There is a steady stream of research on how learners deal with mistakes. The idea that we “learn from our errors” sounds intuitive, although you sometimes hear the opposite claim as well: mistakes might inspire more mistakes. Earlier research discussed when errors can be productive. This was mainly when learners explicitly reflected on their own thinking.

A new meta-analysis by Alemdag, Eichelmann and Narciss in Review of Educational Research connects neatly to that earlier discussion, but shifts the focus to the mistakes of others. More specifically, it looks at what happens when learners study so-called “erroneous examples” and shows how crucial explanation and guidance become in that context.

What are “erroneous examples”?

They are worked examples in which one or more steps are deliberately wrong. Learners are expected to:

• detect the error,

• understand why it is wrong,

• and/or correct it.

The aim is straightforward: make common misconceptions visible so that learners don’t get stuck in them. Across 42 studies with 177 effect sizes, the overall effect is small but positive (g = .136). Yes, erroneous examples can help. However, they will not suddenly boost performance in dramatic ways. Think of it as a refinement rather than a transformation.

What actually seems to work?

This is the main takeaway: erroneous examples only help when learners are supported in understanding the mistake.

More specifically, two forms of instructional support make the difference:

• Self-explanation prompts: questions that require learners to articulate or justify the error.

• Instructional explanations: explicit clarification from the teacher or the material about why the error occurs.

Without such guidance, the effect more or less disappears. Simply showing a mistake is not enough. The analysis also shows that it doesn’t matter much whether the examples come from mathematics, STEM, social sciences, or whether they are used in higher or lower education. The effect stays small but consistently positive.

What seems less effective?

Interestingly, several approaches that might appear promising do not consistently help:

• highlighting or flagging errors,

• asking learners to correct the error,

• comparing erroneous and correct examples,

• or relying on differences between subject domains.

The effects were also larger in quasi-experimental studies and smaller in more rigorous experiments — a reminder to take the more conservative reading seriously.

So what should we do with this?

Three practical implications stand out:

1. Erroneous examples can be helpful, but only when well designed.
   The learning happens through thinking about the error, not through the mistake itself.

2. They are especially helpful for addressing common misconceptions safely.
   Using someone else’s mistake can reduce emotional or organisational barriers compared with using learners’ own errors.

3. Treat this as a targeted technique, not a broad instructional philosophy.
   It works a little, and only under particular conditions.

Overall, this meta-analysis echoes what we have seen in other areas: learning from errors works, but not automatically. Instruction needs to be designed so that learners don’t just see that something is wrong, but actually understand why. That is probably the main message here: mistakes become meaningful only when we give them the time and attention they require.