seats) had come true, NDA would have ended up with 225 seats (and not 185) and BJP would have got an outside possibility of forming the government. An error of projection in these two large states, and many others with multi-cornered contests, can completely alter the trajectory on May 16 and India could be in for some serious surprises.
The writer is a political analyst and psephologist. Views expressed are personal