Do instructions help you learn a task for good, or just at first?
Being told the rules of a simple response task gave people a faster start, but once they had practised enough, they performed no better than people who learned by trial and error.
Source
The Curve of Learning With and Without Instructions
Study at a glance
- Design
- Human experiment — Exp 1: within-subject instruction+feedback vs feedback-only blocks across four set-size conditions; Exp 2: online between-subject instruction manipulation with 60 repetitions per item; exponential curve and drift-diffusion modelling.
- N
- N=204 · One hundred students in Experiment 1; 104 participants retained in Experiment 2 (140 collected).
- Population
- University students (Ghent) and online adult participants
- Outcome
- Reaction time learning curve parameters (initial gap, learning rate, asymptote), accuracy, and diffusion model parameters in the final repetitions
Structured fields used in claim comparison tables when every cited study has a complete layer.
What they did
Participants pressed left or right keys to pictures of objects, with the correct key for each picture either told in advance plus feedback, or discoverable only through feedback. In Experiment 1, students experienced both conditions in separate blocks, with the number of picture-key pairs to learn varied across groups. The researchers fit an exponential curve to each person's reaction times over repetitions, separating the starting gap, learning speed and final (asymptotic) speed, and fit a diffusion decision model to the last repetitions. Experiment 2 tested the hardest condition online with instruction given to one group only and 60 repetitions per picture.
What they found
In Experiment 1, instructions shrank the gap between starting and final performance, mainly when there were many pairs to learn, but did not change learning rate or final speed; Bayesian evidence favoured no effect on asymptotic speed. A small drift-rate advantage for instructed items in the hardest conditions came from people who had not yet finished learning. In Experiment 2, with longer practice, no instruction effect remained on final reaction times, accuracy, or diffusion parameters, and even the head-start effect was not replicated.
The limits
What it doesn't show
The task was a very simple arbitrary key-mapping task, so the results may not apply to complex, hierarchical skills where instructions have been found to matter for days. Experiment 1 set-size conditions differed in block structure and spacing, and Experiment 2 switched to a between-subject online design with many exclusions, which the authors suggest may explain why the head start did not replicate. Several Bayes factors were only anecdotal, so the evidence for 'no lasting effect' is moderate rather than decisive.
Key terms
- Stimulus-response (SR) mapping
- An arbitrary rule linking a stimulus to an action, such as 'if you see a clock, press left'.
- Exponential learning curve
- A function where reaction time falls quickly then levels off, described by a starting gap, a constant learning rate and a final asymptote.
- Asymptote
- The stable best performance a learner approaches after extensive practice.
- Diffusion decision model
- A model of two-choice decisions in which evidence accumulates until it hits a boundary; it splits reaction time and accuracy into drift rate, boundary separation and non-decision time.
- Bayes factor
- A ratio showing how much more the data support one hypothesis over another, which allows evidence for a null effect.
Flashcards
0 of 11 answers reviewed
Research intelligence for this paper
See its role on concept claims, tensions it is part of, placement history, and related discoveries.
Quiz yourself
Which learning curve parameter did instructions reduce in Experiment 1?
Common questions
If instructions don't help in the end, why give them?
They cut early errors and time wasted on trial and error, which is valuable; the study only shows that for simple tasks the final level of skill is the same.
Why model accuracy as well as speed?
People can trade accuracy for speed, so a reaction time curve alone could hide real differences; the diffusion model combines both.
Why did the head start disappear in Experiment 2?
The authors speculate that experiencing both learning modes within the same session in Experiment 1 may have amplified the benefit of instructed blocks.
More on Learning