Learning Science

A reward is what you meant to give. Reinforcement is what actually happened.

You only know a treat was reinforcing if the behavior gets more likely afterward. If you hand your dog a biscuit every time they sit and sitting does not increase, you rewarded them, but you did not reinforce the sit. Reinforcement is defined by its effect on future behavior, never by how much the dog appeared to enjoy it, and never by your intention.

Start here, because this one sentence decides how you read everything below it. It is also the sentence that dismantles "my shock collar is just information, he does not even mind it." If a dog's apparent liking decided the label, anyone could relabel anything. Effect decides. Something that reduces a behavior is punishment regardless of what it is called and what the dog seems to feel about it.

In Push Drop Stick

This is what the pass and fail tally is for. You are not grading your dog, you are measuring whether what you did last set is working. Rising pass rates mean you found a reinforcer. Flat or falling ones mean you have not yet, and it is usually the treat, the rate, or the size of the step.

Often confused with

Bribery. Not a technical term, and mostly used as an accusation. The distinction that matters is timing: food that appears before the behavior is a lure or prompt, and food that appears after it is reinforcement. Luring is a legitimate way to start a new behavior, and it gets faded fast. What people are right to worry about is the version where food has to be visible before your dog will respond, because then the food has become part of the cue.

How your dog learns what works. Behaviors that reliably produce a good consequence happen more often. Behaviors that produce a bad consequence, or that stop producing a good one, happen less.

Operant learning is about voluntary behavior and its consequences: sitting, coming when called, counter surfing, barking at the window. It is organized into four quadrants, and the names are less confusing than they look once you know they are two words glued together:

• Positive means something was added. Negative means something was taken away. Neither word means good or bad.
• Reinforcement means the behavior went up. Punishment means it went down.

That gives you the four entries that follow. Two things are worth knowing before you read them.

First, the quadrants are a description of how learning works, not a menu a trainer should order from. Knowing all four exist does not mean a good trainer draws on all four, any more than knowing about infection means a surgeon should cause one. We train in one of them.

Second, your dog is learning operantly all day whether or not you are training, and the environment reinforces plenty of things you would rather it did not. When a behavior will not go away, the useful first question is not how do I stop it, but what is paying for it.

In Push Drop Stick

Prompting, Shaping, Capturing and Errorless steps are all operant: your dog does something, and you pay it. Errorless steps are operant too. What makes them different is not the learning, it is the setup: the antecedent is arranged so the right answer is very nearly the only one available, which is why the first step of Leave It is built that way. Classical steps are the only ones that are not operant.

Often confused with

Classical conditioning, which is about involuntary emotional responses rather than chosen behavior. Most real training involves both at once. Trainers put it as: Pavlov is always on your shoulder. Whatever you teach your dog to do, they are also learning how to feel about it, about you, and about the place you did it in.

You add something your dog wants right after a behavior, and the behavior becomes more likely. Something is added, and the behavior increases.

This is the quadrant we train in, and the only one you need.

Both halves of the name are required, and the second half is a measurement rather than a plan: if the behavior does not increase, no reinforcement happened, however generous you felt. Plenty of things humans assume dogs love, a pat on the head, a cheerful "good boy", a hug, reliably fail to reinforce, and some actively suppress behavior. Watch the data, not your intentions. See Reinforcement vs Reward.

Common reinforcers: food, play, toys, sniffing, access to a room, permission to greet a friend, being let off leash. Food is the workhorse because it is fast, small, repeatable and easy to rank by value, which lets you match payment to difficulty.

In Push Drop Stick

Every single pass. The app's whole job is to keep the behavior easy enough that your dog earns reinforcement often, and to notice, from your honest pass and fail marks, when it has stopped being easy enough.

Often confused with

Permissiveness. Positive reinforcement is not the absence of rules, it is how the rules get taught. The alternative to punishing an unwanted behavior is teaching and paying a better one, plus management so the unwanted one gets no practice. See Differential Reinforcement.

Something your dog wants is taken away after a behavior, and the behavior becomes less likely. Something is removed, and the behavior decreases.

Examples: ending a game when the puppy's teeth land on skin, stepping away when your dog jumps up, briefly closing a door between you and your dog.

This is the one quadrant outside positive reinforcement that a thoughtful trainer will occasionally reach for, so it deserves a straight answer rather than a wave of the hand. It still depends on your dog losing something they want, which produces frustration, and frustration is a common reason a behavior gets bigger before it gets smaller. It also requires that your dog have something to lose in the first place, so it offers nothing at all for a dog who is already worried.

Our position: it is never the right tool for a behavior driven by fear or anxiety. Removing attention from a dog who barks at visitors because visitors frighten them adds a second problem to the first.

Where it is reasonable, it works for a specific reason: it removes the exact thing that was paying for the behavior. Hard puppy teeth end the game. A dog who steals shoes to be chased loses the chase, and the audience, for a few minutes. So the best sign that a time-out will work is that you can name what the behavior is earning and the time-out takes precisely that away. Done well it is brief, calm, boring and predictable, delivered without scolding or dragging, and always paired with two other things: teaching and paying what you want instead, and a legal outlet for the same fun, such as a real game of keep-away with a toy your dog is allowed to win.

The version we prefer is a touchless time-out. Rather than moving your dog, you remove yourself, or the thing they want. Walk back out the door when your dog jumps at you in the hallway. Stand up and leave the room when puppy teeth land on skin. Let the tug toy go dead and end the game when your dog will not drop it. Close the curtain on the window. Nothing is grabbed, nothing is dragged, no collar is taken hold of, and there is no moment where a frustrated dog has to be handled, which is exactly the moment these things go wrong. If a time-out is called for, start here.

Two practical notes. Repeated time-outs for the same behavior are information rather than failure: usually the situation is too hard, or the alternative behavior is not being paid well enough. But if you are going to use one, one slightly longer, calm time-out often does more than a string of very short ones your dog has learned to sit out. And if you do have to move your dog somewhere, choose the place with care. The whole value of a time-out is that it is calm and boring, so it has to happen somewhere your dog is genuinely relaxed. A crate is fine if, and only if, your dog is properly crate trained and content in there. Well crate-trained dogs will often wander back in for a nap once the time-out is over, which tells you the crate is still a good place. If your dog is not crate trained, do not use the crate. And never use anywhere your dog finds frightening: a time-out somewhere scary is not negative punishment at all, it is positive punishment wearing a kinder name. A boring, safe, familiar room is the reliable choice.

One thing that is not negative punishment: taking something out of your dog's mouth. Removing a chew or a stolen item by hand is not a time-out, it is a good way to build guarding, and it is how people get bitten. Trade instead.

Worth knowing where this entry comes from: there is very little dog-specific research on time-out procedures. What exists is decades of behavior-analytic work in other species plus clinical experience, so read this one as a considered position rather than a finding.

In Push Drop Stick

No plan in the app uses it, and nothing in the app takes anything away from your dog. Ending a session early because your dog is tired is not a time-out, it is good practice. It is still a fair thing to talk through with your coach when a real household problem calls for it.

Often confused with

Negative reinforcement. Both remove something. Here the removal follows the behavior you want less of, and behavior goes down.

Something your dog dislikes is added after a behavior, and the behavior becomes less likely. Something is added, and the behavior decreases.

As with reinforcement, the decrease is the definition, not the intent and not the severity. A raised voice that actually reduces the behavior is positive punishment. A shock that does not reduce it is not punishment at all, just pain.

Examples include leash corrections, yelling, a squirt bottle, a thrown object, and the stimulation from an electric collar.

We never train with positive punishment. It can suppress a behavior in the moment, which is exactly what makes it feel like it worked, but it teaches your dog nothing about what to do instead, and the evidence on what it costs is consistent. Dogs trained at aversive-based schools show far more stress behavior during training, higher stress hormones afterwards, and a more pessimistic outlook in an unrelated test outside of training, with the effect growing as the proportion of aversives goes up (Vieira de Castro et al., 2020). Owners using confrontational methods report their dogs responding aggressively at striking rates (Herron, Shofer and Reisner, 2009).

Nor does it buy better obedience. In a controlled field comparison, professional reward-based trainers got faster and more reliable first-cue responses for sit and come than professional e-collar trainers following best practice, and the authors found no evidence e-collar training was necessary (China, Mills and Cooper, 2020). Being straight about it: reward-based methods have matched or beaten aversive ones in most direct comparisons but not every one, and the effectiveness literature is thinner than the welfare literature (Ziv, 2017). The welfare evidence points one way only.

It also has to be near perfect to do what it promises. Punishment only works the way it is sold when it arrives within about a second of the behavior, every single time the behavior happens, at an intensity that stops it on the first try, and when it is aimed at the behavior you actually meant. Those conditions come from the laboratory, and veterinary behavior bodies point out that people with ordinary lives and ordinary reflexes almost never meet them (Masson et al., 2018, for the European Society of Veterinary Clinical Ethology). Which gives you the shortest argument against the whole approach: the skill it would take to punish well is the same skill that makes punishing unnecessary.

In a real house the timing slips, and what gets punished is whatever your dog happened to be doing, or whatever your dog happened to be looking at, which is often another dog, a child, or you.

A side note, because it is the most expensive version of that mistake: if you use positive punishment on aggression you risk a silent bite. Growling is information. Punish it and you have suppressed the warning system without touching the cause, so next time there may be no warning at all.

Honesty about this one: nobody has run the experiment, because deliberately punishing dogs for growling to see who gets bitten is not a study anyone should run. It is the consensus of behavior specialists rather than a measured finding. What the record does show is that discipline was among the most common triggers in a series of bites to children, and that most of those dogs had never bitten a child before (Reisner, Shofer and Nance, 2007).

In Push Drop Stick

Never. Marking a rep as a fail is not a punishment. Nothing happens to your dog: the treat simply does not arrive, and the app gets a data point. A fail is a measurement, never an instruction.

Often confused with

A fail, as above. Also management, which prevents the behavior from happening rather than adding a consequence once it has.

Something your dog dislikes stops when they do a behavior, so they do that behavior more. Something is removed, and the behavior increases.

Notice what has to be true first: something unpleasant must already be happening, and it has to keep happening until your dog gets it right. That is the whole problem with this quadrant, and it is why it is also called escape or avoidance learning. There is no version of it that does not begin with your dog being uncomfortable.

Examples: a dog learns to walk beside the handler because that is what relieves the steady pressure of a tight leash. A dog learns to sit because that is what turns the collar's stimulation off. A frightened dog learns that a hard stare, a growl or a snap makes the scary person back away, which is negative reinforcement quietly teaching a behavior nobody meant to teach.

We do not train this way. Positive reinforcement produces the same behaviors without making your dog uncomfortable first, and it does not teach your dog that you are the source of the thing they are escaping.

In Push Drop Stick

Nowhere, by design. One place it can sneak in: an exposure your dog cannot leave. This is why fear work keeps escape freely available at all times and never makes it something your dog has to perform to earn.

Often confused with

Negative punishment. Both remove something. Here the removal comes when your dog gets it right, and behavior goes up.

Every behavior sits in a sequence: something sets it up, the behavior happens, something follows it.

The four quadrants above are all about that third part. This entry is about the first, which is where the leverage actually is.

The antecedent is everything present beforehand: where you are, who is there, what you said, how long since your dog ate, whether the counter has a sandwich on it. The behavior is what your dog does, described so precisely that a stranger could count it. The consequence is what happens next, which determines whether the behavior grows or fades. The whole three-part relationship is called a contingency.

Most owners try to change behavior by working on the consequence, usually by adding an unpleasant one. Trainers get further by working on the antecedent, because it is under your control, it takes effect immediately, and it does not require your dog to be wrong first. Put the trash behind a door. Practise recall before the park fills up. Feed the dog before the guests arrive. Then reinforce the behavior you want in that easier setup.

In Push Drop Stick

Every step in every plan is an antecedent arrangement. A step's setup is the antecedent, its pass and fail criteria define the behavior observably, and the treat is the consequence. When a step turns out to be too hard, the app changes the antecedent, by dropping back or splitting, rather than asking more of your dog.

Often confused with

Triggers. A trigger is one kind of antecedent, the kind that sets off an emotional response.

Anything that changes how much your dog wants a reinforcer right now.

Cheese matters more before dinner than after it. A tennis ball matters more after a day of rain than at the end of an hour of fetch. Shade matters more when it is hot. Nothing about the cheese, the ball or the shade changed, only how badly your dog wanted it, and that is enough to decide whether your training works today.

It runs in two directions. Deprivation raises value: train before a meal, keep the special treats for training only, put the toy away between sessions. Satiation lowers it: a dog who just ate a large breakfast will not work for kibble, and a dog who has been playing with other dogs for an hour is far easier to recall away from them than one who just arrived. That second one is useful in both directions, and it is why deliberately saturating your dog with a distraction first can turn an impossible rep into an easy one.

In Push Drop Stick

The reason two identical sessions can go completely differently. If your dog is refusing treats they normally love, check hunger and arousal before concluding the step is too hard, and note that a frightened dog often will not eat at all.

Often confused with

Treat value. Value is the ranking, roughly stable for a given dog. A motivating operation is what shifts the whole ranking up or down today.

A signal that tells your dog a specific behavior is available to earn reinforcement right now.

Cues have to be trained, and they are usually a spoken word or a hand signal. They can also be environmental: a doorbell cueing your dog to go to their mat, you stopping at a curb cueing a sit. Those need teaching through consistent practice too, they just do not feel like cues because you are not saying anything.

The technical name is a discriminative stimulus, and the precise meaning is worth having, because it quietly settles a lot of arguments: a cue does not make behavior happen, it signals that the behavior will pay off. Your dog is not obeying, they are taking an offer. That is why a cue built with reinforcement keeps working, and why a cue that has stopped predicting reinforcement stops working.

Give the cue once. Repeating it, "sit, sit, siiit, SIT", teaches your dog that the first one was optional. Give it, then wait as long as your dog stays engaged.

In Push Drop Stick

Hand signal goals show a hand icon, verbal cue goals a talking head icon. Plans usually teach the behavior first, add a hand signal, then add the word before the hand signal.

Often confused with

A command, which implies your dog must comply, and a prompt, which is something your dog responds to without any training, like a food lure.

A signal that tells your dog a behavior will not pay off right now, so there is no point offering it.

If a cue means the offer is open, an S-delta means the offer is closed. This matters because a dog left to guess will keep trying, and keep being frustrated. A clear S-delta is kinder than silence: your dog gets an answer instead of an ignored request.

The rule that makes it humane is that the signal must be visible and consistent. The potty bells tied up out of reach. The treat pouch off the counter. The closed door to the training room. The food bowl put away. Your dog can see that the behavior is off the table, learns it quickly, and settles.

Compare that with the version that goes wrong: leaving everything exactly as it was and simply not answering. That is not an S-delta, that is extinction, and it buys you the burst, the frustration, and a weaker behavior once you do want it back.

In Push Drop Stick

Not a step type. It appears inside plans as a change to the setup, and it is the right way to teach "the bells are not an option right now" or "the mat is put away."

Often confused with

A correction. Nothing unpleasant happens. It is information, and the behavior becomes available again the moment the signal changes back.

What people mean when they say "he knows it."

A behavior is under stimulus control when four things are true. Your dog does it when you cue it. Your dog does not do it when you have not cued it. Your dog does not do it when you cue something else. And your dog does not offer something else when you cue this. A dog who sits beautifully, but also sits when you say down, and also sits hopefully at you all evening, does not have a sit under stimulus control. They have a favourite trick.

This is the honest standard behind the word "trained", and the middle two conditions matter more than they sound. Most apparent disobedience is missing stimulus control rather than stubbornness: the cue was never the thing controlling the behavior, the context was. Kitchen, plus treat pouch, plus your body facing them, was the real cue, and the word was decoration. Change the room and it falls apart.

In Push Drop Stick

This is what proofing games are built for. The app shuffles earned goals from different plans and calls them in random order, so your dog cannot succeed by guessing the pattern and has to listen to the actual cue. Goals they nail stay green, goals they miss turn blue for review.

Often confused with

Fluency, which is about how well and how quickly, not about the cue being in charge. Also generalization, which is about where.

A behavior that used to work stops working, so eventually your dog stops doing it.

If the counter has never once had food on it, counter surfing fades. If a door has never opened when scratched, scratching fades. That fading is extinction, and it is not punishment: nothing is done to your dog, the payment simply is not there any more.

Two things make it harder than it sounds.

First, the extinction burst. When something that used to work stops working, animals often try harder before they give up, so the behavior gets louder, longer and more frantic before it fades. This is the point where most owners cave, and one cave teaches your dog that persistence pays, which leaves the behavior far more durable than it was before you started.

It is worth knowing that the burst is not inevitable, despite how often it is described that way. The largest review of real applied cases found one in about a quarter of them, and in only about one in eight when extinction was combined with reinforcing an alternative behavior (Lerman and Iwata, 1995). That research is with people rather than dogs, and it points at exactly the plan below: do not just remove the payoff, pay something else instead.

Second, it is frustrating, and frustration in some dogs comes out as mouthing, barking or snapping.

So extinction on its own is a poor plan. What works is management, so the behavior gets no practice at all, plus differential reinforcement, so there is a well-paid alternative that does work. Take the payoff away from one option and hand it to another.

In Push Drop Stick

Deliberately rare. Plans teach a replacement behavior rather than waiting out an old one, so your dog always has something available that pays.

Often confused with

"Just ignore it." Ignoring a behavior is only extinction if your attention was the thing paying for it. If the reinforcer is the smell of the sandwich, ignoring changes nothing.

The humane answer to "how do I stop it": pay a different behavior generously enough that the unwanted one stops being worth doing.

Instead of adding a consequence for the behavior you dislike, you pick a behavior you like, make it easy in that exact situation, and pay it well. DRA, differential reinforcement of an alternative, pays any acceptable alternative. DRI, differential reinforcement of an incompatible behavior, picks one your dog physically cannot do at the same time as the problem, which is why it is the stronger choice.

Some worked examples:

• Jumping on guests — pay four feet on the floor, or a sit. A sitting dog cannot jump.
• Barking out the window — pay coming to find you, and manage the window.
• Bolting out the door — pay a mat station beside the door.
• Pulling on leash — pay walking beside you.

Notice that in every case your dog still gets what they were after, the attention, the information, the walk. They just get it a different way.

For this to work the alternative has to pay better than the problem behavior does, which usually means higher value treats and a higher rate than feels necessary at first, plus management so the old option stops working while the new one is being built.

This is one of the best-established techniques in behavior analysis, though most of the controlled work is with people rather than dogs. The same review that counted extinction bursts found them roughly three times less common when an alternative behavior was being reinforced alongside (Lerman and Iwata, 1995).

In Push Drop Stick

Most Basic Manners plans are differential reinforcement with the structure filled in. Teaching a station, a settle, or four on the floor is a DRI for the thing you actually want gone.

Often confused with

Distraction. You are not interrupting your dog in the moment, you are permanently changing which behavior pays in that situation.

One event reliably predicts another, and your dog starts responding to the predictor.

Everything above this entry is about what your dog does. This one is about how your dog feels, and it is the bridge to all the fear and confidence work in the app.

You do not have to teach it and your dog does not have to try. Picking up the leash predicts a walk, so the leash itself starts producing excitement. The sound of the treat pouch predicts food, so the sound alone starts the drooling. Nothing was cued and nothing was earned. The association simply formed, because one thing kept coming before the other.

It runs in the unpleasant direction just as easily. If the sight of a stranger has reliably predicted something frightening, the stranger alone becomes frightening. And it is also how you change that, by making the trigger predict something wonderful instead. See Counter-Conditioning and Desensitization.

One consequence matters more than any other, because owners are so often told the opposite: you cannot reinforce a feeling. You cannot make a fear worse by feeding a frightened dog. Food arriving after your dog notices a scary thing does not reward the fear, it changes what the scary thing predicts.

That one follows from the difference between the two kinds of learning rather than from a single study you can point at: reinforcement acts on behavior your dog chooses, and being afraid is not chosen. It is the settled position of behavior professionals, and it is the reasoning every counter-conditioning protocol is built on.

In Push Drop Stick

Classical steps, marked with a party popper icon, earning comfort goals. These steps Drop after a single fail, because trust is far slower to rebuild than it is to keep.

Often confused with

Operant conditioning. Ask what changed: how your dog feels about something, or what your dog does to get something.