INDIA. Bihar: Almost all exit polls failed in the 2020 Bihar election with their margin of error exceeding 20 percent. Generally, an acceptable margin of error is under 5 percent. Only one organisation, The Centre for Socio-Economic and Political Research (CSEPR), a New Delhi based socio-economic and political research institute, conducted exit polls that reflected the final result, with only 1 percent error margin.
Most predicted a clear win for the Mahagatbandhan, but the alliance eventually lost out in a close contest with very close margins in some constituencies. Surveys measure vote share and then use an algorithm to convert this into seats. The conversion usually causes the erroneous projections.
Why most exit polls failed
Predicting poll outcomes in a politically volatile state like Bihar is always risky due to its complex region, caste, and culture. Many organisations conducted exit polls by just looking at the crowd coming to rallies and failed to understand whether they are real voters or not.
Analysing exit polling by looking only at only the most vocal and influential crowds skewed the sample size and led to erroneous predictions. The pollsters who cover more constituencies usually conduct phone-based surveys that tend to leave out women and marginalized voters, particularly the poor. Most pollsters use statistical models to convert vote estimates into seats on the basis of previous results.
International models of polling cannot work in India as most states have a multi-party system and the society is more plural in nature. It is hard to convert insights from the ground into predictions because of the potential for ecological fallacy.
Large and inclusive sample size reduces margin of error in exit polling
CSEPR’s sample size was huge with even distribution across sex, religion, caste, region, occupation etc. CSEPR focused more on silent voters, swinging voters, flotation voters, women, backward caste, and supporters of small parties. Understanding the performance of smaller parties and including factors (such as gaps in turnout of different groups that one comes to know only a day or two after the last voting round) ensures that exit polling results are a useful prediction of actual outcomes.