Skip to content
Menu
Rock formations in the Reserva Nacional Salinas, Arequipa, Perú.

Methodology

The rules by which Koyuntura captures, classifies and delivers its data.

Unit of analysis

The unit of analysis is the conflict, the place and the event. Never the person.

  • Conflict

    The social conflict being discussed, taken from a catalog built from the report of the Defensoría del Pueblo (Peru's Ombudsman Office).

  • Place

    Where what is mentioned takes place. The maps by province and the detail by area come from it.

  • Event

    What kind of act is announced or reported, with its reference to radio station, program, date and minute.

Never the person

  • There is no search by the name of private individuals.
  • No voiceprints are used.
  • Listeners who call the radio are anonymized by default.
  • Only public office holders and organizations are recorded as entities.
  • No profiles of leaders or radio hosts are drawn up.

Codebook

The classification rules, by category. Each fragment is coded with them and only doubtful cases go to human review.

Relevance

The first decision: whether the fragment is about a social conflict.

Conflict

Which conflict it is about. Either a known one is chosen from the catalog, which is built from the report of the Defensoría del Pueblo, or a new one is flagged. For each conflict, the catalog records its location, its collective actors, its type and its status.

Type of act

What kind of act the statement is.

  • Call to action
  • Blockade announcement
  • Complaint
  • Rumor

Tone

The tone in which the conflict is discussed.

Place

Where it takes place. The names of communities and projects are recognized with a dictionary of their own, using approximate matching.

Status and escalation

What status the conflict is in, and whether what is said points to its reactivation or its escalation.

Signal sources

Regional radio is the main input source. These are the rules by which it is captured.

  • Internet streaming

    The main source is the direct internet stream of the station.

  • Receiver in the area

    Where a station does not stream over the internet, capturing its signal requires a physical receiver in the area.

  • Station and channels

    A single station can air on more than one channel. The catalog separates the station from its channels, and its identity does not depend on the direction of the transmission.

  • Relays

    A network with repeater stations is captured once and attributed to each locality. Separating them from the original stations prevents inflating the trends.

  • Coverage denominator

    Coverage is measured against the stations authorized by the Ministry of Transport and Communications, cross-checked by frequency and district.

Known limitations

What the data does not cover, or covers with errors.

  • Coverage

    Many of the stations in the mining areas do not stream over the internet. Capturing them requires a physical receiver in the area, and where there is no capture, there is no data.

  • Source bias

    The stations that can be captured may not represent all sides of a conflict equally: what is reached over the internet may lean toward one side.

  • Quechua and Aymara

    Part of the rural programming in the south is in Quechua or Aymara, which are not transcribed. The language of each segment is detected, and only those containing keywords pass to a Quechua-speaking reviewer; the rest in those languages is not covered.

  • Transcription errors

    Automatic transcription of regional radio has errors, and what fails most are proper names, such as those of communities.

  • Automatic classification

    Classification is automatic and can be wrong; its accuracy is measured on fragments labeled by hand. Next to each evaluated rule, the verbatim quote is stored, so the decision can be audited, and only doubtful cases go to human review.

  • Schedules and outages

    Many stations switch off at night, and transmissions drop out. A shutdown outside broadcast hours is not an incident; an outage leaves a gap in the data, and the gap is recorded.

  • History

    The history of each station starts on the day it is added to capture. Almost no local station keeps its programming, so what came before cannot be recovered.

Data, not audio

What is delivered is mentions, counts, trends, indexes and our own summaries. No audio clips are delivered. Traceability is the reference to radio station, program, date and minute, not the audio.