Every election season produces a flood of numbers: "55% back this alliance", "leader X is preferred by 41%". Some come from careful surveys, some from online polls that anyone can vote in, and some from nowhere in particular. This article explains what separates them, and what you can safely conclude from each. It also explains the limits of the polls we run on this site.

Question 1: who was asked?

The most important fact about a poll is how its respondents were chosen.

  • In a random-sample survey, every person in the target group (say, adult voters in a state) has a known chance of being picked, for example through random selection of households or phone numbers. The pollster then adjusts the results so that groups such as age, gender, region, religion and community are represented in the right proportions. This is called weighting.
  • In a quota sample, interviewers are told to find a certain number of people of each type. It is cheaper but less rigorous, because the interviewer decides whom to approach.
  • In a self-selected (open) poll, anyone who sees it can answer. Examples include a poll on a website, a social-media vote or a phone-in. The people who respond are those who happened to see it and cared enough to click.

The first type can be used to estimate what a population thinks. The third cannot, however many people vote, because the group that answers can differ greatly from the group you care about. A famous US example is the Literary Digest poll of 1936, which collected over two million responses and still wrongly predicted the presidential result, because its mailing lists over-represented wealthier households.

Question 2: how many were asked?

For a genuine random sample, accuracy depends on the size of the sample, not on the size of the population. A rough guide for the margin of error at the usual 95% confidence level is about 98 divided by the square root of the sample size, in percentage points. So 400 respondents gives roughly ±5 points, 1,000 gives about ±3 points and 2,500 about ±2 points. A reported lead of 3 points in a poll of 1,000 people is therefore well within the margin of error; it does not tell you who is ahead.

Two cautions apply. First, the margin of error applies to the whole sample; for sub-groups, such as "women in Madurai district", the sample is much smaller and the margin much wider. Second, the margin of error only covers random sampling variation. It says nothing about bias from who was reached or who refused to answer.

Question 3: how was the question asked?

Small changes in wording can move answers considerably. A question that mentions a leader's name tends to get more support than one that does not. A question that offers only "support" or "oppose" gives a different picture from one that adds "no opinion". The order in which options are listed also matters. A trustworthy poll publishes the exact question and the options.

Question 4: when, and by whom?

Opinion moves. A poll taken before a major event can be out of date a week later. Check the dates of the fieldwork, and who paid for the poll. A survey commissioned by a party or campaign is not necessarily wrong, but you should look for the full methodology before trusting the headline.

Seats are not votes

Even an accurate vote-share poll does not directly tell you how many seats a party will win. Under first past the post, seats depend on how votes are distributed across 234 constituencies. Forecasting seats requires a separate model, and it is common for seat projections to miss even when the vote-share estimate was close.

What our polls are, and are not

The polls on SuperstarElection.com are open online polls. Anyone who finds a poll can vote in it, and we record one vote per poll from each network address to discourage casual repeat voting. That is a deterrent, not a guarantee. Because the people who respond are self-selected, our results do not represent all voters in Tamil Nadu, and they should not be used to predict an election. What they do show, honestly, is how the people who chose to take part voted on a given date: useful for gauging the strength of feeling among an engaged audience, and for seeing how that audience's views change over time.

When you quote one of our results, please describe it accurately: "of the 1,200 people who voted in an open online poll on SuperstarElection.com". We publish live totals and we do not edit counts. Each poll shows its end date, after which voting stops and the result remains as a permanent record.

A short checklist

  1. Who was surveyed, and how were they chosen?
  2. How many people responded, and what is the margin of error?
  3. What exactly was asked?
  4. When was the fieldwork done, and who paid for it?
  5. Does the poll claim more than its method can support?

Sources and further reading

Last reviewed: 18 November 2026. Laws and procedures can change; check the official sources below before relying on a detail.