Methods and limits
Methodology
The Observatory uses governed master workbooks, public-source case coding and documented social-context sources. Missing, unavailable and non-comparable values are not treated as zero.
Observatory v1.1.1 · Case Master v1.0.0 QC reconciled · Social Context Master v1.0.0 QC reconciled
Read this first
The site supports descriptive comparison, hypothesis generation and teaching. A pattern may be important without being causal. Case discovery, public reporting, coding judgement, national definitions, denominator choice and time trends can all affect an apparent relationship.
Case inclusion, exclusion and analytical population
The governed case workbook contains 347 internal records. One is excluded under the workbook’s calculation rule, leaving 346 public calculation-included records. The public export omits internal governance commentary and the redundant air_causation compatibility field. That field duplicated the governed AIR-level label in every included record; removing it from the public release does not change a case value or classification.
AIR and CIVM classifications
AIR level summarises the strength of documented pathway evidence in public sources. The separate pathway evidence score structures evidence; neither measure is a danger score, severity scale or prediction. CIVM identifies coercive ideological violence and mobilisation that may fall outside formal terrorism definitions.
Social-context data
The complete public CSV contains 9,054 governed observations. The interactive selector intentionally uses a smaller numeric runtime: it includes only populated observations whose indicator is approved for selector display. The difference is therefore a governed display boundary, not missing data. Methodological and future source-strengthening entries remain outside public selectors until defensible observations exist.
Twenty-five World Economic Forum Executive Opinion Survey observations are ranked categorical responses rather than a numeric time series. They remain in the public CSV with their observation identifier, rank subkey and category label, but are excluded from the numeric selector until a suitable categorical display is implemented. Indicators otherwise retain their source unit, year, age band, source and comparability note.
Age, sex and intersectional comparison
Source age bands are preserved. The Observatory does not silently substitute all-ages values for age-specific measures or infer unavailable intersections. Age and sex views appear only where governed observations support them.
Rates, averages and standardisation
Raw counts and population-adjusted rates answer different questions. Standardised values use the selected Five Eyes observations and place the selected-group mean at zero. They are shown only where units and observations support a meaningful calculation; public tables do not expose internal concern-adjusted scores.
Milestones, event windows and causation
The 2014, 2019, 2020, 2022–2023 and October 2023 markers are pre-specified teaching reference points. Before-and-after differences may reveal temporal association, but they do not isolate a single cause. Platform change, reporting practices, case mix and secular time trends remain alternative explanations.
Evidence quality
Indicator ratings consider source quality, country coverage, temporal coverage, comparability and interpretive limits. Stars are a communication aid, not a substitute for the underlying source and caveat.
Test Yourself: complete scoring method
The educational self-assessment is not formally validated and does not collect participant data. It uses adaptive questions across eight dimensions. Responses are processed only in the user’s browser.
Response values
Frequency items are scored Never or rarely = 0, Sometimes = 1, Often = 2 and Very often = 3. Extent items are scored Not at all = 0, A little = 1, Quite a lot = 2 and Very much = 3. Protective items use the same response order in reverse, so stronger protective practice receives a lower concern-pattern value. Device-reported time is benchmarked descriptively but does not itself determine the AIR-related result.
Questions, values and branching
| Dimension | Item | Scoring or branch |
|---|---|---|
| Time, routine and displacement | Checking within five minutes of waking or before sleep; checking during queues, toilet use, meals or time alone in public; displacement of sleep, study, work, exercise or time with people. | Frequency, frequency and extent scales. |
| Information diet and feed diversity | Use of genuinely different news sources outside social media—including outside YouTube, X, TikTok and similar platforms—such as independent news apps, broadcaster websites or programmes, printed newspapers, full articles, official documents, books or podcasts. | Protective reverse-coded scales. |
| Influencers and algorithmic repetition | Up to three principal platforms; repeated influencer exposure; acceptance of the influencer’s framing; recommendation-led or autoplay use. | Platform selection is unscored. Influencer acceptance appears after Often or Very often repeated exposure. Other items use frequency or extent scales. |
| Verification and misinformation resilience | Checking original and independent sources before sharing; confidence recognising manipulated, decontextualised or AI-generated material. | Protective reverse-coded scales. |
| Emotional and grievance amplification | Anger, fear, humiliation, disgust or group hostility after feed use; whether one issue dominates interpretation of unrelated events. | Frequency and extent scales. A free-text issue prompt appears after Quite a lot or Very much and is unscored. |
| Closed-network and identity reinforcement | Time in private, encrypted or highly like-minded communities; pressure to prove loyalty, denounce others or adopt stronger positions. | Frequency and extent scales. |
| Relationship rupture and ideological sorting | Avoiding or severing a friendship or family relationship because of views; relationship length; replacement by online contacts or newer like-minded friends. | No = 0, considered = 1, temporary rupture = 2, ended = 3. The two follow-ups appear after a temporary or ended relationship. Length scores are 1, 2, 3 and 3 across the four displayed bands; replacement uses the extent scale. |
| Offline connection, activity and agency | Live contact with friends or family; sport, exercise, worship, music, hobbies, reading, meditation or community activity. | Protective reverse-coded scales. |
Dimension weights
| Dimension | Weight |
|---|---|
| Influencers and algorithmic repetition | 1.5 |
| Emotional and grievance amplification | 1.5 |
| Closed-network and identity reinforcement | 1.5 |
| Relationship rupture and ideological sorting | 1.5 |
| Time, routine and displacement | 1.0 |
| Information diet and feed diversity | 1.0 |
| Verification and misinformation resilience | 1.0 |
| Offline connection, activity and agency | 1.0 |
Thresholds
Each dimension is normalised to its available response range: Lower below 34%; Moderate from 34% to below 67%; Higher at 67% or above. The overall result is the weighted mean using the weights above and the same thresholds. Unasked adaptive branches are excluded rather than scored as zero.
Benchmarks
The tool first seeks a genuinely compatible country, age-band and sex observation. It does not silently substitute a broader or different age band. If no compatible subgroup exists, it uses the selected country’s latest suitable all-population observation and identifies that fallback. A Five Eyes average is used only when no suitable national observation exists. Mismatched years, source age bands or device scopes are identified; no comparison is shown when it would mislead.
Versioning and updates
Generated citations include the active release and access date. Record-level update dates reflect the latest published release, not an unpublished administrative draft.