19 Jan
As Bangladesh enters the final three to four weeks before its national election, polls are no longer a background feature of political conversation, they are already shaping it. Survey results are being published by research agencies, discussed on television talk shows, and widely circulated on social media. Increasingly, these numbers are treated not as snapshots of public opinion, but as indicators of momentum, inevitability, and likely outcomes. For many citizens, polls are quietly beginning to answer questions they were never designed to settle: who is winning, by how much, and whether an individual vote still matters.
This moment is not unique to Bangladesh. It reflects a broader global pattern, one in which election polls are highly visible, heavily consumed, and often deeply misunderstood.
Over the past decade, polling has struggled across a wide range of democracies. In the United States, Donald Trump outperformed pre-election polls in three consecutive presidential elections in 2016, 2020, and again in 2024. While some of these errors technically fell within published margins, the pattern was consistent: Trump’s support was systematically underestimated, particularly among certain demographic and geographic groups. Post-election reviews by professional polling bodies linked these misses to declining response rates, faulty turnout assumptions, and the reluctance of some voters to disclose their true preferences (AAPOR, 2017; Pew Research Center, 2018; BBC, 2024).
In India’s 2024 general election, the failure was even more visible. Almost all major exit polls predicted a sweeping landslide for the ruling coalition. When results revealed a far narrower outcome, markets reacted sharply and public confidence in polling took a hit. Pollsters later acknowledged that they failed to capture voter discontent and late swings in key states such as Uttar Pradesh, Maharashtra, and West Bengal, particularly among economically and socially marginalised groups (Reuters, 2024; Frontline/Bloomberg, 2024).
The United Kingdom offers another cautionary tale. Polls failed to anticipate the Brexit referendum result in 2016 and misjudged margins in several subsequent general elections. Official inquiries pointed to persistent weaknesses in turnout modelling, sampling frames, and so-called “shy voter” effects, where respondents conceal preferences they believe may be socially judged (British Polling Council, 2016).
Argentina’s 2019 presidential election was perhaps the most dramatic example. No major pollster predicted the decisive 16-point victory of Alberto Fernández over the incumbent. Financial markets, which had priced in a much closer contest based on polling consensus, suffered a historic overnight collapse once results were announced. Later academic analysis of raw and longitudinal survey data revealed that the problem was not small errors, but missing voters, demographic misrepresentation, and widespread concealment of true voting intentions before the election (Makse et al., 2020).
These cases differ in political culture, institutions, and polling methods. What unites them is not incompetence, but a growing mismatch between how polls are designed and how voters behave today.
Polls often appear more precise than they truly are. Terms like “margin of error” and “confidence level” sound reassuring, but large comparative studies show that election polls routinely overstate their certainty. One comprehensive analysis of thousands of U.S. election polls found that surveys claiming a 95 percent confidence level actually captured the true result only about 60 percent of the time (Kotak & Moore, 2022). Put simply, polling uncertainty is far larger than it looks, especially in the final weeks of an election.
This gap exists because polling error is not only about sample size. It is also about who refuses to respond, who is under-represented, and who chooses not to tell the truth.
Across countries, response rates to political surveys have collapsed. In the United States, only about five to six percent of contacted individuals now respond to telephone polls, with similar challenges in online and automated surveys (Pew Research Center, 2018). Those who decline are not random. They differ systematically in trust, political alienation, and willingness to engage. Entire voter blocs may never enter the dataset at all, and no amount of statistical re-weighting can fully correct for opinions that were never observed.
Face-to-face interviews, often used in countries like Bangladesh, reduce some access barriers but introduce others. In politically charged environments, respondents may fear disclosure, social pressure, or local repercussions. This can lead to polite but inaccurate answers, or silence altogether.
Turnout modelling presents another challenge. Polls typically weight samples to census demographics, but elections are decided not by the census population, but by those who actually vote. Predicting who will turn out by age, income, gender, or region has become increasingly difficult as mobilisation becomes uneven and late-breaking. Even in countries with compulsory voting, such as Argentina, polling failures have occurred because key groups, particularly younger voters, were poorly captured in raw samples despite high turnout (Makse et al., 2020).
Then there is social-desirability bias. In polarised elections, some voters conceal their true preferences, especially when supporting candidates perceived as controversial or socially risky. This is not a theoretical concern. In Argentina’s 2019 election, longitudinal surveys tracking the same respondents before and after voting showed that nearly one in five voters did not disclose their true preference before the election, with concealment concentrated among supporters of the eventual winner (Makse et al., 2020). Similar dynamics have been documented in the United States and the United Kingdom under the labels “shy Trump” and “shy Tory” effects (AAPOR, 2017; British Polling Council, 2016).
These distortions matter because even small errors can have large consequences. In first-past-the-post systems, a two- or three-point polling miss across dozens of competitive constituencies can transform a tight race into a perceived landslide. Polls may “get the winner right” while still misleading voters, campaigns, and markets about margins, momentum, and inevitability.
Despite these failures, abandoning polls altogether would be a mistake. Polls remain valuable because they reveal what voters care about, how issue priorities shift over time, and where dissatisfaction or enthusiasm is building. They inform journalism, policy debate, and campaign strategy. The problem lies not in polling itself, but in how polls are interpreted, communicated, and used.
One overlooked consequence is behavioural. Research shows that voters sometimes decide not to vote because they believe polls already reveal the outcome (Kotak & Moore, 2022). In this sense, polls do not merely observe democracy, they can shape it.
This brings the discussion directly to Bangladesh.
At present, Bangladesh has no clear, codified framework governing the conduct, timing, and publication of election surveys. Yet pre-election opinion polls are already being conducted and disseminated by research agencies and media houses. As election day approaches, it is also likely that exit polls will emerge. Exit polls differ from opinion polls in a critical way: they are conducted immediately after voters cast their ballots and ask whom they actually voted for. Because they claim to measure real behaviour rather than intention, they are far more sensitive and far more capable of influencing post-poll narratives and voter confidence.
Many democracies recognise this risk. In India, exit polls are regulated under Section 126A of the Representation of the People Act, 1951, which prohibits their publication until all phases of voting are complete. Violations carry legal penalties. Across the European Union, most countries impose blackout periods on opinion or exit polls immediately before voting day, ranging from 24 hours to several days. These rules are not about suppressing research, but about preventing undue influence on voters and safeguarding trust in the electoral process.
Bangladesh currently lacks comparable guidelines. As polling activity grows, this absence is no longer neutral. It creates space for confusion, selective amplification, and unintended harm.
The Election Commission of Bangladesh should therefore consider issuing clear, time-bound guidelines that distinguish opinion polls from exit polls, regulate when exit polls may be conducted and published, require transparent disclosure of methodology and uncertainty, and set standards for responsible media reporting of survey results. Such guidance would not restrict legitimate research or free expression. It would protect voters, strengthen confidence, and ensure that polls remain what they are meant to be: inputs to democratic debate, not substitutes for the ballot box.
Polls are tools, not oracles. They do not cast votes. They do not count ballots. And they should never be mistaken for advance verdicts on democratic choice. The lesson from recent elections, globally and increasingly relevant for Bangladesh, is not that polls are useless, but that they must be read with restraint, context, and a healthy pinch of salt.
References
AAPOR. (2017). An evaluation of 2016 election polls in the United States.
https://www.aapor.org/Education-Resources/Reports/2016-Election-Polls.aspx
BBC News. (2024). Why the polls underestimated Trump again.
https://www.bbc.com/news
British Polling Council. (2016). Inquiry into the 2015 general election polling miss.
https://www.britishpollingcouncil.org/past-inquiries/
Frontline / Bloomberg. (2024). How India’s exit polls got the 2024 Lok Sabha election wrong.
https://frontline.thehindu.com
Kotak, J., & Moore, J. (2022). Election polls are 95% confident but only 60% accurate.
https://www.kotak.com/content/dam/Kotak/article-images/election-polls-are-95-confident-but-only-60-…
Makse, H. A., et al. (2020). Why traditional polls fail: Evidence from Argentina’s 2019 election. Nature Human Behaviour.
https://www.nature.com/articles/s41562-020-00964-1
Pew Research Center. (2018). Assessing the representativeness of public opinion surveys.
https://www.pewresearch.org/methods/2018/01/26/assessing-the-representativeness-of-public-opinion-s…
Reuters. (2024, June 5). India exit polls failed to capture voter discontent in key states.
https://www.reuters.com/world/india/india-exit-polls-failed-capture-voter-discontent-key-states-202…
Election Commission of India. (2024). Guidelines on exit polls and media reporting.
https://eci.gov.in
European Parliament. (2020). Regulation of opinion polls in EU member states.
https://www.europarl.europa.eu
