Methodology

1What is measured

Every broadcast is coded in full, chronologically, in five layers.

• Voice: who speaks, and for how many seconds — a speaker, the newsroom or the rest of the programme.
• Indirect presence: the time in which others speak about someone. It excludes segments that are already the voice of the same unit, so for an electoral list voice and indirect presence add up to presence.
• Items: for every item, its topic, occasion, place in the headlines and an overall assessment (balanced, one-sided, constructed), with the reasoning.
• Editorial practice: spin and editorial framing, with origin (editorial, actor-originated, combined), type and verbatim quotation.
• Tone: how the newsroom presents those it speaks about, and how others present an electoral list in its indirect presence. Tone is the model’s assessment and carries Level B.

The denominator is always the full duration of the broadcast.

2Sample

The main news programmes of five television channels with national coverage: RTS 1, Pink, Prva, Happy and B92. From 12 to 25 September, RTS 1 was measured every day, together with two channels in a rotation published in advance. Since then, reports are published without a fixed schedule, most often for RTS 1. Cable channels are outside the frame.

3How a side is assigned

Political side is read from the public register of actors as it stood on the date of broadcast — neither the model nor the coder determines it. Where a speaker is not named, the coder watches the recording and classifies the speaker by a visible marker; where there is no such marker, the speaker stays outside the balance of sides.

4Reliability levels

Level A

measured: the figure can be traced back to the speech segment.

Level B

reading of the measurement: a grouping or comparison that the measurement does not perform by itself.

Level C

interpretation: the compiler’s judgement on what was measured.

5What is not claimed

• Durations are computed from the line ranges of the transcript. Political side is read from the public register of actors as it stood on the date of broadcast — neither the model nor the coder determines it. Where the register cannot answer, because the speaker is not named, the coder decides by watching the recording. No inter-coder agreement coefficient has been computed; an independent check of a sample is planned and has not been carried out.
• A single broadcast describes neither that channel, nor that week, nor its editorial policy — only that broadcast. A single day carries no trend.
• No claim is made about a channel’s editorial policy in general, or about a newsroom’s intent.
• Voice and indirect presence are two separate measures. Indirect presence excludes segments that are already the voice of the same unit — an actor speaking about themselves, a member of a party or list speaking about their own party or list, an actor of the governing side or the opposition speaking about an actor of the same side — so for an electoral list the two add up to presence.
• The origin of spin is not summed without saying so; the measure of newsroom bias is editorially originated spin.
• The truth of what is said is not assessed. Journalistic procedure is recorded, not the accuracy of a claim.
• The measured programmes are the main news programmes of television channels with national coverage. Cable channels are outside the frame, and no finding extends to them.
• Seconds are not audience. Broadcast time is measured, not people reached.
• The supplementary layer of the sample is shown separately and never enters the percentages computed on the core.

6Data on request

Journalists and researchers can obtain aggregate findings, per-broadcast data and interpretations on request: jovan@aitematika.org.

7Documents

The Methodological note v1.0 (in English) and the Campaign sampling plan v1.0 (in Serbian) were deposited on Zenodo before the first data.

Cite asJovanović, J. (2026). Measuring Political Airtime in Television News: Protocol, the denominator problem, and reliability design. Version 1.0. Zenodo. https://doi.org/10.5281/zenodo.22691447