The Power of Feedback: What Hattie and Timperley (2007) Found About Which Feedback Works
By Noam Shemla · The Studies
TL;DR Feedback is among the strongest influences on learning, but the kind decides whether it helps. Pooling 12 earlier meta-analyses, Hattie and Timperley found an average effect of 0.79, about twice the typical effect of schooling. Their model says good feedback answers three questions: where am I going, how am I going, and where next. Praise of the person helped least.
- Paper: Hattie, J., & Timperley, H. (2007). The power of feedback. Review of Educational Research, 77(1), 81–112.
- Type: A review that draws on earlier meta-analyses, mainly of feedback in classrooms, and proposes a model of feedback.
- Evidence: 12 meta-analyses of feedback (196 studies, 6,972 effect sizes), set against a synthesis of over 500 meta-analyses of influences on achievement.
- Access: Read in full from a posted copy of the journal article. Every figure below is the paper’s, taken from its text.
Teachers, managers and coaches all give feedback, and most assume it helps. John Hattie and Helen Timperley asked how much, and found that the answer depends on what the feedback says.1
How big is the effect?
The authors start by setting feedback against everything else that influences how well students do. They draw on Hattie’s earlier synthesis of over 500 meta-analyses, in which the typical effect of schooling was 0.40. Twelve earlier meta-analyses had looked at feedback in classrooms. Pooled, they covered 196 studies and 6,972 effect sizes, and the average effect was 0.79, twice the typical effect of schooling.
Where feedback sits among other influences on achievement
| Influence | Average effect |
|---|---|
| Direct instruction | 0.93 |
| Reciprocal teaching | 0.86 |
| Feedback | 0.79 |
| Students’ prior cognitive ability | 0.71 |
| Homework | 0.41 |
| Reducing class size | 0.12 |
The spread behind the average is wide. The studies with the highest effects involved information about a task and how to do it more effectively. The lowest were about praise, rewards and punishment.
Three questions, four levels
Their model says that the point of feedback is to reduce the gap between where a learner is and where they are trying to get to. For that, feedback has to answer three questions.
- Where am I going? What are the goals.
- How am I going? What progress is being made towards them.
- Where to next? What to do to make better progress.
They call these feed up, feed back and feed forward. How well the answers close the gap depends partly on the level the feedback works at.
The four levels of feedback
| Level | What the feedback is about |
|---|---|
| Task | How well the task is being performed or understood |
| Process | The main processes needed to understand or perform the task |
| Self-regulation | How the learner monitors, directs and regulates their own actions |
| Self | The learner as a person, unrelated to the specifics of the task |
What the paper says about praise
Praise sits at the self level, and the paper is blunt about it. It says praise often carries little information about the task and deflects attention away from it. It cites a meta-analysis of teacher praise that found an overall effect of 0.12, and another source that found an effect of 0.09 for praise, against 0.34 for no praise at all.
The paper separates two kinds. Praise aimed at the person has little learning value. Praise aimed at effort, self-regulation or the way the task was done can help, because the learner can turn it back into work on the task. The authors add that the reviews they drew on did not always tell the two kinds apart, and they stress that none of this means students dislike being praised.
Timing
The paper treats timing as dependent on the level. Immediate feedback helps while a task is being learned, and it can get in the way when the aim is fluency. Citing a meta-analysis of 53 studies, it reports that at the task level some delay was beneficial (0.36), while at the process level immediate feedback was beneficial (0.28).
How it fits with Kluger and DeNisi
Hattie and Timperley lean on Kluger and DeNisi’s meta-analysis, which we cover in The Effects of Feedback Interventions on Performance. They describe it as 470 effect sizes from 131 studies, with an average of 0.38, and with 32% of the effects negative. Both give 23,663 observations, but the original paper’s abstract gives 607 effect sizes and an average of .41. We have not reconciled the two. The message is the same in both: feedback helps on average, and a third of it does not.2
What this paper cannot tell you
- Speaking. Most of the evidence is from classrooms, in which feedback comes from a teacher on schoolwork. Speeches are not schoolwork.
- A single cause. The 0.79 is an average over many designs and many kinds of feedback, which is exactly why the authors argue the kind of feedback matters.
- A tested model. The three questions and four levels are a framework built from earlier research, offered to organise it.
- Anything after 2007.
What a speaker can take from it
Ask the three questions of your own practice. Where am I going: write down what this take should achieve, such as two minutes at a steady pace. How am I going: measure it. The free speaking speed test gives you your words per minute for a minute of talking, and the speech timer gives you the length. Where to next: choose one change for the next take, not five.
When you ask others for feedback, ask for the task and the process, not a verdict on you. “Your example came after the point it was meant to prove” is something to work with. “Great job” feels good and gives you nothing to do. For how to ask, see feedback someone can actually use and how to practise a presentation alone. The step from the paper’s praise finding to these two asks is our application, not something the authors tested on speakers.
This article summarises an open copy of the published paper. Every effect size, count and quotation of a number is the authors’ own, from the text. Where the paper reports another study’s figure, the article says so.
Measured
- The average effect of feedback across 12 earlier meta-analyses (196 studies, 6,972 effect sizes)
- Effect sizes for types of feedback, praise and timing, as reported from earlier studies
Inferred, not measured
- That the model’s levels explain why some feedback fails: the authors’ argument, assembled from earlier research, not tested in a new study
- That the findings carry over to feedback on speaking: our application
References
- Hattie, J., & Timperley, H. (2007). The Power of Feedback. Review of Educational Research, 77(1), 81–112. https://doi.org/10.3102/003465430298487
- Kluger, A. N., & DeNisi, A. (1996). The effects of feedback interventions on performance: A historical review, a meta-analysis, and a preliminary feedback intervention theory. Psychological Bulletin, 119(2), 254–284. https://doi.org/10.1037/0033-2909.119.2.254