Self-study materials

Self-study materials are lessons written once and read by many people who study alone, with no tutor to check their work: textbooks, workbooks, web pages and self-paced online modules. Because nobody responds to the individual reader, the author must build into the material what a tutor would supply: examples, questions, hints, feedback and ways to fit readers of different prior knowledge. The research that bears on these features comes mostly from classrooms, laboratories and computer tutors rather than from static material itself, and direct tests are few. The clearest findings are that prompting readers to explain works in static form, that self-testing aids retention, and that an explanation beats a bare answer in feedback. Whether readers should attempt a problem before seeing its solution, and how much guidance suits readers with different prior knowledge, remain contested. Where a human or AI tutor responds to each learner, see Tutoring interfaces and AI tutoring.

History

History of self-instructional material, including programmed instruction, still to be researched.

Core concepts

Worked examples and expertise reversal

A worked example shows a problem with its solution steps, for the learner to study instead of solving. Expertise reversal is the finding that guidance which helps learners with little knowledge, such as worked examples, can stop helping or even hurt learners with more prior knowledge, who do better with less guidance (Wikipedia contributors, 2026d). The proposed response is adaptive fading: removing worked-out steps as the learner shows understanding (Wikipedia contributors, 2026d). Static material cannot measure understanding as it goes, so it can only offer this adaptation through reader choice or placement (see below).

A literature review reports a related pattern for erroneous examples, which are worked solutions that contain a mistake the learner must find: learners with low prior knowledge tended to benefit more from correct examples and learners with high prior knowledge from erroneous ones, with some interactions significant and some not (Wesenberg et al., 2022). Studies generally find that erroneous examples can help compared with correct ones, but the effect has moderators and results are mixed; one 2022 experiment found no effect on learning performance from the order or congruency of correct and erroneous examples (Wesenberg et al., 2022).

Self-explanation prompts

A meta-analysis of prompts that ask learners to explain material to themselves found a pooled effect of g = .55 across 69 effect sizes (Bisra et al., 2018). Print and digital media gave identical pooled effects (g = .550), and environments with no interactivity (g = .575) did as well as adaptive ones, so such prompts work in static material (Bisra et al., 2018). The effect did not differ detectably whether researchers verified that participants actually self-explained or not (g = .409 unverified, .613 verified) (Bisra et al., 2018). Timing, content specificity and most formats rarely mattered (Bisra et al., 2018).

Two kinds of prompt did worse. Prompts asking learners to conceptualise or explain the content worked better than metacognitive prompts about planning or performance, and metacognitive prompts showed no detectable effect (Bisra et al., 2018). Multiple-choice prompts (2 studies) had an effect not different from zero, possibly because strong cueing undermines self-explanation, while fill-in-the-blank prompts had g = .895 (2 studies) (Bisra et al., 2018). All reviewed prompts were scripted by researchers or instructors, mostly derived from the task content, and the authors say such prompts suit material assigned to many students, which is the static case (Bisra et al., 2018).

Self-tests and retrieval practice

Self-tests help retention even with no feedback, but feedback adds benefit and successful retrieval is critical (Shanks et al., 2023). Learners underuse self-testing and tend to underappreciate practice tests; they choose it more when it is likely to succeed, or when feedback or restudy is offered during testing (Shanks et al., 2023). The negative effect of picking a wrong multiple-choice option on a practice test depends on corrective feedback being withheld (Shanks et al., 2023).

The benefit matches the type of question. In a study summarised by its author, fact quizzes helped a fact test, higher-order quizzes helped a higher-order test, and mixed quizzes helped both (Agarwal, 2019). This suggests that review attached to static material should mix recall and application items. The Transfer of learning article covers retrieval practice and spacing for use over time, and Memory covers the underlying memory theory.

Feedback

A meta-analysis of computer-based feedback found that elaborated feedback (an explanation) had a mean effect of 0.49, against 0.05 for correctness-only feedback and 0.32 for giving the correct answer (van der Kleij et al., 2015). The advantage of elaboration was greatest for higher-order learning outcomes (van der Kleij et al., 2015). A key that gives only the answer is therefore weaker than one that explains it, though the evidence comes from computer-based settings.

Placement and learner choice

Readers of unknown prior knowledge raise the expertise-reversal problem directly, since one version of the lesson cannot suit all of them. In one digital environment (N = 102), letting learners choose their own content order did not change learning compared with an assigned order, while scaffolding improved recall and self-efficacy (Herold & Seufert, 2025).

Module structure and completion

In four edX MOOCs, shorter videos were much more engaging, with engagement measured as watch time and attempts at post-video problems; this is evidence about video and about engagement, not learning (Guo et al., 2014). Completion in online courses is often very low, under 10% in MOOCs, though low completion does not necessarily mean failure (Patterson, 2018). In a randomised MOOC test, a commitment device raised completion by 40%, while alerts and distraction blocking did nothing detectable (Patterson, 2018).

Questions, controversies, and new developments

Should readers attempt before seeing the solution?

A design choice in every static lesson is whether the answer is on the page, behind a click, or absent. Attempting an answer before being shown it, even a failed guess, improved later recall compared with studying only, according to a summary of experiments by Kornell, Hays and Bjork; the authors advise trying textbook questions before reading (Roediger & Finn, 2009). When answers sit behind a click, most learners do attempt first: in a study of 550 students in four classes, students made a genuine attempt before viewing the answer on average 84% of the time, with earnestness falling as the course progressed (Yuen et al., 2016).

Whether attempting first is better than studying an example first is less settled. In one study, students who observed the complete productive-failure process, meaning another learner’s attempts, performed better after instruction than students who attempted the problem themselves first (Hartmann et al., 2021). In grade-8 mathematics, an early advantage of problem solving over worked examples for students with high performance expectancies did not last three months, and the pattern reversed in the delayed test (Reinhard et al., 2019). A meta-analysis of 53 studies found that productive failure (problem solving before instruction) works better for secondary and undergraduate students than for primary students and needs some prior knowledge; this concerns attempts before instruction, not access to answer keys (Elhardt, 2021).

When should feedback arrive?

In an engineering course, homework feedback given one week late led to better performance on later exam problems on the same concepts than immediate feedback, although students believed immediate feedback helped most (Mullet et al., 2014). This is one study.

Do hints help without a tutor?

The evidence on hints comes from computer tutors, not static material. In Geometry Cognitive Tutor data, asking for help on hard steps went with learning, and overuse of help with poorer learning (Roll et al., 2014). For steps where prior knowledge was low, avoiding help and failing repeatedly went with better learning than seeking help, so novices may need to attempt before help makes sense (Roll et al., 2014). Feedback on help-seeking made students use on-demand help more deliberately but did not improve domain learning, and on-demand help is still judged useful in some circumstances (Aleven et al., 2016). A related finding for self-explanation prompts is that making an instructional explanation available on demand alongside them reduced the benefit in one study, because it undermines the learner’s incentive to explain (Bisra et al., 2018).

Fitting the cue to the reader

Cue strength may need to fit knowledge: novices may benefit from strongly cued prompts and more knowledgeable learners from less cued ones. This is a hypothesis (Bisra et al., 2018). The same logic underlies expertise reversal.

Broader connections

Self-study material is the fixed end of a range that runs through Tutoring interfaces to responsive AI tutoring. A tutor can adapt hints and fading to the learner; a static page can offer only the choice of when to look, which is why the help-seeking and learner-control findings matter. The question of whether what is studied is later used belongs to Transfer of learning.

Further reading

• Bisra et al. (2018): the meta-analysis of self-explanation prompts, with comparisons of format, medium and cue.
• Shanks et al. (2023): an encyclopedia review of test-enhanced learning, including feedback and learner use of self-tests.
• van der Kleij et al. (2015): a meta-analysis comparing correctness-only, answer and explanation feedback.
• Roll et al. (2014): help-seeking and help-avoidance in a computer tutor, relevant to when hints help.
• Mullet et al. (2014): a real-course test of delayed against immediate feedback.