New Delhi, Nov 13 (IANS) Inaccurate predictions by a number of exit polls on Bihar elections has brought the reliability of such surveys into sharp focus, say experts.


It was vital, they suggest, that survey organisers choose the correct sample reflecting social, geographic, demographic and caste realities, to avoid going so wrong.

Experts IANS spoke to said that exit polls should follow a scientific method and ensure reliability of the data, to ensure the credibility of polls. They also need a moral code of conduct.

V. B. Singh, a former director of Centre for the Study of Developing Societies (CSDS), said there were a few "non-serious" agencies who may resort to short-cuts without following proper sampling procedure.

"They are not choosing the right kind of sample. Investigators are not being properly trained and there is improper monitoring to see if the field work was being done honestly," Singh told IANS.

Singh, who has decades of experience in voting behaviour studies, said early voter may have different orientation from those who come in the afternoon and evening.

"There is need for day-long work. My hunch is that there is less coverage and it is extrapolated. If a scientific method is not followed strictly, they are bound to fail," he said.

While a few exit poll surveys had predicted victory for the BJP-led alliance, some others predicted a slender victory for the Grand Alliance. Today's Chanakya stood out with its prediction that BJP-led National Democratic Alliance would get 46 percent of the vote (with a margin of error of three percent) and 155 seats (plus/minus 11 seats).

It had also said that the Grand Alliance -including Janata Dal-United, Rashtriay Janata Dal and Congress - would get 39 percent vote (with a margin of error of three percent) and get 83 seats (plus/minus nine seats).

News television channel NDTV, whose exit poll had a sample size of 76,000, had predicted 120-130 seats for the BJP alliance and 105-115 for the Grand Alliance.

The Grand Alliance eventually triumphed with 178 of the 243 seats with a 42 per cent vote share while the BJP and its allies could get only 58 seats with 34 percent of the votes.

Dainik Jagran survey had given 130 seats to the BJP combine and 97 to the JD-U and its allies. The ABP News-Nielsen predicted 130 and 108 seats, respectively while India Today-Cicero gave 120 and 117 seats, indicating a hung house.

Ironically, the exit poll which came out with the nearest prediction - Axis Ad-Print-Media (India) Ltd - was not carried by CNN-IBN which had ordered it, on the ground that they did not get the raw data and that the pollster could not explain the numbers.

Axis, which published the results briefly on its website, had give 58-70 seats to NDA and 169-183 seats to the Grand Alliance.

Both Today's Chanakya and NDTV apologised for getting it so wrong. Prannoy Roy, founder-editor, who had generally predicted several elections correctly, said on NDTV website that "the data from the fieldwork agency, normally a very reliable agency, was incorrect and this happens." He said they were looking into why it went wrong.

Today's Chanakya attributed its huge error to "interchanging" of computer template code marking of two alliances but the explanation seemed a bit stretched, leaving many questions unanswered.

How could the agency claim an error at the aggregate level, that is, at the level of voting percentages for Grand Alliance and NDA when the aggregate number is nothing more than a sum of the data at the disaggregate level, ask experts.

An official spokesman for Chanakya had earlier sought to explain the apparent contradictions by saying "when my whole coding is wrong, interpretation is (also) wrong."

Those who predicted narrow victory for Grand Alliance included Times Now-CVoter (111 seats for BJP+, 122 seats for JD-U+), News Nation (117 BJP+, 122 JD-U+) and the NewsX/CNX (95 BJP+, 135 JD-U+) .

Former Aam Aadmi Party leader Yogendra Yadav had forecast a clear majority for the Grand Alliance.

The post-poll survey done by Lokniti-CSDS (not an exit poll) for The Indian Express, had indicated a distinct turnaround in favour of the Grand Alliance. The survey released before the counting on Sunday (November 8) said that NDA seems to have slipped while the Grand Alliance seems to have a lead of four per cent over the BJP-led alliance.

Sanjay Kumar, director CSDS, said there appeared to be urban bias in samplings of surveys that went off the mark.

He said there is always a rush to show exit poll results on TV after polling ends and there could be a tendency to not go to villages.

"It is basically a problem of sampling," he said.

Kumar also said that he personally wanted the exit poll agencies to give explanation for their predictions.

"There should be some explanation on what went wrong," he said, adding that such explanation should also be meaningful.

He said the agencies doing exit polls should evolve "some kind of moral code of conduct."

BJP spokesperson G.V.L. Narasimha Rao said that in an election in a state like Bihar it was important that the sample closely corresponds to the caste structure of the population.

"In case the sample is not representative of different castes, it could lead to misleading findings," he said.

Rao, who is also a well-known pollster, said certain states reflect unique behaviour in terms of caste.

"A sampling technique that does not help maintain social mosaic of electorate is fraught with risk of projecting a wrong winner," Rao told IANS.