The authors have declared that no competing interests exist.
Anecdotal reports of horses opening fastened doors and gates are an intriguing way of exploring the possible scope of horses’ problem-solving capacities. The species’ natural environment has no analogues of the mechanisms involved. Scientific studies on the topic are missing, because the rate of occurrence is too low for exploration under controlled conditions. Therefore, we compiled from lay persons case reports of horses opening closed doors and gates. Additionally, we collected video documentations at the internet platform YouTube, taking care to select raw data footage of unedited, clearly described and clearly visible cases of animals with no distinct signs of training or reduced welfare. The data included individuals opening 513 doors or gates on hinges, 49 sliding doors, and 33 barred doors and gateways; mechanisms included 260 cases of horizontal and 155 vertical bars, 43 twist locks, 42 door handles, 34 electric fence handles, 40 carabiners, and 2 locks with keys. Opening was usually for escape, but also for access to food or stable-mates, or out of curiosity or playfulness. While 56 percent of the horses opened a single mechanism at one location, 44 percent opened several types of mechanism (median = 2, min. = 1, max. = 5) at different locations (median = 2, min. = 1, max. = 4). The more complex the mechanism was, the more movements were applied, varying from median 2 for door handles to 10 for carabiners. Mechanisms requiring head- or lip-twisting needed more movements, with significant variation between individuals. 74 horses reported in the questionnaire had options for observing the behaviour in stable mates, 183 did not, which indicates that the latter learned to open doors and gates either individually or from observing humans. Experience favours opening efficiency; subjects which opened several door types applied fewer movements per lock than horses which opened only one door type. We failed to identify a level of complexity of door-fastening mechanism that was beyond the learning capacity of the horse to open. Thus, all devices in frequent use, even carabiners and electric fence handles, are potentially vulnerable to opening by horses, something which needs to be considered in relation to keeping horses safely.
It is unlikely that skills specific to opening door and gate mechanisms have evolved in equids: non-prehensile, hoofed animals which feed on easily accessible, distributed resources
So far, the opening of fastening mechanisms has only been investigated in animals with claws and paws. Reported motivations and aims for opening were various: to gain access to food (chicken
To study infrequent behaviour, analysing a comprehensive collection of anecdotes offers a good starting point. For instance, crowdsourcing studies have analysed the range of flexibility of animal problem-solving abilities
Data-mining of this kind runs the risk of collecting biased and occasionally false reports
We employed crowdsourcing
In analysing this data, we aimed to evaluate how and why horses opened door and gate mechanisms of different sorts. We asked whether animals learn to open locked doors and gates individually, or socially by observing stable mates
We invited owners and caretakers to report on door and gate opening by horses, mules, and donkeys, by means of a website we set up (
In addition, we sampled video material from the internet platform YouTube which we found with the key words “open door”, “open gate”, “escape”, “run-away”, “clever”, “horse”, “donkey”, and “mule”. These videos were analysed without downloading them (data and video links see
We obtained informed consent from all persons who answered the questionnaire. On the website, all responders agreed to the anonymous publication of their data, including pictures and videos for scientific purposes: reasonable requests for access to anonymous agreements can be obtained from the corresponding author. Only the agreed information on the equids and no data of the reporting persons was used for the present study. Some videos were published on YouTube with a Creative Commons
In total, we found 419 cases in which equids opened doors or gates. The cases ran through a selection process (
Material from YouTube added 68 videos, giving 77 videos in total. Videos were similarly run through the selection process (
The remaining cases were rated by three independent persons, one professor and two bachelors in equine science, as to whether further reports and videos should be excluded from the study based on reduced welfare, and they agreed in all but one case (inter observer agreement: κ = 0.98). The case in question was therefore excluded as well (for remaining raw data see
We used a quantitative–qualitative mixed questionnaire approach
The specific questionnaire on door and gate opening asked three open questions, two semi-closed questions, and 25 closed questions in a semi-random order to prevent order biases in the responses
The 402 horses reported were of various breeds which we summarised in breed types, as used in genetic studies
In principle, individual animals might open doors or gates at only one location or at several (e.g. their own or other animals’ box doors, feed room doors, house doors, or pasture and paddock gates), and the fastening mechanisms might be of the same or of different types.
We categorized doors and gates into 3 types (
door or gate suspended on hinges and pushed open
sliding door, suspended on wheels above the horses’ heads, and rolled open
barred door or gateways, subcategorized in:
electric fence gate, gate supplied with electric power, consisting of electric fencing and power-free handles which have to be unhooked from the fence to create a gateway gate with horizontal wooden poles, wooden fencing with wooden poles that slide out of an impression or frame to create a gateway doorway chained so that chain blocks an opening, usually by attachment to the other side of a fence or stall door with a carabiner
The arrows indicate orthogonal door and gate opening directions. Doors and gates on hinges were opened in the same or opposite direction to the animals’ movement direction. The doors and gates c1)–c3) are subcategories of the category c) barred door or gateways. At b), horses had to grasp and pull down a pipe with their mouth at the dotted arrow and then move the door into the direction of the continuous arrow.
Mechanism types were categorized as follows (
horizontal bar, which had to be moved backwards and forwards, including wooden gate poles
vertical bar, which had to be moved up or down, including bars which secured sliding doors
twist mechanisms controlling a bolt, where a disc had to be turned to the left or right to close or open the bolt
door handle
electric fence handle
carabiner, which connects chains or secures other devices. Carabiners could have clip or screw mechanisms
security chain, which secures gates or other mechanism types
wooden boards that were pulled out of a door frame
locks with keys, including padlock with key
Arrows indicate the direction of the animals’ head movements.
We evaluated the “complexity” of the doors, gates and mechanisms in three ways, with respect to: the direction of overall movement; the plane of head movement; and the number of head movements.
We categorized the direction in which the doors, gates and mechanisms opened in relation to the animals’ movement direction after the opening (see
*The horizontal movement plane was further distinguished into left-right and backwards–forwards.
From preliminary data analysis it was evident that all the horses opened the locked doors and gates with their mouths. We categorized the movement of the animal’s head or lips (for simplicity, “head movements”) according to linear movement planes and whether twisting was involved (see Figs
For each opening of the doors, gates and mechanisms we counted:
The total number of head movements the individuals had to apply to open the different doors, gates, and mechanism types (Figs
The frequency in which horses displayed series of head movements. We considered responses to form a series when the individual was in contact with a door, gate or mechanism and displayed one or more continuous movements until it turned away from the door, gate or mechanism to at least 30° and /or interrupted its movements for at least 3 seconds. A series of actions was counted as part of the same door or gate opening instance, if the horse continued with further manipulations of the door, gate or mechanism within 1 minute of the preceding sequence. Movements within a series were mostly, but not always, ordered in sequence—some animals displayed opening movements which were not necessarily needed to open a door, gate or mechanism.
The efficiency of the movements. We measured the number of head movements performed to open the doors, gates and mechanisms, and compared it to the minimum number of movements which would be needed by a person (efficiency = minimum number of movements / actual number of movements).
Opening carabiners (median = 10, min. = 4, max. = 10, GLM: N = 89, Std.E. = 2.27, z = 3.91, p < 0.001), locks with keys (median = 8, min = 8, max = 8, GLM: N = 89, Std.E. = 2.85, z = 2.41, p = 0.02), security chains (median = 7, min. = 7, max. = 7, GLM: N = 89, Std.E. = 2.67, z = 2.12, p = 0.03), and bolts which had to be pulled vertically (median = 4, min. = 1, max. = 10, GLM: N = 89, Std.E. = 0.63, z = 4.43, p < 0.001) and electric fence handles (median = 3.5, min. = 2, max. = 6, GLM: N = 89, Std.E. = 0.97, z = 2.96, p = 0.003) needed more movements than opening handles (median = 2, min. = 1, max. = 7, GLM: N = 89, Std.E. = 0.62, z = 1.76, p = 0.08) and locks with bolts which had to be moved horizontally (median = 3, min. = 1, max. = 10, GLM: N = 89, Std.E. = 0.46, z = 6.19, p < 0.001). Twist mechanisms (c) were not documented. The boxplots visualize the quartiles of the data per lock type. The box comprises 50% and the lower and upper whisker 25% of the variability each. The dots visualize outliers. *** = p < 0.001, ** = p < 0.01, * = p ≤ 0.05.
Doors, gates and mechanisms that opened in one direction, in one plane and with not more than two movements were termed “simple”. Doors, gates and mechanisms which required movement in several directions of the same movement plane, or on several movement planes, or twisting head or lips were considered more “complex”
For statistical analysis and the depiction of the data we used the package R commander of the R-Project statistical environment, Libre Office 4.3.3.2 and Photo shop CC2017. Most of the data were not normally distributed (Kolmogorov-Smirnov test). Therefore, we applied tests suitable for non-parametric data throughout. In cases where clear cut descriptive data were offered, we did not apply inferential statistics. For more complex data and research questions, we applied Generalized Linear Models (GLM) and Generalized Linear Mixed Models (GLMM).
We applied GLMs when only fixed effects were calculated. GLMs set at ‘family Poisson’ were calculated for the dependent variables a) door and gate types, b) the mechanism types, and c) the mechanism positions; fixed factors were age of the horses and whether other horses showed the same behaviour. A separate GLM was set at ‘family Binomial’ for the now dependent variable whether other horses showed the same behaviour; fixed factors were door and gate types, the mechanism types. Further GLMs set at ‘family Binomial’ were analysed for the depended variables horses a) stayed in the stable, b) visited other equids, c) moved around freely, d) broke into other places and e) freed other equids; fixed factors were the management of the animals, i.e. single or group housing, daily access to pasture or on a limited number of days per week, permanent or temporary contact with other equids, and limited or unlimited roughage.
We applied GLMMs when several mechanisms opened by particular horses were part of the analysis and considered the ID of the horse as random factor. Dependent variables of the GLMMS were a) the total number of head movements, b) the frequency in which horses displayed series of head movements, and c) the efficiency of the movements. Fixed factors were the door and gate types, the mechanism types, the number of head and mouth movements, and the number of head movement directions (for definition see above). The GLMMs were set at family ‘Poisson’.
The model with the best fit (with the lowest AIC index) was chosen after stepwise removal of factors. In cases reduced models were used the AIC of the particular model is given in the results and complete and reduced models are provided at
In 257 of the 335 cases, people reported on whether the animals had the option to observe the behaviour in stable mates and whether the behaviour was shown by further stable mates after the focus animal displayed it. Seventy-four animals had the option to observe the behaviour in a stable mate. In 61 reported cases the behaviour was later shown by others in the same stable. Whether animals were reported to have the option for observing the door and gate lock manipulations in stable mates did not have any effect on how many different doors and gates, how many different locks, and how many locks in different locations were opened (GLM: N = 333, all p > 0.05). Subjects which had the option to observe the manipulation in conspecifics did not open any particular door, gate or lock types preferentially (GLM: N = 333, all p > 0.05); only twist locks were opened more often by animals which had no option to observe the opening (GLM: N = 333, SE = 0.51, t = -2.08, p = 0.04). Five animals (7%) with the option to observe conspecifics at the same lock type and 29 (16%) with no option to observe the manipulation opened twist locks.
We found no effect of the animals’ age (mean = 10.14, SD = 6.27) on the number of doors and gates or number of mechanism types that were opened, or on the number of locations in which the mechanisms were opened (GLM: N = 333, all p > 0.05). Many of the gate and door openers were castrated male animals (N = 212) and many of them were warmblood horses (N = 207), but as the sex and breed distribution may be biased by preferences of the persons who responded to our query, we do not consider the breed and the sex for further comparisons.
Management conditions were approximately equally distributed for the reported animals. People kept 52% of the animals in single housing and 48% in group housing; 56% of the horses had daily access to pasture, 44% access to pasture on a limited number of days per week; 57% were in unrestricted contact with other horses, 43% in restricted contact; 47% of the horses received roughage ad libitum, 53% received restricted roughage.
After opening the doors and gates of a box, enclosure or pasture, 87% of the animals walked out, 62% ran around in the area surrounding their stable, 22% went into other horse boxes or stables, 15% freed other horses, and 22% broke into other places such as feed storage rooms or human houses. Horses tended to stayed in their boxes or pastures after opening the door or gates when other horses in the same stable showed the same behaviour before (GLM, AIC = 303.36: N = 333, SE = 0.24, z = -1.82, p = 0.07) and were more likely to have daily access to pasture than those that exited the boxes or pastures (GLM, AIC = 303.36: N = 333, SE = 0.28, z = -2.515, p = 0.01). Furthermore, those that went into feed storage rooms were more likely to have daily access to pasture (GLM, AIC = 312.21: N = 333, SE = 0.27, z = 2.41, p = 0.02) and were older (mean: 11.5 years, SD = 6.9) than horses that did not open feed storage rooms (mean: 9.6 years, SD = 5.8). All other management restrictions were unrelated to actions after opening doors and gates (GLM: N = 333, all p > 0.05).
The case reports describe the opening of 520 door or gate types, and the videos show the opening of 75 doors and gates (for frequencies see
Most reports described the opening of one door or gate per animal (median, min. = 1, max. = 4). In the survey, the animals typically opened one door or gate type (median, min. = 1, max. = 2) and two mechanism types (median, min. = 1, max. = 5), at two locations (median, min. = 1, max. = 4). In the videos, the animals opened a median of one door or gate type (min. = 1, max. = 2), one mechanism type (min. = 1, max. = 3), at one location (min. = 1, max. = 6).
Most of the analysed doors and gates (N = 75) opened in the direction of travel of the horse manipulating the mechanism (52 away from the horse =
All mechanisms (N = 89) had to be opened orthogonally to the travel directions of the animals, except one bar which had to be pulled in the direction opposite to that of travel. Most mechanisms were opened with a linear head movement: 19 bars were moved horizontally (14 left and right, 5 backwards and forwards); 20 bars were moved vertically (upwards or downwards); 32 bars were moved horizontally and vertically, 9 door handles were pulled vertically downwards; 5 electric fence handles were moved horizontally left or right towards the supporting pole of the electric fence and then upwards out of wire loops at the pole; 8 wooden boards were pulled vertically upwards out of a door frame. However, some mechanisms had to be opened with twisting movements: 4 sideways and 5 upwards/downwards moving bolts, 3 carabiners, one security chain, and one key. The 3 carabiners had to be twisted circularly and 1 key had to be turned in a padlock and the padlock then lifted vertically upwards.
To open doors or gates (Figs
For opening barred doorways horses performed fewer movements than for opening sliding doors (GLMM: N = 75, SE = 2.89, z = -3.3, p < 0.001) or doors on hinges (GLMM: N = 75, SE = 2.75, z = -3.93, p < 0.001). Thus, horses used the most movement sequences to open barred doorways, fewer for sliding doors (GLMM: N = 75, SE = 2.34, z = -3.14, p = 0.002) and fewer still when the doors were on hinges (GLMM: N = 75, SE = 2.24, z = -3.43, p < 0.001). Individuals differed significantly in the number of movements (GLMM: N = 75, SE = 0.05, z = -2.13, p = 0.03) and number of movement sequences they performed (GLMM: N = 75, SE = 0.04, z = -2.48, p = 0.01) to open barred door and gates.
The efficiency of the horses in opening did not differ between door and gate types (GLMM: N = 75, all p > 0.05) and was very close to the best efficiency possible for humans (efficiency = 1). However, subjects which opened several door and gate types needed fewer movements (GLM: AIC = 395.97, N = 89, z = - 2.23, p = 0.03), fewer movement sequences (GLM: AIC = 287.09, N = 89, z = -1.99, p = 0.05) than horses which opened only one door or gate type.
To open the mechanisms (Figs
Mechanisms differed in the number of movements the horses elicited to open them (GLMM: N = 89, SE = 0.16, z = -2.24, p = 0.02,
The animals needed more movements to open a lock when they had to twist them (median = 8, min. = 4, max. = 10; GLM: N = 89, Std.E. = 1.77, z = 2.59, p = 0.0096), or move them in several directions within one movement plane (left–right, back—forth, vertical) or on several movement planes (linear and twisted): i.e. back or forth and left or right (median = 10, min. = 10, max. = 10; GLM: N = 89, Std.E. = 3.27, z = 2.22, p = 0.03), left or right and vertical (median = 4, min. = 2, max. = 10; GLM: N = 89, Std.E. = 0.89, z = 1.67, p = 0.09), vertical and twisted (median = 8.5, min = 7, max. = 10, GLM: N = 89, Std.E. = 2.22, z = 2.59, p = 0.009), and back or forth and left or right and vertical (median = 6, min. = 6, max. = 6; but this is not significant: GLM: N = 89, Std.E. = 2.59, z = 1.26, p = 0.2), rather than moving them only left or right (median = 3, min. = 1, max. = 10), only back or forth (median = 2.5, min. = 1, max. = 5) or only vertically (median = 2, min. = 1, max. = 8) (GLM: N = 89, all p > 0.05). The boxplots visualize the quartiles of the data per lock type. The box comprises 50% and the lower and upper whisker 25% of the variability each. The dots visualize outliers. ** = p < 0.01 and * = p ≤ 0.05.
Crowdsourcing resulted in a large sample of cases of door and gate opening in horses. Most of the horses opened only one door, gate or mechanism type at a single location. However, some individuals opened the same type of door or gate mechanism at several locations, some operated several types of mechanisms, and some were even able to open doors and gates secured with several mechanism types at different positions. These horses seemed to have understood
In 74 out of 257 cases, other animals in the same stable demonstrated the same behaviour, offering options for learning the manipulation of the same types of locked doors and gates through observation of conspecifics
A beneficial effect of experience was found: animals which opened more locked door and gate types applied fewer head movements and fewer movement sequences to opening locks than horses which opened only one door or gate type.
Horses favoured their mouth for prehension of door and gate mechanisms, which were, of course, made for human hand use. Similarly, most horses chose to use their mouth when pressing a button to open a feeding apparatus
In our survey, most horses were reported to open doors and gates on hinges, with bars or handles which could be opened on one plane with only a few head movements. However, an impressive number of horses handled more complicated mechanisms, which required movements in more than one plane and specific sequences of actions to be applied. In the main, horses applied similar numbers of movements to those needed by humans to open doors and gates, but twice as many when opening locks. The locks included carabiners, twist mechanisms, keys at doors and padlocks, and electric fence handles; the latter also needed to be handled precisely if possible electric shocks were to be avoided. Horses applied more movements the more complex the mechanism was, but their efficiency, in terms of the number of actions applied compared to the minimum necessary, was similar for opening simple and more complex mechanisms. The range of fastening devices that horses have learnt to open apparently spans the gamut of devices in frequent use in the nations participating in the study: thus, we found no obvious limit to the complexity which horses can learn to master
Door and gate opening in horses is not generally associated with the quality of human-imposed environmental conditions
We found individual differences in the number of movements and the efficiency in dealing with barred doors and gateways and mechanisms which needed twisting movements, which may have been the result of construction variations between the different mechanisms. In addition, the horses may have differed in their past opportunities to practice opening techniques
The low number of reports on trained door opening horses in the present study was not unexpected, as animal escapes may have serious consequences for the animal, the animal owners, caretakers and the environment, for example when they hit the traffic
As with any questionnaire, the present study ran the risk of collecting biased reports
Horses open a far wider range of human-made mechanical devices on doors and gates than previously reported, generally handling the mechanisms with their mouths. Although most horses are confined by simple bolts or handles, and most reports were of opening such devices, a surprising range of fastenings, including carabiners and electric fence handles, proved vulnerable to opening by horses confined by them. Indeed, within the range of locking devices in frequent use for restraining horses, we found no clear cognitive limit to horses’ ability to open them and some evidence for experience improving the horses’ skills. The ability of horses and other ungulates to open human-made fastenings therefore needs to be reconsidered to minimise damage caused by escapes.
a) data request: people were requested to send information on unusual behaviour (general questionnaire;
(PDF)
Please enlarge Pdf for viewing the data.
(PDF)
Please enlarge Pdf for viewing the data.
(PDF)
(PDF)
Please enlarge Pdf for viewing the data.
(PDF)
Please enlarge Pdf for viewing the data.
(PDF)
Please enlarge Pdf for viewing the data.
(PDF)
Please enlarge Pdf for viewing the data.
(PDF)
We wish to thank Kevin Laland and Adam Miklosi for discussion, three anonymous referees and the editor for helping to improve the manuscript, Michel-Antoine Leblanc and Anja Zollinger for helping with the French survey, Kate Farmer for correcting the language, Knut Krueger for helping with the figures, 2 Student project groups, Simone Weil, and Helena Hollenhorst for helping to raise the data, various journalists for spreading information about the project and all the horse, donkey and mule owners and keepers for sending us their reports, as well as two anonymous referees and two editors for helpful suggestions on the manuscript.