Exit polls have lost much of their sheen after some priceless flops. But Yogendra Yadav and his team at the centre for study of developing societies (CSDS) came closest to predicting the Uttar Pradesh elections right, for CNN-IBN, with just a sample of 7,000 in a state of nearly 15 crore people.
Image: courtesy Mail Today
IJR 
Was this an exit poll or a real pre-poll public opinion survey? I had the impression that it was the latter, which is what enabled Yogendra Yadav to stratify the sample by various parameters. But finally, I think Yogendra went by a hunch in converting share in the popular vote into seats won. By any criterion of logic, if you have a good sample design, the share in total votes should be more accurately predictable than number of seats. But Yogen got his figure on popular vote share seriously askew — giving the SP a ten point advantage over the BSP. Fortunately, he proved rather conservative in converting this vote share advantage into seats advantage. That is why, despite the SP having only half the vote share advantage over the BSP that his sample indicated, the seats advantage has worked out pretty much as he predicted.. All it proves is that election predictions are a gamble, especially in a mutiply polarised political context such as Uttar Pradesh.
Sukumar is absolutely right in saying that we over-estimated the lead in terms of votes and under-estimated its impact on seats. These two compensating errors cancelled each other and our final forecast was closer to the outcome than others.
Two minor corrections: ours was a post-poll survey, not a pre-poll (all the more reason why we should not have over-estimated the votes so much).
Headlines today’s was a political estimate made by my cooleague Mr Sanjay Kumar. This was not a survey based projection.
Yogendra Yadav, CSDS
If others are miles away then a mile away would be considered closest.
Yogendra Yadav was spot on, even in his predictions of the jayalalitha wave in last year’s assembly elections.
I think his strength is his deep understanding of political processes, and the nature of the ferment in society at given points of time which helps him marry sociology with statistics.
Most others who pass off as psepholgists deal with voters like they were just digits in a vast statistical operation.
The survey however predicted a vote share of 34% for SP, way off the figure of 29%. They got the number of seats in right range just by fluke. Such a wide off the mark figure indicates problem with sample selection which should actually have been corrected then itself. Should we go ahead with a sample size of just 7000 to predict the behavior of 150 million voters.
[…] The ideal scenario is one in which you accurately capture the vote share of different parties and convert these into the correct number of seats. But you can also get the right answer if you make two opposite errors that cancel each other out. This is what occurred with CSDS’ 2012 Uttar Pradesh post-election poll, in which CSDS was the only pollster to correctly predict the number of seats that the Samajwadi Party and the Bahujan Samaj Party would win in that complex four-cornered contest. As Yogendra Yadav later wrote: […]