Rescorla-Wagner Model
Rescorla-Wagner Model of Associative Learning · Also known as: Rescorla-Wagner Theory, Delta Rule, Error-Correction Learning
The Rescorla-Wagner Model is a quantitative theory of associative learning that predicts how organisms learn associations between stimuli (e.g., tone and shock in fear conditioning). The model proposes that learning is driven by prediction error—the difference between what is expected to occur and what actually occurs. When prediction error is large, learning is rapid; when prediction error is small, learning slows. The model captures asymptotic learning curves, blocking effects, and stimulus interactions, providing a principled framework for understanding learning dynamics.
Read the full method
Sign in with a free account to read this section.
When to use it
Use the Rescorla-Wagner Model when developing quantitative theories of learning, predicting learning curves in conditioning experiments, understanding how organisms learn predictions, or examining whether simple error-correction learning explains observed learning phenomena. It is foundational in computational neuroscience, animal learning research, and clinical understanding of conditioning-related disorders.
Strengths & limitations
- Mathematically simple yet powerful: few parameters, clear equations, yet captures diverse phenomena
- Empirically robust: predictions align with observed conditioning curves, blocking effects, and extinction patterns
- Theoretically integrative: grounds learning in prediction error, a principle now supported by neural recording and neurotransmitter studies
- Generative: spawned decades of extensions and variants, advancing learning science
- Assumes linear associative strength: real learning may involve nonlinear dynamics or threshold processes
- Does not account for stimulus representation changes: treats all stimuli equivalently after salience parameterization
- Limited to paired-stimulus paradigms: extensions needed for multi-stimulus or spatial learning contexts
- Weak on extinction and recovery: model predicts extinction as unlearning of associations, but data suggest extinction builds inhibitory learning—a distinction later models address
Frequently asked
What is prediction error in the Rescorla-Wagner model?
Prediction error (λ - V) is the difference between what the organism expects to occur (V, associative strength) and what actually occurs (λ, reinforcer presence). Large errors indicate surprise; small errors indicate predictability. Learning is proportional to error: surprising outcomes teach more than expected outcomes.
Why does blocking occur in the Rescorla-Wagner model?
When one stimulus (A) predicts the US well (high V for A), adding a second stimulus (B) does not increase prediction error. Since error is small, learning about B is slow or absent—it is 'blocked' by A. This explains empirically observed blocking: pre-training on one cue reduces learning about a second cue.
How does the model explain extinction?
During extinction (CS without US), the expected value (V) is positive but the actual outcome is zero (λ = 0), creating a prediction error. Learning reduces V: ΔV = α·β·(0 - V) = -α·β·V, decreasing associative strength. However, data suggest extinction involves learning inhibition, not unlearning; contemporary models add inhibitory learning.
Can I use the model to predict individual learning differences?
Yes, by fitting parameters (α, β) per individual. Individuals showing slow learning may have low α (low stimulus salience sensitivity) or low β (low reinforcer sensitivity). However, fitting few parameters to few observations can overfit; cross-validation and comparison to simple baselines are essential.
Sources
- Rescorla, R. A., & Wagner, A. R. (1972). A theory of Pavlovian conditioning: Variations in the effectiveness of reinforcement and non-reinforcement. In A. H. Black & W. F. Prokasy (Eds.), Classical conditioning II (pp. 64-99). Appleton-Century-Crofts. link ↗
- Simonetta, S. H., Schaafsma, S. M., & Meffert, H. (2010). The Rescorla-Wagner model of Pavlovian conditioning: Some current issues and applications. Neuroscience & Biobehavioral Reviews, 34(6), 821-835. link ↗
- Gluck, M. A., & Myers, C. E. (1993). Hippocampal mediation of stimulus representation: A computational theory. Hippocampus, 3(4), 491-516. DOI: 10.1002/hipo.450030410 ↗
How to cite this page
ScholarGate. (2026, June 3). Rescorla-Wagner Model of Associative Learning. ScholarGate. https://scholargate.app/en/psychology/rescorla-wagner-model