1 Concept and definition

1.1 Basic meaning

Prediction error is the difference between what is expected and what actually happens. It can refer to a mismatch in perception, outcome, or measurement, and it is used in many fields to describe how systems respond when reality does not match prior expectations. The concept is central to understanding adaptation, because new information often becomes meaningful when it departs from what was anticipated.

1.2 Expected outcome versus actual outcome

The idea depends on two elements: a prediction and an observed result. The expected outcome may come from experience, a learned model, a forecast, or a belief, while the actual outcome is the event or data point that occurs. When the two differ, the size and direction of the gap can guide later judgments, decisions, or learning processes.

1.3 Positive and negative prediction error

A positive prediction error occurs when the outcome is better or greater than expected. A negative prediction error occurs when the outcome is worse or smaller than expected. In many applications, these two forms are treated as signals with different implications: one can strengthen future expectations, while the other can weaken them or prompt revision.

1.4 Prediction error and surprise

Prediction error is often associated with surprise, but the two are not identical. Surprise is a subjective or experiential reaction to an unexpected event, whereas prediction error refers more specifically to the measurable discrepancy between predicted and observed states. A highly surprising event usually contains a large prediction error, though the term may be used more technically in formal models than in everyday language.

2 Theoretical foundations

2.1 Learning theory

In learning theory, prediction error is a basic mechanism for updating behavior. When an outcome differs from what was expected, the mismatch provides information that can strengthen, weaken, or redirect future responses. This principle appears in classical conditioning, operant learning, and broader accounts of associative change.

2.2 Reinforcement learning

Reinforcement learning uses prediction error to guide improvement over time. An agent compares an expected reward with the reward actually received, then adjusts future choices accordingly. This framework has been influential in psychology, neuroscience, and computer science because it offers a compact explanation of how performance can improve through feedback.

2.3 Expectation updating

Expectation updating describes the process by which beliefs or forecasts change after new evidence appears. Prediction error acts as the trigger for this revision, with larger discrepancies often producing larger adjustments. In practice, the amount of updating also depends on confidence, prior experience, and the perceived reliability of the new information.

2.4 Relationship to uncertainty

Prediction error is related to uncertainty, but the two concepts are distinct. Uncertainty concerns how unsure a system is about an outcome, whereas prediction error concerns the mismatch after the outcome occurs. High uncertainty can make prediction errors harder to interpret, since a missed expectation may not signal that a model is wrong so much as incomplete.

3 Prediction error in psychology

3.1 Cognitive processing

In psychology, prediction error is used to explain how the mind compares incoming information with internal expectations. This comparison influences judgment, perception, and choice. When discrepancies are repeated, people often alter their mental models to better fit experience.

3.2 Attention and salience

Unexpected events tend to capture attention more strongly than routine ones. Prediction error contributes to salience by marking information as noteworthy or informative. As a result, people often allocate more processing resources to events that violate expectations, especially when the mismatch is large or personally relevant.

3.3 Memory and learning

Prediction error can enhance memory formation by highlighting informative experiences. Events that differ from expectation may be encoded more deeply because they offer useful updates for future behavior. This helps explain why unusual, emotionally charged, or inconsistent experiences are often remembered well.

3.4 Belief revision

Beliefs are more likely to change when evidence repeatedly generates prediction error. If a person expects one result but repeatedly encounters another, confidence in the original belief may decline. Belief revision can be gradual or abrupt, depending on how strongly the new evidence is interpreted and how central the belief is to the individual’s broader view.

4 Prediction error in neuroscience

4.1 Neural signaling

Neuroscience treats prediction error as a signal carried by neural activity that indicates a difference between expected and obtained outcomes. Such signals are thought to support learning by informing the brain when to strengthen or weaken associations. Research in this area often focuses on reward, but similar principles may apply to sensory and cognitive processing.

4.2 Reward prediction error

Reward prediction error is a prominent subtype in which the brain compares expected reward with actual reward. If the reward is greater than anticipated, neural activity associated with positive error may increase; if the reward is less than expected, the signal may decrease or shift in another direction. This mechanism helps explain how rewards shape future choices and habits.

4.3 Dopamine and reinforcement

Dopamine has been closely linked to reward prediction error, especially in models of reinforcement learning. Dopaminergic activity is often described as signaling that an outcome is better or worse than predicted, thereby influencing learning rates and action selection. While the relationship is complex, this account has been highly influential in explaining reward-based adaptation.

4.4 Brain regions associated with prediction error

4.4.1 Midbrain pathways

Midbrain structures are often implicated in the generation of prediction error signals, particularly in reward-related learning. These pathways help distribute information about unexpected outcomes to other parts of the brain, supporting updates in behavior and expectation.

4.4.2 Striatal processing

The striatum is frequently associated with learning from rewards, habits, and action selection. Prediction error signals in this region are thought to contribute to the strengthening or weakening of response tendencies, especially when feedback is immediate and meaningful.

4.4.3 Cortical contributions

Cortical areas participate in evaluating expectations, integrating context, and revising higher-level beliefs. They are important when prediction errors involve more than simple reward, such as social inference, language comprehension, or complex decision-making. These contributions allow prediction error to influence broader reasoning rather than only basic learning.

5 Prediction error in economics and decision-making

5.1 Consumer expectations

In economics, prediction error can describe the gap between expected and actual outcomes in consumer behavior. When product quality, price, or service differs from anticipation, consumers may adjust future purchases or change brand loyalty. Such errors can shape satisfaction and market response.

5.2 Risk and choice behavior

Decision-making under risk often involves estimating likely outcomes and then comparing them with what happens. Prediction error matters because people learn not only from gains and losses but also from mismatches between expected and realized results. This makes it relevant to models of risk preference and adaptive choice.

5.3 Forecasting and model adjustment

Forecasting relies on continually comparing predictions with observed data. Prediction errors reveal where a model is performing well and where it needs refinement. In economic analysis, repeated discrepancies may indicate changes in consumer behavior, market conditions, or underlying assumptions.

5.4 Behavioral economics applications

Behavioral economics uses prediction error to explain how people revise expectations in response to incentives and feedback. It helps account for phenomena such as loss sensitivity, changing reference points, and persistence of mistaken beliefs. The concept is useful because it connects learning from outcomes with the psychological limits of judgment.

6 Statistical and methodological uses

6.1 Residuals and forecast errors

In statistics, prediction error is often discussed as a residual or forecast error: the difference between a model’s predicted value and the observed value. This form is essential for evaluating how well a model fits the data. Smaller errors generally indicate better predictive performance, although interpretation depends on the context and measurement scale.

6.2 Model evaluation

Prediction error is used to assess accuracy, compare models, and detect overfitting. A model that performs well on training data but poorly on new data may have low apparent error in one setting and high error in another. For this reason, analysts often examine error across multiple samples or validation sets.

6.3 Measurement and interpretation

The meaning of prediction error depends on how the variable is measured and what the model is designed to predict. Errors may arise from noise, missing variables, poor assumptions, or genuine unpredictability in the system. Careful interpretation is needed to avoid treating every discrepancy as evidence that the model is fundamentally wrong.

6.4 Error decomposition

Prediction errors can sometimes be separated into components such as bias and variance, or into systematic and random parts. This decomposition helps identify whether a model is consistently off in one direction or merely fluctuating around the true value. Such analyses are useful in both empirical research and practical forecasting.

7 Applications

7.1 Education and skill learning

Prediction error plays a major role in education because learners improve when feedback reveals gaps between expectation and performance. In skill acquisition, errors help refine motor actions, conceptual understanding, and problem-solving strategies. Teachers and training systems often use feedback that makes these gaps visible and actionable.

7.2 Habit formation

Habits are shaped through repeated feedback and reinforcement. When an action produces an outcome better or worse than expected, prediction error can alter the likelihood of repeating that action. Over time, consistent outcomes may make behavior more automatic, while unexpected outcomes can interrupt or redirect a routine.

7.3 Clinical and behavioral research

In clinical and behavioral research, prediction error is studied to understand learning differences, mood processes, and decision patterns. Researchers may examine how people with particular conditions respond to reward, punishment, or expectancy violation. The concept is valuable because it links subjective experience with measurable patterns of adaptation.

7.4 Artificial intelligence and machine learning

Artificial intelligence uses prediction error as a core learning signal. Many algorithms reduce error by comparing predicted outputs with actual targets and adjusting internal parameters accordingly. This approach appears in supervised learning, reinforcement learning, and adaptive control systems, making prediction error a foundational concept in modern computation.

8.1 Expectancy violation

Expectancy violation is the experience of encountering something that does not match what was anticipated. It is closely related to prediction error, though it often emphasizes the psychological or social impact of the mismatch rather than its formal measurement.

8.2 Prediction interval

A prediction interval is a statistical range within which a future observation is expected to fall. Unlike prediction error, which refers to the realized discrepancy, a prediction interval expresses uncertainty before the outcome is observed.

8.3 Forecast error

Forecast error is the difference between a predicted value and the actual observed value, especially in statistics, economics, and planning. It is a common technical cousin of prediction error and is often used when discussing models or time-series estimates.

8.4 Reinforcement signal

A reinforcement signal is feedback that influences the probability of future behavior. In learning models, prediction error can function as such a signal by indicating whether an outcome should strengthen or weaken an association.