Review: reading order
This page lists, with what it found, each source on the order and depth in which a document is read. Its six strands run from overview before detail to practice.
Levels: MA meta-analysis · RCT randomised experiments · S single study · R review without pooling · T theory or model with no controlled test.
(secondary) marks a number read in a summary, not in the paper.
Surveys to read first:
- Ponce 2025 (organizers);
- McCrudden 2007 (relevance);
- Rittle-Johnson 2015 (order of lessons);
- Loibl 2017 (problem first);
- Wickens 2013 (part-task training);
- Rayner 2016 (skimming);
- Brunmair 2019 (interleaving).
1. Overview before detail
| Source | Finding | Level |
|---|---|---|
| Luiten 1980; Stone 1983 | advance organizer: learning d = 0.21 over 135 studies, retention 0.19 to 0.38 by interval, rising with delay (Luiten, read in full); median 0.48, mean 0.66 over 112 effect sizes from 29 studies, on tests a week or more later (Stone, read in full); Mayer's vote count found stronger effects for learners lacking prerequisite knowledge, and Stone found no advantage for low-knowledge learners (k of 4 to 6) | MA |
| Luiten 1980 (moderators) | written organizers 0.17 (k = 89), aural 0.37 (k = 21); high ability 0.23, middle 0.08, low 0.13; no study measured retention beyond 15 weeks (read in full) | MA |
| Alfieri 2013 (principle first) | giving the principle before the cases did no better than giving none (read in full) | MA |
| Ponce 2025 | instructor-provided graphic organizer: memory g = 0.70, comprehension 0.53, no moderator found | MA |
| Ponce 2023 | outline given with the text: memory g+ 0.61 (k = 18, CI 0.33 to 0.88); comprehension g+ 0.34 (k = 10, CI -0.14 to 0.81, I2 72%), not significant. Learner-generated outlining: memory 0.59 (k = 9, prediction interval -0.62 to 1.80), comprehension 0.59 (k = 10, prediction interval 0.02 to 1.16), writing 0.52 (k = 11). Provided outlines: 20 of 22 effect sizes are college (read in full) | MA |
| Mayer 1979 | review of organizer studies; predicts organizers help most on unfamiliar material | R |
| Bransford and Johnson 1972 | topic given before a passage raises comprehension and recall; given after, no gain | RCT |
| DeStefano and LeFevre 2007; Amadieu 2009 | a map matching the domain hierarchy helps low prior knowledge; a network map did worse than a hierarchical one | R; S |
| Reigeluth and Stein 1983; Van Patten 1986 | epitome and zoom-lens sequencing; the few tests are null | T |
| Harden and Stamper 1999 | spiral curriculum; single-site, quasi-experimental evidence | T |
| Duggan and Payne 2009, 2011 | time for half the text: skimming beat reading half on important ideas (Exp 1), not reading half of each paragraph (Exp 2), and only with easy navigation; proposed account: readers leave a section when its gain rate drops (2009 experiments; 2011 reanalysis) | S |
| Rayner 2016 | speed costs comprehension; skimming gives moderate comprehension | R |
| McDaniel 2009 | SQ3R: only the read-recite-review core tested; beats rereading, equals note-taking in less time | RCT |
| Wickens 2013; Swanson and Law 1993 | parts trained apart transfer badly when they run together; adaptive ramps work, fixed steps fail; experts gain less (37 comparisons) | MA; T |
2. Depth control
| Source | Finding | Level |
|---|---|---|
| McCrudden 2005; McCrudden and Schraw 2007 | relevance instructions: relevant parts learned better at no extra reading time; the rest learned less | RCT; R |
| Klauer 1984; St. Hilaire 2023 | objectives and prequestions: targeted content up; the rest down with objectives (Klauer), unchanged with prequestions (g = 0.54 vs 0.04, St. Hilaire) | MA |
| Dee-Lucas and Larkin 1995, 1999 | hypertext in small units, read with a goal: deep on the goal, narrow overall; overviews widen recall | S |
| Ariel 2009 | when the experimenter's reward favoured some word pairs, learners chose which half of 30 to restudy by reward more than by difficulty; weaker for sequential choice and low working-memory span | RCT (4 experiments) |
| Metcalfe and Kornell 2005 | learners skip items they judge known, take the nearly known first, stop when judged learning rate nears zero; describes behaviour, efficacy untested (8 experiments) | RCT |
| Nelson and Leonesio 1988 | extra time on the hardest items gives little gain; correlational (secondary) | S |
| Pirolli and Card 1999 | leave a source when its gain rate falls to the average elsewhere | T |
| Pyc and Rawson 2009; Rawson and Dunlosky 2011 | correct recalls required before moving on: diminishing returns past a few; word pairs | RCT |
| Dunlosky and Rawson 2012 | overconfident learners stop early and score lower; checking against the key's idea units corrected it (secondary) | RCT |
| Rozenblit and Keil 2002 | felt understanding of a mechanism exceeds the explanation; asking for it exposes the gap (12 studies) | RCT |
| Kalyuga and Sweller 2004 | a rapid first-step test correlates up to r = .92 with full knowledge tests and is 2.5 to 4.9 times faster (abstract only; the first-step method stays secondary) | validation studies (Experiments 1 and 2); RCT (Experiments 3 and 4) |
| Tetzlaff 2025 | expertise reversal: assistance helps novices (+0.51), hurts the advanced (-0.43); the reversal is asymmetric, with a larger gain for novices | MA |
| Kester 2001, 2004 | just-in-time model: theory before practice, how-to during it; the experiment (n = 72) found the reverse most efficient | T; S |
| Doignon and Falmagne 1985 | study what all prerequisites are mastered for (outer fringe) | T |
3. Order within a unit
| Source | Finding | Level |
|---|---|---|
| Rittle-Johnson and Koedinger 2009; Rittle-Johnson 2015 | alternating concept and procedure lessons beat concepts first on procedures and transfer; concepts equal (N = 77, 26) | RCT; R |
| van Gog 2011 | example then paired problem beat problems only; problem then example did not | RCT |
| Ashman 2020 | when many elements interact, instruction then problem beat the reverse (2 experiments) | RCT |
| Sinha and Kapur 2021; Loibl 2017 | problem first g = 0.36 on concepts and transfer, -0.03 on procedures; larger with instruction built on the attempt (0.56 vs 0.20); Loibl's review: only with contrasting cases or built-on instruction | MA; R |
| Fyfe 2014; McNeil and Fyfe 2012 | concrete then abstract: no pooled effect; one positive transfer experiment | R; S |
4. Order across units
| Source | Finding | Level |
|---|---|---|
| Gagné 1968; White 1973 | prerequisite order supported; transfer between levels hard to show | R |
| Costa 2022 | whole 4C/ID programmes, built on simple-to-complex task classes, mostly against traditional teaching: performance d = 0.79, 12 quasi- or non-experimental studies; sequencing not isolated | MA |
| Yan 2017 | blocked then interleaved equals interleaved, beats blocked | RCT |
| Carvalho and Goldstone 2014 | interleave categories similar within and between, block categories dissimilar within and between (3 experiments, mostly sequential presentation) | RCT |
| Brunmair 2019; Rohrer 2020 | interleaving: g = 0.42 overall, reversed for words (g = -0.39), n.s. for expository texts; one classroom cluster RCT positive (d = 0.83) | MA; RCT |
| Schwartz 2009 | at least one topic in depth for a month or more went with higher college science grades; covering all topics: no gain in chemistry and physics, a disadvantage in biology; n = 8,310, 55 colleges, correlational | S |
5. Breadth-first and depth-first, by name
| Source | Finding | Level |
|---|---|---|
| Salmerón 2005, 2006 | next section = the most related to the last helps low-knowledge readers; high-knowledge readers gain from low coherence | RCT |
| Jáñez and Rosales 2016 | novices meeting the whole first: a goal takes effect only after an exploration phase | S |
| Lawless and Kulikowich 1996; Chen 2006 | navigation set by domain knowledge; novices observed breadth-first, experts depth-first (secondary) | S; R |
| Weber 2015 | proof reading: skim for structure, try the proof, split into parts, test steps on examples (from interviews, secondary) | S |
| Leron 1983; Fuller 2014 | structured proofs, top level first: better key-idea summaries, slightly worse justification, mostly n.s. (secondary) | S |
| Hodds 2014 | self-explaining each line of a proof: comprehension d = 0.95; a 15-minute version worked in a lecture (3 experiments) | RCT |
Forward citations
- Reigeluth and Stein 1983: Van Patten 1986 reviews its tests.
- Rittle-Johnson and Koedinger 2009: Rittle-Johnson 2015 reviews the line of work.
- Duggan and Payne 2009: Duggan and Payne 2011 adds the stopping-rule account (satisficing).
- Sinha and Kapur 2021 and Loibl 2017 review problem-first designs from two sides.
6. Practice
Kinds: standard · company research · thesis · review. Practice is graded convention, untested unless a study measured its effect.
| Source | What it prescribes or found | Kind | Adoption |
|---|---|---|---|
| ISO 2023 | plain language organised by four principles: relevant, findable, understandable, usable; usable is confirmed by evaluating with readers (principles and clauses 5.1 to 5.2.3 read in the 11-page preview) | standard | clauses 5.2.4 to 5.4 and the annexes not read, so evaluation with readers is known from a title only |
| Nielsen 2006 | eyetracking of 232 users; a rough F shape; "The first two paragraphs must state the most important information." Its owner sells consulting | company research | descriptive |
| Morkes 1997 | three studies, 81 users; on a fictional site, concise text scored 58% better, scannable 47%, objective 27%, combined 124%, all against a promotional control; its owner's own claim | company research | one site, web only |
| Nildam 1998 | 320 readers, mock newspaper page: narrative stories gave better understanding and recall than inverted pyramid, and more interest; credibility did not differ | thesis | Thai readers, crime and science stories, not peer reviewed |
| Sternadori 2008 | 58 readers, four stories each read in both orders (within subjects): no difference in recognition, cued recall or enjoyment; comprehension marginally favoured the pyramid (.66 against .64, p = .08); chronological stories slowed reactions on a secondary task (413 against 396 ms, p = .05) (full text, pp. 48 to 51) | thesis | not peer reviewed |
| Jansen 2010 | two experiments on Dutch bad-news letters (N = 450) and emails (N = 574), convenience samples, 83% graduates in the first, hypothetical scenarios: explanation first was judged more persuasive and more polite, with a better sender image; readability rating was slightly higher in the first experiment only (p = .044, not significant in the second); the comprehension test showed no difference, with a ceiling effect | experiment | bad news only; ratings, not behaviour |
| Hartley 2004 | structured abstracts hold more information and are longer (+21% over all 11 studies, about 35% in three independent comparisons); a recall gain appeared only when they were longer and more readable, and recall evidence is thin | review | many student samples |
The direct tests of the inverted pyramid do not support importance-first order. Neither thesis is peer reviewed, so they do not refute it either.