For ten days on the streets of West Bengal, India, researchers from the Indian Institute of Science Education and Research (IISER) Kolkata staged a quiet contest between two of the oldest currencies in the dog–human relationship: food and affection. Two strangers approached free-roaming street dogs. One carried pieces of raw chicken; the other offered nothing but six gentle strokes along the head and neck. On the first day, the dogs chose the chicken decisively. By the second day, the preference had dissolved into a coin flip — and it never came back. The study, published on 3 March 2026 in the journal Animal Cognition, is the first to pit food against physical affection head-to-head in repeated field trials with unowned dogs, and its verdict is more nuanced than either “dogs love food” or “dogs just want love” would suggest.

Free-ranging dogs are far from a marginal case. Roughly 80 percent of the world’s dogs live this way: breeding freely, scavenging from human refuse, begging at street corners and answering to no owner. Dogs were the first species humans domesticated, a process generally placed between 40,000 and 15,000 years ago across Europe and Asia, yet the free-ranging populations of the Global South retain a lifestyle closer to that original contract than any kennel-bred companion does. In India they have shared human settlements for centuries, occupying everything from villages to dense cities, and their relationship with people is starkly ambivalent: some residents feed them daily, while others beat or kill them — humans are, in fact, the leading cause of death for young free-ranging dogs. Because these animals evolved under natural selection rather than the artificial selection that sculpted pedigree breeds, they carry greater genetic diversity and serve as a living laboratory for studying how the bond between two species assembles itself without leashes, food bowls or owners.

The question the researchers posed was deceptively simple: when a street dog decides whom to trust, does food or petting carry more weight? Decades of reinforcement research have treated food as the most powerful reinforcer in dog training, yet a body of work dating back to the 1950s argues that social interaction can match or even exceed it. A 2017 study from the same Indian group reported that free-ranging dogs preferred petting over food in repeated interactions with unfamiliar humans, while a 2020 study found no clear first-day preference between a food giver and a social-contact giver. Because Indian street dogs survive largely by scavenging and begging, the team hypothesized that food would dominate. They predicted that dogs would choose the food provider more often, approach her faster, and display warmer behaviour toward her across repeated trials.

The fieldwork, conducted between February and June 2023 in Gayeshpur, Kataganj and Kalyani in the Nadia district, began with a filter. Ninety-nine visibly healthy adult dogs, all strangers to the experimenters, were screened in an approachability test: an experimenter stood four to five metres away, called the dog with a friendly “ae-ae-ae” vocalization, held a neutral posture with steady eye contact, and waited up to sixty seconds. Only dogs that closed the distance to within one body length — roughly 0.8 metres — without showing fear or aggression advanced. Sixty-one did. Over the following ten days, fifty-one completed every trial, and forty-five made a clear, scorable choice in all ten choice tests; these forty-five formed the analytical core. Three experimenters ran the sessions, dressed identically and matched for height and build, and hid between trials inside a curtained electric vehicle so dogs could not anticipate who would emerge. Each dog was photographed, GPS-tagged and identified by distinctive scars, ear shape and coat patches, and sessions ran in morning and evening windows aligned with the dogs’ natural activity peaks.

Days one through five formed the familiarization phase. For each dog, one experimenter was randomly assigned as the food provider, dropping an eight-to-ten-gram piece of raw chicken — a rare but highly preferred food in the street-dog diet — once the dog approached; the other served as the petting provider, leaning in to stroke the dog six times from the crown of the head to the neck while holding its gaze for about ten seconds. Rewards were delivered roughly two minutes apart, with the order randomized across days, and dogs were given up to three attempts per phase, though nearly all chose on the first call. Choice tests followed after a two-minute gap: both experimenters stood four to five metres from the dog and one to one-and-a-half metres from each other in a triangular arrangement, calling the dog in unison for up to sixty seconds. No rewards were ever offered during tests. A “clear choice” required approaching within one body length of an experimenter and directing at least a gaze at her; tail wags and body contact counted as stronger affiliative signals. From days six to ten, only these memory-driven tests ran — no food, no petting, just recognition.

The team quantified two behaviours throughout: approach latency, measured in seconds from the first call, and a Socialization Index (SI) ranging from zero to twenty-nine, which aggregates affiliative behaviours such as gazing, wagging, proximity-seeking and body contact, normalized to a zero-to-one scale for statistical modelling. Choice was analysed with binomial generalized linear models; approach times were fitted with Cox proportional hazards models, switching to log-logistic accelerated failure time models whenever the hazards assumption was violated, while the bounded SI scores required beta regression. An independent rater who took no part in the experiment coded twenty percent of the videos, achieving near-perfect agreement (Cohen’s kappa = 0.947). On day zero itself, neither sex, nor the level of human foot traffic, nor sociability predicted how quickly dogs approached a stranger, with a mean latency of about 11.7 seconds. On day one, however, the result was unambiguous: dogs chose the food provider significantly more often (chi-squared = 5.00, p = 0.025). From day two onward, the preference vanished — choices became statistically indistinguishable from chance, and day-to-day consistency fell so low that agreement was below what chance alone would predict.

Beneath the headline choice, the learning curves were revealing. Approach latency declined day after day during familiarization, and the Socialization Index jumped roughly 1.4- to 1.7-fold from day one to day two before stabilizing — evidence that a handful of positive encounters were enough to convert strangers into known quantities. Crucially, a dog’s sociability toward a stranger on day zero predicted its sociability throughout the study (beta = 0.093, p < 0.001), suggesting the index captures a stable behavioural trait rather than a passing mood. Females consistently outscored males, showing about 1.8 times the baseline socialization index, in line with earlier reports of sex differences in canine sociability. Whoever visited second — chicken or no chicken — received about twenty percent less affiliative behaviour, hinting at dampened enthusiasm for repeat engagements within a single session. A transient left-side bias in approaching the experimenters also emerged between days one and five before dissolving, while dogs that had been warmer toward the petting provider during familiarization were correspondingly less likely to pick the food provider in the early choice tests.

The reward-free phase delivered perhaps the study’s most striking finding. With all reinforcement withdrawn from days six to ten, dogs continued to approach both experimenters at low latency — until day ten, when approach times rose about 1.6-fold compared with day six, a textbook signature of extinction as motivation gradually fades. Yet the Socialization Index stayed high throughout, indicating that the pull toward familiar humans outlasted the rewards that built it, echoing anecdotal reports of Indian street dogs greeting known feeders after months of absence. Dogs living in areas with less human foot traffic showed higher sociability during this phase (beta = -0.014, p = 0.028), hinting that animals from quieter streets may harbour a stronger drive for social contact once a person becomes familiar, even if they are warier of strangers. A subtler signal also surfaced: the identity of the specific experimenter began to matter, with substantial random-effect variance suggesting the dogs were choosing between people, not merely between reward types.

The authors interpret the pattern as a division of labour between two reward systems. Food is a fast-acting but short-lived currency — a powerful lever on first contact that quickly loses its edge once both humans are known. Petting accrues value with repetition until both rewards purchase roughly equal loyalty, a dynamic the researchers liken to a serial reversal task in which two continuously reinforced stimuli eventually elicit equivalent responses, making random choice the cognitively cheapest option. The result also complicates the 2017 finding of a petting preference, which relied on a different paradigm in which trust was inferred from a dog’s willingness to take food from a human hand; here, each individual dog weighed two distinct providers against each other, a stiffer test of preference. One caveat applies: because only dogs bold enough to approach a stranger on day zero were included, the conclusions speak primarily for the more sociable slice of street-dog society.

All sessions were approved by the IISER Kolkata Animal Ethics Committee, and every experimenter was vaccinated against rabies — a necessary precaution given that the disease remains endemic among free-ranging dogs in India and other parts of the Global South. Notably, none of the selected dogs showed hesitation, fear or aggression in response to petting on any day of the study, a testament to the screening protocol. Beyond its street-side methodology, the work carries practical weight: it challenges the entrenched assumption that food is the universal motivator of dog behaviour, informs sterilization and vaccination teams who depend on street dogs tolerating human handling, and suggests that simple tactile kindness can build durable social bonds with animals that no one owns. The team now aims to untangle the ecological and individual factors that tilt these preferences further. Fifteen thousand years after domestication began, the experiment reads like a field note on how the ancient contract between the two species was probably drafted — food opened the negotiation, but affection signed it.