State of the art, per sub-question
Two records are kept apart (method): best evidence (a research record) and most adopted (a practice record). They have different holders.
Best evidence
| Sub-question | Record holder | Measured how |
|---|---|---|
| Q1 Redundancy of on-screen text and speech | Adesope 2012 | meta-analysis, 57 studies, retention and transfer |
| Q1 Modality, segmenting | Ginns 2005; Rey 2019 | meta-analyses, system-paced lab lessons |
| Q1 Coherence, signalling | Sundararajan 2020; Schneider 2018 | meta-analyses (see learning science) |
| Q1 Does design matter more when pacing is fixed | Noetel 2022 | overview of 29 reviews, 1,189 studies; moderator "system-paced" |
| Q2 Does showing slides help | Baker 2018 | meta-analysis, 48 studies, no average difference |
| Q2 Which slide design | Garner 2013 | randomised experiment, n = 110, essay and delayed test, d about 0.8 to 0.9 |
| Q2 How common are slide faults | Kosslyn 2012; Le Blanc 2025 | three studies of slideshows; an audit of one medical school's lectures |
| Q3 Gesture | Dargue 2019; Hostetter 2011 | meta-analyses, 83 and 63 samples |
| Q3 Instructor visible in video | Henderson 2021; Polat 2022 | systematic reviews, 12 and 41 studies, no consistent learning effect |
| Q6 Does fixed pacing (20 by 20) raise learning | Bakcek 2020 | randomised trial, n = 134, test scores equal |
| Q4 Contrast floor | WCAG 2.2 (W3C 2023b) | standard (a convention), not an experiment |
| Q5 How talks are built | Rowley-Jolivet 2005 | move analysis of 44 talks in three fields |
| Q7 Build steps that follow the voice | Liu 2022; Ito 2026 | randomised, n = 99; n = 40 with eye tracking, d = 0.49; both recorded narration |
| Q7 Who sets the pace | Tabbers 2009 | learner-paced beat system-paced on transfer, at a large cost in time |
| Q7 Does video help | Noetel 2021 | 105 randomised trials, g = 0.28 as a substitute and 0.80 as an addition |
| Q7 Live coding against static code | Shah 2023; Shah 2025 | static code: two remote sections, no learning difference (Shah 2023); active against traditional live coding: about 530 students, no learning difference (Shah 2025) |
| Q7 Animated explainer against slides | Pichler 2023; Li 2023 | groups of 29 to 42; n = 290; engagement differs, comprehension mostly does not |
| Q8 Colour as a signal | Ozcelik 2009 | one eye-tracking experiment, n = 52 |
| Q8 Pleasant colours and faces | Brom 2018; Wong 2020 | meta-analyses, 33 samples and 28 articles, d+ 0.33 to 0.39 (Brom 2018), g+ 0.27 to 0.35 (Wong 2020) on retention and transfer |
| Q8 Colour maps for data | Crameri 2020; Liu 2018 | an argued guide with a distortion measure; a user study of colormaps |
| Q8 Light or dark when projected | Greco 2008 | three experiments, dark text on light best |
| Q8 Harmony | Schloss 2011 | preference experiments only |
Most adopted
| Sub-question | Practice holder | Evidence of use |
|---|---|---|
| Slide template | PowerPoint, with Designer | Designer: over 1 billion slides, owner's claim (Microsoft 2019) |
| Typeset talks in computer science and mathematics | Beamer | a CTAN package with a long user guide; a usage count was not found |
| Short talks | TED and TEDx guidance; PechaKucha; Ignite | formats run in many cities; counts are the owners' claims (TED 2026, PechaKucha 2026 and Ignite 2026 score 9.5, 6 and 5.5 of 20: examples only) |
| Accessible talks | Rector 2019 and the conference guides derived from it | three conference pages seen citing it; no evidence cited by the guide |
| Replacing slides | Amazon narrative memo | one company's own report (Bezos 2018) |
| Headlines as sentences | Assertion-evidence workshops; consulting action titles | no count of users found |
| Build steps and Morph | PowerPoint's animation pane and Morph | a vendor page with no learning claim (Microsoft 2026e, 6 of 20: example only); no count of users |
| Live coding in teaching | Software Carpentry style workshops | a practice guide that says it is taught in instructor training (Nederbragt 2016, 11.5) |
| Animated maths videos | 3Blue1Brown and its Manim library | over 3.3 million views of one video (Radmehr 2024); the owner's aim (Sanderson 2026, 8: example only) |
| Colour defaults | the viridis family as a library default; the Okabe-Ito eight colours | a software manual (Matplotlib 2026, 14.5); the palette is copied from a column (Wong 2011) |
| Colour accessibility | WCAG 1.4.1 and the guides built on it | a standard (W3C 2023c, 17 of 20) |
Upper bounds (achieved)
Key points shown in step with the narration beat the whole slide in one randomised study (Liu 2022).
Pleasant colours and faces gave small gains in retention and transfer across 33 samples (Brom 2018).
A perceptually uniform map keeps data variation honest where a jet-like map distorts over 7% of it (Crameri 2020).
A bundle of principles in a slide set raised delayed recall and essay quality in one lab setting (Garner 2013).
Gesture helps comprehension across many samples (Dargue 2019).
Verbatim text on screen with speech does not beat text alone (Adesope 2012).
Lower bounds (not shown)
- No independent randomised replication of the assertion-evidence effect was found in this search. Closest: a non-randomised veterinary course with better retention at one month (Root Kustritz 2014), a bundled geology course (Alley 2007), and small studies (Alzayed 2023, Sripradith 2023) or pilots (Wolfe 2023).
- No study was found that varies one slide factor (headline form, text amount, graphic) in isolation within a live talk.
- No study was found of projected text size and comprehension in a room.
- No controlled comparison of a memo-first meeting against a slide meeting was found.
- The 20 by 20 and 15-second pacing rules have three small outcome studies (Liao 2020, Warmuth 2021, Bakcek 2020); the one randomised trial found no test-score gain. None tests Ignite's 15 seconds.
- No study was found of accessibility advice (say every slide word aloud) against the redundancy result for blind or low-vision listeners.
- No study was found of slide transitions, of an embedded clip in a live conference talk, of a live demo, or of build steps in front of a live audience.
- No live-coding study found a learning gain over static code (Shah 2023); active live coding did not beat traditional live coding (Shah 2025).
- Colour alone, without faces or shape, is 5 samples of 33 in the emotional-design meta-analysis (Brom 2018, Table 3); no isolated effect is established.
- No study of colour harmony and learning, of hue meaning on slides, or of projected colour at room distance was found.