Ethnographic Methods

Two notebooks in one tent

Through the winter of 1917 and into 1918, in a tent on the lagoon side of Kiriwina in the Trobriand Islands, Bronislaw Malinowski kept two kinds of notebook. One held village censuses, garden magic formulae, genealogies, Kiriwinian texts taken down as they were spoken. The other was a personal diary, written in Polish, recording seasickness, lechery, novels he wished he had, quarrels with traders, and a running commentary on how much the people around him annoyed him that day. The first set of notebooks became the discipline's model of what fieldwork produces. The second was published in 1967, twenty-five years after his death, and made a good number of anthropologists wish it had stayed in the box.

Both notebooks matter here. Fieldwork asks a person to be the instrument, and instruments have moods, politics and preferences; the discipline spent its first fifty years pretending otherwise and its next fifty admitting it.

What you could learn by post

Before roughly 1900 the people who wrote the general books about human societies did not visit them. Edward Burnett Tylor in Oxford and James George Frazer in Cambridge built vast comparative works out of other people's reports: missionary memoirs, colonial officers' notes, travellers' accounts, replies to questionnaires. Frazer's The Golden Bough went through three editions and twelve volumes without its author spending a night in a village he wrote about. This is what later generations mocked as armchair anthropology, and the mockery is a little too easy.

The armchair had a real instrument. Notes and Queries on Anthropology, first issued by the British Association for the Advancement of Science in 1874, was a field manual for people already abroad for other reasons: it told a district officer what to measure, which questions about descent to ask, how to record a vocabulary. Six editions later it was still in use in the mid twentieth century, and comparison across the resulting pile was the whole intellectual point.

Here is the flaw, and it is not that the armchair theorists were lazy. It is that a questionnaire can only return answers to questions somebody already thought to ask. If your form has a line for "marriage payment" and the society you are asking about moves valuables in a long chain of gifts with no single moment of payment, the reply comes back either blank or wrong, and you cannot tell which from Oxford. Worse, the person filling in the form is a missionary who wants converts or an officer who wants taxes, and neither is neutral about what the answer should be.

Key idea: the deep problem with the armchair was not distance. It was that the categories were fixed before the encounter, so the data could never talk back.

Torres Strait, 1898: a survey that stayed a little longer

The Cambridge Anthropological Expedition to Torres Straits, led by Alfred Cort Haddon, is the usual hinge in this story. Haddon, trained as a zoologist, took a team to the islands between Australia and New Guinea: W. H. R. Rivers, Charles Seligman, Sidney Ray the linguist, William McDougall and Charles Myers on psychology, and Anthony Wilkin. They stayed months, not years, and they moved between islands. They ran experiments on visual acuity and colour naming with equipment carried from Cambridge.

Two things came out of it that matter to you. First, Rivers invented the genealogical method there, which you will learn to run in Module 3: instead of asking abstract questions about kinship rules, he collected actual pedigrees, name by name, and then read the rules off the pedigrees. Second, Rivers argued in print, in 1913, that the future of the work lay in what he called intensive study: one worker, one small area, a year or more, the local language, and personal contact with every person in the community. He named the method before anybody had really done it.

Malinowski became the founding figure rather than Rivers because the Torres Strait team still worked in survey mode, because he wrote the manifesto and trained students at the London School of Economics, and because of the accident of a war.

An enemy alien with a tent

Malinowski, a subject of the Austro-Hungarian Empire, was in Australia when the First World War broke out in August 1914. Technically he was an enemy alien in British territory. Australian authorities let him carry on with fieldwork in the Territory of Papua rather than intern him, which is how a man who had planned a short study ended up with three long field periods: Mailu in 1914 and 1915, then the Trobriands from June 1915, and again from late 1917 into 1918. Long residence began as an administrative accident and was afterwards written up as a method.

The programme appears in the introduction to Argonauts of the Western Pacific (1922), and it is still the clearest short statement of what an ethnographer is trying to collect. There are three strands, and they are not the same kind of thing:

1. The skeleton. The organisation of the society, documented concretely: censuses, genealogies, maps of gardens and villages, tables of who did what in an exchange. The framework written down as evidence a reader could check, not as an author's judgement.
2. The imponderabilia of actual life. The texture nobody thinks worth reporting: how people eat, the tone of a joke, who sits where, what happens on a boring afternoon. Only presence catches it, recorded in a diary of observation kept beside the formal materials.
3. The corpus inscriptionum. Statements, formulae and turns of phrase taken down verbatim in the local language, so the reader meets the people's own categories rather than the ethnographer's translation of them.

Put together, the aim is the sentence everybody quotes: to grasp, as Malinowski put it, "the native's point of view, his relation to life". The phrasing has aged; the ambition has not.

Boas, and a different answer to the same problem

Fieldwork was not invented once. While Malinowski was in the Pacific, Franz Boas had been building an American version for thirty years out of a different starting point. Boas went to Baffin Island in 1883 to study seawater colour and the migration of Inuit families and came back convinced that geography did not determine culture. From 1886 he worked on the Northwest Coast, above all with the Kwakwaka'wakw, and there he did something Malinowski never did: he trained a collaborator, George Hunt, to write.

Hunt, whose mother was Tlingit and whose father was English, wrote thousands of pages of Kwak'wala texts, in the Kwak'wala language, and Boas edited and published them. Much of what is credited to Boas is a joint product of a fieldworker and a literate insider being paid by the page. That relationship, its unequal credit and its editorial control, is now a whole research literature of its own.

The two schools bequeathed different instincts. Malinowski's asks how the parts of a living society fit together now, and prefers one deep case. The Boasian tradition asks how this particular history produced this particular culture, distrusts grand comparison and pours effort into texts and language: historical particularism. Boas also built the instinct that became cultural relativism: judge a practice by the standards of its own system before judging it by yours.

What the kula found

Take one finding to see the method earn its keep. The kula is a system of ceremonial exchange linking a ring of islands off eastern New Guinea. Red shell necklaces travel one way around the ring, white shell armbands the other. A man gives a necklace to a partner overseas and receives an armband in return, later, with no bargaining and no fixed rate. The objects are not consumed, not worn much, and not sold. They circulate for decades, gathering names and histories.

A questionnaire would have destroyed this. Asked "what is the currency, and what is the rate of exchange?", an informant either answers in the terms the form supplies, which are wrong, or says something that gets recorded as confused. Malinowski could see instead that the voyaging, the magic, the canoe building, the partnerships and the informal trade in yams and pots that travelled alongside the valuables were one institution, and that the kula was the political spine of a region with no state. Marcel Mauss then took the material, put it beside the Maori hau and the Northwest Coast potlatch, and wrote The Gift in 1925. The intensive case fed the comparison; the armchair could not have generated the case.

What the tent could not see

Now the other half of the ledger, and this is the habit this course wants you to build: every method has a shape, and the shape has a shadow.

The Trobriand monographs are strikingly quiet about the colonial situation that put Malinowski there. There were traders, missionaries, a resident magistrate, plantation labour recruiters and pearl buyers in the islands. His personal diary is full of them: he ate with them, borrowed their boats, was bored by them. His ethnography treats Trobriand society as a self-contained working system, described in a timeless present tense as though the events of 1918 were what always happens. That habit acquired a name, the ethnographic present, and it is a specific analytic cost: it hides change and it hides the outside world.

The sharper omission was found by returning. Annette Weiner went to Kiriwina in 1971 and 1972, to the same villages, and studied what women did with wealth. Bundles of dried banana leaves and fibre skirts, made by women and distributed in enormous quantities at mortuary ceremonies, turned out to be a wealth system that ran parallel to the men's kula and was necessary to it: the bundles pay out the claims of the dead person's relatives, and a man's standing depends on his wife's ability to produce them. Malinowski had seen the bundles. He described them as trivia. He was a man doing fieldwork in a society where men and women lived substantially separate working lives, and the sampling consequence of his own body was invisible to him.

The upshot: the strongest critique of a piece of fieldwork usually comes from more fieldwork, not from theory. Weiner did not argue that Malinowski was biased. She went back and counted the bundles.

The diary, published

In 1967 Malinowski's widow, Valetta Malinowska, published the Polish diaries in translation as A Diary in the Strict Sense of the Term, with an introduction by Raymond Firth, one of his own students. The book showed a man frequently sick, frequently bored, dosing himself with arsenic compounds, resentful of the people he lived among, referring to them with a racial epithet, and at one low moment quoting Kurtz from Conrad's Heart of Darkness: "Exterminate the brutes."

Reaction split three ways, and all three positions are still live. Some read it as a betrayal of a dead man's private paper. Some read it as proof that ethnographic authority was a pose: if this was the man behind the impersonal prose, why believe the prose? Clifford Geertz, writing in 1988, took the more useful line. The diary does not falsify Argonauts. It shows the gap between the person doing the work and the narrator constructed to report it, and that gap is a permanent feature of the genre rather than Malinowski's private failing.

Treat it as a methodological document: evidence about the working conditions of participant observation, from exhaustion and dependence on a few English speakers to the daily labour of seeming more patient than you feel. None of that appears in the monograph, and all of it shaped the data.

Common misconceptions

• Malinowski invented fieldwork. He codified and publicised it. Rivers named intensive study in 1913, Boas had been in the field since 1883, and missionaries and colonial officers had lived in communities for decades, without the analytic framework that made residence into evidence.
• The armchair theorists had no data. They had enormous quantities of it. What they lacked was any way for the data to contradict the categories the questionnaire imposed.
• The diary discredits the ethnography. It complicates the persona, not the census figures. If you want to challenge the Trobriand material you must do what Weiner did and go back.

Looking back

Fieldwork became the discipline's method when three things arrived together: long residence, work in the local language, and a written charter saying what residence was supposed to produce. Rivers supplied the prescription, Boas the linguistic seriousness and the collaborator, Malinowski the manifesto and the students. Carry the charter's three parts permanently, because everything in this course hangs on one of them: the documented skeleton, the imponderabilia, and what people said in their own words.

Worth holding on to: the method's founding case is also its founding demonstration of blind spots. Two years in the Trobriands produced the kula and missed the banana-leaf bundles, and it took another fieldworker, differently positioned, to see them.

Sources

1. Malinowski, B. (1922). Argonauts of the Western Pacific: An Account of Native Enterprise and Adventure in the Archipelagoes of Melanesian New Guinea. Routledge and Kegan Paul. archive.org
2. Boas, F. (1911). The Mind of Primitive Man. Macmillan. archive.org
3. Malinowski, B. (1967). A Diary in the Strict Sense of the Term (N. Guterman, Trans.; introduction by R. Firth). Harcourt, Brace and World.
4. Weiner, A. B. (1976). Women of Value, Men of Renown: New Perspectives in Trobriand Exchange. University of Texas Press.
5. Stocking, G. W. (1983). The ethnographer's magic: Fieldwork in British anthropology from Tylor to Malinowski. In G. W. Stocking (Ed.), Observers Observed: Essays on Ethnographic Fieldwork (pp. 70-120). University of Wisconsin Press.

A clipboard that lasted one afternoon

In 1989 a first-year graduate student at the University of Chicago walked into the Robert Taylor Homes carrying a survey his department had given him. One item asked how it felt to be black and poor. The response options ran from very bad to very good. Sudhir Venkatesh put the question to a group of young men in a stairwell of a high-rise controlled by a gang, was held there for the night while they worked out what he was, and was eventually told, more or less, that the question was ridiculous and he should come back and just watch. He did, for years, and the resulting work made his career.

The instrument failed for the same reason the questionnaires of the 1880s failed. Somebody who had never been in that stairwell decided in advance what the possible answers were. But do not draw the romantic conclusion that design is the enemy of fieldwork. Venkatesh spent the next several years making decisions about where to be, whom to follow, what to count and what to promise, and those decisions are a design whether or not anybody writes them down. Writing them down is better, because a design on paper can be criticised before it wastes a year.

Before you read on: write your current research interest in one sentence, on paper, now. Keep it in front of you. By the end of the lesson you will have taken it through four transformations, and the first version is usually unrecognisable by the end.

Topic, puzzle, question

A topic is a region of interest: eviction, midwifery, esports, land titling. It is not a question, and you cannot answer it. Between the topic and a workable question sit two moves.

The first move is to find the puzzle: something in the topic that is not what you would expect, or where two reasonable expectations collide. Matthew Desmond began with a topic, poverty in American cities, and found a puzzle in the housing court records: eviction was being treated in the literature as a symptom of poverty, a thing that happens to people who are already poor, when the sequence in the records suggested it might also be a cause, an event that makes people poorer, sicker and more likely to lose jobs. A puzzle has that shape. It says: the obvious ordering may be backwards, or two things that should not coexist do.

The second move is to turn the puzzle into a question with four properties. It must be empirically answerable, which rules out asking whether landlords are bad people. It must be bounded, which means a site and a period, not a civilisation. It must be capable of coming out otherwise, so that some findable state of the world would count as a no. And it must be worth asking after somebody has already looked: check the Human Relations Area Files and the regional literature before you decide your question is new.

Watch a topic go through the mill:

1. Topic. Eviction in poor neighbourhoods.
2. Puzzle. Eviction is treated as an outcome of poverty, but tenants describe it as the thing that started the collapse.
3. First question. What effect does eviction have on poor families? Still too big, and not answerable by one person watching.
4. Researchable question. In two Milwaukee neighbourhoods over eighteen months, how do landlords decide whom to evict, how do tenants absorb the loss of a home, and what does each side's account leave out of the other's?

Notice what the last version buys. It names a period. It requires two populations, which is a design decision with teeth: Desmond studied landlords as well as tenants, and the landlords are why the book can say anything about how evictions are produced rather than only how they are suffered. It also states an ignorance, which is the part most proposals omit.

The point: a good question tells you what to do on Tuesday morning. If your question does not imply a place to stand and a person to talk to, it is still a topic.

Three questions that will waste your year

• The normative one. Should traditional healers be integrated into the national health system? No amount of fieldwork settles a should. Convert it: how do clinic staff and healers in one district actually refer patients to each other, and what happens to patients who move between them?
• The settled one. Do migrants send money home? Yes, extensively documented since the 1970s. Convert it: which household members claim a right to name what remittances are spent on, and how is that claim enforced at a distance?
• The unfalsifiable one. Is the community resilient? Resilient is not observable; it is a verdict. Convert it to a mechanism: when the cannery closed, which households moved and which stayed, and what did the stayers live on in the first two years?

Choosing a site: the logic of the case

Once you have a question, the site has to earn its place. Four justifications are respectable, and the fifth is not.

• Typical. The site is unremarkable in the relevant respects, so what you find plausibly holds elsewhere. You owe the reader the numbers that establish typicality.
• Extreme. The process you study runs hot here, so it is visible. Eviction rates in Milwaukee in 2009 made the mechanism legible in a way a low-turnover suburb never would.
• Revelatory. Nobody has been able to get in before, and access is itself the contribution.
• Contrastive. Two sites chosen so that one factor differs. Desmond took a mostly white trailer park on the South Side and a rooming house and rented flats on the mostly Black North Side, which lets race and neighbourhood do analytic work instead of sitting in the background.
• Convenient. Your cousin lives there. This is not a justification, though it is often the true reason. If convenience is doing the work, say so in the proposal and state what it costs the argument.

Prior relationship is a legitimate reason with a different name. Nancy Scheper-Hughes returned in 1982 to a hillside shantytown in north-east Brazil where she had worked as a young health volunteer in the 1960s. Twenty years of relationships bought her a depth of access no newcomer could have had; they also meant she arrived with obligations, an established role and a set of people who remembered her opinions. Both halves are design facts.

Feasibility: the arithmetic nobody writes down

Most failed projects fail on arithmetic, not on ideas. Work it out before you commit, in hours.

Two lines in that table decide most designs. The writing-up ratio is the reason a fieldworker who observes eight hours a day produces unusable notes: there is no time left to write them, and unwritten observation evaporates within about forty-eight hours. And the transcription ratio is why an interview-heavy design needs either money or a much smaller sample than the proposal usually claims.

What the ethics board will ask, and what it will miss

In the United States a university institutional review board reviews human-subjects research under the federal regulations at 45 CFR 46, whose principles come from the Belmont Report of 1979: respect for persons, beneficence, justice. Other countries have research ethics committees with similar structures and different paperwork.

The system was built around clinical trials, and the fit with fieldwork is poor in specific, predictable ways. A trial knows its subjects in advance; you will meet yours by walking around. A trial has an intervention with a defined risk; your risk is mostly the disclosure of information. A trial ends; your relationships continue after the funding stops.

Three practical consequences. First, most ethnographic protocols qualify as exempt or expedited under the revised Common Rule rather than needing full board review, but the determination is the board's to make, not yours. Second, ask about a waiver of documented consent where a signature is the main risk to the participant: for undocumented workers or members of a criminalised group, the consent form is the only piece of paper linking a name to the study, and oral consent with a witnessed log is safer. Third, write the protocol so that it describes real fieldwork. A protocol promising that every person observed in a public market will sign a form describes a study nobody could run, and a board that approves it has approved a fiction.

Approval is a floor. The board will not ask what happens to your friend when your book appears, whether the community wants the study at all, or who owns the recordings in ten years. Those questions belong to you, and they get a lesson of their own later in the course.

The proposal, and the test it has to pass

A working proposal runs to about ten pages and has eight parts: the question and the puzzle behind it; what is already known and the gap; the site and the case logic; methods, stated with numbers; a timeline by month; the ethics protocol and consent procedure; risks with mitigations, including the risk that access collapses; and what will be produced and for whom, including anything owed to the community.

The methods section is where proposals go vague, and vagueness there is a reliable predictor of a bad year. Compare two versions. Weak: "I will conduct participant observation and interviews with key informants." Strong: "Three days a week at the depot across the three shifts, from March to November; a freelist of job categories with 30 workers in month two; 25 semi-structured interviews sampled to cover both shifts and both contract types; a time-allocation sample of 200 spot observations; all notes expanded within 24 hours."

The test to apply before you send it: can a hostile reader say what result would surprise you? If nothing in the design could produce an unwelcome answer, you have written a plan to illustrate a conclusion, and reviewers who have seen a hundred of these will spot it immediately.

Common misconceptions

• Ethnography is inductive, so a design would prejudge the findings. Design commits you to a site, a period and a set of methods, not to an answer. The alternative to an explicit design is an implicit one you cannot criticise.
• A hypothesis is unethnographic. A stated expectation is useful precisely because the field can contradict it, and a contradiction you can name is a finding.
• Ethics review is a bureaucratic formality to be got past. It is a floor, badly shaped for fieldwork, and it will not catch the dilemmas that actually arrive.
• You can decide the sample once you get there. You can revise it there; you cannot invent it there. Without a sampling logic on paper you will end up with the people who liked you, which is the commonest defect in fieldwork data.

Putting it together

Take the sentence you wrote at the start. Find its puzzle, then bound it by site and period, then check that some findable state of the world would count as a no. Justify the site by case logic and name the justification out loud. Do the arithmetic in hours before you promise anything, remembering that writing notes costs more time than making them. Draft the protocol so that it describes fieldwork as it will really be done, and reserve the harder ethical questions for yourself, because the board will not ask them.

Worth holding on to: Venkatesh's survey was not too rigid because it was designed. It was too rigid because its designer had never stood in the stairwell. Design as far ahead as you can, then let the site rewrite the parts it can prove wrong.

Sources

1. Bernard, H. R. (2017). Research Methods in Anthropology: Qualitative and Quantitative Approaches (6th ed.). Rowman and Littlefield.
2. Desmond, M. (2012). Eviction and the reproduction of urban poverty. American Journal of Sociology, 118(1), 88-133.
3. Venkatesh, S. (2008). Gang Leader for a Day: A Rogue Sociologist Takes to the Streets. Penguin Press. Wikipedia
4. National Commission for the Protection of Human Subjects of Biomedical and Behavioral Research. (1979). The Belmont Report. U.S. Department of Health and Human Services. hhs.gov
5. Scheper-Hughes, N. (1992). Death Without Weeping: The Violence of Everyday Life in Brazil. University of California Press.

A police raid solves a problem in ten minutes

In April 1958 Clifford Geertz and Hildred Geertz had been in a Balinese village for about ten days and were being treated, as Geertz later wrote, as though they were not there. People looked through them. Conversation stopped and restarted around them. Then they went to a cockfight in the village square, which was illegal, and the police arrived in a truck. Everybody ran, and the two anthropologists ran with them, into a compound where a man's wife pulled out a table, produced tea, and had all three of them sitting there deep in conversation by the time an officer came through the gate. She told him the visitors had been drinking tea and talking about culture all afternoon and had heard nothing about a cockfight. The officer left.

The next morning the village was open. People who had ignored them for a week and a half now wanted to tell them exactly how fast they had run, and to hear the story back, repeatedly. Geertz made this the opening of Deep Play: Notes on the Balinese Cockfight, published in Daedalus in 1972, and it became the most retold entry story in the discipline.

It is also useless as advice. You cannot schedule a police raid. If you strip the drama out, though, the incident did four specific things, and each of them can be produced deliberately.

What the raid actually did

1. It put them on a side. They ran with the villagers rather than standing still and showing their papers, which any foreigner with a research permit could safely have done. The choice was legible instantly.
2. It made them vulnerable. A protected outsider became someone who could be in trouble, and who could be protected by a villager's lie. Reciprocity had a direction for the first time, and it ran towards them.
3. It gave the village a story about them. Communities need a way to talk about a stranger. Before the raid there was no story except suspicion. Afterwards there was a comic one everybody could tell.
4. It was witnessed. Access is a public fact, not a private agreement. Dozens of people saw it happen.

So the practical question is not how to be lucky. It is how, in the first weeks, to become someone with a side, a vulnerability, a story and witnesses.

Two kinds of gatekeeper, two kinds of trap

A gatekeeper is anyone whose refusal ends your project. Formal gatekeepers hold offices: the ministry that issues research clearance, the hospital director, the union branch, the head teacher, the platform's legal department. Informal gatekeepers hold standing: the man everyone consults before agreeing to anything, the shift supervisor, the woman who runs the burial society.

The formal trap is thinking permission is consent. A chief, a director or a village council may sincerely grant access on behalf of everyone below them, and the people below them may feel unable to refuse afterwards. Permission from above obliges you to work harder at individual consent, not less, because your presence now carries an implied instruction from someone with power over the people you are studying.

The informal trap is subtler and worse. Whoever befriends you first will be someone with a reason to befriend a stranger, and that reason is usually a position in a local conflict. In taking their help you take their enemies.

Doc, Herman, and the sponsor problem

The founding case is William Foote Whyte in the Italian North End of Boston in the late 1930s. After a couple of comic failures, including walking into a bar with a vague plan to talk to somebody, a social worker introduced him to a man he called Doc, the leader of a street corner gang. Doc took him everywhere, vouched for him and told him, in a line every fieldworker should memorise, that if he asked questions people would clam up, whereas if he hung around they would explain things without being asked. Street Corner Society (1943) came out of that friendship.

Doc also read the drafts. He shaped what Whyte thought was going on, and he was the source of a great deal of the interpretation as well as the access. In 1992 a Dutch sociologist, W. A. Marianne Boelen, published a restudy after interviews with surviving North End residents, several of whom disputed Whyte's picture. Her account has itself been contested, but the general point survives: a study built on one sponsor inherits that sponsor's angle of view, and there is no way to detect this from inside the study.

Elijah Anderson hit the same structure at a South Side Chicago bar in the 1970s. His entry came through one regular, and his eventual analysis of the bar's status order, the categories of respectable, non-respectable and regular, is visible from the position that man's friendship gave him. He could describe the world of the corner because someone on the corner adopted him.

What matters here: a sponsor is not a neutral door. Plan from day one to acquire a second and third sponsor in different parts of the social map, even at the cost of appearing less loyal to the first.

The role you get whether you want it or not

You do not introduce yourself into an empty space. The community already has a set of roles for outsiders, and you will be sorted into one within days: government spy, missionary, journalist, tax inspector, aid worker, buyer, rich fool, student, or a person whose family has failed to arrange a marriage for them.

Philippe Bourgois, who moved into East Harlem in 1985 to study a crack-selling network, spent months being read as either an undercover officer or an addict, and was stopped by police who could not construct any other account of a white man on that block at that hour. His eventual position, the harmless obsessive who writes everything down and lives here with his family, was a role the neighbourhood built for him and then explained to newcomers on his behalf.

You cannot choose your role, but you can supply materials for it. Three things help. Be able to say what you are doing in one sentence, in the local language, in words a twelve-year-old would understand, and say the same sentence to everyone. Be visibly bad at something and let people teach you: language, cooking, repairing a net, sorting recycling. Be boringly consistent about where you are and when, so that your presence stops being an event.

The first six weeks, in order

1. Week 1. Formal permissions in hand, courtesy calls to every office you have been told matters, plus the ones you have not. Walk the whole site. Draw a map. Say your sentence a hundred times.
2. Week 2. Census what you can see: households, shops, shifts, stalls, machines. This is Malinowski's skeleton, and it doubles as a reason to speak to everyone once without an agenda.
3. Weeks 3-4. Establish a routine that puts you in a public place at a fixed time. Take no notes in front of anyone yet unless you have already asked. Start recording who talks to you and who conspicuously does not.
4. Week 5. Deliberately cross to a second network, ideally one your first friends mildly disapprove of. This is uncomfortable and it is the single highest-value move of the entry period.
5. Week 6. First formal interviews, with people who have already known you for a month. Ask permission to write while people talk, and notice who says yes with a hesitation.

Every week of that plan has a note-taking discipline attached, which is the subject of the next lesson. Entry is not a phase you complete; it repeats every time you reach a new part of the field.

Rapport is not friendship

Rapport is often described as though the goal were to be loved. The working definition is duller and more useful: enough trust that people behave in your presence roughly as they would in your absence, and tolerate the odd questions. That can coexist with being found irritating.

The classic failure mode has an unfortunate old name, going native, and it describes the loss of the doubled position that makes fieldwork productive: you stop noticing, because nothing is strange any more, and you stop writing, because everything feels obvious. Michael Agar's phrase for the alternative, the professional stranger, catches the required awkwardness. You are permanently a person who asks about things everyone else takes for granted.

The opposite failure is rarer in books and common in practice: staying so protected that you only ever meet people in arranged settings. If everything you know arrived by appointment, you have conducted an interview study with a long commute.

What Deep Play leaves out

Now turn the essay round, because the way it is usually taught hides three things.

First, the institutional frame. The Geertzes were in Bali as part of a funded American research project, with permits, affiliations and a national context in which foreign scholars were legible objects. The story reads as two vulnerable strangers taken in by villagers, and that is true at the scale of the afternoon and misleading at the scale of the year.

Second, the missing voices. Vincent Crapanzano, writing in 1986, pointed out that the essay's Balinese never speak. Geertz tells you what the cockfight means to them, in confident detail, using a reading of symbols rather than any recorded statement by a Balinese man about his own bird. The interpretation may be right. Nothing in the text lets you check it.

Third, and heaviest, the politics. The essay presents cockfighting as status play whose stakes are symbolic, a dramatisation of rivalry with no real consequence. Seven years after that raid, in 1965 and 1966, Bali suffered some of the worst of the Indonesian mass killings, with tens of thousands dead in an island of under two million people, neighbours killing neighbours along lines of faction and party. Geoffrey Robinson's 1995 history argued that the picture of a harmonious, aestheticised Bali, to which anthropology had contributed, had made that violence appear inexplicable when it was in fact the outcome of decades of documented political conflict. A study can be observant about symbols and blind about power at the same time, and it will not feel blind while you are writing it.

Common misconceptions

• Once you have official permission you have access. Permission removes an obstacle. Access is granted person by person and can be withdrawn silently, by people simply not being where they said they would be.
• Good rapport means people like you. It means they act normally around you. Being liked by one faction can cost you the rest of the site.
• The first weeks are wasted time before the real data starts. The entry period is when you can still see the setting as strange, and those notes are irreplaceable. Write hard while you are still confused.
• An entry story like the cockfight shows a method working. It shows a windfall, narrated by its beneficiary, in a text that gives no one else a voice.

Where this leaves us

Getting in is not one event. It is a sequence: formal permission that binds nobody, a sponsor who brings a faction with them, a role the community assigns you out of its own stock, and a slow accumulation of ordinary presence until you stop being news. Plan for the sponsor problem by acquiring rivals of your first friend. Plan for the role by giving people better material than the rumour they would otherwise use. Expect to redo all of it whenever you move to a new corner of the field.

In short: the cockfight story is worth teaching precisely because it is untypical. Almost nobody gets a raid. Everybody gets six slow weeks, and what you do in them decides which half of the site will ever talk to you.

Sources

1. Geertz, C. (1972). Deep play: Notes on the Balinese cockfight. Daedalus, 101(1), 1-37. Wikipedia
2. Whyte, W. F. (1943). Street Corner Society: The Social Structure of an Italian Slum. University of Chicago Press. Wikipedia
3. Crapanzano, V. (1986). Hermes' dilemma: The masking of subversion in ethnographic description. In J. Clifford and G. E. Marcus (Eds.), Writing Culture: The Poetics and Politics of Ethnography (pp. 51-76). University of California Press.
4. Bourgois, P. (1995). In Search of Respect: Selling Crack in El Barrio. Cambridge University Press.
5. Robinson, G. (1995). The Dark Side of Paradise: Political Violence in Bali. Cornell University Press.

Nine words in a pocket notebook

Here is a jotting, of the kind that fits on half a line while you are standing up: 9:40 blue truck. R shouts at C re pallet. C laughs, walks off. "you always do this".

That scrap is worth almost nothing tomorrow and nothing at all next month, unless it is expanded the same day into a page of prose while the scene is still in your head. This lesson is about the machinery that turns the first into the second, and about the third thing, the memory that never gets written down at all and still does most of your analysis.

What participation buys that watching does not

In 1988 Loic Wacquant, a French graduate student in Chicago, walked into a boxing gym on the South Side intending to use it as an observation post on the neighbourhood. He ended up training there for three years, sparring, dieting, and finally fighting in the Golden Gloves. Body and Soul (2004) is built directly out of the notebooks he kept.

The finding could not have come any other way. Boxers cannot tell you how they learned to slip a punch, because the knowledge is not stored in a form that survives translation into speech; it is a set of dispositions built into the body by repetition, a habitus. Ask a fighter why he moved left and he will say it felt right. Train for a year and your own body starts to make the same decisions, and now you have something to describe.

What it missed is equally instructive. The gym is treated as a moral island, a place of discipline set against a disordered neighbourhood, and the neighbourhood itself stays out of focus: the wage labour, the households, the women who sustain fighters' careers. Deep participation buys unmatched access to one bounded practice and quietly narrows the frame to whatever that practice contains.

Four positions, and what each one costs

Raymond Gold set out the classic typology in 1958, and it still sorts most designs.

1. Complete participant. You are a member and your research role is concealed. Access is maximal, ethics are worst, and you cannot ask a naive question without breaking character.
2. Participant as observer. You take part fully and everyone knows why you are there. This is the standard modern position and Wacquant's.
3. Observer as participant. Brief involvement, mainly interviews and short visits. Cheap, and it produces the study that never gets past what people say about themselves.
4. Complete observer. No interaction: counting people through a gate, watching a public square. Useful for behaviour sampling, useless for meaning.

Most fieldwork moves between positions in a single day. What matters is knowing which one you were in when a given note was made, because it determines what the note can support.

Jottings: what goes in the pocket

Jottings are memory triggers written in the moment or at the first opportunity, and they obey different rules from prose. Write down verbatim phrases inside quotation marks, and only phrases you actually heard: the exact words are the most perishable thing in the scene and the most valuable later. Write numbers: how many people, what time, how much it cost, how long it took. Write the sequence, in order, even in fragments. Write where people were standing, because spatial arrangement encodes relationships and is impossible to recall accurately. Write the moment you were surprised, marked with a symbol you can find again.

Two things stay out. Your interpretation, which belongs in a memo where it can be dated and later disowned. And anything that would harm someone if the notebook were read by a police officer, a manager or a rival, because notebooks are lost, seized and borrowed. Use initials or invented names from the first day, keep the key elsewhere, and never write an address next to a name.

One day at the depot, worked all the way through

The following field example is invented, so that all three layers can be shown against the same ten minutes; the technique is exactly what you would use at a real site. Setting: a municipal recycling depot, second week, morning shift, sorting line.

Stage one, the jotting, written standing near the baler: the nine words at the top of this lesson.

Stage two, the expanded fieldnote, written that evening, in the past tense, first person, with description and quotation but no verdicts:

At 9:40 the blue truck backed into bay two, twenty minutes later than the schedule taped inside the office door. R, the shift supervisor, came out of the office as it stopped and shouted across the yard at C, who drives it, that the pallet of flattened board had been left on the wrong side again and the loader could not reach it. C did not shout back. He laughed, said "you always do this", and walked past R towards the canteen without turning his head. R stood for a moment, then went back inside. Two of the sorters on the line, both women in their fifties, watched this without stopping work, and one of them said something to the other that I could not hear. The line did not slow. Nobody moved the pallet for the rest of the morning; the loader driver reversed twice to get around it at 11:15.

Stage three, the analytic memo, written after the note and kept in a separate file, dated, and clearly speculative:

Memo, day 9. Third time this week an instruction from R has been answered with a joke rather than compliance or refusal. Laughing seems to be the available move for drivers, who are on a different contract from the sorters and cannot be disciplined by R directly. Watch whether sorters ever use it. If they do not, the joke is a privilege of the contract rather than a workplace style. Test: ask the loader driver why the pallet stayed there, and see whether the answer is about the pallet or about R.

Remember: the fieldnote contains the pallet, the words and the times. The memo contains the theory. Keeping them apart is what lets you change your mind six months later without contaminating the evidence you would need in order to notice you were wrong.

The four kinds of record, after Sanjek

Roger Sanjek, in the 1990 collection that made fieldnotes a subject in their own right, distinguished the kinds of paper that pile up in a field project. Keep them physically separate.

The log looks like bureaucracy and turns out to be the most quietly valuable of the five. Six months in, it is the only thing that can tell you that you have spent forty mornings with the day shift and four with the night shift, which is a sampling fact you will otherwise discover from a reviewer.

Then there are headnotes, Simon Ottenberg's term for the memories that never reach paper and that you nonetheless think with when you write. Headnotes are why an ethnographer can read their own notes years later and see far more than a stranger would. They are also why nobody else can fully use your notes, and why a lost notebook is not the whole loss and a lost year is.

The rules that make it work

• Twenty-four hours. Expand every jotting the same day. Memory for sequence and wording decays fast, and after a night's sleep you reconstruct rather than recall, which feels identical from the inside.
• Two to three hours per hour. Budget writing time at two to three times observation time. If you cannot, observe less. Eight hours in the field and no notes produces nothing but headnotes, which cannot be quoted or checked.
• Describe, do not summarise. Not "the meeting was tense" but who interrupted whom, who left early, what the room did when the director spoke. Verdicts are unusable as evidence; the details behind them can be re-read against a later idea.
• Write the ordinary. The temptation is to record only the dramatic. What later turns out to matter is usually the routine: what time people arrive, who greets whom first, what gets thrown away.

Being watched while you watch

The notebook is a visible object with a meaning of its own. Writing while someone talks marks their words as important, which can be a courtesy or a threat depending on where you are; in a workplace with a disciplinary process, a stranger writing during an argument reads as evidence-gathering. The Hawthorne effect gave a name to the reactivity of observed people, though the original studies at the Western Electric plant in the 1920s and 1930s turn out to support the idea much more weakly than the textbooks claim.

The practical response is not to hide. It is to make the notebook boring: visible, ordinary, opened in front of people, sometimes read aloud from, and occasionally shown to somebody who asks. Reactivity fades with familiarity, and familiarity is cheaper than concealment, which is both an ethical problem and a fragile plan.

Common misconceptions

• Good notes record everything. Nobody records everything. Notes are selective by necessity, so the discipline is to know your selection principle and to write the routine as well as the dramatic.
• Notes should be purely objective, with no interpretation. Interpretation is unavoidable and useful; the requirement is that it be separated into dated memos rather than mixed into description.
• A recorder makes notes unnecessary. Audio captures words and loses the room: who was present, where they stood, what was pointed at, what happened after. It also produces transcription hours you may not have.
• Typing notes at the scene is more efficient. Screens change how people behave near you and make it harder to look up. Paper first, keyboard later, is still the standard for a reason.

What to carry forward

The notebook is the instrument, and the instrument has three parts. Jottings preserve the perishable things: exact words, numbers, positions, the moment of surprise. Fieldnotes, written within the day at a cost of two to three hours per observed hour, turn triggers into evidence that can still be read by a stranger. Memos hold your thinking where it can be dated and disowned. Keep them apart, keep a log so you can audit your own coverage, and remember that headnotes are doing work you cannot see.

The upshot: Wacquant learned things in that gym that no interview could have produced, and he learned them because he wrote every night about a body that was slowly becoming a boxer's. The participation is only half the method. The writing is the other half, and it is the half people skip.

Sources

1. Emerson, R. M., Fretz, R. I., and Shaw, L. L. (2011). Writing Ethnographic Fieldnotes (2nd ed.). University of Chicago Press. University of Chicago Press
2. Sanjek, R. (Ed.). (1990). Fieldnotes: The Makings of Anthropology. Cornell University Press. Wikipedia
3. Gold, R. L. (1958). Roles in sociological field observations. Social Forces, 36(3), 217-223.
4. Wacquant, L. (2004). Body and Soul: Notebooks of an Apprentice Boxer. Oxford University Press.
5. Bernard, H. R. (2017). Research Methods in Anthropology: Qualitative and Quantitative Approaches (6th ed.). Rowman and Littlefield.

Forty minutes of "it is difficult sometimes"

A researcher sits down with a midwife and asks what the main challenges in her work are. The midwife says it is difficult sometimes, especially at night. The researcher asks what makes it difficult. The midwife says there are many things. The researcher asks for an example, and is told that every case is different. Forty minutes later the recording contains nothing that could not have been guessed before the meeting.

Nothing went wrong with rapport. The question was wrong. It asked for a summary judgement in the researcher's vocabulary, which obliges the informant to do the analysis, and she declined. Now try: "Tell me about the last birth you attended. Start when the call came." Ten minutes of continuous speech follow, containing the names of three things you had never heard of, a complaint about the ambulance rota, and the reason the night shift is different.

Five instruments, and what each returns

Read the last column before the others. Each form fails in a specific, predictable way, and a serious design uses two forms whose failures differ. Unstructured interviews early to find the categories, then structured tasks built out of those categories, is the standard and strongest combination in anthropology, because the fixed instrument is no longer made of your words.

Spradley's questions, and the flops of Seattle

James Spradley worked in the 1960s with men caught in the Seattle drunk-arrest cycle, repeatedly jailed for public drunkenness. He went in expecting to study alcoholism. What he found, by asking questions in the men's own terms, was an elaborate body of practical expertise about where to sleep, which they called making a flop. He collected more than a hundred named kinds of flop, arranged in a folk taxonomy: flops you paid for, flops in boxcars, flops in bottle gangs, flops that risked a roll, flops that risked arrest. The taxonomy reorganised the research question. These were not men without competence. They were men whose competence was entirely invisible from the clinic or the court.

His 1979 handbook turned the procedure into three kinds of question, and it is still the best scaffolding available.

• Descriptive questions get speech flowing in the informant's vocabulary. The grand tour: "Describe a typical day from waking up." The mini tour: "Walk me through everything that happens when a new load arrives." Example: "You mentioned a bad flop. Can you give me an example?" Experience: "What is the worst thing that has happened to you in a hotel?" Native language: "What would you call that? Would you ever say X?"
• Structural questions map the categories you have now heard. "You have mentioned three ways of getting a bed. Are there others? What kinds of flop are there?" Asked repeatedly, gently, until the list stops growing.
• Contrast questions find the dimensions that distinguish terms. "What is the difference between a paid flop and a bottle-gang flop?" The answer gives you the attributes that organise the whole domain, and those attributes are the thing you could never have written on a form.

In short: descriptive questions get words, structural questions get lists, contrast questions get the logic behind the lists. Run them in that order and the instrument ends up made of the informants' categories instead of yours.

Probing without leading

Most of the skill of interviewing is in what you do after the informant stops talking. Five probes, in rough order of safety:

1. Silence. Count to five. Uncomfortable, and the single most productive technique in the repertoire.
2. The echo. Repeat their last phrase as a question. "The ambulance rota?"
3. The neutral prompt. "Go on." "What happened then?" "Tell me more about that."
4. The long question. Ask a deliberately wordy question about something you already understand. It signals that long answers are welcome and resets the rhythm after a run of short ones.
5. The baited probe. Offer a mildly wrong version and let them correct you. Powerful, and it contaminates the answer with your frame, so use it late and note that you used it.

What to avoid: questions with the answer inside them ("So that must have been frustrating?"), double questions, and any question that can be answered yes. Also avoid why, early. Why invites justification and people defend themselves; how and when and who invite description.

Life histories: Oscar Lewis and the tape recorder

In the 1950s Oscar Lewis recorded hundreds of hours with one Mexico City family, then published the results in 1961 as the separate, overlapping autobiographies of a father and four children. The structure was the argument: four siblings narrate the same household and do not agree about it, and the reader is left holding the disagreement. The book was also a scandal. A Mexican learned society denounced it as obscene and defamatory, and the publisher faced legal action, which is worth remembering as a fact about what verbatim speech can do when it reaches print.

Two lessons. The method's strength is that a life history shows how a structure is inhabited: the same eviction, wage or illness refracted through four tellers. Its weakness is twofold. People impose retrospective coherence, converting a series of accidents into a trajectory with a moral, and the ethnographer then edits hundreds of hours down to a publishable shape, making invisible decisions about what a person's life was about. Lewis's further move, explaining the family through a culture of poverty transmitted across generations, has been heavily criticised for converting the effects of poverty into its causes, and that critique is aimed at his framing rather than at the recordings.

The interview is an event, not a window

In the 1970s Charles Briggs went to a Mexicano community in northern New Mexico to study wood carving and local history, and his interviews kept failing in ways that rapport could not fix. People answered questions briefly and then told long stories that seemed to be about something else. His eventual conclusion, published in 1986 as Learning How to Ask, was that the interview is itself a speech genre with rules, and those rules are not universal. In that community, knowledge was transmitted by watching, by doing and by narrative at the right moment, not by a junior person interrogating a senior one. His questions were not neutral requests for information; they were a breach of etiquette that his hosts politely absorbed.

The practical consequences are large. Who else is in the room changes the record: a husband present, a supervisor audible next door, a teenager translating for a parent. Where it happens matters: a clinic office produces patient speech, a kitchen produces something else. And the payment question is unavoidable. If people are giving you two hours of working time, consider what that is worth locally, and decide before you start rather than improvising at the door.

What people cannot tell you

There is a hard limit on interviewing that no amount of technique removes. In a 1984 review, Russell Bernard and colleagues surveyed decades of studies comparing what people reported about their own behaviour with independent records of that behaviour: communication, time use, consumption, clinic visits. The pattern was consistent and unflattering. People misreport frequencies, compress and round, move events in time, and report the normative rather than the actual, without any intention to deceive.

This is not a reason to stop interviewing. It is a reason to stop asking interviews to do the work of observation and counting. If you want to know how much time women in a household spend fetching water, do not ask; sample behaviour, which is the subject of Lesson 8. If you want to know why it matters who fetches it, you must ask, because no amount of observation yields a reason.

Common misconceptions

• A good interview feels like a natural conversation. It is asymmetric by design: one person talks most, the other steers and shuts up. Conversational reciprocity fills your recording with your own voice.
• With enough rapport, people report their behaviour accurately. Accuracy limits are cognitive, not motivational. Trust improves candour about opinions and barely touches recall of frequencies.
• Structured interviews are more scientific. They are more comparable. Their validity depends entirely on whether the categories came from the field or from your desk.
• Recording everything means you can listen later. Every recorded hour costs four to six hours to transcribe. Unless you have budgeted that, recording is a way of deferring work you will never do.

The short version

Match the instrument to the question and stack instruments whose failures differ. Open unstructured to learn the vocabulary, build semi-structured guides to cover your topics, then make structured tasks out of the terms people actually used. Use descriptive questions to get speech, structural questions to get lists, contrast questions to get the logic. Probe with silence before anything else. Treat the interview as an event with its own etiquette, and record who was in the room. And keep one limit permanently in view: what people can tell you reliably is what things mean to them, not how often they do them.

The core of it: Spradley did not discover a hundred kinds of flop by asking better questions about alcoholism. He discovered them by letting the men's own word for the problem replace his.

Sources

1. Spradley, J. P. (1970). You Owe Yourself a Drunk: An Ethnography of Urban Nomads. Little, Brown. archive.org
2. Lewis, O. (1961). The Children of Sanchez: Autobiography of a Mexican Family. Random House. archive.org
3. Spradley, J. P. (1979). The Ethnographic Interview. Holt, Rinehart and Winston.
4. Briggs, C. L. (1986). Learning How to Ask: A Sociolinguistic Appraisal of the Role of the Interview in Social Science Research. Cambridge University Press.
5. Bernard, H. R., Killworth, P., Kronenfeld, D., and Sailer, L. (1984). The problem of informant accuracy: The validity of retrospective data. Annual Review of Anthropology, 13, 495-517.

An anthropologist tells Hamlet to the elders

Sometime in the 1950s, in a Tiv homestead in central Nigeria, Laura Bohannan was rained in with a compound full of elders who had been drinking beer since morning and wanted a story. She had a copy of Hamlet. She had also been told, before she left Oxford, that Hamlet had one universally obvious interpretation, and she decided to test it.

It went wrong immediately, and the wrongness is a short course in translation. There was no Tiv word for scholar, so Horatio became a man who knew things. There was, the elders informed her, no such thing as a ghost: what the sentries saw must have been an omen sent by a witch. When she reported that Claudius had married his dead brother's widow within a month, the old men were delighted rather than scandalised, because marrying a dead brother's widow is exactly what a good brother does, and the only fault was not waiting the customary period. Hamlet's reluctance to accept his uncle as father became incomprehensible. His madness could only be bewitchment, and since only kin bewitch, the witch had to be Claudius. By the end the elders were explaining to her, kindly, that she had the story slightly wrong and that she should tell her own elders back home, who would set her right.

Bohannan published the episode in 1966 as Shakespeare in the Bush, and it is usually taught as a parable about cultural relativity. Read it again as a methods problem and it says something narrower and more useful: every translation smuggles in a set of institutional assumptions, and the assumptions travel invisibly inside ordinary words. Ghost, father, madness and uncle each carry a social structure on their backs.

Two honest limits on the parable

The essay is a magazine piece written by a very good writer, and it flatters the reader. Two cautions. The Tiv elders were not attempting literary criticism; they were instructing a young woman in correct narration, which is a different speech act and tells you about Tiv pedagogy rather than Tiv hermeneutics. And Bohannan is the only witness, writing in English, years later, with a point to make. Take it as an illustration, not as data: the same scepticism you will apply to your own vivid anecdotes.

How much language do you need

There is a respectable answer, and it is uncomfortable: enough to conduct an ordinary transaction unaided, to follow a conversation you are not part of, and to hear when somebody switches register or language. Less than that and you will systematically over-sample the bilingual, the educated, the young and the men, because those are the people who can talk to you.

Three practical milestones are worth planning towards. Greeting and small talk, which buys ordinary presence. Enough grammar to ask a question you did not prepare, which is what lets an interview follow an unexpected answer. And listening comprehension well beyond your speaking ability, which is the cheapest high-value investment in fieldwork: you do not need to speak well to benefit enormously from understanding what is said around you.

If the gap is unbridgeable in the time available, say so in writing in your methods section, and name the consequence. A study conducted entirely through English in a setting where English marks education is a study of a stratum, not a community.

The interpreter is a person, not a medium

Working with an interpreter is normal and respectable. Pretending the interpreter is a window is not.

An interpreter has an age, a gender, a clan, a religion, a dialect, a neighbourhood and often a job. Each attribute opens some doors and closes others. A young man interpreting for interviews with older women will produce shorter answers on anything intimate, every time. A government-supplied interpreter in a politically tense district may be a reporting channel, whether or not they intend to be. An interpreter who belongs to the dominant lineage will be treated as that lineage's representative by everyone you speak to.

Four working rules. Record your interpreter's social position in your notes as you would record an informant's, because it is a sampling parameter. Train explicitly: ask for literal rather than tidy translation, and agree a signal for stop, I want that word in the original. Debrief after every session, separately, and write what they tell you as a memo with their reading marked as theirs. And consider credit and pay as an ethical matter: an interpreter who is also explaining, introducing and advising is a research collaborator, and increasingly gets named as one.

The upshot: every interpreted interview has three participants, and the transcript shows only two. Put the third one in your notes.

Instruments that survive the crossing

For any fixed instrument, a wording list, a survey item, a consent script, the standard machinery is back-translation. One bilingual speaker translates your item into the local language. A second, who has not seen the original, translates it back into English. You compare. The mismatches are where the concept does not exist, is impolite, or has split into two words.

Expect three kinds of finding. A term with no counterpart, such as a question about household income in a setting where money is not pooled by household. A term that fissions: English cousin becomes four words, and the question no longer means one thing. And a term that carries unwanted force: privacy, consent and rights all translate into vocabularies with political histories attached.

The repair is usually not a better word but a better question. If income cannot be asked, ask who paid for the roof, the school fees and the last funeral. Specific transactions translate far better than abstract nouns.

Quoting people in a language your reader does not have

When you publish somebody's words, you choose between smooth English and visible difference. The serious compromise, borrowed from linguistic anthropology, is interlinear glossing: the original line, a morpheme-by-morpheme gloss, then a free translation. In Swahili, for example, Nimekwisha kula analyses as ni-me-kwisha ku-la, literally I-have-finished to-eat, and freely as I have already eaten. The gloss shows the reader the machinery, including that the completeness lives in a verb rather than in a word like already.

Full glossing is for linguistic claims. For ordinary ethnography, the cheap version is enough and is becoming standard: keep the key term in the original throughout, italicised, defined once, and never translated away. If your informants say one word for a relationship that English splits into obligation, friendship and debt, keeping their word is not decoration. It is the only way to avoid deciding, by translation, the question you are supposed to be investigating.

The Hopi time dispute, and why it matters to your methods

Benjamin Lee Whorf wrote in the 1930s and 1940s that Hopi had no words or grammatical forms referring to time as Europeans understand it, and the claim became the most-cited illustration of linguistic relativity in every introductory textbook for forty years. In 1983 Ekkehart Malotki published a volume of Hopi temporal material running to several hundred pages: units of time, metaphors, tense and aspect marking, calendrical vocabulary. The strong version of Whorf's claim did not survive, and the Hopi time controversy has since become an argument about what exactly Whorf said and how much of a weaker version stands.

Two methodological morals, and the second is the one that should worry you. First, a striking claim about a language is checkable, and someone will eventually check it with a corpus. Second, Whorf was working largely from an expert consultant in New York rather than from sustained fieldwork in Hopi communities, and the gap between elicited material and everyday speech is exactly where this kind of error breeds. You will make the same mistake in miniature every time you take a term's meaning from the one articulate bilingual person who enjoys explaining things to you.

Common misconceptions

• A good interpreter makes language learning unnecessary. Interpreters make interviews possible; they do not let you overhear, read a joke, or notice a switch of register, and those are where much of the analysis comes from.
• If two languages lack matching words, the speakers lack the concept. Lexical gaps are facts about vocabulary, not about thought. The Hopi case is the standard cautionary tale.
• Translating smoothly respects the reader. It also quietly resolves ambiguities that were your actual findings. Keep contested terms untranslated.
• Back-translation verifies a translation. It locates mismatch. Agreement between two translators can still be wrong in the same direction, which is why you also pilot the item with real informants.

What you now know

Translation failure rarely looks like a wrong word. It looks like a perfectly fluent sentence that has imported your institutions: a ghost where an omen belongs, a scandal where a duty belongs. Budget the language seriously, and state in print what your level of it cost the study. Treat the interpreter as a third participant with a social position that shapes the data, and record it. Use back-translation to find the places where your instrument does not exist in the other language, then replace abstractions with transactions. Keep the key terms in the original. And remember that elicitation from one expert, in one room, far from ordinary speech, is how famous errors are made.

Worth holding on to: the Tiv elders were not failing to understand Hamlet. They were understanding it correctly given their institutions, which is precisely what your informants will do with your questions.

Sources

1. Bohannan, L. (1966). Shakespeare in the bush. Natural History, 75(7), 28-33. Wikipedia
2. Malotki, E. (1983). Hopi Time: A Linguistic Analysis of the Temporal Concepts in the Hopi Language. Mouton. Wikipedia
3. Brislin, R. W. (1970). Back-translation for cross-cultural research. Journal of Cross-Cultural Psychology, 1(3), 185-216.
4. Bowen, E. S. [Bohannan, L.]. (1954). Return to Laughter: An Anthropological Novel. Harper and Brothers.
5. Bernard, H. R. (2017). Research Methods in Anthropology: Qualitative and Quantitative Approaches (6th ed.). Rowman and Littlefield.

A triangle, a circle, and everything you have to decide

The notation is nearly two centuries old and takes ten seconds to learn. A triangle is male, a circle is female, a square is a person whose gender is unspecified or irrelevant. An equals sign between two of them is a marriage, a vertical line below runs down to a horizontal sibling bar, and every child hangs off that bar in birth order, oldest on the left. Shade or cross out the dead. Mark the person whose chart it is, called ego, with a small letter.

Now try to draw the household you grew up in, and notice what the notation forces. Where does a stepfather go? What line connects an adopted child? What do you do with a father's second wife, or with a man who raised you and is not on any register? The chart is not a neutral recording device. It is a theory of relatedness, made in nineteenth-century Europe, that assumes reproduction is the spine of kinship. Everything difficult in this lesson lives in that gap.

The genealogical method, step by step

W. H. R. Rivers worked out the procedure in the Torres Strait in 1898 and formalised it in 1910. It survives because it does something unusual: it collects facts a person can supply accurately, then derives from them the rules the person could not have stated.

1. Ask for people, not rules. Start with ego and record parents, siblings, spouses, children, then the same for each parent. Never open with a question like how does descent work here.
2. Take names, and take them in the local form. Also sex, approximate age or birth order, marital status, clan or lineage if there is one, current residence, and whether living.
3. Record the vernacular term for every link. Not uncle, but the exact word ego uses for that man, written as heard.
4. Push outward until recall fails, usually three to four generations up and out to second cousins, and note where it fails, because the boundary of remembered kinship is itself a finding.
5. Cross-check across informants. Collect the same section of the network from a sibling and a cousin. Disagreements are not errors to be resolved; they are claims, usually about land, inheritance or legitimacy.
6. Derive, do not ask. With thirty charts in hand you can read off marriage patterns, post-marital residence, adoption frequency, lineage exogamy and inheritance, each as a count rather than an assertion.

That last step is the whole point of the genealogical method. If you ask people where a couple lives after marriage, you get the norm. If you take fifty actual marriages off your charts and count, you might get seventy percent with the husband's people, twenty percent with the wife's and ten percent elsewhere, and the exceptions are where the interesting fieldwork is.

Kin terms and what they bundle

Now the second half of the method, which is about words rather than people. Elicit the terms directly, in two directions. Point along the chart: what do you call this man, your mother's brother? Then reverse it: who else do you call by that same word? The second question is the one that produces discoveries, because it reveals how far a term reaches.

Lewis Henry Morgan collected this material by circular letter and consular contact in the 1860s and published a vast comparative table in 1871. His central observation still holds: societies bundle relatives differently, in a small number of recurring patterns.

Read the middle column as a set of predictions you can test on your own charts. If father and father's brother share a term, you should expect them to share obligations too, and you can check that against who paid for a funeral.

What the charts cannot see

In 1984 David Schneider, who had built a career on kinship studies, published an argument that the whole enterprise rested on an assumption imported from Europe: that biological reproduction is the natural basis of relatedness, so that every society must have a kinship system whose terms map onto genealogical positions. On Yap, where he had worked, the relevant relationships were held to be created by exchange of land and food, and the genealogical grid distorted them.

You do not have to accept the strong version to take the practical warning. Ask separately, in every setting: who feeds this child, who is responsible for this person, who will be buried where, who inherits, who may not marry whom. Compare that list against the pedigree. The mismatches are the interesting part, and in most contemporary fieldwork they are large: fostering, migration, remarriage, assisted reproduction, adoption and long-distance obligation all produce relatedness that the triangle-and-circle chart cannot express.

Worth holding on to: collect the genealogy anyway. It is cheap, it produces countable facts, and it is the fastest way to find the people whose lives do not fit it.

Personal networks: the name generator

Kinship gives you one kind of structure. Network methods give you another, and they work in settings where kinship is not the organising principle: a hospital ward, a trading estate, an online guild.

The instrument has two parts. A name generator is a question that produces a list of alters: who do you discuss important matters with, who would you ask for a loan of a week's wages, who have you worked a shift with in the last month, who did you exchange seed with last season. Then a set of name interpreters collects attributes of each alter (sex, age, occupation, kin relation, how long known, how often seen) and, crucially, the ties among them: for each pair of alters, does A know B?

Choose the generator to match the resource you care about. Confidants, money, labour, information and childcare produce different lists from the same person, and a study that uses one generator and then talks about someone's social support in general has overreached.

Density, worked by hand

Suppose Ana names five people: B, C, D, E and F. You ask, for each pair, whether they know each other. The pairs are BC, BD, BE, BF, CD, CE, CF, DE, DF, EF, which is ten. The number of pairs among n alters is n times n minus one, divided by two: five times four is twenty, divided by two is ten.

Ana reports that B knows C, B knows D, C knows D, and E knows F. That is four ties out of ten possible, so the density of her alter network is four divided by ten, or 0.40.

Now read it. There is a triangle of B, C and D who all know each other, and a separate pair, E and F, connected to the rest only through Ana. Ana sits in what network analysts call a structural hole: she is the only bridge between two clusters. That position typically brings non-redundant information and brokerage opportunities, and it also brings strain, because the two clusters may make incompatible demands.

Compare a second respondent, Bala, who names five people with nine of the ten pairs connected: density 0.90. Bala is inside a dense, closed group, which is good for enforcing norms and mutual aid and poor for novel information. Two numbers, computed from twenty questions, and you have a structural contrast worth an entire chapter of interviews.

What a famous network finding got wrong

In 2006 three sociologists compared answers to the same discuss important matters question in the American General Social Survey from 1985 and 2004 and reported that the average number of confidants had fallen from about 2.9 to about 2.1, and that the share of people naming nobody at all had risen from roughly one in ten to about one in four. The finding travelled worldwide as evidence of social isolation.

Claude Fischer then examined the 2004 data as an artefact problem and argued in 2009 that the drop was too abrupt and too uniform across groups to be real, and that it bore the signature of an instrument effect: a change in how the question sat in the survey and how interviewers handled it, so that later respondents were more likely to stop naming people. The debate continues, and what matters for you is the shape of it. A name generator is a fragile instrument. Where it sits in the interview, how tired the respondent is, whether the interviewer probes for more names, and how you define important matters can move the mean by a whole person.

The defence is procedural: fix the wording, fix the position in the schedule, always probe once for additional names, record the interviewer, and report the generator verbatim in your methods section so that a reader can tell what your number is a number of.

The consent problem built into both methods

Genealogies and name generators share a property that ordinary interviewing does not: they collect detailed data about people who are not present and have not agreed to anything. A pedigree records paternity, illegitimacy, divorce and death. A network roster records who lends money to whom. In a small community these data are re-identifiable no matter what you do to the names, and in a criminalised or surveilled setting they can be dangerous. Collect the minimum you need, code names at the moment of recording rather than later, keep the key separate, and think hard before recording any tie that would embarrass or endanger a third party if the file were read.

Common misconceptions

• Genealogies tell you the rules of kinship. They tell you the practice; the rules are what people say when asked, and the interesting work is in the distance between the two.
• A network is a picture of somebody's real relationships. It is a picture of the answers to one specific question. Change the generator and you get a different network from the same person.
• Density is always better when higher. High density supports enforcement and mutual aid; low density supports novel information and brokerage. Which is better depends on what you are asking.
• Kinship charting is obsolete in urban fieldwork. It is cheap and fast anywhere, and in migrant, fostered and blended households it is the quickest way to locate the relationships that no register records.

Pulling it together

Both methods work by refusing to ask for the general rule. You collect particular people, particular ties and particular words, then derive the pattern by counting. The genealogical method gets you descent, residence and marriage as frequencies, plus a set of vernacular terms whose reach you test by asking who else the term covers. The name generator gets you a roster and the ties among it, from which density and brokerage fall out with arithmetic you can do on paper. Both then need the same corrective: ask what relationship in this setting is not captured by the instrument, and go looking for the people who prove it.

The point: a density of 0.40 and a density of 0.90 are two different lives, and you can establish which one a person is living in about twenty questions.

Sources

1. Morgan, L. H. (1871). Systems of Consanguinity and Affinity of the Human Family. Smithsonian Institution. archive.org
2. Rivers, W. H. R. (1910). The genealogical method of anthropological inquiry. The Sociological Review, 3(1), 1-12. Wikipedia
3. Schneider, D. M. (1984). A Critique of the Study of Kinship. University of Michigan Press.
4. McPherson, M., Smith-Lovin, L., and Brashears, M. E. (2006). Social isolation in America: Changes in core discussion networks over two decades. American Sociological Review, 71(3), 353-375.
5. Fischer, C. S. (2009). The 2004 GSS finding of shrunken social networks: An artifact? American Sociological Review, 74(4), 657-669.

Thirteen visits before breakfast

In the early 1970s, in a Machiguenga settlement on a tributary of the upper Amazon, Allen Johnson and his family spent a year walking into households at times chosen in advance by a random procedure, writing down what each person present was doing at the instant of arrival, and leaving. Over a year the visits accumulated into several thousand momentary records.

What came out contradicted two things everyone believed. First, the ceaseless-toil picture of subsistence horticulture: hours actually spent in food production and manufacture were a modest fraction of the waking day, with a great deal of time in rest, visiting and eating. Second, the idea that men worked and women kept house: once food processing, water, firewood and childcare were counted in the same units as gardening and hunting, women's total working time equalled or exceeded men's. Neither finding could have come from interviews, because neither men nor women had any reason to have counted.

Four designs, four bias profiles

Read the bias column across the rows and the design principle becomes obvious: pair methods whose biases point in opposite directions. Spot observation undercounts what happens behind doors; diaries are strongest precisely there. Continuous monitoring catches sequence and loses breadth; spot observation catches breadth and destroys sequence. A study with one behavioural method has one bias and no way to detect it.

The arithmetic of spot observation

Spot sampling works because of a simple equivalence: if your observation moments are a random sample of the moments in a day, then the proportion of observations showing an activity is an unbiased estimate of the proportion of time spent in it. No stopwatch is needed, and each observation costs seconds.

Work an example. You make 200 spot observations of adult women across four months, at times drawn randomly from the twelve daylight hours. Thirty-four of them show water collection.

• Proportion: 34 divided by 200 is 0.17, so 17 percent of daylight time.
• In hours: 0.17 times 12 hours is 2.04 hours a day.
• Margin of error: the standard error of a proportion is the square root of p times one minus p, divided by n. Here that is the square root of 0.17 times 0.83 divided by 200, which is the square root of 0.000706, which is 0.027. Multiply by 1.96 for a 95 percent interval: about 5.2 percentage points.
• So the estimate is 17 percent plus or minus 5 points, or between about 1.4 and 2.7 hours of daylight a day.

That interval is the honest version, and it is wide. To halve it you need four times the observations, because the error falls with the square root of n. Eight hundred observations would give roughly plus or minus 2.6 points. Knowing this before you start is what separates a design from a hope: if you need to detect a difference of three percentage points between two groups, 200 observations per group will not do it, and no amount of careful writing afterwards will repair that.

Bottom line: behaviour sampling converts presence into numbers, and the numbers come with an interval you are obliged to compute and report.

The five ways spot sampling goes wrong

1. Non-random times. Nobody wants to sample at 5 am or during a downpour, and the activities concentrated there vanish from the record. Draw the times mechanically, in advance, and keep the misses as a logged category rather than silently replacing them.
2. Reactivity on approach. If people hear you coming, you record the activity they switch to. Record what you saw at the first instant, note whether you were seen first, and treat seen-first observations as a separate stratum you can compare.
3. Visibility bias. Cooking inside a closed house, sex, prayer, quarrels, illicit work and screen time are all systematically undercounted. Pair with diaries and say so in print.
4. Who is present. Sampling households at home undercounts people who are elsewhere, which is exactly the population doing wage labour and migration. Count absences explicitly as an activity category, with a destination where you can get one.
5. Simultaneity. A woman shelling beans while minding a toddler and talking to a neighbour is doing three things. A single-code scheme forces a choice, and the choice usually erases childcare. Record a primary and a secondary activity from the start, because you cannot reconstruct it later.

Sampling people, not just moments

Behaviour sampling solves the when. You still have to solve the who, and this is where fieldwork is routinely weakest, because the people who end up in your notes are the ones who liked you.

• Probability sampling is possible more often than ethnographers assume. If you can census a site, you can draw a simple random sample of households or workers and knock on those doors, which changes the standing of every number you report.
• Purposive sampling for range is the ethnographic workhorse: choose cases to span the dimensions you think matter, such as both shifts, both contract types, long-service and new. The claim it supports is about range and mechanism, not proportion, and you must write it that way.
• Chain referral reaches hidden populations by asking participants for introductions. Its weakness is that it travels along existing ties and over-represents the well connected. Respondent-driven sampling formalises it with fixed numbers of coupons per participant and statistical adjustment for network size, and it is now the standard approach for populations that cannot be enumerated.
• Theoretical sampling selects the next case to test an emerging idea: you think the joke is a privilege of the driver's contract, so you deliberately seek a sorter who does it. Sampling becomes part of analysis rather than a preliminary.

How many interviews is enough? The honest answer is that it depends on how varied the population is and how narrow the question is, and the useful empirical answer comes from a 2006 study by Guest, Bunce and Johnson, who coded sixty interviews in a relatively homogeneous population and found that the great majority of their codes had appeared within the first twelve, with almost nothing new after that. Treat twelve as a floor for a narrow question in a homogeneous group, and expect a heterogeneous population or a comparison across categories to multiply it.

What the numbers cannot do

Time allocation data are powerful and narrow. They tell you that women in a settlement spend two hours a day carrying water. They cannot tell you whether that walk is drudgery or the only hour of the day away from a mother-in-law, and those two readings imply opposite interventions. The coding categories are yours, not the community's, and an activity system does not cut itself at the joints you bring.

This is why the systematic module sits between the interviewing module and the interpretive ones. The count tells you what is there and in what quantity; it cannot tell you what it is for. Fieldwork that reports only counts has the authority of measurement and the blindness of a form, and fieldwork that reports only meanings has the opposite pair.

Common misconceptions

• You need long observation sessions to measure time use. The opposite: thousands of instants sampled at random beat a handful of long watches, and cost less.
• Ethnography cannot produce representative numbers. It can, whenever the site can be enumerated. The usual barrier is that nobody drew a sampling frame, not that fieldwork forbids one.
• Saturation means no new information at all. It means no new codes for your specific question. It is a property of a question and a population, not a universal count of interviews.
• If the sample was not random, numbers are pointless. Counts from a purposive sample are still informative about range and mechanism, provided you never present them as proportions of a population.

Recap

Random spot observation turns momentary presence into unbiased time estimates, and the estimate comes with an interval: 34 of 200 observations is 17 percent, plus or minus about five points, and narrowing that by half costs four times the work. Continuous monitoring buys sequence at the price of reactivity and small samples. Diaries reach behind the doors you cannot enter. Pair designs whose biases oppose each other, record primary and secondary activities from day one, log your misses rather than moving them, and count absence as a category. Then sample people with a stated logic, purposive for range or chain referral for hidden populations, and write the claim the design can actually carry.

What matters here: Johnson's thousands of one-second observations overturned two durable beliefs about who works in a horticultural society. The method was a notebook, a randomised list of times and the discipline to use it on wet mornings.

Sources

1. Johnson, A. (1975). Time allocation in a Machiguenga community. Ethnology, 14(3), 301-310. Wikipedia
2. Bernard, H. R. (2017). Research Methods in Anthropology: Qualitative and Quantitative Approaches (6th ed.). Rowman and Littlefield.
3. Guest, G., Bunce, A., and Johnson, L. (2006). How many interviews are enough? An experiment with data saturation and variability. Field Methods, 18(1), 59-82.
4. Heckathorn, D. D. (1997). Respondent-driven sampling: A new approach to the study of hidden populations. Social Problems, 44(2), 174-199. Wikipedia
5. Gross, D. R. (1984). Time allocation: A tool for the study of cultural behavior. Annual Review of Anthropology, 13, 519-558.

Twenty lists and a curve that flattens

Ask twenty people in one town to name every illness children get, writing as they speak, prompting only with what else. The first person gives fourteen items. The second adds four you have not seen. By the eighth person, new items arrive one at a time; by the fifteenth, hardly at all. Plot the cumulative count of distinct items against the number of respondents and you get a curve that rises fast and then flattens. That flattening is your evidence that you have found the boundaries of a shared domain, and it took an afternoon.

This lesson is the systematic core of the course: three linked techniques that take you from an unknown domain to a defensible statement about what a group collectively knows and how much any individual shares it. All three were designed to be run in the field with paper.

Step one: freelisting, and the arithmetic of salience

The prompt matters more than anything else. Use the local term for the domain, keep it short, and never offer examples, because an example is a category imposed. Write items in the order spoken; order is data.

Two things make an item important: how many people mention it, and how early. Combine them with Smith's salience index. For one respondent, an item in position k on a list of length n has salience (n - k + 1) / n. So the first item on a list of five scores (5 - 1 + 1) / 5 = 1.0, the third scores 0.6, the last 0.2. Average each item's score across all respondents, counting zero for respondents who never mentioned it.

Work three lists, on a domain of childhood illnesses:

• Respondent A (n = 4): fever, cough, worms, rash.
• Respondent B (n = 3): worms, fever, evil eye.
• Respondent C (n = 5): fever, diarrhoea, worms, cough, evil eye.

Take fever. For A it is first of four: (4 - 1 + 1) / 4 = 1.00. For B it is second of three: (3 - 2 + 1) / 3 = 0.67. For C it is first of five: 1.00. Average over all three respondents: (1.00 + 0.67 + 1.00) / 3 = 0.89.

Now worms. A: third of four, (4 - 3 + 1) / 4 = 0.50. B: first of three, 1.00. C: third of five, 0.60. Average: 0.70.

And evil eye. A: absent, 0. B: third of three, 0.33. C: fifth of five, 0.20. Average: (0 + 0.33 + 0.20) / 3 = 0.18.

The ranking, fever 0.89, worms 0.70, evil eye 0.18, is a measurement of cultural salience rather than an impression. Note what it does not say: evil eye may be enormously consequential when it does occur. Salience measures cognitive prominence in a domain, not importance in life.

So what?: the top ten or twenty items by salience are your card set for the next technique, and they are made of the community's words, which is the whole reason the sequence is worth running.

Step two: pile sorts, and the similarity matrix

Write each salient item on a card. Hand the shuffled deck to one person with an unconstrained instruction: put these into piles so that the ones that go together are together, as many or as few piles as you like. Then ask them to name each pile and explain one card you find surprising. Repeat with thirty people.

Each person's sort becomes a matrix of ones and zeros: one if two items were in the same pile. Add the matrices across people and divide by the number of people, and you have a similarity matrix whose cells are the proportion of respondents who grouped that pair.

Six items, three sorters, worked:

• Sorter 1: {fever, cough, rash} and {worms, diarrhoea} and {evil eye}.
• Sorter 2: {fever, cough} and {worms, diarrhoea, rash} and {evil eye}.
• Sorter 3: {fever, cough, rash, evil eye} and {worms, diarrhoea}.

Fever with cough: together in all three, so 3 / 3 = 1.00. Worms with diarrhoea: together in all three, 1.00. Fever with worms: never, 0.00. Rash with fever: sorters 1 and 3, so 0.67. Evil eye with fever: sorter 3 only, 0.33.

Even with three sorters a structure is visible: a respiratory-and-surface cluster, a gut cluster, and evil eye sitting mostly outside both. With thirty sorters you would feed the matrix to multidimensional scaling and get a map in which distance represents dissimilarity, plus a cluster analysis to cut it into groups. The software is trivial. The judgement is in the interpretation: a map shows how items group, and it does not tell you why, which is what the pile-naming question is for.

Step three: the consensus model, and competence by hand

Now the harder question. Suppose you ask thirty people twenty true-or-false questions about childhood illness. They disagree. Is there a shared body of knowledge that individuals know to varying degrees, or are there two or three different bodies of knowledge in the same town? Cultural consensus theory, formalised by Romney, Weller and Batchelder in 1986, answers that question and, if the answer is yes, recovers the shared answers without the researcher knowing them in advance.

The logic is a test theory turned inside out. In an exam, you know the key and you score the students. Here you do not know the key, so you use agreement among respondents to estimate how competent each of them is, and then use competence to reconstruct the key. Two people who both know a lot will agree often; two who are guessing will agree at chance.

Here is the arithmetic, done by hand for three informants and true-or-false items. Let M be the observed proportion of items on which two people gave the same answer. Because two people can agree by both guessing, correct for chance: the corrected agreement a is 2M - 1. Under the model, that corrected agreement equals the product of the two people's competences.

• Observed: M(1,2) = 0.80, M(1,3) = 0.75, M(2,3) = 0.70.
• Corrected: a(1,2) = 0.60, a(1,3) = 0.50, a(2,3) = 0.40.
• Since a(1,2) = c1 times c2, and so on, c1 = the square root of a(1,2) times a(1,3) divided by a(2,3) = the square root of (0.60 times 0.50) / 0.40 = the square root of 0.75 = 0.87.
• c2 = the square root of (0.60 times 0.40) / 0.50 = the square root of 0.48 = 0.69.
• c3 = the square root of (0.50 times 0.40) / 0.60 = the square root of 0.33 = 0.58.

Check it: 0.87 times 0.69 is 0.60, and 0.87 times 0.58 is 0.50, and 0.69 times 0.58 is 0.40. The three competences reproduce all three agreements, which is the model fitting. With thirty people you do the same thing with factor analysis of the corrected agreement matrix, and the first-factor loadings are the competence scores.

Then the answer key: for each question, count the votes weighted by competence rather than by head. If informant one, with competence 0.87, says true and two informants with competences 0.30 say false, the weighted evidence still favours true, and the formal version computes a posterior probability for each answer.

When the model says no

The model's most useful output is its refusal. Three conditions have to hold before you may report a consensus, and each one is checkable.

1. One large factor. The first eigenvalue should be at least about three times the second. If the second factor is comparable, there are two answer keys in your sample, which means two subcultures, and averaging them describes nobody.
2. All loadings positive. A negative loading means someone is systematically giving the opposite of the consensus answer, which usually signals a different frame of reference rather than incompetence.
3. Items from one domain, asked the same way. The model assumes each question has a single culturally correct answer and that respondents answer independently, so a group interview breaks it.

Failure is a finding. A ratio of 1.4 with two clean groups of loadings is the discovery that the town holds two theories of childhood illness, which is more interesting than a consensus would have been, and it points you straight back at qualitative work to find out who holds which and why.

What consensus analysis found: Aguaruna manioc

James Boster gave Aguaruna women in the Peruvian Amazon the task of identifying manioc varieties growing in gardens, and analysed their agreement. Agreement was high: there was a shared code for identifying and naming varieties, not a private idiosyncrasy per gardener. Competence varied systematically, and it varied with social position: women who grew more varieties and who were connected through kinship to more gardens agreed more closely with the consensus. Knowledge tracked the paths along which planting material moved.

What that design could not see is equally clear. It measures agreement about naming, not skill at cultivation, and a woman could in principle name every variety correctly and grow them badly. And the consensus it recovers is the consensus of the women sampled; if men classify manioc differently, and they were not asked, the model will happily report a strong single culture.

Common misconceptions

• Consensus analysis proves a culture is homogeneous. It tests homogeneity for one domain, with one instrument, in one sample, and its failures are the point.
• Competence means intelligence or authority. It means agreement with the group's shared answers. A dissenting expert scores low, which is why a low score is a prompt to interview rather than a verdict.
• Salience means importance. Salience is cognitive prominence within a domain. Rare and catastrophic things score low and may matter more than anything else on the list.
• You need software to use these methods. Freelist salience, pile-sort similarity and three-person competence are all arithmetic you can do on paper in the field, which is exactly when you need the result.

The takeaway

The three techniques form one pipeline. Freelists find the domain's items and rank them by salience, using the community's words rather than yours. Pile sorts turn those items into a similarity matrix that shows how the domain is organised, with the pile names supplying the reasons. The consensus model then asks whether the group shares one body of knowledge about the domain, estimates how much each person shares it, and reconstructs the shared answers. Each step is arithmetic you can check, and each has a specific failure that is itself informative: a list that never flattens, a matrix with no structure, an eigenvalue ratio near one.

Key idea: corrected agreement between two people is the product of their competences. That single equation is why a researcher who does not know the answers can still find out who does.

Sources

1. Romney, A. K., Weller, S. C., and Batchelder, W. H. (1986). Culture as consensus: A theory of culture and informant accuracy. American Anthropologist, 88(2), 313-338. Wikipedia
2. Boster, J. S. (1986). Exchange of varieties and information between Aguaruna manioc cultivators. American Anthropologist, 88(2), 428-436. Wikipedia
3. Weller, S. C. (2007). Cultural consensus theory: Applications and frequently asked questions. Field Methods, 19(4), 339-368.
4. Smith, J. J. (1993). Using ANTHOPAC 3.5 and a spreadsheet to compute a free-list salience index. Cultural Anthropology Methods, 5(3), 1-3.
5. Bernard, H. R. (2017). Research Methods in Anthropology: Qualitative and Quantitative Approaches (6th ed.). Rowman and Littlefield.

A father who would not let his daughter travel alone

In 1978 a graduate student flew to Egypt to study Awlad Ali Bedouin communities in the Western Desert, and her father insisted on coming with her to the first village. She was embarrassed. He was adamant, and he was right: he spent the visit establishing with the senior men that she was somebody's daughter, under a named man's protection, and that framing settled her position for the next two years. Lila Abu-Lughod has written that without it she would have been unplaceable, and an unplaceable young woman in that setting would have collected very little.

Hold the consequences in view. Her Palestinian father's name, her gender and her age did not merely colour her interpretation. They determined which tents she could sit in, which conversations she could hear, which poems women would recite to her and which political talk among men she would never attend. Veiled Sentiments (1986) is built on the oral poetry women performed in private settings, which is exactly the material a male fieldworker of any nationality could not have gathered. The positional facts are the data collection method.

That much is not in dispute. The dispute is about what belongs in the book, and it has two serious sides.

Position one: without the ethnographer in the text, the reader cannot evaluate anything

The case rests on specifics rather than principle. Malinowski's monographs describe Trobriand society with no statement of how much time he spent on traders' verandahs, and the 1967 diary shows it mattered. Weiner's restudy found a parallel wealth system that a male fieldworker had dismissed, which is a gender-of-researcher effect with a concrete missing chapter attached. The 1970 collection Women in the Field assembled case after case in which a woman's access ran orthogonally to a man's in the same village.

Paul Rabinow pushed this furthest in 1977 by making the fieldwork relationship itself the subject. His Moroccan book describes quarrels with his main informant, moments when he used his money or his temper to get compliance, and the recognition that every piece of data he held had been produced inside a relationship with a power gradient running through it. The claim is not that this is confessional colour. It is that an account which hides the relationship hides the instrument.

Zora Neale Hurston had made a version of the point in 1935, in the introduction to Mules and Men, describing her return to the Florida town she grew up in and needing what she called "the spy-glass of Anthropology" to see a life she had worn too closely to notice. Being from a place is a position too, with its own blind spots, and the remedy is not detachment but a stated vantage point.

Position two: the author expands until the subjects disappear

The counter-case is not a defence of fake objectivity, and the strongest version of it is empirical: look at what the reflexive turn produced. A genre arrived in which the ethnographer's feelings about fieldwork occupy the foreground and the people of the field appear as occasions for the author's self-examination. A reader who wants to know how Awlad Ali women used poetry to say things prose could not say is given instead an account of the researcher's discomfort.

There is also the alibi problem, which is the sharper objection. A positionality paragraph in the introduction, listing race, class, gender and nationality, costs nothing and changes nothing if no subsequent claim is qualified by it. It functions as a ritual disclaimer that purchases authority rather than limiting it: I have acknowledged my position, therefore the remaining three hundred pages proceed as before. Declaring a bias is not the same as correcting for it, and a declaration that does no analytic work is decoration.

The practical version of this complaint comes from anthropologists working in the communities they write about, who have noted that intense authorial self-presence is a luxury of researchers whose careers do not depend on the community reading the book.

What would settle it

The dispute resolves if you stop treating reflexivity as a virtue to be displayed and treat it as a claim to be tested. Ask of any reflexive passage: which finding in this book would be stated differently without it?

• Abu-Lughod's account of her placement as a daughter passes. It explains why her corpus is women's poetry, and therefore what the book can and cannot say about men's politics.
• Weiner's account of her position passes. It names the mechanism by which a previous study missed an entire wealth system.
• A paragraph stating that the author is a middle-class European who acknowledges the colonial history of the discipline, followed by no qualified claim anywhere, fails. Nothing downstream depends on it.

The core of it: reflexivity earns its place when it changes a conclusion, bounds a claim, or explains a gap. If you can delete the passage and every other sentence stands unaltered, it was never doing epistemic work.

The insider and outsider pair does not survive contact

Kirin Narayan took the binary apart in 1993 with a simple observation about her own fieldwork in India: she was read as a local by some people and as a foreigner by others, in the same village, on the same day, depending on which of her attributes was salient to whom. Language, caste, education, accent, religion, dress, marital status and the institution paying for the trip each pull in different directions, and none of them is stable across a year.

Abu-Lughod's term for researchers whose identity straddles the field is halfies, and her 1991 essay argues that such positions make visible what the standard opposition conceals: that every fieldworker is partially positioned, and the self-and-other division is a convention of the writing rather than a fact of the encounter.

Being read as an insider has specific costs that the romantic version never mentions. You are expected to know things and so cannot ask. You are held responsible for the community's reputation in print. Your own faction is assumed. And your competence is discounted by colleagues who will describe your work as advocacy.

Being read as an outsider has compensating advantages, which is uncomfortable but true. Ignorance is permitted. Factional neutrality is sometimes conceded. Questions that would be rude from a member are tolerated from a foreigner. Neither position is better; they are differently shaped, and the useful sentence in a methods section names which attributes were salient to whom and when, rather than claiming membership or distance.

How to do this in practice, without writing a memoir

1. Keep a separate field diary. Your state and your relationships belong in writing, daily, and not inside the fieldnotes. Rabinow's book was possible because the material existed.
2. Log access refusals. Every time you are excluded from something, record it with the reason you were given and the reason you suspect. A list of exclusions is the most honest positionality statement available, and it is made of evidence.
3. Name the salient attributes, not the full census. In a given setting, two or three of your characteristics do nearly all the work. Identify which, and say how you know.
4. Attach positionality to claims, not to the preface. Write the limiting clause in the paragraph where the claim appears: my account of the men's meetings comes entirely from what women reported afterwards.
5. Check the gap with a differently positioned person. A research assistant, a collaborator, or a colleague of another gender, generation or background will see what you structurally cannot.

Common misconceptions

• Reflexivity means writing about your feelings. It means accounting for how your position produced the data. Feelings are relevant only where they altered access or interpretation.
• Insiders get better data. They get different data, with an obligation structure attached and a standing inability to ask obvious questions.
• Acknowledging bias neutralises it. Acknowledgement without a downstream qualification is an alibi; the test is whether any claim is actually bounded by it.
• Positionality is a fixed list of demographic facts. It is situational: which of your attributes matters changes by setting, by interlocutor and across the months of fieldwork.

Where this leaves us

Both sides of this argument are defending something real. The reflexive case rests on documented instances where the fieldworker's position determined the findings, from the banana-leaf bundles Malinowski dismissed to the poems only a daughter of the household could hear. The counter-case rests on an equally observable genre failure, in which the author's presence crowds out the people and a positionality paragraph buys authority without paying for it. The way through is the test: keep the reflexive material that bounds a claim, explains a gap or changes a conclusion, and cut the rest. Record refusals as data, name only the attributes that were doing work, and put the limiting clause next to the claim it limits.

Bottom line: Narayan's point is the one to keep. You are not an insider or an outsider. You are a particular person, read differently by different people on the same afternoon, and the job is to say which reading produced which data.

Sources

1. Abu-Lughod, L. (1986). Veiled Sentiments: Honor and Poetry in a Bedouin Society. University of California Press. Wikipedia
2. Hurston, Z. N. (1935). Mules and Men. J. B. Lippincott. Wikipedia
3. Narayan, K. (1993). How native is a native anthropologist? American Anthropologist, 95(3), 671-686.
4. Rabinow, P. (1977). Reflections on Fieldwork in Morocco. University of California Press.
5. Abu-Lughod, L. (1991). Writing against culture. In R. G. Fox (Ed.), Recapturing Anthropology: Working in the Present (pp. 137-162). School of American Research Press.

A town that recognised itself

In 1958 two sociologists published a study of a village in upstate New York under the pseudonym Springdale. Names had been changed. It made no difference. Residents worked out within weeks who everyone was, because a small community's positions are unique: there is exactly one man who chairs that committee and owns that shop. The town's response was a Fourth of July parade in which residents rode in a float dressed as the book's characters, followed by an effigy of one of the authors bent over a manure spreader.

Springdale is the standard teaching case for a reason that has nothing to do with malice. Vidich and Bensman broke no regulation that existed in 1958. They anonymised competently by the standards of the day. And the protection failed completely, because ethnographic detail is what defeats anonymisation: the richer the description, the more identifying it is. That tension does not go away, and this lesson is about running an ethics practice that takes it seriously.

Where the professional obligations came from

Anthropology's ethics rules were written after scandals, in a specific order that is worth knowing.

In 1919 Franz Boas published a letter accusing four unnamed anthropologists of using research as a cover for espionage in Central America during the war. The American Anthropological Association censured him rather than them. Half a century later the discipline arrived at the opposite view. Project Camelot, an army-funded study of the causes of insurgency, collapsed in 1965 after Chilean scholars discovered its sponsorship, and revelations about anthropologists working on counterinsurgency in Thailand followed. Out of that came the 1971 Principles of Professional Responsibility, whose central rule the discipline has kept ever since: when obligations conflict, the primary obligation is to the people studied.

The current AAA principles, adopted in 2012, list seven: do no harm; be open and honest about your work; obtain informed consent and necessary permissions; weigh competing ethical obligations to collaborators and affected parties; make results accessible; protect and preserve your records; and maintain respectful professional relationships. Note the second one. Covert research is not merely risky; openness is a stated professional requirement, which is why Tearoom Trade, Laud Humphreys' 1970 study in which he served as a lookout during anonymous sexual encounters and later traced participants to their homes through their licence plates to interview them under a false pretext, remains the canonical example of what the rules exist to prevent.

Consent as a process, in seven steps

The informed consent form is a legal instrument built for a procedure that happens once. Fieldwork is a relationship that lasts months, in which the researcher's understanding of the project changes and the participant's exposure changes with it. Run it as a process.

1. Permissions before consent. National clearance, institutional approval, community-level approval where a body exists to give it. Some Indigenous nations run their own review boards, and their approval is a requirement, not a courtesy.
2. An opening statement everyone gets. The same eight-second sentence you use for entry, plus who funds you, what will be written, and that saying no costs nothing. Repeat it to newcomers for as long as you are there.
3. Consent per use, not per person. Agreeing to talk is not agreeing to be recorded, quoted by name, photographed, filmed, archived or written about in a book that will be sold. Ask separately for each, and record which were granted.
4. Written where writing is safe, oral where it is not. In many settings a signature is the only document tying a name to the study. Oral consent with a witnessed log entry protects better; boards can waive documentation for exactly this reason.
5. Re-consent at transitions. When the project changes, when a relationship becomes intimate enough that people forget you are working, and again before publication for anyone who features heavily.
6. A consent log. Date, person's code, what was explained, what was agreed, what was declined, in what language, who else was present. This is your evidence, and it takes a minute per entry.
7. An exit. Tell people how to withdraw and what withdrawal can and cannot undo. Be honest: once a book is printed nothing can be retracted, so the meaningful window is before submission.

Why this matters: a signature obtained in week one covers a study the participant could not have imagined and you had not yet designed. Process consent is not extra paperwork; it is the only version that describes what actually happened.

Anonymisation, and its four limits

Changing names is the beginning of anonymisation and nowhere near the end. Four limits recur.

• Uniqueness of position. The only woman on the night shift, the only imam in the valley, the only nurse who trained abroad. Role plus place identifies a person even with the name removed.
• Insiders read differently. The people who can identify your informants are their neighbours, and they know what you cannot disguise: who was at the meeting, whose child was ill that month.
• Combination. Age, occupation and neighbourhood are individually harmless and jointly identifying. This is the ordinary logic of re-identification, and ethnographic detail supplies more attributes than any dataset.
• Site pseudonyms leak. A disguised place name plus a real regional detail plus a publication date is usually enough for a determined reader, and search engines have made that reader ordinary.

The available responses are blunter than students expect: alter non-essential identifying details and say in a note that you have; composite characters, which some journals permit and others treat as fabrication, so declare it either way; delay publication; omit the case entirely; or negotiate with the person about what may appear. The last is the strongest and the least used.

Promises you may not be able to keep

Confidentiality is a promise about your future conduct, and your conduct is not the only variable. Field notes have been subpoenaed. In the most instructive case, Boston College ran an oral history project recording former paramilitaries in Northern Ireland, with interviewees assured that their tapes would stay sealed until their deaths. Between 2011 and 2013, British authorities obtained material from the Belfast Project through a treaty request and United States courts, and the college lost most of the litigation. The promise had been made by researchers who had no power to enforce it against a subpoena.

Practical consequences. Do not promise absolute confidentiality; promise what you control, and say what you do not. Consider legal protections where they exist, such as certificates of confidentiality in some United States federally funded research, and read what they actually cover. Keep the pseudonym key on separate media from the notes. Destroy recordings on the schedule you promised, and put the schedule in the consent script. And think about jurisdiction before you archive anything: material in a university repository is reachable by processes that may not exist where you did the fieldwork.

Whose data is it

The final question is the one funders answer by default if you do not. Open-data requirements now push researchers to deposit interview transcripts and field materials, and the assumption that data belong to the researcher and their institution is being challenged by Indigenous data governance frameworks. The Canadian OCAP principles assert First Nations ownership, control, access and possession of data about their communities, and the CARE principles for Indigenous data governance set collective benefit, authority to control, responsibility and ethics alongside the technical open-data norms.

What this means in a design: agree in advance who may use the material and for what, whether the community holds a copy, who is consulted before secondary use, and what returns to the community in a usable form. Write it into the protocol, because a general promise to share results turns into nothing once you are two jobs away.

Common misconceptions

• A signed form means consent has been obtained. It documents a moment. Consent is a state that has to be maintained, and it is use-specific.
• Changing names protects informants. Position, place and combined attributes identify people, and neighbours are the readers who matter.
• Ethics review approval settles the ethics. Boards check protocols against regulations; the primary obligation to the people studied is yours to honour when a conflict arrives.
• Covert research is acceptable when there is no other way in. Openness is a stated professional principle, and the burden of proof against concealment is heavy for the reasons Tearoom Trade demonstrated.

What to remember

Ethics in fieldwork is a set of running procedures, not a document you clear before departure. Get permissions from every body entitled to give them, including community boards. Ask consent for each use separately and log it. Re-consent when the project or the relationship changes, and again before publication for anyone prominent in the text. Treat anonymisation as partial by default, tell people it is partial, and negotiate about specific passages rather than promising a blanket disguise you cannot deliver. Promise only the confidentiality you can enforce. And settle ownership, archiving and return of the material while you still have the leverage of a project that has not yet been funded.

Remember: the Springdale float was not a failure of anonymisation technique. It was a failure to notice that a detailed description of a small community is, by construction, a description of identifiable people.

Sources

1. Vidich, A. J., and Bensman, J. (1958). Small Town in Mass Society: Class, Power and Religion in a Rural Community. Princeton University Press. archive.org
2. Office for Human Research Protections. (2018). 45 CFR 46: Federal Policy for the Protection of Human Subjects. U.S. Department of Health and Human Services. hhs.gov
3. American Anthropological Association. (2012). Principles of Professional Responsibility. AAA Ethics Statement, adopted 2012, revised 2020.
4. Humphreys, L. (1970). Tearoom Trade: Impersonal Sex in Public Places. Aldine.
5. Carroll, S. R., et al. (2020). The CARE principles for Indigenous data governance. Data Science Journal, 19(43), 1-12.

An email that arrived before the book

In September 2000 two senior anthropologists sent officers of their professional association a warning about a book they had read in manuscript. The book was Patrick Tierney's Darkness in El Dorado, and the email said the scandal it described would be unparalleled in the discipline's history. Within days the claims were in newspapers on three continents.

The most explosive of them ran roughly as follows. During a measles epidemic among the Yanomami of Venezuela in 1968, the geneticist James Neel and the anthropologist Napoleon Chagnon had administered an outdated vaccine that caused or intensified the epidemic, and had done so as part of a research programme, withholding medical care to observe the results.

That is the plausible wrong answer this lesson works from. It is plausible because the surrounding context is true: a research expedition, a devastating epidemic, blood samples taken from a vulnerable population, an anthropologist with a reputation for high-handedness. Watch what happens when you take the accusation apart into checkable pieces.

Step one: separate the claims

A public accusation is almost never one claim. This one contained at least six, and they are of different kinds and different difficulty.

1. The vaccine caused or worsened the epidemic. Empirical, medical, checkable.
2. Medical care was deliberately withheld to observe outcomes. Empirical, historical, checkable against records and witnesses.
3. Blood samples were taken without adequate consent. Empirical, and dependent on what consent could mean across that language barrier in 1968.
4. Filming and gift-giving staged events, including fights, for the camera. Empirical, partially checkable.
5. Chagnon's published findings about violence were methodologically unsound. A scientific dispute, not an ethics charge.
6. His work was used politically in ways that harmed the Yanomami. Historical and contested.

Bundling these produces the effect the accusation had. If claim one is true, everything else is coloured by it. If claim one is false, defenders sweep away claims three and six along with it. Neither move is legitimate.

Step two: trace the chain on the biggest claim

The vaccine used was Edmonston B, a live attenuated measles vaccine that was standard in the 1960s. It commonly produced fever and rash, which is exactly why an observer in 1968 might record vaccinated people appearing ill. The decisive question is different: can a person vaccinated with Edmonston B transmit measles to someone else?

The answer, established well before the controversy and restated by the vaccine's own developers when asked, is no. Attenuated measles vaccine does not produce transmissible infection. That single fact breaks the causal chain the accusation required. Medical bodies, including the American Society of Human Genetics and researchers who had worked on the vaccine, examined the charge and rejected it. So did the association's own investigating task force, which reported in 2002.

Notice the form of the refutation. It is not a judgement about anyone's character, and it does not depend on trusting either side. It depends on a property of an attenuated virus, checkable in the immunological literature, that determines whether the alleged mechanism could operate at all. Always look for that kind of hinge.

The point: when an accusation names a mechanism, test the mechanism first. If the mechanism is impossible, the remaining claims must be judged separately rather than inherited.

Step three: judge the remaining claims on their own evidence

The 2002 task force did not clear everything. Its findings on other counts were mixed, and some criticisms survived: about the adequacy of consent for blood collection, about representations of the Yanomami that were used by others in Brazilian and Venezuelan political disputes over Yanomami land, and about the conduct of filming.

Then the discipline complicated the picture again. In 2005 the association's members voted to rescind acceptance of the task force report, on the grounds that the inquiry itself had been improperly constituted and had judged colleagues without due process. Both things are now true: the central accusation failed, and the profession's own investigation was repudiated by its members.

The blood samples had the longest afterlife. Thousands of vials collected in the 1960s remained in American laboratory freezers for decades, while Yanomami representatives asked for their return, holding that the blood of the dead should not be kept. Repatriations began in the 2010s. Nothing about that outcome depended on the vaccine claim; it followed from a question the original researchers had not asked, about who owns biological material after the study ends.

Step four: keep science and ethics apart

Chagnon's 1988 paper in Science reported that Yanomami men who had participated in a killing had, on average, more wives and more children than men who had not, and he drew evolutionary implications from it. R. Brian Ferguson argued in 1995 that Yanomami warfare was substantially shaped by the intrusion of steel tools, missions and outposts, so that the violence being explained was itself partly a product of contact, including contact by anthropologists distributing goods. Others attacked the age controls, the classification of killers, and the comparison groups.

That is a scientific dispute about analysis and confounding, and it should be settled by evidence and reanalysis, not by the ethics charges. Weak inference is not misconduct, and misconduct does not make an inference weak. Keeping the two questions separate is the single most useful discipline in reading a controversy of this kind.

The same standard applied to a live case

In 2007 the United States Army began embedding social scientists, including anthropologists, with combat brigades in Iraq and Afghanistan under the Human Terrain System. The stated purpose was to give commanders social and cultural knowledge that would reduce violence. The association's executive board declared the programme an unacceptable application of anthropological expertise, and a commission spent two years examining it.

Run the same decomposition. What are the checkable claims?

• Consent. Can a villager decline to speak to a researcher who arrives with an armed patrol? If refusal is not available, the ethical basis of the entire method is absent, whatever the researcher intends.
• Confidentiality. Information entering a military system can be retained, shared and used for purposes the collector does not control, including targeting. A promise not to do so is not enforceable by the person making it, which is the Belfast Project lesson in a harsher setting.
• Openness. Working in uniform, or alongside people in uniform, makes the ordinary declaration of purpose impossible to hear as a free invitation.
• Harm. Identifying which families support which faction produces information that can be used against them, regardless of intent.

The counter-argument deserves its strongest form: soldiers with better cultural knowledge may kill fewer people, and refusing to participate leaves those decisions to people with no training at all. Several participants argued exactly this, and three programme staff were killed in the field. The commission's answer was not that the goal was bad but that the conditions of the work made consent, confidentiality and openness unavailable at once. The programme was closed in 2014.

Common misconceptions

• Where there is smoke there is fire. Some charges in this affair were false and some were sound. Treating them as one bundle guarantees a wrong verdict in one direction or the other.
• A professional inquiry settles the matter. This one was repudiated by a vote of the same association's members three years later.
• Bad ethics implies bad science. They are separate audits with separate evidence, and running them together lets each hide the other.
• Applied work with a military is the same as any other client. The distinguishing feature is that refusal by the researched is not freely available, which removes the basis of consent rather than complicating it.

Summing up

An accusation arrives as a story and has to be turned into a list. Separate the claims, sort them into empirical, historical and scientific, and test any named mechanism first, because an impossible mechanism collapses one claim and only that one. Judge what remains on its own evidence, remembering that the norms of 1968 are the relevant standard for what people should have done then, while the question of who owns the blood is answerable now. Keep the scientific dispute in its own file. Then apply the same procedure to the case in front of you, which for anthropology in the 2000s was a programme where the people being studied could not decline, could not be promised confidentiality, and could not hear the declaration of purpose as anything but an order.

Worth holding on to: the most useful question in any of these controversies is not who is the villain. It is what would have to be true for this claim to hold, and can that be checked.

Sources

1. Tierney, P. (2000). Darkness in El Dorado: How Scientists and Journalists Devastated the Amazon. W. W. Norton. Wikipedia
2. Chagnon, N. A. (1988). Life histories, blood revenge, and warfare in a tribal population. Science, 239(4843), 985-992.
3. Ferguson, R. B. (1995). Yanomami Warfare: A Political History. School of American Research Press. Wikipedia
4. American Anthropological Association. (2002). El Dorado Task Force Papers (2 vols.). AAA. Acceptance of the report was rescinded by member vote in 2005.
5. American Anthropological Association Commission on the Engagement of Anthropology with the US Security and Intelligence Communities. (2009). Final Report on the Army's Human Terrain System Proof of Concept Program. AAA.

Eleven months, in a box

Say you have finished: 1,340 pages of fieldnotes, 87 recorded interviews of which 40 are transcribed, three notebooks of jottings, a log, a freelist dataset, 400 photographs and a memo file you have not opened since August. You have nine months to produce something. The commonest failure at this point is not laziness; it is reading the whole pile repeatedly while waiting for a pattern to announce itself, and running out of time.

Analysis is a procedure. Here it is, applied to a single paragraph you already know: the fieldnote from Lesson 4, where the supervisor shouted about a pallet and the driver laughed and walked to the canteen.

Before coding: the inventory and the unread pass

Inventory first. One spreadsheet: every document, its date, its type, who it concerns, whether it is transcribed. You need this because your claims will eventually require statements like thirty-one of forty-three interviews, and you cannot reconstruct that later from memory.

Then read everything once, in order, taking no codes and making no highlights. At the end write a memo of about two pages headed what I think I know. This is the best three days you will spend, because it captures your headnotes while they are still live, and because it gives you something specific to be wrong about. Date it and do not revise it; it is the baseline against which you will later see your own analysis move.

Open coding, on one paragraph

Open coding means attaching short labels to small pieces of data, staying close to what is there. Two kinds of label. In vivo codes use the participants' own words. Researcher codes name an action or a relation.

Six codes from one paragraph is a normal density for a first pass. Notice that joking refusal is not in the data; it is a name for a pattern you suspect. That is legitimate in open coding and dangerous if it never gets tested, which is what the rest of the procedure is for.

The codebook, and why boundaries matter more than names

By the time you have coded fifty pages you will have 120 codes, many of them near-duplicates. Build a codebook with four fields per code: the label, a one-sentence definition, what is included, and what is excluded, plus one example from the data.

The exclusion rule is the field that does the work. For joking refusal: include instances where an instruction is answered with humour and not complied with; exclude humour that accompanies compliance, and exclude flat refusal without humour. Without that boundary you will code forty passages and discover that the category contains three different phenomena.

If more than one person codes, test agreement on the same ten pages and talk through every disagreement. The number matters less than the conversation, which is where vague definitions get exposed.

Key idea: a code is a question in compressed form. Its definition and exclusion rule are what make it answerable.

Focused coding and constant comparison

Now collapse and promote. Focused coding takes the codes that recur or that carry analytic weight and applies them systematically across the whole corpus, while merging the near-duplicates. Joking refusal, public reprimand at distance and cost absorbed rather than resolved might combine into a category called authority without sanction.

The engine underneath is constant comparison, from grounded theory as set out by Barney Glaser and Anselm Strauss in 1967: every new instance is compared with previous instances of the same code, and with instances of neighbouring codes, and the comparison forces the definition to sharpen. Incident against incident, then incident against category.

Write memos throughout. Kathy Charmaz is emphatic that memo-writing, not coding, is where analysis happens: the codes are handles, and the memo is where you say what the category consists of, how it varies, what conditions produce it and what it explains. A project with 200 codes and no memos has an index, not an argument.

From category to claim: three tests

1. Distribution. How many distinct people, on how many occasions, in how many settings? A category resting on one vivid episode is an anecdote with a label. Count it, and write the count into the text: in eleven of the nineteen instances I recorded.
2. Negative cases. Go looking for the instances that should exist if you are wrong. If joking refusal is a privilege of the driver's contract, then sorters should not do it. Find a sorter who does, and either the category changes or you have discovered a condition.
3. Alternative explanation. Write the best competing account of the same data in three sentences, then say what evidence separates them. Perhaps the laughing is about R personally rather than about contracts, in which case it should follow R to a new yard, and it does not.

What Charmaz's own work shows is how far this can go. Coding interviews with chronically ill people, she built a category she called loss of self, a form of suffering distinct from pain or disability, in which people watch their former identities become unavailable while still being required to account for themselves. It is grounded in coded interview material and it reorganised how a literature understood chronic illness. It is also bounded in ways her method makes visible: the sample was built from particular illnesses and a particular American setting, and the analysis is centred on identity rather than on the insurance and employment structures that shape it. Both halves belong in a methods section.

Thick description is not thick detail

The most misused term in the field. Thick description came to anthropology from the philosopher Gilbert Ryle, who asked what distinguishes a blink from a wink. Physically, nothing: the same contraction of the same eyelid. A wink is a signal, a conspiratorial one, in a code shared with one other person, which can also be parodied, rehearsed, faked, or misread. A thin description records a rapid eyelid contraction. A thick description says which of those things was happening, and on what grounds.

So when you write that the driver laughed, thin description is satisfied. Thick description asks what laughing meant there: who could do it, to whom, in front of whom, and what it accomplished that silence or argument would not. That is an interpretive claim, and it needs the same three tests as any other.

Where theory goes

Two bad options are common. Theory first, where a framework is chosen in advance and the data are trawled for illustrations. Theory last, where a concept is bolted on in the final draft to make the paper publishable. Both produce the same smell: a claim that would be unchanged if the theory were deleted.

The workable option is abduction, reasoning from a surprising observation to the explanation that would best account for it. You need theory to be surprised: the joking refusal is only puzzling if you hold some expectation about how instructions and authority work. Then existing concepts become a resource for asking what kind of thing this is, and your finding either extends, bounds or contradicts them. Timmermans and Tavory argued in 2012 that this, rather than grounded theory's image of a mind emptied of prior ideas, is what good qualitative analysis actually does.

Software, and what it will not do

Qualitative analysis software stores, codes, retrieves and counts. It is genuinely useful for a large corpus, for team projects and for reporting distributions honestly. It will not tell you what a category is, and it has one characteristic failure: code-and-retrieve invites you to read all 43 passages coded joking refusal in a row, stripped of the days they sat in, and sequence is often the thing that explains them. Re-read whole documents at intervals, not just coded extracts.

Common misconceptions

• Coding is analysis. Coding is preparation. The analysis is in the memos, the comparisons and the negative cases.
• Grounded theory means starting with no ideas. Nobody does, and the attempt mainly conceals the ideas you have. Make expectations explicit so the data can contradict them.
• Thick description means lots of detail. It means layered interpretation of what an act meant to the participants, supported by evidence.
• Counting betrays qualitative work. Reporting how many people and occasions support a category is elementary honesty; what would betray it is presenting those counts as population proportions.

Putting it together

Inventory the corpus, read it once without coding, and write down what you think you know. Open-code closely, mixing in vivo labels with your own, then build a codebook whose exclusion rules make each code answerable. Promote and merge into categories through constant comparison, and write memos continuously, because that is where the analysis lives. Test every category three ways: distribution across people and occasions, deliberate search for negative cases, and a written alternative explanation with the evidence that separates it. Use theory to make observations surprising rather than to decorate conclusions. And keep re-reading whole documents, because code-and-retrieve quietly destroys sequence.

The upshot: the distance between a box of notes and an argument is crossed by a named category with a boundary rule, a count attached and a failed attempt to refute it.

Sources

1. Charmaz, K. (2006). Constructing Grounded Theory: A Practical Guide Through Qualitative Analysis. Sage. Wikipedia
2. Glaser, B. G., and Strauss, A. L. (1967). The Discovery of Grounded Theory: Strategies for Qualitative Research. Aldine. Wikipedia
3. Geertz, C. (1973). Thick description: Toward an interpretive theory of culture. In The Interpretation of Cultures (pp. 3-30). Basic Books.
4. Charmaz, K. (1983). Loss of self: A fundamental form of suffering in the chronically ill. Sociology of Health and Illness, 5(2), 168-195.
5. Timmermans, S., and Tavory, I. (2012). Theory construction in qualitative research: From grounded theory to abductive analysis. Sociological Theory, 30(3), 167-186.

Ten scholars in Santa Fe, and the people who were not there

In April 1984 a seminar convened at the School of American Research in Santa Fe to discuss ethnographic writing as writing. The book that followed in 1986, Writing Culture, edited by James Clifford and George Marcus, became the most cited methodological book in late twentieth century anthropology. Among its contributors were no feminist anthropologists, an absence the introduction acknowledged and explained in terms that satisfied almost nobody.

So the dispute begins with the volume's own composition, and that is not a debating point. It shapes both sides of what follows.

The case: ethnographic authority is manufactured, and the workshop is visible

The critique is textual and it is specific. Read a classic monograph and you can name the devices that produce its authority. The ethnographic present, which converts a year of observation into a timeless account. The disappearance of the fieldworker after an opening anecdote, so that the description seems to have no author. The composite informant. Free indirect style, in which the ethnographer reports what people think without saying how that was established. The arrival scene, whose function is to certify presence and then get out of the way.

Each device does epistemic work: together they make an interpretation look like a report. Clifford's argument was not that ethnographers lie. It is that a monograph is constructed, that construction is invisible by convention, and a reader who cannot see the construction cannot evaluate the claim.

Behind the textual case sits a historical one. Talal Asad's 1973 collection documented how British social anthropology worked inside colonial administrations that supplied access, funding and safety. Edward Said's Orientalism (1978) traced how a body of European scholarship about the Middle East produced an object it then claimed to describe, in institutions bound to imperial power. Johannes Fabian, in 1983, named a habit he called the denial of coevalness: anthropologists share time with the people they study during fieldwork, then write about them in a tense that places them in another era, as earlier, traditional, not-yet-modern. That is a grammatical fact with a political consequence, and once you see it you cannot unsee it in older texts.

The replies, at full strength

Three serious answers came back, and none of them is a defence of naive objectivity.

The timing objection. Feminist anthropologists pointed out that the discovery that the observer's position shapes the account was not new in 1986; women had been documenting exactly that since the 1960s and had been treated as producing special-interest work rather than method. Mascia-Lees, Sharpe and Cohen argued in 1989 that a critique of authorship arrived, and authorship became suspect, at precisely the moment when women and formerly colonised scholars were acquiring the authority to write. Ruth Behar and Deborah Gordon assembled Women Writing Culture in 1995 to put on record the lineage the Santa Fe volume had passed over, from Hurston and Benedict onward.

The paralysis objection. If every representation is an exercise of power and every text a construction, what does a fieldworker do on Monday? The critique generated a decade of prose about the impossibility of writing and comparatively few new ways of finding things out. A graduate student who absorbed only this literature could arrive in the field unable to justify asking a question.

The displacement objection. Attention moved from the people studied to the anthropologist's own textual difficulties, which is a strange outcome for a critique launched in the name of the represented. The people whose representation was at issue were rarely in the room in Santa Fe either.

What would settle it: look at what changed

Arguments about whether a critique was correct are less useful than an audit of what it produced. Four things came out of it that are now ordinary practice, and each can be judged on results.

• Multi-sited ethnography. Marcus argued in 1995 that if your object is a commodity chain, a diaspora or a policy, one village cannot contain it, and proposed following: follow the people, the thing, the metaphor, the story, the life, the conflict. Anna Tsing's study of matsutake mushrooms follows a single fungus from Oregon forests through buyers and Japanese gift economies, and the account of value it produces could not have been written from any one of those places. The cost is real and is named by its own practitioners: thinner presence at each site.
• Autoethnography. Behar's argument in 1996 was not that anthropologists should write memoirs but that vulnerability should be admitted where it is doing analytic work. Judged by the test from Lesson 10, the good examples bound a claim and the weak ones do not.
• Collaborative ethnography. Luke Eric Lassiter and colleagues restudied Muncie, Indiana, with African American residents as co-researchers and co-authors, on the ground that the classic Middletown studies of the 1920s and 1930s had written a city as though its Black population were not there. Consultants read drafts and changed them. The cost is time and negotiated compromise; the gain is a text the community can use and a set of errors caught before publication.
• Indigenous and decolonial methodologies. Linda Tuhiwai Smith's 1999 book begins from the proposition that research is, for many Indigenous communities, a word associated with extraction, and sets out an agenda in which communities set questions, control data and receive results. That is now embedded in tribal review boards and data governance frameworks of the kind covered in Lesson 11.

In short: the critique's durable legacy is procedural, not stylistic. Positionality statements, co-authorship, consultant review of drafts, community data governance and multi-sited designs all survive. The experimental prose forms of the late 1980s mostly did not.

Reading the two sides against each other

The table is where this lesson wants you to end up. Both columns contain something you should carry into your own project, and the second column is the one that generates tasks.

Common misconceptions

• The crisis showed that ethnography is just literature. It showed that ethnography is also writing, with conventions that can be examined. Fieldnotes, counts and language competence remain evidence.
• Writing Culture invented reflexivity. Feminist and formerly colonised scholars had been working on position and authority for years before Santa Fe, which is the substance of the timing objection.
• Multi-sited means several field sites. It means following an object, a person or a conflict across the sites it actually crosses, with the design justified by the object rather than by variety.
• Collaboration means letting participants approve everything. It means negotiated authorship and draft review, with disagreements documented rather than resolved by either party's fiat.

Looking back

The dispute was about whether ethnographic authority is earned or manufactured, and the honest answer is that it is both, in proportions you can inspect text by text. The critique's evidence was real: devices that hide the author, a colonial history of access, a grammar that put the people described in another time. The replies were also real: the critique arrived exactly as new authors gained standing, it produced more prose about writing than new ways of knowing, and it drifted towards the anthropologist's own difficulties. What settles it, for a working fieldworker, is the audit of practice. Multi-sited designs, collaborative authorship, consultant review, community data governance and honest positionality all descend from this argument and all give you something to do.

What matters here: when you write your methods section, the useful question is not whether you can legitimately represent anyone. It is which specific practices in your project make your construction visible and correctable.

Sources

1. Clifford, J., and Marcus, G. E. (Eds.). (1986). Writing Culture: The Poetics and Politics of Ethnography. University of California Press. Wikipedia
2. Said, E. W. (1978). Orientalism. Pantheon Books. Wikipedia
3. Marcus, G. E. (1995). Ethnography in/of the world system: The emergence of multi-sited ethnography. Annual Review of Anthropology, 24, 95-117.
4. Mascia-Lees, F. E., Sharpe, P., and Cohen, C. B. (1989). The postmodernist turn in anthropology: Cautions from a feminist perspective. Signs, 15(1), 7-33.
5. Smith, L. T. (1999). Decolonizing Methodologies: Research and Indigenous Peoples. Zed Books.

The same forty seconds, told three ways

One scene, the pallet in bay two from Lesson 4, written in three registers. Read them aloud; the differences are audible.

Realist. Drivers at the depot routinely answer supervisors' instructions with humour rather than compliance. When the blue truck arrived twenty minutes late, the shift supervisor shouted across the yard about a misplaced pallet; the driver laughed, said "you always do this", and continued to the canteen. The pallet remained in place for the rest of the morning, and the loader driver reversed twice to avoid it. In eleven of nineteen recorded instances, instructions to drivers were met this way.

Confessional. I had been at the depot nine days and still did not know whether I was allowed in the yard. When R started shouting I assumed a fight was beginning and felt my notebook become conspicuous in my hand. Nothing happened. C laughed, walked past him and went for tea. It took me three more weeks to understand that I had been watching the normal operation of a boundary between two contracts, and that my own alarm had been the only unusual event in the yard that morning.

Impressionist. A shout crosses forty yards of wet concrete and lands on nobody. The laugh that answers it is short, almost kind. Two women on the sorting line do not look up, and their hands do not slow. The pallet stays where it is, and all morning the loader swings wide around it, the way traffic goes around a pothole everyone has agreed not to mention.

All three are honest. They do different jobs, and the choice among them is an argumentative decision rather than a matter of taste.

Van Maanen's tales, compared

John Van Maanen, who did his fieldwork with police officers, set out the typology in 1988 and it remains the most useful map of the options.

Read the failure column and the practical rule follows: a monograph needs more than one register, deployed on purpose. Use impressionist passages to establish a scene, realist prose for the claims that must be checkable, confessional passages exactly where access explains a limitation, and formal analysis where you are making the case travel. What you must not do is let the register drift, so that an atmospheric sentence carries the weight of an empirical claim.

What matters here: every register is a promise to the reader about what kind of evidence is behind the sentence. Breaking that promise is the one writing error that is also a research error.

Tense, and where to put the ethnographer

Three concrete rules will improve almost any draft.

1. Date your observations. Write in March 2019, or in my second month, rather than in a timeless present that converts one year into eternity. Reserve the present tense for claims you intend as current generalisations, and then own them as such.
2. Attribute every mental state. Not the drivers resented the new rota, but three drivers said the rota was designed to catch them out, and two others described it without complaint. If you inferred it rather than heard it, say so in the sentence.
3. Put yourself where you were. If you were in the office when the argument happened in the yard, the sentence should say so. A reader cannot correct for a vantage point you have concealed.

The author-function

Michel Foucault argued in 1969 that an author's name does more than identify who wrote a text: it organises how the text is read, classified and granted authority. What Is an Author? is worth knowing here because ethnographic authority works exactly this way. A monograph by an established anthropologist on a society they have visited for thirty years is read as testimony; the identical sentences in a student thesis are read as a report.

Geertz, in 1988, read four anthropologists as authors rather than as sources, and described the signature problem of the genre as the need to be convincing about two things at once: that the writer was really there, and that the writer is here now, making an argument to colleagues. Malinowski solved it with the confessional I of the introduction and the absent narrator of the chapters. Benedict solved it by style. Levi-Strauss solved it by writing a travel book and a structural analysis in the same volume. None of the solutions is neutral, and you will adopt one whether or not you notice.

Evocation against explanation

The sharpest disagreement about ethnographic writing is whether the goal is to explain or to evoke. The explanatory position, Geertz's, is that the ethnographer interprets: you say what the wink meant and you show the grounds. The evocative position, argued most strongly in the 1986 Writing Culture volume, is that interpretation imposes a false coherence, and that a text should instead produce in the reader an experience analogous to the fieldworker's, fragmentary and unresolved.

Behar's version is narrower and more usable: write with vulnerability where vulnerability is doing analytic work. Her formulation, that anthropology which does not break your heart is not worth doing, is a claim about what a reader must feel in order to understand certain findings, particularly about suffering. Applied carelessly it licenses self-display. Applied with the Lesson 10 test, it produces passages that do something no realist summary could: they make a reader understand why a person acted in a way that looks irrational in a table.

Where does this leave a graduate student with a thesis to write? With a division of labour. Evocation earns its place when the finding is about experience and the reader's incomprehension is the obstacle. Explanation is required wherever you are making a claim someone might dispute, because an evocative passage cannot be argued with, and a claim that cannot be argued with cannot be supported either.

Quoting people, and cleaning up speech

Transcribed speech on the page is full of repetitions, false starts and fillers. Printing it raw can make a fluent person look incoherent, especially across a class or dialect difference; cleaning it silently misrepresents what was said. The workable convention, now common, is light editing declared once in a note: false starts and fillers removed, no words added, pauses marked where they matter. Never standardise dialect into your own register; it changes who the speaker is on the page.

Two further rules. Quote at length when the speaker's own categories are the finding, and paraphrase when you only need the content, because a page of block quotation is often an abdication of analysis. And check long quotations with the speaker where you can, which is both an ethical courtesy and the cheapest accuracy check available.

The methods appendix, and why Whyte started it

Street Corner Society acquired an appendix in its 1955 edition in which Whyte described how he got in, what went wrong, who sponsored him and how the study changed shape. It became a genre, and at its best it is the single most useful part of an ethnography: Mitchell Duneier's Sidewalk (1999) includes a long methodological statement covering how he paid people, what he recorded, who read drafts and where his account might be wrong.

Write yours to answer six questions. How long were you there and doing what. How were people sampled, and who refused. What was recorded and how, including the share of material that is notes rather than transcript. What you could not access, with the reason. What you promised participants and what you did about anonymity. And what in the account you consider least secure. The last one costs nothing and buys more credibility than any amount of assertion.

Common misconceptions

• Good ethnographic writing means literary writing. It means a register matched to the kind of claim. An unfalsifiable beautiful page is a research failure wearing good clothes.
• The first person undermines authority. It locates the observation. What undermines authority is an unlocated claim about what people think.
• Quotations speak for themselves. Selection, trimming and framing are analytic acts, and block quotation often substitutes for the analysis it appears to provide.
• A methods appendix is a confession of weakness. Naming what is least secure is how a reader learns to trust the rest, and it has been a respected convention since 1955.

The short version

Writing is part of the method, not a wrapper around it. Choose registers on purpose: impressionist to put the reader in the yard, realist for claims that must be checkable, confessional where access explains a limit, formal where the case has to travel, jointly told where disagreement is the finding. Date observations, attribute every mental state, and say where you were standing. Accept that your name and position will do authority work whatever you write, and compensate with a methods appendix that answers the six questions, including what you consider least secure. Quote at length only when the words are the finding, edit lightly and say that you did, and never translate a person's speech into your own register.

Bottom line: the same forty seconds can be three true paragraphs. The error is not choosing among them; it is letting a sentence claim more evidence than its register actually carries.

Sources

1. Van Maanen, J. (2011). Tales of the Field: On Writing Ethnography (2nd ed.). University of Chicago Press. Wikipedia
2. Foucault, M. (1977). What is an author? In D. F. Bouchard (Ed.), Language, Counter-Memory, Practice (pp. 113-138). Cornell University Press. Wikipedia
3. Geertz, C. (1988). Works and Lives: The Anthropologist as Author. Stanford University Press.
4. Behar, R. (1996). The Vulnerable Observer: Anthropology That Breaks Your Heart. Beacon Press.
5. Duneier, M. (1999). Sidewalk. Farrar, Straus and Giroux.

An anthropologist buys land in a world that does not exist

Between 2004 and 2007 an anthropologist logged into Second Life as an avatar called Tom Bukowski, acquired a plot of virtual land, built a house on it that he named Ethnographia, and held office hours there. Tom Boellstorff did participant observation the way Module 2 describes it: long residence, jottings, interviews, a census of a community, all conducted inside the world. His 2008 book took its title from Margaret Mead's Coming of Age in Samoa, and the echo was the argument: this is a place where people have a culture, and the discipline's ordinary method applies.

He made one strict methodological decision that is still argued about. He studied Second Life in its own terms, which meant he did not routinely seek out residents' offline identities, travel to meet them or treat the offline person as the real one behind the avatar. If two residents held a wedding, that was a wedding in the place where it occurred.

The payoff was a demonstration. Residents built kinship-like relations, property norms, labour markets, etiquette, grief practices for departed friends, and a sense of place dense enough that land disputes mattered. None of that could have been established by a survey of users, and the avatar-to-avatar interviews produced talk that an offline interview would have reframed.

The cost was equally clear, and critics named it quickly. By declining to follow people offline, the study could not connect what residents did in-world to the lives that paid for their computers, and questions about who can afford the time, who is performing gender or disability differently online, and what offline constraints show up in-world were left open by design. A second limit was invisible at the time: Second Life was governed by a company, and every norm Boellstorff described sat on top of terms of service, pricing and software decisions made by Linden Lab. When user numbers fell away, the field site itself contracted, which is a thing villages rarely do.

Nine sites, fifteen months each

The second landmark took the opposite approach to scale. Daniel Miller assembled a team of nine anthropologists who each spent about fifteen months in a different place, from an industrial Chinese town to a Syrian-Turkish border city, an English village, a Brazilian town, a Chilean mining settlement, an Italian town, a south Indian village and Trinidad, asking the same question: what are people actually doing with social media?

Three findings came out of the comparison that a single site could not have produced. First, the platforms mean different things in different places: in one site Facebook was where family obligations were performed and intimacy happened elsewhere, while in another the pattern was reversed, so the same software supported opposite social functions. Second, what matters is the whole repertoire rather than any one app: Miller and Mirca Madianou's concept of polymedia holds that once several channels are available at negligible cost, the choice of channel itself becomes a social statement, so sending a message by text rather than voice call is itself the message. Third, the team proposed scalable sociality: social media let people calibrate group size and intimacy along a continuum between private dyads and broadcast publics, which older categories of mass and interpersonal communication could not describe.

The costs of that design are the mirror image of Boellstorff's. Standardising a question across nine sites required compromises that each ethnographer could have spent the time pursuing locally, and the results aged fast, because the platforms described in 2012 and 2013 fieldwork had changed their products and their populations before the books were widely read. The team's response, publishing openly and quickly, is itself a methodological lesson about doing fieldwork on moving objects.

What counts as the field

The practical question both projects answer differently is where the boundary goes. Three workable definitions, each with a consequence.

• The platform as site. One world, game, forum or server, studied in its own terms. Strong on internal norms, weak on the relation to offline life, and dependent on a company's continued existence.
• The practice as site. Follow an activity, such as remittance messaging or raid organising, across whatever channels people use. This is multi-sited logic applied to media, and it matches how people actually behave, at the cost of never having a complete view of any one channel.
• The person as site. Follow a set of people across their devices and platforms and into their households. Most informative about the relation between online and offline, most expensive, and the hardest to get consent for, because it means reading a person's whole communicative life.

Whichever you choose, state it and state what it excludes. A study of one subreddit that makes claims about a political movement has smuggled in the practice-as-site design without doing the work.

Methods that transfer, and the ones that need rebuilding

1. Participant observation transfers, but lurking is not participation. Reading a forum for six months without posting is archival research with a live feed. Fieldwork means taking part and accepting the consequences of being known, which online includes being blocked, mocked and screenshotted.
2. Jottings and notes still matter. The temptation is to rely on the archive, since everything is already text. Resist it: what you need is the record of what you noticed and what you were doing when, which the platform does not store.
3. Interviews need a channel decision. Text interviews give people time to compose and produce quotable prose; voice gives hesitation and tone; in-world or in-game interviews keep the register of the site. Choose deliberately and report the choice, since it changes the data as much as an interpreter does.
4. New techniques are worth learning. Scroll-back, in which you go through a person's own timeline with them and ask about specific posts, turns the archive into elicitation. The app walkthrough, in which you document an interface step by step as a designed environment, makes the platform's own priorities visible before you start attributing everything to users.
5. Remember whose feed you are seeing. Algorithmic ranking means your screen is not your informant's screen. Ask people to show you theirs, and never treat your own feed as the field's content.

Ethics online, where the old rules break

Guidance from the Association of Internet Researchers has been developed precisely because the standard protocol assumptions fail here. Four points are non-negotiable in a protocol.

• Public is not published consent. A post visible to anyone was written for an imagined audience, and that audience was not a researcher with a grant. Accessibility is not permission.
• Verbatim quotation is identifying. A pseudonym protects nothing if the quoted string can be pasted into a search box. Paraphrase, or seek permission to quote, or alter non-essential wording and say that you have.
• Terms of service are not ethics. A platform's rules may permit scraping that your obligations to participants forbid, and may forbid research that is ethically sound. Decide separately, and document both.
• Aggregated data re-identify. Published research datasets drawn from social and dating platforms have repeatedly been re-identified after release, and the same logic that defeats anonymisation in a small village operates on a profile with a few attributes.

Add one more that is specific to the present decade: you cannot assume an account is a person. Automated and synthetic accounts are common enough that identity verification has become part of sampling, and a study of sentiment in a hashtag may be a study of software.

Common misconceptions

• Online fieldwork is cheaper and faster. It removes travel and adds archive management, channel decisions, platform change and harder consent. Miller's team still spent fifteen months each.
• Virtual worlds are not real field sites. Boellstorff's demonstration is that practices with their own norms and consequences can be studied where they occur. The open question is what is gained by also following people offline.
• Scraping a million posts beats interviewing thirty people. They answer different questions. Trace data show distribution; only people can tell you what a post was for.
• A public post can be quoted freely because it is public. Searchable verbatim text is identifying, and writing for an open audience is not consent to be studied.

What to carry forward

Two landmark designs, two opposite strategies, both defensible. Boellstorff treated one platform as a place and proved that participant observation transfers intact, accepting that he could not link the world to the lives outside it. Miller's team traded depth at each site for a comparison across nine, and discovered that the same software performs opposite social functions in different towns, that the repertoire matters more than the app, and that findings about platforms decay fast. For your own design: say whether your site is a platform, a practice or a person, and what that excludes. Participate rather than lurk. Keep notes even though the archive exists. Choose your interview channel deliberately and report it. And treat public posts as other people's speech in their own context, which means quotation is an ethical decision and not a matter of access.

Remember: a company can close your field site. Build a design, and an archive of your own notes, that survives the platform you are studying.

Sources

1. Boellstorff, T. (2008). Coming of Age in Second Life: An Anthropologist Explores the Virtually Human. Princeton University Press. Wikipedia
2. Miller, D., et al. (2016). How the World Changed Social Media. UCL Press. Wikipedia
3. Madianou, M., and Miller, D. (2013). Polymedia: Towards a new theory of digital media in interpersonal communication. International Journal of Cultural Studies, 16(2), 169-187.
4. Hine, C. (2015). Ethnography for the Internet: Embedded, Embodied and Everyday. Bloomsbury.
5. Light, B., Burgess, J., and Duguay, S. (2018). The walkthrough method: An approach to the study of apps. New Media and Society, 20(3), 881-900.

A queue at twenty to seven

At 6:40 on a Tuesday morning, outside a food bank in a town of about forty thousand people, the queue is thirty people long and roughly a third of them are wearing work uniforms: care tabards, a supermarket fleece, high-visibility jackets. They will be at work by nine. This is the worked case for the whole lesson, and it is invented for teaching purposes, though the pattern it describes is ordinary in many countries.

The puzzle arrives immediately, and it has the right shape: emergency food aid is designed and described as a response to the absence of income, and a third of this queue has income. Either the aid system is being used for something other than what it was built for, or wages have stopped doing what everyone assumes wages do, or both. Now build the whole project.

Step one: from puzzle to question

Draft: what is the experience of in-work poverty? Unanswerable and unbounded. Second draft: how do employed people in this town come to use emergency food aid? Better, still vague about what would count as an answer.

Working question: over twelve months in one town, how do employed households come to rely on emergency food provision, what sequences of income, cost and obligation produce that reliance, and how do the people distributing food understand the households they serve? Site and period are fixed. It requires two populations, users and distributors, in the manner of Lesson 2. And it can come out otherwise: if reliance turns out to follow single catastrophic events rather than sequences, the framing is wrong and you will know it.

Before going further, check whether it is answered. The regional literature, government statistics on in-work poverty, charity sector reports, and comparative material in the Human Relations Area Files for household provisioning under stress. If somebody has done this in a comparable town, your contribution is a comparison, and you should say so.

Step two: sites, chosen for contrast

Two distribution points rather than one, selected to differ on the factor you think matters: a church-run food bank requiring a referral voucher, and a community pantry with a small weekly membership fee and no referral. Contrastive case logic makes the referral system itself analysable rather than background. Add a third setting where the households are: two employers in the town who consent to workplace access, giving you the wage side of the sequence.

Step three: methods, with numbers attached

Two design properties are worth naming. The blind column is populated on purpose, and no cell is empty: every method is paired with another whose blindness differs. And the systematic components are built out of the qualitative ones, in the order Module 3 taught: freelists first so the pile sort uses local terms, spot observation only once you know which categories are worth counting.

Step four: the arithmetic that cuts the design

Now cost it in hours per week, using the ratios from Lessons 2 and 4.

• Observation: 6 hours. Notes at two and a half times: 15 hours. Subtotal 21.
• Interviews: 45 across the year, averaging 1.5 hours recorded. At five hours of transcription and coding per recorded hour, that is roughly 340 hours over 36 weeks, or about 9.5 hours a week. Subtotal 30.5.
• Systematic tasks: freelists, pile sorts, 200 spot observations and 25 network interviews, spread across the year, about 4 hours a week including travel and entry. Subtotal 34.5.
• Analysis, memos, admin, reading and reporting: at least 6 hours. Total about 40.

That is a full-time year with nothing left for illness, transport, dead days or the second half of the analysis. So cut, deliberately and on the record: reduce interviews from 45 to 32, transcribe 20 in full and use timed summaries with selective transcription for the rest, and drop the network component to 15 respondents. Write down what was cut and why, because that paragraph becomes your limitations section, and a limitations section written in advance is honest in a way that one written afterwards never is.

Bottom line: a design that has not been costed in hours is a wish. The cut list is part of the design, not an admission of failure.

Step five: ethics, as procedures

1. Permissions: both charities' trustees, the two employers, and institutional ethics review under the framework in 45 CFR 46 or its local equivalent.
2. Consent: oral, logged, and use-specific, with a waiver of documentation requested on the grounds that a signature linking a named person to food bank use is itself the risk.
3. Repeat consent before any interview and again before publication for anyone whose case is described at length.
4. Anonymisation stated as partial: one town, two named-in-practice charities, distinctive employment histories. Negotiate specific passages rather than promising disguise.
5. Data plan: coded names at the point of recording, key held separately, recordings destroyed on a stated schedule, nothing in a cloud service the participants were not told about.
6. Return: a short plain-language report to both charities before submission, and a commitment to send anything quoting a person to that person first.

Step six: positionality with consequences

Four sentences, each ending in a consequence, per the Lesson 10 rule. If you have never used a food bank, say so and note that your questions about shame will be answered by people managing your reaction as well as their own. If you are the same age and background as most volunteers, expect to be absorbed into the volunteer side and plan a deliberate crossing to the queue side, at a cost in volunteer trust. If you are visibly not from the town, expect the practice-as-usual explanations people give outsiders, and plan to check them against observation. If you are from the town, expect to know people in the queue and write down how you will handle recognition before it happens.

Step seven: analysis and writing, planned in advance

Analysis follows Lesson 13: inventory, unread pass with a dated what-I-think-I-know memo, open coding of the first fifty pages, a codebook with exclusion rules, focused coding, continuous memos, then the three tests on every category, distribution, negative cases, and a written alternative explanation. Add one systematic check the design makes possible: run consensus analysis on the eligibility questions asked of distributors, because if the eigenvalue ratio is low you have discovered that the volunteers do not share one theory of who deserves help, which is a finding about the institution.

Writing follows Lesson 15. Impressionist passages for the queue at 6:40, realist prose for every claim about sequences and proportions, one confessional passage about the crossing from volunteer side to queue side because it explains an access limit, and a methods appendix answering the six questions including what is least secure.

Step eight: the pre-mortem

Before starting, write the failure. It is eighteen months from now and the project produced nothing usable. Why?

• Access collapsed because one charity's trustees changed their minds after a newspaper story. Mitigation: two sites from the outset, a written agreement, and a named contact at each.
• The queue would not talk because people are ashamed and in a hurry. Mitigation: volunteer for six weeks before asking anything; conduct interviews away from the site; pay for time.
• You over-sampled the willing. Mitigation: a sampling matrix filled in weekly, and a target of two refusals or hard-to-reach households approached for every easy one.
• Notes never got written. Mitigation: the 6-hour observation ceiling exists for this reason; treat it as a hard limit rather than an aspiration.
• The town changed. A benefit rule changed halfway through and the population shifted. Mitigation: this is data; date everything and treat the change as a natural experiment.

What this design cannot see

State it before a reviewer does. It cannot estimate how many employed households in the town use food aid, because the sample is not a probability sample of the town. It cannot see households that never approach either site, which is the population most likely to be in the worst position. It cannot follow money inside a household beyond what people report, and Lesson 5 says those reports will be unreliable on frequencies. It cannot say what happens after the fieldwork year. And it will over-represent people who queue at times you can be present.

Common misconceptions

• A good design is fixed before departure. It is written before departure so that changes are visible as changes, dated in the log and defended in the appendix.
• The limitations section is written at the end. The honest version is written at the design stage, from the cut list.
• More methods make a stronger study. Only if their blindnesses differ and you have costed the hours. Three well-run methods beat seven half-run ones.
• The pre-mortem is pessimism. It is the cheapest available insurance: every failure you can name in advance has a mitigation that costs almost nothing at the design stage.

What you now know

You can now take a puzzle from a queue at twenty to seven and turn it into a project: a bounded question with two populations, contrastive sites, a methods table whose blind spots are deliberately mismatched, a budget in hours that forces a cut list, an ethics procedure built on logged use-specific consent rather than a signature, a positionality statement whose sentences end in consequences, an analysis plan with three tests per category, a writing plan that assigns registers to kinds of claim, and a pre-mortem with mitigations. That is the whole course, applied once.

Worth holding on to: the sequence that matters is question, then site, then methods, then hours, then cuts. Most failed projects reverse it, choosing methods they like and discovering the question in month nine.

Sources

1. Bernard, H. R. (2017). Research Methods in Anthropology: Qualitative and Quantitative Approaches (6th ed.). Rowman and Littlefield.
2. LeCompte, M. D., and Schensul, J. J. (2010). Designing and Conducting Ethnographic Research: An Introduction (2nd ed.). AltaMira Press.
3. Human Relations Area Files. (n.d.). eHRAF World Cultures. Yale University. hraf.yale.edu
4. Office for Human Research Protections. (2018). 45 CFR 46: Federal Policy for the Protection of Human Subjects. U.S. Department of Health and Human Services. hhs.gov
5. Desmond, M. (2016). Evicted: Poverty and Profit in the American City. Crown.