Positivism, Interpretivism and What Lies Between

Six Parts of this course have made claims about society. This Part asks the question that should have been nagging at you the whole time: how does anybody know?

And before we can talk about surveys and interviews and experiments, we have to settle something prior. Is studying people like studying anything else?

Your answer to that decides what you think evidence is, what you think an explanation is, and what you will accept as proof. It is not an abstract question. It changes what you do on a Tuesday morning with a notebook.

Two researchers, one problem, two entirely different days.

A university wants to know why students drop out. It hires two researchers.

The first researcher gets the records. Forty thousand students over ten years. She has, for each one, their entry marks, their family income bracket, whether their parents went to university, their commute distance, their first-semester grades, whether they had a scholarship, and whether they finished.

She builds a model. She finds that first-semester grades predict dropout more strongly than entry marks do — meaning something happens after arrival. She finds that commute distance matters, but only for students in the lowest income bracket. She finds that the scholarship effect is large, and that it is much larger for first-generation students. She can put a number on each of these, and she can say how confident she is in each number.

What she cannot tell you is what it was like.

The second researcher takes a different day. He sits in the canteen. He gets to know thirty first-year students and follows them for a year. He goes to their rooms. He reads their group chats, with permission. He is there in March when one of them stops coming.

He finds that almost nobody experiences dropping out as a decision. They describe a slow accumulation: a missed class that made the next class harder to walk into, a tutor's comment that landed badly, the discovery that everyone else seemed to already know how to do this. He finds that the students who leave almost all describe a moment of feeling like an impostor — and that this moment is usually about class or language or accent, not about ability. He finds that the ones who stay usually name one person who noticed them.

What he cannot tell you is how common any of this is.

Both researchers wrote a report. Both reports are true. Neither can be converted into the other, and the university has no idea what to do with them.

Rocks do not have opinions about being rocks.

Here is the difficulty that started the whole argument, and it is not a small one.

When a chemist studies a compound, the compound has no view about the experiment. It has no reasons. It does not act differently because it read the last paper. It does not lie to the chemist, or try to help, or find the question offensive.

People do all of those things. And more importantly: people's behaviour is partly constituted by what they think they are doing.

Raising your arm and signalling a bid at an auction are the same physical movement. What makes one a bid is a set of shared meanings and rules. You cannot see the bid. You can only see the arm. To know it was a bid, you have to know what the movement meant — to the bidder, and to the auctioneer.

So the question is:

Is a science of society just a science that happens to have people as its subject matter — same logic, harder data? Or does the presence of meaning change the kind of knowledge that is possible?

Every methods argument in sociology descends from this one.

The first answer: the science of society is a science

Positivism, in its original and non-insulting sense.

Auguste Comte (1798–1857) coined the word sociology and proposed that the study of society could become a science like the others — later than the others, because it is more complex, but the same in kind.

His law of three stages said that human thought about any subject passes through three phases: the theological (events are explained by the will of gods), the metaphysical (events are explained by abstract essences and forces), and the positive (events are explained by observed regularities between observable things). Physics had reached the positive stage. Society had not. Comte's project was to bring it there.

"Positive" here means "given" — what is posited, laid down, observable. It does not mean optimistic and it does not mean approving. It means: start from what can be observed, and do not go beyond it.

The core commitments of a positivist social science:

One — the same logic. There is one scientific method, and it applies to atoms and to societies alike. This is sometimes called naturalism (the position that social science is continuous with natural science).

Two — observable evidence. Claims are settled by what can be observed and measured, not by intuition, authority or introspection.

Three — regularities. The aim is to establish patterns of the form when this, then that — reliably, across cases.

Four — explanation as subsumption. To explain something is to show it as an instance of a general pattern. Why did this rate rise? Because rates of that kind rise under those conditions.

Five — the separation of fact and value. What is the case and what ought to be the case are different questions, and science answers only the first.

Two corrections that matter, because this is where most textbooks go wrong.

First: Durkheim objected to being called a positivist. In the preface to the second edition of The Rules of Sociological Method he complains about the label directly, and describes his own position as rationalist rather than positivist. He was not signing up to Comte's system; he was arguing that social facts have to be treated as real and studied from the outside — which is a narrower claim.

And his own practice was less austere than the slogan. The Elementary Forms is a study of meanings — of what the sacred does to people's minds. Durkheim the positivist and Durkheim the analyst of collective representations are the same person.

Second: "logical positivism" is a different animal. The Vienna Circle of the 1920s and 1930s — Schlick, Carnap, Neurath and others — held the verification principle: a statement is meaningful only if it can be verified by observation, or is true by definition. Everything else — ethics, metaphysics, theology — is not false but meaningless.

It collapsed, and it collapsed on its own terms. The verification principle is not verifiable by observation, and it is not true by definition. By its own rule it is meaningless. There were other problems too — no finite set of observations verifies a universal law — but that one was fatal.

Almost nobody in sociology has ever been a logical positivist. Which is worth remembering, because in sociological argument "positivist" is now mostly used as an accusation, aimed at people who would not accept it, on the basis of a doctrine they never held.

The second answer: understanding is not the same as explaining

Interpretivism, and the German argument behind it.

While Comte was building a natural science of society in France, a different argument was under way in Germany about what kind of knowledge the human studies produce at all.

Wilhelm Dilthey (1833–1911) drew the line that everything else hangs on. He distinguished the Naturwissenschaften — the sciences of nature — from the Geisteswissenschaften, the sciences of mind or spirit, meaning history, philology, law, the study of human works.

The natural sciences explain (erklären): they bring an event under a law. The human sciences understand (verstehen): they grasp a meaning from the inside.

And you cannot substitute one for the other. You can give a complete physical account of a man writing a letter — the muscle movements, the ink — and have said nothing about the letter.

Wilhelm Windelband added a second distinction with the same shape: nomothetic knowledge seeks general laws; idiographic knowledge seeks to characterise the individual case in its particularity. Physics is nomothetic. History is idiographic. The question for sociology is whether it has to choose.

Weber's actual position, which is the one almost everybody misstates.

Weber is claimed by the interpretivists, and the claim is half right. Here is his own definition, from the opening page of Economy and Society:

Sociology is a science which attempts the interpretive understanding of social action in order thereby to arrive at a causal explanation of its course and effects.

Read that again, because it contains both halves. Interpretive understanding — and causal explanation. Weber did not think you had to choose. He thought a sociological account had to be adequate twice over:

Adequate at the level of meaning (Sinnadäquanz): the account makes the action intelligible as something a person could have meant to do, given their situation and beliefs.

Adequate at the level of cause (Kausaladäquanz): the account is supported by evidence that, in cases of this kind, this sort of situation is in fact followed by this sort of action.

Meaning without causation is a story. A plausible narrative about why people did something, with no evidence that they generally do. Causation without meaning is a correlation you cannot interpret. You know that A is followed by B and you have no idea what is happening.

Weber wanted both, and said that either alone was a failure. This is the most important single sentence in this lesson.

The Protestant ethic (see 4.3.7) was built to this specification.

The meaning half: what it was like to hold a doctrine of predestination — the anxiety, the search for signs of election, the reinterpretation of methodical work as a calling. Weber reconstructs it from sermons and devotional manuals, from the inside.

The cause half: the comparative and statistical claims — the occupational composition of Protestant and Catholic regions, and the long comparative studies of China, India and ancient Judaism, designed to show that the combination did not arise where the doctrine was absent.

Modern research has damaged the causal half — Becker and Woessmann's work using Prussian county data finds the Protestant economic advantage is better explained by literacy, which Protestantism promoted for the sake of Bible reading, than by the ethic itself. But notice the shape of the disagreement: it is being settled on evidence, using the second criterion. The argument is inside Weber's framework, not outside it.

Interpretivism after Weber, and how far it goes

Two developments, one moderate and one radical.

Alfred Schutz (1899–1959) asked how Verstehen is actually possible, and answered with typification. We do not encounter the social world raw. We encounter it through ready-made types — "a customer", "a policeman", "the sort of person who says that". These types are the ordinary equipment of everyday life; we use them without noticing.

His key methodological point: the social scientist's concepts are second-order constructs — constructs of the constructs made by the people being studied. The people have already interpreted their world before the researcher arrives. Sociology interprets an interpretation.

From which follows Schutz's postulate of adequacy: a scientific account of an action should be one that the actor could, in principle, recognise as an account of what they were doing. Not that they would agree with the explanation — that they would recognise the description.

Peter Winch (1926–1997) took the argument much further. In The Idea of a Social Science (1958) he argued that social behaviour is rule-following, that understanding a rule is not the same as observing a regularity, and therefore that the relation between a social scientist's account and the behaviour is conceptual, not causal. On this view social science is much closer to philosophy or translation than to physics, and the search for causal laws is a category mistake.

This is the strong version, and it has a serious problem. If understanding a practice means grasping it entirely in its own terms, then you can never say that a society is wrong about itself — that its account of what it is doing is mistaken, or ideological, or self-serving. The Marxist tradition, the Freudian tradition, feminist analysis and Bourdieu's whole project all depend on being able to say exactly that (see 5.8.1). A method that rules it out in advance has ruled out most of sociology.

The mistake to stop making right now: interpretivism is not "qualitative" and positivism is not "quantitative".

This equation is in a great many textbooks and it is wrong in both directions.

Quantitative work can be interpretive. Content analysis counts meanings. Large cross-national values surveys are attempts to measure how people understand their world. Conversation analysis is fiercely systematic and produces counts.

Qualitative work can be thoroughly positivist in spirit. A great deal of interview research treats what respondents say as a report on facts — what happened, how often, what caused what — with the meaning treated as data rather than as the object. Comparative-historical work (see 7.5.4) uses cases the way an experimenter uses conditions.

Methods are tools. Philosophies are about what the tools are for. You can hold almost any of these positions and use almost any of these methods. Confusing the two produces sterile arguments in which people defend a technique when they mean to defend a purpose.

What happened to the middle

The natural sciences turned out not to work the way the positivists said either.

This is the development that reframed the whole quarrel, and it came from the philosophy of natural science.

Karl Popper attacked verification and put falsification in its place. No number of confirming instances proves a universal claim; one genuine counter-instance refutes it. So the mark of a scientific theory is not that it can be confirmed but that it forbids something — that it is testable, and could fail. A theory that explains every possible outcome explains nothing.

Popper was not a positivist and said so repeatedly — he thought he had killed logical positivism. Which makes it one of the odder facts in the discipline's history that when he debated the Frankfurt School in Germany in the 1960s, the exchange became known as the Positivismusstreit — the positivism dispute — with Popper cast as the positivist (see 5.4.1).

Thomas Kuhn did more damage. The Structure of Scientific Revolutions (1962) described science as it is actually practised: most of the time researchers work inside a paradigm — a framework of assumptions, exemplary problems and accepted methods — doing normal science, which is puzzle-solving, not testing the framework. Anomalies accumulate, are absorbed or ignored, and then at some point the framework breaks and is replaced. Scientists do not abandon a theory the moment it is falsified, because there is nothing to move to until there is.

Neither of these is an argument against studying society scientifically. They are arguments that the natural sciences were never the tidy machine that positivism held up as the model. The gap between the sciences of nature and the sciences of society is narrower than it looked — but it narrowed from the other end.

Critical realism: the most useful middle position on offer.

Roy Bhaskar's argument, developed in the 1970s, starts from a question about experiments: why do scientists need laboratories?

Because outside them, causes do not produce regular results. Gravity does not stop working when a leaf falls unpredictably; it is that many mechanisms are operating at once. The laboratory is a device for artificially closing a system so that one mechanism can show itself.

From which Bhaskar draws a picture of reality in three layers:

The empirical — what is experienced and observed. The actual — the events that happen, whether or not anybody observes them. The real — the underlying structures and mechanisms with the power to produce events.

Science is the search for mechanisms. Not for regularities: regularities are what mechanisms produce in closed systems, and society is never a closed system. In an open system, mechanisms operate as tendencies, and several operate at once, sometimes cancelling each other.

This buys sociology something valuable and gives up something real. It explains why the discipline has almost no laws and yet is not therefore failing: with many mechanisms operating in an open system, stable constant conjunctions are not to be expected, and their absence is not evidence that there is nothing there. It also explains why prediction is much harder than explanation in social life — you can often show convincingly why something happened and still not have been able to say in advance that it would.

What it gives up is the simple test. If your theory does not have to produce regularities, it becomes much easier to protect it from evidence, and critical realism has been criticised — fairly — for producing more philosophy than research.

Bhaskar's own three-part slogan is the useful part: ontological realism (there is a world independent of our beliefs about it), epistemic relativism (all our knowledge of it is produced socially, in some language, from some position), and judgemental rationality (nevertheless, we can have good reasons for preferring one account to another). The third clause is what stops the second from becoming "anything goes".

Standpoint epistemology: knowledge from somewhere.

A different challenge to the positivist picture came from feminist philosophers of science, and it is not the one it is usually taken to be.

The positivist ideal of objectivity is a view from nowhere: the observer's identity, position and interests should make no difference to the result. Donna Haraway called this "the god trick" — the claim to see everything from nowhere — and argued that all knowledge is situated, produced by particular people with particular equipment from particular positions.

Sandra Harding pushed further, and her move is the interesting one. She argued that the standard ideal is not too objective but not objective enough — she called it weak objectivity. It scrutinises the evidence and the method, but exempts from scrutiny the background assumptions of the community doing the research, which are precisely the assumptions least likely to be visible from inside. Strong objectivity requires putting the researcher and the research community into the frame as objects of examination too.

The standpoint claim proper is that starting research from the lives of subordinated groups can produce less partial knowledge, because people who have to navigate a system they did not design generally have to understand it better than those who did.

Two things are routinely got wrong about this. First, a standpoint is achieved, not given — Harding and Hartsock both insist it is a product of critical reflection and shared struggle, not something you automatically have by belonging to a category. Second, the claim is about the starting point of research, not about who is allowed to speak.

And the real difficulty is unresolved: when standpoints conflict — and they do, including among the subordinated — the theory does not by itself say how to adjudicate. Which pushes you back towards needing shared standards of evidence, and therefore back towards the middle (see 6.8.1 on the symmetry criterion).

Pragmatism's contribution is the shortest and the most usable.

The American pragmatists — Peirce, James, Dewey, and in sociology the Chicago tradition (see 5.3.1) — declined the whole question. Their position: an idea's meaning is what difference it makes in practice, and a method's value is whether it answers the question you have.

Peirce added a third form of inference to the usual two. Deduction moves from a general rule to a case. Induction moves from cases to a rule. Abduction moves from a surprising observation to the best available explanation of it — what would have to be true for this to make sense? It is not certain, and it is what researchers actually do most of the time.

The pragmatist rule of thumb, and it will serve you well: fit the method to the question, not the question to the method. If you want to know how common something is, count. If you want to know what it is, go and look at it closely. If you want to know both, do both, and be honest about which part rests on which (see 7.8.3).

What this argument decides in practice

Four things follow from where you stand, and they are all practical.

One — what counts as a good explanation. A positivist explanation identifies the conditions under which an outcome reliably occurs. An interpretive explanation shows what the outcome meant to the people involved, such that they acted as they did. A realist explanation identifies a mechanism and shows how it operated in this case. Three different bars, and a study cleared for one will look weak against another.

Two — what you do with what people say. For a positivist, a respondent's account is data about the world, to be checked against other evidence. For an interpretivist it is data about the respondent's world, and its accuracy is a different question from its meaning. A great deal of methodological argument is really this disagreement, unstated.

Three — how much your own position matters. Standard research reports write the researcher out. Interpretive and standpoint traditions require writing them back in — not as confession but because the researcher's position shaped what they could see (see 7.1.2).

Four — what to do when a finding does not replicate. If you expect law-like regularities, failure to replicate is a scandal. If you expect mechanisms operating as tendencies in open systems, it is expected, and the question becomes what was different about the second context — which is a research question rather than an embarrassment (see 7.8.2).

Three ways this argument goes wrong.

One — using "positivist" as an insult. It has largely stopped denoting a position and become a way of dismissing quantitative work without engaging with it. If you find yourself about to use the word, try stating the actual objection instead: this measures the wrong thing, or this treats a category as natural, or this cannot support a causal claim. Those are arguments. The label is not.

Two — the slide from situated to arbitrary. "All knowledge is produced from a position" is true and important. "Therefore all accounts are equally good" does not follow, and nobody who advances the first claim seriously believes the second — they are, after all, arguing that their account is better. Bhaskar's judgemental rationality is the name for the missing step; Harding's strong objectivity is another route to it. Do not let anyone take the first clause and drop the third.

Three — treating the choice as a personality. Researchers acquire an identity — "I'm a qualitative person" — and then select questions that suit the identity. This is backwards, and it quietly determines what the discipline knows: questions that no one's identity fits go unasked.

Because it decides what a study is allowed to claim, and that is the thing you most need to be able to judge.

Go back to the two researchers. The first can tell you that scholarships reduce dropout by a certain amount, with a certain confidence, for a certain population. She cannot tell you why, and if she offers a why, she is speculating — respectably, but speculating.

The second can tell you what dropping out consists of, from the inside, and he has probably identified the mechanism. He cannot tell you how common it is, and if he offers a prevalence, he is guessing from thirty people.

Almost every bad reading of research is a failure to notice which of these you are holding. A newspaper reports an ethnography as though it established a rate. A policy document reports a regression as though it established a reason. A commentator dismisses a survey because "you can't reduce people to numbers", or dismisses an ethnography because "it's only thirty people" — each demanding of a study something it never claimed to do.

The whole of Part 7 is training in that one discrimination: what did this study actually establish, and what is being read into it.

Rocks do not mean anything; people do, and their meanings are part of what they are doing. That is the fact the whole argument sits on.

Positivism holds that social science is continuous with natural science: observable evidence, regularities, explanation by subsumption, fact separated from value. Comte named it; Durkheim demonstrated it on suicide and objected to the label; logical positivism was a stricter doctrine that collapsed under its own verification principle and that sociologists almost never held.

Interpretivism holds that meaning changes the kind of knowledge available. Dilthey split erklären from verstehen; Schutz showed that sociology's concepts are constructs of the actors' constructs; Winch's strong version makes it impossible to say a society is wrong about itself, which is too high a price.

Weber wanted both — a sociological account must be adequate at the level of meaning and at the level of cause, and either alone is a failure.

The middle exists because the natural sciences turned out not to be positivist either. Popper replaced verification with falsification; Kuhn showed science working inside paradigms rather than testing them. Critical realism offers a stratified reality of mechanisms operating as tendencies in open systems — which explains the absence of social laws without concluding there is nothing there — held together by ontological realism, epistemic relativism and judgemental rationality. Standpoint epistemology adds that objectivity as usually practised is too weak, because it exempts the research community's own assumptions from scrutiny. Pragmatism says fit the method to the question.

And positivist does not mean quantitative. Methods are tools; these positions are about what the tools are for.

Positivism — the position that social science shares the logic of natural science: observation, measurable regularities, explanation by subsumption under general patterns.

Naturalism — the specific claim that social and natural science are continuous in method.

Verification principle — the logical positivists' rule that a statement is meaningful only if empirically verifiable or true by definition; self-defeating, and abandoned.

Interpretivism — the position that social action must be understood through the meanings it has for those performing it.

Verstehen — interpretive understanding; reconstructing an action's meaning within the actor's frame of reference. A method, not empathy.

Erklären / Verstehen — Dilthey's distinction between explaining an event by law and understanding it by meaning.

Nomothetic / idiographic — Windelband's distinction between seeking general laws and characterising the individual case.

Adequacy at the level of meaning / causal adequacy — Weber's two requirements; an account must satisfy both.

Typification — Schutz's term for the ready-made types through which people encounter the social world.

Second-order constructs — the researcher's concepts, built on the concepts the actors already use.

Falsification — Popper's criterion: a scientific claim must forbid something and could therefore fail.

Paradigm / normal science — Kuhn's terms for the framework researchers work inside, and the puzzle-solving they do there.

Critical realism — Bhaskar's position: a stratified reality (empirical, actual, real) in which mechanisms operate as tendencies in open systems.

Ontological realism / epistemic relativism / judgemental rationality — the world is independent of our beliefs; our knowledge of it is socially produced; we can still have good reasons to prefer one account.

Situated knowledge — Haraway's term; all knowledge is produced from a particular position, against "the god trick".

Strong objectivity — Harding's demand that the research community's own assumptions be examined too.

Standpoint — a critically achieved perspective, arising from a subordinated position, argued to yield less partial knowledge.

Abduction — inference from a surprising observation to the best available explanation.

One — sort your own instincts. Read a news report of a study. Before you evaluate it, ask: does it claim how much or what it is like? Then ask whether the evidence described could support the other kind of claim. It usually cannot.

Two — apply Weber's double test. Take any explanation you believe about why people behave a certain way. Is it adequate at the level of meaning — could a person have meant to do that, for those reasons? Is it adequate at the level of cause — what evidence is there that people in that situation generally do? Most popular explanations fail the second.

Three — find the mechanism. Take a correlation you have seen reported. Write down two different mechanisms that would produce it, and one thing that would distinguish between them. This is critical realism as a daily exercise.

Four — practise strong objectivity on yourself. Pick a question you would like to research. Write down three assumptions you hold that would be invisible to you if everyone you discussed it with shared them.

Five — run the two-researcher test on something. Choose any social question you care about — why people leave jobs, why a neighbourhood changed, why a rule is ignored. Sketch how each of the two researchers in the story would study it, and write down one thing each would find that the other could not.