How can single units be made out of a multiple? How can this unity be decomposed to make a new diversity? And to what end?
These three different but inseparable questions recur throughout the slow development of the statistical tools used in objectifying the social world.
1Although the American and European social sciences have developed several international surveys on values and political opinions, as well as a range of highly sophisticated methodologies and statistical models for analyzing them, they are rarely used in mainstream French sociology or political science. Economists to the contrary, though still often mocked for the narrowness of their theory of the rational actor, capture these data on a large scale to work on questions such as happiness, values, or even identity, encroaching all the more easily on the territory of sociologists, because the latter leave it open to them in their marked disdain for these data. Opinion polls are not as such always more popular among the “quantitativists” than the “qualitativists” in so far as they seem to present a subjective, and therefore diminished or unattainable, reality that has no place in a scientific understanding of the world.
2I propose firstly to identify four specific approaches underlying the criticism of opinion data. The first is rooted in a form of objectivism that we could call “statistical materialism,” which refuses to accept opinions or subjective perceptions as “objects” truly worthy of scientific investigation or measurement in the same way as “behaviors”—such as voting or economic activities—would be. Conversely, a subjectivist approach might consider the production of figures using closed-ended questions to be reductive and thus incapable of describing the richness of human subjectivity. A nominalist approach to the criticism of opinion data argues that these polls concern or construct objects that do not exist, such as public opinion. Lastly, this criticism becomes political when it identifies polls as opinion-makers that are generally supportive of neoliberalism (Champagne 1990). This last form tends to morph into a battle between professions, given that it often involves asserting the superiority of academic political scientists [politistes] or sociologists over political analysts [politologues] working in institutes of public opinion, who do not share the same political conceptions and who are guilty of the “illegal misuse of science.” These different criticisms—objectivist, subjectivist, nominalist, and political—can be found in Pierre Bourdieu’s renowned lecture “Public Opinion Does Not Exist.” We can also find the foundational elements in Distinction and his Pascalian Meditations. They are taken up again in works that have become classics in the criticism of polls, such as that of Patrick Champagne (1990) or Daniel Gaxie’s more recent criticism of the Eurobarometers (2011).
3We are not concerned here with political criticism, which is not really based on methodological reflection, but rather on a social study of the media that evaluates whether opinion polls actually “shape opinions.” What interests us here is therefore not the political criticism of pollsters or even the misuse of opinion data, but the epistemological criticism of their foundations and their scientific legitimacy.
4Bourdieu’s criticism of opinion data is primarily based on the argument around imposing a problematic (or conceptual framework) and the construction of artefacts:
The effect of the imposition of a problematic, an effect produced by all opinion polls and all political questioning (not least by elections), results from the fact that the questions asked in an opinion survey are not questions which arise spontaneously for the people questioned, and that the responses are not interpreted in terms of the problematic actually referred to in their answers by the different categories of respondents.
6Opinion polls supposedly impose a particular wording of the questions and their responses on the respondents and the resulting statistics are consequently artefactual in two respects: they aggregate responses that are not truly those of the respondents (they have been imposed on them) and they give them a different meaning from that given by the respondents. This critical approach has become a common ground in political science, bringing together the different critical approaches that I have identified: objectivism, subjectivism, and nominalism. For writers such as Daniel Gaxie and Philippe Aldrin, Bourdieu’s arguments have become weapons used to defend the use of qualitative surveys against the quantitative polls conducted by the European Commission, the Eurobarometers, which are accused both of epistemological vacuity and of political bias (Aldrin 2010; Gaxie et al. 2011).
7We are not concerned here with examining the history of polls and their use, or the difficulty of defining the concept of public opinion,  but rather with reflecting, in the light of these criticisms, upon particular arguments and many of the ideas that can be found in methodological textbooks, without subjecting them to explicit theorization, as is the case with Pierre Bourdieu’s criticism. As we shall see, the criticism of opinion data, or often more simply the resistance to them, raises fundamental questions about the challenge of describing subjective perceptions in the social sciences and of quantifying them, the issue of aggregating statistical data to construct macrosocial indicators, and, finally, a subject dear to Pierre Bourdieu, the construction of the object and, more specifically, what constitutes a scientific object in the social sciences. I will attempt therefore to identify the weakness of this criticism by addressing the three underlying approaches in turn: objectivism, subjectivism, and nominalism. This will enable me to elucidate the epistemological and methodological principles that justify the use of these data and more generally to reformulate certain classic epistemological questions in the social sciences.
The objectivist criticism: Behaviors and opinions
8Data on happiness, opinions, or identity are often rejected on the grounds that practices or behaviors would be more objective indicators. Statistics are not therefore rejected on principle, but they must concern practices considered to be “real,” such as economic activities, voting, or even cultural practices. Through this approach, statistics and the social sciences adopt a form of behaviorism, an influential approach in psychology that has made observable behavior the only object of scientific value. Scientific psychology however long ago abandoned this idea, which nevertheless regularly reappears through theoretical or philosophical references that are often more developed and more acceptable in the field of sociology than is behavioral psychology, which reduces the scientific study of humans to the relationship between stimuli and responses. The underlying intention is nevertheless the same: to dispense with these strange objects called thoughts, feelings, and intentions, which seem to have no place in a true scientific understanding of the world and much less in a quantified understanding of it.
9In order to simplify the explanation, I do not distinguish here between concepts like intention, opinion, belief, or perception; I instead group them together as the manifestation of the use of subjective, as opposed to objective, language to describe human activity. This grouping is justified as part of a particular approach: demonstrating that statistical data must take account of this way of describing individual actions and that we can in no event dispense with these descriptions, which are perceived to be subjective, in order to prioritize behavior or practice.
The subjectivity of behavior
10I will first argue that this distinction between objective data measuring behaviors like economic phenomena and subjective data concerning opinions, beliefs, intentions, or happiness must be put into perspective. Behaviors cannot in fact be measured as solely objective and substantive facts stripped of all subjective interpretation. The most objective statistics distinguish between behaviors according to the meaning that the actors give them, imbuing these objective data with a high level of subjectivity. We can demonstrate this by examining any quantitative data within the social sciences.
11Firstly, remaining within a field of political science, we can take the example of voting, which is often presented as more objective data than that obtained from opinion polls on the grounds that it is a real behavior. It seems for some to symbolize the actual choice of the individual, which would make it a much more legitimate scientific object.  Voting is however anything but a phenomenon that we can content ourselves with observing without any description of its inherent subjectivity. In this respect, studying it is not fundamentally different from studying the responses to an opinion poll.
12We cannot in fact easily equate two like votes with two like actions. People could vote for the same candidate for extremely different reasons, in support of the candidate or rather to block another candidate, or even because they are attracted by different proposals made by this candidate. If we consider a vote like the 2005 French European Constitution referendum, we cannot understand the “no” victory by considering only the share of people who voted “yes” or “no” without considering the different groups of voters, with doubtlessly divergent intentions, that formed a majority. To the contrary, we must attempt to reconstruct the reasons for this vote and so distinguish, for example, among “no” voters, those in favor of the European project and another treaty, from those that more directly oppose the whole idea of European federalism. Nevertheless, in order to understand the distribution of the different reasons explaining this vote, individuals must be asked about their political preferences, which involves the use of opinion polls. Although statistical voting data have the advantage of appearing more objective, they undoubtedly have the shortcoming of presenting only a summary and as such of preventing this vote from being contextualized in a broader system of preferences and political beliefs. Rejecting any data on the reasons behind the votes and the voters’ preferences or political opinions would amount to the large-scale amputation of the social reality and would certainly result in incorrect predictions about future votes.
13It is simply illusory to think that voting data could entirely replace opinion data. Yet more fundamentally, and more philosophically, it must be said that characterizing this behavior as a vote is already to describe it in relation to its subjectivity. We assume that by casting their ballots, individuals have expressed their intentions: their desire that the person named on their ballots will win. More simply, voting is not casting a ballot, it is rather choosing a candidate or a project and it is this subjective action that we seek to measure through the act of individuals casting their ballots. We therefore saturate the description with subjectivity and intentionality by calling it a vote.
14Voting may seem to be an example that makes it a little too easy to demonstrate that it is illusory to content ourselves with measuring behaviors. The reasoning can nevertheless be extended to any practice or behavior that could be considered a more difficult example. Cultural practices, for example, are often employed as examples of more objective quantitative data, but they too are behaviors that are not so straightforwardly objective. For example, the act of “going to the cinema” could describe different practices depending on the intention that accompanies the practice. Is it the same to go to the cinema as a movie buff who wants to see the latest film by a favorite director or as a couple seeking to fill an evening? Is it the same to go with friends or as a pretext with the hope of starting a relationship? In these different examples, we can produce very different statistics that take into account the intentions associated with these actions and that enable us to better understand the social reality by distinguishing between the different practices encompassed by the behavior of “going to the cinema.”
15Finally, we can extend this reasoning to examples that could appear to be more challenging: the economic statistics produced by public bodies. If we look, for example, at the issue of unemployment, the problem is effectively the same. The definition of an unemployed person by the Institut national de la statistique et des études économiques(INSEE) (French National Institute of Statistics and Economic Studies) clearly incorporates the idea that an unemployed person seeks employment, thus referring to an intention. We can of course attempt to define this search “objectively” (registration at a job center, submitting applications), but in doing so we characterize and group these behaviors merely as indicators of a single intention: the desire to be employed, which is in itself part of the definition of an unemployed person. An unemployed person is clearly someone who is seeking employment and not simply someone who submits applications. It is also this intention that is described by unemployment statistics, which therefore also include a subjective dimension.
16We could give endless examples: violence is not simply a physical act devoid of the subjectivity of those who perpetrate or suffer it (Crettiez 2010). The statistical measurement of a behavior therefore requires that the associated intention be characterized in order to be able to describe it, statistically, in a way that is relevant to sociologists. This cannot be done using only that which is easily visible to the naked eye. A purely behavioral statistic would soon find its object reduced to near irrelevance and would certainly not provide sociologists with adequate data to understand the range of political, cultural, and social phenomena.
The objectivity of opinions
17What applies in one direction, also applies in the other: if data on practices are less objective than they appear, opinion polls are likewise not as subjective as their critics assume. In fact, what we measure in such polls is not public opinion, values, or happiness, but the responses given to questions and we can therefore certainly consider them objectively: we tick a box in the same way as we cast a ballot, go to the cinema, or submit a job application. The difficulty then lies entirely in how to interpret these responses and what they say about the intentions of their actor, and his or her preferences or perceptions. What is most surprising in Bourdieu’s criticism is that it presents these interpretations as predetermined by the method in asserting that opinion polls impose a problematic and measure an artefact by aggregating responses with different meanings. While there is certainly no reason to assume that all the respondents assign the same meaning to responses that could appear identical (ticking the same box), there is likewise no reason to assume that all researchers will interpret these responses as identical. It is precisely because these data measure the responses to questions and not an opinion or value directly that the criticism concerning the imposition of a problematic is often confused. The responses to these questions are always understood through the reconstruction of intentions or perceptions and it is unclear on what basis it is assumed that it is the survey that imposes a problematic, as if different researchers have no authority to have different interpretations of the responses and the problematic underlying the question.
18The supposed contrast between objective practices and behaviors and subjective opinions is therefore far from clear. The most objective statistics always relate to behaviors that they identify and define by attributing intentions to them, in the same way as the person interpreting the responses to an opinion poll. Sharing the behaviorism that underlies the criticism of opinion polls, the neo-positive philosophers of the Vienna Circle tend to advocate the reduction of the discussion of psychology and sociology to a physicalist language that makes the objects described by the physical sciences the only real objects.  This is in fact the conclusion that must be reached by the materialist approach at the root of behaviorism. Sociologists seek perhaps to escape neo-positivist reduction, social matter comprised of behaviors seeming a little more “solid” than the psychological matter of thoughts. They do not however see that studying an individual without ideas, without intentions, without perceptions, and without identity is ultimately nothing more than studying a physical body displaced by the laws of the universe. As highlighted by the theories of comprehensive sociology, debates around social sciences methods in nineteenth-century Germany, or more recently the sociology of Raymond Boudon (2003), the attribution of rationality to individuals is a fundamental dimension of the social sciences, a rationality that clearly goes beyond the maximization of the homo economicus. The need to attribute rationality to individuals, in particular through the “principle of charity,” has also been advocated extensively by the philosophies of language and of mind (Davidson 2004).
19I maintain therefore that it is the use of subjective language—mobilizing concepts such as “beliefs,” “perceptions,” “intentions,” or even “feelings”—that distinguishes the study of individuals from the study of the matter from which they are made and that lays the foundations for the social sciences’ approach to research. Consequently, the quantitative approach cannot in any event overlook data on this subject and must confront the difficult issue of the quantification of processes that appear to us to be subjective. Often, when faced with this difficulty, the desire for or the belief in the greater objectivity of observing practices and behaviors arises. Asserting the primacy of practices or behaviors is in fact to become trapped in mind-body dualism, since speaking of primacy implies that there is something over which supremacy can be asserted rather than understanding that we are talking about one and same thing, as previously asserted by the pragmatist philosopher John Dewey.  There are not in fact observable behaviors and practices on one side and perceptions and intentions on the other. The description of human activity inextricably combines these two orders of phenomena: to describe an individual’s action, a practice, or a behavior is to describe an action using concepts like perceptions or intentions. Statistical descriptions must therefore allow these subjective phenomena to be presented and the use of opinion polls is, among others, one means of getting an indication of these subjective phenomena.
The subjectivist criticism
20The criticism of opinion polls, when it does not seek to do away with any intentionality or subjectivity, considers conversely that such polls cannot convey the richness of this subjectivity in the way that interviews or more qualitative data can. This same instinct is also at the root of the argument on the imposition of the problematic, which reproaches questionnaires for asking the respondents questions that they do not ask themselves and therefore for not capturing their true terms of reference. This criticism raises on the one hand the issue of “bad questions” that supposedly do not measure the real thoughts of the respondents and, on the other hand, the issue of the empiricist illusion of science as simply a strict record of reality.
Good and bad questions?
21Philippe Aldrin writes of one Eurobarometer question on further integration of European politics: 
Yet, in order to be able to formulate informed answers to such questions—i.e. being aware of the political issues involved and being able to take a stance on that basis—respondents require pre-existing knowledge of the problems raised by the institutional situations mentioned and a structured vision on the alternative positions or solutions available in the debate. Overall, on such a political question, the likelihood of collecting an actual opinion depends not only on the comprehension of the question’s terms and concepts, but also on the knowledge of European institutional mechanisms.
23The main problem with such an assertion certainly lies in the desire to collect a “true opinion,” but before coming to this, we will consider the argument that respondents must “formulate informed answers to a question.” Why would a respondent’s answer not be a “true opinion” on the grounds that the respondent would not have answered the question in the context of the issues that the person who wrote the question or the best specialists on Europe would have in mind? The same difficulty arises in any field of social sciences. To a question such as “do you like action films?” or even “do you go to see action films?”, would respondents be unable to answer because they do not know all the internal workings of the cinema, or action films, and they will undoubtedly not attribute the same meaning to the question as a movie buff or an industry professional, or not be completely sure of everything covered by the expression “action film” among industry professionals?
24We could further add that one of Pierre Bourdieu’s major works, Distinction, discusses such an imposition of a problematic by asking the respondents whether they prefer The Blue Danube or the Piano Concerto for the Left Hand. We can in fact ask to what extent we are imposing an aesthetic judgment on the individual in the same way that we impose a political judgment or why it would be more methodologically rigorous to ask the respondents whether they trust one politician or another than whether they like the Piano Concerto for the Left Hand. But above all there is no guarantee that the beginning of Distinction does not aggregate ultimately different opinions and therefore that Bourdieu’s commentary concerns nothing more than “artefacts” through the questions posed to individuals who do not always have the knowledge required to distinguish between these different works.
25The problem with this criticism is that it assumes that there is a correct way to answer the question. Critics of polls would like without doubt for respondents to answer with knowledge of all the issues underlying the question, as if they were taking a political science exam in which the student must understand the problematic that the professor who wrote it had in mind. It is true that, depending on the political competence of the respondent, identical responses to a questionnaire will have different meanings, but it is quite simply part of the work of research to be able to identify those responses in the event that these distinctions are important, and we will see that they are not always so.
26Everyone responds based on their own terms of reference and it cannot be said that there are informed and uninformed responses, because an opinion poll is not merely a political science exam. A respondent does not, in order to respond to a question, necessarily have to understand it in the same way as the person who asked it. Those who use the argument of the imposition of a problematic seem to consider it obvious that respondents should interpret the question in the same way as the researcher. However, when we undertake a secondary analysis of the data, we only have a vague idea anyway of what the researcher intended the meaning of the question to be. The respondents respond to the question by translating it into their own problematics and it is this that must be uncovered and understood. But this criticism of polls that maintains that they do not capture the real opinions of the respondents nevertheless raises an even more fundamental issue: the failure to take into account the role of experimentation in the social sciences and the role of the scientific question we seek to answer.
Observation and experimentation, or the confusion between “real” and “natural”
27It is surely worth noting, uncontroversially, that if we were to actually go along with this criticism that advocates the primacy of freely expressed perceptions, we would end up believing that the work of social scientists amounts to nothing more than letting the respondents speak or act and finding a way to record them without the least interference. Engaging someone in discussion or activities in any format at all is in fact always a form of imposing a situation and a way of recording the data. It is similarly evident that in a qualitative interview, the interviewer is never “neutral,” no more than their questions are. Incognito observation without disturbing the “scene” would therefore become the only acceptable methodological tool, the only one that would not impose a particular formulation of the questions or a framework on the respondents. And indeed, the supremacy of ethnography over methods like interviews is commonly asserted on the pretext that practices are more real than words. The root of this criticism is again the empiricist illusion that reduces science to a strict record of reality as it would be from the point of view of an entirely neutral observer.
28The argument accuses polls of distorting reality, of putting the respondents in positions or confronting them with alternatives and problematics that are foreign to them, and of producing artefacts in doing so. This argument applies no more to opinion polls than to any other form of scientific experimentation and this is precisely the problem with this abusive use of the term “artefact.” In the case of opinion polls, the researchers construct the questions and impose the formulation of them on the respondent. They consequently manipulate several parameters of the situation: the position of the questions in the poll, the wording of the questions, the choice of how to answer them, etc. This situation has therefore almost every feature of an experimental situation, in that it is certainly a way to examine the social reality and not simply to observe it, a situation upon which the results are indeed dependent. Conversely, the epistemological ideal that poll critics seem to advocate is that of the raw observation of the social reality in contrast to data that would manipulate it by creating an artefactual situation.
29The results are not however artefacts simply because they are the product of a scientist’s activity and not the raw observation of reality. Experimental physicists would be rather amazed to learn that by examining nature in their laboratories and by testing different protocols, they produce “artefactual” measurements that are therefore without scientific value, while their colleagues responsible for measuring a natural phenomenon outside the laboratory have access to a “true reality.” Critics of opinion data seem therefore to adopt a perspective that is in fact based on a naive form of empiricism, a long way from the “construction of the object” or the “epistemological break.” The defense of observation as the only method allowing access to the social truth is an epistemological step backwards for the social sciences. In the mid-nineteenth century, in his Introduction to the Study of Experimental Medicine, Claude Bernard noted the contrast between observational science and experimental science, highlighting that it was through the latter that knowledge could be furthered more rapidly, with the ability to act on nature offering a far superior capacity for research. The same applies to the social sciences: the fact that a situation is not natural, i.e., independent of the observer, in no way means that it is not real or that it provides no avenue for improving our knowledge of the social reality. To the contrary, the ability to manipulate the facets of a social situation indeed enables certain theoretical hypotheses to be put to the test.
30The use of the term artefact thus relies on the illusion of science as merely observing reality without changing or influencing it. This modification of reality apparently produces biased data or artefacts: a “false reality”—false because it is changed by people. Indeed, it is precisely the modification of this reality as a result of our controlled activity that enables us to understand it and it is actually the point of the experimental approach to examine reality in this way, to manipulate it so that it can answer the questions we are asking, which would rarely happen if we contented ourselves with “watching” the world, waiting for it to respond to our questions of its own accord, as is advocated by proponents of observation as the sole scientific method.
Scientific questions and data production
31Criticisms of polls that rely on the argument of the imposition of a problematic and that denounce “bad questions” neglect another significant methodological issue: the need to take account of the problematic inherent in the use of data and to evaluate a method in relation to this problematic. In a quantitative survey, there is no reason to think that one question would be better than another a priori on the pretext that it would address more “buried” or “better established” convictions, or even a “true opinion.” When respondents answer questions that they have not considered before, the conviction they attach to their responses is, without doubt, not to be taken at face value, but, even if weak, it nevertheless has the potential to reveal the nature of their political opinions, their prejudices, the first things that come to their minds, or even what they think the researcher expects of them. These data are just as significant as well-established convictions, which are not in fact necessarily of greater interest and may merely parrot theories publicized by opinion leaders or newspapers that the respondents have internalized and adopted.
32But above all, why would the scientific contribution be made more significant if a political question were formulated, as some would wish, in the respondents’ own terms? Do we learn anything by asking radical left-wing activists “Do you think that left-wing governments should cut ties with the financial community?” or by asking them questions that reference the arguments of their political adversaries, which they will already know well: “Do you support the policy of reducing the public deficit in order to reduce the interest rate on debt and reestablish confidence in the markets?”? Such targeted questions would without doubt be far more present in the thinking of these activists and, if we go along with the criticism of the imposition of a problematic, these are actually the only questions that we would be able to ask. Would it not be just as interesting to confront respondents with ideas that are unfamiliar to them?
33Although we could be tempted to respond in the affirmative to this last question, I maintain in fact that it makes no sense to try to respond to it: empirical data are of interest within the framework of a question and we therefore cannot evaluate them independently thereof. It is impossible to declare a priori, as criticisms of opinion polls and a number of methodological handbooks do, that a question is good or bad. It is the empirical test or experiment that must be evaluated.
Aggregation and problematic
34The criticism of opinion polls that maintains that it is impossible to aggregate responses that are superficially the same, but that have a different meaning, again relies on the false empiricism that is expressed very well in this passage by Daniel Gaxie:
It is potentially thorny to add up formally identical responses to close-ended questions without precaution. When researchers make such totalisations, they presuppose that all the interviewees who have given formally identical responses have not only understood the question, but have also understood it in the same way.
36This criticism does not recognize that the question of knowing whether we can proceed with such an operation does not depend on the data and therefore whether individuals have understood the questions in the same way, but on the scientific question at hand or on what we want to demonstrate. Returning to the example of the different cinema-related practices (going as a couple, as a movie buff, with friends, etc.), we would not aggregate them in the same way if we wanted to conduct an economic evaluation of the expected profits of the film industry (in which case we would without doubt aggregate them), to study the cultural practices of French people or young people, to study relationships between couples, or if we were interested in the success of a particular genre of film. The same applies to opinion polls. We may or may not want to combine different opinions depending on what we want to do: predict the outcome of an election, describe the diversity of opinions on the left or the right, or even examine the impact of socioeconomic characteristics or education levels on attitudes to the redistribution of wealth or to the death penalty.
37Practices, like responses to questions, are in themselves never the same or different, but only insofar as they are described in one way or another. This is a philosophical and epistemological point that, admittedly, is not always entirely obvious or intuitive. It is however a well-known difficulty, in particular among sociologists and anthropologists. For example, Mary Douglas wrote “no superficial sameness of properties explains how items get assigned to classes” (Douglas 1986, 58). She was of course thinking more of material objects, but we can say the same of actions or opinions. Identifying the objects, classifying them, or categorizing them is done in a theoretical context, in a particular problematic, and with a given intention. To single out the issue of aggregation or classification as a special case that would in essence apply only to opinion data, is to fail to grasp a fundamental philosophical issue. As stated by Mary Douglas, inspired by philosophers Quine and Goodman, sameness is not something that can be recognized in things themselves; it is only conferred upon elements through their insertion into a coherent scheme (1986). It therefore makes no sense to assert that we cannot, ever, aggregate formally identical responses that have a different meaning. We will always find one interpretation of two responses, and more generally of two actions, that allows them to be grouped under the same concept or, conversely, an interpretation that distinguishes between them.
38The aggregation of responses and the production of a figure to summarize the diversity of opinions only has value within the framework of a specific question and cannot be judged independently thereof. Returning to the example of questions on the European Union, without doubt a significant proportion of voters do not have all the knowledge required to understand all the facets of these questions, but, whether we rejoice in it or deplore it, this is also how they will judge the majority of policies and how they will vote. As voters, we never have the means to truly understand the policies implemented and we have to form an opinion and make a choice based on extremely piecemeal knowledge. Nevertheless, this is precisely what political scientists must understand: the reasons and mechanisms that can explain and predict the political choices made in these far from perfect decision-making circumstances. As such, there is merit in political scientists studying the responses to questions often formulated on the basis of terms of reference that are foreign to the respondents. This is the only way that they can understand how democracy currently works.
39Whether it is more interesting to study behaviors, values, opinions, literature or polls, and qualitative or quantitative data is a question to which there is but one banal response: “it all depends on what you want to test or understand.” More generally, it is rather superficial to think that the scientific value of data would be one of its intrinsic qualities and not a characteristic of its use. We could not know before the formulation of Einstein’s theory that the simple observation of an eclipse would bring Newtonian mechanics crashing down, a pillar and symbol of scientific successes. In any other field, logicians and analytic philosophers entertain themselves by imagining the most absurd situations, because it is these situations that allow us to question our terms of reference or to put them to the test. The simple story of the barber unable to shave himself, conceived by Bertrand Russell, for example, ruined the logical foundation of arithmetic that the philosopher and logician Gottlob Frege wanted to establish. Cognitive psychology is also overflowing with examples of experimentation, and it is precisely by putting respondents in the most unexpected and unusual situations that we move beyond the common knowledge provided to us by everyday social situations. Finally, taking one last example, economists, using the method of instrumental variables, create their own experiments to study correlations that could a priori seem absurd, but that are in fact better able to uncover causal links than correlations that are more immediately comprehensible.
40Polls should therefore certainly not be seen as photographs of opinion, but rather as experiments and, as such, questions with counter-intuitive formulations are to be encouraged in order to test the reaction of the respondents. It is not, to the contrary, of great interest to let respondents talk about their own natural terms of reference, as these often constitute knowledge to which we already have access via other avenues. Empiricism in science is not simply measuring or observing reality, but testing it, whether this reality is natural or social. Consequently, empirical data have no value or wealth of their own, they only have significance to the extent that they enable certain theories or, at least, certain proposals to be tested. We must therefore put an end to the myth that there are “good questions” and “bad questions.” There are only bad empirical tests of theories.
41Science does not always seek a “rich” or profound description of reality and, as such, a poet may well dispute an oceanographer’s description of the sea or a meteorologist’s description of a storm, and it is unlikely that a sociologist would have much to contribute to literature. Statistics on suicide and Émile Durkheim’s analysis of them certainly do not describe the internal experiences of an individual about to end their life. Durkheim’s objective is not however to give a phenomenological description, but to identify the social determinants of suicide, the differences between countries or societies, and the relationships between types of suicide and social integration. In the same way, it is not the role of opinion data to plumb the depths of the soul or the identity of the respondents, but rather to seek to test hypotheses relating to social phenomena.
Public opinion and nominalism
From methodology to social ontology
42As highlighted in the introduction, the criticism of the use of opinion data has an ontological dimension: public opinion is allegedly no more than an artefact constructed by illegitimately aggregating responses. As such, the description of public opinion as a collective reality is also in the firing line. The criticism amounts therefore to submitting public opinion to Occam’s razor, the central tenet of which is not to employ more entities than necessary. In one sense, this is indeed one of the first “uses” of this razor, which was to reject the existence of the collective realities of the social world. It was Occam who asserted that the Pope could not give his property to the Franciscan Order because the order considered as a whole did not exist; it was nothing more than a name for the Franciscan individuals (Desrosières 2002 ). Should we subject public opinion to the same fate? And consider it a fiction constructed by opinion polls?
43The standard criticism of opinion polls is the assertion that public opinion does not exist because it is no more than an “aggregation,” with this term usually having a pejorative meaning in the social sciences. This assertion has significant consequences for social theory given that it calls into question its objects and collective phenomena, and the possibility of understanding them through statistical methods that aggregate measures concerning individuals. This willingness to definitively condemn opinion data thus leads to the mobilization of the same old arguments against using quantitative methods in social sciences. Pierre Bourdieu, for example, asserts that a percentage is unable to express the reality of public opinion because it is unable in turn to express the reality of conflicts within society.
The “public opinion” that is manifested on the front pages of newspapers (“60 percent of French people are in favour of …”) is a pure and simple artefact whose function is to disguise the fact that the state of opinion at a given time is a system of forces, tensions, and that nothing more inadequately expresses the state of opinion than a percentage.
45Firstly, we must consider two points. Opinion polls do not create objects like “public opinion” by aggregating responses as a percentage. It is those who process the data who do so and who choose to represent a variable or distribution of responses in this way. This distinction is important because opinion polls only produce the variable and it is the analyst who chooses to summarize the information that it contains as a percentage. The production of a single figure (“60 percent of French people think that”) is therefore in no way an imperative and the researcher may well study the distribution of the responses to a question in the different social classes, their correlations with other responses, etc.
46The second point to consider is that we often pass off as epistemological criticism arguments that are ultimately just a criticism of formulations that are a little brief. Of course, if someone were to write that “French people think that,” we would doubt whether all French people unanimously share the same opinion and this is indeed most often shorthand for saying that “the majority of French people responded that.” We encounter the same problem when we write “young people,” “workers,” etc. A criticism that is intended to be epistemological and scientific is thus reduced to a simple battle against the abuse of language and often simplistic turns of phrase, or even a battle against the pragmatism of language that does not seek to make everything explicit in the articulation of a phrase, in particular that which is abundantly evident or that which the context would usually enable to be understood.
47But if we want to look further and examine the basis of the criticism, the use of a single figure to report on a complex reality, we could certainly stumble upon a real problem. It is not however one unique to opinion data. In the same way, we could say that an unemployment rate is inadequate because it does not directly reflect the power relations or the tensions in the economic and social system. These arguments are mirrored throughout the history of statistics: Hoover disputed the unemployment rate on the grounds that it did nothing more than add up different and incomparable realities (Desrosières 2002 ). 
48This is in fact the question asked by Alain Desrosières throughout his work on the politics of large numbers: How do we construct equivalence, i.e., consider as identical behaviors that are not identical, in order to produce, by adding them together, a single figure describing the social world? How do we go from the diversity of unemployed people’s situations to a single figure characterizing unemployment? (Desrosières 2002 , 67–102). Behind this problem we find Durkheim, who analyzed suicide trends through suicide rates, which are never anything more than aggregations of individual actions that Durkheim fought to have recognized as sociological realities. By asserting that aggregating opinions and producing a percentage in this way merely constructs an artefactual reality, we do far more than criticize opinion data and public opinion; we rekindle a debate concerning sociological theory and its methodology as much as ontology and we come back around to the ideas with which Durkheim was trying to dispense.
49Far from the usual arguments, Pierre Bourdieu’s assertion that “public opinion does not exist” is often based on individualism and nominalism, which mistrust abstract entities and prefer the more tangible reality of the diversity of behaviors. Moreover, he does not apply this nominalism only to public opinion. It recurs throughout his work when he asserts, for example, that youth does not exist. In some of his assertions, Pierre Bourdieu even rejects any possibility of the scientific use of a collective concept: trying to determine whether social classes or regions exist would, according to him, usurp a “divine right” (Bourdieu 1982, 12–13).
50If we really were to go along with this criticism, we would end up condemning all statistical aggregation: the unemployment rate, gross domestic product, birth and migration rates are just “aggregations” of individual events that each have their own specific characteristics. The stories of each unemployed person, each economic transaction, each family expecting a child, or each migrant are distinctive, just like the stories behind each suicide studied by Durkheim. We could likewise claim that it is never possible to aggregate and that aggregation only produces a statistical artefact compared with the actual observation of each unique event. In contrast to this perspective, I have maintained above that aggregation is only of value in the context of a specific question and that it cannot be judged to be impossible or possible a priori. I would now however like to examine this question more directly from the perspective of the reality of social phenomena measured in this way.
A pragmatist epistemology
51I advocate here for a pragmatic approach  in order to understand this issue of the production of statistical measures of sociological realities and, more specifically, of public opinion. With pragmatism we depart from the description of real sociological practice and ask the question more in terms of success than of reality. This is, for example, supported by the philosopher Hilary Putnam:
Like the great pragmatists, these thinkers [Quine, Davidson and Goodman] have urged us to reject the spectator point of view in meta-physics and epistemology. Quine has urged us to accept the existence of abstract entities on the ground that these are indispensable in mathematics, and of microparticles and spacetime points on the ground that these are indispensable in physics; and what better justification is there for accepting an ontology than its indispensability in our scientific practice? he asks.
53It is therefore first and foremost an argument based on fact: the radical nominalism that effectively rejects concepts that describe collective social realities does not exist in the vocabulary of contemporary social sciences, which is full of terms like state, unemployment rate, labor market, the government, the working classes, etc. While the ontological status of these collective realities is not the subject of unanimous agreement in philosophy, they at least undeniably exist in the language of sociology to express the fact that individuals act together in specific social situations and that our concepts do not refer to these individuals taken one by one, but rather to the social reality that they construct together. Considering this, we must conclude that the description of social and individual realities is one and the same thing. But to think that we could now dispense with these famous concepts, as Weber advocated in his famous letter to the economist, Robert Liefmann, is utterly illusory. Even today it would mean discarding the majority of the terms used in sociology. Saying that public opinion is an artefact has no more value than calling the state or social classes artefacts. In any event, these are concepts that enable us to describe collective realities.
54Sociological practice obliges us therefore to describe social realities that are more difficult to observe with the naked eye than simple behaviors. In a survey on social mobility, for example, we summarize, based on only a few figures, the mobility of entire generations or social groups within a given society. We thus use many concepts in reference to collective realities. In the same way, when we study public opinion, we are not merely seeking to understand each individual situation, but rather the overarching phenomenon: the more or less pronounced prevalence of an opinion in different social groups, its evolution over time, or even its variations between countries. The aggregated rates obtained summarize a reality that, admittedly, is without doubt much more complex. But this is how science advances: it does not seek an irreducible description for every single entity.
55Methodological constructivism and individualism allow us to assert that the actions of individuals are not the result of social laws or of structures defined beyond them that constrain them to acting in a certain way like how the laws of magnetic fields constrain the particles that cross them. These epistemological approaches also prevent us from anthropomorphizing collective realities, i.e., attributing an intention to them in the same way that we would attribute an intention to an individual. While we should not therefore talk about the state as an individual that acts, has intentions, or changes its mind and we should to the contrary make visible the diversity of actors that form it with their competing interests, this does not however mean that we must banish the concept of the state from language. In the same way, talking about public opinion should not be done as if this opinion were the parallel of an individual opinion, but rather as the description of a collective reality: the fact that an opinion is apparently more or less accepted or rejected by the population of a particular country, by certain social groups, in certain eras, or in specific political situations. Public opinion is therefore a way to describe collective phenomena, which is what many historians studying attitudes and sociologists do. Jean Stoetzel, who introduced polls to France, similarly understood opinion polls as a way to empirically study broad social trends. Despite his criticism of sociological realism, which he associated with Durkheim, he described public opinion as having a permanence that surpasses basic individual existences.  Polls are thus a way to study the shared values upon which, according to Stoetzel, the social glue relies (Stoetzel 1943, 356). Talking about public opinion therefore enables us to describe the persistence of certain opinions in a group and how the sharing of these opinions—the fact that they become public—plays a central role in how society works.
56The notion of public opinion could therefore have a more significant role in the sociological theory of political phenomena. In a 2012 article published in the American journal Sociological Theory, Jeff Manza and Clem Brooks similarly noted that sociology had “lost public opinion,”  with the concept never used by academics working in the field of political sociology. It is even often the subject of rather aggressive attacks. Manza and Brooks maintain nevertheless that it is in fact employed in other forms. With regard to how a political power imposes its legitimacy, as discussed by Michael Mann, or how citizens revolt, as discussed by Charles Tilly, we always form important hypotheses about the political attitudes of the population we are studying. We are therefore, according to Manza and Brooks, very close to a theory of public opinion.
57The concept of public opinion is a tool to identify a reality that is difficult to grasp: the macrosociological consequences of the same opinions being shared by a significant proportion of the population. This reality lays the foundations for numerous theories like those concerning political legitimacy or control and that assume that individuals accept the power and control exercised over them. To describe this “acceptance” and the reasons for it, we can certainly use concepts other than “opinion,” a term that has the flaw of not “sounding” very theoretical or scientific. We can as such maintain that this acceptance is explained by values or a political stance. That is not to say that there is no difference between these concepts or the theories to which they relate, but it does highlight that they are attempting to do the same thing: to describe how the political perspectives of a significant proportion of the population contribute to the functioning of overarching social mechanisms, such as support for a public policy, the legitimacy of state power, or even the reproduction of domination.
58We do not seek here to develop a theory or a rigorous definition of public opinion, or even to defend the concept of public opinion in particular, but rather to show that the epistemological criticism of public opinion relies itself on a false epistemology in that it claims that public opinion can be nothing more than an artefact because it is constructed through the aggregation of diverse opinions. We can of course use a term other than public opinion, but we must still, in any case, try to describe the collective dimension of the phenomenon we study—which we may think of as public opinion, shared values, or even a doxa—and not content ourselves with descriptions, as rich as they may be, of individual cases. As such, returning to a pragmatist epistemology, we are actually talking about accepting concepts that seek to describe the collective reality of this phenomenon and attempt to measure it.
Scientific language and everyday language
59Pierre Bourdieu’s epistemological approach is underpinned by the construction of the object, an imperative that has achieved success, considering that it appears in the majority of methodology textbooks. This constructivist perspective is based on the Bachelardian idea of a fact that is first and foremost “conquered” in opposition to the everyday meaning, or even the Durkheimian idea of a break with preconceptions. There is indeed a problem when researchers believe that a scientific description and measure can be provided for every term used in everyday language. But as important as this criticism and the underlying epistemological approach are, we must also uncover the limits of this approach, which adopts the natural sciences as the epistemological paradigm, neglecting the fact that the social sciences are, whether we like it or not, far from having actually accomplished such an epistemological break, i.e., having constructed a language to describe the social world that is as different from everyday language as the language of theoretical physics is from our everyday understanding of the natural world. In other words, physicists talk about objects beyond our everyday perception —bosons, neutrinos, etc.—and that can undoubtedly explain the fact that we talk about an epistemological break. There is however no close equivalent in the social sciences of this type of transformation of the language used to describe reality. The social sciences more often seek to elucidate the meaning of concepts that we use on a daily basis, like social class, identity, citizenship, integration, or, here, public opinion. There are not, moreover, any theories that would actually provide a conceptual framework that could entirely replace everyday language. What indeed would sociology become if it were to discard all commonly understood concepts and all the objects as identified in the social reality in order to adopt only those that are “scientifically constructed”?
60This position is obviously flawed in that it introduces too great a break between everyday knowledge and scientific knowledge, and between scientifically constructed objects and others that are apparently nothing more than the illusions of everyday language. Constructing the object is not a separate and independent scientific act, it is the construction of a theory that might replace the everyday theory, i.e., not the construction of a specific object, but rather the construction of a network of concepts for understanding reality and asking questions about these concepts. As such, condemning the failure to construct the object should involve proposing the use of a more satisfactory alternative theory. In our case, it is pointless to condemn researchers for talking about public opinion without proposing a system of concepts that enables them to describe the diversity of opinions as well as the situation in which an opinion becomes more or less the consensus across a social group. Replacing opinion with behavior is no more “scientific.” Behavior in particular is in no way a scientific concept; it is a concept that is similarly based on its everyday understanding, but one that seems more scientific because it is more tangible.
61The problem is with the abusive use of the term “break.” It is through empirical research and theoretical reflection that the object is built, an object that was at the outset merely poorly defined by everyday understanding, just like, for example, planets and unemployment. The imperative to construct the object has its limits, in the impression that it gives that this construction could be, at a given moment, complete if we were to successfully construct this object in opposition to everyday knowledge and finally grasp our scientific object. There is not however truly an opposition between supposedly scientific objects and others that are supposedly not scientific. If the language used to describe the social reality is changing, this is not due to a break; it is being constructed gradually by the empirical research that is, step by step, uncovering the limits of our everyday concepts. This approach therefore often turns to misplaced arrogance when it asserts that public opinion is merely an artefact compared with scientific concepts that themselves are actually constructed, as if we could not find, in any social sciences article, a multitude of everyday concepts that have not been subjected to the razor of scientific construction or the epistemological break. This is not necessarily a problem, precisely because these concepts are only starting points. As highlighted by Quine, our examination of the world can only begin with the concepts that we have at our disposal.
Neurath has likened science to a boat which, if we are to rebuild it, we must rebuild plank by plank while staying afloat in it. […] Our boat stays afloat because at each alteration we keep the bulk of it intact as a going concern. Our words continue to make passable sense because of continuity of change of theory: we warp usage gradually enough to avoid rupture. […] our questioning of objects can coherently begin only in relation to a system of theory which is itself predicated on our interim acceptances of objects. We are limited in how we can start even if not in where we may end up. To vary Neurath’s figure with Wittgenstein’s, we may kick away our ladder only after we have climbed it.
63We must therefore employ everyday concepts such as identity, nation, ethnicity, social class, or public opinion even if research increasingly demonstrates the limits of their use and makes it necessary to rework them to advance our knowledge. Conversely, the condemnation of some concepts that are supposedly not scientific because they supposedly do not exist or have far too many diverse meanings makes little sense if it does not suggest the use of an alternative theory that would advance scientific language. Wiping the concept of “public opinion” from language could only be done if we were to propose other concepts that would enable us to better describe the diversity of attitudes and opinion in a country, as well as their possible unity and potential consensus over a given period. We only change our ontology when a system of description offers us the option of describing reality in a clearer and simpler way.  But when this is not the case, we can, as Quine proposes, allow ourselves to endure their use safe in the knowledge that there is no reason to stagnate. And it is too bad if sociology must wait before it succeeds in entirely eradicating everyday language and thus achieve the Durkheimian dream of a discipline that assumes the esoteric character which, according to him, befits all science.
64The position defended here is not intended merely as an appeal in defense of opinion data, but rather as an appeal for methodological pluralism. The social sciences are today broken apart by artificial objections that, rather than acknowledging the diversity of ways to produce knowledge and enrich our theories, regularly proclaim the supposed superiority of one method, whether qualitative or quantitative, over another or even of a trend in sociology, instead of examining their respective contributions. Nevertheless, observing the natural sciences should demonstrate to us the infinite diversity of methods for producing knowledge. This does not of course mean that we must yield to epistemological anarchism, but conversely that we must subject every method to critical examination and be skeptical when confronted by those who proclaim the superiority of their own method, which often ends up simply promoting one way among many of understanding the social reality that, if we were really to content ourselves with it alone, would only offer us a rather narrow understanding. Opinion data are therefore one way among many of generating knowledge. They provide us with pointers for understanding individuals’ attitudes and the spread of certain perspectives that the mere observation of their “natural” practices could not. They are clearly nothing more than data among other data.
65Lastly, I am not particularly concerned with the political question or criticism of the media world, which are in fact at the center of a work like that of Patrick Champagne. It is clear that people can find themselves frustrated by the everyday practice of polling that consists of legitimizing a policy by asserting that it is supported by a majority of the population. But this would be to raise an issue that has little to do with the epistemology of the social sciences and opinion polls. The use of opinion polls on television often amounts to a simplistic conception of what constitutes a democratic political regime. But by focusing their criticism on the epistemological foundations of polls rather than on the political philosophy of politicians or the columnists of major newspapers, critics of polls have largely obscured the debate. There are without doubt good political and philosophical reasons to oppose the idea that a democracy should be governed by polls. Yet we do a disservice to both science and politics when we seek to pass off this criticism of the use of polls and their political opinions in public spaces as the defense of something that is supposedly the only legitimate scientific practice.
I would like to thank Pierre-Marie Chauvin and Jeremy Ward for their comments, suggestions, and critiques. I alone am responsible for the opinions defended here.
Translator’s note: Unless otherwise stated, all translations of cited foreign language material in this article are our own.
For a “social history of polls,” see Blondiaux 1998.
This is not however the case according to Pierre Bourdieu, who argues that voting is also the result of the imposition of a problematic (1993 ).
See for example the collection of Otto Neurath’s writings (1973).
“Dewey was explicit on this point: ‘Meaning… is not a psychic existence; it is primarily a property of behavior.’” (Quine 1969, 26–27).
“For some time there has been talk of a “Two Speed Europe.” This means that some countries would be ready to intensify the development of a common European policy in certain important areas, while other countries would not. Please tell me, for each of the following countries, whether or not you see it as being ready to intensity the development of a common European policy in certain important areas.”
Furthermore, this criticism often simply relies on the classic criticism of the average, which amalgamates situations that are too diverse.
I draw here on the work of the American pragmatist philosophers and more specifically of Hilary Putnam. For an overview, see Putnam 1990, Realism with a Human Face. I will not establish any connection with pragmatist sociology.
“Public opinion should be recognized, like the public itself, as having permanence, a longevity that overcomes fluctuations and even individual appearances or disappearances” (Stoetzel, 1943, 362). For a discussion of this passage by Stoetzel and its relationship with Maurice Halbwachs’ theories, see Marcel 1998, 30.
An article by Burstein in the journal Social Forces (Burstein 1998) observed that in sociology journals, none of the articles published between 1980 and 1990 on the consequences of public policies included a measure of public opinion.
“At any rate the ontology of abstract objects is part of the ship which, in Neurath’s figure, we are rebuilding at sea. We may revise the scheme, but only in favor of some clearer or simpler and no less adequate overall account of what goes on in the world.” (Quine 1969, 16).