The Gold Standard of Anthropology | Cross-Site Ethnography, Protocol Drift and Distributed Interpretation

1. Begin with one phenomenon, not one script

Back to contents

A cross-site project needs a common object of inquiry: for example, how supply delays are managed, how families navigate a service, or how authority is negotiated at a boundary. It does not need every researcher to collect identical sentences in identical order.

Write the common phenomenon, comparison unit and intended inference before writing the field guide. A protocol is useful when it helps each team member recognise evidence relevant to that question. It becomes harmful when procedural sameness substitutes for conceptual clarity.

2. Distributed interpretation is part of the design

Back to contents

Researchers are not interchangeable sensors. Language, biography, institutional role, relationships and time in the field influence what becomes visible and what participants will say. Those differences create risks, but they also reveal how the field is organised.

Record who observed, through which relationship, in what language and under which access conditions. Team discussion should ask not only whether accounts agree, but why a particular researcher could encounter one account and not another.

3. Build a shared core with local extensions

Back to contents

Use a small set of required elements across all sites: common research questions, event boundaries, participant roles, minimum context fields, ethical rules and memo prompts. Add site modules for local institutions, categories, calendars and risks.

Mark each field as required, optional or site-specific. This separation prevents local learning from being mistaken for protocol failure and prevents an improvised local question from silently becoming a universal measure.

4. Version protocol changes as evidence accumulates

Back to contents

Protocol drift is any change in what is asked, noticed, recorded or coded across time or sites. Some drift is accidental: a researcher shortens a prompt or uses a familiar but narrower translation. Some is warranted learning: a local distinction proves essential and the guide must change.

Keep a dated change log with the trigger, decision, sites affected and consequences for earlier material. If a new distinction matters, decide whether to recode earlier notes, collect a bridging sample or limit comparisons to the period after the change.

5. Calibration meetings need concrete cases

Back to contents

General reminders to “take rich notes” do little. Bring the same vignette, transcript excerpt or observation sequence to the team. Ask each researcher to mark the event start and end, actors, action, local term, inferred meaning and uncertainty.

Compare the resulting records. Disagreement may reveal an ambiguous rule, a site difference, a translation problem or a valuable alternative interpretation. Resolve procedural ambiguity while preserving substantive disagreement in the analytic record.

6. Worked example: waiting is not one event

Back to contents

Two researchers study supply delays. At a wholesale market, one measures waiting from arrival until goods are loaded. At a neighbourhood shop, another begins when the owner first calls a supplier and includes credit negotiation, substitute sourcing and relationship repair. Their first comparison suggests much longer waiting at the shop.

The team discovers that the protocols identify different events. It creates a shared event record: trigger, first request, physical queue, interruption, credit relation, substitute action and resolution. Earlier notes are recoded where possible. The final account compares elapsed time and social work separately, retaining the shop’s local category for obligation rather than translating it simply as delay.

7. Construct the cross-site object explicitly

Back to contents

A global or multi-site phenomenon does not appear by pooling notes. It is constructed through stated links: a shared supply chain, policy, professional standard, platform, migration route or recurring relation. Show how each site contributes to that larger object.

Use matrices or arrays to place comparable events alongside local context, then write site memos before synthesis. The matrix makes gaps visible; the memo prevents the cells from becoming detached fragments.

8. Changed-context transfer check

Back to contents

Move the design to a digital ethnography of moderation teams in three online communities. The shared core still covers event boundaries, roles, rule invocation and outcome. Local extensions now include platform tools, asynchronous time, deleted material and pseudonymous identities.

A screenshot is not automatically equivalent to an in-person observation, and a missing message may reflect deletion rather than absence. The transfer succeeds only when the observation conditions and platform affordances are added to the common record.

9. Limits and common misconceptions

Back to contents

A single manual cannot remove researcher effects, and high inter-coder agreement cannot prove that the shared category is meaningful in every site. More sites do not automatically increase representativeness when access follows the same organisational gatekeepers.

Do not treat local adaptation as contamination, or every difference as cultural. Do not merge translated terms merely because they share an English gloss. Do not let a headquarters team resolve disagreements without the field researchers who know the contexts.

10. Audit the distributed evidence chain

Back to contents

For each cross-site claim, trace the contributing excerpts, site memo, code definition, protocol version, researcher and translation decision. Inspect which sites supply most examples and which appear mainly as absence.

Reanalyse a sample under the previous and revised rules. Run a leave-one-site-out check on headline interpretations. If the conclusion disappears without the most intensively observed site, report that dependence rather than presenting it as a stable multi-site pattern.

11. Frequently asked questions

Back to contents

Must every site use the same interview guide?

No. A common core supports comparison, while justified local prompts protect relevance and meaning. The differences must be documented.

Is protocol drift always an error?

No. Drift can reflect learning. It becomes a problem when the change is invisible or its effect on comparability is ignored.

Who has the final interpretive authority?

The project should define a transparent process. Site researchers, translators and participants may hold evidence unavailable to a central analyst; unresolved differences can remain in the report.

12. The gold standard coordinates without flattening

Back to contents

Strong cross-site ethnography uses coordination to make evidence comparable and reflexivity to keep that comparison honest. It shares questions, not assumptions; records changes, not just final rules; and treats researcher differences as part of the evidence.

Before accepting a cross-site claim, ask whether each site observed the same kind of event, whether protocol versions can be traced and whether local meanings remain visible in the synthesis. A defensible answer should survive all three questions.