In the run up to the Tamil Nadu election results day, four exit polls have predicted a victory for DMK, while only one, C Voter, has projected an astounding for the AIADMK.
There has been much criticism directed towards polls (opinion and exit) due to the discrepancies. We speak to Yashwant Deskhmukh, Founder Editor of C Voter on why their exit poll gave Jayalalithaa’s incumbent government an edge and the basis of the findings.
1. Out of five pollsters, C voter is the only one to predict a clean victory for Jayalalithaa. What is the exact basis of this analysis?
This is nothing different from what we do regularly or what we have done in other 4 states. It’s a routine data process that we have been doing for 20 years; can’t really see any deviation from the same.
I don’t see any positive swing towards DMK + Congress barring Chennai region. In other words our data did not capture any huge anti-incumbent wave which would change the equation of 2014. Practically speaking there is no change in political alliances as such in last two years. If DMDK would have contested along with DMK then the opposition unity would have been full and intense. Same goes with PMK. Even if you add 2014 votes of Congress to DMK; just these two are nowhere near to 44% that Jaya polled all alone. In order to change the equations dramatically between 2014 and 2016, we must have a huge anti-incumbency wave built up; and that should have been evident in the data.
As we could not sense any such huge anti-incumbency wave in our data; so the projections remain same. There was no variation in terms of important demographics. Females and first time voters also were showing similar patterns as of the rest of the population. So we are just going by what the data is reading or telling us. It can go wrong only in one condition; that the DMDK voters voted for DMK + Congress, but did not tell the same to researchers. If that has happened; we might go wrong. If not; we will get it right.
2. Why do you think the polls differ so much? Are you becoming uneasy? Why should yours be closer to truth and not of the others?
No. I am not uneasy at all. This is the 3 rd election in a row that we have differed from others on Tamil Nadu. We got it right each time and there is no reason for us to question ourselves. Odd thing is that four other agencies who have a track record of polling in Tamil Nadu: CSDS; ORG; C-Fore and Nielsen; have NOT polled Tamil Nadu this time. The agencies who have done the polls have never polled in Tamil Nadu in earlier elections, so it is really difficult to compare in that perspective.
We also asked Deshmukh few questions on their earlier polls and on West Bengal exit polls. Here are the answers.
1. Given the dreaded muscle power of Didi's men in West Bengal, is it not possible that respondents sought to avoid stating the truth?
The fear-factor or lying-factor or "spiral of silence" is a global phenomenon. There is hardly anyone who can do anything about that. Even in absence of fear, sometime the peer-pressure or the pressure to be seen correctly gets a wrong answer from the respondents. Big time Modi-supporters did vote for Nitish in Bihar Assembly and Kejriwal in Delhi assembly. But they didn't say so in the interviews outside the booth as there were family/friends around them. That resulted in underestimating the extent of sweep that Nitish and Kejriwal had. But even then we picked the winner correctly. These factors are the risks of this business; can't really help.
2. There have been a spate of controversies in TN over the opinion polls; inevitably people are sceptical of exit polls. Would you suggest some regulatory authority, even if non-binding, just to keep a tab. and thus make the process transparent?
We are working on the formation of IPC (Indian Polling Council) on the lines of British Polling Council. The 6 major agencies have agreed to cooperate and take this forward as internal regulator. Problem is not if polls go wrong. Polls go wrong many a times the world over; even in countries where it’s been done for almost 80 years. We have seen polls go wrong in recent UK parliamentary elections, and also in the ongoing US primaries. Polls can go wrong for variety of reason. That's not the problem. The real problem is when we don't learn from mistakes, when we are unable to tell our clients the "limitations" of the polls. When there is virtually no transparency. When we are unable to distinguish between a genuine polling exercise and a non-genuine one. When we are unable to explain what you should expect from us and what you shouldn't in terms of pure statistics. When there is no dialogue between peers of polling industry. When we are unable to share our experiences which could help everyone grow.
All these issues are important and critical. That is why creation of IPC will be a landmark in the direction to overcome or find answers to all these issue.
Last but not the least, talk to any damn pollster in the world, they will refuse to poll in Tamil Nadu, simply because of the complexities of demography. It’s the most dangerous playing field full of stats-mines for any pollster. I believe we are over-critical of what our pollsters are trying to do. Talk to any pollsters outside India and they will tell you how and why it’s really difficult to poll in India. So we should stop comparisons. Others do well not just because they have been doing that for a hundred years. Others do well because they have a much, much simpler demography to cover and a simpler society to map. Give the equation of Caste based parties in any one single assembly seat of Tamil Nadu to any American or British polling expert and then check his/her reaction. You will know. It is very easy to criticise ourselves and very difficult to appreciate what home-grown talents are achieving in this line of studies.
3. If most pollsters failed in Bihar, could you not be wrong now? Have you sharpened your methodology?
I don’t consider Bihar as a failure. Not even Delhi. Pollsters’ job is to pick a winner, and they did it correctly in Bihar (barring one odd pollster) as well as Delhi. The extent of "seat-share" is NEVER a part of polling science. There is no direct correlation of Votes and Seats in a FPTP (First-Past-The-Post) system. Remember, Jayalalitha polled more votes that Karunanidhi in 2006 election but ended up with much less seats. In 2004 Lok Sabha NDA got more votes than UPA but lost power. In 2008 elections of Karnataka BJP polled much less votes than Congress, but still got the majority!
Our science stops at Vote Share estimate, that too within a statistical margin of error. The Indian media is absolutely obsessed with the "Seats-Projection". So much so that they don’t even show the vote share, even if they show they don’t discuss in. I had to struggle for 4 hours yesterday to compile the vote share projections by different channels. They were only flashing the Seats-Projections all the time. This is not the way it should be.