THE GOLD STANDARD OF ANTHROPOLOGY
Find your next learning step
A static network says who was connected. An event log says who did what, with whom, when, and in what order. That distinction matters when an anthropologist is studying coordination, care, exchange or authority as a process rather than a finished diagram.
Start with the mechanism, follow the worked case, then change the context and see whether the explanation still holds.
Return to the Gold Standard of Anthropology · Learning that transfers into life · The Gold Standard hub · How X Works
Chapter contents
Route A · Events, ties and observation
Route B · Aggregation and worked analysis
Route C · Transfer, limits and misconceptions
Route D · Audit, FAQs and sources
CHAPTER 1 OF 12 · UNDERSTAND
1. A tie is not the same thing as an event
Start with the record
A tie is a summary: two people are connected over some chosen period. An event is an occurrence: A asks B for help at 09:12, B gives C a tool at 09:17, or a household joins a meeting at 19:00. Event data preserve timing and sequence, so they can support questions about reciprocity, escalation and response.
The first design decision is therefore not which network statistic to calculate. It is what counts as an event, how its actors are identified, and what clock and boundary define the observation.
CHAPTER 2 OF 12 · UNDERSTAND
2. Separate interaction, co-presence and opportunity
Two people can share a room without speaking, exchange messages without sharing a place, or appear co-present because a sensor has coarse spatial resolution. Co-presence is often an opportunity for interaction, not proof that an interaction occurred.
| Record | What it supports | What it does not prove |
|---|---|---|
| Observed conversation | A witnessed interaction | Every exchange outside the observer’s view |
| Bluetooth proximity | Devices were near within a technical threshold | Attention, conversation or social meaning |
| Attendance roster | People were registered in the same setting | Simultaneous presence throughout the session |
Contents · Previous chapter · Back to contents · Next chapter
CHAPTER 3 OF 12 · UNDERSTAND
3. Design the event schema before collection
A useful log normally records sender or initiator, receiver or target, event type, start time, end time or duration, setting, collection channel and uncertainty. Undirected events are possible, but direction should not be discarded merely because the later analysis is simpler without it.
Write operational rules before fieldwork: how pauses split conversations, how group interactions are represented, how simultaneous events are ordered, and how corrections are logged. Version the coding guide when the rules change.
Contents · Previous chapter · Back to contents · Next chapter
CHAPTER 4 OF 12 · BUILD
4. Choose a time resolution that matches the mechanism
The clock changes the network
A five-second sensor stream, a five-minute observation sheet and a daily diary do not describe the same process. Fine windows may be sparse and noisy. Wide windows can connect people who were never present at the same moment and can erase the order that made a response possible.
Choose the smallest resolution the evidence can support, then test at least one wider and one narrower aggregation. If the substantive conclusion changes, report that dependence instead of presenting one window as natural.
Contents · Previous chapter · Back to contents · Next chapter
CHAPTER 5 OF 12 · BUILD
5. Worked example: four people, four events
Consider a fictional morning log: 08:00 A asks B for information; 08:04 B contacts C; 08:10 C delivers a file to D; 08:50 A checks with D. An hour-wide static network contains A–B, B–C, C–D and A–D. It suggests a connected group but hides the relay A→B→C→D.
| Window | Edges visible together | Interpretive risk |
|---|---|---|
| Five minutes | A→B, then B→C in separate snapshots | Sparse snapshots may conceal the longer relay |
| One hour | All four edges in one graph | Order and response delays disappear |
| Event sequence | Four timestamped events | Preserves order but requires more careful modelling |
The arithmetic is simple; the inferential point is not. A path in an aggregated graph does not prove that information could travel along it in a feasible temporal order.
Contents · Previous chapter · Back to contents · Next chapter
CHAPTER 6 OF 12 · BUILD
6. Compare event rates, not only edge counts
Suppose A contacts B twelve times and C once. A binary graph gives both ties the same value. A weighted graph retains frequency but may still hide bursts, long silences and whether one actor always initiates. Event-rate models ask how the chance of the next event changes with prior events and contextual predictors.
Before fitting a model, plot event counts over time, durations, inter-event gaps and the number of eligible actor pairs. Sudden changes may reflect a schedule, a device outage or a fieldwork boundary rather than a social mechanism.
Contents · Previous chapter · Back to contents · Next chapter
CHAPTER 7 OF 12 · TRANSFER
7. Changed-context transfer: a disaster-support desk
Move the method from a campus to a disaster-support desk. Calls arrive from households, are routed to volunteers and sometimes escalate to an agency. The relevant event is not mere co-presence; it is a directed hand-off with a timestamp, issue type and resolution state.
| Question | Campus example | Support-desk version |
|---|---|---|
| Actor boundary | Enrolled participants | Households, volunteers and agencies |
| Event | Observed interaction | Call, referral or escalation |
| Time window | Minutes or hours | Minutes, shifts and response deadlines |
| Ethical risk | Re-identifying social routines | Exposing vulnerable households |
The structure transfers, but the privacy and consequence profile changes. A technically rich log may be inappropriate if it creates avoidable risk.
Contents · Previous chapter · Back to contents · Next chapter
CHAPTER 8 OF 12 · TRANSFER
8. Test conclusions across aggregation windows
A stability check
Recompute the descriptive claim at several defensible windows. If ‘central helpers’ are the same at five minutes, one hour and one day, the conclusion is more stable. If the ranking flips, investigate whether shift boundaries, repeated bursts or missing data explain the change.
Do not average away the disagreement. Window sensitivity is evidence about the timescale of the process and about how much confidence the result deserves.
Contents · Previous chapter · Back to contents · Next chapter
CHAPTER 9 OF 12 · LIMITS
9. Misconceptions and evidence limits
- A dense aggregate is not proof of simultaneous community cohesion.
- Co-presence is not automatically communication, trust or influence.
- A timestamp from a device is not necessarily the time a socially meaningful event began.
- More temporal precision does not repair selective participation or missing devices.
- A fitted event model describes patterns conditional on its observation and modelling assumptions; it does not by itself establish causation.
Ethnographic notes, participant accounts and event traces can complement one another. None should be treated as the hidden ‘truth’ that automatically overrules the others.
Contents · Previous chapter · Back to contents · Next chapter
CHAPTER 10 OF 12 · VERIFY
10. Practical audit before interpretation
Freeze the raw event log, retain a transparent cleaning script or decision trail, and record the observation boundary. Summarise missing time by actor and setting. Check clock synchronisation, duplicate events, impossible durations and changes in device or observer coverage.
| Check | Evidence to retain | Failure signal |
|---|---|---|
| Clock | Synchronisation and timezone record | Events appear out of order |
| Coverage | Actor-hours observed | Centrality tracks device uptime |
| Aggregation | Results at several windows | Main conclusion depends on one arbitrary window |
| Meaning | Field notes or coding definitions | Sensor edge is interpreted as a social act |
Contents · Previous chapter · Back to contents · Next chapter
CHAPTER 11 OF 12 · VERIFY
11. Useful FAQs
Does a relational event model require sensors?
No. Events may come from structured observation, messages, calls, transactions or coded archival records, provided their meaning and timing are defensible.
Can I convert events into a normal network?
Yes, but state the time window, direction, weighting and information lost. Keep the event-level data for checks.
What if several events share the same timestamp?
Use the measurement resolution honestly. Do not invent an order. Apply a documented rule or a method that handles ties.
Is more detail always better?
No. Collect only the precision needed for the research question and justified by consent, privacy and data quality.
Contents · Previous chapter · Back to contents · Next chapter
CHAPTER 12 OF 12 · VERIFY
12. Sources and a reproducible reading route
These sources explain event sequences, temporal aggregation and high-resolution proximity measurement. Read their methods and limitations, not only their diagrams.
- Relational event modelling and social interaction sequences
- Dynamic relational event modelling
- High-resolution measurement of social networks
- SocioPatterns temporal proximity datasets
Contents · Previous chapter · Back to contents · Return to the Anthropology root
Connected reading across eduKate
Continue through the routes that materially extend this method: