A person reaches for a cup while a screen shifts the apparent position of their hand a few degrees to the side. On the next reach, their movement changes. That adjustment may reveal learning from an unexpected sensory result, but it does not give an unfiltered reading of a prediction inside the brain. The person might have corrected the movement deliberately, changed attention, altered the movement goal or reacted to the last seen error. To measure sensorimotor prediction, researchers must specify what outcome was expected, what was observed, and which alternative route could produce the same behavior.
Sensorimotor prediction refers to estimating the sensory consequences of an action. It is related to motor planning and feedback control, but neither is a synonym. A predicted feeling when one presses a surface, an expected visual position of the hand and an anticipated sound after a button press are different events. This Methods & Measures entry compares task families that make predictions experimentally testable. It remains distinct from measuring proprioception, which concerns information about body position and movement, and motor imagery, which can involve simulated action without overt execution.
Describe the prediction before observing the error
A useful experiment begins by saying which information is available before an action. A participant might see a target, learn a consistent relation between movement and cursor displacement, and prepare a reach. The predicted cursor location can then be contrasted with the observed location. If the experimenter imposes a rotation, the difference becomes a candidate sensory prediction error. But it is only interpretable if the task shows that the movement and expected mapping were sufficiently stable to make a prediction plausible.
One can ask participants to report expected outcomes explicitly, but a verbal expectation is not necessarily the signal used for motor control. A person may predict a cursor position accurately while the movement system updates differently. Conversely, behavior can adapt without a person articulating the altered mapping. Researchers should name whether their target is conscious expectation, trial-by-trial motor updating or neural response to an action outcome. A single word, “prediction,” should not make those levels interchangeable.
Blakemore, Goodbody and Wolpert studied how sensorimotor context supports anticipation of action consequences. Their work illustrates why self-generation alone is insufficient: the system must estimate what movement and environmental circumstances will produce. A motor command to pull on an object, for example, can be accompanied by anticipatory grip-force adjustments. Such adjustments are evidence of predictive control when their timing precedes feedback, but their size also depends on object properties, practice and the movement selected.
Expected outcomes are often learned. Before a novel cursor rotation, a participant expects a familiar hand-to-screen relation. After repeated exposure, that expectation may shift. A study should record the learning history, number of trials and any strategic instructions. The same mismatch early and late in a session may have different meaning because the internal estimate has changed. Treating all perturbed trials as equivalent can hide the process the experiment seeks to measure.
Distinguish feedforward change from online correction
In a reaching task, a movement can be adjusted during its execution as visual feedback arrives. That online correction is not the same as changing the next movement’s prediction. Researchers can study initial movement direction before feedback can plausibly affect it, and compare that with later trajectory corrections. They can also remove feedback on selected trials to estimate whether an updated movement is expressed without immediate visual guidance. Neither technique is perfect, but both sharpen the temporal contrast.
Tseng and colleagues examined adaptation to visuomotor perturbations and distinguished sensory prediction error from the motor correction a person makes to hit a target. Their result is important for construct validity: a participant can reduce target error through a compensatory strategy without making the same kind of predictive update. If the outcome measure is simply how close the hand or cursor came to the target, these processes are blended. Experimental designs can vary target error and predicted sensory error separately to ask which drives adaptation.
Trial-by-trial change is often summarized as a learning rate. That parameter belongs to a model whose assumptions should be stated. Does the model assume one stable state, several timescales, or a constant sensitivity to error? Are errors measured in visual, hand-centered or goal-centered coordinates? A fitted rate can be useful for comparing conditions, but it is not a direct biological quantity. Different models can fit similar trajectories while assigning change to different processes.
When conditions include catch trials or no-feedback trials, participants may notice the manipulation and change strategy. An aftereffect when the perturbation is removed can support an implicit adaptation component, yet even an aftereffect needs interpretation. People may retain a habit of aiming away from the target, or experience residual perceptual shifts. Asking about strategy, measuring reach angles and varying feedback together gives a fuller picture than one final displacement score.
Use sensory attenuation with its controls
Self-produced touch may feel weaker than comparable externally produced touch. Blakemore, Wolpert and Frith’s work on self-generated tactile sensation has been influential because predicted action consequences can attenuate sensory responses. A force-matching task can compare how participants reproduce an externally applied force through direct touch or an indirect device. If direct self-produced force is overproduced, attenuation is one possible interpretation. The conditions must be matched carefully enough that mechanics, delay and attention do not provide a simpler answer.
Shergill and colleagues used force-matching methods to investigate why self-generated forces can be perceived differently. The task gives a behavioral signature, not a literal readout of a forward model. The participant’s skin location, motor effort, force transmission, timing and response rule all affect the match. A design that changes direct contact to a joystick changes more than whether feedback is predicted; it also changes the action and sensory channel.
Timing manipulations can sharpen inference. If a tactile consequence arrives later than expected or at a different location, attenuation may change. An experiment can compare expected and unexpected consequences while keeping the physical stimulation as similar as possible. Bays and colleagues tested whether tactile attenuation is predictive rather than a later inference based on self-generation. Their result supports a prospective component in their task, while leaving open how much different kinds of touch or motor contexts recruit the same mechanism.
Neuroimaging adds another measure: responses to predictable self-produced stimuli may differ from responses to externally generated or unexpected stimuli. Blakemore, Rees and Frith investigated this using action-linked tones. Yet a neural response difference can reflect attention, salience or task context alongside prediction. A region’s activity does not identify the computation on its own. Behavioral contrasts and the exact sensory event are needed to interpret it.
Do not equate agency judgments with prediction
People often feel more or less responsible for an outcome depending on whether it followed their action as expected. That makes agency tasks tempting proxies for sensorimotor prediction. They are imperfect proxies. A person can predict an event but deny causing it, or cause an event they did not foresee. Social information and post-event inference can influence agency judgments. Studies should keep a motor prediction measure separate from a question about authorship or control.
Intentional binding is a shift in the perceived timing of an action and its effect. Moore and Haggard separated predictive and inferential contributions under specific contingency conditions. The measure is valuable because the effect can be manipulated, but a temporal judgment also depends on attention, probability learning and the occurrence of the outcome. It should not be treated as a pure meter of the forward model or of a person’s enduring sense of agency.
One strong design varies the probability of an action effect and occasionally omits it. If temporal judgments change even when the expected effect is omitted, that supports an anticipatory contribution. If they change only after an effect occurs, post-event inference is implicated. Both can operate. The study must report the actual contingency participants learned, not merely the probability programmed by the investigator, because participants can miss patterns or acquire different beliefs.
Self-report can be useful in parallel. Asking what a participant expected, whether an outcome felt self-caused, and how certain they are can reveal whether the manipulation was understood. The answers are not interchangeable with movement kinematics or sensory attenuation. Disagreement between measures may signal different mechanisms, measurement noise or a task whose response demands are poorly understood. Treating disagreement as a finding is better than averaging it away.
Identify common confounds before fitting a model
Motor variability is not always prediction error. A person may miss a target because of tremor, fatigue, an awkward posture or changing speed. A sensory observation can be noisy because of visual blur or delayed display. If a task attributes every miss to an incorrect internal prediction, it overstates what was learned. Repeated baseline trials and estimates of movement and sensory variability help separate unpredictable noise from systematic bias.
Proprioception is another confound. A participant’s estimate of their hand position may shift during adaptation. A cursor error could therefore reflect a mismatch between visual and proprioceptive estimates, not just a change in predicted visual feedback. The companion proprioception measure asks how position and movement estimates are obtained. A sensorimotor prediction study should at least acknowledge the body-state estimate on which its model depends and avoid labeling every recalibration as an efference-copy effect.
Strategy and instruction matter. If a researcher tells participants to aim at a neighboring target to counter a screen rotation, later success may be deliberate compensation. If the instruction is withheld, participants may discover a strategy anyway. Asking about aim on selected trials can help, though the report itself may alter behavior. Researchers should predefine whether the central outcome is performance, implicit adaptation, explicit strategy or a combination, then choose controls suited to it.
Expectation can be manipulated without movement. A familiar warning cue may make a tone predictable even if the person does not act. To claim a motor-specific prediction effect, compare action-linked prediction with non-motor statistical prediction where feasible. Differences in attention, timing and arousal still need control. “Self-generated versus external” is a useful first contrast but not a complete account of why sensations differ.
Population comparisons require particular caution. Children and adults can differ in task comprehension, movement speed and attention. Clinical groups may differ in sensory access, medication or fatigue. A group difference in force matching, binding or adaptation does not diagnose a predictive-system deficit by itself. Reliability across sessions, enough trials, and convergence with other measures are needed before a person-level score is treated as stable.
Interpret the result at the level tested
A careful report says, for example, that after a visual rotation the first part of participants’ reaches shifted across trials, even on no-feedback trials. It states the rotation, feedback schedule, target configuration and strategy instructions. It does not jump from that effect to a claim about all action prediction in daily life. A force-matching result should name the contact and reproduction conditions. An imaging result should name the sensory event and neural contrast.
Uncertainty should travel with the estimate. A model may assign a higher learning rate to one group, but that difference can depend on the number of trials and the chosen state-space model. Show individual trajectories, parameter intervals and model checks. If a simpler strategic account explains the same data, the interpretation should remain open. Prediction is inferred from a pattern of behavior and physiology under constraints, not observed as a labeled internal variable.
The Institute’s measurement questions synthesize the cited motor-control, sensory-attenuation and agency studies. They are not a validated assessment scale. Their purpose is to keep predicted sensory consequences separate from correction, sensory estimation, conscious expectation and authorship judgments. A study becomes more informative when these constructs are made to disagree in planned contrasts rather than left to overlap in one score.
What the evidence does not show
These studies do not establish a single universal sensorimotor-prediction ability. Reaching adaptation, tactile attenuation, action-effect timing and explicit expectation have different tasks and sources of error. A participant may show one effect and not another without inconsistency. The research does not show that a larger sensory attenuation effect is always healthier or that a particular adaptation rate is optimal across environments.
Successful movement does not prove that the predicted sensory result was accurate; good online correction can rescue an initial error. Conversely, a movement error does not prove a failed prediction; the body and environment are noisy. Neural attenuation does not by itself reveal a cerebellar computation. Intentional binding does not, on its own, diagnose agency or intention. Each proxy requires its own validity argument.
The cited evidence is mostly from constrained laboratory actions and specific sensory channels. It does not directly establish how people anticipate every consequence of complex sport, social action or technology use. Extending findings to those settings requires tasks that preserve their relevant uncertainty and opportunities for learning. The bounded conclusion is often enough: one experimental contrast changed a particular signature consistent with prediction under stated assumptions.
Return to the shifted cursor
The reach toward the cup becomes informative when the experimenter records where the hand started, where it aimed, what feedback appeared, and how the next reach began. If the first movement direction changes even when feedback is absent, the case for updated feedforward control grows. If success improves only through late corrections, a different process may dominate. The visible adjustment is the start of the measurement argument, not its end.
Sensorimotor prediction is worth measuring because action and sensation continually inform one another. The strongest study makes that loop temporarily legible: it states the expected result, perturbs the actual result, observes the response and tests alternative explanations. It finishes with the exact prediction relation the task revealed, leaving the wider capacities of a person outside the reach of one laboratory score.
Research references
- Blakemore, S.-J., Goodbody, S. J., & Wolpert, D. M. (1998). Predicting the consequences of our own actions: The role of sensorimotor context estimation. Journal of Neuroscience, 18, 7511–7518. View source
- Tseng, Y.-W., Diedrichsen, J., Krakauer, J. W., Shadmehr, R., & Bastian, A. J. (2007). Sensory prediction errors drive cerebellum-dependent adaptation of reaching. Journal of Neurophysiology, 98, 54–62. View source
- Blakemore, S.-J., Wolpert, D. M., & Frith, C. D. (1998). Central cancellation of self-produced tickle sensation. Nature Neuroscience, 1, 635–640. View source
- Shergill, S. S., Bays, P. M., Frith, C. D., & Wolpert, D. M. (2003). Two eyes for an eye: The neuroscience of force escalation. Science, 301, 187. View source
- Bays, P. M., Flanagan, J. R., & Wolpert, D. M. (2006). Attenuation of self-generated tactile sensations is predictive, not postdictive. PLoS Biology. View source
- Blakemore, S.-J., Rees, G., & Frith, C. D. (1998). How do we predict the consequences of our actions? A functional imaging study. Neuropsychologia, 36, 521–529. View source
- Moore, J., & Haggard, P. (2008). Awareness of action: Inference and prediction. Consciousness and Cognition, 17, 136–144. View source