Why our barometer is not an opinion poll

This site’s barometer shows, from the answers of people who take the test, how many have responded, their average position on the political compass in each region, which candidate most often comes out closest to their answers and the issues on which those answers are most divided. Once there are thousands of tests, it will be tempting to read it as an opinion poll. It is not one, and it is worth explaining exactly why.

What the barometer shows

The barometer has four parts, and the data are updated daily:

• Participation: the number of completed tests, how many of them in the last 30 days, and how many regions already have publishable data.
• Territory: the average position on the compass (economic axis from −100 on the left to +100 on the right; social axis from −100 progressive to +100 conservative) in each of the 13 regions of metropolitan France and the 5 overseas departments and regions, based on the approximate location of the connection. The compass does not count questions on Europe, foreign affairs and defence, or institutions.
• Closest candidate: the share of completed tests in which each candidate came out closest to the participant’s answers.
• Polarisation: for each topic, an index from 0 (everyone answers the same) to 100 (answers are split as far as possible between the two ends of the scale). It is calculated from all answers to the test’s current questions, including those from unfinished tests.

Every chart carries the same warning, in French: “Ce baromètre n’est pas un sondage au sens de la loi n° 77-808 du 19 juillet 1977…” (“This barometer is not an opinion poll within the meaning of Act No. 77-808 of 19 July 1977. It aggregates the anonymous results of the people who chose to take the test: it is not representative and does not offer the methodological and representativeness guarantees required of an election poll.”) The readings of the barometer that we publish in our analyses follow the same rules.

The difference lies in how respondents are chosen

A serious poll starts from a sampling design. The polling company decides how many people to ask, and with what spread by sex, age, occupation, region or size of municipality, so that the result can be generalised. It contacts those people and, if the final spread does not match the population, weights the answers to correct it. That work is what makes it possible to say that a given share of the French think something, with a margin of error.

French law recognises this difference. Under Article 1 of Act No. 77-808, an opinion poll (“sondage”) is, “whatever it is called”, a statistical survey intended to give a quantitative indication of the opinions, wishes, attitudes or behaviour of a population “by questioning a sample”, and the people questioned are chosen by the organisation carrying it out “so as to obtain a representative sample of the population concerned”. Since 2016 the law has applied to any poll on matters directly or indirectly linked to electoral debate, at any time and not only during the campaign.

None of that happens with the barometer. The people in these data came on their own: they searched for a political test, clicked a link on social media or were sent one by someone. Nobody was selected. There are no quotas and no weighting, and there is no margin of error to calculate, because a margin of error assumes a random draw that never took place.

This group can be expected not to resemble the electorate: it will probably be younger, more online and more interested in politics than average. The test is also open to anyone, including people who do not vote in the presidential election: only French citizens registered on an electoral roll do. That is selection bias, and more answers do not fix it. A million voluntary answers are still a million voluntary answers.

What the law says, and how we apply it

Act No. 77-808 sets precise obligations for opinion polls: their first publication must include mandatory information, such as the name of the organisation that carried it out and of the one that commissioned it, the number of people questioned, the dates of the survey, the wording of the questions and the margins of error (Article 2). The Commission des sondages, the polling watchdog, ensures they are respected.

The barometer does not meet the definition in Article 1, and the Commission says so about this kind of survey at every presidential election: in 2017 it wrote that “people who freely answer the questions asked do not constitute representative samples and the results of these surveys are not opinion polls within the meaning of the law”. But that brings another obligation. Article 12 punishes with a fine of €75,000 the use of the word “sondage” for surveys linked to electoral debate that do not meet the definition in Article 1. In 2022 the Commission specified that explaining where the respondents come from is not enough: the word “sondage” must not appear, and the data must not be presented as a poll’s estimates would be.

Hence our rules, as we read the law and the Commission’s practice:

• we never call the barometer an opinion poll, or its figures estimates of the election result;
• every chart carries the warning quoted above, which follows the one the Commission has had published for non-representative surveys;
• the barometer publishes no data on the candidate participants say they would vote for, and does not compare itself with any poll;
• the share of tests in which each candidate comes out closest is always accompanied by a reminder that it says nothing about how people will vote.

So what is it good for?

For describing that group, which is no small thing. If most participants agree with a statement, or if an issue splits their answers sharply, that says something real about the people who take the test. What it does not allow is a leap to “the French think”.

It is also useful for comparison, with care. Differences between regions within the same self-selected group are informative, even though the absolute level cannot be generalised and the bias need not be the same everywhere.

Nor is the distribution of the closest candidate a vote. It says which candidate the test answers were closest to, not whom those people will vote for. Anyone may vote for a candidate other than the one closest to their answers, and that is their right.

How we protect participants

The test does not ask for a name or an email address, and it does not require registration. Each result is stored with a random identifier. It optionally asks for your age group, your sex and the candidate you would vote for: none of these appears in the link for sharing your result, and the vote you declare is never published, not even aggregated with other people’s answers; it only serves to show you, and you alone, where you agree with that choice. Your region is not asked: it is estimated roughly from your connection, using a geolocation database installed on our own server.

Aggregates are subject to a minimum threshold: no cell is published unless it contains at least 100 tests. That is why, while there are few tests, the barometer will be almost empty, and a region that does not reach the minimum will not appear in the table. It is not that the data are missing: with few cases, an aggregate starts to say too much about specific people.

We never publish individual data or combine aggregates in a way that would allow anyone’s profile to be reconstructed. The details are in the privacy policy.

What about the election weekend?

Article 11 of Act No. 77-808 prohibits publishing, disseminating or commenting on any election poll “by any means whatsoever” on the eve and the day of each vote. For the presidential election, the ban applies throughout the national territory from 00:00 on the Saturday before the vote and ends when the last polling station in metropolitan France closes, at 8 pm on Sunday. Saturday voting in the French territories in the Americas, in French Polynesia and at consulates in the Americas does not bring the ban forward. At the same time, all electoral campaigning is banned, online included (Electoral Code, Article L49).

The barometer is not an opinion poll, but some of its figures could be read as one. That is why the whole barometer is switched off on the weekend of each round, and for a little longer than the law requires: from 00:00 on Saturday 17 April 2027 to midnight on Sunday 18 April, and from 00:00 on Saturday 1 May to midnight on Sunday 2 May, Paris time. The other places where we publish aggregated data from answers (the home page and the analyses that use them) are switched off at the same times, and we do not publish or comment on their figures anywhere else, social media included. Nothing is deleted; the data return after the vote.

The test itself stays open. Your individual result is neither an opinion poll nor an estimate: it only shows you how far you are from each candidate.

How to cite it

If you use these data, the correct way to describe them is as aggregates from a self-selected sample of users of a political affinity test, stating the period, the number of tests and the warning that the sample is not representative. The barometer’s methodology note includes all of these elements.

We ask that they not be presented as an opinion poll or as an estimate of the election result, because they are neither; Article 12 of the French act also penalises using the word “sondage” for this kind of survey. And on the weekend of each round, please do not publish them at all.