Upper and lower bounds are kept apart, as in every topic (method).
Upper bounds, by sub-question
| Does testing beat rereading? |
Rowland 2014, Yang 2021 |
meta-analysis, classrooms |
g = 0.33 against restudy (Yang); 0.73 with feedback (Rowland) |
| Does spacing help? |
Latimier 2021 |
meta-analysis |
g = 0.74 spaced vs massed |
| Best gap |
Cepeda 2008 |
one large study, n above 1,350 |
20-40% of delay at 1 week |
| Do examples help novices? |
Barbieri 2023 |
meta-analysis, maths |
g = 0.48 |
| When does assistance stop helping? (indirect evidence for examples) |
Tetzlaff 2025 |
meta-analysis, 60 studies on instructional assistance in general |
+0.51 novices, -0.43 advanced |
| Draw or study a concept map? |
Schroeder 2018 |
meta-analysis, 142 effects |
0.72 vs 0.43 |
| Words with their picture? |
Ginns 2006 |
meta-analysis |
d = 0.72 (spatial, 37 effects); 0.85 pooled with temporal (50 effects, 2,375 novices) |
| Visual cues? |
Schneider 2018 |
meta-analysis, 103 studies |
g = 0.53 retention (N = 12,201) |
| Which multimedia rules matter most? |
Noetel 2022 |
meta-meta-analysis, 1,189 studies |
captioning second-language video, contiguity, signalling |
| Comparing cases? |
Alfieri 2013 |
meta-analysis, 57 experiments (336 tests) |
d = 0.50 per test |
| Prequestions? |
St. Hilaire 2023 |
meta-analysis |
g = 0.54 targeted, 0.04 other |
| Skeletal notes? |
Kiewra 1985 |
randomised experiments |
less writing, more recall |
Lower bounds (limits)
- Expanding gaps do not beat uniform gaps (g = 0.03).
- Testing gains shrink as element interactivity grows (van Gog 2015).
- Surface outcomes gain more than deep ones across techniques (0.60 vs 0.26).
- Seductive details cost a small to medium amount, g = -0.33, and the effect varies widely by moderator (Sundararajan 2020).
- No pooled effect exists for subgoal labels, atomic notes or maps of content.
Most adopted, kept apart from best evidence
| Scheduling reviews |
Anki (SM-2, FSRS option), Duolingo's own model |
recall-prediction benchmarks by the schedulers' own community; no trial on exam scores found |
| Edtech efficacy |
Khan Academy, Duolingo pages by their owners |
Oreopoulos 2026, an academic RCT, small gains |
| Note-taking |
Zettelkasten, second brain |
skeletal notes (Kiewra 1985); linked notes untested |
| Matching teaching to style |
belief in 89.1% of educators (Newton 2020) |
not supported (Pashler 2008) |