Do Dogs Learn by Watching? One Study, Two Measures
Table of contents
When dogs watch people work through a problem, gaze and performance can tell different parts of the story. A 2026 detour study measured both: how long each dog watched a human demonstrate a route, and how quickly the dog reached a reward on later attempts. In this sample, the two measures formed distinct patterns.
What researchers asked
The researchers analyzed video from 187 privately owned, adult purebred dogs in a social-learning task. Each dog faced a transparent V-shaped barrier with a reward visible beyond it. The question was not simply whether a dog looked at a person. The team also tracked whether the dog's route to the reward became quicker across repeated attempts.
The first measure described where the dog looked. The second described how long the task took. Reading them separately keeps attention and task performance from becoming interchangeable labels.
How the detour task worked
Each dog took part in one session of three trials, each capped at one minute. In the first trial, the dog saw the reward placed at the inside corner of the V-shaped fence and then tried to reach it by going around the outside. Before the second and third trials, a human demonstrator showed the detour.
The demonstration included more than movement: the person held the reward, said the dog's name and used attention-getting words. The owner could encourage the dog but could not point out the route. A favourite food reward was used in most cases; a toy was used for a very small number of dogs that were not motivated by food.
| Dogs | 187 privately owned, adult purebred dogs older than one year, drawn from working breeds |
|---|---|
| Testing period | September 2022 to December 2023 |
| Environment | Outdoor detour task; the article does not identify the exact testing location or where participating dogs came from |
| Design | One session, three trials, with a person demonstrating the route before Trials 2 and 3 |
| Analysis | Video-coded behavior, principal component analysis and six-group cluster analysis |
Two measures, not one
The researchers built two separate measures from the videos:
- Relative demonstration watching: the time a dog looked at the demonstrator divided by the length of the demonstration. This is a ratio for these two demonstrations, not a general score for a dog's attention.
- Latency improvement: the time to reach the reward in later trials compared with the previous trial. Each latency was measured in seconds, and the one-minute trial limit matters when interpreting the ratios.
In the analysis, these became two statistical dimensions. Together they accounted for 78.048% of the variation in the coded measures. That percentage describes how much variation those components summarized in this dataset; it is not an accuracy score, a prediction rate or a measure of how well any dog learns in everyday life.
What the six groups revealed
A second analysis grouped dogs according to combinations of watching and improvement. The groups were uneven in size, and their patterns show why the two measures should not be collapsed into a single label:
- Four dogs across two of the smallest groups watched the person for a relatively long time but did not get quicker at the detour.
- One group of 18 dogs, representing 16 breeds, generally improved despite relatively little demonstration-watching; 12 of the dogs were classified as independent working types.
- The largest group included 68 dogs and showed both high relative watching and the strongest improvement in this particular task.
These patterns describe statistical groups within this sample. They do not rank breeds or sort individual pets into fixed categories. The researchers also found variation among dogs of the same breed, showing how much a breed label can leave out about individual behavior.
What the study cannot tell us
The 187 dogs were not a representative sample of all dogs. Researchers pooled eligible dogs from three earlier experiments, kept the transparent V-fence and human-demonstration conditions, and excluded some no-demonstration, neutral-intonation and straight-barrier conditions. Toy breeds and companion-only breeds were not tested. The authors were affiliated with a university in Budapest, Hungary, but the paper does not state the exact test site or where the participating dogs came from.
A quicker second or third attempt does not show that watching the person caused the change. The authors note that practice and trial and error could explain some improvement. The current analysis focused on dogs who saw a demonstration, so it cannot isolate observation as the only explanation for every change in speed.
The result says little about a dog's general intelligence, trainability or learning style. It describes one structured problem-solving task with a particular barrier, reward and human demonstration. It cannot assess an individual pet, diagnose a behavior concern or compare products.
What the measures mean together
In this experiment, watching and doing answered different questions. Watching described where a dog's gaze went during a short demonstration; latency described how long it took to reach a reward. Some dogs showed one pattern without the other, while many showed both. Keeping those observations separate gives a clearer account of what happened than calling every attentive-looking dog a successful learner.
For dog families, a look at the demonstrator and a faster route are separate observations, even when they happen together. This experiment tracked both in a controlled task; it cannot say how a particular dog will respond in a different situation.
Why the distinction matters outside the task
The study is easiest to misunderstand when attention is treated as a shortcut for learning. A dog can watch a demonstration because the person, food or movement is salient. A faster route can reflect practice, trial and error or a useful change in strategy. Those possibilities can coexist. The two measures in this paper keep them visible instead of forcing one label to stand in for both.
Three conclusions the study does not support
- A dog that watches longer is not automatically more trainable.
- A dog that improves quickly is not automatically copying the person's route.
- A group pattern in a V-shaped barrier is not a stable learning type for a household dog.
These limits follow from the design. The dogs saw one kind of demonstration, crossed one kind of barrier and worked for a reward within a short session. The sample included working breeds and excluded some companion-only and toy-breed populations. The paper also notes that the source data came from earlier experiments with different conditions removed. That makes the result useful for a defined comparison, while leaving everyday transfer open.
What a stronger follow-up would add
A follow-up could test whether the same two measures stay separate when the task changes. It could include more companion-only and toy breeds, report the exact testing locations and recruitment frame, and repeat the task with new barriers, new demonstrators and new rewards. Repeated sessions would help separate a one-session practice effect from a more durable change. A preregistered comparison could also make it easier to tell whether watching predicts later performance once motivation and prior experience are accounted for.
Those additions would not turn the study into a general intelligence test. They would answer a narrower question: which parts of the pattern travel to another task, another dog and another setting?
A careful way to observe learning at home
Pet owners can borrow the study's separation without copying its apparatus. When teaching a new behavior, note what the dog watches, what cue is available, how long the dog takes to complete the task and whether the result changes across repetitions. Keep the reward and difficulty reasonably consistent. A dog that looks at the person but takes longer may need a clearer route. A dog that moves quickly without watching may already know the task or may be using trial and error.
One observation should not become a verdict about ability. Look for a pattern across ordinary sessions, and change one feature at a time. If a behavior change is sudden, distressed or unsafe, a qualified professional is the right next step. The Pet Wellness library can provide general care context, but this research brief cannot assess an individual dog.
The best reading of the paper is therefore specific: in this sample, attention during a demonstration and improvement on a detour task captured related but distinct behavior. Keeping them separate gives a better question for future research and a calmer way to interpret a pet's learning at home.
How to test learning without overreading it
A simple home check can keep the distinction visible. First note whether the dog watches and orients to the demonstration. Later, change the position or timing slightly and see whether the dog attempts the action without the same prompt. The two moments answer different questions, so they should not be collapsed into one impression of intelligence or obedience.
Owners should also record the conditions that make the task easier or harder. A familiar room, a quiet pause and a clear reward can change performance. Repeating a small task across days helps separate a useful pattern from a single successful attempt. The study supports better observation; it does not promise that every dog will learn a household cue in the same way.
Source
Pongracz P, Dobos P. Attention and improvement: social learning reveals two behavioural dimensions behind dog-human interactions. Evolutionary Human Sciences. Published online June 11, 2026; 8:e26. DOI: 10.1017/ehs.2026.10057.
The article reports that study data and supplementary information are available through Mendeley Data.