Abstract
Habits are a type of behavior: learned dispositions in which a cue reliably triggers a response without the response being selected for its current goal value. Cognitive psychology models the habit as one pole of a dual control architecture. A goal-directed system chooses actions from a represented response–outcome contingency and the current worth of the outcome; a habitual system stores direct stimulus–response links that fire automatically once a behavior has been repeated enough in a stable context. The experimental signature that separates them is outcome devaluation: a goal-directed agent stops responding when the outcome loses value, whereas a habit persists. This article sets out the dual-system account, its neural basis in the dorsolateral striatum, the real-world time course of habit formation, and the arbitration that decides which system controls behavior, with a worked automaticity-growth example and three interactive demonstrations.
Keywords: habit, outcome devaluation, stimulus–response learning, basal ganglia, automaticity
Habits is the Medical Subject Headings (MeSH) descriptor for the acquired behavior patterns that become regular and often unconscious through repetition. In the laboratory the term has a precise operational meaning that everyday usage blurs: a habit is behavior that has become insensitive to the value of its outcome and is instead evoked directly by antecedent cues. That definition is what lets researchers decide, experimentally, whether a given behavior is a goal-directed action or a true habit, and it organizes the whole modern literature on how behaviors migrate from one category to the other with practice.
Key Takeaways
- A habit is operationally defined as behavior that is insensitive to outcome value: it persists when the goal is devalued, unlike a goal-directed action.
- Two systems compete for control — a goal-directed system computing response–outcome contingencies, and a habitual system storing stimulus–response links — and behavior shifts toward the habit with repetition in a stable context.
- The habitual controller depends on the dorsolateral (sensorimotor) striatum; the goal-directed controller depends on the dorsomedial striatum and prefrontal cortex.
- In everyday settings automaticity grows as an asymptotic curve: repetition in a consistent context raises it steeply at first and then plateaus, with a reported median of about 66 days to reach the asymptote.
- Habit mechanisms are central to clinical conditions — drug addiction and obsessive–compulsive disorder are both framed as pathological over-reliance on stimulus–response control.
What Habits Are
The word habit in cognitive psychology does not mean merely a frequent or familiar behavior. It names a behavior under a specific kind of control. Following the associative framework set out by Dickinson (1985), an instrumental behavior can be governed in either of two ways. It can be a goal-directed action, selected because the agent represents a response–outcome (R–O) contingency and currently values that outcome; or it can be a habit, elicited directly by a stimulus–response (S–R) link that was strengthened through repeated reinforcement and that no longer consults the outcome's value at all. The behavior may look identical from outside — the same lever press, the same turn of the car toward the office — but the controlling representation differs (Figure 1).
Figure 1
The dual control architecture that produces instrumental behavior.
The experiment that operationalizes the distinction is outcome devaluation. An animal or person is trained to perform a response for a reward, the reward is then made worthless (by feeding the animal to satiety on it, or by pairing it with illness), and responding is tested in extinction. A goal-directed action collapses immediately, because the controller reads the now-zero value of the outcome. A habit continues at near-baseline rate, because the S–R link does not reference the outcome. This behavioral autonomy from the goal is the defining mark of a habit and the measure every subsequent study inherits (Dickinson, 1985; Wood & Rünger, 2016). Perez and Dickinson (2020) formalize the transition as a competition between a rate-correlation system that tracks the action–outcome relationship and a contiguity system that strengthens direct associations, with extended training shifting the balance toward the latter.
Types of Habits
MeSH classifies Habits under the broader descriptor Behavior and files four narrower descriptors beneath it. These subtypes are an indexing classification, not a theory of habit: MeSH groups them because each is a repetitive, cue-triggered oral or manual behavior of clinical interest, and the grouping is orthogonal to the goal-directed/habitual control distinction that organizes the experimental literature. A behavior can be any of these subtypes and still, in a given person, be more or less automatic.
| Subtype | Description |
|---|---|
| Bruxism | Habitual grinding or clenching of the teeth, awake or during sleep; the one child descriptor with a dedicated article on this site. |
| Fingersucking | Persistent sucking of a finger or thumb, a common early-childhood oral habit that usually self-resolves. |
| Nail Biting | Onychophagia — repetitive biting of the fingernails, often placed among the body-focused repetitive behaviors. |
| Tongue Habits | Habitual tongue postures and movements, such as tongue thrusting, with consequences for dental occlusion. |
Goal-Directed Versus Habitual Control
The clearest way to see the two controllers is to watch what each does when the outcome is devalued. The goal-directed system holds an explicit model: this response yields that outcome, and that outcome is worth this much now. Devalue the outcome and the computed value of the response drops to zero, so responding stops at once. The habitual system holds only this stimulus has been followed by this response, a link built up by reinforcement; devaluation leaves that link untouched, so responding continues. The degree to which responding survives devaluation is therefore a direct readout of how habitual the behavior has become.
This maps onto the computational distinction between model-based and model-free reinforcement learning. A model-based controller plans using a learned model of the environment's transitions and rewards, reproducing goal-directed sensitivity to devaluation; a model-free controller caches a scalar value or action tendency updated by prediction error, reproducing habitual insensitivity. Extended or over-trained behavior, and behavior acquired under conditions that discourage deliberation, leans on the model-free/habitual controller (Wood & Rünger, 2016; Robbins & Costa, 2017).
Demo 1 · Outcome devaluation
Train a lever-press response over several sessions, then test it when the outcome is still valued or has been devalued (made worthless). Early in training the response is goal-directed and collapses under devaluation; with more training a stimulus–response habit takes over and persists.
The Neuroscience of Habit
The two controllers are dissociable in the brain, and the basal ganglia carry much of the load. Across lesion, recording, and inactivation studies, goal-directed control depends on the dorsomedial striatum (caudate, in primates) together with prefrontal cortex, whereas habitual control depends on the dorsolateral (sensorimotor) striatum and its dopaminergic input (Graybiel, 2008; Smith & Graybiel, 2016). As a behavior becomes automatic, control migrates along this medial-to-lateral gradient.
A striking neural correlate is task-bracketing: as rats over-train a runway or maze habit, dorsolateral striatal activity reorganizes to fire at the start and end of the behavioral sequence rather than throughout it, as if the whole sequence has been packaged by chunking into a single retrievable unit (Graybiel, 2008; Smith & Graybiel, 2016). In humans, Knowlton, Mangels, and Squire (1996) showed the system's independence from declarative memory: amnesic patients with hippocampal damage learned a probabilistic classification habit normally while unable to remember the training episodes, whereas patients with striatal (Parkinson's) pathology showed the reverse. Habit learning is a distinct memory system, not simply well-rehearsed explicit knowledge (Robbins & Costa, 2017).
How Habits Form
Habits are built by repetition, but repetition alone is not enough: the behavior must be repeated in the presence of a stable context cue so that the cue can come to trigger it directly. Neal, Wood, and Quinn (2006) argue that once this cue–response binding is established, perception of the cue is sufficient to activate the response, which is why habits are so robust to changes in intention. Wood and Rünger (2016) synthesize the field around this context-cue principle, noting that the most effective habit interventions work by changing the environment rather than by exhorting the person.
The real-world time course was measured by Lally, van Jaarsveld, Potts, and Wardle (2010), who had participants adopt a new eating, drinking, or activity behavior in a constant daily context and rate its automaticity each day. Automaticity rose as a negatively accelerated (asymptotic) curve — steeply at first, then flattening — and the median time to reach the plateau was about 66 days, far longer than the folk figure of 21 days. Missing an occasional repetition did not derail the process. Measurement of the resulting habit strength relies on instruments such as the Self-Report Habit Index (Verplanken & Orbell, 2003), whose items probe automaticity, lack of awareness, and identity rather than mere frequency; Gardner (2015) reviews how the construct has been used to predict and change health behavior.
Demo 2 · How automaticity grows
Repeating a behavior in a stable context raises its automaticity along a negatively accelerated curve — steep at first, then flattening. Context consistency sets how fast the curve climbs. The model is A(n) = 1 − e−kn, fitted by Lally et al. (2010).
Arbitration Between Systems
Because both controllers are always present, behavior requires arbitration: a mechanism that decides, moment to moment, which system drives the response. The influential proposal is that the brain weights each controller by its reliability or its cost. The goal-directed/model-based system is accurate but slow and effortful; the habitual/model-free system is fast and cheap but rigid. When the environment is stable and the behavior well-practiced, the habitual controller is reliable enough to take over, freeing cognitive resources — the practical payoff of automaticity emphasized by Haith and Krakauer (2018), who distinguish the reduced cognitive load of a habit from motor skill per se. When the environment becomes volatile or the stakes rise, control reverts to the goal-directed system (Wood & Rünger, 2016). Watson and de Wit (2018) caution that laboratory devaluation tasks often struggle to produce robust habits in humans, so the balance observed depends heavily on training amount, context stability, and the sensitivity of the test.
Demo 3 · Which system is in control?
Both controllers are always present; arbitration weights each by its reliability. Over-training makes the habit reliable; a volatile environment makes the goal-directed model unreliable. A softmax over the two reliabilities sets the share of control.
Worked Example
Suppose a person adopts a new behavior — a daily walk after lunch — and performs it in a constant context. Model its automaticity A after n days with the asymptotic curve fitted by Lally et al. (2010):
A(n) = Amax × (1 − e−kn)
Take the asymptote Amax = 1.0 (fully automatic) and set the growth rate k so that automaticity reaches 95% of its asymptote at the reported median of 66 days. That requires k = ln(20) / 66 = 0.0454 per day. The behavior's automaticity then climbs as follows.
| Day n | Automaticity, consistent context | Automaticity, inconsistent context |
|---|---|---|
| 1 | 4.4% | 2.2% |
| 7 | 27.2% | 14.7% |
| 21 | 61.4% | 37.9% |
| 30 | 74.4% | 48.4% |
| 66 | 95.0% | 77.6% |
| 132 | 99.8% | 95.0% |
Two lessons fall out of the arithmetic. First, the curve is steep early and flat late: the behavior gains more than half its eventual automaticity in the first three weeks, which is why early repetitions feel the most effortful and matter the most. Second, context consistency sets the rate, not just the endpoint. Halving the effective repetition rate — performing the walk at an unpredictable time or place — pushes the 95% plateau from 66 days out to 132. The half-life of the consistent curve is ln(2)/k = 15.3 days; automaticity is already at 50% before three weeks are out, then yields diminishing returns. The model assumes an undisrupted context; a change of environment resets the cue that drives the behavior and the curve must, in effect, start again.
Discussion
The dual-system account has become the standard framework, but it is contested at its edges. Watson and de Wit (2018) document how hard it is to produce outcome-insensitive responding in healthy humans within a single session, which has led some to argue that human instrumental behavior is more persistently goal-directed than the animal literature implies, and that apparent habits sometimes reflect the cost of deliberation rather than a separate S–R system. The model-based/model-free dichotomy, likewise, is increasingly treated as a continuum with intermediate successor-representation schemes rather than two boxes.
What is not contested is the clinical reach of the construct. Everitt and Robbins (2005; 2016) frame drug addiction as a progressive shift from goal-directed drug-seeking to compulsive, stimulus-bound habit, mediated by a ventral-to-dorsal migration of striatal control that mirrors the medial-to-lateral shift seen in normal habit learning. Gillan et al. (2014) show that patients with obsessive–compulsive disorder form avoidance habits that persist after the threat is removed, and that this over-reliance on the habit system — rather than a primary deficit in goal representation — may generate compulsions. In both cases the pathology is not a new mechanism but the ordinary habit system operating where goal-directed control should prevail.
Current Directions
Three fronts are active. The first is computational arbitration: formal models that specify exactly how the brain weights model-based against model-free control — by relative uncertainty, by reward rate, or by the cost of computation — and tests of whether that weighting is itself learned. Perez and Dickinson (2020) offer one such dual-process formalization grounded in the classic associative data. The second is translational, using the addiction and OCD findings to ask whether strengthening goal-directed control, or disrupting maladaptive S–R links during reconsolidation, can treat compulsive disorders (Gillan et al., 2014; Everitt & Robbins, 2016). The third is measurement and ecology: whether laboratory devaluation and self-report indices such as the Self-Report Habit Index capture the same construct, and how to measure habit strength robustly in everyday life, given how fragile experimental habits have proved (Watson & de Wit, 2018; Gardner, 2015). The relationship between habit and motor skill — overlapping but not identical, since skill improves performance while habit removes the need to deliberate — is also being sharpened (Haith & Krakauer, 2018; Robbins & Costa, 2017).
Cognitive Implications
Habits are the cognitive system's solution to a resource problem. Deliberate, goal-directed control is accurate but expensive; offloading well-practiced behavior to a cue-driven controller frees working memory and attention for whatever is novel. The cost of that bargain is rigidity — the habit keeps running when circumstances change, which is adaptive in a stable world and maladaptive in a shifting one. This is why behavior change is so hard: intention operates through the goal-directed system, but entrenched behavior is controlled by cues the intention does not touch. The practical corollary, drawn repeatedly from the research, is that durable change comes from engineering the context — removing or adding the cues that trigger behavior — rather than from willpower alone (Neal et al., 2006; Wood & Rünger, 2016). Understanding habit also reframes clinical compulsion: treating it as a disorder of the balance between two normal control systems, rather than as a failure of insight or motivation.
Glossary
- Arbitration.
- The process that determines, moment to moment, whether the goal-directed or the habitual controller drives behavior, typically by weighting each system's reliability or cost.
- Automaticity.
- The quality of a behavior that runs quickly, efficiently, and without conscious control once triggered by its cue; the dimension measured by habit-strength indices.
- Basal ganglia.
- A group of subcortical nuclei, including the striatum, that are central to action selection and to the storage of habits.
- Behavioral autonomy.
- Dickinson's term for the defining mark of a habit: behavior that has become independent of the current value of its outcome.
- Chunking.
- The packaging of a repeated action sequence into a single retrievable unit, reflected in the task-bracketing pattern of dorsolateral striatal activity.
- Context cue.
- A stable feature of the environment that, after repetition, comes to trigger a habitual response directly, without reference to the goal.
- Dorsolateral striatum.
- The sensorimotor region of the striatum whose activity is necessary for habitual control; damage here restores goal-directed sensitivity.
- Dual-system theory.
- The framework holding that instrumental behavior is produced by two parallel controllers — goal-directed and habitual — that compete for expression.
- Goal-directed action.
- Behavior selected from a represented response–outcome contingency and the current value of the outcome; it stops when the outcome is devalued.
- Habit.
- Behavior that has become insensitive to the value of its outcome and is evoked directly by antecedent cues through a stimulus–response link.
- Model-based control.
- Reinforcement-learning control that plans using a learned model of transitions and rewards; the computational counterpart of goal-directed action.
- Model-free control.
- Reinforcement-learning control that caches values or action tendencies updated by prediction error; the computational counterpart of the habit.
- Outcome devaluation.
- The diagnostic procedure that reduces an outcome's value after training and tests whether responding persists; persistence indicates habitual control.
- Response–outcome (R–O) association.
- The learned link between an action and its consequence that underlies goal-directed control.
- Self-Report Habit Index (SRHI).
- A questionnaire by Verplanken and Orbell that measures habit strength through items on automaticity, lack of awareness, and behavioral identity rather than frequency alone.
- Stimulus–response (S–R) association.
- The direct cue-to-response link that underlies habitual control; it fires without consulting the outcome's value.
Key Researchers
Anthony Dickinson
Emeritus Professor of Comparative Psychology at the University of Cambridge, whose associative-learning analysis of the action–habit distinction and behavioral autonomy underlies the modern dual-system account. Wikipedia · Google Scholar · Faculty page
Claire M. Gillan
Associate Professor at Trinity College Dublin, whose work on goal-directed versus habitual control and transdiagnostic compulsivity shows how over-reliance on habit contributes to obsessive–compulsive disorder. ORCID
Ann M. Graybiel
Institute Professor at the Massachusetts Institute of Technology, whose recordings of the dorsolateral striatum established task-bracketing and the chunking of action sequences as neural signatures of habit formation. Wikipedia · Faculty page
Barbara J. Knowlton
Professor at the University of California, Los Angeles, whose patient studies demonstrated that habit (procedural) learning is a memory system distinct from the declarative memory supported by the hippocampus. Google Scholar · Faculty page
Trevor W. Robbins
Professor at the University of Cambridge, whose work on the striatal mechanisms of actions, habits, and compulsion links normal habit learning to addiction and compulsive disorders. ORCID · Wikipedia · Faculty page
Wendy Wood
Provost Professor Emerita at the University of Southern California, whose synthesis of the psychology of habit established the central role of stable context cues and the habit–goal interface in everyday behavior. ORCID · Wikipedia · Google Scholar · Faculty page
Frequently Asked Questions
Is a habit just a frequently repeated behavior?
Not in the technical sense. Frequency is neither necessary nor sufficient. A habit is behavior that has become insensitive to the value of its outcome: it is triggered by a cue and persists even when the goal it once served is devalued. A behavior can be frequent yet still goal-directed, or relatively infrequent yet habitual.
How is a habit distinguished from a goal-directed action experimentally?
By outcome devaluation. After training, the outcome is made worthless and responding is tested in extinction. A goal-directed action stops immediately because its controller reads the outcome's new value; a habit continues because its stimulus-response link does not consult the outcome. The survival of responding after devaluation is the operational measure of a habit.
Does it really take 21 days to form a habit?
No. The '21 days' figure has no solid basis. Lally and colleagues (2010) measured a median of about 66 days for a behavior to reach its automaticity plateau in a consistent context, with wide individual variation. The curve is steep early and flat late, so the first few weeks matter most, but reaching full automaticity typically takes two to three months.
Which part of the brain stores habits?
The dorsolateral (sensorimotor) striatum, part of the basal ganglia, is necessary for habitual control, while the dorsomedial striatum and prefrontal cortex support goal-directed control. As a behavior becomes automatic, control shifts from the medial toward the lateral striatum.
Why are habits so hard to break?
Because intention works through the goal-directed system, but an entrenched habit is driven by context cues that the intention does not touch. As long as the triggering cues remain in the environment, perceiving them tends to activate the response automatically. This is why changing the environment is usually more effective than relying on willpower.
What is the Self-Report Habit Index?
A questionnaire developed by Verplanken and Orbell (2003) that measures habit strength through items tapping automaticity, lack of awareness, difficulty of control, and behavioral identity, rather than how often the behavior occurs. It is the most widely used self-report measure of habit in health and social psychology.
How are habits related to addiction and OCD?
Both are framed as a pathological over-reliance on the habit system. Addiction is modeled as a progressive shift from goal-directed drug-seeking to compulsive, cue-driven habit; obsessive-compulsive disorder involves avoidance habits that persist after the threat is gone. In each, the ordinary habit mechanism operates where goal-directed control should prevail.
Is a habit the same as a motor skill?
They overlap but are not identical. A motor skill is improved performance of a movement; a habit is a change in the control of behavior, the removal of the need to deliberate. Practice can build both at once, but the reduced cognitive load of a habit is conceptually separate from the increased precision of a skill.
References
Dickinson, A. (1985). Actions and habits: The development of behavioural autonomy. Philosophical Transactions of the Royal Society B, 308(1135), 67–78. https://doi.org/10.1098/rstb.1985.0010
Everitt, B. J., & Robbins, T. W. (2005). Neural systems of reinforcement for drug addiction: From actions to habits to compulsion. Nature Neuroscience, 8(11), 1481–1489. https://doi.org/10.1038/nn1579
Everitt, B. J., & Robbins, T. W. (2016). Drug addiction: Updating actions to habits to compulsions ten years on. Annual Review of Psychology, 67, 23–50. https://doi.org/10.1146/annurev-psych-122414-033457
Gardner, B. (2015). A review and analysis of the use of 'habit' in understanding, predicting and influencing health-related behaviour. Health Psychology Review, 9(3), 277–295. https://doi.org/10.1080/17437199.2013.876238
Gillan, C. M., Morein-Zamir, S., Urcelay, G. P., Sule, A., Voon, V., Apergis-Schoute, A. M., Fineberg, N. A., Sahakian, B. J., & Robbins, T. W. (2014). Enhanced avoidance habits in obsessive-compulsive disorder. Biological Psychiatry, 75(8), 631–638. https://doi.org/10.1016/j.biopsych.2013.02.002
Graybiel, A. M. (2008). Habits, rituals, and the evaluative brain. Annual Review of Neuroscience, 31, 359–387. https://doi.org/10.1146/annurev.neuro.29.051605.112851
Haith, A. M., & Krakauer, J. W. (2018). The multiple effects of practice: Skill, habit and reduced cognitive load. Current Opinion in Behavioral Sciences, 20, 196–201. https://doi.org/10.1016/j.cobeha.2018.01.015
Knowlton, B. J., Mangels, J. A., & Squire, L. R. (1996). A neostriatal habit learning system in humans. Science, 273(5280), 1399–1402. https://doi.org/10.1126/science.273.5280.1399
Lally, P., van Jaarsveld, C. H. M., Potts, H. W. W., & Wardle, J. (2010). How are habits formed: Modelling habit formation in the real world. European Journal of Social Psychology, 40(6), 998–1009. https://doi.org/10.1002/ejsp.674
Neal, D. T., Wood, W., & Quinn, J. M. (2006). Habits—A repeat performance. Current Directions in Psychological Science, 15(4), 198–202. https://doi.org/10.1111/j.1467-8721.2006.00435.x
Perez, O. D., & Dickinson, A. (2020). A theory of actions and habits: The interaction of rate correlation and contiguity systems in free-operant behavior. Psychological Review, 127(6), 945–971. https://doi.org/10.1037/rev0000201
Robbins, T. W., & Costa, R. M. (2017). Habits. Current Biology, 27(22), R1200–R1206. https://doi.org/10.1016/j.cub.2017.09.060
Smith, K. S., & Graybiel, A. M. (2016). Habit formation. Dialogues in Clinical Neuroscience, 18(1), 33–43. https://doi.org/10.31887/dcns.2016.18.1/ksmith
Verplanken, B., & Orbell, S. (2003). Reflections on past behavior: A self-report index of habit strength. Journal of Applied Social Psychology, 33(6), 1313–1330. https://doi.org/10.1111/j.1559-1816.2003.tb01951.x
Watson, P., & de Wit, S. (2018). Current limits of experimental research into habits and future directions. Current Opinion in Behavioral Sciences, 20, 33–39. https://doi.org/10.1016/j.cobeha.2017.09.012
Wood, W., & Rünger, D. (2016). Psychology of habit. Annual Review of Psychology, 67, 289–314. https://doi.org/10.1146/annurev-psych-122414-033417