A hunter told the dog to leave a piece of food on the floor alone, then sat down on a chair a metre and a half from it, facing the food. On some trials the hunter watched. On others the hunter’s eyes were closed. Earlier work found that dogs take the forbidden food more readily when the person who forbade it cannot see, which has been read as evidence that a dog tracks what a human can and cannot observe. Run on hunting dogs in five countries, the effect did not appear.
The null sits inside a paper titled “Striking global similarities in dog–human interactions”, published in Scientific Reports on 23 June 2026 by the cognitive psychologist Juliane Bräuer and colleagues. Four of its six experiments replicated what Western labs had found. This one did not, and the authors say so themselves.
The paper’s central finding holds up. It is also more mixed than its own summary, and every one of the mixtures is printed in the results section.
The test that did not work
The paradigm the team borrowed comes from a 2003 experiment by Call and colleagues, and it is one of the standard tools for asking whether a dog models what a person can see. The version here had owners issue the prohibition themselves, then sit either watching or with their eyes shut.
Whether the eyes were open made no difference to whether the dog ate the food, and there was no interaction between country and condition. The section opens: “We did not replicate previous findings that dogs are differentiate between humans’ attentional state,” typo in the original.
The authors do not present this as overturning anything. They attribute it to their own setup, in which owners rather than an experimenter issued the prohibition, and did so less consistently than in earlier studies. That is a fair reading, and also the sort of caveat that vanishes by a finding’s third retelling.
Four cups, one dog, and an owner who had to guess
Bräuer’s team at Friedrich Schiller University Jena, working with the Max Planck Institute for Evolutionary Anthropology in Leipzig, ran six standardised tests and a 31-question interview with at least 30 hunting dog and owner pairs at each of five sites: rural Germany, Efate in Vanuatu, Khentii Province in Mongolia, the Andasibe region of Madagascar, and Shipibo-Konibo villages in the Peruvian Amazon. Jena puts the total at 164 dog-human teams, tested between October 2022 and September 2024. Almost everything known about dog cognition comes from Western pet dogs, and roughly 75 percent of the world’s dogs are not kept that way.
The showing test produced the study’s most striking cross-country result. A researcher hid food in one of four cups, set in a row at a height the dog could not reach, while the owner was out of sight. The owner came back and had to pick the cup with nothing to go on but the dog’s behaviour. Hunters on Efate got it right 81.7 percent of the time. German owners managed 56.3 percent. Chance was 25 percent, so every group beat it, and the Vanuatu advantage was significant against Germany, Mongolia and Peru.
The authors supply a reason: on Efate, dogs are essential for tracking wild pigs through dense undergrowth, which demands fine coordination between hunter and animal. Jena’s release puts it that hunters there are much better at reading their dogs’ cues than hunters anywhere else in the study, which goes a step past the paper, where the comparison with Madagascar did not reach significance.
The dogs’ half of that exchange did not register. When researchers coded which cup the dogs looked at first, the dogs did not signal the correct one significantly above chance.
Elsewhere the similarities are real. Dogs everywhere looked to humans when stuck and took cues from their owners when confronted with something alarming. More than 90 percent of owners in every country said they could at least sometimes rely on their dog to be there for them, and at least sometimes expected it to protect them, with no significant variation between countries. Every owner at all five sites said they enjoyed having the dog around at least some of the time. One owner, in Peru, was the only one who did not also say life was sometimes better for it.
Following a human point held up in the sense that mattered most, which is that dogs performed similarly everywhere, with no significant country variation. It did not hold up uniformly. Dogs beat chance in Germany at 75.6 percent, Madagascar at 67.4 and Mongolia at 70.2, but not in Vanuatu at 58.6 or Peru at 67.9. The Peruvian figure sits above Madagascar’s and still fails the test, because it rests on eight dogs with valid trials, which the authors flag themselves. The paper never prints a chance level for this task.
Eleven of fourteen measures varied by country
The paper tests for a country effect on fourteen behavioural outcome measures. Eleven come back significant.
Some of the differences are wide. German dogs were more persistent on the unsolvable box than dogs in Madagascar, Peru or Vanuatu. They played with their owners around the frightening toy far more often than dogs at any other site. Faced with the sealed box, they were less likely than dogs in Madagascar and Vanuatu to direct their first look at the unfamiliar researcher.
The similarity claim survives partly on how the statistics were corrected. Of the fourteen measures, only seven produced a country pair that stayed significantly different once comparisons were corrected for multiple testing. On four, the overall country effect was significant while no individual pair was, leaving a difference real in aggregate and unlocatable in detail. Three measures showed no country variation at all.
Two places where the discussion runs ahead of the results are worth flagging for anyone citing the paper. The discussion says German dogs were more persistent than dogs from other societies, where the results name Madagascar, Peru and Vanuatu only and do not report a significant Germany-Mongolia difference. It also states that two thirds of Peruvian dogs disobeyed the food prohibition, a figure that appears nowhere in the results and carries no accompanying test.
There is also analysis described but never reported. The statistical methods specify a model for a perspective-taking proportion-of-time measure, and a perspective-taking latency figure surfaces in the list of overdispersion checks. No model output or test statistic for either appears in the paper or its supplementary tables, and the table of coded behaviours defines neither variable, although two supplementary tables do carry an unexplained duration figure for the test.
Feeling close to a dog barely predicted anything the dog did
The most useful result got the least attention. The team built a relationship quality score from twelve interview questions and tested it as a predictor on all fourteen behavioural measures. It reached significance on one.
Dogs whose owners scored higher were likelier to approach the frightening object alongside them. On the other thirteen, including every measure of communication, obedience and persistence, how close an owner felt to their dog predicted essentially nothing about what the dog did, with joint play the only near miss. The authors state it plainly: the score and the individual interview answers “were, for the most part, not significant predictors of performance in the tests.”
Their explanation is that the scores were high and similar nearly everywhere. Peru is the exception, consistently: its owners scored lowest on relationship quality, were likeliest to say they sometimes regretted having a dog, and were less likely than German or Mongolian owners to say the dog understood their feelings. Every significant questionnaire difference between countries involves Peru, which the authors tie to hunting practice, it being the only site besides Germany where hunters sometimes go out without dogs.
The paper is candid about the limits. Interview questions came through local assistants in translation, which can shift meaning, and owners may have given the answer they thought was wanted. The tests were built for Western pet dogs, and in Peru, Vanuatu and Madagascar some had to be dropped from the pointing and unsolvable-problem tests because the dogs were not used to being held back or to eating from cups and plastic boxes. Germany was the only Western site, though the authors note the German results match what the same paradigms produced elsewhere in the West. Every dog was a hunting dog, though most also guarded: only eight of 35 Mongolian and five of 32 Peruvian dogs hunted exclusively, and even in Germany 16 of 34 had other jobs. Bräuer ran the tests for every non-Western dog and most German ones.
Where that leaves the evolutionary story
The framing that travelled furthest was that the consistency points to shared evolutionary history rather than culture. Jena’s release opens on that question and says the bond “may be deeply rooted in their shared evolutionary history”; the same release, whose English version is machine-translated from the German, quotes Bräuer saying the team found the dog-human relationship surprisingly universal across the globe. The Max Planck release gives the idea its own section, while closing on senior author Russell Gray’s rather different emphasis, that this is a relationship which has adapted to many different ways of life.
The paper is more guarded than the coverage that followed. It reports the evolutionary account as something “some authors have speculated,” says cooperative hunting “may have favored” the bond’s development, and on the tighter German dog-to-single-owner attachment declines to choose between explanations: it “could have proximate reasons,” meaning individual learning, “but also ultimate reasons,” meaning selection. Its explanations for the differences lean mostly on hunting technique and Western social psychology, though selection for persistence in Western dogs is floated too.
The strongest similarity claim in the paper is also narrower than the headline, and carries a conditional clause: cultural differences do not appear to influence the bond much, “particularly when dogs are still used for hunting, which requires cooperation.”
What survives is worth having. Dogs looked to people when they were stuck, in five very different places. Their owners, wherever they lived, mostly said they could count on them. And the owners who read their dogs best were not the ones whose dogs go through obedience training and sit a hunting exam, but hunters on Efate working four cups with nothing but a dog to tell them.
