Treats or affection? Understanding reward preferences in Indian free-ranging dogs

Abstract

Free-ranging dogs (FRDs) constitute approximately 80% of the global dog population. They are freely breeding and live without direct human supervision, making them ideal for studying how factors such as the lack of supervision, unmanaged breeding, and variable human contact shape dog-human relationships. Living in proximity to humans, FRDs in India frequently interact with people, and previous studies suggest humans to be a crucial part of their social environment. Positive reinforcement in the form of food and petting is commonly received from humans. In this study, we investigated which reward, food or petting, is preferred more during short-term and repeated interactions. Field trials were conducted on 61 adult FRDs. During the familiarization phase (Days 1 to 5), two unfamiliar individuals each provided either food or petting to the dogs. This was followed by a series of choice tests (Days 1 to 10), in which dogs could choose between the two individuals. On the first day, dogs significantly preferred the food provider. However, from the second day onward, preference was no different from chance, suggesting that the strength of food as a reward was reduced. These findings suggest that while food is a stronger short-term motivator, repeated interactions involving either food or petting contribute equally to the formation of positive social associations over time. This study sheds light on the development of the dog-human relationship in Indian FRD populations and highlights the nuanced role of different rewards in fostering affiliative associations.

Supplementary Information

The online version contains supplementary material available at 10.1007/s10071-026-02046-4.

Introduction

The domestic dog, Canis lupus familiaris (Linnaeus 1758), is the first species to have been domesticated by humans (Clutton-Brock 1995) and is widely regarded as man’s best friend (Ostrander and Giniger 1997). It is believed that the domestication of dogs took place between 40,000 and 15,000 years ago in various regions of Europe and/or Asia (Larson et al. 2012; Ovodov et al. 2012; Frantz et al. 2016; Botigué et al. 2017; Bergstrom et al. 2020; Perri et al. 2021). Hunter-gatherers of the Pleistocene epoch are thought to be responsible for this process. While the domestication of approximately twenty animal species to date has been for food, clothing, and transportation (Beaver 2009), the reasons behind the domestication of dogs remain unclear (Larson et al. 2012; Tancredi and Cardinali 2023). Since their domestication, dogs have occupied a special place in human society (Nagasawa et al. 2009), serving purposes such as hunting, guarding, and herding in early human societies. In more recent times, they have taken on additional roles, including service dogs, sniffer dogs, and companion animals (Coppinger and Schneider 1995). Dogs have become integral members of human households, forming a unique and special bond with our species. Several traits have contributed to their esteemed status, particularly their ability to respond to human gestures and language (Bhattacharjee et al. 2017a, b, 2020) and their hypersocial behaviour (Bhattacharjee et al. 2017a, b; VonHoldt et al. 2017). Caring for dogs evokes positive emotions in humans that are similar to the bond shared between mothers and infants (Serpell 2003). Likewise, dogs form attachments to their owners just like children do with their parents (Horn et al. 2013).

There is an uncertainty regarding the relative roles of different reinforcers in the development of social bonds between dogs and humans. Food is often cited as the most powerful reinforcer (Fukuzawa and Hayashi 2013; Feuerbacher and Wynne 2012; Elliot and King 1960). However, some studies suggest that social interaction, such as petting, is just as effective, or even more so, in establishing bonds (Brodbeck 1954; Igel and Calvin 1960; Fonberg et al. 1981; Bhattacharjee et al. 2017a, b).

In the Global North, it is common to keep dogs as companion animals. However, the situation is different in countries of the Global South, such as India, where dogs are not only owned as companions, but are also found to live freely, sharing human habitats (Lord et al. 2013). These dogs, known as free-ranging, make up about 80% of the world’s dog population. They live on the streets, breed freely, and are not under direct human supervision (Boitani and Ciucci 1995; Bonanni and Cafazzo 2014; Hughes and Macdonald 2013; Serpell 1995). They are primarily scavengers dependent on human-generated waste, but also beg and rely on handouts from humans (Bhadra et al. 2016). Unlike most companion dogs (except adopted FRDs), shaped through artificial selection, FRDs have evolved under natural selection and exhibit greater genetic diversity. Their ongoing exposure to environmental selection pressures makes them especially suitable for research (Akey et al. 2010; Shannon et al. 2015). In India, they have lived alongside humans for centuries (Thapar 1990) and can be found in all human settlements, from rural areas to urban centres (Vanak and Gompper 2009a). However, although they serve as effective guardians in many rural areas (Martinez et al. 2022; Sepúlveda et al. 2014), free-ranging dogs also create disturbances by scattering garbage, defecating in public spaces, barking at night, and acting as reservoirs for zoonotic diseases such as rabies (Fekadu 1982), thereby becoming a nuisance for local communities.

In the Indian society, there is a stark contrast in how people perceive and interact with dogs. While some people care for and regularly feed them, others subject them to violence, with incidents of dogs being beaten or killed by humans not being uncommon. In fact, humans are the leading cause of mortality for dogs, particularly in their early life (Paul et al. 2016). Thus, dogs need to assess the intentions of unfamiliar humans before interacting with them (Bhattacharjee et al. 2018; Ruiz-Izaguirre et al. 2014).

Two common rewards that FRDs in India receive from humans are food and petting, with food being a common (Chaudhari et al. 2022) and more frequent reward compared to petting (unpublished data). In this study, we aimed to explore the differing roles that food and petting played in fostering positive social association between humans and FRDs, using a choice test, where each dog had the option to choose between two humans providing two kinds of rewards. Lazzaroni et al. (2020) reported that FRDs showed no clear preference between a person providing social contact and one offering food during the first day of interaction, suggesting that they may not prioritize one reward type over the other in the short term. Building on this, we examined both short- and longer-term effects of different reinforcers on dogs’ social responses. Since FRDs in India are scavengers and frequently beg for food from humans, we hypothesized that food would be a stronger driver in the development of positive social associations with humans. We predicted that dogs would (1) show higher preference for the food provider, (2) exhibit shorter latencies to approach the food provider, and (3) display more affiliative behaviours towards the food provider across trials.

Methods

Subjects

A total of 99 randomly selected adult FRDs unfamiliar to the experimenters were tested. A male dog was considered an adult if his testes had descended, while a female was considered an adult if her nipples were dark in colour (Nandi et al. 2024). We also consulted the local residents regarding the approximate age of each dog and classified an individual as an adult only if it was estimated to be at least one year old (Pal 2001). The dogs’ sex was determined by observing their genitalia. Only visibly healthy dogs, with no signs of injury or illness, were selected for the experiments. Since dogs are typically social animals, most of the selected dogs were part of a group. Non-focal dogs in the group were lured away by the experimenter performing the video recording. The dogs were photographed for tracking purposes, and their locations were marked using GPS on cell phones. They were identified based on their location, sex, body colour, distinctive patches, ear and tail shape, and other unique features, such as scars from previous injuries. The experiments were conducted between February and June, 2023.

Study areas

The experiments were conducted in Gayeshpur (22°57’19.40"N, 88°29’45.88"E), Kataganj (22°57’2.61"N, 88°28’31.82"E) and Kalyani (22°58′30″N, 88°26′04″E) in Nadia district, West Bengal, India (Figure S1).

Experimenters involved

Three adults, all unfamiliar to the dogs were involved. Two were females (Experimenter 1 [E1] and Experimenter 2 [E2]), and one was male (Experimenter 3 [E3]). E1 and E2 gave rewards to the dogs and participated in the choice tests by acting either as the food provider (FP) or petting provider (PP), while E3 performed the approachability test on the first day of the experiment (Day 0) and recorded videos on the subsequent days.

Experimental procedure

Approachability test

The approachability test was conducted to assess the sociability of the dogs, which can be defined as their tendency to approach an unfamiliar experimenter (Bhattacharjee et al. 2021). On the first day of the experiment (Day 0), the experimenters walked through an area, and upon spotting one or more dogs, they conducted the test on individual dogs rather than groups. E3 stood approximately 4–5 m from the dog and called out with a positive vocalization (ae-ae-ae) for up to 60 s. The experimenter maintained a neutral posture, with arms held alongside the body and the head oriented straight ahead. Since the dogs were free-ranging and not on a leash, E3 had to adjust his position in relation to the dog by eye estimation. E3 also maintained eye contact with the dog throughout the experiment. The dog was given a maximum of 60 s to approach E3 within about one body length (approximately 0.8 m). Only dogs that approached E3 within approximately 0.8 m within 60 s and showed no signs of fear (e.g., crouched body, tail droop) or aggression (e.g., growling, barking, snarling) were selected for further experimentation. Video S1 shows the protocol used during the approachability test.

Familiarization phase

The familiarization phase took place from Day 1 to Day 5. The goal was to help the dogs associate the experimenters (E1 and E2) with the specific reward they received from them. For each dog, one experimenter provided food rewards, while the other provided petting rewards across all five days. The assignment of rewards was randomized for each dog: for some, E1 provided food while E2 gave petting, and for others, the rewards were reversed. On any given day, both E1 and E2 gave their pre-determined rewards to the dog, ensuring there was approximately 2-minute gap between the rewards. The order of reward delivery was randomized across days; on some days the food provider delivered the reward first followed by the petting provider, and on other days this order was reversed.

Familiarization involving food

The experimenter providing food, the food provider (FP) stood in front of the dog and called it using the same protocol as the approachability test. If the dog approached within 1 body length from the experimenter within 60 s, a piece of raw chicken (weighing approximately 8–10 g) was dropped in front of the dog. The experimenter stood in front of the dog and gazed at it until the food was eaten, after which the experimenter left and hid in an e-vehicle parked away from the dog. Video S2 shows the protocol used by the food provider.

Familiarization involving petting

The protocol followed by the petting provider (PP) was the same as that followed by FP. The experimenter leaned forward and petted the dog 6 times by running the fingers of her right hand from the top of the dog’s head to the neck, and also gazed at it for approximately 10 s.

If the dog did not approach within approximately 0.8 m of the experimenter within 60 s during either the food or petting trials, the trial was considered unsuccessful, and no reward was provided for that trial. A maximum of three food trials and three petting trials were permitted per dog per day, contingent on the dog having failed to receive the reward in earlier trials. Throughout the procedure, E3 remained close to the dog at all times. Video S3 shows the protocol used by the petting provider.

Test phase

The test phase was conducted to assess the dogs’ preference for the person providing food or petting rewards and involved a simple choice test involving E1 and E2. In each trial, E1 and E2 stood in front of the dog at a distance of approximately 4–5 m, maintaining a distance of about 1–1.5 m between each other. The dog, E1, and E2 were positioned in a triangular arrangement. The trial began with both experimenters calling the dog using the same vocalization, which was used in the approachability test and the familiarization phase. They called for a maximum of 60 s, maintaining eye contact with the dog throughout this period. If the dog did not approach within one body length of one of the experimenters within 60 s, the trial was considered unsuccessful. No reward was provided by any of the experimenters during the test phase. Video S4 shows the protocol used during the test phase.

The test phase took place from Day 1 to Day 10, with a maximum of three trials per day. More than one trial per day was conducted only if the dog did not approach either of the experimenters or if the choice was unclear. By ‘clear choice,’ we refer to instances in which the dog approached within 0.8 m of either E1 or E2 and displayed distinct attentional and/or affiliative cues, such as gazing, tail wagging, or making body contact with the chosen experimenter, with gazing toward the chosen experimenter considered the minimum criterion. An ‘unclear choice’ was noted when the dog approached within 0.8 m of an experimenter but failed to display the minimum attentional cue (gazing) or any affiliative behaviour, rendering the choice ambiguous. From Days 1 to 5, the test phase followed the familiarization phase involving the food provider (FP) and the petting provider (PP). On Days 6 to 10, only the test phase was conducted. The time gap between the familiarization and test phase was 2 min.

The posture maintained by both experimenters during the familiarization and test phases was identical to that used during the approachability test on Day 0. In order to minimize potential bias toward a particular experimenter (FP or PP), they wore the same-coloured clothing and shoes throughout all phases of the experiment. The experimenters were also of a similar height and build. The side on which FP and PP stood during the trials was also randomized across days. All experiments were conducted either in the morning (06:00–09:00 h) or evening (15:00–18:00 h) to align with the activity patterns of the dogs, the availability of daylight, and the convenience of the experimenters. Only daylight hours were selected for the experiments to ensure adequate visibility. The dogs were held by E3 during both the familiarization and test phases and were released only when the experimenters called them for the first time. This procedure was implemented because, as observed during the pilot studies, the dogs became increasingly familiar with the experimenters and began approaching them even before being called. E1 and E2 remained seated inside an e-vehicle, which was equipped with curtains on the sides and back, preventing the dog from seeing the experimenters. The experimental protocol is outlined in Fig. 1.

Fig. 1.

Schematic representation of the experimental protocol. Abbreviations: (a) E1, E2, E3: Experimenter 1, 2 and 3, (b) bl: body length of dog, i.e. the length from the nose tip to the tail base (1bl = ~ 0.8 m). Illustration by Arpan Bhattacharyya

Human flux Estimation

The human flux in the areas where the dogs were located was measured using the protocol outlined in Bhattacharjee et al. (2021). A total of 12 flux videos were recorded for each location where the dogs lived, across four randomly selected days. Two videos were recorded during the 06:00–09:00 h slot, and the remaining two were recorded during the 15:00–18:00 h slot. The mean flux for each location was calculated using these 12 videos. Video S5 shows the protocol used for recording flux.

Video decoding and analysis

All the videos were coded by S.N., and the obtained data were used for further analysis. An independent rater, S.M., who did not take part in the experiment, coded 20% of the videos for choice. Inter-rater reliability was high (Cohen’s κ = 0.947, z = 14.7, p < 0.001). Only the dogs that completed and showed a clear choice in all ten choice tests (up to Day 10) were used for analysis. We checked four parameters across different days of the experiment which were namely choice, latency of approach (in seconds), trial number in which the dog approached and socialization index (SI).

Choice

A choice was scored if a focal dog approached close to i.e., within one body length (approximately 0.8 m) of the experimenter and at least gazed at her during the choice tests. Only “clear choices”, described earlier, were scored as choices.

Trial number

This variable represented the sequential number of the trial in which the dog approached, either in the familiarization phase or the test phase (choice test). Trial number could have a maximum of three levels (1, 2 and 3 representing the first, second and third trial, respectively). This parameter was however, not considered in the analysis due to the very infrequent occurrence of trial numbers 2 and 3, both in the familiarization and test phases.

Approach latency

Approach latency was measured in seconds and was defined as the time taken by the dogs to approach within one body length of the experimenter after the dog was called for the first time and released by E3.

Socialization index (SI)

The Socialization Index (SI) was scored following the method described in Table 1 of Nandi et al. (2024), with the addition of one new behaviour observed in the current study (Table 1).

Table 1.

Behaviours exhibited by the tested dogs and their corresponding scores used to calculate the socialization index (SI)

Behaviour	Score
No tail wag	0
Slow tail wag	1
Fast tail wag (without back movement)	2
Rapid tail wag (with back movement)	3
Licking experimenter	4
Pawing experimenter (using either the left or right paw to touch the experimenter)/Nudging experimenter (using nose to touch the experimenter)	5
Jumping on experimenter’s body/Affiliative vocalizations while looking at the experimenter	6

Choice of the food provider and petting provider was coded as 1 and 0, respectively. A generalized linear model (GLM) with a binomial distribution was used to model choice behaviour. Approach latency was analyzed using a Cox proportional hazards model when the proportional hazards assumption was met. In cases where this assumption was violated, an accelerated failure time model was used. The SI ranged from 0 to 29. To normalize the scores, each SI value was divided by the maximum possible SI score (29), scaling it to a range between 0 and 1. Since the number of zero values was minimal, a small non-zero value was added to ensure all transformed SI values fell strictly within this range, allowing for a beta regression. Effect size and associated confidence intervals were mentioned for different parameters. Model-predicted estimated marginal means were also reported along with the confidence intervals.

All statistical analyses were conducted using R Studio (R Development Core Team, 2022). Survival analyses (Cox proportional hazards models, and AFT models) were implemented using survival, coxme, MASS, and beta regression was implemented using glmmTMB. Mixed-effects modelling was additionally supported by lme4. Model diagnostics, including checks for proportional hazards, model fit, multicollinearity, dispersion, and residual structure, were carried out using the coxme, performance package, and DHARMa. Post-hoc comparisons and estimated marginal means were obtained using emmeans. Visualisations were generated using ggplot2, ggthemes, ggsurvfit, and survminer with high-resolution outputs rendered via Cairo. An alpha level of 0.05 was used for all the analyses conducted.

Hypotheses and predictions

Based on prior work on Indian free-ranging dog behaviour and human-dog interactions, we formulated a set of broad hypotheses and model-specific predictions. These hypotheses guided the structure of all statistical analyses.

H1. Individual traits influence dogs’ initial responses to unfamiliar humans (Day 0)

We predicted that dogs with higher SI scores and residing in areas with higher human flux would approach faster on Day 0, and that approach latency and sociability would not differ between male and female dogs.

H2. Reward type influences dogs’ choice behaviour upon first exposure (Day 1)

We predicted that dogs would be more likely to choose the food provider (FP) than the petting provider (PP) during the first encounter on Day 1.

H3. Repeated exposure across the familiarization phase (Day 1–5) modifies sociability and approach behaviour

We predicted that approach latency would decrease and SI would increase across days as dogs became familiar with the experimenters. We further predicted that individual characteristics (sex, approach latency and SI on Day 0), as well as human flux, reward sequence and the type of reward provided, could influence these measures.

H4. Dogs show stable preferences across days in choice behaviour

We predicted that dogs’ choices would be more consistent than expected by chance across both the familiarization phase (Days 1–5) and the test phase (Days 6–10).

H5. Behaviour during later test days (Day 6–10) reflects both stable individual traits and accumulated familiarity

We predicted that SI and approach latency during Days 6–10 would be positively associated with each dog’s initial sociability (Day 0 SI) and influenced by sex, human flux, the type of reward provider chosen and day. Choice during these days was also expected to be shaped by the same factors.

Model-specific predictions

Each statistical model corresponded to one or more of the hypotheses above:

• Models A1–A2 (Day 0): Test H1 regarding effects of sex, flux, SI, and initial approach behaviour.

• Models B1–B2 (Day 1): Test H2 for first-trial choice behaviour and its predictors.

• Models C1–C6 (Day 1–5): Test H3 and H4 regarding changes in latency/SI in familiarization phase and changes in latency/SI and consistency of choice during test phase.

• Models D1–D4 (Day 6–10): Test H4 and H5 regarding consistency and predictors of behaviour across later test days.

Results

Of the 99 dogs that participated in the approachability test on Day 0, 61 (30 females, 31 males) approached within one body length of E3 and were studied over the following 10 days. Among them, 51 dogs (27 females, 24 males) completed all trials, while 45 (21 females, 24 males) not only completed all trials but also demonstrated a clear preference between the food provider (FP) and petting provider (PP). Only these 45 dogs, who consistently made a clear choice across all 10 trials, were included in the final analysis. Table 2 presents the summary of statistical models and tests used for each experimental day, including outcome variables, predictor variables, model types, and model formulas.

Table 2.

Summary of statistical models and tests used for each experimental day, including outcome variables, predictor variables, model types, and model formulas

Model No.	Day	Outcome variable	Predictor variables	Name of statistical test/model	Formula with abbreviations of the variables used
A1	Day 0	Approach latency on Day 0	Sex, flux, SI on Day 0	Cox proportional hazards model	approach_latency_day0 ~ sex + flux + si_day0
A2	Day 0	New SI on Day 0	Approach latency on Day 0, sex, flux	Beta regression (glmmTMB) model	new_si_day0 ~ approach_latency_day0 + sex + flux
B1	Day 1	Choice	–	Chi-squared goodness-of-fit test	–
B2	Day 1	Choice	Sex, flux, SI displayed towards PP, SI displayed towards FP, sequence in which FP and PP provided reward, side chosen, identity of experimenter chosen	Binomial regression (GLMM) model	choice_day1 ~ sex + flux + si_pp + si_fp + sequence_pp + side + (1|experimenter_chosen)
C1	Day 1–5	Approach latency during familiarization phase	Sex, flux, reward provider involved (FP or PP), SI during familiarization, approach latency on Day 0, day of experiment, sequence in which FP and PP provided reward	AFT model	approach_latency_day1_5 ~ sex + flux + reward_provider + si_day1_5 + approach_latency_day0 + day + sequence_pp
C2	Day 1–5	New SI during familiarization phase	Sex, reward provider involved (FP or PP), sequence in which FP and PP provided reward, flux, SI on Day 0, day of experiment	Beta regression (glmmTMB) model	si_new_day1_5 ~ sex + reward_provider + sequence_pp + flux + si_day0 + day
C3	Day 1–5	Approach latency during test phase	Sex, flux, choice (FP or PP), approach latency on Day 0, day of experiment, SI during test phase	AFT model	approach_latency_test_day1_5 ~ sex + flux + choice + approach_latency_day0 + day + si_test_day1_5
C4	Day 1–5	New SI during test phase	Sex, flux, choice (FP or PP), day of experiment, SI on Day 0	Beta regression (glmmTMB) model	si_new_test_day1_5 ~ sex + flux + choice + day + si_day0
C5	Day 1–5	Consistency of choice	–	Fleiss’ Kappa	–
C6	Day 1–5	Choice	Sex, flux, day of experiment, side chosen, sequence in which FP and PP provided reward, SI displayed towards PP, SI displayed towards FP, identity of experimenter chosen	binomial regression (glmm) model	choice_day1_5 ~ sex + flux + day + side + sequence_pp + si_pp + si_fp + (1|experimenter_chosen)
D1	Day 6–10	Approach latency during test phase	Sex, choice (FP or PP), SI during test phase, flux, approach latency on Day 0, day of experiment	AFT model	approach_latency_day6_10 ~ sex + choice + si_day6_10 + flux + approach_latency_day0 + day
D2	Day 6–10	SI during test phase	Sex, flux, day of experiment, SI on Day 0, choice (FP or PP)	Beta regression (glmmTMB) model	si_new_day6_10_ ~ sex + flux + day + si_day0 + choice
D3	Day 6–10	Consistency of choice	–	Fleiss’ Kappa	–
D4	Day 6–10	Choice	Sex, flux, day, side chosen, identity of experimenter chosen	Binomial regression (GLMM) model	choice ~ sex + flux + day + side + (1|experimenter_chosen)

A. Approachability test (Day 0)

A1. Effect of dog sex, flux and SI on approach latency

On Day 0, the overall approach latency was 11.689 ± 12.901 s. Males approached in 11.083 ± 13.590 s on average, while females took 12.381 ± 12.363 s.

The analysis included 45 observations, all of which recorded an event (i.e., all dogs approached; no censored cases). Assumptions of proportional hazards were tested and met (all p > 0.05; Table S1). None of the predictors significantly influenced approach latency (all p > 0.05; Table S2). The model fit was not significant (Likelihood ratio test: χ²(3) = 1.10, p = 0.80), and the concordance index was 0.573, indicating limited predictive ability.

A2. Effect of dog sex, flux and approach latency on SI

The overall SI on Day 0 was 1.844 ± 1.678. The values for males were 1.333 ± 0.761 and for females were 2.428 ± 2.204. Sex was found to significantly influence SI (β = −0.606, SE = 0.255, p = 0.018, exp(β) = 0.546, 95% CI: 0.334–0.901), with females showing approximately 1.8 times the SI as that of males. Consequently, the model-estimated marginal mean SI for females on the response scale was 0.081 (95% CI: 0.057–0.113), while for males, it was 0.046 (95% CI: 0.030–0.068), reflecting a higher estimated SI for females. However, approach latency and flux were not statistically significant in predicting SI (Table S3). The pseudo R² was 0.089, indicating that fixed effects explained 8.9% of the variance in the response variable. The model’s prediction error, as measured by RMSE was 0.056, indicating a reasonable fit.

B. Familiarization phase followed by test phase (Day 1)

B1. Do FRDs prefer the food or petting provider on day 1?

Results indicated a significant preference for FP compared to PP on Day 1 (χ² = 5.00, p = 0.025). Choice on Day 1 is given in Fig. 2a.

Fig. 2.

A bar graph representing percentage of trials in which food provider (FP) and petting provider (PP) were chosen during the test phase conducted from (a) Day 1 to 5: when the test phase was conducted following the familiarization phase and (b) Day 6 to 10: when only the test phase was conducted

B2. What factors affect the person chosen on day 1?

None of the predictors was significant in determining the person chosen, and the variance due to the random factor (identity of person chosen) was found to be 0 (Table S4).

C. Familiarization phase followed by test phase (Day 1 to 5)

C1. What factors affect approach latency during the familiarization phase from day 1 to 5?

Dogs that did not approach either the PP or FP during Trial 1 of the familiarization phase were censored. First, a Cox proportional hazards model was run. This model violated the proportional hazards assumption (Table S5).

A log-logistic accelerated failure time (AFT) model was then fitted which revealed a significant negative association between the SI (β = −0.043, SE = 0.012, p < 0.001, TR = 0.958, 95% CI: 0.935–0.981) and approach latency, indicating that higher SI values were associated with faster approach times. Similarly, approach latency significantly decreased over time, with Day 2, Day 3, Day 4 and Day 5 (Table 4) showing progressively shorter latencies compared to Day 1. Other predictors, including sex, reward provider, flux, sequence, and approach latency on Day 0, were not significant in determining approach latency during familiarization phase from Day 1 to 5 (Table S6). The estimated scale parameter was 0.336, and the model fit was significant (χ2 = 107.39, df = 10, p < 0.001). These results suggest that socialization and repeated exposure over days strongly influenced approach latency, while other factors had no impact. A Kaplan-Meier survival curve illustrating approach time during familiarization phase from Day 1 to 5 is presented in Fig. 3a.

Table 4.

Results of a beta regression model predicting SI displayed during familiarization phase from day 1 to 5

Term	Estimate	Std. Error	p	exp (estimate)	95% CI
Day 2	0.379	0.123	0.002	1.461	1.147–1.860
Day 3	0.433	0.123	< 0.001	1.542	1.211–1.963
Day 4	0.515	0.122	< 0.001	1.673	1.317–2.127
Day 5	0.349	0.124	0.005	1.418	1.111–1.809

Fig. 3.

Visual summary of the familiarization phase conducted from Day 1 to Day 5: (a) Kaplan–Meier survival curves illustrating the time to approach (seconds). Statistical comparison across days was performed using the log-rank test (p < 0.001), (b) Scatterplot showing the effect of SI displayed on Day 0 on new SI (SI normalized to lie between 0 and 1). The blue line represents the linear regression line, and the blue area depicts its 95% CI, illustrating the uncertainty around the estimated relationship, (c) Effect plot showing the impact of sequence in which reward was given on new SI, (d) Effect plot showing the impact of day of experiment on new SI. The dots represent the model-fitted mean while the whiskers represent the uncertainty (95% CI)

C2. What factors affect SI during the familiarization phase from day 1 to 5?

The beta regression model predicting the SI showed a significant effect of several predictors. Compared to Day 1, dogs displayed approximately 1.5 times more SI on Day 2, 1.5 times more SI on Day 3, 1.7 times more SI on Day 4, and 1.4 times more SI on Day 5 (Table 3). Pairwise comparisons revealed no significant differences among Days 2, 3, 4, and 5 (all p-values > 0.05 after adjustment), suggesting that SI remained relatively stable after the initial increase from Day 1. SI on Day 0 was positively associated with the displayed SI during the familiarization phase (β = 0.093, SE = 0.020, p < 0.001, exp(β) = 1.097, 95% CI: 1.054–1.142), indicating that one-unit increase in SI on Day 0 made dogs approximately 1.1 times more likely to display a higher SI during the familiarization phase. Sequence in which the reward provider visited the dog had a significant influence on SI (β = −0.161, SE = 0.075, p = 0.033, exp(β) = 0.851, 95% CI: 0.734–0.987). SI displayed towards the person visiting the second was 0.8 times as that for the first. Flux, sex and person did not have significant effects (Table S7). The pseudo R² was 0.087, indicating that fixed effects explained 8.7% of the variance in the response variable. Model prediction error was 0.078 RMSE, indicating a reasonable fit. The effects of SI on Day 0, sequence in which the reward was given and the day of experiment is presented in Fig. 3b, c and d respectively.

Table 3.

Results of an accelerated failure time (AFT) model predicting survival time during the familiarization phase and test phase conducted from day 1 to 5

Phase of the experiment	Term	Estimate	Std. Error	p	TR	95% CI
Familiarization phase	Day 2	−0.307	0.091	< 0.001	0.735	0.615–0.879
	Day 3	−0.590	0.087	< 0.001	0.554	0.467–0.658
	Day 4	−0.620	0.087	< 0.001	0.538	0.453–0.638
	Day 5	−0.760	0.087	< 0.001	0.468	0.394–0.554
Test phase	Day 2	−0.429	0.142	0.002	0.651	0.493–0.860
	Day 3	−0.478	0.138	< 0.001	0.620	0.473–0.813
	Day 4	−0.663	0.140	< 0.001	0.515	0.391–0.679
	Day 5	−0.693	0.140	0.005	0.500	0.380–0.658

Consequently, the model-estimated marginal mean new_SI for Day 1 on the response scale was 0.063 (95% CI: 0.053–0.075), for Day 2 was 0.089 (95% CI: 0.077–0.105), for Day 3 was 0.094 (95% CI: 0.081–0.110), for Day 4 was 0.101 (95% CI: 0.087–0.117) and for Day 5 was 0.087 (95% CI: 0.074–0.103). SI displayed for the person visiting first was 0.093 (95% CI: 0.084–0.093), while for the one visiting later was 0.080 (95% CI: 0.072–0.089).

C3. What factors affect approach latency during the test phase from day 1 to 5?

A Cox proportional hazards model was initially fitted. The test for proportional hazards assumption indicated some possibility of violation (Table S8).

A log-logistic accelerated failure time (AFT) model was then fitted which revealed a significant negative association between the SI (β = −0.070, SE = 0.019, p < 0.001, TR = 0.933, 95% CI: 0.897, 0.969) and approach latency, indicating that higher SI values were associated with faster approach times. Similarly, approach latency significantly decreased over time, with Day 2, Day 3, Day 4 and Day 5 showing progressively shorter latencies compared to Day 1 (Table 3). Other predictors, including sex, person chosen, flux and approach latency on Day 0, were not significant in determining approach latency. The estimated scale parameter was 0.372, and the model fit was significant (χ2 = 44.02, df = 9, p < 0.001). These results suggest that socialization and repeated exposure over days strongly influenced approach latency, while other factors had no impact (Table S9). A Kaplan-Meier survival curve illustrating approach time during the test phase from Day 1 to 5 is presented in Fig. 4a.

Fig. 4.

Visual summary of the test phase conducted from Day 6 to Day 10: (a) Kaplan–Meier survival curves illustrating the time to approach (seconds). Statistical comparison across days was performed using the log-rank test (p < 0.001). (b) Scatterplot showing the effect of SI displayed on Day 0 on new SI (SI normalized to lie between 0 and 1). The blue line represents the linear regression line, and the blue area depicts its 95% CI, illustrating the uncertainty around the estimated relationship

C4. What factors affect SI during the test phase from day 1 to 5?

Among the predictors, SI on Day 0 had a significant positive effect (β = 0.096, SE = 0.026, p < 0.001, exp(β) = 1.101, 95% CI: 1.046–1.159), suggesting that an increase in SI on Day 0 was associated with higher si_new. Flux, sex, reward chosen and the day of the experiment did not have significant effects on si_new (Table S10). The pseudo R² was 0.081, indicating that fixed effects explained 8.1% of the variance in the response variable. Model prediction error was 0.074 RMSE, indicating a reasonable fit. The effect of SI on Day 0 is presented in Fig. 4b.

C5. Are dogs consistent in their choice from day 1 to 5?

The agreement was found to be κ = −0.22, indicating lower agreement than expected by chance. A z-score of −1.06 was obtained, with a p-value of 0.29, suggesting that the observed agreement was not statistically significant. This result implies a lack of consistency in choices across the test phase conducted from Day 1 to 5. Figure 2a shows the choice displayed by the dogs from Day 1 to 5.

C6. What factors affect choice from day 1 to 5?

Results indicated higher SI displayed towards PP to significantly decrease the probability of choosing FP (β = −0.151, SE = 0.061, p = 0.013, OR = 0.859, 95% CI: 0.763–0.969). Also, a left side bias was observed (β = −0.758, SE = 0.285, p = 0.008, OR = 0.469, 95% CI: 0.268–0.820). The other predictors did not show significant effects (Table S11). Variance due to the random factor was found to be 0. The Tjur’s R² was 0.075, indicating that fixed effects explained 7.5% of the variance in the response variable. Model prediction error was 0.477 RMSE, indicating a moderate prediction error.

D. Test phase (Day 6 to 10)

D1. What factors affect approach latency during the test phase from day 6 to 10?

A Cox proportional hazards model was initially fitted. The test for proportional hazards assumption showed a violation (Table S12).

A log-logistic accelerated failure time (AFT) model was then fitted which revealed a significant negative association between the SI (β = −0.064, SE = 0.019, p = 0.001, TR = 0.938, 95% CI: 0.903–0.975) and approach latency, indicating that higher SI values were associated with faster approach times. Similarly, approach latency significantly increased over time, however, it was significantly higher only on Day 10 compared to Day 6 (β = 0.469, SE = 0.148, p = 0.002, TR = 1.598, 95% CI: 1.195–2.136). Approach latency on Day 0, sex, person chosen and flux were not significant in determining approach latency (Table S13). The estimated scale parameter was 0.402, and the model fit was significant (χ2 = 26.23, df = 9, p = 0.002). A Kaplan-Meier survival curve illustrating approach time across Day 6 to 10 is presented in Fig. 5a.

Fig. 5.

(a) Kaplan–Meier survival curves illustrating the time to approach (seconds). Statistical comparison across days was performed using the log-rank test (p < 0.001), (b) Effect plot showing the impact of sex of dog on new SI. The dots represent the model-fitted mean while the whiskers represent the uncertainty (95% CI), (c) Scatterplot showing the effect of flux on new SI (SI normalized to lie between 0 and 1), (d) Scatterplot showing the effect of SI displayed on Day 0 on new SI (SI normalized to lie between 0 and 1). The blue line represents the linear regression line, and the blue area depicts its 95% CI, illustrating the uncertainty around the estimated relationship

D2. What factors affect SI during the test phase from day 6 to 10?

Sex had a significant effect, with males having lower SI scores than females (β = −0.323, SE = 0.102, p = 0.002, exp(β) = 0.724, 95% CI: 0.592–0.885). SI on Day 0 was positively associated with SI displayed from Day 6 to 10 (β = 0.109, SE = 0.025, p < 0.001, exp(β) = 1.115, 95% CI: 1.060–1.172), indicating consistency over time. Flux also had a significant negative effect (β = −0.014, SE = 0.006, p = 0.028, exp(β) = 0.986, 95% CI: 0.973–0.998), indicating decreasing values of SI with an increase in flux. However, day and the reward person chosen did not have significant effects (p > 0.05; Table S14). The pseudo R² was 0.124, indicating that fixed effects explained 12.4% of the variance in the response variable. Model prediction error was 0.079 RMSE, indicating a reasonable fit. The effects of sex, flux, and SI on Day 0 are presented in Fig. 5b and c, and 5d, respectively.

The estimated marginal means analysis on the response scale revealed that the SI displayed by females (EMM = 0.128, SE = 0.008, 95% CI: 0.113–0.144) was higher than males (EMM = 0.096, SE = 0.006, 95% CI: 0.084–0.109).

D3. Are dogs consistent in their choice from day 6 to 10?

The agreement was found to be κ = −0.351, indicating lower agreement than expected by chance. A p-value of 0.18 suggests that the observed agreement was not statistically significant. This result implies a lack of consistency in choices across the test phase from Day 6 to 10. Figure 2b shows the choice displayed by the dogs from Day 6 to 10.

D4. What factors affect choice from day 6 to 10?

This model showed no significant effect of any of the predictors (Table S15). However, the variance due to the random factor was found to be 5.384.

Discussion

Our findings show that Indian free-ranging dogs initially preferred the human who provided food, indicating that food serves as a highly salient reward during first-time encounters. However, with repeated interactions, dogs showed comparable preference toward both the food and petting providers. This shift suggests that although food plays a dominant role in shaping early interactions, social reinforcement through petting becomes equally important in sustaining and strengthening dog-human affiliative associations over time.

Attachment in humans is often assessed through proximity-seeking behaviours (Bowlby 1973). A similar concept applies to dogs, where proximity to a human is an indicator of sociability (Barrera et al. 2010). The time taken to approach a human reflects the level of interest in dogs (Bhattacharjee et al. 2017a, b), while their behaviours provide insights into the nature of social associations. In the present study, the approach latency to an unfamiliar person on Day 0 was not influenced by the dog’s sex. However, SI values were affected, with male dogs exhibiting lower SI values than females. The Socialization Index (SI) is a measure of the sociability of the dogs towards humans. Hence, the results indicate that while the initial level of interest in approaching an unfamiliar human does not differ between the sexes, the propensity to engage in social interaction does. This finding aligns with previous studies that have reported sex-based differences in sociability with humans in pet dogs (Lore and Eisenberg 1986; Scandurra et al. 2018).

Flux, a measure of human movement in the dog’s environment, did not influence approach latency or SI toward the unfamiliar human. This contrasts with the findings of Bhattacharjee et al. (2021), who reported greater sociability in dogs from high and intermediate flux areas compared to those from low flux areas. The discrepancy may be attributed to differences in the experimental design: the current study included only dogs that approached the unfamiliar person on Day 0, thereby excluding such individuals who would be very low in sociability. The previous study assessed a mixed group that included both responsive and unresponsive individuals, and obtained a mean response of 34.5% in low flux zones, as opposed to 75.5% and 78.5% in high and intermediate flux zones respectively. Moreover, the lack of correlation between SI and approach latency on Day 0 suggests that a higher level of interest does not necessarily translate into greater sociability toward an unfamiliar human.

When reward preference was assessed, dogs showed an initial preference for the food provider (FP) on Day 1, when both individuals were unfamiliar. This contrasts with Lazzaroni et al. (2020), who reported no preference between food and cuddle providers on Day 1, potentially due to differences in geographic context or reward type. In this study, raw chicken, a rarely provisioned but preferred food among FRDs (Bhadra et al. 2016) was used, possibly explaining the initial preference. From Day 2 onwards, as interactions with both providers increased, the dogs’ choices became random, suggesting that while food served as a strong short-term motivator, repeated positive interactions eventually led to a lack of preference, thereby confirming the importance of petting as a reward for FRDs.

No side preference at the population level was observed on Day 1, but it emerged on subsequent days and influenced choices until Day 5, before disappearing again from Day 6 to 10, when rewards were no longer provided. The cause of this disappearance remains unclear and warrants further investigation. Furthermore, choice between providers from Day 1 to 5 was significantly influenced by the SI displayed toward the petting provider during the familiarization phase, with higher SI values predicting a greater likelihood of choosing her during the choice tests. This suggests that more sociable dogs might have a higher, albeit less skewed, preference for petting as a reward.

Approach latency and SI during the familiarization phase (Days 1–5) showed significant temporal changes. Approach latency decreased while SI increased substantially from Day 1 to Day 2, after which both stabilized. These findings are in line with Nandi et al. (2024), who reported decreased approach latency after a single interaction and increased SI following two interactions. The faster rise in SI in the present study may be due to methodological differences. Notably, approach latency during familiarization was not predicted by latency toward the unfamiliar person on Day 0. However, SI during familiarization was positively correlated with SI toward the unfamiliar person on Day 0, suggesting that SI may be a more stable behavioural trait. The sequence of visits by the reward providers did not affect approach latency, but SI was lower for the second visitor, regardless of the reward type, possibly reflecting reduced enthusiasm during subsequent interactions.

Dogs with higher SI demonstrated shorter approach latencies, reinforcing the link between sociability and approach motivation. This, however, contradicts Day 0 findings, where SI and approach latency were not negatively associated. This is likely due to the familiarity the dogs developed with the experimenters on Day 1, in contrast to Day 0 when E3 was still unfamiliar, and/or the effects of positive reinforcement. Neither sex nor flux influenced approach latency or SI during this phase, diverging from Day 0 results where males were found to be less sociable, reinforcing the suggestion of familiarity and rewards influencing latency. Furthermore, no differences in approach latency or SI were observed between the two reward providers, indicating comparable motivation to engage with both.

During the test phase (Days 1–5), when choice tests followed the familiarization phases, approach latency continued to decrease from Day 1 to Day 2, following trends seen during familiarization. However, SI did not change significantly across these days, potentially due to a decline on Day 1 (immediately after the familiarization sessions), leading to a level of saturation.

From Days 6 to 10, when no rewards were offered and choices were based on memory, dogs maintained low approach latencies, except on Day 10, when latency increased significantly. This suggests a gradual reduction in motivation in the absence of rewards, a pattern consistent with reward extinction (Skinner 1938). Despite this, SI remained high throughout, indicating a persistent tendency to engage with familiar humans. This is consistent with anecdotal reports of Indian FRDs maintaining affiliative behaviours toward familiar individuals even after long periods without interaction or reinforcement. Furthermore, from Days 6–10, the identity of the person was found to influence the dogs’ choices, as indicated by a variance of 5.8, suggesting differential preferences toward the experimenters despite efforts to standardize their appearance and calling procedure.

Male dogs continued to show lower SI than females during Day 6–10, in line with earlier observations. Interestingly, flux emerged as a significant negative predictor of SI during this reward-absent phase, dogs from lower flux areas exhibited higher SI. Bhattacharjee et al. (2021) found that dogs from low-flux areas were more anxious around unfamiliar humans. In this study, the experimenters were no longer unfamiliar to the dogs in the reward-absent phase. Also, as mentioned earlier, we included only those dogs that responded to the experimenters on Day 0, thereby considering a pool of bolder and more sociable individuals. Although flux did not affect SI on Day 0, its influence during the reward-absent phase suggests that dogs from areas with less human activity may have a greater drive for social contact, but may be more reluctant to approach unfamiliar humans. This would lead to more sustained affiliative behaviour in such dogs once familiarized.

It is known that some free-ranging dogs may not prefer to be petted on the head or neck. However, because we included only those individuals that voluntarily approached within 1 body length of E1 on Day 0, our sample inherently consisted of the bolder or more sociable dogs. Among these selected individuals, none displayed hesitation, fear, anxiety, or aggression in response to petting on any subsequent day of the experiment, indicating that the tactile interaction used in our study was well tolerated by all participating dogs.

Overall, dogs did not exhibit a consistent preference for either reward provider across the study. This inconsistency may reflect several factors: the increasing equivalence of reward value over time, habituation to the experimental setup, the influence of side preference, and limited discrimination between the two providers. Nevertheless, the initial strong preference for the food provider highlights the powerful short-term motivational role of food. The observed pattern resembles a serial reversal task, in which subjects begin to respond equally to multiple stimuli after repeated exposures; in this context, because both options continued to provide reinforcement, random choice may have required less cognitive effort.

Our findings contrast with Bhattacharjee et al. (2017a, b), who reported that FRDs preferred petting over food in building trust during repeated interactions with unfamiliar humans. This discrepancy likely reflects differences in experimental paradigms. In the earlier study, food was the primary reward and could be accessed either from the experimenter’s hand or the ground, with petting or food given as an additional motivator, prior to the choice test. The decision of the dogs to eat from the hand was considered at trust. In contrast, this study used two distinct experimenters providing separate rewards, presenting a different context for reward-based preference. The choice of the dog was determined by its approach towards an experimenter within one body length and gazing towards the experimenter, which would require a higher level of motivation and sociability. There was no requirement to make actual contact with the experimenters in this paradigm. Further, the earlier study was conducted in separate sets of dogs, and the individual dogs did not have a choice between the two kinds of rewards. Hence, it reflects the population-level response to petting in FRDs, while this study tests the hierarchy of preference of individual dogs for reward givers of two types.

It is important to note that all experiments were conducted by trained experimenters experienced in handling FRDs, and who had been previously vaccinated against rabies. Given that rabies is a common and fatal zoonotic disease in India and other regions of the Global South where FRDs are prevalent, prior vaccination of experimenters is a critical ethical and safety consideration when conducting such studies.

In conclusion, this study indicates that Indian FRDs initially prefer food from unfamiliar humans but, over repeated interactions, value both food and petting rewards equally, a pattern observed primarily among highly sociable and bold individuals, highlighting the importance of social rewards in the co-existence of humans and dogs. This offers valuable insights into their social behaviour, underscoring the nuanced interplay between immediate and long-term reinforcement in shaping social preferences. Future research should aim to explore additional ecological and individual-level factors that modulate reward preferences in dog-human interactions.

Significance statement

This study is the first to investigate the preference between humans providing food and social rewards (petting) in FRDs, with a focus on how these preferences change over time. While earlier research has explored reward preferences in dogs, such studies have primarily been limited to short-term assessments or conducted in controlled environments where food was the sole reward. Most existing work has involved companion dogs in human-managed settings, thereby limiting its relevance to understanding the behaviour of FRDs, who live independently of direct human ownership. By observing these dogs in their natural environment, our study provides ecologically valid insights into how they value different forms of human interaction and how repeated exposure influences these preferences. It challenges the assumption that food is universally the primary motivator in dog behaviour and highlights the potential value of social bonding even among non-owned dogs.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Srijaya Nandi: Conceptualization, Methodology, Formal analysis, Investigation, Data curation, Writing – Original Draft, Visualization, Project administration. Aesha Lahiri: Investigation. Tuhin Subhra Pal: Investigation. Anamitra Roy: Investigation. Rittika Bairagya: Investigation. Anindita Bhadra: Conceptualization, Methodology, Resources, Writing – Review & Editing, Supervision, Funding acquisition.

Funding

This research was supported by the Department of Science and Technology, Ministry of Science and Technology (INSPIRE fellowship to S.N.) and the Janaki Ammal – National Women Bioscientist Award (BT/HRD/NBA-NWB/39/2020-21 (YC-1)), Department of Biotechnology, India. PhD fellowship was granted to T.S.P. by UGC.

Data availability

Supplementary data associated with this article can be found in the online version at doi: 10.17632/ykknkytdth.1.

Declarations

Conflict of interest

The authors declare that they have no conflict of interest.

Ethical approval

The study design did not violate the Animal Ethics regulations of the Government of India (Prevention of Cruelty to Animals Act 1960, Amendment 1982). The experimental protocol was approved by the IISER Kolkata Animal Ethics Committee, as part of a larger project sanctioned by the SERB (EMR/2016/000595).