Bangalore: The delimitation exercise that has redrawn India’s political constituencies could skew the various projections for the 2009 general election, defying refined statistical tools, a larger sample size and even the factoring in for political alliances.
The 15th general election is the first parliamentary poll to be conducted after the delimitation exercise of 2007 based on population density, the first in three decades.
For example, in Karnataka, though the number of constituencies since the 2004 election is the same at 28, Bangalore, its capital, has got an additional representation due to an increase in population.
Also See PDF
As a result, psephologists analysing data from exit polls and drawing up forecasts ahead of the results on 16 May have no base to fall back on.
Typically, statistical tools and models for election forecasts take into account past voter behaviour in a constituency. But delimitation has confounded the forecasts because of the change in caste, religion and language combinations of the voters, and multiple parties contesting in each constituency.
“This election is as blind as (in) 1952. Because there is no electoral history for any seat,” said Yashwant Deshmukh, director of the Centre for Voting Opinion and Trends in Election Research (CVOTER), which conducted exit polls for television channels such as India TV and UTVi using its software. “If things really go haywire, I wouldn’t be surprised that delimitation would have played a big role.”
India held its first general election between October 1951 and February 1952.
Rajeeva Karandikar, executive vice-president at Bangalore-based Cranes Software International Ltd, which analysed raw poll data for the CNN-IBN television news channel, agrees. “Yes, absolutely, (delimitation can take) everybody by surprise. We have to tried to make our best assessment, through old to new constituencies, (and) make political judgement. This (delimitation) has been a challenge and deviation from the past.”
Karandikar, a former professor at the Indian Statistical Institute, New Delhi, and his team at Cranes use their in-house statistical software, Systat, to analyse data to arrive at projections. “We believe very strongly our methodology of doing it (collecting data) a day after the poll is a lot more robust, because we can then do proper randomization and go to homes rather than end up at the booth and talk to every seventh person,” he said. “Previously, that never gave the correct social profile (and) we used to do adjustments for that.”
In the 2004 Lok Sabha elections, most pollsters had got their predictions wrong in forecasting the Bharatiya Janata Party (BJP)-led National Democratic Alliance to emerge as the single largest party, only to see the Congress party chalk up more seats and form the government at the Centre in an alliance with the Left parties.
Since then, though, the pollsters have hit the target in various state elections by using more sophisticated tools and a broader sample.
“There are simplistic versions of this (software tools), but what happens is that they work primarily in a two-party system, not in a multi-party system like ours,” said Deshmukh of CVOTER.
Indian election verdicts are not just based on who is voting for whom, but also have to factor in the turnout, caste affiliations and local issues. Besides, unlike in two-party systems as in the US, where a simple swing of votes could determine a winner, in India, political parties sew alliances that also influence voter decisions.
G.V.L. Narasimha Rao, a psephologist and BJP national executive member, who conducted exit polls for his party in this election, said that though the delimitation of constituencies poses a challenge in making forecasts, pollsters have overcome this in recent assembly elections in Karnataka, Madhya Pradesh, Delhi and Rajasthan.
Karnataka was the first state to hold elections last year after the delimitation exercise. “Delimitation is not such a spoiler as we had originally thought, because it could prove to be a major confounding factor. But it has turned out to be not a major problem,” Rao said.
A larger sample size from each constituency could help improve seat projections in the absence of historical data, he added.
Gallup Inc., which does poll projections in the US and has gone wrong only in two American presidential polls since it began the exercise in 1935, also said that if the sampling is large and done right, projections could be accurate in countries such as India.
“It is more expensive in a country like India because of the larger sample size,” said Prashant Srivastava, country head of Gallup. The company, which doesn’t take payments for its surveys, doesn’t make poll projections in India.
PDF by Sandeep Bhatnagar / Mint