The Token Economy: Definition, History, Components, Evidence, and How to Build One

How a token economy works

Reinforcement works best when the consequence arrives within seconds, and most consequences people care about — a day pass off a ward, a paycheck — cannot be delivered that fast. A token solves the timing problem. It can be handed over the instant the behavior occurs, and it carries the value of whatever it will later buy, the way a clicker bridges the gap between a dog's sit and the treat.^[1]

What gives a token its value is its history. A poker chip is nothing to a chimpanzee until it has gone into a vending machine and produced a grape. Paired with one backup reinforcer, the token becomes a conditioned reinforcer; paired with many, it becomes a generalized conditioned reinforcer, and that is what makes a token economy work. A single reinforcer satiates; a token that buys candy, or free time, or a phone call home, is worth something in almost any state the person is in. Skinner made the same point about money.^[2] Primary and secondary reinforcers ›

Behavior analysts describe a token economy as three interlocking schedules of reinforcement.^[3] The token-production schedule says how much behavior earns a token: one point per completed problem. The exchange-production schedule says when exchange becomes available: at the end of the lesson, after twenty tokens. The token-exchange schedule is the price list.

The components of a token economy

Every token economy has the same parts. Ivy and colleagues found that many published programs leave one or more unspecified — most often the schedules of token production and exchange, and how backup reinforcers were identified — which makes them hard to replicate.^[4]

The fine runs in the opposite direction. Removing tokens contingent on a behavior is negative punishment, which behavior analysts call response cost. Token loss reliably suppresses behavior in the laboratory,^[3] and Achievement Place used fines throughout.^[5] But a fine only works while the person has tokens to lose.

History: from poker chips to hospital wards

Chimpanzees and poker chips

The first token economies had no name and no human subjects. In the 1930s John Wolfe taught chimpanzees to insert poker chips into a vending machine, the "Chimp-o-mat," that delivered a grape per chip, and then to earn chips by lifting a weighted handle. The animals worked for chips much as for grapes, chose chips that bought two grapes over chips that bought one, ignored brass slugs that bought nothing, and kept working when the machine was unavailable, though less the longer the exchange was delayed.^[6] John Cowles then showed that chips could teach as well as maintain: chimpanzees learned new discriminations when the only immediate consequence of a correct choice was a token to be spent later, and performed almost as well as for food itself.^[7]

The ward at Anna State Hospital

Teodoro Ayllon and Nathan Azrin built the first systematic token economy for people in the early 1960s, on a ward of Anna State Hospital in Illinois whose patients were women, most hospitalized for years with diagnoses of schizophrenia. Patients earned tokens for self-care and ward jobs — serving meals, cleaning, laundry — and spent them on privacy, walks on the grounds, trips into town, time with staff, and commissary items.^[8] Six experiments tested whether the tokens were doing the work. When the tokens were moved from the job each patient preferred to the one she had avoided, the patients moved with them; when tokens were handed out daily regardless of work, the ward's total labor fell from about 45 hours a day to about one, and recovered when the contingency was restored.^[8] Their 1968 book laid out the method as rules, two of which every later program has had to relearn: let people sample a backup reinforcer before asking them to work for it, and reinforce only behaviors that will go on being reinforced once the program ends.^[1]

Achievement Place and the social-learning program

In 1967 a house opened in Lawrence, Kansas, for boys the courts had labeled "pre-delinquent," run by a married couple as teaching-parents. Points were earned for chores, homework, punctuality, and appropriate speech; lost for aggressive statements, poor grammar, and lateness; and exchanged for the following week's privileges — allowance, snacks, television, permission to go downtown. Elery Phillips's 1968 report used reversal designs to show that each behavior rose or fell with the points attached to it.^[5] The Teaching-Family Model that grew from it was replicated widely, and its follow-up evaluation, discussed below, is the clearest demonstration of the token economy's limits.

The most rigorous test remains Gordon Paul and Robert Lentz's multi-year comparison of three treatments for long-term psychiatric inpatients: a social-learning program built around a token economy and skills training, milieu therapy, and standard hospital care. The social-learning program produced the largest improvements in functioning and the most successful community discharges, at the lowest cost.^[9] Classrooms adopted tokens almost as soon as the ward did; by 1972 the published programs spanned schools, wards, prisons, and homes.^[10] Tokens and the Good Behavior Game in the classroom ›

What the evidence shows

The applied reviews

Kazdin and Bootzin's 1972 review, and Kazdin's follow-up a decade later, judged the effect of token economies while they were running to be well established across psychiatric wards, institutions for people with intellectual disabilities, delinquency programs, and classrooms. They also named the problems that have not gone away: behavior fell when tokens were withdrawn, did not transfer to settings without tokens, some individuals never responded, and staff often stopped delivering tokens consistently.^[10][11] By 1982 the psychiatric programs were in decline, as deinstitutionalization emptied the wards and court rulings made meals, beds, and ground privileges rights rather than things to be earned.^[11]

Two recent reviews of the classroom literature reach a mixed verdict. Maggin and colleagues applied the What Works Clearinghouse standards to studies with students with challenging behavior and found the effects generally positive but the studies too weak in design and reporting to qualify as evidence-based under those standards.^[12] Soares and colleagues' meta-analysis of single-case classroom studies found positive effects in most cases, varying with the students and how the systems were built.^[13] The gap in the record is not evidence that the programs failed; it is evidence that too many were never described well enough to be copied.^[4]

The laboratory evidence

Timothy Hackenberg's reviews of token reinforcement in the laboratory — pigeons earning lights, rats earning marbles, chimpanzees earning chips — establish what a practitioner needs to know.^[3][14] Tokens function as reinforcers: behavior that produces them increases, and behavior that costs them decreases. Their power depends on the exchange schedule: responding is weakest when exchange is far away and rises as it approaches, and tokens that cannot be exchanged lose their effect. Tokens also serve as discriminative stimuli marking distance from exchange, which is one reason a visible token board works where an invisible tally does not. And animals given the choice will often accumulate tokens before exchanging them, the laboratory version of saving.^[3]

Where token economies are used

The clinical case with the strongest trial evidence is contingency management: people in treatment for substance use earn vouchers or prize draws for drug-negative urine samples, with the value escalating across consecutive negatives and resetting after a positive — a token economy with a laboratory test as the target behavior. A meta-analysis of the controlled trials found moderate, reliable effects, strongest for stimulants and opioids, that shrink after the incentives end.^[15] All applications, with the evidence rated ›

The last row is the one most people live in. A stamp card, an airline's miles, and a language app's streak all have the token economy's structure, and whether they change behavior depends on the same variables: whether the token arrives immediately, buys something the person wants, and can be spent soon enough to matter.^[14] Sticker charts and token systems at home ›

How to set up a token economy

The steps follow Ayllon and Azrin's rules and the maintenance strategies Kazdin's reviews recommended.^[1][10]

1. Choose three to five target behaviors, stated as things to do. "Starts the worksheet within a minute," not "stays focused." Prefer behaviors the world will eventually reinforce on its own — Ayllon and Azrin's relevance-of-behavior rule — because those will survive the program.
2. Pick a token you can deliver in a second. A chip, a tally, a sticker. For young children, keep the token tangible and the board in view.
3. Build the menu from observation, not guesses. Watch what the person does when free to choose — the Premack principle in use — and let people sample items they have never had. Price it so a single good hour buys something, and refresh it, because backup reinforcers satiate.
4. Deliver immediately, with specific praise every time. "You started on your own — one point." The praise inherits the token's power and keeps working when the tokens are gone.
5. Exchange early and often at first. Same day, or same lesson for young children; laboratory tokens far from exchange are weak ones.^[3] Lengthen the interval only after the behavior is steady.
6. Use fines rarely, if at all. Start without them. If you add response cost, define the fined behaviors in advance, keep each fine small relative to a day's earnings, and never let a balance go below zero.
7. Count before you start, review weekly, and plan the fade from day one. A baseline is the only way to know whether the tokens did anything. Then thin the token-production schedule, space the exchanges, raise prices, shift from tokens to praise to natural consequences, and hand the counting to the earner.

Common mistakes

• Tokens with no backup value. A chart with nothing to buy, or a menu of things the earner never chose, is a list of tallies; tokens that cannot be exchanged stop working.^[3]
• Exchange delayed too long. A weekly prize for a six-year-old asks the child to work for something five days away, and behavior early in a long exchange cycle is the weakest in the system.^[3]
• Fines that exceed earnings. A person at zero has nothing to work for and nothing to lose. When fines dominate, the token economy has become a punishment procedure with a bookkeeping layer, and the rational response is to stop participating.
• Never fading. If the system is identical in June and September, the behavior belongs to the system and disappears with the tokens.^[10]
• Inconsistent delivery. Staff who stop handing out tokens are a failure mode Kazdin's reviews found repeatedly.^[10][11] A system that depends on adults remembering needs prompts for the adults.
• Paying for behavior that was already happening for its own sake. Tokens are for building behavior that is not occurring; see the next section.

The intrinsic-motivation debate

The standing objection to any token system is that paying people for a behavior makes them stop doing it for its own sake. The evidence is a long-running dispute between two meta-analyses of largely the same experiments.

Deci, Koestner, and Ryan's analysis of 128 experiments found that expected, tangible rewards for an activity people already found interesting reduced their later free-choice engagement with it and their self-reported interest, most strongly when the reward was given merely for engaging in or completing the task. Verbal rewards — praise — increased intrinsic motivation, and unexpected rewards did not undermine it.^[16] Cameron and Pierce, five years earlier, concluded that rewards do not, on the whole, reduce intrinsic motivation. The one reliable negative effect they found was the same one: expected tangible rewards for simply doing an interesting task, regardless of how well, reduced free-choice time afterward. Rewards tied to performance did not.^[17]

Read side by side, the two agree on the boundary of the effect and disagree on how much it matters: on how large and general it is, which studies belong in the analysis, and which outcome measures count. For a token economy the practical conclusions follow from either reading. Use tokens for behavior that is not happening, not for activities the person already does for pleasure. Pay for quality or completion rather than mere engagement, pair every token with specific praise, and fade the tokens as the behavior meets the consequences the world already provides.

What the evidence does not show

It does not show that gains survive the program. This has been the weak point since the first review.^[10][11] The clearest case is Achievement Place. When Teaching-Family group homes were compared with other community homes for juvenile offenders, the youths in Teaching-Family homes had fewer recorded offenses during treatment; in the year after they left, the difference was gone.^[18] Contingency management shows the same shape: effects that shrink after the vouchers stop.^[15] Behavior follows the contingencies in force; when the tokens end and nothing replaces them, the behavior is on extinction. Maintenance has to be engineered: by fading, by pairing tokens with praise, by training where the behavior must eventually occur, and by choosing target behaviors the natural environment will pay for.^[1][10]

It does not show that tokens teach skills. Ayllon and Azrin subtitled their book "a motivational system," and that is what a token economy is. It increases behavior already in the repertoire; a student who cannot do the problems will not do them for points. Skills need shaping and instruction; the token economy pays for the practice.

It does not show large effects in ordinary classrooms. Most of the classroom evidence comes from single-case studies of students with challenging behavior, and the best of it is methodologically modest.^[12][13] The strongest controlled trial concerns long-term psychiatric inpatients in a hospital system that no longer exists in that form.^[9] And some people do not respond to any menu on offer; the reviews have said so from the beginning.^[10] What the evidence does show is narrower and still useful: while a well-built token economy is running, target behaviors rise, and the components that make it work are the ones the chimpanzees demonstrated in 1936.

Key takeaways

• A token economy delivers tokens immediately after target behaviors and lets the earner exchange them later for backup reinforcers. The token works because it is a generalized conditioned reinforcer, like money.
• The components are target behaviors, tokens, backup reinforcers, an exchange rate, a schedule of exchange, and optionally response cost. Most failures trace to a menu nobody chose, an exchange too far away, fines that exceed earnings, or a fade that never happened.
• The lineage runs from Wolfe and Cowles's chimpanzees working for poker chips, through Ayllon and Azrin's ward and Achievement Place, to Paul and Lentz's social-learning program, still the most rigorous test.
• Expected tangible rewards for an already interesting activity can reduce later free-choice engagement; praise does not. Build token economies for behavior that is not happening, pair tokens with praise, and fade them.
• Gains do not survive the program by default. Maintenance has to be engineered: fade the tokens, pair them with praise, train where the behavior must occur, and choose behaviors the world will go on reinforcing.

Check yourself

A teacher runs a point system all year. Points are tallied on a chart the students cannot see and exchanged for prizes on the last Friday of each month. On-task behavior has not changed. Which components are most likely at fault?

The schedule of exchange and the visibility of the token. A month between exchanges puts most of the behavior far from the reinforcer, where laboratory tokens are weakest, and a tally the students cannot see gives them no signal of how close they are. Move exchange to the same day, put the tally where the student can watch it, and check that the prizes are things the students actually choose when free.

A group home fines residents heavily for rule violations. Several residents are in debt and have stopped doing chores at all. What has gone wrong?

Fines have exceeded earnings, so the token economy has become a punishment procedure. A resident below zero has nothing to lose and nothing to work toward, and the rational response is to stop participating. Fines should be rare, small relative to a day's earnings, and never allowed to take a balance below zero; the system should run on earning, not on loss.

A residential program for adolescents shows excellent behavior while youths are enrolled, but a year after discharge they are no different from youths who went elsewhere. Does this mean the program did not work?

It means the program changed behavior while its contingencies were in force and did not arrange for anything to replace them, which is what the Achievement Place evaluation found. A token economy is a set of contingencies; when they end without a fade, without pairing to praise, and without training in the settings where the behavior must occur, the behavior is on extinction. The procedure worked; maintenance was never engineered.

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