G Model

APPLAN-3916; No. of Pages 12 ARTICLE IN PRESS

Applied Animal Behaviour Science xxx (2014) xxx–xxx

Contents lists available at ScienceDirect

Applied Animal Behaviour Science

journal homepage: www.elsevier.com/locate/applanim

Behaviour of horses in a judgment bias test associated with

positive or negative reinforcement
Sabrina Briefer Freymond a,∗,1 , Elodie F. Briefer b,1 , Anja Zollinger a,c ,
Yveline Gindrat-von Allmen c , Christa Wyss a , Iris Bachmann a
a
Agroscope – Swiss National Stud Farm, Les Longs Prés, P.O. Box 191, CH-1580 Avenches, Switzerland
b
Institute of Agricultural Sciences, ETH Zürich, Universitätstrasse 2, 8092 Zürich, Switzerland
c
Bern University of Applied Sciences, Agricultural Sciences, Länggasse 85, 3052 Zollikofen, Switzerland

a r t i c l e i n f o a b s t r a c t

Article history: Moods can influence our judgment of ambiguous stimuli as positive or negative. Measuring
Accepted 13 June 2014 judgment bias in animals is a promising method to objectively assess their emotional states.
Available online xxx Our study aimed to develop a cognitive bias test in horses, in order to assess the effect of
training using positive reinforcement (PR) or negative reinforcement (NR) on their emo-
Keywords: tional states. We trained 12 mares to discriminate between a rewarded and a non-rewarded
Cognitive bias location situated on each side of a paddock. The mares were then trained during five days to
Equus caballus
perform several exercises using PR (n = 6) for one group, and NR (n = 6) for the other (treat-
Emotions
ment). Finally, we compared the responses of the two groups to three ambiguous locations
Optimism
Training method situated between the rewarded and non-rewarded locations (judgment bias test). During
the training exercises, according to our predictions, behavioural measures suggested that
NR mares experienced more negative emotions than PR mares. Surprisingly, the results of
the judgment bias test suggest that NR mares were in a more optimistic mood compared to
PR mares, despite previously experiencing more negative emotions during the treatment.
NR mares could have been more motivated to obtain a food reward than PR mares, which
had been rewarded throughout the treatment phase. Alternatively, NR mares could have
developed optimistic bias triggered by release from the negative state experienced during
treatment. This first attempt to test judgment bias in horses suggests that this is a promis-
ing method to measure horse mood. Knowledge about the effect of training methods on
the mental health of domesticated animals can add a new dimension to animal welfare, in
order to promote better ways to work with animals.
© 2014 Elsevier B.V. All rights reserved.

1. Introduction and Paul, 2004), and animal welfare is nowadays assessed

via both physical and mental health (Dawkins, 2008).
Public concern about animal welfare is strongly based Negative emotional states, and particularly mood disor-
on the attribution of mental states to animals (Mendl ders, should be minimized. Good welfare thus implies a
reduction in animal suffering and other negative emo-
tions (e.g. fear; Forkman et al., 2007). In recent years,
the importance of promoting positive emotions has also
∗ Corresponding author. Tel.: +41 58 482 61 01.
been suggested (Boissy et al., 2007). However, ani-
E-mail address: [email protected]
(S. Briefer Freymond). mal welfare legislation in many countries focusses only
1
These authors contributed equally to this work. on how animals are housed and not on how people

http://dx.doi.org/10.1016/j.applanim.2014.06.006
0168-1591/© 2014 Elsevier B.V. All rights reserved.

Please cite this article in press as: Briefer Freymond, S., et al., Behaviour of horses in a judgment bias test associated with
positive or negative reinforcement. Appl. Anim. Behav. Sci. (2014), http://dx.doi.org/10.1016/j.applanim.2014.06.006
G Model
APPLAN-3916; No. of Pages 12 ARTICLE IN PRESS
2 S. Briefer Freymond et al. / Applied Animal Behaviour Science xxx (2014) xxx–xxx

interact with them (e.g. during training). Few studies have (Canis lupus Mendl et al., 2010b) and goats (Capra hircus;
investigated the impact of human interactions on animal Briefer and McElligott, 2013; see also review Mendl et al.,
mental health (Hausberger et al., 2009; Sankey et al., 2010a, 2009).
2010b). The mental health of domestic animals is likely not
Emotions are intense but short-lived affective reac- only affected by their housing conditions (Bateson and
tions to events, and have an important function for Matheson, 2007; Brydges et al., 2011; Douglas et al., 2012;
animals’ lives. They act as “detectors” to identify situa- Matheson et al., 2008), but also by the way that people
tions and select appropriate behavioural decisions (Mendl behave towards them. Specifically, the way animals are
et al., 2010a, Nettle and Bateson, 2012). Emotions can trained (e.g. shepherding in dogs, riding in donkeys (Equus
be defined using two dimensions: their arousal (bodily asinus), horses (Equus caballus) or camels (Camelus spp.))
activation or excitation) and their valence (positive or could influence their mood (Hall et al., 2008; Hausberger
negative), and four components: cognitive, behavioural, et al., 2009). This may especially be true for horses,
neurophysiological and subjective component (Keltner and which are trained for various tasks (e.g. jumping, dres-
Lerner, 2010; Scherer, 1984). The subjective component sage) on a regular basis. For instance, controversial or
of emotions cannot yet be shown to occur in animals. unclear commands from the rider can lead to frustra-
However, cognitive, behavioural and neurophysiological tion, neurosis and even chronic stress in horses, resulting
components are increasingly used as indicators of animal in an apathetic state or stereotypies (Hausberger et al.,
emotions (Mendl et al., 2010a). The behavioural param- 2009).
eters that can be affected by emotions (among other Animals can be trained using positive (PR) or neg-
factors) include body, tail and ear postures, movements and ative reinforcement (NR). PR consists of rewarding the
also vocalizations (e.g. Boissy et al., 2011; Briefer, 2012; animal, typically with food, when it performs the requested
Reefmann et al., 2009; Young et al., 2012). Many studies behaviour. Conversely, NR consists of applying an aversive
have revealed indicators of emotional arousal, and these stimulus and ending it when the requested behaviour is
largely focused on arousal during situations of negative achieved (Skinner, 1938). NR is often mistaken for pun-
rather than positive valence. In contrast, the evidence sug- ishment and, if not carefully applied, this training method
gests that it might be difficult to find clear indicators of can trigger fear (Waran et al., 2002). This method is more
valence (Mendl et al., 2010a). Yet, indicators of valence commonly used in animal training than PR (Innes and
would be especially useful to differentiate between neg- McBride, 2008). Recent evidence suggests that training
ative (e.g. behavioural despair) and positive (e.g. relaxed using PR triggers a more positive perception of humans
animal) states of low arousal, which are characterized than NR (Sankey et al., 2010a). When PR is used, the
by quite similar physiological and behavioural expres- animal learns a requested behaviour through motivation
sions (e.g. lack of movement, low head posture; Fureix to obtain and consume a reward. This type of train-
et al., 2012; Hall et al., 2008). Therefore, more research ing thus involves appetitive-anticipation, consumption
is needed to highlight indicators of emotional states, and and post-consumption of a reward, which can induce a
particularly of emotional valence, in order to assess animal positive emotional state (e.g. satisfaction/contentment)
welfare. if the anticipated reward is never unexpectedly absent
One promising method to differentiate between pos- (Burman et al., 2011; Moe et al., 2009; Spruijt et al.,
itive and negative emotional states is the cognitive bias 2001).
approach (Désiré et al., 2002; Mendl et al., 2009; Paul et al., This study investigated differences in behavioural
2005). This method assesses the valence of long-term emo- measures that could suggest underlying emotional states
tional states (i.e. mood) through their impact on cognitive between horses trained using PR and NR. We also inves-
processes (i.e. attention, learning, memory and decision- tigated if these two training methods could induce more
making). Moods are diffuse emotional states that arise as a long-term changes in emotional states (moods). We
result of an accumulation of shorter term emotional states hypothesized that training with PR would trigger posi-
(Mendl et al., 2010a, Nettle and Bateson, 2012). In humans, tive emotions linked to the anticipation and consumption
individuals in positive moods will be biased towards posi- of food rewards (e.g. Moe et al., 2009). By contrast, we
tive events and be more attentive to them, learn them more hypothesized that NR would trigger negative emotions,
easily, memorize them better and expect them more often or at least less positive emotions than PR, as this train-
than negative events (“optimistic bias”), whereas the oppo- ing method involves the cessation of a negative stimulus,
site happens for people in negative moods (“pessimistic which is presumably less positively rewarding than a food
bias”; MacLeod and Byrne, 1996; Sharot, 2011; Strunk et al., reward. Then, we tested horse mood using a spatial judg-
2006; Wright and Bower, 1992). In non-human animals, ment task (Burman et al., 2008a, 2009). Given that mood
“pessimism” and “optimism” can be assessed by examin- states arise as a result of an accumulation of shorter-term
ing how animals perceive “ambiguous” stimuli that are emotions (Mendl et al., 2010a), we also expected that
intermediate between two “reference” stimuli, such as a horses trained with PR would show more optimistic bias,
stimulus known to be rewarded and another one known to indicating more positive mood than horses trained with
be unrewarded or aversive (“judgment bias”; Harding et al., NR. Knowledge about the effects of training methods on
2004; Mendl et al., 2009). This judgment bias technique has the mental health of domesticated animals could add a new
now been successfully tested in several species, from hon- dimension to animal welfare, in order to promote not only
eybees (Apis mellifera carnica; Bateson et al., 2011), to Euro- better housing systems, but also better ways to work with
pean starlings (Sturnus vulgaris; Brilot et al., 2010), dogs animals.

Please cite this article in press as: Briefer Freymond, S., et al., Behaviour of horses in a judgment bias test associated with
positive or negative reinforcement. Appl. Anim. Behav. Sci. (2014), http://dx.doi.org/10.1016/j.applanim.2014.06.006
G Model
APPLAN-3916; No. of Pages 12 ARTICLE IN PRESS
S. Briefer Freymond et al. / Applied Animal Behaviour Science xxx (2014) xxx–xxx 3

Table 1 2.2. Experimental design

Characteristics of the horses used in the experiment (age and breed;
FM = Franche-Montagne, TR = Trotter; SHB = Swiss half-bred), along with
the treatment group (PR = positive reinforcement, NR = negative reinforce-
The experiment consisted of several phases (Table 2):
ment). The side (L = left, R = right) on which they were rewarded during the (1) the habituation phase (described in Section 2.3.2),
judgment bias experiment (positive location) is also indicated (rewarded aimed at habituating the horses to the judgment bias set-
side). up; (2) the judgment bias training phase (described in
Mare Treatment Rewarded Age Breed Section 2.3.3), during which the mares learned to dis-
group side criminate the side where they were rewarded with food
1 PR R 11 SHB (“positive location”; left or right depending on the horse)
2 PR R 12 SHB from the non-rewarded side (“negative locations”; oppo-
3 PR R 17 SHB site direction); (3) the treatment (described in Section 2.4),
4 PR L 9 TR during which mares were trained to perform several exer-
5 PR L 12 FM
6 PR L 20 SHB
cises using either positive or negative reinforcement; (4)
a reminder session for the judgment bias test (described
7 NR R 11 FM
in Section 2.3.4); and (5) the judgment bias testing phase
8 NR R 12 TR
9 NR R 13 SHB (described in Section 2.3.4), during which mares were
10 NR L 10 FM exposed to ambiguous locations situated between the pos-
11 NR L 15 SHB itive and negative reference locations.
12 NR L 15 TR

2.3. Judgment bias

2. Methods We used a judgment bias test, which uses spatial loca-

tion as a stimulus (Burman et al., 2008a, 2009), to assess
2.1. Subjects and management conditions horse mood following treatment.

The study was carried out at the Swiss National Stud 2.3.1. Experimental setup
Farm, Avenches, in February 2012. Subjects were 12 mares The judgment bias experimental setup was adapted
of three different breeds (three Franches-Montagnes, three from Burman et al. (2008a) and Doyle et al. (2011, 2010).
Trotters and six Swiss half-bred) and aged 9–20 years Experiments were carried out in a paddock with rubber
(Table 1). They had been at the National Stud Farm for at mats on the ground (28 m × 12 m; Fig. 1). The horses had to
least two years and they were not used for riding. They cross the paddock, in order to reach a black bucket (50 cm
were all previously owned horses that retired to breeding, wide and 30 cm deep), which always contained food (car-
after withdrawal from sport. They were used for breed- rots, corn and oat) when it was situated at the positive
ing as surrogate (foster) mares, but were not pregnant or location (always rewarded), or never contained food when
manipulated for breeding purpose during the time of the it was situated at the negative location or at the ambiguous
study. locations (never rewarded). The ambiguous locations were
At all times, the 12 mares, which were split into two situated at intermediate distances between the positive
treatment groups (PR and NR, see Section 2.4), were housed and negative locations (ambiguous +, middle and ambigu-
all together in a large group housing system, which could ous −; Fig. 1). They were never rewarded in order to avoid
normally contain up to 14 horses, with several compart- any association between the location and the presence of
ments. Routine care of the study animals was provided by a food reward (Burman et al., 2009). The paddock was split
employees of the National Stud Farm, who were blind to in three main areas, which had been traced on the ground
the purpose of the study and to the treatment groups. They using a coloured spray: (1) an 8 m long start area delim-
were fed in foraging stalls with haylage and a mix of corn ited by the starting line; (2) a 15 m long central area that
and oats twice per day, 1 h before the experiments in the the horses had to cross to reach the bucket; and (3) a 3 m
morning, and after the experiments in the evening. They long end area, delimited by a stop line and divided into five
had ad libitum access to straw in a rack and in laying areas, smaller areas corresponding to the various locations (pos-
as well as water through an automatic water dispenser. itive, ambiguous +, middle, ambiguous −, negative). The
Mares had access to an outside field 2–3 times per week bucket was placed 90 cm behind the stop line and 90 cm
depending on weather conditions. All subjects were lead from the right and left lines delimitating each of the five
from their home pen to the paddocks used for the treatment areas corresponding to the test locations (Fig. 1). At any of
and judgment bias on a halter and rope. the locations, the bucket was thus situated at 15.9 m from

Table 2
Description of the five phases of the study, along with duration and aims.

Phases Duration (day) Aim

1 Habituation to the judgment bias set-up 1 Habituation

2 Training phase of the judgment bias experiment 3 Discrimination between rewarded and non-rewarded locations
3 Treatment 5 Inducing different moods using negative or positive reinforcement
4 Reminder session for the judgment bias experiment 1 Reminding reference locations
5 Testing phase of the judgment bias experiment 2 Assessing horse mood following treatment

Please cite this article in press as: Briefer Freymond, S., et al., Behaviour of horses in a judgment bias test associated with
positive or negative reinforcement. Appl. Anim. Behav. Sci. (2014), http://dx.doi.org/10.1016/j.applanim.2014.06.006
G Model
APPLAN-3916; No. of Pages 12 ARTICLE IN PRESS
4 S. Briefer Freymond et al. / Applied Animal Behaviour Science xxx (2014) xxx–xxx

Fig. 1. Experimental setup for the judgment bias experiment. Position of the positive bucket (rewarded; right or left depending on the horses), the negative
bucket (non-rewarded; opposite side as the rewarded one), the ambiguous buckets (equidistant angles between the positive and negative buckets), the
waiting, start, central and end areas. The latency to reach the location was measured as the time to go from the starting line to the stop line next to the
tested bucket.

the middle of the starting line, where the horse was placed and one negative (i.e. ++−−+−). During the following ses-
at the beginning of the trial. The bucket was always covered sions, we used a pseudo-random sequence with no more
with a 75 cm long and 1.5 cm thick wooden lid, in order to than two consecutive positive or negative trials, and with
prevent any visual and olfactory cues indicating the loca- the same number of positive and negative trials per session
tion of the food reward (Briefer and McElligott, 2013). (e.g. ++−+−−, −++−−+ or +−−++−). Each mare from the PR
group was randomly paired with a mare from the NR group
2.3.2. Habituation and these two mares received the same sequence over the
On the first day of the experiments, horses were habitu- three training days. The training ended after three days,
ated individually to the experimental paddock twice, once when a significantly shorter latency to reach the positive
in the morning and once in the afternoon (two habitua- location than the negative one had been obtained during
tion sessions per horse). They were taken individually from two following days (Fig. 2).
their home enclosure to the paddock on a halter and rope.
They were walked around the paddock and released there 2.3.4. Judgment bias testing
for 2 min. The black bucket used for the judgment bias The judgment bias testing took place after the treat-
experiment was placed in the start area (Fig. 1) and the ment phase (PR and NR; described in Section 2.4). The
horse was allowed to eat from it. The experimenter then day after the end of treatment phase, and prior to the
placed the wooden lid on the bucket and allowed the horse judgment bias test, each horse received one session of four
to eat again, which enabled them to learn how to remove trials, as a reminder. During this session, all mares received
the lid in order to access the food. During the following
20 min (morning session) or 10 min (afternoon session), the
horse was walked by the experimenter on a halter and rope
to the other end of the paddock and released to allow the
horse to go back to the bucket, remove the lid and feed.

2.3.3. Judgment bias training

The judgment bias training lasted three days and
consisted of training the mares to discriminate between
the positive and negative locations situated on each side of
the paddock. Half of the horses assigned to the PR treatment
group (n = 3 horses) and half of the horses assigned to the
NR treatment group (n = 3 horses) were trained to expect
food on the left (positive side = left of the end area). The
other half (n = 3 PR and 3 NR mares) were trained to expect
food on the right (positive side = right of the end area;
Table 1; Fig. 1). The training procedure was adapted from
Burman et al. (2009) and Briefer and McElligott (2013).
On each training day, the mares received one session of Fig. 2. Results of the judgment bias training phase. Latency to reach
six training trials: three positive and three negative. To the rewarded location (+) and the non-rewarded location (−) over the
three days of training for the treatment group subsequently trained with
facilitate learning, during the first session of six training
positive reinforcement (black) and the group subsequently trained with
trials (first training day), all mares received two positive negative reinforcement (grey; mean ± SE per group and training day; n = 6
trials, followed by two negative trials, then one positive horse per group).

Please cite this article in press as: Briefer Freymond, S., et al., Behaviour of horses in a judgment bias test associated with
positive or negative reinforcement. Appl. Anim. Behav. Sci. (2014), http://dx.doi.org/10.1016/j.applanim.2014.06.006
G Model
APPLAN-3916; No. of Pages 12 ARTICLE IN PRESS
S. Briefer Freymond et al. / Applied Animal Behaviour Science xxx (2014) xxx–xxx 5

two positive trials, followed by two negative trials (i.e. day, 13/108 occasions during negative training trials, 17/48
++−−), to facilitate memory. The judgment bias testing occasions during negative testing trial and 3/12 occasions
phase was carried out on the two following days. On when testing the ambiguous location next to the negative
each testing day, each mare received one session of seven one. During the first training day, if the horse did not reach
testing trials. Each session consisted of two trials with the the positive location, Exp 1 brought the horse to the bucket
positive location, two trials with the negative location, and and showed it the food. At the end of the daily session of
one trial with each of the ambiguous location. Following 6 (training) or 7 (testing) trials, the subject was taken back
Briefer and McElligott (2013), each session started with to the home enclosure by Exp 1, while Exp 2 was refilling
one trial with the positive location and one trial with the the positive bucket for the next horse.
negative location, or the opposite, as a reminder. Then,
mares were tested with the three ambiguous locations 2.4. Treatments
in a random order, interspersed by the positive and the
negative locations. For each mare, the order of testing with 2.4.1. Training exercises
the ambiguous, positive and negative locations was coun- To study the effect of PR or NR on horse mood, the horses
terbalanced over the two days, so that each ambiguous were trained, during five days to perform several exercises
location (“A”) would be tested one day after the positive using only PR for one group (“PR group”) and NR for the
location, and the other day after the negative location (i.e. other group (“NR group”). The training was carried out in a
day 1 = +−A3 + A2 − A1; day 2 = −+A1 − A2 + A3). Pairs of different paddock than the one used for the judgment bias
mares from the PR and NR groups, which had been allocated experiment. It was situated close to the home enclosure,
to the same sequence of trials during the training phase, in order to avoid stress linked to isolation. The treatment
also received the same sequence over the two testing days. started on the day after the judgment bias training phase
and lasted for five days, after which the reminder and the
2.3.5. Judgment bias training and testing procedure judgment bias testing phase were carried out (Table 2).
Two experimenters were present during the training Each day, horses received three sessions (days 1 and 2)
and testing phases. The first experimenter (Exp 1), who was or two sessions (days 3–5) of 5 min each, interspersed by
blind to the side at which subjects were rewarded (right or 2.5 min breaks (12.5–20 min per horse each day includ-
left), walked the horse on a halter and rope from the home ing breaks). Each day, all the horses from one group were
pen to the paddock, while the second experimenter (Exp 2) trained, followed by all the horses from the other group, in
prepared the first trial by placing the bucket with or with- a counterbalanced order (PR–NR or NR–PR depending on
out food, covered with a lid, at the appropriate location for the days). All the horses were trained by the same exper-
the trial. Two black buckets, identical in size and shape, imenter (Exp 2), while Exp 1 was in charge of moving the
were used for the experiment; one bucket was filled with horses to the training paddock and back, and collecting the
food and used for all the positive trials and one bucket was data during the training.
empty and used for all the negative and ambiguous trials. The exercises used for the treatment were designed to
For each horse, the positive bucket was fully filled with be easily trained using either PR or NR. Exp 2 trained all
food at the beginning of the session and the food quantity horses to perform the same following sequence of exer-
lasted for the whole session. For each trial, Exp 2 always cises on demand, while standing afoot on the left side of
presented only one of the two buckets at the appropri- the horse;
ate location, while the other bucket was placed outside
the paddock, out of sight of the tested horse. Exp 1 then 1) Stand still next to the trainer, with the head forward.
entered the paddock, released the horse in the middle of 2) Walk next to the trainer.
the start line and stood still during the whole time, head 3) Stop next to the trainer.
down, while the horse was going away from her, towards 4) Move backward.
the buckets (Fig. 1). Both experimenters then waited for the 5) Flex the neck alternatively on the right and on the left.
horse to access and eat a few bites of the food (positive loca- 6) Lower the head down towards the ground.
tion) or to reach the appropriate area where the bucket was 7) Walk on a tarp (1.5 m × 2.5 m) that was covered by
located (negative and ambiguous locations). A 2 min inter- plastic bottles.
trial interval followed, during which Exp 1 returned the 8) Climb on a wooden podium (70 cm large × 70 cm
horse to the waiting area (Fig. 1). While the horse was facing long × 50 cm high).
the opposite side, Exp 2 placed the bucket with or without 9) Push a blue balloon (diameter 1 m) with the nose or
food at the appropriate location for the next trial. During leg.
each trial, Exp 2 scored the latency to reach the bucket as 10) Lift one of the front legs upward.
the time from when one of the horse’s front legs passed
the start line, until one of its front legs passed one of the Each day, this sequence of exercises was trained or
lines (stop line or lines on each side of the bucket) delin- rehearsed (when successfully learned) in the same order
eating the appropriate area where the bucket was located for all horses. Each exercise was trained during 5 min max-
(Fig. 1). If the horses did not reach the area where the bucket imum, unless well executed, before moving to the next
was situated, it was brought back to the start area after one, and so on, until the end of the daily training session.
3 min by Exp 1, and the training/testing session continued. An exercise was considered to be well executed when
This happened on 6/108 occasions during a positive trial on easily performed on demand. Depending on individual per-
the first (five occasions) or second (one occasion) training formances, and because the daily training sessions were

Please cite this article in press as: Briefer Freymond, S., et al., Behaviour of horses in a judgment bias test associated with
positive or negative reinforcement. Appl. Anim. Behav. Sci. (2014), http://dx.doi.org/10.1016/j.applanim.2014.06.006
G Model
APPLAN-3916; No. of Pages 12 ARTICLE IN PRESS
6 S. Briefer Freymond et al. / Applied Animal Behaviour Science xxx (2014) xxx–xxx

limited in duration for each horse, the number of exercises of score 1 for head level). Scores of 2 for all behaviours
performed on each of the training day could differ between indicated an intermediate between negative and positive
horses. A daily session always ended up with the execu- states (Table 3). Therefore, we calculated a general score
tion of the last exercise in the sequence that was correctly per horse for each of these behaviours as follows: (% score
performed by the horse. 1 × 1) + (% score 2 × 2) + (% score 3 × 3). According to this
equation, a horse would obtain a general score of 300 if it
2.4.2. Positive and negative reinforcement always scored 3 for one behaviour, a general score of 200
All horses were trained individually, unmounted, on a if it always scores 2, and a general score of 100 if it always
halter and rope. The same procedure was applied for both scores 1 (range = 100–300; for the example above, the gen-
PR and NR groups (Heleski et al., 2008; Hendriksen et al., eral score would be (4.3 × 1) + (2.1 × 2) + (93.6 × 3) = 289.4).
2011; Innes and McBride, 2008; Sankey et al., 2010a); each Higher scores for body tension (more motivation), for atti-
movement towards a correct execution of the exercise was tude towards trainer (more contact with the trainer) and
immediately rewarded (“shaping procedure”; McGreevy for ear position (ears more forward) thus suggested a more
and McLean, 2010). For the PR group, the reward was a often positive emotional state, while higher scores for head
click using a clicker, followed by a piece of food (food pel- level (head higher) suggested a more often negative emo-
let for horse; “clicker method”; Kurland, 1999). To guide tional state (Briefer Freymond et al., 2013; McDonnell and
their movement, PR horses were trained to follow a target Haviland, 1995; Rietmann et al., 2004; Visser et al., 2009).
(55 cm long stick, with a tennis ball at the end), by initially
rewarding them when they were touching the target with 2.5. Data analyses
their nose. Conversely, for the NR group, the reward was the
cessation of an uncomfortable stimulus. The stimulus was a For the behavioural measures taken during the treat-
gentle pressure on the halter, gently shaking or pulling the ment phase (training exercises), the analyses were carried
lead or/and a gentle pressure on the horse’s flank using a out on the average values for each subject and for each exer-
stick, depending on the exercise. The intensity of the stimu- cise. All horses reached exercise 7, eight horses from both
lus was gradually increased if the horse was not responding groups (three PR and five NR) reached exercise 8, and two
to it or not responding correctly. As soon as the expected NR mares reached exercise 9 (exercise reached per horse:
response was shown, the experimenter stopped the stim- range = exercise 7–9; mean = exercise 7.83 ± 0.21; n = 12
ulus immediately and the horse could rest for a short break horses). We therefore decided to focus on behavioural
(5 s) before continuing the training. The movements of NR measures collected during exercises 1–8, which contained
horses were guided using such gradual pressure intensity, data from both groups, for our analyses.
which was removed as soon as they moved in the correct For the judgment bias training phase, the analyses were
direction. Exp 2 followed a detailed training procedure that carried out on the average latency for each subject to reach
was identical for all horses of a given group. The two groups the positive and negative location on each of the three
received the same amount of training (mean ± SE total training day. The data on the reminder day (day after the
training time per horse and per exercise (exercises 1–9): PR, treatment and before starting the test; Table 2) were ana-
485.56 ± 52.73 s; NR, 441.02 ± 37.01 s; linear mixed-effect lysed separately. For the judgment bias testing phase, the
model, with horse identity nested within exercise number analyses were carried out on the latency to reach each
as a random factor, and group as a fixed factor: F1,84 = 0.81, of the locations (ambiguous, positive and negative), aver-
p = 0.37). aged over the two test days for each subject. If a mare did
not reach the area where the bucket was situated during
2.4.3. Behavioural measures the training or testing phases of the judgment bias experi-
All training sessions were video recorded using ment, we attributed, for the trial, a latency corresponding
a camcorder (Sanyo VPC-WH1, Waterproof, High- to the maximum time taken by this individual, over the
Definition). From the videos, the following behaviours days of training and testing, to reach any of the locations.
were scored, based on a 1–3 scale (Table 3), and using This allowed us to avoid replacing these values by either
an instantaneous time sampling every 10 s; body tension, missing data or by an artificial maximum of 3 min that had
attitude towards the trainer, head level and ear position. been decided by us during the planning of the experiments.
Exercises varied in duration between horses, and also It also maintained the individual variability in latencies to
within horse between the training days. Thus, we calcu- reach the locations (Briefer and McElligott, 2013).
lated the percentage of time each horse was given a score We analysed behavioural measures from the treatment
of 1, 2 or 3 on the established scale for each behaviour, by (training exercises), and the latency data from the training
dividing the occurrence of each score by the number of and testing phases of the judgment bias using linear mixed-
10 s samples for the whole exercise (e.g. two occurrences effects models (LMM; lme function in R; Bates, 2005). This
of score 1, one occurrence of score 2 and 44 occurrences allowed us to investigate or control for the effect of sev-
of score 3 = 4.3% for score 1, 2.1% for score 2 and 93.6% for eral factors (age, breed, training side, training day, group,
score 3). For each parameter, the scale was designed as location). For the treatment phase, the initial models for
a continuum between a behaviour suggesting a negative the behavioural measures included body tension score,
emotional state (score of 1 for body tension, attitude attitude towards trainer score, head level score or ears posi-
towards trainer and ear position, and score of 3 for head tion score calculated for the horses as a response variable
level) and a positive emotional state (score of 3 for body (four different models), as well as the breed (Franches-
tension, attitude towards trainer and ear position, and Montagnes, Trotters and Swiss half-bred) and age (9–20)

Please cite this article in press as: Briefer Freymond, S., et al., Behaviour of horses in a judgment bias test associated with
positive or negative reinforcement. Appl. Anim. Behav. Sci. (2014), http://dx.doi.org/10.1016/j.applanim.2014.06.006
G Model
APPLAN-3916; No. of Pages 12 ARTICLE IN PRESS
S. Briefer Freymond et al. / Applied Animal Behaviour Science xxx (2014) xxx–xxx 7

Table 3
Description of the scale used to describe the behavioural parameters during the training exercises.

Parameter Score

1 2 3

Body tension Negative tension (tensed, Indifferent (calm, but lack of Positive tension (motivated, responds
nervous, scared) motivation, relaxed hind leg) to stimulations, focused)
Attitude towards Avoids contact, moves away Keeps constant distance with Tries to establish contact, comes closer
the trainer from trainer the trainer to the trainer
Head level Low (nose below the belly Intermediate (nose between High (nose at or above elbow level)
level) the belly and middle-elbow
level)
Ear position Ears backward Ears asymmetrical Ears forward

of the mares as fixed factors to control for age and breed 3. Results
differences. The training group (PR or NR) and the exer-
cise number (Ex1–Ex8) were included with the interaction 3.1. Judgment bias training
between them, as fixed factors, to test for their effect. For
the judgment bias experiment, models for both the training Mares went faster to the positive (rewarded) location
and testing phases included the latency to reach the var- than to the negative (non-rewarded) location, and thus suc-
ious locations as a response variable, as well as the age, cessfully learned the task (Fig. 2). There was no significance
breed and rewarded side (left or right) of the mares as difference between the two treatment groups (PR and NR)
fixed factors, in order to control for age and breed differ- in their latencies to reach the locations (Fig. 2).
ences and for potential laterality bias (e.g. De Boyer Des The model selection procedure for the training session
Roches et al., 2008). The day of training (1–3; for the train- revealed an effect of day of training on the latencies to
ing phase only), the location (positive and negative for the reach the locations (LMM: F1,58 = 6.30, p = 0.015; Fig. 1).
training phase; positive, negative, ambiguous +, middle and The general latency to reach the locations changed over
ambiguous − for the testing phase) and the group were the training phase (latency: day 1 = 35.16 ± 4.10 s; day
included, with all possible interactions between them, as 2 = 16.95 ± 2.98 s; day 3 = 21.94 ± 5.51 s; n = 12 horses).
fixed factors, in order to test for their effect. Finally, for all The effect of the treatment group was not significant
models, the identity of the mares was included as a random and was removed during model selection (likelihood-
factor to control for repeated measurements of the same ratio test: 2 1 = 0.37, p = 0.54). The latencies of PR mares
subjects. were therefore not different to the latencies of NR
We then removed non-significant terms using a mares during training (Fig. 2). There was a general
standard model simplification procedure. Each non- effect of location on the latencies (LMM: F1,58 = 21.97,
significant term was removed if the deletion did not cause p < 0.0001); mares reached the positive location substan-
any significant reduction in goodness of fit. The two models tially faster (latency = 14.56 ± 2.46 s) than the negative
with and without each term, both fitted with the maximum location (latency = 34.81 ± 4.00 s; n = 12 horses; Fig. 2).
likelihood method (ML), were compared using a likelihood The control factors (rewarded side, age and breed)
ratio test. We present the results after model simplification and the interaction terms included in the initial model
and with restricted maximum likelihood method (RELM). that are not mentioned above (day–location–group;
When the interaction effect between two factors was sig- location–group; day–group; day–location) did not have
nificant, further post hoc comparisons were carried out significant effects on the latencies to reach the locations.
using LMM including control factors that remained in the They were thus removed during model selection.
final models. The residuals of the models were checked On the reminder day, horses could still correctly
graphically for normal distributions and homoscedasticity, discriminate between the two reference locations, as
and were log-transformed if necessary (body relaxation shown by the significantly shorter latency to reach
score, attitude towards trainer score and latencies during the positive location (5.57 ± 0.41 s) than the negative
the reminder day and during testing). Statistical analyses one (35.38 ± 10.97 s; n = 12 horses; LMM: F1,11 = 17.04,
were carried out using R v. 2.15.0 (R Development Core p = 0.0017). The effect of the treatment group was not
Team, 2012). The significance level was set at ˛ = 0.05. No significant and was removed during model selection
Bonferroni correction was applied for the post hoc compar- (likelihood-ratio test: 2 1 = 1.74, p = 0.19). The control fac-
isons due to the small sample sizes (Nakagawa, 2004). All tors (training side, age and breed) and the interaction term
means are given with standard errors. between group and location were not significant either and
were also removed during model selection.

2.6. Ethics 3.2. Behavioural measures during treatment

Animal care and all experimental procedures were Differences between treatment groups were consistent
in accordance with the International Society for Applied for all variables; PR mares showed a more motivated body
Ethology guidelines. tension, a more positive attitude towards the trainer, a

Please cite this article in press as: Briefer Freymond, S., et al., Behaviour of horses in a judgment bias test associated with
positive or negative reinforcement. Appl. Anim. Behav. Sci. (2014), http://dx.doi.org/10.1016/j.applanim.2014.06.006
G Model
APPLAN-3916; No. of Pages 12 ARTICLE IN PRESS
8 S. Briefer Freymond et al. / Applied Animal Behaviour Science xxx (2014) xxx–xxx

Fig. 3. Behavioural responses to the treatment. (a) Body tension score (residuals of log-transformed scores controlled for age and breed), (b) attitude
towards trainer score (residuals of log-transformed scores controlled for age and breed), (c) head level score and (d) ear position score measured during the
training exercises 1–8 with positive reinforcement (black) and negative reinforcement (grey; mean ± SE per group and per exercise; 12 horses for exercises
1–7 and 8 horses for exercise 8). More positive body tension score’s residuals indicated a more positive tension (+) and more negative residuals a more
negative/indifferent tension (−). Similarly, for the attitude towards trainer score, more positive residuals indicated a more positive attitude (+) and more
negative residuals a more negative attitude (−). High head level scores indicated a higher position of the head and lower scores a low position. Finally,
high ear position scores indicated that the ears were more often forward and low scores that the ears were more often backward. Differences between
the two treatment groups (negative and positive reinforcement) for each exercise are indicated as follows (when the interaction term between group and
exercise was significant): *p < 0.05; **p < 0.01; ***p < 0.001; non-significant otherwise (linear mixed-effect models). Dotted lines indicate mean ± SE across
the exercise for horses trained with negative reinforcement (grey) and positive reinforcement (black).

lower head position and more forward ears than NR mares 3.2.2. Attitude towards trainer
(Fig. 3). There was a significant interaction effect between exer-
cise and treatment group for the attitude towards trainer
score (LMM: F7,66 = 4.70, p = 0.0003). Post hoc comparisons
3.2.1. Body tension revealed that PR mares had higher attitude towards trainer
There were no significant differences between the scores, indicating a more positive attitude than NR mares,
two treatment groups, depending on the type of exer- for all exercises (Fig. 3b). Concerning the control factors,
cise, for the body tension score (LMM: interaction effect, breed had a significant effect on attitude towards trainer
F7,66 = 1.88, p = 0.086). However, this interaction term could scores (LMM: F2,7 = 30.10, p = 0.0004). Age did not have
not be removed during the model selection, because a significant effect in the final model (LMM: F1,7 = 2.40,
doing so caused a significant reduction in goodness of p = 0.16), but this term could not be removed during the
fit (likelihood-ratio test: 2 1 = 14.54, p = 0.042). The model model selection, because its deletion caused a signifi-
selection procedure did not reveal any significant variation cant reduction in goodness of fit (likelihood-ratio test:
in body tension between the exercises (LMM: F7,66 = 0.73, 2 1 = 4.39, p = 0.036).
p = 0.64; Fig. 3a). There was a general effect of the group,
with PR mares having higher body tension scores, indi-
cating more motivated body tension, than NR mares 3.2.3. Head level
(LMM: F1,7 = 246.64, p < 0.0001; Fig. 3a). The two con- There were significant differences between the two
trol factors had a significant effect on body tension groups depending on the type of exercise for head level
scores (LMM: age, F1,7 = 6.38, p = 0.040; breed, F2,7 = 11.58, scores (LMM: interaction term, F7,66 = 5.48, p = 0.0001). Post
p = 0.006). hoc comparisons revealed that the head level scores of PR

Please cite this article in press as: Briefer Freymond, S., et al., Behaviour of horses in a judgment bias test associated with
positive or negative reinforcement. Appl. Anim. Behav. Sci. (2014), http://dx.doi.org/10.1016/j.applanim.2014.06.006
G Model
APPLAN-3916; No. of Pages 12 ARTICLE IN PRESS
S. Briefer Freymond et al. / Applied Animal Behaviour Science xxx (2014) xxx–xxx 9

mares were significantly lower than the scores of NR mares significant difference between the latencies taken by PR
for all exercises, except exercises 5 (flex the neck) and 6 and NR mares to reach the other locations (Fig. 4).
(lower the head; Fig. 3b). Age and breed (control factors) Age did not significantly affect the latencies to reach the
did not significantly affect head level scores and they were locations (LMM: F1,9 = 2.14, p = 0.18), but this term could
removed during model selection. not be removed during model selection, because doing so
caused a significant reduction in goodness of fit (likelihood-
ratio test: 2 1 = 3.86, p = 0.049). The other control factors
3.2.4. Ear position
(breed and training side) included in the initial model did
The interaction term between treatment group and
not have a significant effect on the latencies to reach the
exercise was not significant and was removed during
locations, and were removed during model selection.
model selection for ear position score (likelihood-ratio test:
2 1 = 9.16, p = 0.24). Ear position scores varied significantly
4. Discussion
between exercises (LMM: F7,73 = 8.68, p < 0.0001; Fig. 3d).
There were significant differences between the two groups;
Knowledge about the effect of training methods on
The ears of PR mares were forward more often than the
the emotional states of domesticated animals could help
ears of NR mares (LMM: F1,10 = 88.69, p < 0.0001; Fig. 3d).
improve welfare, not only in terms of how animals are
The two control factors, age and breed did not significantly
housed, but also how people interact with animals. We
affect ear position scores and were removed during model
assessed behavioural measures that could suggest under-
selection.
lying emotional states during training sessions using either
positive or negative reinforcement in horses. We also
3.3. Judgment bias testing tested if the training method (PR or NR) had a long-
term impact on horse emotional states, using a judgment
The results of the testing phase of the judgment bias bias experiment. As predicted, behavioural measures sug-
experiment revealed that mares reached the ambiguous gested that mares trained with PR experienced more
locations between the positive and negative locations with positive emotions (i.e. more motivated, more positive
intermediate latencies (Fig. 4). PR mares were generally attitude towards the trainer, head lower and ears more
slower than NR mares to reach locations; they went sig- forward), than mares trained with NR (i.e. more tensed,
nificantly slower than NR mares to the negative location more negative/indifferent attitude towards the trainer,
and the adjacent ambiguous location (Fig. 4). head higher and ears more backwards; McDonnell and
The interaction effect between group and location was Poulin, 2002; McDonnell and Haviland, 1995; Rietmann
significant (LMM: F4,40 = 4.19, p = 0.006), indicating dif- et al., 2004; Sankey et al., 2010a, 2010b; Visser et al.,
ferences between the groups depending on the locations. 2009). However, during the judgment bias test, NR mares
Post hoc comparisons revealed that PR mares reached were then more optimistic compared to PR mares, despite
the negative location and the ambiguous location next to their behaviour suggesting more negative emotions dur-
it (“ambiguous −”) slower than NR mares. There was no ing training. NR mares could have been more motivated
to obtain food rewards than PR mares, which had been
rewarded throughout the training exercises (Burman et al.,
2011). Alternatively, NR mares could have developed an
optimistic bias, suggesting more positive moods, triggered
by a release from stress experienced during the training
exercises (Briefer and McElligott, 2013; Doyle et al., 2010;
Sanger et al., 2011). This first attempt to measure judgment
biases in horses suggests that this is a promising method
to assess horse optimism.

4.1. Behavioural indicators of emotions during treatment

Behavioural indicators of emotions were remarkably

different between the two treatment groups. Body ten-
sion scores were lower, indicating more motivation in PR
mares, whereas it indicated more stress and tension/lack
of motivation in NR mares, for all exercises. Our meas-
ures of attitude towards the trainer showed that, for all
exercises, PR mares tended to establish contact with the
Fig. 4. Results of the judgment bias experiment. Latency (residuals of log-
transformed latency controlled for age) to reach the five locations (see trainer, whereas NR mares tended to avoid contact with/be
Fig. 1 for details) during the two days of test for the treatment group for- indifferent towards the trainer. PR mares had a lower head
merly trained with positive reinforcement (black) and the group trained level than NR mares for all exercises, except exercises 5
negative reinforcement (grey; mean ± SE per group per location; n = 6 and 6, which consisted of flexing the neck and lower-
horses per group). More positive residuals indicated longer latencies.
ing the head and resulted obviously in similar head levels
Differences between the two groups (negative and positive reinforce-
ment) for each location are indicated as follows: *p < 0.05; **p < 0.01; for both groups. Finally, the ears of PR mares were more
non-significant otherwise (linear mixed-effect models). forward than the ears of NR mares, for all exercises. All

Please cite this article in press as: Briefer Freymond, S., et al., Behaviour of horses in a judgment bias test associated with
positive or negative reinforcement. Appl. Anim. Behav. Sci. (2014), http://dx.doi.org/10.1016/j.applanim.2014.06.006
G Model
APPLAN-3916; No. of Pages 12 ARTICLE IN PRESS
10 S. Briefer Freymond et al. / Applied Animal Behaviour Science xxx (2014) xxx–xxx

these measures suggest that PR mares were experiencing 2011). NR mares, which were rewarded during the judg-
more positive emotions than NR mares during the treat- ment bias test, but not during the treatment phase (NP and
ment (Boissy et al., 2011; Reefmann et al., 2009; Sankey NR exercises), could have been more motivated to obtain
et al., 2010a, 2010b; Young et al., 2012), although some food than PR mare. By contrast, PR mares, which obtained
of our measures, such as the attitude towards the trainer, food only at the reference location during the judgment
could indicate motivation for food more than underlying bias test, instead of throughout the treatment phase (rein-
emotions. Similar differences have been found in horses forcer = clicker + food), could have been less motivated to
trained with PR or NR in other studies. For instance, Sankey check in the ambiguous and negative locations (“incen-
et al. (2010a) observed more forward ears, less head lift- tive contrast”; Flaherty, 1996; Burman et al., 2008b). This
ing, lower neck position and faster contact initiation with explanation does not necessarily imply underlying treat-
human in PR compared to NR horses trained to walk ment group difference in mood. Similarly, non-rewarded
backward. Hendriksen et al. (2011) found less discom- dogs were shown to be more optimistic than rewarded
fort behaviour (eye and nostril widening, tail whipping) dogs (Burman et al., 2011).
and less avoidance towards the training in PR than NR Alternatively, our treatments could have really induced
horses trained to enter a trailer. Innes and McBride (2008) different moods in PR and NR horses, but in the oppo-
showed that PR horses made more active contact with site direction to that expected. We predicted that, as a
the trainer than NR horses. Finally, Warren-Smith and result of accumulation of short-term positive emotional
McGreevy (2007) found that horses trained with both PR states during treatment, PR horses would be experienc-
and NR to halt, shook their heads vertically less than horses ing positive moods. Similarly, we predicted that NR horses
trained only with NR. All these results strongly suggest that would be in a negative mood, following an accumulation of
PR could be associated with more positive emotions, more short-term negative emotional states. Yet, our results show
positive contact with the trainer, and thus better horse wel- more optimism in NR than PR mares, following negative
fare at work than NR. PR could even be a more efficient emotions during treatment. If this optimism really indi-
form of training (i.e. higher success rate, faster) than NR cates underlying moods, then NR mares were in a more
(Hendriksen et al., 2011; Sankey et al., 2010a). positive mood than PR mares. Similar unexpected opti-
mism has been found in sheep following stressful shearing
4.2. Judgment bias test and restraint (Doyle et al., 2010; Sanger et al., 2011), and
in female goats rescued from poor welfare (Briefer and
We found significant differences between the two McElligott, 2013). This optimism has been suggested to
groups in their responses to the judgment bias exper- be triggered by a release from a stressful situation (nega-
iment. As mares from the two groups were all housed tive treatment; Doyle et al., 2010), and could be present in
together in a large housing system, and as National Stud females more than males (Briefer and McElligott, 2013). It
employees who were providing routine care of the ani- could also be compared to optimistic judgment bias follow-
mals were blind to the purpose of the study and to the ing a change in environment, as shown in pigs (Sus scrofa)
treatment groups, it is highly likely that cognitive bias and rats, which were transferred from an unenriched to an
differences observed between the two groups resulted enriched environment (Brydges et al., 2011; Douglas et al.,
from our treatment (PR versus NR), and not from other 2012). Animals that have been stressed might seek posi-
confounding factors (e.g. housing, routing care, feeding). tive events to balance their situation once a negative event
Despite displaying behavioural responses suggesting more ceases (Spruijt et al., 2001). Therefore, our results suggest
negative emotions during training, NR mares went signifi- that release from the negative emotions experienced dur-
cantly faster than PR mares to the negative location and the ing NR training induces optimism, at least in female horses.
ambiguous location next to it (ambiguous −). These results We propose that this effect could be quite common; when a
suggest that NR mares expected less negative events to short-term (Doyle et al., 2010; Sanger et al., 2011), or long-
occur and were therefore less pessimistic than PR mares. term stressor ends (Briefer and McElligott, 2013), females
According to Mendl et al. (2010a), NR mares could have might develop optimistic bias. After a particularly long-
been in a mood appropriate to an intermediate reward term stressor (i.e. lasting > 1 year), this optimistic bias could
opportunity environment, and were optimistic about the even last for years (Briefer and McElligott, 2013). Further
expectation of negative events (absence of food in our case), judgment bias experiments testing reactions to ambiguous
compared to PR mares. locations after the negative treatment ends could be com-
Our results cannot be explained by PR horses being fully pared to situations where the same negative treatment is
satiated, because the two groups were fed in the same way still applied, in order to verify this hypothesis.
in their home enclosure, and were tested on the second
day following the end of the treatment. Furthermore, we 5. Conclusion
did not find any difference between groups in their latency
to reach the positive location, suggesting no obvious differ- The judgment bias approach is a promising method
ence in motivation to participate in the test (Burman et al., for measuring optimism and pessimism in horses. The
2011). However, we found a difference between PR and NR behaviour of horses during training using NR suggests more
horses in their latency to reach the negative reference loca- negative emotions than when using PR. Yet, horses trained
tion, despite this difference not being present during the using NR develop more optimistic bias than horses trained
reminder day. It is possible that NR mares differed in their using PR, after the training stops. This optimistic bias could
feeding motivation compared to PR mares (Burman et al., be due to higher feeding motivation in NR compared to PR

Please cite this article in press as: Briefer Freymond, S., et al., Behaviour of horses in a judgment bias test associated with
positive or negative reinforcement. Appl. Anim. Behav. Sci. (2014), http://dx.doi.org/10.1016/j.applanim.2014.06.006
G Model
APPLAN-3916; No. of Pages 12 ARTICLE IN PRESS
S. Briefer Freymond et al. / Applied Animal Behaviour Science xxx (2014) xxx–xxx 11

mares, or to the cessation of a negative event. We therefore Doyle, R.E., Fisher, A.D., Hinch, G.N., Boissy, A., Lee, C., 2010. Release from
recommend training method using only PR, or combining restraint generates a positive judgement bias in sheep. Appl. Anim.
Behav. Sci. 122, 28–34.
NR and PR (Warren-Smith and McGreevy, 2007). Train- Doyle, R.E., Hinch, G.N., Fisher, A.D., Boissy, A., Henshall, J.M., Lee, C., 2011.
ing methods that trigger good mental health should be as Administration of serotonin inhibitor p-chlorophenylalanine induces
important for animal welfare as appropriate housing sys- pessimistic-like judgement bias in sheep. Psychoneuroendocrinology
36, 279–288.
tems. Flaherty, C.F., 1996. Incentive Relativity. Cambridge University Press,
Cambridge, UK.
Forkman, B., Boissy, A., Meunier-Salaün, M.-C., Canali, E., Jones, R.B., 2007.
Conflict of interest A critical review of fear tests used on cattle, pigs, sheep, poultry and
horses. Physiol. Behav. 92, 340–374.
The authors declare that there is no conflict of interest. Fureix, C., Jego, P., Henry, S., Lansade, L., Hausberger, M., 2012. Towards
an ethological animal model of depression? A study on horses. PLoS
ONE 7, e39280.
Acknowledgements Hall, C., Goodwin, D., Heleski, C., Randle, H., Waran, N., 2008. Is there evi-
dence of learned helplessness in horses? J. Appl. Anim. Welf. Sci. 11,
249–266.
We are grateful to H. Zurkinden, to the vet team of ISME Harding, E.J., Paul, E.S., Mendl, M., 2004. Animal behaviour: cognitive bias
at Agroscope Swiss National Stud Farm, and to all the staff and affective state. Nature 427, 312.
Hausberger, M., Gautier, E., Biquand, V., Lunel, C., Jégo, P., 2009. Could work
of Agroscope Swiss National Stud Farm for assistance. We be a source of behavioural disorders? A study in horses. PLoS ONE 4,
are also grateful to S. Hintze and A. McElligott for helpful e7625.
comments on the manuscript. E. Briefer is funded by a Swiss Heleski, C., Bauson, L., Bello, N., 2008. Evaluating the addition of posi-
tive reinforcement for learning a frightening task: a pilot study with
National Science Foundation (PZ00P3 148200) fellowship. horses. J. Appl. Anim. Welf. Sci. 11, 213–222.
Hendriksen, P., Elmgreen, K., Ladewig, J., 2011. Trailer-loading of horses:
is there a difference between positive and negative reinforcement
References concerning effectiveness and stress-related signs? J. Vet. Behav. 6,
261–266.
Bates, D., 2005. Fitting linear mixed models in R. R News 5, 27–30. Innes, L., McBride, S., 2008. Negative versus positive reinforcement: an
Bateson, M., Matheson, S.M., 2007. Performance on a categorisation task evaluation of training strategies for rehabilitated horses. Appl. Anim.
suggests that removal of environmental enrichment induces pes- Behav. Sci. 112, 357–368.
simism in captive European starlings (Sturnus vulgaris). Anim. Welf. Keltner, D., Lerner, J.S., 2010. Emotion. In: Handbook of Social Psychology.
16, 33–36. John Wiley & Sons, London, pp. 317–352.
Bateson, M., Desire, S., Gartside, S.E., Wright, G.A., 2011. Agitated honey- Kurland, A., 1999. Clicker Training for Your Horse. Sunshine Books,
bees exhibit pessimistic cognitive biases. Curr. Biol. 21, 1070–1073. Waltham, MA.
Boissy, A., Manteuffel, G., Jensen, M.B., Moe, R.O., Spruijt, B., Keeling, MacLeod, C., Byrne, A., 1996. Anxiety, depression, and the anticipation
L.J., Winckler, C., Forkman, B., Dimitrov, I., Langbein, J., Bakken, M., of future positive and negative experiences. J. Abnorm. Psychol. 105,
Veissier, I., Aubert, A., 2007. Assessment of positive emotions in ani- 286–289.
mals to improve their welfare. Physiol. Behav. 92, 375–397. Matheson, S.M., Asher, L., Bateson, M., 2008. Larger, enriched cages
Boissy, A., Aubert, A., Désiré, L., Greiveldinger, L., Delval, E., Veissier, I., are associated with optimistic response biases in captive Euro-
2011. Cognitive sciences to relate ear postures to emotions in sheep. pean starlings (Sturnus vulgaris). Appl. Anim. Behav. Sci. 109,
Anim. Welf. 20, 47–56. 374–383.
Briefer, E.F., 2012. Vocal expression of emotions in mammals: mechanisms McDonnell, S.M., Haviland, J.C.S., 1995. Agonistic ethogram of the equid
of production and evidence. J. Zool. 288, 1–20. bachelor band. Appl. Anim. Behav. Sci. 43, 147–188.
Briefer, E.F., McElligott, A.G., 2013. Rescued goats at a sanctuary display McDonnell, S., Poulin, A., 2002. Equid play ethogram. Appl. Anim. Behav.
positive mood after former neglect. Appl. Anim. Behav. Sci. 146, 45–55. Sci. 78, 263–290.
Briefer Freymond, S., Briefer, E.F., Niederhäusern, R.V., Bachmann, I., 2013. McGreevy, P., McLean, A., 2010. Equitation Science. Wiley-Blackwell Pub-
Pattern of social interactions after group integration: a possibility to lishing, United Kingdom.
keep stallions in group. PLoS ONE 8, e54688. Mendl, M., Paul, E.S., 2004. Consciousness, emotion and animal welfare:
Brilot, B., Asher, L., Bateson, M., 2010. Stereotyping starlings are more insights from cognitive science. Anim. Welf. 13, 17–25.
pessimistic. Anim. Cogn. 13, 721–731. Mendl, M., Burman, O.H.P., Parker, R.M.A., Paul, E.S., 2009. Cognitive bias
Brydges, N.M., Leach, M., Nicol, K., Wright, R., Bateson, M., 2011. Environ- as an indicator of animal emotion and welfare: emerging evidence
mental enrichment induces optimistic cognitive bias in rats. Anim. and underlying mechanisms. Appl. Anim. Behav. Sci. 118, 161–181.
Behav. 81, 169–175. Mendl, M., Burman, O.H.P., Paul, E.S., 2010a. An integrative and functional
Burman, O.H.P., Parker, R., Paul Elizabeth, S., Mendl, M., 2008a. A spatial framework for the study of animal emotion and mood. Proc. R. Soc.
judgement task to determine background emotional state in labora- Lond. B 277, 2895–2904.
tory rats, Rattus norvegicus. Anim. Behav. 76, 801–809. Mendl, M., Brooks, J., Basse, C., Burman, O., Paul, E., Blackwell, E., Casey,
Burman, O.H.P., Parker, R.M.A., Paul, E.S., Mendl, M., 2008b. Sensitivity to R., 2010b. Dogs showing separation-related behaviour exhibit a pes-
reward loss as an indicator of animal emotion and welfare. Biol. Lett. simistic cognitive bias. Curr. Biol. 20, R839–R840.
4, 330–333. Moe, R.O., Nordgreen, J., Janczak, A.M., Spruijt, B.M., Zanella, A.J., Bakken,
Burman, O.H.P., Parker, R.M.A., Paul, E.S., Mendl, M.T., 2009. Anxiety- M., 2009. Trace classical conditioning as an approach to the study
induced cognitive bias in non-human animals. Physiol. Behav. 98, of reward-related behaviour in laying hens: a methodological study.
345–350. Appl. Anim. Behav. Sci. 121, 171–178.
Burman, O., McGowan, R., Mendl, M., Norling, Y., Paul, E., Rehn, T., Keeling, Nakagawa, S., 2004. A farewell to Bonferroni: the problems of low statis-
L., 2011. Using judgement bias to measure positive affective state in tical power and publication bias. Behav. Ecol. 15, 1044–1045.
dogs. Appl. Anim. Behav. Sci. 132, 160–168. Nettle, D., Bateson, M., 2012. The evolutionary origins of mood and its
Dawkins, M.S., 2008. The science of animal suffering. Ethology 114, disorders. Curr. Biol. 22, R712–R721.
937–945. Paul, E., Harding, E.J., Mendl, M., 2005. Measuring emotional processes in
De Boyer Des Roches, A., Richard-Yris, M.-A., Henry, S., Ezzaouïa, M., animals: the utility of a cognitive approach. Neurosci. Biobehav. Rev.
Hausberger, M., 2008. Laterality and emotions: visual laterality in the 29, 469–491.
domestic horse (Equus caballus) differs with objects’ emotional value. R Development Core Team, 2012. R Foundation for Statistical Computing,
Physiol. Behav. 94, 487–490. Vienna, Austria.
Désiré, L., Boissy, A., Veissier, I., 2002. Emotions in farm animals: a new Reefmann, N., Wechsler, B., Gygax, L., 2009. Behavioural and physiological
approach to animal welfare in applied ethology. Behav. Processes 60, assessment of positive and negative emotion in sheep. Anim. Behav.
165–180. 78, 651–659.
Douglas, C., Bateson, M., Walsh, C., Bédué, A., Edwards, S.A., 2012. Environ- Rietmann, T.R., Stuart, A.E.A., Bernasconi, P., Stauffacher, M., Auer, J.A.,
mental enrichment induces optimistic cognitive biases in pigs. Appl. Weishaupt, M.A., 2004. Assessment of mental stress in warm-
Anim. Behav. Sci. 139, 65–73. blood horses: heart rate variability in comparison to heart rate

Please cite this article in press as: Briefer Freymond, S., et al., Behaviour of horses in a judgment bias test associated with
positive or negative reinforcement. Appl. Anim. Behav. Sci. (2014), http://dx.doi.org/10.1016/j.applanim.2014.06.006
G Model
APPLAN-3916; No. of Pages 12 ARTICLE IN PRESS
12 S. Briefer Freymond et al. / Applied Animal Behaviour Science xxx (2014) xxx–xxx

and selected behavioural parameters. Appl. Anim. Behav. Sci. 88, behaviour as an indicator for the state of reward systems. Appl. Anim.
121–136. Behav. Sci. 72, 145–171.
Sanger, M.E., Doyle, R.E., Hinch, G.N., Lee, C., 2011. Sheep exhibit a positive Strunk, D.R., Lopez, H., DeRubeis, R.J., 2006. Depressive symptoms are
judgement bias and stress-induced hyperthermia following shearing. associated with unrealistic negative predictions of future life events.
Appl. Anim. Behav. Sci. 131, 94–103. Behav. Res. Ther. 44, 861–882.
Sankey, C., Richard-Yris, M.-A., Henry, S., Fureix, C., Nassur, F., Visser, E.K., van Dierendonck, M., Ellis, A.D., Rijksen, C., van Reenen,
Hausberger, M., 2010a. Reinforcement as a mediator of the per- C.G., 2009. A comparison of sympathetic and conventional train-
ception of humans by horses (Equus caballus). Anim. Cogn. 13, ing methods on responses to initial horse training. Vet. J. 181,
753–764. 48–52.
Sankey, C., Henry, S., Górecka-Bruzda, A., Richard-Yris, M.A., Hausberger, Waran, N., McGreevy, P.D., Casey, R.A., 2002. Training methods and horse
M., 2010b. The way to man’s heart is through his stomach: what about welfare. In: The Welfare of Horses. Kluwer Academic Publishers, The
horses. PLoS ONE 5, e15446. Netherlands, pp. 151–180.
Scherer, K.R., 1984. On the nature and function of emotion: a component Warren-Smith, A.K., McGreevy, P.D., 2007. The use of blended positive and
process approach. In: Approaches to Emotion. Erlbaum, Hillsdale, NJ, negative reinforcement in shaping the halt response of horses (Equus
pp. 293–317. caballus). Anim. Welf. 16, 481–488.
Sharot, T., 2011. The optimism bias. Curr. Biol. 21, R941–R945. Wright, W.F., Bower, G.H., 1992. Mood effects on subjective probability
Skinner, B.F., 1938. The Behavior of Organisms: An Experimental Analysis. assessment. Organ. Behav. Hum. Decis. Process. 52, 276–291.
Appleton-Century-Crofts, New York. Young, T., Creighton, E., Smith, T., Hosie, C., 2012. A novel scale of
Spruijt, B.M., van den Bos, R., Pijlman, F.T.A., 2001. A concept of welfare behavioural indicators of stress for use with domestic horses. Appl.
based on reward evaluating mechanisms in the brain: anticipatory Anim. Behav. Sci. 140, 33–43.

Please cite this article in press as: Briefer Freymond, S., et al., Behaviour of horses in a judgment bias test associated with
positive or negative reinforcement. Appl. Anim. Behav. Sci. (2014), http://dx.doi.org/10.1016/j.applanim.2014.06.006