Are Polls and Surveys Objective?

Presented as objective data, polls nonetheless shape the reality they claim to measure. An analysis of the mechanisms at work.

The Illusion of Media Objectivity

Studies, surveys, projections, and polls are almost always presented as objective data by contemporary journalism. This presentation as raw, indisputable fact participates in a staging of neutrality that masks the conditions under which this data is produced. Journalists readily present themselves as great defenders of critical thinking, yet they frequently relay information without any critical distance regarding sources, protocols, and the biases inherent to any production of knowledge.

The Covid period offered a striking illustration of this phenomenon. We were told that “lives had been saved” based on projections of the death toll that would have occurred without the liberty-restricting “health” measures. These projections relied on partial data and forecasting models whose reliability was itself highly questionable. But it was virtually forbidden, under penalty of being dismissed as a “conspiracy theorist,” to question the production of this “data.” These estimates were presented as established, indisputable realities, and still are today for some. We supposedly saved lives, but this was in comparison to the imaginary of a projection based on models most often funded by private pharmaceutical actors who had an interest in selling treatments.

I am not claiming that this data was entirely imaginary and deceitful, nor that the measures restricting public freedoms were completely useless. My point is different: it is to emphasize that this data was situated, produced in a particular context, with interests at stake, and that it should have been presented as such. The critical thinking I defend here consists precisely in questioning the devices that fabricate what presents itself as information. And if one is not allowed to do this or is stigmatized for doing so, what does this signify in terms of the exercise of democracy, freedom, and pluralism?

Bourdieu and the Critique of Opinion Polls

Pierre Bourdieu laid the foundations for a radical critique of opinion polls. In his article “Public Opinion Does Not Exist” (1973), he identifies three implicit postulates of any poll:

• that everyone can have an opinion,
• that all opinions are equal,
• and that asking the same question to everyone is meaningful.

Yet these postulates are highly contestable. Asking a question to someone who has never asked it of themselves means imposing a problematic that is not their own.

Bourdieu speaks of “imposition of problematic”: the poll artificially creates opinions where there were none, or forces binary responses on complex subjects. This critique remains urgently relevant. Political polls, in particular, “measure” voting intentions regarding candidates sometimes unknown to the public, or adherence to proposals whose concrete implications respondents are unaware of.

The non-consideration of “no response” for example constitutes another major bias. When a poll does not offer the possibility of answering “I don’t know” or “I prefer not to answer,” it compels those surveyed to take a position, thus essentializing positions that do not truly exist. As Bourdieu writes, “the most important question that polls leave in the shadows is that of knowing what question those who respond are actually asking themselves” (Questions of Sociology, 1984).

Anatomy of a “Bad Poll”: Mayer and Tiberj’s Analysis

The article by Nonna Mayer and Vincent Tiberj, “Investigating Antisemitism: Autopsy of a Bad Poll” (AOC, November 2025), offers an exemplary demonstration of these mechanisms. The two researchers, whose scientific integrity is recognized (Mayer is emeritus research director at CNRS and a member of the CNCDH, Tiberj is a professor at Sciences Po Bordeaux), methodically dissect the biases of a survey commissioned by the Ministry of Higher Education on antisemitism at universities.

• First problem: the constitution of the sample. The choice of hierarchical channels to distribute the survey (a letter asking university presidents to encourage their staff to participate) casts doubt on the independence of the approach. This mode of distribution risks biasing responses through a “social desirability effect,” with certain people tending to give responses conforming to what they think the commissioner expects. The sample constituted on a voluntary basis also poses problems: representativeness is not measured by the number of responses but by the quality of the sampling. This is the lesson of the controversy between the Literary Digest and George Gallup during the 1936 American election. The magazine had collected 2.4 million responses and predicted Landon’s victory; Gallup, with only a few tens of thousands of people surveyed according to a sampling plan constructed to reflect the structure of the electorate, had correctly announced Roosevelt’s reelection. The Literary Digest’s base, consisting of telephone subscribers and car owners, overrepresented wealthy and Republican voters. A large biased sample gives a precise estimate... of the wrong value.
• Second problem: the formulation of questions. Mayer and Tiberj note a problematic “mixing of registers.” Where a standard poll asks about opinions (“agree/disagree”), several questions in this survey ask whether certain propositions concerning Jews in France are “completely true, rather true, rather false, or completely false.” This shift from the register of opinion to the register of truth is dangerous: suggesting that stereotypes can be “true” confers upon them a legitimacy they do not have.
• Third problem: the impossibility of not responding. “The possibility of saying ’I don’t know’ or ’I don’t want to answer’ is rarely offered,” note the authors. One most often must respond to move to the next question, even if one has no opinion or if the proposed response options are unsuitable. This constraint essentializes positions that do not exist among respondents.

The Halo Effect and Framing Biases

Mayer and Tiberj’s analysis also highlights more subtle framing biases. Certain questions mix elements of very different natures, creating a “halo effect” where the response to one question contaminates the following ones. Placing “displaying one’s support for the Palestinian cause” and “displaying one’s support for Hamas” on the same level risks suggesting an equivalence between these two positions, with the second contaminating the perception of the first.

The authors also emphasize that the questionnaire is strongly colored by a particular thesis, that of Pierre-André Taguieff’s “new Judeophobia,” even though their own work shows that “anti-Zionism and antisemitism are two distinct dimensions that only partially overlap” and that “the old antisemitism remains predominant”. This prior ideological prism orients the questions and, consequently, the responses.

A particularly poorly constructed question asks respondents to choose between two propositions about Israel without offering an alternative or possibility of refusal. This type of binary question, on a subject as complex as Middle Eastern geopolitics, can only produce a caricature of respondents’ thinking. Many do not have in-depth knowledge of international relations in this region, and forcing them to choose between two simplistic statements tells us nothing about their actual opinions.

Media Coverage: The Absence of Critical Distance

It would be interesting to examine how this type of poll is picked up by the media. Mayer and Tiberj anticipate that “these figures will be scrutinized closely, blown out of proportion by the media, while public debate has shifted significantly to the right”. Experience shows that poll results are generally relayed without the slightest contextualization of the conditions of their production.
Beyond the questions and results themselves, there is what the authors call “a politics of polls”: only certain figures will be highlighted, not others. The poll then becomes an instrument for legitimizing a pre-existing discourse rather than a tool for knowledge. Whatever its results, this survey “risks being mobilized to criticize the university,” note the researchers, even though the university is likely less affected by prejudices than other social spaces.

This political instrumentalization of polls is not new, but it seems to have intensified. The media, caught in a logic of continuous information flow, often have neither the time nor the skills to question methodologies. The figure carries authority in itself, regardless of its scientific value. Let us recall the blatant lie brandished as an “ethical” standard by major media during the Covid period: “Everything can be discussed except the figures.”. It’s enough to make you weep... Some people had fun, instead of adding up the daily epidemic deaths in absolute value, mentioning the percentage of deaths relative to the general population (180 deaths per day on average in 2020, which were added up each day with a figure that thus seemed to swell day by day, whereas this represented 0.00030% of the population per day – 3 deaths per million, roughly equivalent in number of deaths to the flu during annual winter epidemics, and I’m not even questioning here the very large number of false Covid death declarations by hospital centers that benefited from them – if you died and tested positive for Covid, sick or not from Covid, you were declared dead from Covid). The number of deaths in France in absolute value over the two years 2020 and 2021 was lower than in the previous 10 years (cf. official data from the DREES). Covid killed a large number of people in France, certainly (164,000 over this period according to DREES, without specifying whether with or because of Covid), but no more in number than other diseases in previous years. On these interpretations of figures, statisticians all equally competent, but with different biases, can assert opposite conclusions! It all depends on which figures are compared with which other figures, at what scales, from which studies, using which statistical methods, etc.

Deleuze: Information as Order-Word

To deepen this analysis, I propose to invoke Gilles Deleuze and his reflection on the nature of information. In his filmed lecture “What Is the Creative Act?” (1987), Deleuze states that “information is the system of control”. Information is not a simple transmission of neutral content: it is an “order-word”, meaning it prescribes what must be believed, what must be thought, what must be done.

This Deleuzian conception singularly illuminates the function of polls. When a media outlet announces that “67% of French people think that...,” it is not merely describing a state of opinion: it prescribes a norm, it indicates what one is supposed to think, it creates conformist pressure. The poll, in this sense, is indeed an order-word: it says what must be thought while claiming to say what people think.

Deleuze distinguishes information from “counter-information”. If information is a system of control, counter-information consists in questioning the devices that produce information, in revealing their conditions of possibility, their biases, their interests. This is precisely what Mayer and Tiberj do in their article, and it is what I am endeavoring to do here: to produce counter-information by deconstructing the mechanisms by which the poll presents itself as objective data.

The poll as order-word participates in what Deleuze calls, with Guattari, “societies of control” (Negotiations, 1990). In these societies, control is no longer exercised through disciplinary confinement but through the continuous modulation of behaviors and opinions. The poll plays a central role: it modulates opinion by measuring it, it shapes it by claiming to reflect it.

For Genuine Critical Thinking

What conclusions can we draw? Not that we should abandon surveys and polls – they remain valuable tools for social knowledge – but that we must practice and receive them with awareness of their limits and biases. As Mayer and Tiberj write, this data “tells us as much about the responses as about its own biases and the reasons why people respond in one way or another”.

The sociologist must maintain critical distance regarding protocols themselves. This requirement applies to poll producers, but also to those who disseminate them and those who receive them. Authentic critical thinking does not consist in rejecting all quantified data, but in systematically questioning the conditions under which this data is produced: who commissioned the survey? How was the sample constituted? How were the questions formulated? What response options were offered?

This is what I call critical thinking: questioning the devices that fabricate what presents itself as information. Counter to journalism that merely relays figures as self-evident truths, the aim is to reveal that all data is situated, partial, constructed – not to disqualify it, but to restore it to its proper place: that of one point of view among others on reality, and not reality itself.