Pollsters continue missing the mark: Why election forecasts go wrong

In India, the experience with Lok Sabha electoral poll forecasts has been frustrating in recent years. In 2004, almost all pollsters predicted a victory for the Bharatiya Janata Party-led National Democratic Alliance (NDA), but the Congress-led United Progressive Alliance (UPA) won instead. In 2009, while predictions pointed to a clear UPA victory, the actual outcome was a hung Parliament. Most polls failed to foresee the decisive BJP victory in 2014.

 

In 2019, while most opinion polls suggested the NDA had run out of electoral steam and would fall short of the majority mark, the exit polls predicted the NDA crossing the 300-mark, and partially redeemed the pollsters’ honour and credibility. But even this redemption proved to be temporary. In 2024, most polls projected a sweeping NDA victory, estimating 362-411 seats, but the alliance secured only 293.

 

Even in recent Vidhan Sabha elections, pollsters have fared poorly. Take the last four in Delhi in 2013, 2015, 2020 and 2025.  In 2013, although many of them foresaw the debut of the startup Aam Aadmi Party (AAP), they underestimated the extent of its victory (28 out of 70 seats).  In 2015 and 2020, while they predicted an AAP victory, most failed to detect its sweeping nature—67 out of 70 seats in 2015 and 62 in 2020.

 

The mistake was repeated in 2025, but in reverse—they hugely underestimated both AAP’s defeat (down from 62 in 2020 to 22) and BJP’s victory (up from 8 in 2020 to 48).

 

The story is similar in Vidhan Sabha elections in Bihar (2015), Uttar Pradesh (2017), Chhattisgarh (2023), and Haryana (2024). In 2015 in Bihar, a majority of pollsters failed to predict a substantive victory for the Mahagatabandhan against the NDA.  In 2017, in Uttar Pradesh, pollsters failed to gauge the massive resurgence of the BJP.  In Chhattisgarh in 2023 and  Haryana in 2024, pollsters failed to predict the BJP’s victory.  

 

Surprisingly, even many exit polls have been off the mark in recent years. Opinion polls conducted before balloting suffer from the problem of surveying people who may not vote on polling day. Exit polls, on the other hand, survey voters as they exit polling booths after casting their votes and do not have this limitation. Opinion polls are also prone to late swings—prospective voters changing their minds between the poll date and election day. Exit polls are not affected by such late swings. Thus, exit polls might be expected to be more robust than opinion polls in forecasting election outcomes.

 

Polls are probabilistic and have gone wrong in the past not only in India but also in other countries, including the US.  Even George Horace Gallup, the inventor of the Gallup poll, who in 1936 started predicting the US presidential election on the basis of survey-sampled measure of public opinion, famously went wrong in 1948.   In that year’s US Presidential election, Gallup, along with two other major polls, wrongly predicted Republican Thomas Dewey would defeat Democrat Harry Truman.  Polls have performed poorly even in predicting the last few US Presidential elections. 

 

When polls go wrong with unusual regularity, pollsters come under attack, facing charges of incompetence, faulty sampling and methodology, or even deliberate fraud.  But the factors that can motivate pollsters to come out with fraud poll predictions appear unconvincing.  Undoubtedly, a favourable poll prediction boosts the morale of party workers and energises the electoral campaign.  A “bandwagon effect” may lead people to cast their votes in favour of the possible winner.  No one wants to “waste” her vote by casting it for a loser.  But leaving aside fly-by-night operators, for any pollster with ambition to stay and grow in the business of polling, the adverse reputational impact of a wrong prediction, particularly close to the polling date, should act as a strong deterrent against fraudulent prediction.

 

Finding a rational motive for a fraudulent exit poll seems even more difficult.  Any mistake in an exit poll prediction is likely to be exposed within 72 hours when the votes are counted. The only plausible rationale is stock market manipulation. For instance, if a market-friendly party is expected to win, falsely predicting a victory for a less market-friendly party could depress stock prices, allowing one to buy stocks cheaply before the actual results are out.  As markets boom with the results coming out, you can make a fortune by selling these stocks.   But this is a mischievous game that you can play only once, the danger of being caught is considerable and you will not be able to repeat the game more than once or perhaps twice.

 

In recent years, proposed solutions to prevent opinion and exit polls from making inaccurate predictions have ranged from outright bans to regulation.  While the possibility of some “suspicious” polls, say from astrologers, cannot be ruled out, we must respect the constitutional right to freedom of speech and have faith in the superior judgement of the consumers of polls in separating the wheat from the chaff.

 

Appropriate light-touch regulation, including full disclosure of ownership, address, experience and track record, may be considered.  But beyond that, let us wait for pollsters to improve their methodology, minimise errors in estimating vote shares of different parties, and then accurately convert votes into seats for different political parties.  Psephology, the science and art of election forecasting as it is popularly known in India, is still a relatively young discipline. With experts in political science, statistics, and sampling methodology contributing, it will evolve and mature over time.  

 

The author is an economist and a member of the West Bengal Legislative Assembly