A 48-market survey is not 48 equally reliable samples

The revenue mechanism

Publishers use the Reuters Institute's annual Digital News Report to benchmark how readers find and pay for news, and every figure in it depends on a survey method the Institute discloses alongside its findings. The 2025 edition, published 17 June 2025, states it is based on data from six continents and 48 markets, including Serbia for the first time. The fieldwork, its own methodology page states, was commissioned by the Institute and carried out by YouGov as an online questionnaire fielded from mid-January to the end of February 2025, with samples assembled using nationally representative quotas for age, gender and region in every market, plus education quotas in most and political quotas based on recalled vote choice in about a third of markets.

What the documents show

The methodology page states plainly what this design does and does not represent. Because the survey is conducted online, results are described as representative of the online population in each market, not the whole national population; where internet penetration is lower the gap between the two can be large, while in Northern and Western Europe, where penetration typically exceeds 95%, that gap is described as small. India, Kenya, Nigeria and South Africa are flagged as exceptions: their samples represent younger, English-speaking internet users, not the national population, because reaching other groups representatively online is not possible there. The page also states that because the sampling approach is non-probability, it cannot generate a conventional margin of error, and that the Institute treats differences of roughly 2 percentage points as too small to be meaningful.

The assumptions exposed

The report's scale, 48 markets across six continents, is often read as one global sample with uniform reliability, but the Institute's own document distinguishes markets where online samples track the general population from markets where they do not. A quota-based online panel is not a probability sample; the page notes this design cannot generate a standard margin of error, so a two-point year-over-year shift is, by the report's own threshold, not distinguishable from noise. The report is a self-reported behavior survey, not a traffic log, so a platform-usage figure measures what respondents say, filtered through recall, not what analytics would record.

What to check before you rely on it

This is an editorial checklist. Before citing a country-level figure, check whether that country is one of the four flagged as representing only younger, English-speaking internet users; before citing a year-over-year change, check whether it clears the Institute's own 2-point threshold; and before treating the sample as a census, check the figure describes the online news audience, not everyone in that country.

• Is this a country where the report itself warns the sample is not nationally representative?
• Is the year-over-year change larger than the roughly 2 points the Institute treats as noise?
• Am I citing a self-reported behavior, such as claimed platform use, as if it were measured traffic?

The report remains a disclosed, methodologically transparent survey rather than an audited traffic log, and its own documentation is the right place to check what a figure can and cannot support.