Biased data, personal choices: Why predictions of exit polls go astray

  • HT Correspondent, Hindustan Times, New Delhi
  • Updated: Feb 11, 2015 13:39 IST

Six exit polls predicted an AAP win but none of them could gauge the scale of the party’s landslide victory or the magnitude of BJP’s drubbing. One, by a little known firm called Data Mineria, even predicted a BJP win.

The exit polls gave AAP between 31 and 53 seats, the BJP between 17 and 35 seats and the Congress between zero and five seats. Over the last six years, exit polls in India have rarely got the final tally right, though most have accurately predicted the winner.

“Our analysis shows that only three out of the 72 exit polls since 2009 got the leading party wrong. However, the record on seat forecast has been patchy, with only 42% of exit poll seat forecasts coming close to the final tally,” said Subhash Chandra, consultant, political polling, Ipsos India, a leading multinational market research firm.

“A scientific exit poll cannot go wrong unless pollsters use their personal choices, biases or historical data to create a tailor-made survey,” added Raghu Roy, managing director, MaRS, a market survey firm.

Then, many voters, especially from economically or socially-challenged backgrounds, mislead pollsters about their voting choices. This makes predicting polls in India doubly difficult.

“There are many reasons for the wide range of variations (in the number of seats projected by exit polls). It could be on account of methodological errors, choice of interview locations or the basis of calculating the conversion of votes into seats,” said Prem Chand Palety, chief executive at C-Fore, which conducts several such polls for political parties as well as newspapers.

In the US and other western countries, exit polls have mostly been more accurate than in India. "In the US, market research firms often pool their findings. This is the way forward as in India most polling agencies reach at most 7,000 voters due to limited budget and time. If all seven pollsters had come together, the sample size would have exceeded 40,000 and would have fetched more accurate results," Chandra explained.

also read

Show comments