Exit poll forecasters have been proven wrong in the Karnataka Assembly elections. The BJP has won — a verdict which few forecasters saw coming.
Is exit poll forecasting an impossible task? Is it futile to predict election results in India?
In order to get answers to the above and more, IBNLive.com organised a chat with psephologist Prof Rajeeva Karandikar.
Sameer: Some of the other surveys have predicted the results to a greater accuracy this time. And you had done better in some of the other surveys. Why does each survey throw up different results, adopting a similar process?
Prof Rajeeva Karandikar: The processes are similar, but samples are different and then methodology used in predicting seats is also different.
Sandeep Khanna: We all understand that your poll is a representation of the sample of the society, then how come most of your poll results show BJP as always getting lesser seats than they actually do (I do not have selective memory and let’s not count Bihar?
Prof Rajeeva Karandikar: Why not count Bihar! If you leave out states where we predicted BJP victory and harp on those where we underestimated, what we can say! I will only say that we are not motivated and do try our best to give the picture as we see from the data.
Sankar: Do you find any possible way to correct this predictions?
Prof Rajeeva Karandikar: There is always scope for improvement. We will try, but cannot guarantee that we will get seat projections on the dot.
Rahul: You are saying that after the delimitation, the redrawan areas was the reason for the wrong forecasting. Then how come the private channel in Karnataka (Suvarna TV) was able to predict near perfect exit poll? And there was no delimitation in Gujarat, how did you go wrong there?
Prof Rajeeva Karandikar: Delimitation was one of the reasons. Along with that, the error in vote estimates for BJP was the other main reason. In Gujarat, we had predicted the winner correctly, only the extent of his win could not be picked up by us. Full marks to Suvarna TV and Cfore for getting Karnataka poll correctly.
GV: Why is it that exit polls in India are so off target, while the exit polls in countries like US are so accurate — improper techniques or incomplete respondents’ data?
Prof Rajeeva Karandikar: The ground realities are different. Presidential election is easier to predict than the parliamentary elections. Techniques are chosen taking into account the situation and we try to do the best.
Muralidhara Bhat M: How do you take the data from different places and believe that they are perfect?
Prof Rajeeva Karandikar: The data is sent to us via phone, fax etc. And we do look at demographic profile of the sample and compare it with the population profile. Usually, these do match. If they don’t then we might have to apply correction. But this does not happen often.
Anwesha: Thanks to Prof Yadav and you for saying 'we have got the exit polls wrong!' Exit poll has always been a matter of debate. Still, I love to see the exit polls. How difficult is it to translate the seats on basis of vote share? Which one is more scientific — exit poll or post poll? Earlier, in most cases you have got most figures and trends right; hope next election prediction will hit the bull's eye!
Prof Rajeeva Karandikar: The vote to seat translation is somewhat difficult, in the sense that best model also may still give error. Our assessment is that even if we had got Congress and BJP votes shares correctly in each region, we would have still predicted lesser seats for BJP — 99 to be precise.
Rushabh Doshi: Does the increase in sample size, reduce the error in exit poll, and secondly, won't it be better to take samples from all constituencies?
PAGE_BREAK
Prof Rajeeva Karandikar: Since we have to keep it within a fixed cost we try to do the best, and we feel sampling about a third of constituencies is enough.
Ravi: Starting with 2004 General Elections, exit polls have been proved wrong most of the time in India. (No exit poll predicted clear majority for Mayavathi, no exit poll predicted such a big victory for Gujarat etc). Why? Are Indian voters difficult to predict?
Prof Rajeeva Karandikar: You have selective memory. We had predicted Bihar, West Bengal and Tamil Nadu very accurately. Also, Assam and Kerala we got it essentially right — all this in last two and a half years since we (Yogendra Yadav and I) started working with CNN-IBN.
M N Sane: How do you get data for exit poll? Do you question individuals (a very large number, of course)? If so, how come no one ever in several years wanted to know my friends or my view?
Prof Rajeeva Karandikar: We select some polling booths and our people stand at the polling booth and interview people as they "exit" the polling booth. The proportion of people who are interviewed is very tiny, less than 1 in 5000 usually.
Santosh: Everyone was surprised by the analyst report since most of the Karnataka people wanted BJP in power and in fact your analyst report helped Congress to gain 16 seats else it would have been to BJP. I don’t think the analyst had taken any vote from rural area or from every city in Karnataka.
Prof Rajeeva Karandikar: You are giving too much importance to our poll. I don’t think in today's time, there is this kind of feedback effect. And we did our poll with representation from all sections of society.
R Bhaargav: How important is sample size for the accuracy of exit poll prediction?
Prof Rajeeva Karandikar: Sample size and proper sampling design (choice of polling booths and respondents) are important factors. Higher the sample size, better it is.
Raja Ram Mohan Roy: Why was the error in exit poll? Is it due to the voter changing his decision according to circumstances or is it due to the redrawn constituencies ?
Prof Rajeeva Karandikar: Redrawn constituencies was a source of error as we did not have last election data. Sampling error was the other reason.
Sungeo: I have been given to understand that there is a lot of mathematical modelling and statistical analysis that goes on for exit poll data interpretations. Do you use any software package for the same? If yes which one?
Prof Rajeeva Karandikar: There is some mathematical modelling and statistical analysis that goes on for exit poll data analysis. I have software that I have created myself and each time I need to tweak it a little bit before usage.
Ashfaq Nishtar: To convert the vote share distributions among political parties into seats, what kind of mathematical models do you use? Actual vote percentage in every constituency might be different. Do you assume any particular distribution of votes by a political party across the constituencies?
PAGE_BREAK
Prof Rajeeva Karandikar: We use the past data to estimate the distribution of votes for a party across the state. The assumption is that the socio-economic profiles of voters have not changed drastically. Delimitation has made things more difficult for us as the last election data is also missing!
Simon: Ever since the reservation controversy started, Congress appears to keep losing elections. Has is really got any negative impact in voter base of Congress?
Prof Rajeeva Karandikar: We have not seen any evidence of reservation issue causing a negative impact. All parties toe the same line in any case on reservation issue.
Kuttappa: Don't you think the pre-poll survey and the exit polls (before completion of other phases like the one we had in Karnataka) confuse the people to a great extent and the channels which give a wrong figure play a role in getting votes to a certain party?
Prof Rajeeva Karandikar: I agree to some extent. However, with different channels giving different results, voters do not give it much importance. It would be very bad if only one agency is doing the survey and prediction.
Saakshi Bhat: A survey data analysis seems like a relative simple application of statistics. Where is the mathematical challenge that fascinates an ace statistician like you? Do you do any sophisticated modelling?
Prof Rajeeva Karandikar: The challenge is in the vote to seat conversion. We have had several successes in the past. Somehow, in Gujarat and Karnataka our errors in vote estimate and vote-to-seat conversion went in the same direction and compounded the error. Also, when I see simple methods yielding good results, it makes me happier.
GV: The major reason why I see exit polls failing in India is due to lack of clear demographics of voters. In countries like US, the political system is issue based like pro life, outsourcing, illegal immigrants etc. Subsequently voters fall into these demographics of people for and against these issues. Since Indian politics is not issue based the voter segments are not clear, hence exit polls are not representative and will fail more often than not.
Prof Rajeeva Karandikar: I do not agree. In the US, the system is presidential and it is relatively easier to predict than in a parliamentary system.
Rahul: What is the range of experimental error in exit polls? If the error is miles away from the range, like in the case of Gujarat and Karnataka, isn't it means that that there are serious flaws in the experement (methodology) itself?
Prof Rajeeva Karandikar: As I wrote in my previous answer, the error margin can be 3 to 4 percent in vote share estimate. This can translate in huge error in seat forecast. Vote to seat translation is not an easy task: we do have methodology and that has its own built in error.
Karthik Srivatsan: What has been the impact of Narendra Modi in the constituencies he campaigned? Has it benefited the BJP?
PAGE_BREAK
Prof Rajeeva Karandikar: Political analysts do believe that he has had a positive impact. But we have no evidence to that effect.
V Sriharsha: Is it not possible to have a representative sample survey instead of just limiting your survey in certain pockets where there could be a big margin of error going by the pocket's political affiliation and biases?
Prof Rajeeva Karandikar: Indeed, we do use methods that give us a representative sample. We do not just limit our survey to convenient locations. In spite of that, we can be off by 3-4 percent in vote share estimate. This is what happened in Karnataka.
SK: Why is there a significant difference between vote share and seats won?
Prof Rajeeva Karandikar: In the Indian democratic system, (as elsewhere) the candidate getting largest votes in each seat gets elected. Some with by huge margins and some by small margins. If there are two parties, each getting nearly equal seats, if one party wins several seats by large margins, and the other wins most seats by small margins, the first party will have larger vote share. In Karnataka, Congress got higher vote share but BJP got more seats!
Comments
0 comment