Characteristics of Effective Feedback: What the Research Says

Ask anyone who builds courses whether feedback matters and they’ll say yes. Ask what makes feedback effective and the answers get vague: be specific, be timely, be kind. Those lists aren’t wrong, but they were written for a teacher handing back marked essays. In a self-paced online course there often is no teacher in the loop. Most of the feedback a learner gets was written months earlier, by you, and attached to a quiz answer or an assignment rubric.

That changes the question. It’s no longer “how do I give good feedback?” but “how do I design it in advance, for mistakes I haven’t seen yet?” This entry covers what four decades of research say about the characteristics of effective feedback, including the uncomfortable finding that a lot of feedback makes performance worse, and how to build the kind that helps into a course.

What the research says

Feedback is powerful, and unusually variable

The most cited framework comes from John Hattie and Helen Timperley, who define feedback as “information provided by an agent (e.g., teacher, peer, book, parent, self, experience)” about someone’s performance or understanding (Hattie & Timperley, 2007). Note the list: a book, or in our case a course, counts as an agent. Pooling earlier meta-analyses, they reported an average effect of 0.79, about twice the 0.40 they use as a benchmark for the typical effect of schooling.

A more recent and more careful re-analysis brought that figure down. Benedikt Wisniewski, Klaus Zierer and Hattie pooled 435 studies, with 994 effect sizes and more than 61,000 learners, and found a medium average effect of feedback on learning, d = 0.48 (Wisniewski et al., 2020). The more important finding was the spread around that average. The authors say the heterogeneity shows “feedback cannot be understood as a single consistent form of treatment”. Some feedback helps a great deal; some barely helps at all.

Not all feedback helps

The starkest evidence comes from organisational psychology. Avraham Kluger and Angelo DeNisi analysed 607 effect sizes covering 23,663 observations of feedback interventions. Feedback improved performance on average (d = .41), “but that over 1/3 of the FIs decreased performance” (Kluger & DeNisi, 1996).

Their explanation is about attention. Feedback works when it keeps attention on the task. The further it pulls attention up towards the self (how am I doing as a person, how do I compare), the weaker its effect becomes. That is the research behind a piece of advice every teacher has heard: comment on the work, not the person.

Three questions, four levels

Hattie and Timperley’s model explains what the helpful kind looks like. Good feedback closes the gap between where a learner is and where they’re meant to be, and to do that it has to answer three questions.

They then sort feedback into four levels: about the task (is this answer right?), about the process (what strategy produced it?), about self-regulation (how is the learner monitoring their own work?) and about the self (usually praise: “well done”, “you’re a natural”). Process and self-regulation feedback are the most powerful for deeper processing and mastery. Task feedback helps most when learners can use it to improve their strategy. Feedback about the self is “rarely effective”, they conclude, because praise rarely answers any of the three questions (Hattie & Timperley, 2007).

More information, larger effects

Wisniewski and colleagues tested this directly by coding each study for how much information the feedback contained. Rewards and punishment carry almost none. Corrective feedback says right or wrong and gives the correct answer. High-information feedback adds information about the process or the learner’s self-regulation as well (Wisniewski et al., 2020).

Feedback rich in information had an effect size of 0.99, about four times that of rewards or punishment

Average effect on learning by the information content of the feedback, from a meta-analysis of 435 studies · effect size (d)

The same pattern shows up in research on computer-based learning, which is closest to what a self-paced course does. Fabienne Van der Kleij and colleagues pooled 40 studies (70 effect sizes) of feedback delivered by computer (Van der Kleij et al., 2015).

Telling learners only whether they were right had an effect close to zero. Giving the correct answer helped somewhat. Explaining why helped most, and the abstract notes that explanations had a particular advantage for higher-order learning outcomes. For anyone who has set up a quiz that just shows a green tick or a red cross, that is the most practical finding in this entry.

Timing: it depends, but don’t skip it

Timing is where the evidence is least settled. In the computer-based meta-analysis, delayed feedback was associated with smaller effects (Van der Kleij et al., 2015). Hattie and Timperley argue that earlier studies gave inconsistent results partly because they ignored the level of the feedback: the best timing for correcting a simple task answer may not be the best for feedback on strategy. Valerie Shute’s review recommends immediate feedback for difficult tasks and for learners who are struggling, and delayed feedback for simpler tasks and for transfer (Shute, 2008).

What is clear is that some feedback beats none. In a lab study of multiple-choice testing, Andrew Butler and Henry Roediger found that both immediate and delayed feedback increased correct answers on a later test compared with no feedback, and reduced the number of wrong options learners carried over from the quiz. Their advice is blunt: “Educators should provide feedback when using multiple-choice tests” (Butler & Roediger, 2008).

How strong is the evidence?

Strong for the core claims, with real limits. Several large meta-analyses agree that feedback improves learning on average and that more informative feedback works better. But the studies vary enormously, Wisniewski and colleagues found an asymmetry in the results from journal-published studies that can signal publication bias, and they caution against leaning too hard on any average. Some subgroups rest on few studies, and much of the research is in schools and universities rather than adult, self-paced courses. We’d treat the direction of the findings as well supported and the exact effect sizes as rough guides.

How to apply it in a course

The research boils down to a test you can apply to every piece of feedback in a course: does it tell the learner what they were aiming for, how their attempt compares, and what to do next? Here’s how that plays out in a build.

We’ll use one invented example: a self-paced course teaching new hospitality supervisors to plan a weekly staff roster within a wage budget. It’s typical of practical skills where learners make predictable mistakes.

1. Make the goal visible before the task

Feedback can’t say “how am I going?” unless the learner knows where they’re going. Before each practice task, state what a good answer looks like: “A workable roster covers every shift with the right mix of skills and stays within the weekly wage budget.” This is the job your learning objectives should already be doing; the feedback refers back to them.

2. Write feedback for each wrong answer, not just the right one

For every quiz question, list the two or three mistakes learners are most likely to make, and write a short explanation for each. In the rostering course, a question asks which draft roster works for Saturday lunch. One wrong option leaves two junior staff alone at the peak. Its feedback says what goes wrong, gives the rule, and stops there:

Not quite. This roster stays within budget, but it leaves two first-month staff alone between 12 and 2, the busiest window. Cover the peak with at least one experienced person, then fill the quieter shifts.

That is elaborated feedback: the what, how and why that Shute recommends, in one small piece (Shute, 2008). “Incorrect. The answer is C.” is corrective feedback, and it leaves most of the learning on the table.

3. Keep each piece small enough to use

Shute’s guidelines warn against feedback so long or complex that learners skip it. Two or three sentences per answer is usually enough. If the explanation needs more, link to the lesson section that covers it rather than rewriting the lesson in the feedback box. This is the same logic as cognitive load theory: feedback that overloads working memory won’t be processed.

4. Always give a next step

“Where to next?” is the question most course feedback forgets. After each explanation, send the learner somewhere: a similar question to try, a 90-second clip that shows the rule, or a worked example. In a self-paced course, a retake with a different version of the question is a natural next step, and it doubles as retrieval practice.

5. Leave out the comparisons and go easy on praise

Leaderboards, class averages and “you scored in the top 20%” all push attention towards the self and away from the task, which is exactly where Kluger and DeNisi found feedback lost its effect. Shute’s review also advises avoiding comparisons with other learners. A brief “good” on a correct answer is fine, but make the useful part a sentence about why it was right, so learners who guessed find out.

6. Use people and rubrics for open-ended work

Automated feedback works for questions with knowable answers. For open-ended work, such as a learner’s own roster for their own venue, you need a person, a rubric, or both. A rubric written against the course objectives lets a reviewer (you, a tutor or a peer) give feedback at the process level (“your Sunday is short-staffed because you rostered by headcount rather than by expected covers”) rather than just a mark. If you use peer review, give reviewers the rubric and a model answer.

7. Ask learners to act on it

Feedback only works if it’s used. Build a short step after each marked assignment: learners fix one thing the feedback named and resubmit, or write a sentence on what they’ll do differently in the next task. This closes the loop that the “where to next?” question opens.

Feedback design belongs in the design phase of the ADDIE model, alongside the assessments themselves, not as an afterthought in the platform’s settings. Our online course design checklist includes it as a check for each module.

How the platforms support it

Most of what makes feedback effective is in the writing: the explanations, the next steps, the rubric. The platform decides how much of that you can attach and where learners see it. We haven’t yet published our reviews of Thinkific, LearnWorlds, Teachable and Kajabi, and quiz features change often, so check these in your platform’s current documentation, on the plan you’re paying for:

• Feedback per answer option. Can you write a different explanation for each wrong answer, or only one message for “correct” and one for “incorrect”?
• When feedback appears. Can learners see the explanation straight after each question, or only at the end of the quiz? Can you choose?
• Retakes and question pools. Can learners retake a quiz, and can each attempt draw different questions, so the next step isn’t just memorising the last answer?
• Assignment submissions with comments. Can learners upload work, and can you or a tutor reply with written, audio or video comments and ask for a resubmission?
• A link from feedback to the lesson. Can feedback text include a link back to the relevant lesson or clip?

Whatever the platform, the gains come from what you write into those boxes. A tick or a cross is the least useful feedback a course can give.

Further reading

• Hattie & Timperley (2007), “The power of feedback”. The source of the three questions and the four levels. Long, but the clearest framework there is.
• Wisniewski, Zierer & Hattie (2020), “The power of feedback revisited”. The most up-to-date large meta-analysis, with an honest discussion of its limits. Open access.
• Shute (2008), “Focus on formative feedback”. The most practical review: its tables of things to do and things to avoid translate directly into course design.
• Van der Kleij, Feskens & Eggen (2015). The meta-analysis closest to self-paced online courses: feedback delivered by computer.
• Kluger & DeNisi (1996). The paper that showed feedback can make performance worse, and why.

For the rest of the principles behind good course design, browse our research library.