What if the reward is not as yummy? Study of the effects of Successive Negative Contrast in domestic dogs in two different tasks

Abstract

Successive Negative Contrast (SNC) occurs when there is a reduction in the quantity or quality of a reward that is expected according to the presence of contextual cues. This induces an emotional response of frustration that is similar to stress. While this phenomenon has been observed in several mammal species, findings in domestic dogs have been inconsistent, although this issue has strong relevance in dog training. The aim of this study was to assess the effects of Successive Negative Contrast in two responses that had already been studied in this species, but with an increase in the methodological rigor and variations in the experimental conditions to examine the generalizability of the phenomenon. To this end, experimental dogs experienced a pre-shift phase in which they received a high-value reward (liver), followed by a post-shift phase in which they obtained a low-value reward (dry dog food), and then a re-shift phase in which the high-value reward was available again. Control dogs received dry food in all phases. The results show a contrast effect on the behavior of following human pointing to obtain food (Study 1). On the contrary, there were no differences in problem solving behavior after the de- and re-evaluation of the reward during a non-social task (Study 2). The results support that Successive Negative Contrast is not a consistent phenomenon in pet dogs. It is possible that certain characteristics of dogs such as the great availability of high-value rewards in their daily lives could attenuate the effects of a reduction in incentive value.

Introduction

Frustration is a negative emotional state that occurs when an expected reward is lost or diminished (e.g., Papini et al., 2006). In dogs, this negative state has been correlated with physiological and behavioral stress responses such as an increase in cortisol, vocalizations and lunging (McPeake et al., 2021). Moreover, frustration has been associated with negative facial expressions such as ears flattened, blinking, parted lips, jaw-dropping, nose licking, and behavioral displays such as looking away upon being denied an expected food reward or toy (Bremhorst et al., 2019; 2021; Pedretti et al., 2022). Furthermore, ears flattened, blinking, and nose licking were associated with the presence of an audience, suggesting these behaviors have a communicative function (Pedretti et al., 2022).

One of the most commonly used protocols to study frustration in animal models is Successive Negative Contrast (SNC). This protocol comprises a pre-shift phase in which the experimental group receives a high-value reward, followed by a post-shift phase in which they unexpectedly receive a low-value reward for doing the same task. The performance of the experimental group is compared to a control group which has always received low-value rewards. Due to the unexpected devaluation in the amount or the quality of the reward, the operant responses of the experimental group often drop to levels below those of the control group (Papini, 2006). In this sense, SNC shows that animals adjust their behavior not only according to the absolute value of reinforcers but they also consider their relative value according to previous experiences (Amsel and Penick, 1962).

The phenomenon of SNC has been extensively observed in several mammalian species (e.g., marsupials, rats, mice, Mongolian gerbils, sheep, deer, monkeys, chimpanzees and humans) but has not been found in non-mammalian vertebrates (e.g., fish, reptiles, birds and amphibians, although exceptions have been observed in starlings and bees; for a review see Papini, 2014).

Studying SNC in dogs is of great relevance, as they often receive different qualities and quantities of rewards from their owners during training sessions, which could affect their behavior and performance. This is particularly relevant for dogs that receive advanced training, such as working dogs or those participating in competitive sports. These studies provide valuable information regarding the strength of the SNC effect in domestic dogs, according to its occurrence in different experimental situations. Furthermore, the SNC phenomenon has been suggested as an indicator of animal welfare, as it has been shown to increase with poor welfare (Burman et al., 2008). For instance, unenriched rats showed a more prolonged SNC response to reward loss than enriched rats, suggesting a relationship between poor welfare and SNC (Burman et al., 2008).

Surprisingly, prior studies of SNC in domestic dogs have yielded contradictory results, both in the case of non-social and social tasks in which there is interaction with a person (Bentosela et al., 2009; Pongrácz et al., 2013; Riemer et al., 2016; Riemer et al., 2018a,b; Dzik et al., 2019; see Table 3).

Table 3. Summary of the previous SNC studies.

	Evidence of SNC	High/low value reward	Kind of task	Presence of the owner	Testing place
Bentosela et al. (2009)	Yes	Liver/regular dry food	Social	No	Familiar
Riemer et al. (2018b)	Yes	Dog treat/Novel dry food	Non-social	Yes	Unfamiliar
Dzik et al. (2019)	Yes	Liver/regular dry food	Non-social	No	Familiar
	No	Sausage/regular dry food	Non-social	No	Familiar
Pongrácz et al. (2013)	No	Sausage/novel dry food	Social	Yes	Unfamiliar
Riemer et al. (2016)	No	Sausage/novel dry food	Social	-	Familiar
Riemer et al. (2018a)	No	5 pieces/1 piece of novel dry food	Non-social	No	Unfamiliar
	No	Sausage/novel dry food	Non-social	No	Unfamiliar

Regarding non-social tasks, Riemer et al. (2018a) found no evidence of SNC when they measured the time it took dogs to complete a runway to obtain a food reward after the reward received decreased in both quality and quantity. Conversely, Riemer et al. (2018b) found SNC in dogs during a foraging task in which food was available on four boards. Dogs experienced an unshifted condition in which there was a low-value reward in all trials and a shifted condition in which the reward value changed between high and low. After a reward downshift, dogs switched boards more frequently. Similar findings were observed in a problem solving task using a commercial toy in which dogs had to remove plastic bones to obtain food hidden underneath. The dogs that received a decrease in the quality of the reward picked up fewer plastic bones than dogs from the control group, exhibiting SNC (Dzik et al., 2019).

Social tests, including human interactions, also yielded contradictory results. Bentosela et al. (2009) found that dogs gazed less towards an unfamiliar human to request inaccessible food and exhibited a higher rate of food refusals when they experienced a devaluation in the quality of the reward compared to a control group that remained unchanged. Nevertheless, Riemer et al. (2016) could not replicate these findings using a similar protocol. Finally, Pongrácz et al. (2013) examined another communicative response, point following to find hidden food, employing distal momentary pointing cues. The results showed that dogs did not modify their behavior after a decrease in the reward quality. However, it is important to consider that this study did not include a control group that remained unchanged, which hinders the interpretation of the results.

Considering this conflicting evidence, Riemer et al. (2018a) suggested that dogs may be less sensitive to decreases in reinforcement value compared to other mammalian species. However, as shown above, the SNC phenomenon has been observed in dogs in some circumstances (Bentosela et al., 2009; Riemer et al., 2018b; Dzik et al., 2019). Several factors could underlie these contradictory findings, such as methodological differences (e.g., the type of rewards selected for each task, Dzik et al., 2019), the fact that dogs have not been assessed under controlled laboratory conditions as it has been done in other species (Riemer et al., 2016), their living environment which is more enriched than that of captive or laboratory animals (Burman et al., 2008), ecological factors such as the scavenger lifestyle of dogs (Pongrácz et al., 2013), the exposure to schedules of intermittent reinforcement during their lives (Amsel and Penick, 1962), the presence or absence of the owner during the task and the social component of the situation which may act as an alternative source of reinforcement or make dogs less sensitive to the downshift in food quality (Pongrácz et al., 2013).

Taking these findings into account, the general aim of this study is to assess the presence of SNC in domestic dogs in both a non-social and a social task to examine the strength of this phenomenon and increase the methodological rigor of previous studies. In both situations, dogs from the experimental group received a pre-shift phase in which a high-value (liver) reward was available and a post-shift phase in which they received a less valuable option (dry dog food). Then, they received a re-shift phase in which the high-value reward was available again to control for the effects of satiety and fatigue (Bentosela et al., 2009). Meanwhile, dogs from the control group always received the low-value reward (dry dog food). Modifying the conditions as well as increasing experimental rigor allows for assessing the replicability of this phenomenon to examine the robustness of SNC in domestic dogs.

Study 1 (social task) aimed to examine the effect of SNC on point following, a social behavior of great relevance for dogs. Point following is often assessed using an object choice task in which there are two opaque bowls, and dogs must follow the human pointing cue to find which one contains food (Soproni et al., 2001). In this study, the protocol was similar to that of Pongrácz et al. (2013) with the following changes to increase the chance of observing SNC: the pointing cue was proximal static instead of distal momentary; a control group that did not experience changes in the reward quality was included; a higher number of training trials were used; and we added a re-shift phase in which dogs received the high-quality reward again to discard effects of satiety and fatigue (Bentosela et al., 2009). In this case, it is not possible to make a precise prediction of the results since SNC has not been observed during a point following task after a decrease in reward value (Pongrácz et al., 2013).

Study 2 (non-social task) aimed to replicate the SNC effect observed in Dzik et al. (2019), employing the same problem solving task but including a series of methodological changes, such as the dogs being evaluated with their owners present and in a novel place instead of their home. These changes made the testing condition more similar to those carried out by Riemer et al. (2018b). Therefore, if SNC was a robust phenomenon in dogs, similar findings to that of Riemer et al. (2018b) and Dzik et al. (2019) should be expected.

These studies will add to the knowledge about SNC in dogs, a species in which the SNC phenomenon seems to have particular characteristics that differ from what has been observed in other mammals. This becomes even more relevant considering that dogs are exposed to unexpected changes in the quality and quantity of rewards both in their daily lives and during training sessions, which could entail frustration effects that might counteract their performance.

Study 1: Social task, point following

Ethics approval

Study 1 was run in Argentina, and it complied with the current Argentinean law of animal protection (Law 14.346). The procedure was in accordance with the ethical standards and the approval of the CICUAL (Institutional Commission for the Care and Use of Laboratory Animals) from the Medical Research Institute IDIM CONICET (Res. N° 100-18). All owners expressed their written consent for participation in this study.

Method

Subjects

We assessed 55 domestic dogs. However, 22 dogs had to be eliminated from the study as they were not interested in the task, were fearful of the experimenters or showed signs of stress when the owner left the experimental set. Hence, the final sample consisted of 33 dogs, which were semi-randomly assigned to one of the two possible groups. Groups were matched, as far as possible, in regards to sex, age and FCI breed group.

The experimental group consisted of 16 dogs, 7 females and 9 males, ranging from 9 months to 9 years old (mean age ± SD = 3.89 ± 2.49 years old); 10 mixed breeds, 2 Poodles, 1 Chihuahua, 1 Chow Chow, 1 Labrador Retriever, 1 Maltese.

The control group consisted of 17 dogs, 8 females and 9 males, ranging from 2 to 10 years old (mean age ± SD = 5.4 ± 2.61 years old); 10 mixed breeds, 3 Labrador Retrievers, 1 Boston Terrier, 1 Dachshund, 1 German Shepherd, 1 Poodle.

All dogs were healthy and had no owner-reported aggression or excessive fearfulness towards humans. They were recruited through flyers on social media and personal contacts. The owners were asked to refrain from feeding them for 4 h prior to the test while water was available ad libitum.

Materials

Dogs were evaluated in a quiet room, either within their homes or in a dog daycare familiar to them. The situation was filmed with a Sony DCR-SR 88 camera placed on a tripod in a corner of the room. During the task, one experimenter (E) was present who gave the pointing cues, and one handler (H) held the dog by the leash.

Two opaque bowls (base diameter 9 cm, diameter of the opening 23 cm, depth 10 cm) were used for the task. To control for odor cues, both bowls were fitted with double bottoms in which pieces of the reward could be hidden out of reach from the dog. The reward that was hidden in the bottom was the same that was being used in each phase. Different bowls were used for the high (liver) and low (dry food) value rewards. A series of holes in the top bowl facilitated the liberation of the smell.

According to the size of the dog, the bowls were placed on two chairs of approximately 50 cm in height at the seat level or, in the case of small dogs (<30 cm at the shoulders), two circular polystyrene platforms (diameter 24 cm, height 10 cm). The chairs/platforms were placed 1 m apart from each other's center, in between which the E stood during trials. The starting point for the dog was delimited between 1.25 m and 1.5 m from where the E stood (see Figure 1a).

Figure 1. Experimental set up of Study 1 (a) and Study 2 (b).

Small pieces of cooked liver were used as a high-value reward, while pieces of the dry food the dogs usually consumed were used as a low-value reward. These rewards were chosen because they were the most and least preferred on a preference test by Bentosela et al. (2008), and prior studies found SNC in dogs using these rewards on two different tasks (Bentosela et al., 2008; Dzik et al., 2019).

Procedure

Dogs were divided into two groups: dogs in the experimental group received a high-value reward (liver) during the pre-training, pre-shift and re-shift phases, but they received a low-value reward (dry food) during the post-shift phase (see the description of each phase below). Dogs in the control group always received the low-value reward (dry food).

The test consisted of an object choice task in which only one of the bowls contained the reward, so the dog had to follow a pointing cue in order to find it. All pointing cues were proximal (i.e., the E's finger was at a distance of 10 cm from the bowl) and static (i.e., the cue was maintained until the dog made a choice or the time ran out). The E always gazed at the dog and avoided directing her gaze to either bowl. The pointed side was semi-randomized across trials, with no more than 2 consecutive trials on the same side.

Dogs were free to explore the room for approximately 3 min to get familiarized with the situation. The task began immediately afterwards and comprised 4 phases:

Warm-up: This phase comprised 1 session of 2 trials and had the aim of showing the dog that the bowls could contain food. After H took the dog to the starting point, E walked in a straight line carrying one bowl in each hand. Once E reached the middle of the chairs/platforms, she turned and placed one bowl on each one and stayed standing/kneeling there. Both bowls contained a piece of reward (liver in the experimental group, dry food in the control). The H guided the dog by the leash to one bowl and allowed them to eat while effusively saying "very good!", and then repeated this for the other side. Once the dog ate from both bowls, E picked them up and left. Another pre-training trial began immediately. In total, the dog ate two times from each side, and the side which was approached first was counterbalanced across trials.

Pre-shift: This phase started immediately after the warm-up and comprised 2 sessions of 10 trials each, with an intertrial interval of 20 s and an intersession interval of 2 min. After the H took the dog to the starting point, E walked in a straight line carrying one bowl in each hand. Once E reached the middle of the chairs/platforms, she turned and placed one bowl on each one while standing/kneeling. Only one bowl contained a piece of reward (liver in the experimental group, dry food in the control). Then she called the dog's name and, after making eye contact with the dog, she pointed towards the baited bowl. Then the H released the dog, letting it free to choose.

Post-shift: There was a 25 min interval between phases to avoid an effect of sensory carry-over across phases (i.e., the sensorial properties of the reward used in the previous phase affecting the perception of the characteristics of the new reward). This phase comprised 2 sessions of 10 trials each, with an intertrial interval of 20 s and an intersession interval of 2 min. The procedure was similar to the pre-shift, except the reward was dry food for both groups.

Re-shift: This phase occurred 20 s after the last post-shift trial and comprised 4 trials with an intertrial interval of 20 s. These trials were similar to the pre-shift phase (dogs received liver in the experimental group and dry food in the control), in order to discard the effects of satiety or fatigue.

A choice was scored when the dog approached a bowl with its muzzle at less than 10 cm. If it was the correct one, the dog was allowed to eat, and the H said, "very good!" in a positive tone while using the leash to return the dog to the starting point. Dogs were not allowed to approach the other bowl. If the dog chose incorrectly, the H said "no", while the E showed that the bowl was empty and that the reward was in the other one, allowing the dog to see but not eat it. If the dog did not make a choice within 15 s from the moment the pointing gesture started, it was registered as a no choice. If the dog made 4 consecutive no choices during the post-shift phase, that phase was interrupted, and the re-shift phase started.

Analyzed variables

The dependent variables were the number of correct, incorrect and no choices during the pre-shift, post-shift and re-shift phases, as well as the number of rewards eaten. In addition, the latency to choose was analyzed for the last 10 pre-shift trials and all post-shift and re-shift trials. Latency was measured from the moment the E emitted the pointing cue until the dog was less than 10 cm from one of the bowls. In the case of “no choices”, the maximum latency was 15 s, and these cases were coded as censored events.

Statistical analysis

R studio (https://www.rstudio.com/) was used for the statistical analysis. For analyzing the choice latency, we used mixed effects cox regression (coxme) and identified a parsimonious model with AIC-based backwards elimination. 'No choice' was treated as censored events with maximum latency (15 s), and terminated trials were treated as missing values. Tukey post-hoc test was applied for pairwise comparisons.

Regarding the choice success, we applied Generalized Linear Mixed effects model with AIC-based backwards elimination to identify a parsimonious model. Again, Tukey post-hoc test was used for pairwise comparisons.

In both analyses, in the initial model condition, phase, condition x phase interaction, age, size, sex, neutered status, sex x neutered status interaction as fixed effects, and dog name as random intercept were included.

Results

Choice latency

In the final model, we found a significant interaction between phase and condition (LR test: χ2(3) = 28.203; p < 0.001). According to the post-hoc test, there was no difference between phases in the control group, while in the test group, dogs chose more slowly in post-shift phases (when they received dry food instead of liver) than in the pre- and re-shift phases (see Table 2, Figure 2).

Table 2. Comparison of choice latency for the two groups (control and experimental) between the phases of Study 1 (social task-pointing test). Tukey post-hoc tests.

Control						Experimental
Contrast	Odds ratio	-2.5% CI	+2.5 CI	z ratio	p value	Odds ratio	-2.5% CI	+2.5 CI	z ratio	p value
Pre2 / Post1	0.970	0.710	1.326	-0.248	0.995	1.338	0.890	2.011	1.834	0.257
Pre2 / Post2	1.121	0.823	1.528	0.950	0.778	2.415	1.578	3.696	5.323	0.000
Pre2 / Re	1.258	0.842	1.878	1.469	0.457	0.742	0.450	1.224	-1.532	0.418
Post1 / Post2	1.156	0.847	1.577	1.195	0.630	1.805	1.202	2.711	3.734	0.001
Post1 / Re	1.296	0.865	1.943	1.648	0.352	0.555	0.334	0.922	-2.982	0.015
Post2 / Re	1.122	0.748	1.682	0.728	0.886	0.307	0.182	0.518	-5.796	0.000

Note: Pre 2: session 2 of the pre-shift phase. Post1: session 1 of the Post-shift phase. Post2: session 2 of the Post-shift phase. Re: Re-shift.

Figure 2.

Survival plots for the latency of choice in the control (ctrl) and experimental (exp) groups of Study 1 (point following). Different phases are indicated by different colors. Pre 2: session 2 of the Pre-shift phase. Post1: session 1 of the Post-shift phase. Post2: session 2 of the Post-shift phase. Re: Re-shift. Shading around the lines shows 95% confidence interval. In the control group, there was no difference between the phases. In the test group, dogs chose more slowly in the two post-shift phases.

Choice success

None of the examined variables had an effect.

The null model showed that the dogs chose the correct bowl with a 9.56 times higher chance (β = 2.26+/-0.14; z = 15.82; p < 0.001).

Discussion

The aim of this study was to examine the effects of SNC on point following, increasing the methodological rigor of Pongrácz et al. (2013)’s study. To this end, we followed some of the parameters used in Bentosela et al. (2009)’s study, in which SNC was observed in gazing behavior. Our pointing task comprised three phases (pre, post and re-shift), the same high and low-value rewards were used (liver and the dry food the dogs usually consumed), and a control group was included.

The results show that dogs from the experimental group (i.e., those who experienced a devaluation in the reward) were faster to choose a bowl in the pre and re-shift blocks, while in the two post-shift blocks, they chose significantly slower. Conversely, there were no differences between blocks for the control group. This result suggests that dogs experienced SNC during the post-shift phase. An increase in latency towards the devaluated reward has also been considered in previous studies as an indicator of the phenomenon of contrast (Amsel and Penick, 1962; but see Riemer et al. 2018a,b). These differences could be due to the discrepancy in incentive value that dogs could perceive when consuming the rewards. An alternative explanation, although unlikely, could be that the dry food used as a low-value option was not as odorous as liver, so dogs could have been slower to notice the reward and respond in this case.

These results differ from those of Pongrácz et al. (2013), who found that dogs were not sensitive to changes in the appetitive value of a reward during a point-following task. A series of methodological differences could explain this discrepancy. First, Pongrácz et al. (2013) did not include a control group that always received the low-value reward. The lack of a control group precludes concluding the presence of SNC, as the definition of the phenomenon implies that experimental animals show an abrupt decrease in their responses after the reward is down-shifted, which falls below the performance rate of the control group. Conversely, in species or individuals who do not show SNC, the responses of down-shifted experimental animals gradually adjust to that of the control group (Flaherty, 1982). Moreover, having a control group allows to discard alternative explanations for the behaviour of the experimental group during the down-shift phase, such as fatigue, satiety or boredom. On the other hand, the high-value rewards were different from the ones used in the present study (sausage vs liver), and as Dzik et al. (2019) stated, the magnitude of the discrepancy in incentive value of the high and low-value rewards is crucial for the SNC phenomenon to appear (see Papini, 2022).

On the other hand, these results are in line with prior findings featuring gazing towards the human face, which is another communicative response of dogs that was studied using similar experimental parameters to that of the present study (Bentosela et al., 2009). This reveals the flexibility of dogs’ social skills upon brief changes in the environmental contingencies, which has also been shown with the complete omission of an expected reward in an extinction protocol (Bentosela et al., 2008).

Nevertheless, it is important to consider that the pointing task showed no group differences in the other analyzed variables, particularly in the number of correct responses and the number of no-choices. As it was previously mentioned, this could be attributed to point following being a preponderant response in dogs. In this sense, dogs may have been selected during domestication, among other factors, for their ability to follow human social cues (e.g., Hare and Tomasello, 1998; 2005). Furthermore, point following is a response that is usually reinforced in the lives of dogs since early ontogeny. As such, it is a fruitful behavior which leads them to obtain a great quantity and variety of reinforcers in their everyday life (Bentosela and Mustaca, 2007). Moreover, as prior evidence shows, dogs are highly persistent in their point following behavior, even when this response is not successful (e.g., Szetei et al., 2003; Elgier et al., 2009; Kundey et al., 2010). However, despite this being a preponderant response, prior findings also indicate that dogs are able to stop following human pointing cues that do not lead to a reward (Elgier et al., 2009; Kundey et al., 2010), even choosing the opposite container to the one that is being pointed (Elgier et al., 2009), or following non-social cues that are opposed to the pointing one (Elgier et al., 2012). Taking this into account, a change in latency towards the pointed bowl could be a simpler and earlier manifestation of the SNC phenomenon, and more training trials could be needed to observe greater changes in point following behavior according to a reduction in reward value.

In sum, the present findings indicate an SNC effect in the point following behavior of dogs. Specifically, experimental dogs were slower to choose the pointed bowl after they experienced an unexpected reduction in the reward value. This was not observed in control dogs, who always received a low-value reward.

Study 2: Non-social task, problem solving

Ethics approval

Study 2 was run in Hungary. Ethical approval was obtained through the National Animal Experimentation Ethics Committee of Hungary (PE/EA/2019-5/2017). Owners completed a written consent form, which permitted them to volunteer their dogs to participate in the study.

Method

Subjects

We assessed 28 pet dogs. However, 7 dogs had to be eliminated from the study as they were not interested in the task (see Procedure). Hence, the final sample consisted of 21 dogs, which were semi-randomly assigned to one of the two possible groups. Groups were matched, as far as possible, in regards to sex, age and FCI breed group.

The experimental group consisted of 14 dogs, 8 females and 6 males, ranging from 2 to 12 years old (mean age ± SD = 5.81 ± 3.33 years old); 5 mixed breeds, 2 Golden Retrievers, 1 Australian Shepherd, 1 Beagle, 1 Border Collie, 1 Vizsla, 1 Newfoundland, 1 Schnauzer, 1 Shetland Sheepdog.

The control group consisted of 7 dogs, 5 females and 2 males, ranging from 3 to 11 years old (mean age ± SD = 7.85 ± 2.96 years old); 3 mixed breeds, 1 Belgian Shepherd, 1 Labrador Retriever, 1 Poodle and 1 Puli.

All dogs were healthy and had no owner-reported aggression or excessive fearfulness towards humans. They were recruited from the Family Dog Project database of Eötvös Loránd University. The owners were asked to refrain from feeding them for 4 hs prior to the test while water was available ad libitum.

Materials

Dogs were evaluated in a 3 x 6 m² unfamiliar room at the Department of Ethology, Eötvös Loránd University. The situation was filmed with two wall-mounted cameras. During the task, the owner was present, sitting in a chair while ignoring the dog. Tape marks 2 m in front of the chair indicated the starting point for the dog.

The apparatus was a Nina Ottosson© Dog Magic interactive toy (see Figure 1b), which consisted of a round base of 36 cm in diameter, with nine bone-shaped depressions containing nine removable plastic bones (eight arranged in a circle, and the ninth located in the middle). All bones had a small hole to release the smell of the reward hidden underneath. There was one reward hidden underneath each bone so that dogs could eat a maximum of 9 rewards during each trial.

The apparatus was located in the middle of the experimental room and was fixed to the floor with adhesive gum to prevent it from slipping around. A perimeter was marked at 1 m around the toy to determine the dog's proximity to the toy later.

Small pieces of cooked liver were used as a high-value reward, while pieces of dry food were used as a low-value reward. In this case, it was not possible to obtain the food each dog consumed (as not all dogs consumed dry food), so a popular commercial brand of dry food (Purina Pro Plan) was used.

Procedure

Dogs in the experimental group received a high-value reward (liver) during the pre-shift and re-shift phases, but they received a low-value reward (dry food) during the post-shift phase, see the description of each phase below. Dogs in the control group always received the low-value reward (dry food).

During the trials, the owner was asked to sit on the chair and hold the dog by the collar to keep the dog at the starting point, which was 2 m away from the place of the toy. Then the E entered the room carrying the baited toy, put it down and left the room. Immediately after the E left, the owner released the dog, so the dog was free to interact with the toy, and the trial began. Once the trial ended, the E came back and took the toy to the other room to wipe it and refill it. All trials lasted until the dog removed all the bones or a maximum of 3 min. The intertrial interval was approximately 30 s, in which the toy was taken away and refilled in another room out of the dog's view. The same procedure was repeated for all trials. The toy was washed at the end of each phase.

Before the first trial, dogs were free to explore the room for approximately 2 min to get familiarized with the situation. The task began immediately afterwards and comprised 4 phases:

Warm up: For the first trial, the toy had 3 bones partially removed to help solve the task (facilitation trial). If during the first trial, the dog did not eat at least 7 of the 9 rewards in the available 3 min, a second facilitation trial was included. If in this second trial, the dog again did not eat at least 7 of the 9 rewards, the dog was excluded from the sample due to lack of motivation.

Pre-shift: Dogs in the experimental group received a pre-shift phase of 4 trials in which the toy contained high-value reward (liver).

Interval: A 10-min interval was included to avoid an effect of sensorial carry-over (i.e., the sensorial properties of the reward used in the previous phase affecting the perception of the characteristics of the new reward). across the phases. Unlike Study 1 and Dzik et al. (2019), in which the interval between phases was of 25 min, we decided to use a shorter interval as some authors did not include an interval at all (Riemer et al., 2016).

Post-shift: For the experimenter group, the post-shift phase consisted of 4 trials in which the toy contained the low-value reward (dry food).

Re-shift: 1 trial in which the toy contained again the high-value reward to control for satiety or fatigue.

The procedure for the control group was similar, except that the toy contained low-value reward (dry food) in all trials.

Analyzed variables

The dependent variables were the number of picked-up bones, the number of rewards eaten, and the latency to picking all nine bones. If the dog did not pick up all the bones, a 3 min latency was registered.

Statistical analysis

The statistical analysis was similar to Study 1.

Results

Regarding the latency to picking the bones, we did not find differences between the conditions, only an effect of the phase (LR test: χ2(2) = 17.576; p < 0.001). According to the post-hoc test, in re-shift phase, dogs solve the toy significantly faster than in pre (exp(β) [95%CI] = 0.293 [0.156 - 0.551]; z = -4.557; p < 0.001) and post-shift phases (exp(β) [95%CI] = 0.380 [0.204 - 0.708]; z=-3.645; p < 0.001), independently from the condition (see Figure 3).

Figure 3.

Survival plots for the latency to picking all the bones in the control (ctrl) and experimental (exp) groups of Study 2 (problem solving). Different phases are indicated by different colors. Pre: Pre-shift, Post: Post-shift and Re: Re-shift phases. Shading around the lines shows 95% confidence interval. Dogs were faster in the Re-shift phase.

There were no significant differences in the number of picked-up bones and in the number of rewards eaten (see Figure 4).

Figure 4.

Number of picked-up bones (max. 9) and rewards eaten (max. 9) in Study 2 (problem solving) in the control (CTR) and experimental (EXP) groups. Different phases are indicated with different colors. Dots show the number of bones picked up (A) or number of rewards eaten (B). Note that slight random jitter was added to the points for visibility’s sake. Boxplots represent median (wide horizontal line), IQR (boxes) and range within 1.5 IQR (whiskers). Pre: Pre-shift, Post: Post-shift and Re: Re-shift phases. There was no difference between the groups, only the phases differed.

Discussion

The aim of Study 2 was to examine whether the presence of SNC in non-social tasks is a robust phenomenon in dogs. To this end, a series of methodological changes were included in the protocol used by Dzik et al. (2019), who found SNC using a non-social problem solving task featuring a commercial dog toy. This had the goal of assessing the degree of generalization of this phenomenon under different experimental conditions. The results showed that experimental dogs who experienced a reduction in reward value behaved similarly to control dogs. Moreover, the dogs’ performance did not differ across the different phases of the task. This suggests that, under these conditions, the dogs did not experience an SNC effect.

It is possible that the methodological changes reduced or even eliminated the presence of SNC. First, the low-value reward that was used was not the dry food each dog usually consumed, as in Dzik et al. (2019) and Study 1, but all dogs received the same commercial dry food. As shreds of evidence indicate that dogs tend to be neophiliac (Kaulfuss and Mills, 2008), it is possible that the novel food had a higher reward value than the usual food. In this sense, the discrepancy between the high and low-value rewards may have been smaller, precluding the appearance of SNC (Flaherty, 1982).

Second, dogs from the current study were assessed with their owners present. Prior findings indicate that the presence of the owner reduces stress in dogs (Tuber et al., 1996; Shiverdecker et al., 2013), and conversely, their absence increases it (Payne et al., 2015). The emotional response of frustration that accompanies SNC is similar to stress, and it is related to changes in animal behavior (Papini, 2022). It is possible that the presence of the owner reduced the dogs’ frustration, preventing them from experiencing an effect of SNC.

Third, in this case, dogs were evaluated in an unfamiliar environment while the task was carried out in their home in Dzik et al. (2019). The novel environment may have increased both stress (Hennessy et al., 2006) and exploratory behaviors (Marshall-Pescini et al., 2017; Mettke-Hofmann et al., 2022) in both groups. These effects could have led dogs to be less focused on the task. However, this explanation is unlikely as dogs from both groups were successful in completing the task in all phases.

In conclusion, although it is not possible to pinpoint which of these discrepancies account for the lack of an SNC effect in this sample, the lack of replicability of this phenomenon under new experimental conditions highlights that this is not a robust effect in dogs, in line with prior inconsistent results in the literature.

Finally, dogs solved the problem faster during the re-shift phase compared to the other phases. That their performance improved across trials suggests a learning effect independent of the reward value they received.

General Discussion

The present findings, together with prior literature, suggest that SNC in dogs does not appear consistently across protocols and experimental conditions (see Table 3). In particular, while some studies observed this effect (Bentosela et al., 2009; Riemer et al., 2018b; Dzik et al., 2019), others could not find differences in dogs’ behavior after experiencing a reduction in the value of an expected reward (Pongrácz et al., 2013; Riemer et al., 2016; 2018a). These findings are contradictory in light of the extensive evidence of SNC being a robust phenomenon that appears in a great variety of mammal species (Papini et al., 2022).

There are several possible explanations for the elusiveness of SNC in dogs. To begin with, as suggested by Riemer et al. (2016), it is possible that contrast effects are more reliably observed in controlled laboratory conditions and animals that have been bred and raised under standardized criteria. In the case of dogs, they are evaluated as adults, and their upbringing is not only uncontrolled but also largely unknown. Nevertheless, there is some evidence against this explanation. For instance, SNC has been observed outside of a laboratory in the usual living conditions of human babies (Lewis et al., 1990) and pet rats (Ellis et al., 2020).

In the second place, as it was previously mentioned, it is possible that the selected rewards vary in their discrepancy in incentive value. According to Papini et al. (2022) animals may detect a change in the reward without experiencing frustration if the difference between the obtained and the expected reward is under a certain threshold. In this sense, Dzik et al. (2019), keeping all other methodological aspects similar, found an SNC effect when dogs were shifted from receiving liver to the dry food they usually consumed, but not when the high-value reward was sausage. Moreover, when they carried out a preference test, dogs chose the liver significantly over the sausage (Dzik et al., 2019). These results suggest that the discrepancy between the high and low-value rewards is greater in the case of liver instead of sausage when they are compared to the dry food the dogs usually consumed. Regarding the present studies, for the pointing task, the high and low-value rewards (liver and the dry food the dogs usually ate) were the same ones that effectively produced SNC in previous studies (Bentosela et al., 2009; Dzik et al., 2019). However, this was not the case for the problem solving task, as the dry food was the same one for all dogs, and not all dogs ate dry food regularly.

In the third place, the hedonic value of rewards is relative and depends on the comparison with other rewards, both present and experienced in the past (Flaherty, 1982). Considering that dogs usually experience a great availability of high-value rewards during their daily lives, it is possible that the incentive value of the high-value reward presented during the task was diminished. This could have reduced the discrepancy with the low-value one, thus decreasing the probability of SNC occurring. On the other hand, the reduction in reward value has been shown to increase exploratory behavior (Pecoraro et al., 1999). In this sense, it has been suggested that the adaptative function of frustration is promoting incentive disengagement from a reward that is no longer available due to its devaluation or omission (Papini, 2003). To this end, frustration could facilitate switching from a previously successful response that no longer works, to new ones that could be more effective (Stout et al., 2002). In the case of dogs, being used to the availability of highly appetitive food could reduce the searching behavior that occurs when the incentive value is decreased during the task.

In the fourth place, it should be noted that dogs frequently receive intermittent reinforcement for different behaviors during their daily lives. Partial reinforcement schedules increase tolerance to frustration and thus decrease the contrast phenomenon (Amsel and Penick, 1962).

In the fifth place, the presence of the owner during the task could reduce the frustration response experienced by the dogs (Tuber et al., 1996; Shiverdecker et al., 2013. This hypothesis is supported by the fact that the owners were absent during the pointing task in Study 1, but they were present during the problem solving task of Study 2. Nevertheless, Riemer et al. (2018b) found SNC in dogs evaluated with their owners. It is important to consider that this possible stress-ameliorating effect has not been specifically assessed in a CNS protocol yet, and prior studies have not been consistent regarding this variable.

Finally, in terms of the mechanisms that may be involved, the emotional impact that dogs experience due to the reduction in the value of an expected reward is unclear. Using a different protocol, Bremhorst et al. (2019; 2022) found an increase of facial expressions such as blink, lips part, jaw drop, nose lick and ears flattened in a situation in which frustration was provoked by a delay in obtaining the reward. In addition, Jakovcevic et al. (2013) carried out a social task in which a reward was omitted during an extinction phase. As possible indicators of frustration during this phase, they observed an increase in dogs moving away and orienting their gaze and head away from the food source (a person that previously reinforced their gazing behavior) as well as an increase in lying down, ambulation, sniffing and vocalizations. Finally, McPeake et al. (2021) found that the change in cortisol levels after receiving a battery of frustration-inducing tests, compared to pre-test levels, correlated with the dimension ‘Frustration coping’ of the Canine Frustration Questionnaire. Nevertheless, Riemer et al. (2018b) did not find an association between the intensity of SNC and anxiety-related behaviours, which were measured through a novel object test and a personality questionnaire (Canine Behavioral Assessment and Research Questionnaire, C-BARQ). Although these findings suggest that dogs could experience an emotional response of frustration during SNC, the evidence is unclear regarding its occurrence and intensity and requires further exploration.

Given the characteristic of this species, the emotional impact could be greater if the reduction was in an expected reward of social nature. While there are no studies of SNC using social rewards, there are some evidences that indirectly support this hypothesis. For instance, dogs react, both behaviorally and physiologically, to being separated from their owners, suggesting an increase in stress. This has been observed even after separations of a short duration (Payne et al., 2015; Rehn and Keeling, 2016; Riggio et al., 2021). Recently, Barrera et al. (2021) evaluated dogs who experienced an unexpected decrease in a social reward which was the interaction with an unknown person. Dogs showed less affiliative behaviors when the reward was reduced. Although affiliative behaviors increased again when the interaction was reinstated, their levels were lower than they initially were, suggesting a carry-over effect of the reward devaluation. Unfortunately, that study did not include a control group that always received the low-value reward, so it is not possible to confirm an SNC phenomenon. Nevertheless, the sum of these findings is promising and encourages further research in this area.

Taking all these aspects into account, there are a series of methodological differences between study 1 and study 2 that could explain the inconsistency in the results. To begin with, one task was eminently social while the other was non-social. Although a human provided rewards in both tasks, in study 1 the reinforced behaviour was communicative in nature, while in study 2 the dog had to solve a problem independently. Secondly, the low-value reward differed across studies, which could have affected the extent of the discrepancy of this option and the high-value one. In the third place, the presence or absence of the owner could have modulated the dogs’ stress levels during the task. Finally, the degree of effort required to carry out each task was different. These experimental designs do not allow us to discriminate which of these factors impacted the results and what may have been the relative influence of each of them. Future studies should compare tasks that only differ in one of these aspects, in order to separately examine their effect on SNC. For instance, a social and a non-social task could be used, but the reinforcers, presence of the owner and difficulty of the task should remain the same.

All in all, in the current work, we observed an effect of SNC in dogs during a point following task, but not in a non-social problem solving task. This discrepancy is in line with prior findings in the literature that show that SNC appears in dogs in some situations but not in others. It is possible for this inconsistency to be due to methodological factors as well as characteristics of the species. Given the relevance that this phenomenon has for learning and in the daily interaction of dogs with people as well as during training, further research is needed to deepen our understanding of this phenomenon.

Table 1. Experimental design of Study 1.

Groups	Warm-up (1 session of 2 trials)	Pre-shift (2 sessions of 10 trials)	Post-shift (2 sessions of 10 trials)	Re-shift (1 session of 4 trials)
Experimental	Liver	Liver	Dry food	Liver
Control	Dry food	Dry food	Dry food	Dry food