Why Punishment-Based Dog Training Backfires

What “punishment-based” actually covers

Punishment-based training is a broad label, and most of the people using it would never describe themselves that way. It covers the choke chain that tightens when a dog pulls, the prong collar, the electronic collar, the sharp leash correction, the raised voice, the spray bottle, the scruff shake, pinning a dog on its side, and the alpha roll. What they have in common is a mechanism: something the dog wants to avoid arrives after a behaviour, in the hope the behaviour becomes less likely. In learning-theory terms that is positive punishment, one of the four options we walk through in the four quadrants of dog training.

Intensity varies enormously along that list. A muttered “ah-ah” and a shock collar are not the same thing, and it would be dishonest to pretend the research treats them as identical. They do sit on one line, though, and the studies below sample points along it, from ordinary household scolding through to professional e-collar work.

The obedience data: what a controlled trial found

The cleanest comparison available is China, Mills and Cooper, published in Frontiers in Veterinary Science 7:508 (2020). Sixty-three dogs were trained across three matched groups: two worked with reward-based trainers, one with trainers using electronic collars. The tasks were ordinary pet-dog obedience, chiefly recall and sit.

Dogs in the reward-based groups responded to the first recall cue 82.5% of the time. The e-collar dogs managed 71.0%, a difference that reached statistical significance (p = 0.015). Once they arrived, the reward-trained dogs sat in 1.36 seconds against 1.67 seconds for the e-collar group.

Read that honestly. These are not landslide margins, and the e-collar dogs were not ruined; they came back most of the time. It is also one trial, with professional trainers, on simple tasks. What it does establish is narrow and worth owning: on the everyday skills people buy shock collars to fix, the aversive method was not better. It was slightly worse.

Practical take: If a trainer tells you rewards fall apart off-leash and only a collar gives you reliable recall, ask which study shows that. Do not accept a confident anecdote in place of the one trial that tested the question and found the opposite.

The stress data, and the part that lasts

Vieira de Castro and colleagues, writing in PLOS ONE 15(12) (2020), followed 92 companion dogs enrolled at training schools that used aversive, mixed or reward-based methods, and recorded what the dogs did during sessions.

Lip-licking is one of the standard low-level stress signals in dogs. Observers counted a mean of 55.9 lip-licks in the aversive groups and 4.1 in the reward-based groups. Across stress-related behaviours the effect size was d = 3.9, a separation that is rare in behavioural work. Cortisol was elevated in the aversive group as well.

The finding that should change how you think about the trade-off came later. Weeks after training, the same dogs took a cognitive-bias test, where a bowl is placed in an ambiguous spot between a location that has always held food and one that never has. Dogs from the aversive schools approached more slowly, reading the ambiguous bowl pessimistically. The sessions had finished. The outlook had not.

The caveats are real. Owners chose their own schools, so the groups were not randomly assigned, and it is possible that already-anxious dogs were more likely to end up in certain classes. What the design does not easily explain away is the sheer size of the behavioural gap during training itself.

Why owners keep using it anyway

If the outcomes are no better, why is punishment still everywhere? Cooper and colleagues (2014) found that owners using aversive methods reported higher confidence in those methods. No percentage from that work is available to us, so take it as a direction, not a measurement. The direction alone is striking, because confidence and results are pointing opposite ways.

The mechanism is not mysterious, and it has nothing to do with the dog. Punishment usually does stop the behaviour in the moment. The barking cuts out, the jumping stops, the dog freezes. Something unpleasant for you ended the instant you acted, which is a textbook reinforcer for you. You are being trained, very efficiently, by two seconds of silence. Rewards feel slower by comparison because you are building something rather than interrupting it, and building never delivers that jolt of relief.

Noticing this does not make anyone a bad owner. Almost everyone who has lived with a dog has raised their voice at one. It does mean your own sense that “it worked” is the least reliable instrument in the room.

Punishment and reported problem behaviour

Hiby, Rooney and Bradshaw surveyed 364 owners for Animal Welfare 13 (2004). Of those owners, 66% used vocal punishment, 60% used praise, and 51% used food rewards. The finding is that the more tasks an owner had trained using punishment, the more problem behaviours that owner reported.

That is correlational, and the caveat belongs in the open rather than in a footnote. A survey cannot tell you which came first. Owners of already-difficult dogs may reach for punishment more often, which would produce the same pattern with the causal arrow reversed. If you want the fuller picture of where aggression actually comes from, we go through it in dog aggression and its causes.

What the correlation does undercut is the comfortable assumption that punishment buys you a calmer household. Whatever the direction of cause, the heavily punishing homes were not the peaceful ones. DK's The Complete Dog Breed Book (2nd edn, 2020) states the practical conclusion on p. 322: harsh handling “is simply not effective” and is “likely to cause more problem behaviours, including aggression and anxiety”. The same page is equally blunt about choke chains used for disobedience: not effective.

Why punishment fails mechanically, not just morally

Set welfare aside for a moment and judge punishment purely as a teaching tool. Karen Pryor works through eight methods of ending an unwanted behaviour in Don't Shoot the Dog! (Bantam, 1985), and rates only four of them kind or reliable. Punishment is not one of the four.

Her objection is engineering, not sentiment. Punishment arrives after the behaviour is over. By the time you have registered the counter-surf, crossed the kitchen and delivered your correction, the dog has done several other things, and any of them can pick up the association instead. Dogs corrected for barking at the window sometimes learn that the window predicts trouble. Dogs corrected near another dog sometimes learn that other dogs predict trouble, which is precisely the outcome you were trying to avoid.

Even with perfect timing, a correction carries no information about the alternative. “Not that” leaves every other behaviour on the table, including several you will like considerably less. A dog stopped from jumping may lunge, mouth, or bark instead.

Pryor's fifth method is the way out: train an incompatible behaviour. A dog lying on a mat cannot jump on your visitor. A dog holding a loose lead cannot be dragging you down the street, which is the whole logic behind teaching loose-leash walking rather than correcting pulling. The incompatible behaviour gives the dog something to do, which is the one thing a correction can never supply.

What to do instead

Replacing punishment is less about willpower than sequence. A workable order:

1. Manage the environment first. Bins with lids, a gate at the stairs, a lead on in the garden. Every rehearsal you prevent is one you do not have to untrain.
2. Name the behaviour you want. “Stop barking at the door” is a wish. “Go to the mat when the bell rings” is trainable, and you can tell whether it happened.
3. Pay it far more often than feels necessary at first, then thin the payments out gradually once the behaviour is solid. The full method is in our guide to positive reinforcement dog training.
4. Use a marker if it earns its place. The evidence here is genuinely mixed. Gilchrist and colleagues (PeerJ, 2021) ran two controlled studies with pet dogs on simple tasks and found a clicker gave no advantage over food alone or over a verbal marker. Lazarowski and colleagues (Frontiers in Veterinary Science, 2025) randomly assigned 28 detection dogs to a hard task with a long delay before the reward, and there the marker group reached the final stage in 36 trials against 68, with alert accuracy of 92% against 69%. Simple task, probably no difference. Hard task with a delay, it matters.
5. Keep sessions short. Pryor notes that appetite for reinforcement falls off after roughly a dozen reinforcements, and advises ending each session on a high note. Three good minutes beats twenty grinding ones.

For the specific behaviours people usually reach for a correction to fix, we have collected the alternatives in eight ways to stop unwanted behaviour without punishment, and Tavi's training module sequences the underlying skills so you are not deciding what to work on each evening.

Frequently asked questions

Does punishment ever work in dog training?

It often stops a behaviour in the moment, which is why it feels effective, but that is not the same as teaching anything. In the controlled trial by China, Mills and Cooper (Frontiers in Veterinary Science, 2020), dogs trained with electronic collars responded to a first recall cue 71.0% of the time against 82.5% for reward-trained dogs. The aversive method did not produce better obedience on the tasks it is usually sold for, and it came with measurably more stress in the training session itself.

Are e-collars really worse than reward-based training?

On the evidence we have, they are not better and appear slightly worse. The 2020 Frontiers in Veterinary Science trial matched 63 dogs across reward-based and e-collar trainers and found reward-trained dogs recalled on the first cue more often and sat faster afterwards. That is one study on simple obedience tasks, so treat it as a strong hint rather than a closed case. What it does remove is the common claim that shock collars deliver a reliability rewards cannot match.

Is telling my dog no counted as punishment?

Technically yes, if it reduces the behaviour. Practically, a calm interruption is a long way from a shock or a scruff shake, and pretending otherwise helps nobody. The useful question is what happens next. An interruption followed by showing the dog what you wanted and paying for it is training. An interruption on its own leaves the dog with no alternative behaviour, so the same problem returns, and repeated often enough it teaches your dog that your attention predicts something unpleasant.

Does punishment cause aggression in dogs?

The honest answer is that we have a strong association rather than proof of cause. Hiby, Rooney and Bradshaw surveyed 364 owners in 2004 and found that the more tasks trained by punishment, the more problem behaviours were reported. That is correlational, and difficult dogs may attract more punishment in the first place. DK's Complete Dog Breed Book (2020, p. 322) states that harsh handling is likely to cause more problem behaviours, including aggression and anxiety. Either way, punishment is not buying you calm.

What should I do when my dog is doing something dangerous right now?

Stop the situation physically rather than correcting the dog. Pick the puppy up, shut the door, take hold of the lead, block the stairs. Interrupting an emergency is not training and does not need to be; nothing is learned in three seconds of panic. Once everyone is safe, work out what you want the dog to do instead in that exact situation, train it when nothing is going wrong, and manage the environment so the emergency stops recurring while you build the new behaviour.

Tavi's guides are informational and not a substitute for professional veterinary advice, diagnosis or treatment. When in doubt, call your veterinarian.

Sources

• China, Mills & Cooper, Frontiers in Veterinary Science 7:508 (2020)
• Vieira de Castro et al., PLOS ONE 15(12) (2020)
• Hiby, Rooney & Bradshaw, Animal Welfare 13 (2004)
• Cooper et al. (2014)
• Gilchrist et al., PeerJ (2021)
• Lazarowski et al., Frontiers in Veterinary Science (2025)
• Karen Pryor, Don't Shoot the Dog! (Bantam, 1985)
• DK, The Complete Dog Breed Book, 2nd edn (2020)