Election mystery: How did Delhi exit polls completely miss the AAP tsunami?

Psephology is an inexact science, especially in a diverse voter population like that of India. Even so, one has to wonder whether all is right with the opinion polling industry where no one got it right this time in Delhi – except the broad point of a clear win for the Aam Aadmi Party (AAP). While one can leave pre-election opinion polls out of the analysis, since voter intent can change by voting day, the fact that even the exit polls got it all wrong is worrying. Exit polls should normally be more accurate, since they seek to measure actual voter behaviour, not voter intent. The exit polls this time estimated AAP’s seat tally from a high of 53 seats by India News- Axis to a low of 31-39 at the bottom end by CVoter. Others predicted seats in the mid-range - 48 (News 24-Chanakya), 43 (IT-Cicero), and 39 (ABP News-Nielsen). [caption id=“attachment_2093595” align=“alignleft” width=“380” class=" “]  AAP supporters celebrating party’s landslide victory on Tuesday in Delhi. Photo credit/Naresh Sharma[/caption] The best effort – by India News-Axis – was still a huge 25 percent off from the actual number of seats the party won (67). It is significant that an exit poll was not able to make better predictions even in a mostly two-horse race. Of course, it can be argued that the projection of seats from vote shares can be tricky in a first-past-the-post system, but did they even get it right on the vote share front? Absolutely not. Axis gave it the best shot and gave AAP 49 percent of the vote – which is still more than 10 percent off from the real AAP vote share of 54.3 percent. A 10-11 percent margin of error is far, far beyond any reasonable range of acceptability in an exit poll. It makes no sense to poll with this kind of margin of actual error. You can get a reasonable estimate of which way the wind is blowing merely by talking to cabbies and chaiwalas – which is what visiting journos tend to do during election time. That there was such a wide margin of error in a largely two-horse race seems scandalous. This is not to suggest that some of the pollsters may have fiddled with the figures to get the answers their clients were seeking, but it does raise questions about their methodology, and whether they are missing something else. It is also possible to argue that in “wave” elections, pollsters can’t really get the final seat tally right, but this is only partly valid. In Delhi, they got even the vote shares badly wrong, remember? One also wonders if pollsters ought to take something more into account beyond what people are actually telling them. For example, even as early as September 2014 – five months before the actual polling day in February 2015, when Arvind Kejriwal was presumed to be hors de combat – India Today’s Mood of Delhi poll was showing him as the preferred choice for CM of Delhi even while voters also saying that they would vote BJP. This dichotomy was never explained. In November, when the PM and Amit Shah were campaigning in J&K and Jharkhand, another poll showed the same results: Kejriwal for CM, BJP for win. If pollsters wanted Kejriwal as CM, and were simultaneously indicating a pro-BJP voting intent, what were they really saying? That they would vote BJP if they gave them a leader like Kejriwal? Or that their heart wanted Kejriwal, even though their reason told them the BJP would win anyway? Many polls also indicated that Delhi wanted Kejriwal as CM and Modi as PM as far back as 2013 December and early 2014. This signal too got drowned out in the noise. The voter, in a sense, was sending clear signals, but neither the pollsters nor the media picked it up. There is also the other issue: from poll to poll, the Congress appeared to be getting squeezed out. How come no one suggested which way this vote would swing finally? Psephology clearly is much art as science. The differing signals – on leadership, actual party preference, and estimating which way voters would swing if their own preferred first choice party didn’t look like winning – clearly need to be combined better to give us more credible final results. Seats and vote shares cannot be predicted only by relying on the answer to the main question: will you vote BJP, AAP or Congress? The voter’s real answer may often be: it depends. In Kashmir, for instance, the mere fact that the BJP was seeking power in the Valley brought out more voters to the booths in order to stop it. They voted for parties they thought would be best placed to keep the BJP out of the Valley. In Delhi, once Congress looked like a loser, the ultimate result depended on figuring out how many will stay loyal to the party on voting day, and how many will swing towards AAP. This is what the pollsters failed to do.