The Gold Standard of Anthropology | Cultural Consensus, Informant Competence and Shared-Belief Estimates

1. Begin with a bounded knowledge domain

Back to contents

Consensus is always about a defined set of questions and a defined group. “What this community believes” is too broad. A defensible study may ask how experienced coastal fishers classify local currents, how birth attendants recognise danger signs or how residents rank appropriate condolence practices. The researcher must show that the items belong together and that participants are positioned to know the domain. Agreement outside that frame should not be interpreted as general cultural unity.

2. Shared belief is not objective truth

Back to contents

The estimated answer key is the response pattern that best explains agreement under the model. It need not correspond to biomedical, legal or scientific truth. A group can share an inaccurate belief; subgroups can also disagree for principled reasons. The analysis therefore estimates culturally shared knowledge or belief, not universal correctness. The wording of the claim should preserve that distinction.

3. Understand the three core assumptions

Back to contents

The classic model assumes a common truth or shared answer pattern, local independence of respondents conditional on that pattern, and homogeneity of items within the domain. These are empirical commitments, not formalities. Close relatives who copied one source may agree without independent knowledge. Mixed domains may reward different kinds of expertise. A strong first factor is informative only after these possibilities have been investigated.

4. Read competence as model alignment

Back to contents

An informant competence estimate describes how consistently a person’s responses align with the inferred shared pattern for these items. It is not a score of intelligence, virtue, sincerity or cultural authenticity. Negative or very low estimates can signal guessing, reversed coding, a different interpretation, weak domain knowledge or a distinct subgroup. Researchers should inspect the response process before labelling a person.

5. Diagnose one culture before estimating an answer key

Back to contents

A common diagnostic compares the first and second eigenvalues of the respondent agreement matrix and examines whether first-factor loadings are mostly positive. A dominant first dimension supports—but does not prove—a single shared model. Researchers should inspect residual patterns, alternative codings, item difficulty and subgroup clustering. A convenient ratio cannot rescue a badly framed domain or dependent sample.

6. Worked example: four remedies and six healers

Back to contents

Six healers answer whether four plant preparations are appropriate for a named fever syndrome. Five agree on most items; one reverses nearly every response. The analyst checks the questionnaire and learns that the sixth healer classified the syndrome under a different local name. Recoding nothing would produce a low or negative competence estimate and a misleading answer key. After a follow-up interview, the mismatched item meaning is reported and the two diagnostic categories are analysed separately. Agreement becomes evidence because interpretation came before calculation.

7. Separate shared knowledge from response style

Back to contents

Yes-saying, acquiescence, extreme-category use and shared interview cues can create agreement unrelated to knowledge. Balance item direction where appropriate, pilot translations, record uncertainty and compare response formats. For multiple-choice or rank data, use a model suited to that format. If respondents discuss answers together, note that independence has been weakened rather than treating the extra agreement as stronger consensus.

8. Changed-context transfer check: a school misconception

Back to contents

Suppose teachers want to know which ideas about fractions are shared by a class. A consensus model might recover the most common answer pattern, but calling it an answer key would be wrong because the curriculum supplies an external standard. The transferable method is to compare peer agreement with known correctness: consensus can reveal a shared misconception, while competence relative to the cultural pattern should not replace achievement relative to mathematics.

9. Know when consensus is the wrong model

Back to contents

Do not force one-culture consensus when factions, roles or generations plausibly use different models. Do not treat a convenience sample as a population. Very small item sets, ambiguous questions, severe missingness and deterministic copying can make estimates unstable or uninterpretable. Qualitative disagreement may be the main finding. In those cases, mixture models, subgroup analysis or descriptive ethnography can be more honest than one pooled answer key.

10. Audit the shared-belief inference

Back to contents

Archive item development, translations, sampling logic, response format, missing-data rules and the complete agreement diagnostics. Refit the model after removing suspect items, changing coding assumptions and examining plausible subgroups. Compare the inferred key with interviews, observation and documentary evidence. Report uncertainty and show whether the substantive interpretation survives these checks.

11. Frequently asked questions

Back to contents

Does high agreement prove a shared culture? No. Shared sources, response cues or narrow sampling can also create agreement. Is competence a personal trait? No; it is domain- and model-specific. Can experts and novices be analysed together? Only if one shared model remains defensible; otherwise roles may require separate analyses. Must the inferred key be factually correct? No. Cultural consensus concerns shared responses, which can differ from external truth.

12. The gold standard makes agreement interpretable

Back to contents

Consensus analysis is strongest when it joins formal modelling to careful ethnography. Define the domain, examine who can know it, test whether one shared pattern is plausible, and interpret competence narrowly. Then challenge the result with subgroup, wording and response-process checks. The aim is not to turn majority answers into truth; it is to explain when patterned agreement supports a transparent claim about shared knowledge.