TT Epaper
The Telegraph
Graphiti
 
CIMA Gallary

Enter, exit polls & butterflies

New Delhi, May 11: Mathematician and veteran psephologist Rajeeva Karandikar knows that this week will present a challenge unlike any he has faced over the past five years.

Karandikar had successfully foretold Mamata Banerjee’s triumph in 2011, the DMK’s victory in the 2006 Tamil Nadu polls, and its defeat to Jayalalithaa’s AIADMK in 2011, and Nitish Kumar’s wins in Bihar in 2005 and 2010.

But this week, if he gets his prediction of the winner in the 2014 Lok Sabha elections wrong, those past successes will count for little.

“This is the real challenge,” Karandikar, director of the Chennai Mathematical Institute — one of the country’s top math schools — told The Telegraph.

“People will forget all the state Assembly elections we called right. They’ll just remember the Lok Sabha polls we got wrong and keep reminding us about that error.”

For four days between the end of polling tomorrow and the declaration of results on May 16, these exit poll numbers will rule the economy — determining whether markets surge or subside.

If eventually proved correct, they would have started a process of mentally preparing the nation for its next government. But if proved wrong, they could have triggered false market expectations that could crash equally dramatically on May 16.

“You can use the same methodology and get both accurate and incorrect results,” Pune University political scientist and psephologist Sushas Palshikar said. “The trick is to get the sample robust, and right.”

In 2004, all the exit polls — which capture what voters tell researchers after they have voted — had notoriously predicted a comfortable return to power for the Atal Bihari Vajpayee-led NDA government, and a defeat for the Congress-led UPA.

Sample pitfalls

Between 2004 and 2014 — a period that has seen switch-hits compete with the traditional straight drive in cricket coaching books — the methodology behind the nation’s other major pastime these days, predicting political outcomes, hasn’t changed much.

What has changed is the level of caution that pollsters employ in selecting samples of voters to draw larger conclusions from.

“But there are a number of factors that make predicting Indian elections harder than those in western democracies,” Karandikar said.

For instance, voting patterns are relatively static in America and Britain compared with India, he said. Exit pollsters need to track any change and measure that against earlier election results to draw predictions.

“In India, voting patterns can shift dramatically across elections, so the assumptions pollsters can make in the West just don’t hold (here),” Karandikar said.

Second, block voting, and the starkly different voting timings that different voting groups may employ, can also skew the conclusions of exit polls not sensitive enough to account for these variations.

A researcher for an exit poll firm who spends the morning outside a polling booth, for instance, may meet voters predominantly from one community or voting group, whereas an opposing block may vote late in the afternoon — after work — in large numbers.

An exit poll based on the researcher’s data would then inadvertently be skewed in favour of the first group because the voters had defied the traditional theory that they would turn up to vote randomly.

These challenges have nudged polling agencies into implementing sampling techniques more robustly than earlier.

Day-after polls

One set of such exit polls will try and get the researchers to pick a genuinely random sample of voters as they leave the polling booth. Others, like Karandikar, conduct what they call not exit polls but day-after polls, based on conversations the researchers have with a sample of those who had voted, a day after they cast their ballot, at their homes.

Both have advantages. The day-after polls allow researchers caution and care in picking candidates for the study in such a way that they genuinely represent the socio-economic makeup of the constituency.

Voters meet the researchers at the voters’ homes, in familiar surroundings, and not with other voters or political party representatives around. But exit polls conducted outside polling booths are decidedly quicker.

From Monday evening, pollsters will roll out multiple post-voting polls predicting the outcome of the world’s largest-ever democratic exercise — results that, though unknown to us, would by then have already been locked into voting machines.

Win some, lose some

Overall, exit polls’ accuracy over the years has been significantly superior to that of opinion polls, which try and track voter preferences before they vote.

But although they are built on modern statistics, excel sheets flooded with numbers, and days of sweaty sampling, neither exit polls nor day-after polls are infallible.

Most of the exit polls broadcast on May 6, 2004, had given the NDA 248 or more seats in the 543-seat house, and the Congress no more than 190. Eventually, the UPA won 219 seats, and the NDA just 187.

The closest prediction had been that of AC Nielson, who said the NDA would get no fewer than 230 seats and the UPA no more than 205.

Five years later, all exit pollsters gave the UPA the edge over the NDA — but only just. Eventually, the difference between the seats won by the two major political formations was a whopping 103 — the UPA won 262 and the NDA 159.

Nor are Indian pollsters unique in getting predictions wrong.

In 1948, exit polls in the US had widely predicted that Republican Thomas Dewey would beat Democrat Harry Truman in the presidential elections. Truman won comfortably. And in 2004, exit polls suggested current secretary of state John Kerry would defeat President George W. Bush’s re-election bid. They were wrong.

“Eventually, how far we are correct,” Palshikar said with a nervous laugh, “will be known only on May 16”.