Why Opinion Polls Keep Getting Indian Elections Wrong
The 2004 Indian general election was one of the most dramatic upsets in democratic history. Almost every major poll predicted a comfortable victory for the NDA. The actual result was a shock Congress victory that sent markets into a circuit breaker.
In 2019, polls broadly captured the BJP’s return but significantly underestimated the margin. In several state elections in between, exit polls have been embarrassingly wrong. Bihar 2020. Tamil Nadu 2021. Multiple rounds in Uttar Pradesh.
The polling industry has invested heavily in methodology over the years. Sample sizes have grown. Stratification techniques have improved. Yet the errors persist. The question is why.
The structural problem with opinion polls
An opinion poll asks a sample of voters what they intend to do. This sounds straightforward but contains several deep problems.
The first is social desirability bias. Voters in many parts of India do not tell pollsters their true voting intention. They tell pollsters what feels socially safe to say. In areas where one party is dominant, supporters of the other party often underreport their intention. The poll captures stated preference, not actual behaviour.
The second is sampling. India’s electorate is vast, diverse, and geographically complex. A nationally representative sample is genuinely hard to construct. Getting rural representation right, capturing caste and community distribution accurately, and reaching voters in areas with low digital penetration are all significant methodological challenges. Most polls compromise somewhere.
The third is timing. Voters change their minds. A poll conducted three weeks before an election captures sentiment at that moment. Campaigns, news events, and local dynamics can shift things significantly in the final stretch. The poll is already stale by the time it is published.
The incentive problem
Beyond methodology, there is an incentive problem in Indian political polling. Polling agencies are often commissioned by media organisations with political affiliations. The client has a preferred narrative. The polling agency that consistently produces results that support that narrative gets more business. Over time, this creates subtle pressure on methodology even without explicit manipulation.
This is not unique to India. But in a media ecosystem as politically fragmented as India’s, the effect is amplified.
Why aggregated independent forecasting does better
The alternative to asking people what they will do is asking independent analysts to put a probability on each outcome. When these estimates are aggregated across a large and diverse group, the incentive problems that distort individual polls largely disappear. No single participant has enough influence to move the aggregate. Social desirability bias affects individual responses but cancels out in the aggregate. Timing is less of an issue because participants update their estimates as new information arrives.
The academic evidence on this is strong. Across multiple countries and multiple election cycles, aggregated forecast markets have outperformed traditional polls, particularly in close contests where the polls are most likely to mislead.
The mechanism is simple. Polls measure stated intention. Forecasting aggregates informed probability estimates. The second is a fundamentally better measure of expected outcomes under uncertainty.
What this means for how you consume election coverage
When you see a poll showing one party at 52 percent and another at 48 percent, the honest interpretation is that the election is essentially uncertain. A four point margin is within the margin of error of almost any realistic polling methodology in India. Treating it as a meaningful lead is a mistake that television anchors make confidently every cycle.
A better frame is to ask: what do independent analysts who are tracking their own accuracy give as the probability of each outcome? That number, if it comes from a community with a track record, is more useful than any single poll.
The Strategem360 blog on how forecast exercises differ from opinion polls goes deeper into the mechanics of why aggregated forecasting works and what it requires to do well.
Indian elections will keep surprising people. Not because they are inherently unpredictable, but because the tools most people use to predict them are not fit for purpose. The better tools exist. They are just less widely used.
