Superstitious Behavior: Skinner's Pigeons and Why Rituals Outlast the Evidence

Skinner's 1948 experiment

The procedure was almost nothing. Skinner reduced pigeons to about three-quarters of their free-feeding weight so that food would work as a reinforcer, put each bird alone in a chamber, and arranged for the food hopper to swing into reach for a few seconds once every fifteen seconds, with no reference to what the bird was doing. In the language of schedules this is a fixed-time schedule; unlike a fixed-interval schedule, it requires no response at all.^[1]

The process was not nothing. Six of the eight birds developed a clearly defined response that two observers agreed on. One turned counterclockwise around the cage, two or three turns between deliveries. Another repeatedly thrust its head into an upper corner. A third developed a "tossing" motion, as though lifting an invisible bar with its head. Two swung head and body from side to side like a pendulum, and one made incomplete pecking or brushing movements toward the floor. None had been seen before the food started arriving, and none had anything to do with producing it.^[1]

Skinner's explanation was that reinforcement acts on whatever behavior is in progress when it arrives. At the first delivery the bird was doing something; that something was strengthened, so it was more likely to be under way at the next delivery, when it was strengthened again. A feedback loop turned a random act into a ritual, and the shorter the interval, Skinner noted, the more readily the loop closed. His summary is the line quoted ever since: "The bird behaves as if there were a causal relation between its behavior and the presentation of food, although such a relation is lacking."^[1]

When Skinner lengthened the interval to a minute for one bird, its response became more energetic, a hop from foot to foot; when he then stopped the food altogether, the hopping went through an extinction curve of more than ten thousand responses before it faded, and a single reinstated delivery brought it back. He compared the birds to a bowler who keeps twisting his arm after the ball has left his hand.^[1] In Science and Human Behavior he generalized: a few accidental pairings between a ritual and a favorable result are enough to set the ritual up and carry it through many unrewarded repetitions.^[2]

Richard Herrnstein put the point in its strongest form in 1966. Superstition is not an anomaly; it is a corollary of the principles of operant conditioning. If a reinforcer strengthens whatever precedes it, it must sometimes strengthen behavior that had nothing to do with producing it, and the same process runs inside every ordinary contingency: when a rat's press is reinforced, its force, posture, and accompanying head-turn are strengthened along with the feature the contingency required.^[3] A click delivered a second late reinforces the spin after the sit, which is why marker timing matters so much in shaping and dog training.

The reexamination: what the pigeons were actually doing

Skinner watched his birds and described what he saw. In 1971 John Staddon and Virginia Simmelhag repeated the experiment with one change that altered its meaning: they catalogued every behavior each pigeon emitted, second by second, through the whole interval, on fixed-time schedules like Skinner's and on matched schedules where a peck was actually required. Two classes of behavior fell out of the records. Terminal responses occupied the last part of each interval, just before food was due, and in every bird they were much the same: pecking at the wall where the hopper appeared. Interim activities filled the early part of the interval, when food was least likely: the turning, pacing, wing-flapping, and preening that look so much like Skinner's rituals.^[4]

Neither class fit the accidental-reinforcement story. Terminal pecking emerged in birds that had not been pecking when earlier deliveries arrived, so chance pairings could not have selected it, and it looked the same whether or not a peck was required. Interim activities were almost never adjacent to food, yet they persisted. Periodic food, Staddon and Simmelhag concluded, induces a species-typical pattern: food-directed behavior when food is imminent, other activities when it is not. Behavior, they proposed, is generated by principles of variation, including such induced responses, and then selected by principles of reinforcement, an explicit analogy to evolution.^[4] Reinforcement still selects; it does not select from a blank slate.

William Timberlake and Gary Lucas went further in 1985, setting three explanations against one another: Skinner's chance contingency, Pavlovian stimulus substitution (the bird treating the chamber as if it were food, as in classical conditioning), and their own account in terms of appetitive behavior systems. Under periodic food their pigeons' behavior was strikingly uniform, almost entirely directed at the wall where food appeared, and resembled the pigeon's natural foraging and social repertoire rather than randomly selected acts or misdirected eating. The rituals, on this view, are pieces of an evolved feeding system that free food switches on.^[5]

A third study asked whether an animal can even tell that it caused an outcome. Peter Killeen trained pigeons to report whether a stimulus change had been produced by their own peck or by the apparatus, and they discriminated with high accuracy; the errors that remained moved with the payoffs for each answer. Superstition, Killeen argued, is a matter of bias rather than detectability: organisms can usually tell what they did from what merely happened, and what tilts them toward claiming credit is the relative cost of being wrong.^[6]

Superstitious behavior vs. its neighbors

Several things can happen when reinforcers arrive without depending on behavior, and they are easy to confuse.

The fourth row is the one most often forgotten. Noncontingent reinforcement does not generally build behavior; added on top of a real contingency, free reinforcers make the contingent response decline, because the behavior no longer makes much difference to how often the reinforcer arrives.^[7] Superstition is what accidental reinforcement does to behavior that had no contingency to begin with.

Superstition in the human laboratory

The cleanest human demonstration of accidental reinforcement is one of the earliest. Catania and Cutts gave college students two buttons: presses on one were reinforced on a variable-interval schedule, presses on the other never. Students went on pressing the useless button anyway, because a reinforcer earned by the first sometimes arrived just after a press on the second. A changeover delay, a required pause after switching buttons before any reinforcer could be collected, eliminated the superstitious pressing.^[9]

Koichi Ono's 1987 study is the human version of Skinner's. Twenty university students sat alone in a booth with three levers and a counter, told only that they need not do anything in particular, that doing something might produce points, and that they should try to get as many as possible. Points arrived on fixed-time and variable-time schedules regardless of what they did. Most developed some pattern of lever-pulling; a few developed elaborate, persistent rituals. The best-known is a woman who, after a point happened to follow a jump, climbed onto the table, jumped to touch the ceiling with her slipper, and went on jumping until she was too tired to continue. Superstition does develop in people under noncontingent reinforcement, Ono concluded, but not uniformly and often only for a while.^[10]

Children do it too. Wagner and Morris put preschoolers in a room with a mechanical clown that dispensed marbles on a fixed-time schedule, exchangeable afterward for a toy; seven of the twelve developed a recurring behavior between deliveries, such as touching or kissing the clown's face or swinging their bodies.^[11] A later computer-based study found superstitious responding whether the noncontingent event was the delivery of points or the removal of an aversive condition, so accidental negative reinforcement builds rituals as readily as accidental positive reinforcement.^[12]

That connects superstition to its apparent opposite. Human learned-helplessness experiments expose people to bursts of loud noise they cannot stop, then test whether they give up on a later, solvable task. Helena Matute noticed that the noise always stops eventually, so whatever the person did last is negatively reinforced by its ending. In her 1994 experiments, students told to try to stop the noise kept pressing keys in patterns, reported that they had found a way to control it, and showed no helplessness afterward; helplessness appeared only when the procedure led people to stop responding.^[8] A 1995 follow-up confirmed it: as long as people keep trying, uncontrollable outcomes tend to produce superstition and an illusion of control rather than passivity.^[13]

Superstitions in real life

Baseball

The anthropologist George Gmelch, a former minor-league first baseman, borrowed an observation of Bronislaw Malinowski's: Trobriand Islanders used elaborate magic for open-sea fishing, where catches were uncertain and danger real, and almost none for fishing in the lagoon, where results were reliable. Baseball follows the same rule. Rituals, taboos, and lucky objects cluster in hitting and pitching, where even an excellent player fails much of the time, and are nearly absent in fielding, where players handle the overwhelming majority of chances cleanly. The rituals sit exactly where skill accounts for the least of the outcome.^[14]

Lucky charms

Do the rituals do anything? A widely reported set of experiments by Damisch, Stoberock, and Mussweiler in 2010 said yes. Participants who putted with a golf ball described as lucky, who heard "fingers crossed" before a dexterity task, or who kept their own lucky charm with them during a memory game or an anagram task outperformed controls, and the effect ran through higher confidence in their own ability.^[15] In 2014 Calin-Jageman and Caldwell ran two direct replications of the lucky-ball putting experiment, with larger samples than the original, and found no benefit.^[16] The performance benefit of a lucky charm is unconfirmed.

The cultural angle

Skinner's account explains how an individual acquires a ritual from its own history of accidents. Stuart Vyse, whose Believing in Magic is the standard survey, points out that most human superstitions were not invented that way. Nobody discovers for themselves that thirteen is unlucky; the rituals come ready-made from parents, teammates, and the surrounding culture, and personal accidental reinforcement then keeps them going. Vyse also documents where superstition concentrates: among students before exams, athletes before competition, and gamblers, where outcomes matter and control is incomplete. It is not a sign of low intelligence or of any disorder, and part of what maintains it may be that the ritual reduces anxiety, which is real negative reinforcement even when the effect on the outcome is imaginary.^[17]

Why superstition is a rational-looking mistake

Nothing about a superstition requires stupidity. Four features of ordinary learning are enough to build one and hold it in place.

Reinforcement is blind to causation. A consequence strengthens the behavior it follows whether or not the behavior produced it; that is the whole content of Skinner's demonstration and Herrnstein's corollary.^[1][3] The mechanism that lets a rat find the lever is the one that attaches a ritual to a home run, and a learner that ignored every pairing it could not verify as causal would learn almost nothing.

Accidental reinforcement is intermittent, and intermittent reinforcement is durable. The good outcome follows the ritual sometimes and not other times, which puts the ritual on a lean, irregular schedule from the start. Behavior maintained that way is far more resistant to extinction than behavior reinforced every time, the partial reinforcement extinction effect described on the page comparing continuous and intermittent reinforcement. Skinner's pigeon, still hopping ten thousand responses after the food stopped, is the laboratory version of a fan in the same shirt through eleven losing seasons.^[1]

Extinction is never run. To extinguish a superstition you would have to skip the ritual, repeatedly, and watch the outcomes; but the ritual is cheap and the outcome feels important, so the test is never made. A contingency is a comparison between two rates, good outcomes with the ritual and without it, and someone who never skips the ritual has only one of the numbers. Killeen's pigeons show this is a bias set by payoffs rather than a failure of perception: when a wrong "I didn't cause it" costs more than a wrong "I did," organisms claim the credit.^[6]

Superstition grows where control runs out. Gmelch's ballplayers were not superstitious about fielding. Rituals attach to outcomes that matter and cannot be fully controlled, which is where accidental pairings are most frequent and real contingencies hardest to find.^[14][17] Matute adds the last piece: in that situation, continuing to act produces superstition and stopping produces helplessness, and of the two, superstition is arguably the better deal.^[8][13]

How to test your own superstitions

The pigeon could not run the test. You can. The question is the one behind any contingency: does the outcome depend on the behavior, or does it just sometimes follow it? Catania and Cutts' students stopped pressing the useless button as soon as the accidental pairings were broken, and you can break them deliberately.^[9]

1. Name the ritual and the outcome. Be specific on both sides: "wearing the gray hoodie" and "scoring above 80 on the quiz," not "feeling ready." If you cannot say what outcome the ritual is supposed to produce, it is probably maintained by how it feels rather than by what it does.
2. Record a baseline. For a dozen occasions, note whether you performed the ritual and how the outcome went. Most people find they have never once skipped the ritual before an outcome that mattered, so they have no evidence either way.
3. Withhold the ritual on alternate occasions. Decide in advance, ideally by coin flip, which occasions get the ritual, so that you are not quietly saving it for the hard days. This is extinction arranged as an experiment. It will feel wrong at first; the discomfort is the negative reinforcement that has been helping to maintain the ritual, and it fades.
4. Compare the two rates. Count good outcomes with the ritual and without it. A contingency exists only if the first rate is clearly higher. One bad day without the hoodie proves nothing; you have had bad days with it.
5. Decide what to keep. If the outcome does not depend on the ritual, drop it, or keep it with clear eyes: a routine that steadies you may be worth its thirty seconds, provided it does not stand in for preparation and you can do without it when the hoodie is in the wash.

What the evidence does not show

• It does not show that Skinner's account is the whole story. The 1948 paper was a demonstration, not a controlled experiment: eight birds, no comparison group, behavior described by observers rather than recorded through the interval, and the measured response chosen after it appeared. Recorded continuously, most of the behavior turned out to be induced, species-typical activity organized by the schedule.^[1][4][5] Accidental reinforcement is real, but its clearest demonstrations are the experiments where a real contingency runs alongside.^[3][9]
• It does not show that noncontingent reinforcement generally builds behavior. Added to an existing contingency, free reinforcers weaken the contingent response.^[7] Even with no competing contingency the human results are uneven: Ono's students mostly developed something, but few developed anything lasting.^[10]
• It does not show that superstitious organisms cannot detect contingency. Killeen's pigeons could. The bias toward claiming credit is a decision under uncertainty, and it moves with the payoffs.^[6]
• It does not show that lucky charms improve performance. The 2010 result did not survive direct replication.^[15][16]
• It does not show that everyday superstitions arise from personal accidents. Most are learned from other people and then maintained by accidental reinforcement, anxiety relief, and social approval; the pigeon model explains the maintenance better than the origin.^[17]
• It does not show that superstition and helplessness are opposite kinds of people. In Matute's experiments they were the same people under different instructions; the difference was whether they were still responding when the uncontrollable outcome arrived.^[8][13]

Key takeaways

• Superstitious behavior is behavior maintained by accidental reinforcement, a reinforcer that happened to follow the behavior without depending on it. Skinner produced it in 1948 by feeding pigeons every fifteen seconds regardless of what they did, and six of eight developed a ritual.
• Skinner explained the rituals as reinforcement of whatever the bird happened to be doing when food arrived, and Herrnstein showed that the same process strengthens the incidental features of every reinforced response. Staddon and Simmelhag then found that the pigeons' behavior was mostly species-typical food anticipation organized by the schedule.
• People develop rituals under noncontingent points, marbles, and noise that stops on its own, and when they keep trying to control an uncontrollable outcome they become superstitious rather than helpless. Real-world superstitions cluster where outcomes matter and control is incomplete.
• Superstitions persist because accidental reinforcement is intermittent, the extinction test is never run, and claiming credit is a cheap bet. To test one, record a baseline, withhold the ritual on randomly chosen occasions, and compare the two rates.
• The lucky-charm performance benefit did not replicate, most human superstitions are learned socially rather than built from personal accidents, and the 1948 study was a demonstration rather than a controlled experiment.

Check yourself

A dog spins in a circle before every meal. Its owner puts the bowl down at the same time each evening, whatever the dog is doing. Is the spinning superstitious behavior in Skinner's sense, and how could you tell?

It could be. The bowl arrives on a fixed-time schedule, so whatever the dog was doing just before it, including a spin that once happened by chance, is accidentally reinforced and becomes more likely to be in progress at the next delivery. But Staddon and Simmelhag would ask whether the spin is species-typical excitement induced by food coming due. The test is contingency: put the bowl down at varied times, unrelated to spinning, and see whether the spin stays tied to the moment of delivery or drifts.

A student wears the same gray hoodie to every exam and has done well in most of them. She says the evidence supports the hoodie. What is missing from her evidence?

The other rate. A contingency is a comparison between the outcome rate with the behavior and the rate without it, and she has never taken an exam without the hoodie. Her results are consistent with the hoodie mattering and with it mattering not at all; only occasions without the ritual, chosen in advance rather than saved for easy exams, can tell those apart. Meanwhile the hoodie is cheap and skipping it feels risky, which is exactly Killeen's bias.

In a replication of Skinner's study, every pigeon pecks at the wall near the hopper just before each food delivery. Is this the accidental reinforcement Skinner described?

No. Accidental reinforcement should produce different rituals in different birds, selected from whatever each happened to be doing. A response that is the same in every bird, appears in birds that were not pecking when earlier deliveries arrived, and is directed at the food source is a terminal response induced by the schedule: species-typical food anticipation, not chance selection.

Want more? The 20-question quiz covers every page on the site with instant explanations.