Phasing out the treat: the transition to intermittent reinforcement (and the jackpot)

When a cat responds to a cue 90 percent of the time across short sessions, it is time to lower the treat frequency. The common mistake is pulling the treat all at once: the behavior extinguishes in a week or two and you have to retrain. The correct transition is gradual. You move from continuous reinforcement (1:1, one treat per response) to fixed reinforcement at 1:2, then 1:3, then to variable reinforcement (the average stays 1:3, but the cat cannot predict whether the next correct response pays out). Variable reinforcement produces the behavior most resistant to extinction, the operant principle B. F. Skinner described in the 1950s. The clicker still fires on every correct response; what shrinks is the treat, not the marker.

The woman named Karen Pryor

In 1963, a marine biologist named Karen Pryor started training dolphins at Sea Life Park in Hawaii without coercion. She applied Skinner's principles of operant conditioning to the dolphin: she marked the exact behavior she wanted with a whistle, the equivalent of the clicker in cat work, and delivered a herring right after. The method worked so well that Pryor carried it over to dogs, horses, cats, and, as she wrote in Don't Shoot the Dog (1984, revised 2002), to children and partners too. The book sold millions of copies and opened positive training to a general audience.

Pryor's most interesting contribution was not the clicker itself; Skinner had already described it as a "secondary marker" in the 1950s. It was popularizing the idea that a behavior holds up better on unpredictable rewards than on guaranteed ones. Common intuition says the opposite: "if I want the cat to sit every time, I should reward it every time." That turns out to be false. A cat rewarded every time, then suddenly given no reward, stops sitting within a few attempts. A cat rewarded on a variable schedule keeps sitting for weeks even when some attempts pay nothing.

This article is the mechanics of that transition, applied to cat training.

The operant principle in one page

Skinner sorted reinforcement schedules into four broad types along two axes:

1. Continuous reinforcement (CRF). Every correct response is rewarded. This is the learning phase. It builds the behavior fast but leaves it fragile: when the reward disappears, the behavior extinguishes quickly.
2. Fixed-ratio intermittent reinforcement (FR). Every Nth correct response is rewarded (1 in 2, 1 in 3). It makes the behavior more durable than continuous does.
3. Variable-ratio intermittent reinforcement (VR). The reward arrives after a variable number of responses, around an average (1:3 on average: sometimes on the second, sometimes on the fifth, sometimes on the first). It produces the behavior most resistant to extinction of any schedule. It is the same principle that keeps a gambler feeding coins into a slot machine: you do not know when it will hit, but you know it will.
4. Interval reinforcement (fixed or variable). The reward arrives after a period of time, not after a count of responses. Less useful in training, more common in motivation studies.

To maintain a behavior that is already consolidated in a cat, the goal is to reach variable reinforcement. The path runs through stepped fixed reinforcement first.

When to start the transition

The practical rule is the 90 percent rule: when the cat responds correctly to the cue nine times out of ten across three consecutive sessions, the behavior is consolidated and you can begin the transition. If it responds six out of ten, the behavior is not yet stable, and pulling the treat at that point extinguishes the little you have built.

An easy way to measure it: log each session in a notebook or on your phone. A session of ten "sit" attempts where the cat sits 9 times and misses once is logged "9/10." Three straight sessions of 9/10 or 10/10 are the signal to start stepping down.

The transition step by step, using "sit" as the example

Say you have spent three weeks training "sit" with the clicker. Your cat, Luna, responds to the verbal cue "sit" 9 of 10 times. You are going to shift to variable reinforcement over four weeks.

Week 1: fixed reinforcement 1:2 (reward on one of every two responses)

Daily sessions of one minute, ten attempts. You say "sit." Luna sits. Click. Treat. You say "sit." Luna sits. Click. No treat. A small verbal cue ("good girl") and you continue. You say "sit." Luna sits. Click. Treat. And so on.

The point: the click always fires on a correct response. The click is the marker, not the reward. What gets cut in half is the treat that follows. The cat still knows it did well (from the click) even when no treat arrives on that particular attempt.

After three or four days, the cat no longer shows surprise when a click is not followed by a treat. It keeps responding to the cue at the same rate.

Week 2: fixed reinforcement 1:3 (reward on one of every three responses)

Same session, same format. Now you reward only every third correct response. Click-treat, click-no treat, click-no treat, click-treat, click-no treat, click-no treat, click-treat.

If the cat slips (drops to 7/10), back off to 1:2 for two sessions and try again. The curve is not always linear.

Week 3: variable reinforcement, average 1:3

Here is the qualitative shift. You no longer reward on an exact count of three. You reward unpredictably, with an average of one in three. The sequences might run: treat, none, none, treat, none, treat, none, none, none, treat. Or: none, treat, none, none, none, none, treat, none, treat, none. The average stays 1:3, but the cat cannot predict when it pays.

This is where what Pryor called the "expectation of gain" kicks in: the cat keeps responding because it knows the treat could come at any moment. The behavior turns robust.

Week 4: variable reinforcement, average 1:5

You raise the average to one in five. It stays variable. The first sessions may dip to 80 percent; if it drops below 70 percent, back off to 1:3 for a week. If it holds at 80 to 90 percent, keep going.

From here, the behavior is consolidated on a variable schedule. You can climb to 1:10 over time, but past a certain dilution (beyond 1:15 or 1:20 in a cat) the behavior starts to weaken. The realistic move is to stabilize between 1:5 and 1:8 for the long term.

The jackpot, the other side of the coin

The jackpot is the opposite of intermittent reinforcement, used occasionally to reward an important breakthrough especially well. When the cat does something new, very hard, or particularly clean, you do not give one treat: you give five or six in a row, slowly, while you praise it. It is a "shower of treats" that stamps into the cat's memory that this specific behavior is worth a lot.

When to use it: the first time the cat walks into the carrier on its own, the first time it lets a nail get trimmed without pulling the paw back, the first time it holds still on the vet's scale. The jackpot is used sparingly, not in every session. Overuse it and it stops being a jackpot and becomes continuous reinforcement at high cost.

The jackpot and intermittent reinforcement are not in conflict. They coexist in the same training phase. The cat responds almost every time on a variable schedule, and when it does something exceptional, jackpot. That combination is what good trainers of dolphins, service dogs, and any animal working for positive reinforcement maintain over the long term.

Why most people get it wrong

Mistake 1 is pulling the treat all at once. "The cat already knows how to sit, no need to reward it now." Result: in two weeks the cat stops sitting, because it has learned the cue no longer predicts anything interesting. Retraining takes about three times as long as training the first time, the effect Pryor described as the "learned extinction" phenomenon.

Mistake 2 is dropping the click instead of the treat. If you stop clicking because "it is no longer needed," you lose the marker and the cat stops knowing exactly what you liked. The click is free and stays for life. The treat is what gets reduced to a variable frequency.

Mistake 3 is rewarding with no pattern, "whenever I feel like it." There is no statistical pattern, only pure arbitrariness. The cat detects that the reward depends on your mood, not on its response, and the behavior degrades. Variable reinforcement is planned: average 1:3, random distribution but with that average respected across the session.

Mistake 4 is keeping no record. Without a notebook or a note on the phone, the first two weeks you believe the cat is responding well, then fail to notice it has dropped to 70 percent. Logging correct responses per session is the only way to know whether you can advance a phase or need to step back.

Natural use outside formal sessions

Once the behavior is on a variable schedule, you can cue the cat outside the formal sessions. "Sit" when you are about to put down food, "come" when you walk in the door. Sometimes a treat follows, sometimes a scratch, sometimes only the verbal click. That generalization is what keeps the behavior alive for years, without depending on the physical clicker or the treat jar.

Specific common mistakes

Rushing the transition. Jumping from 1:1 to 1:5 in a week. The cat does not process the leap. Step down gradually, week by week.

Confusing intermittent with forgetful. Rewarding when you remember rather than when it is scheduled. Minimum structure: plan before the session how many responses you will reward and which ones.

Rewarding the slower response. If you hesitate between rewarding or not, reward the faster or cleaner response. Rewarding the slow response reinforces slowness.

Keeping continuous reinforcement for life. It works, but it leaves you dependent on a treat in your pocket forever. A well-trained cat responds to the cue even with no treat in sight.

Dropping the clicker too. The clicker is the marker, not the reward. It stays for whenever you ask for something new or want to reinforce a known behavior. What you vary is the treat that follows.

What to verify

1. The cat responds 9 of 10 times to the cue before you start the transition.
2. A written log of correct responses per session (10 attempts, noted X/10).
3. A weekly phased step-down: 1:1 to fixed 1:2 to fixed 1:3 to variable 1:3 to variable 1:5.
4. The click maintained on every correct response (it is the marker, not the reward).
5. The jackpot reserved for important breakthroughs, not for every session.
6. If the response drops below 70 percent, step back a phase rather than pushing on blind.
7. Generalization outside formal sessions once the variable phase is consolidated.

This protocol applies directly to maintaining name recall and to tolerating nail trims, two behaviors that extinguish within a few months without intermittent reinforcement. The AAFP notes that positive reinforcement, applied consistently, is what keeps trained responses reliable over time.