Event ReportsPublished on Aug 10, 2018
Data quality, not sample-size, matters in opinion polls, says expert

Initiating a discussion on ‘Power and Limitations of Opinion Polls’ at Observer Research Foundation, Chennai, on 21 July, Dr Rajeeva L Karandikar, Director, Chennai Mathematical Institute and a psephologist, said that opinion polls, including exit polls, can give accurate results provided the randomness and unbiased nature of the data can be ensured.  The size of the ‘sample’, or the number of people surveyed did not matter as long as the quality of the ‘sample’ and of the data could be ensured, he said.

Dr Karandikar explained how mathematical and statistical theories like probability coupled with scientific analysis of past electoral behaviour and specific conditions. He however added that opinion poll results were valid ‘as of that date’, explaining why and how the final election results end up deviating from the psephologist’s predictions in some /many cases.

Predicting the election results is both an art and science, Dr Karadikar said that random sample data will give the best results when applied over the pre-defined mathematical model, adding that dynamic factors and variables identified over a period of time also mattered. Ensuring the randomness of the ‘sample’, or the people chosen for ‘opinion polls’ and also that their responses were as genuine as they can come, the questions are ‘open-ended’, so as to derive their overall views over an impending election, issues and stake-holders.

Cognitive trap

‘Psychological anchoring’ of responses through leading questions could prove unhelpful, the speaker said. Likewise, the randomness of sample-selection too is not arbitrary. ‘Big data’ is a ‘cognitive trap’ since much of as through tele-polls and e-mail polling that help the process is invariably restricted to the urban voter, keeping the ‘voting majority’ out. Though by default and not design, such a process defeats the purpose of ‘random-ness’, and can give misleading predictions.

Votes-to-seats conversion

With genuine data and domain knowledge over the history of voter-behaviour in particular packets, mathematical models can help the psephologist arrive at his conclusions. However, the practical problem then arises in the conversion of ‘votes to seats’. It is here greater domain-knowledge like grassroots-level cooperation between alliance partners, leadership relations at different levels and their personal/political acceptance-level between two elections and anti-incumbency, helps.

Victory and defeat is at times decided by ‘swing voters’, whose behaviour cannot be taken for granted.  Post facto, one such error while was notice in the 2016 US presidential polls. More the number of swing voters, newer parties and personalities, there is room for greater complexity, Dr Karandikar said

Pre-election opinion polls are less successful when compared to post –election surveys. On the other hand, pre-election poll results will have a feed-back effect and it has the ability to swing votes. At the same time, too many opinion poll results contradicting one another, their collective impact in influencing swing voters could become negligible on occasions.

This report was prepared by S Sivanesan, Associate, Observer Research Foundation, Chennai

The views expressed above belong to the author(s). ORF research and analyses now available on Telegram! Click here to access our curated content — blogs, longforms and interviews.