Sheet 13 of 17 · Methods
Methods
Most evidence is self-reported and cross-sectional. The field needs to move from measuring adoption to measuring its educational consequences.
Table 13.1: Evidence base
| Source | Population | Strength | Limitation |
|---|---|---|---|
| HEPI / Kortext 2024–26 | Full-time UK undergraduates, ~1,050/yr | Repeated annual series | Excludes part-time and graduate students |
| Tyton Partners 2023–26 | US students, faculty, admins; 750+ institutions | Multi-stakeholder, repeated | Not a panel; thresholds vary by wave |
| Lumina–Gallup 2026 | 3,800 US associate and bachelor's students | Large US sample | Cross-sectional |
| Inside Higher Ed 2025–26 | 1,000–5,000 US students | Direct student voice | Flash surveys |
| College Board 2025 | US high-school students, ~600 per wave | Pipeline and subgroups | Unweighted samples |
| Pew Research 2023–26 | 1,458 US teens (probability panel) | Representative pipeline data | Teens, not college students |
| Middlebury 2025 | ~20% of one liberal-arts college | Detailed task and subgroup data | Single selective institution |
| Dartmouth 2026 | 1,003 US national + 1,533 Dartmouth | Incoming-cohort data | Institutional sample not representative |
| Experiments & meta-analyses | Specific courses and interventions | Causal mechanisms | Short duration, narrow contexts |
Table 13.2: How evidence is classified
Evidence strength depends on design, not on the publisher's prestige. Each reference is tagged with its design and with what kind of finding it supports: causal, association, measured, self-reported or synthesis.
| Design | What it can show | Sources here |
|---|
One row per statistic: source, sample, design, limitations and verification status. Open questions are on Open questions.
How evidence is classified
- Randomized controlled trial: Cause and effect, in the setting and time frame studied.
- Experiment (design not yet checked): Possible cause and effect; design details still need checking.
- Quasi-experimental study: Cause and effect without random assignment. None in this evidence base yet.
- Longitudinal study: How the same people change over time. None in this evidence base yet.
- Systematic review and meta-analysis: Pooled effects across studies. Only as good as the studies included.
- Nationally representative survey: How common a behaviour or view is in a defined population, at one point.
- Repeated cross-sectional survey: Population trends across years, not how individuals change.
- Cross-sectional survey: A snapshot, often from panel or convenience samples.
- Observational study + experiment: Associations in one setting, plus a small embedded experiment.
- Qualitative research: Why and how, in students’ own words. None in this evidence base yet.
- Technical evaluation: How a tool performs on test material.
- Industry or institutional report: Useful signals; methods are often less transparent.
- Source document: The AI-assisted syntheses this site was compiled from. Not evidence on their own.
- News: Context, not evidence.