Object permanence has in recent years, directly or indirectly, been assessed in some domestic animal species such as dogs (Canis familiaris) (Triana and Pasnak, 1981; Fiset and Plourde, 2013; Zentall and Pattison, 2016), pigs (Sus scrofa domestica) (Nawroth et al., 2013), sheep (Ovis orientalis aries), and goats (Capra aegagrus hircus) (Nawroth et al., 2014). As these species are kept under human care, knowledge about their ability to track objects can be relevant for their welfare. Domestic dogs have been found to be able to solve visible displacement tasks where the object disappears into the container (Triana and Pasnak, 1981; Fiset and Plourde, 2013; Zentall and Pattison, 2016), and possess good memory for remembering the location of hidden and displaced object (Zentall and Pattison, 2016). Domestic pigs were tested in a set-up similar to that used for primates but adapted to the behavior of pigs (Nawroth et al., 2013). One of three locations contained a reward and pigs were tested for visible, and invisible displacement of hidden rewards, and solved only the visible displacement tasks. In goats, a later study revealed that dwarf goats performed above chance in some of the displacement experiments, where they tracked the movement of hidden objects and thus could reach stage 6 (Nawroth et al., 2015), suggesting that other domestic ungulates might possess this ability too.

In horses (Equus caballus), studies on object permanence are limited, although knowledge of horses’ abilities to follow hidden objects is important information for humans interacting with horses. Horses may e.g., react fearful toward re-appearing objects, conspecifics or humans, if they do not have the ability to mentally follow the objects. Such fear reactions could be unexpected by humans, which could lead to potentially dangerous situations. Uller and Lewis (2009) provided indirect evidence of object permanence, as horses in this study remembered the amount of hidden items (i.e., food). More sophisticated forms of locating objects e.g., when they are hidden and displaced, have, to the best our knowledge, only been studied by one team. This particular study showed that horses were capable of solving visible displacement of a reward on either two or three possible hiding locations but failed to solve invisible disposition tasks (Trösch et al., 2020). The results, however, need to be reviewed with caution due to the limited sample size (n = 16) and as only one sex was included (only females). Further investigation of object performance abilities of horses is therefore needed, and is highly relevant to the keeping and management of horses.

Assessing horses’ performance in any test situation needs to be viewed in a context of individual characteristic. Individual variation means that individuals can vary in their motivation to engage in cognitive tasks, and thereby affect their performance and hence applicability (and replicability) of the results. The motivation to e.g. engage in a learning task is a crucial factor for learning as it will affect the horse’s willingness to participate, and its attention span. In a recent study by Christensen et al. (2021b), curiosity and exploration of novel objects was positively correlated with learning performance in both positive and negative reinforcement tasks. Motivation may, however, vary with both situational factors and individual characteristics such as age, sex, but also emotional state (Christensen et al., 2012a; Valenchon et al., 2013; Schubiger et al., 2015). Age has been shown to affect performance in food location tasks (Murphy et al., 2004), and in social transmission tasks (Krueger et al., 2014). Young horses also learned a practical task, e.g., the ability to use an automated forage station, quicker than older horses using the same introduction routine (Kjellberg and Morgan, 2021). In the study by Murphy et al. (2004), authors also found an effect of sex in visuo-spatial tests, with females performing consistently in six consecutive trials, whereas males performed slower in the first trial, but faster than females in subsequent trials. This effect is in accordance with findings by Wolff and Hausberger (1996) who demonstrated that females tended to learn faster than males. Collectively, this indicates that female horses are better instrumental problem solvers, while male horses outperform mares in visuo-spatial tasks. Coat colors is another individual characteristic described to be associated with greater tactile sensitivity and reactivity e.g., chestnut colored horses (Rørvang et al., 2020a). Although there is no research on the topic in horses, practitioners often report red or chestnut horses to be more reactive, and research from rodent studies show that red coat color is associated with greater pain sensitivity (Mogil et al., 2005).

In addition to the above mentioned individual characteristics, an individual’s current affective state may also affect learning performance (Schubiger et al., 2015; Ringhofer and Yamamoto, 2017). This is what has been described by the Yerkes and Dodson (1908) theory, stating that performance (in any test) increases with increasing mental arousal state, until a peak point, then decrease rapidly (i.e., the Yerkes-Dodson bell curve). Lansade and Simon (2010) found that temperament, and there the suscibility to stress, influenced learning performance but further noted that this was highly task-dependent. The latter is hypothesized to be a predisposition to react to stimuli involved in learning situations rather than a direct effect. However, it often can remain unclear which is cause and effect i.e., if stress is caused from an inability to succeed in a task, or if the stress hampers performance in the task. For example, unclear cues or unrealistic performance expectations can increase arousal level and can often result in frustration and conflict related behaviors such as kicking, pawing in ridden horses (Górecka-Bruzda et al., 2015; Christensen et al., 2021). Measuring arousal level and occurrence of frustration behavior in horses trying to solve cognitive tasks could therefore improve our understanding of how frustration can be associated with learning performance.

The current study aimed to investigate horses’ performance in visible and invisible displacement tasks and investigated whether specific ontogenetic, behavioral and physiological factors were associated with performance. The hypothesis was that horses would be able to solve visible but not invisible displacement tasks following the results from Trösch et al. (2020). In addition, the study investigated if individual characteristics such as training level, age, sex, and coat color (red or other color) influenced horses’ performance in the tasks, and if horses who were successful in one task also were successful in another. We hypothesized that training level would have a positive influence on the horse’s attentiveness toward the human and thus the task, making horses with higher training level more likely to solve the tasks. Moreover we hypothesized that younger horses would perform better than older horses, that females would perform better than males, and that red coat color would affect performance negatively in the tasks.

Lastly, the study aimed to examine if there was any association between physiological parameters (i.e., heart rate) of the horses and performance in the tests, and between behavior indicative of frustration and performance. While we did not expect any difference in horses’ heart rate between the different tests per se, we would expect unsuccessful horses to experience a certain level of frustration resulting in higher heart rates and more behavior linked to frustration than successful horses during the tests.

Prior to the experiment start, the owner of the stud was informed and agreed to all experimental procedures, data collection and publication (written consent, signed 1st of March 2021). All procedures were conducted in accordance with national legislation on animal experimentation by the Danish Ministry of Justice, Act. no. nr. 253 (8 March 2013) and § 12 in Act. no. 1459 (17 December 2013), as well as met the ARRIVE guidelines (Kilkenny et al., 2010) and the ethical guidelines proposed by the Ethical Committee of the ISAE (International Society of Applied Ethology) (Sherwin et al., 2013).

As the experiment was conducted during the COVID-19 pandemic, measures were taken to comply with the current precautions. The experiment were conducted in Denmark, and hence complied with the Danish COVID-19 regulations effective in the experimental period 1st of March–15th of April 2021. Only one experimenter was part of the experiment to avoid contamination risks between persons. The experimenter was PCR tested for COVID-19 bi-weekly and only traveled to the stud if test results were negative (all tests = negative) before the experiment started and were never in near proximity (less than 2 m) to any other persons at the horse stud. Clothes were washed between each visit to the stud, and hands were disinfected with alcohol upon arrival and frequently during an experimental day. The experiment was conducted in a separate section of the stud, where no other daily activities connected to the stud (training and preparation for training) were performed.

Thirty-nine privately owned Icelandic horses participated in the study. Thirty-two females and 7 males with a mean age of 12 years (range: 3–25 years). All horses were from the same owner and were either born at the stud where the experiment was performed, or had been trained at the stud for minimum 6 months. The stud owner was also the horses’ regular trainer. The experiment was performed in a stable section of the stud, separate from the normal daily activities going on at the stud. This stable had six individual pens with grating openings in the front of each pen toward to stable aisle. All horses were group housed when outdoors, and individually housed when indoors, in individual pens. All horses were thus acquainted with the individual pen(s) used for the experiment beforehand. All horses were also used to being alone in a pen with visual access to at least two familiar peers. Upon arrival to the stud, the owner of the horses provided data on each horse: ID, age, sex, and training level (T0–T5) (Table 1). The scale of training level was made in collaboration with the stud owner:

No changes were made to the horses normal feeding (roughage and concentrates were provided 6–7 am and at 5.30 pm) or training routines (normal training included either round pen training, riding (in- or outdoor) or lunging in the indoor riding arena). All horses had access to ad libitum water at all times.

The experimenter stayed on the stable aisle in front of the particular pen (Figure 1). From inside the pen, the horse had access to the aisle at three points in the pen grating (each 25 cm wide, with 70 cm between each). A white sliding board (1.6 × 0.3 m) made of lacquered wood was placed in front of the pen grating, on the aisle floor approximately 1 m away (out of reach of the horse). In the testing situation, a target (squared black plastic box: 7 × 10.5 × 8 cm, Plast team, Denmark) was placed on the board on either the left or right side. The target was then covered by a white plastic cup (height: 13 cm, Ø: 17.5 cm, Plast team, Denmark), which was also placed on the other side of the sliding board (Figure 2). The sliding board, the target and two white cups had been placed in the horse’s normal environment i.e., the stable, 48 h in advance to ensure no olfactory differences, nor olfactory cueing or novel odors. The placement of the task on floor level was chosen to accommodate the perceptual abilities of horses. Tasks where stimuli are placed at ground level appear to be better suited for the presentations to horses (Hall, 2007), probably due to the visual abilities of horses, with binocular vision only 55–65° in front of the horse (Hughes, 1977), and predominantly below the head, extending down∼75° (Timney and Macuda, 2001).

Horses were habituated, trained and tested in groups of 3–5 horses at a time, with at least one older and calm companion present in a nearby pen to ensure a calm atmosphere. The calm companions were always familiar horses which were physically distanced but within visual contact from the particular individual being trained or tested. The companions were either not part of the experiment (n = 2, who were selected by the stud owner), or horses who had already been tested to exclude any bias due to socially acquiring information about the task. Each group was habituated over a 1–2 days period, depending on progress. All horses were habituated, trained and lastly tested from their respective pen (3.1 × 3.5 m). The individual pens were bedded with wood chip bedding and had an automatic water cup adjacent to the box opening. The horse would initially be habituated to wearing a wireless heart rate monitor (Polar Equine, see section “Heart rate”) on an elastic girth, and heart rate was monitored during habituation, target training, and testing.

To prepare all horses to the procedure and the instruments used, a stepwise habituation procedure was conducted prior to the experiments:

Hereafter, a target training was initiated to teach the horse to follow a target: The horse was presented with a small black box (i.e., the target, 7 × 10.5 × 8 cm) placed in the middle of the sliding board. The target was placed on the board in full view of the horse, and the board was then pushed toward the pen grating. As soon as the horse touched the target it was rewarded with a treat (the treat was delivered by the experimenter on the board by the experimenter’s hand). The sliding board was moved away (approx. 80 cm from the grating, and out of reach of the horse, Figure 1) and the target was lifted and placed again for another trial. When the horse touched the black box within 3 s from presentation, three times in a row, the target was moved to either the right or left of the sliding board in the next trial. Hereafter, the target was presented pseudorandomly on the left or right side. When the horse touched the target 8 times in a row on either the left or right side it was considered ready to progress to testing. Each horse had two sessions of 10 trials each to reach this criterion. A video illustrating the target training can be found in Supplementary Video 1.

Video recording of all habituation, training and tests was conducted from the stable aisle, using a Sony HDR-cx240E handycam on a tripod (150 cm high). After each experimental day, video files were downloaded and raw data was extracted into Excel (version 2016) by an experienced observer. Trials needed to reach the step 1 habituation criterion were counted, and the duration of time needed for the horse to accomplish the step 2 habituation criterion was measured. During target training, horses had different ways of approaching the task. Some were more active and had more trials per minute than others had; hence, trials per minute were noted for each horse to form a variable indicative of the individual’s motivation to the task (mean trials per minute during target training). Success rate in target training was also noted as successful trials out of 10 trials total (as all horses accomplished at least 80% success rate in their first session i.e., 10 trials). For all tests; choice test, and visible and invisible displacement task, success rate (successful trials out of 10) was noted, and behavior potentially indicative of frustration (defined as; scraping or stamping the pen bedding, scratching body, head shaking, or pawing (from Rørvang et al., 2020a,b) was also noted as 1/0 sampling during each trial (hence maximum 10 occurrences per horse per test).

Horse heart rate was recorded using Polar Equine H10 sensor (RR recordings; Polar Electro OY, Kempele, Finland), consisting of a belt/elastic girth with two electrodes, a wireless heart rate sensor and a wristwatch receiver. The belt with electrodes was fitted around the horse’s belly behind the shoulder. Water and ultrasound gel [Apotekets Eksplorations gel (clinic), 211894, Skovlunde, Denmark] were applied for optimal connection between electrode and skin. Data were downloaded using the online software Polar Flow. Average and maximum heart rate from habituation, target training, choice test, visible displacement task, and invisible displacement task, respectively, were noted. Heart rate variability was not considered as the duration of each trial or registration period (either habituation, training or testing period) was shorter than 10 min each (von Borell et al., 2007).

All trials were video recorded and coded live while testing. All choices of the horses (left or right side and corresponding correct or incorrect) were clear and unambiguous; hence, we did not calculate inter-observer reliability.

One horse showed strong neophobic reactions during the habituation period and was thus excluded from the study for welfare reasons. All remaining horses complied with the habituation criterion (both step 1 and step 2) and all but one horse complied with the target training success criterion and thus proceeded to the choice test (i.e., data from habituation n = 39, and target training n = 38). From the choice test, all but one horse (which succeeded only in 50% of trials) complied with the success criterion (Table 2) and hence 37 horses proceeded to the visible displacement task and the invisible displacement task. During the visible displacement task, two horses lost interest in the test (did not approach the board/the task upon presentation), resulting in 35 horses in the analysis of the visible displacement task. In the invisible displacement task, the same two horses did not show an interest upon presentation of the task, and four other horses dropped out after the first five (unsuccessful) trials and hence were excluded from the analysis of the invisible displacement task (Table 2). In addition, there were some errors in the heart rate measures, potentially due to low conductance between skin and electrode when horses moved. This resulted in the following sample sizes for the heart rate data for the different phases of the experiment (Table 2).

Statistical analyses were performed using the R software, version 3.1.2 (R Core Team, 2019) and all p-values evaluated according to a significance level of 5%.

Individual horses needed to complete 9/10 correct trials in order to perform significantly better than expected by chance (binomial test based on 10 trials, p < 0.05). In the first 10 trials of the choice test, 27 horses performed better than expected by chance, and eight horses performed better than expected by chance in the visible displacement task. In the invisible displacement task, none of the horses performed significantly better than expected by chance (Table 3). This picture was also reflected in the group performance where horses performed above chance level in choice test and visible displacement task (One-sample Wilcoxon test, p < 0.05, Figure 3). In the invisible displacement task, the group performed significantly below than expected by chance (one-sample Wilcoxon test, p < 0.05, Figure 3) indicating that horses persistently chose the side where they had last seen the target (i.e., where they had seen the target before it was moved while occluded).

The fitting of the three models showed that none of the fixed effects [i.e., interaction (age:Training level), Sex, color, motivation to the task, baited side, and for visible and invisible displacement tasks: displacement direction] had a noteworthy effect on horse performance in any of the three tests (GLMM, all variables p > 0.05).

There was a positive correlation between success rate in the visible and invisible displacement task, meaning that successful horses in the visible displacement task also were more successful in the invisible displacement task (figure can be found under Supplementary Figure 1).

Overall horses performing well had lower mean and maximum heart rate compared to less successful horses, with the invisible displacement task provoking the highest heart rates and lowest success rates. The model fitting showed that performance were significantly associated with average (F_df = 83.360₆₀ and p < 0.001) and maximum (F_df = 79.558₅₈ and p < 0.001) heart rate during the tests. The model output moreover showed that low average and low maximum heart rate were significantly associated with high performance in the invisible displacement task [Linear Mixed-effects model: (mean) F_df = 9.06₁, p < 0.001 and (max) F_df = 22.28₁ p < 0.001] but not in the visible displacement task [Linear Mixed-effects model: (mean) F_df = 1.84₁, p = 0.18 and (max) F_df = 0.25₁, p = 0.62] and the choice test [Linear Mixed-effects model: (mean) F_df = 1.98₁, p = 0.28 and (max) F_df = 0.47₁, p = 0.45] (Figures 4A,B).

The Pearson correlations showed that horses with high (or low) heart rates in one test also had high (or low) heart rates in the next test (figure can be found under Supplementary Figure 2).

The occurrence of behavior indicative of frustration was generally low, but increased from Choice test to the visible displacement task and again from the visible displacement task to the invisible displacement task. In the choice test, behavior indicative of frustration was noted in 6% of the trials (22 out of 370 trials), and of these, the majority of trials were unsuccessful [5 out of 22 trials were successful, Linear Mixed-effects model: F_df = 52.81₁, p < 0.001]. In the visible displacement task, horses expressed behavior indicative of frustration in 23% (82 of 350 trials) of the trials, and in the invisible displacement task, the percentage further increased to 43% (134 of 310 trials), but the behavior was still mainly linked to unsuccessful trials [Linear Mixed-effects models: (visible displacement task) F_df = 82.79₁, p < 0.001 and (invisible displacement task) F_df = 55.18₁, p < 0.001].

Horses’ ability to solve visible displacements and their ability to remember placement of hidden food is beneficial as it may aid their ability to follow objects, conspecifics and humans in their environment. On the other hand, horses’ inability to follow hidden objects is also important information as it means that one should not expect a horse to know where objects, conspecifics or humans have gone, if the move or displacement was not visible to the individual. As a result, horses may react fearful toward suddenly re-appearing objects, conspecifics or humans, which could be unexpected by humans, which poses a risk of mismatching expectations. Therefore, it is important to know and communicate both the abilities but certainly also the inabilities of horses to ensure welfare and a safe horse training environment for both horses and humans. The frustration shown in both the occurrence of conflict behavior and heart rate associated with non-successful trials further pinpoints this issue.

In terms of practical horse training, horse’s ability to follow and remember the placement of targets is beneficial in situations where the trainer would want to distance him/herself from the horse. This could for instance be in risky situations such as during training for loading to transport. In such close contact situations, horses may react fearfully and unexpected (due to the novelty of the situation), and hence using targets on e.g., a stick, could be a way to train the horse from a safe distance (see e.g., Dai et al., 2019).