AIR Programme Insights

The public release contains a complete 2019–2024 panel of 60 country-year-indicator observations. Within each same-report-year comparison, the United States has the highest estimate for both daily internet time and daily social-media time. This is a bounded descriptive finding, not a general population ranking.

The panel establishes that intensive digital use is a measurable feature of the recent Five Eyes context and that estimates differ between countries and years. It supplies contextual evidence for AIR questions; it does not measure an individual pathway or an algorithmic effect.

Differences could reflect digital infrastructure, work and leisure practices, platform habits, pandemic-period behaviour and other country-specific conditions. The figure does not adjudicate between these possibilities.

Some apparent differences may arise from sampling variability, survey and reporting choices, the defined internet-user population, the five derived 2023 social-media values or the reporting-method change before 2024.

It does not show that more time online caused radicalisation, violence, loneliness or any case outcome. It does not reveal content, recommendation exposure, platform, motive or the experience of any offender or victim.

The figure supplies one context layer for the proposed AIR Framework. Its use is governed by the ecological and non-causal rules in the AIR methodology; it is not empirical validation of AIR.

Hollewell and Longpré (2022) review the relationship between self-radicalisation and the internet, while Bonsaksen et al. (2023) illustrate that social-media associations can depend on motives and study design. Neither source validates national screen-time estimates as a causal proxy for AIR.

Values are estimates for internet users aged 16–64, not all residents. The five 2023 social-media values are derived, the 2024 values begin a documented reporting-method regime, and this visual does not display sampling uncertainty. No line is drawn across that boundary.

Can the country ordering and period patterns be replicated with independent harmonised sources? How sensitive are results to age, population weighting and uncertainty? Which research designs can measure content and recommendation exposure without treating national averages as individual causes?

The 76 public indicators are not one interchangeable comparison pool. The current release deliberately restricts 43 of 76 indicators (56.6%) to within-country trend or snapshot/context use. Of all 76, 33 permit cross-country comparison with caveats, 32 are limited to within-country trends and 11 are restricted to snapshot or contextual display.

A rich dashboard can create false confidence if every available variable is treated as comparable. The permission classes make the limits of each source regime part of the analytical result rather than hiding them behind chart controls.

The distribution reflects differences in national law, survey instruments, population bases, reporting cadence, source coverage and the availability of harmonised series. Some themes therefore support broader comparison than others.

The pattern also reflects the programme’s current indicator selection and conservative governance rules. A different registry, better harmonised sources or revised metadata could change the class counts without any substantive social change.

It does not rank themes, countries or sources by quality. It does not show that an indicator causes AIR, violence or any case outcome, and it does not turn qualified cross-country permission into full construct equivalence.

The figure operationalises the comparison boundaries in the AIR methodology and Research Laboratory. It is a map of what the current release permits, not a social-science effect estimate.

The OECD and European Commission handbook on composite indicators emphasises transparent conceptual, normalisation, weighting and robustness decisions. AIR therefore does not collapse these classes into a country score.

Counts refer to public indicator definitions rather than observations. The denominator excludes two methodology-only indicators and one non-plottable umbrella. The classes are project governance decisions for Social Context v1.5, every included indicator requires a visible caveat, and pairwise compatibility still depends on unit, population, country coverage, period, breaks and source metadata.

Which restricted series can be harmonised without erasing substantive differences? How should uncertainty and source breaks be represented? Which additional metadata would permit stronger pairwise screening while preserving transparent exclusions?