Positive Punishment & Negative Reinforcement: What Those Terms Actually Mean
Why “positive” doesn’t mean kind, “negative” doesn’t mean cruel — and why changing a behavior is only part of what a dog is learning.
If you’ve spent any time around dog training, you’ve probably heard someone say:
“I only use positive training.” Or “I would never use negative reinforcement.”
The trouble is that, in learning theory, positive and negative don’t mean good and bad. They’re closer to arithmetic.
Positive = something is added.
Negative = something is removed.
And reinforcement and punishment don’t describe whether we approve of what happened, either.
Reinforcement = the behavior becomes more likely.
Punishment = the behavior becomes less likely.
That’s it. Which means something called positive punishment isn’t positive in the everyday sense at all. And negative reinforcement isn’t punishment.
Confused yet? Almost everyone is the first time.
The four pieces of the puzzle
Behavior science describes four basic ways consequences can affect future behavior:
Positive reinforcement:
Something desirable is added, and the behavior increases.
Dog sits → dog gets a treat → sitting becomes more likely.
Negative reinforcement:
Something unpleasant is removed, and the behavior increases.
Pressure is applied to the leash → dog performs the desired behavior → pressure stops → that behavior becomes more likely because it made the discomfort disappear.
Positive punishment:
Something unpleasant is added, and the behavior decreases.
Dog performs an unwanted behavior → leash correction, startling noise, physical correction, or another aversive is added → that behavior becomes less likely.
Negative punishment:
Something desirable is removed, and the behavior decreases.
Dog jumps for attention → attention disappears → jumping becomes less likely.
These aren’t philosophical categories. They’re descriptions of learning. In fact, technically, we don’t even know whether something functioned as a reinforcer or punishment simply because a trainer intended it to. We have to look at what happened to the behavior afterward.
If the behavior increased, something reinforced it. If it decreased, something punished it. The American Veterinary Society of Animal Behavior similarly distinguishes reward-based approaches — primarily positive reinforcement and negative punishment — from aversive approaches involving positive punishment and negative reinforcement. And this is where dog training gets much more interesting than the vocabulary lesson.
Positive punishment: “Don’t do that.”
Imagine a dog lunges toward something during a walk. The dog hits the end of the leash and receives a sharp correction. If lunging decreases because that consequence was unpleasant, the correction has functioned as positive punishment:
Something was added
→ to make a behavior decrease.
And here’s an important point: Punishment can work. If it couldn’t change behavior, it wouldn’t be punishment. That fact sometimes gets lost when conversations about humane training become polarized. The question isn’t simply whether discomfort can make an animal stop doing something. Of course it can. The better question is:
What did the dog have to experience in order for that behavior to stop — and what else did the dog learn along the way?
Because “the behavior disappeared” and “the dog understands what I wanted” are not necessarily the same thing.
Negative reinforcement: relief becomes the reward
Negative reinforcement is perhaps the most misunderstood of all four terms. The easiest way to understand it is this:
The reward is relief. Something uncomfortable begins. The dog performs a behavior. The uncomfortable thing stops. So the dog learns:
Do this, and I can make that feeling go away.
For example, imagine leash pressure is maintained until a dog sits. Dog sits. Pressure disappears. If sitting becomes more likely in the future because sitting reliably turns off that pressure, we have negative reinforcement.
The removal of the aversive consequence reinforced the behavior.
This is the mechanics underneath much of what I meant in my previous discussion of the “do this or else” model of training. The “or else” doesn’t necessarily have to be dramatic. It only needs to be something the dog wants to avoid. And once the animal understands the contingency, the aversive may not even need to occur every time. The possibility of it can become enough. The dog has learned how to prevent it.
One moment can contain both
Here’s where things get especially interesting. Positive punishment and negative reinforcement often occur together, depending on which behavior you’re examining. Researchers studying aversive training have used a very simple example: Imagine spraying a dog with water when they jump.
The spray may positively punish jumping because something unpleasant was added and jumping decreases.
But when the dog puts all four paws on the floor and the spraying stops, removal of the spray can simultaneously negatively reinforce standing on the floor.
One aversive event. Two behaviors. Two learning contingencies. The same thing can happen with leash pressure, electronic collars, physical corrections and other aversive techniques. This is why understanding the terminology matters. We’re not arguing about whether a trainer is “positive” or “negative.”
We’re asking: What experience is controlling the behavior?
“But it worked.”
It very well may have. That deserves to be acknowledged. Aversive learning exists because animals are extraordinarily good at learning how to escape and avoid unpleasant experiences. And there are experiments in which punishment has produced rapid behavioral suppression.
A 2024 randomized study, for example, compared electronic-collar training with food-based protocols intended to stop dogs from chasing a moving lure. In that particular experimental setup, the dogs receiving electronic stimulation stopped chasing much more rapidly than the reward-trained groups. Dogs receiving the shocks were also the only dogs reported to yelp during training. The authors themselves called for further research into longer-term effects and the level of expertise necessary to apply the technique.
That study is worth acknowledging precisely because science shouldn’t require us to pretend aversives can’t work. Instead, it lets us ask a much more meaningful question:
Is effectiveness our only measurement of good training?
I don’t think it should be. If two methods can influence behavior, I also want to know:
How much stress accompanied the learning?
What emotional association was created?
What happened to the dog’s willingness to engage?
What happens when the trainer isn’t present?
Did the dog learn what we wanted them to do, or primarily what they needed to avoid?
What happened to the relationship between the dog and the person holding the leash?
And does another effective method exist with less welfare cost?
Those are scientific questions, too.
When we measure more than obedience
Researchers have increasingly tried to measure exactly those things. In a 2020 study of 92 companion dogs attending seven different training schools, researchers directly observed dogs trained primarily with rewards, dogs trained with a mixture of reward and aversive techniques, and dogs trained with a high proportion of aversive techniques.
Dogs in the aversive group showed more stress-related behaviors, spent more time in tense and low body states, panted more during training and showed greater post-training increases in cortisol than dogs in the reward-based group.
The researchers then tested something beyond the training session itself.
They used a cognitive bias test — essentially asking whether dogs approached an ambiguous situation as though they expected something good or something disappointing.
Dogs from the aversive-training group responded more “pessimistically” than those from the reward-based group. Even dogs in the mixed-method group displayed more stress-related behavior during training than dogs in the reward group.
A separate 2021 study compared 50 dogs whose owners reported using two or more aversive methods with 50 matched dogs whose owners did not report using those techniques.
Again, the dogs exposed to aversive methods demonstrated a more pessimistic cognitive bias.
Importantly, the researchers were careful about what that meant: the study showed an association, not proof that aversive training caused the difference. Dogs who are already more fearful or behaviorally difficult could potentially be more likely to provoke owners into using aversive methods.
That distinction matters. Good science should make our conclusions more precise, not merely louder.
Sometimes the aversive doesn’t buy us better behavior anyway
There is another assumption worth examining: That stronger correction produces stronger training. In a controlled 2014 study involving 63 dogs with recall and chasing problems, dogs were trained either with electronic collars or without them. Owners reported improvement across the groups, but researchers found no significant difference in training efficacy between the electronic-collar and non-electronic-collar groups.
Dogs trained with the electronic collars, however, spent more time tense, yawned more frequently and interacted less with their environment than one of the reward-trained groups.
So the scientific picture isn’t: “Punishment doesn’t work.” It’s more interesting than that. Some aversive procedures can suppress behavior, sometimes very effectively. But the broader evidence gives us reason to consider the welfare cost, emotional consequences and availability of effective alternatives, not merely whether the dog eventually complied.
The American Veterinary Society of Animal Behavior currently recommends reward-based training and concludes that the existing literature favors reward-based methods with respect to welfare, training effectiveness and the dog-human relationship.
What else did the dog learn?
This may be my favorite question in animal behavior. Because the lesson we intended isn’t always the only lesson happening.
Imagine a dog barks whenever another dog approaches.
The handler corrects the barking.
Eventually, the barking stops.
From the human perspective: Success. The dog learned not to bark at dogs.
Maybe.
But the approaching dog was present every time the correction happened. Learning doesn’t occur in a vacuum. The dog could potentially learn:
When another dog appears, something unpleasant happens to me.
The visible behavior may be quieter while the emotional response underneath it remains unchanged — or potentially becomes stronger. This is one reason behavior modification requires us to look at the whole animal, not simply the behavior we’re trying to erase.
A quiet dog isn’t automatically a comfortable dog. A still dog isn’t automatically a calm dog. And the absence of a behavior isn’t proof that we’ve changed the emotion that produced it.
So what does reward-based training do differently?
Reward-based training takes another route. Instead of:
How do I stop the behavior I don’t want?
we begin with:
What would I like the dog to do instead?
A dog jumps on visitors? Let’s teach four paws on the floor or a sit and make that behavior valuable. A dog loses their mind when another dog passes? Let’s create enough distance that they can still think, then reinforce attention, disengagement and calm observation. A dog pulls toward something they desperately want? Let’s teach that a loose leash is what makes access to the environment happen. We’re still changing behavior.
We’re still establishing boundaries. We’re still saying yes to some things and no to others. But we’re arranging the environment so the dog has an understandable path toward success. That is very different from requiring the animal to discover which behavior makes discomfort stop.
And no, this doesn’t mean dogs never hear “no”
Reward-based training is sometimes caricatured as tossing cookies at a dog while allowing them to do whatever they please. It certainly is not. Management matters. Boundaries matter. Safety matters. Consequences matter. If jumping makes human attention disappear, that’s a consequence. If pulling means forward movement temporarily stops, that’s a consequence. If calmly sitting makes the front door open, that’s a consequence.
We’re not removing structure. We’re deciding what kind of information will create it. The AVSAB position statement makes the same distinction: reward-based training does not mean an animal is permitted to engage in every behavior they choose; animals still benefit from routine, boundaries and clear guidelines.
Behavior is communication, too
This is where positive punishment and negative reinforcement connect directly back to Cue, Not Command. When I give a dog a cue, I want the animal’s behavior to be built primarily around:
“I know what that means, and I know what happens when I do it.”
Not:
“I know what happens if I don’t.”
Both can influence behavior. But they are not the same learning experience. And when a dog doesn’t respond, I don’t want my first assumption to be that the consequence needs to become stronger. I want to ask:
Did they understand me?
Have we practiced this here?
Are they over threshold?
Is the environment too distracting?
Am I asking for something they’re physically or emotionally struggling to give me?
Have I made the behavior worthwhile?
Because once we stop treating every unwanted behavior as defiance, training becomes less about winning a contest with an animal. It becomes problem solving.
The dog in front of the behavior
Learning theory gives us beautifully clinical language:
Positive.
Negative.
Reinforcement.
Punishment.
But there is a living nervous system underneath those four squares. An animal forming associations. Making predictions. Seeking safety. Avoiding discomfort. Discovering what works. Learning us while we are trying to teach them. So when choosing a training method, “Did it stop the behavior?” is an important question. It just isn’t the only one.
I also want to know:
What did the dog learn about the world?
What did the dog learn about me?
And perhaps most importantly:
What did the dog have to feel in order to learn it?
That is the difference between simply controlling behavior— and building communication.
Sources & Further Reading
Vieira de Castro, A.C., Fuchs, D., Morello, G.M., Pastur, S., de Sousa, L. & Olsson, I.A.S. (2020). Does training method matter? Evidence for the negative impact of aversive-based methods on companion dog welfare. PLOS ONE, 15(12), e0225023. The study examined 92 companion dogs using behavioral observations, salivary cortisol and a cognitive-bias task.
Casey, R.A., Naj-Oleari, M., Campbell, S., et al. (2021). Dogs are more pessimistic if their owners use two or more aversive training methods. Scientific Reports, 11, 19023.
Cooper, J.J., Cracknell, N., Hardiman, J., Wright, H. & Mills, D.S. (2014). The welfare consequences and efficacy of training pet dogs with remote electronic training collars in comparison to reward based training. PLOS ONE, 9(9), e102722.
Vieira de Castro, A.C., Araújo, Â., Fonseca, A. & Olsson, I.A.S. (2021). Improving dog training methods: Efficacy and efficiency of reward and mixed training methods. PLOS ONE, 16(2), e0247321.
Johnson, A.C. & Wynne, C.D.L. (2024). Comparison of the Efficacy and Welfare of Different Training Methods in Stopping Chasing Behavior in Dogs. Animals, 14, 2632.
American Veterinary Society of Animal Behavior. Position Statement on Humane Dog Training. AVSAB recommends reward-based training methods and summarizes evidence regarding welfare, efficacy and the human-dog relationship.