Content uploaded by Anna-Yezica Norling
Author content
All content in this area was uploaded by Anna-Yezica Norling on May 25, 2016
Content may be subject to copyright.
ORIGINAL PAPER
Positive affect and learning: exploring
the ‘‘Eureka Effect’’ in dogs
Ragen T. S. McGowan
•
Therese Rehn
•
Yezica Norling
•
Linda J. Keeling
Received: 31 May 2013 / Revised: 24 September 2013 / Accepted: 25 September 2013 / Published online: 6 October 2013
Ó Springer-Verlag Berlin Heidelberg 2013
Abstract Animals may experience positive affective
states in response to their own achievements. We investi-
gated emotional responses to problem-solving in dogs,
separating these from reactions to rewards per se using a
yoked control design. We also questioned whether the
intensity of reaction would vary with reward type. We
examined the response (behavior and heart rate) of dogs as
they learned to gain access to different rewards: (1) food
(2) human contact, and (3) dog contact. Twelve beagles
were assigned to matched pairs, and each dog served as
both an experimental and a control animal during different
stages of the experiment. We trained all dogs to perform
distinct operant tasks and exposed them to additional
devices to which they were not trained. Later, dogs were
tested in a new context. When acting as an experimental
dog, access to the reward was granted immediately upon
completion of trained operant tasks. When acting as a
control, access to the reward was independent of the dog’s
actions and was instead granted after a delay equal to their
matched partner’s latency to complete their task. Thus,
differences between the two situations could be attributed
to experimental dogs having the opportunity to learn to
control access to the reward. Experimental dogs showed
signs of excitement (e.g., increased tail wagging and
activity) in response to their achievements, whereas con-
trols showed signs of frustration (e.g., chewing of the
operant device) in response to the unpredictability of the
situation. The intensity of emotional response in experi-
mental dogs was influenced by the reward type, i.e.,
greatest response to food and least to another dog. Our
results suggest that dogs react emotionally to problem-
solving opportunities and that tail wagging may be a useful
indicator of positive affective states in dogs.
Keywords Canis familiaris Canine
Positive emotions Problem-solving Behavior
Animal welfare
Introduction
The investigation of emotional states in animals is fast
becoming a key area in animal welfare research (e.g.,
Wemelsfelder 2007; Green and Mellor 2011). Although
experimental studies in this area are still limited, the exact
nature of the emotional experiences of animals remains
poorly understood (Harding et al. 2004; Paul et al. 2005;
Boissy et al. 2007). Affective states are complex, involving
an interplay between behavioral, physiological, and sub-
jective components (Panksepp 1998; Paul et al. 2005;
Panksepp 2011); although the way in which animals
experience emotion is difficult to measure directly,
behavioral and physiological components of affective
responses can serve as useful proxy measures (Burman
et al. 2008; Gygax et al. 2013). From an evolutionary
perspective, there should be positive states associated with
actions where the benefits are in the future (e.g., play).
Problem-solving opportunities should evoke an immediate
positive emotional state in animals, as a means to motivate
the animal to explore and solve problems, even if the true
benefit of the behavior is in the long term. Such a notion is
supported by recent evidence that in order to store
R. T. S. McGowan T. Rehn Y. Norling L. J. Keeling
Department of Animal Environment and Health, Swedish
University of Agricultural Sciences, Box 7068, 750 07 Uppsala,
Sweden
R. T. S. McGowan (& )
Nestle
´
Purina Research, St. Louis, MO, USA
e-mail: Ragen.Trudelle-SchwarzMcGowan@rd.nestle.com
123
Anim Cogn (2014) 17:577–587
DOI 10.1007/s10071-013-0688-x
experiences permanently, the brain must release dopamine
(e.g., Bethus et al. 2010; Chowdhury et al. 2012).
The idea that providing animals with opportunities for
learning and problem-solving could elicit positive emo-
tions has been the topic of much discussion (Meehan and
Mench 2007; Boissy et al. 2007; McGowan et al. 2010);
however, experimental work focusing on the intrinsic
reinforcing properties of problem-solving and the emo-
tional consequences of responding to challenge is rare
(Langbein et al. 2009; Zebunke et al. 2011). One paper
which has specifically addressed this idea is that of Hagen
and Broom (2004), who designed a study with cattle using
a yoked control design to determine whether cattle respond
emotionally to their own achievements in an operant
learning task. They found that cattle who had mastered an
operant task in order to receive a food reward demonstrated
more behavior indicative of excitement than control ani-
mals that received the same reward without the opportunity
to learn the relationship between the task and the reward. A
finding of this nature deserves replication and investigation
of its generality, as well as further study due to its potential
in the assessment of animal emotion. There has been a call
for additional studies to examine the idea that problem-
solving opportunities are intrinsically rewarding, even if
the experiences are also externally rewarded (e.g., Meehan
and Mench 2007).
Measuring emotion in animals is challenging, and
although most work in the past has focused on negative
affective states, there is an emerging view that good wel-
fare requires the presence of positive affective states such
as pleasure (Fraser and Duncan 1998; Boissy et al. 2007).
There is, however, a lack of research concerning the cir-
cumstances under which such positive emotional states
might occur and also on how these positive states can be
measured. In particular, since most work on emotional
states has involved research on negative affect, finding
biologically based paradigms to put animals into positive
states is key to starting the investigation into positive
emotional states.
In this paper, we took on this challenge in an investi-
gation into the emotional responses to learning in dogs,
Canis familiaris. Utilizing a modified version of the yoked
control design outlined by Hagen and Broom (2004), we
aimed to identify the reactions of dogs to their own
learning and to separate these from reactions to the rewards
themselves. We hypothesized that dogs would react emo-
tionally to their own achievements, predicting that the
development of an understanding of the relationship
between an action and a reward would elicit a positive
affective state in dogs. We also took this idea one step
further hypothesizing that the intensity of such emotional
reactions to learning might vary in response to different
reward types. For example, we questioned whether the
excitement of learning to gain access to food is experienced
the same as the excitement for learning to gain access to a
social companion (as suggested in humans, Sakaki et al.
2012). We therefore investigated the emotional response to
learning in dogs in relation to rewards catering to different
motivational systems.
Individual dogs with different experiences and personali-
ties may be expected to solve problems more or less quickly or
experience learning in different ways (Carere and Locurto
2011). Accordingly, we were interested in teasing out the
individual differences that dogs demonstrated during prob-
lem-solving events and relating this back to their past per-
formances with cognitive tasks. In doing so, we felt that it was
important to look at the heart rate in addition to the behavior as
we know that the heart rate can be a measure of arousal or
excitement, providing insight into the internal state of the
animal. In particular, we were interested in trying to identify
the moment at which dogs might experience the ‘‘Eureka
Effect’’ during a problem-solving event by looking closely at
the behavior and the heart rate before and after the dogs learn
the causal relationship between their actions and gaining
access to a reward. Their behavioral and physiological
responses to this learning may be an indicator of what they
experience at this presumably positive moment in time.
Methods
Animals and housing
We conducted trials to investigate emotional responses to
problem-solving with twelve female beagles (age
10–14 months) housed at the Swedish University of Agri-
cultural Sciences in Uppsala. During the day, dogs at this
facility were kept in groups of six in large (145–200 m
2
)
outdoor runs constructed from chain-link fencing. This con-
struction allowed for visual, auditory, and olfactory commu-
nication between groups. Each outdoor run was furnished
with multiple shelters, climbing structures, toys, and water
dispensers. Dogs were moved indoors (in groups of three) at
night into rooms (24.3 m
2
) equipped with blankets, water
dispensers, and toys. Indoor rooms were cleaned each morn-
ing when the dogs were moved to the outdoor runs. Dogs were
fed a maintenance chow diet twice daily (individually in
separate feeding areas) at 08.30 and 15.30 h. Dogs interacted
with human caregivers during routine care, play sessions and
daily walks, and had been used in several behavioral obser-
vation studies exploring the positive affective states.
Training
Training sessions were conducted to familiarize the dogs
with the operant devices and the skills required for
578 Anim Cogn (2014) 17:577–587
123
manipulating them. Since each dog would serve as both an
experimental and a control animal at different stages of the
experiment, we trained all dogs to perform three distinct
operant tasks (experimental treatment) and exposed all
dogs to an additional three operant devices without training
(control treatment). Training sessions were carried out in a
single indoor room to which dogs were habituated prior to
the initiation of the training sessions. In this room, two
training arenas were constructed (1 m 9 2 m), each
equipped with two different operant devices (Fig. 1). The
dogs were assigned to matched pairs. Both dogs of a
matched pair were present in the training room at the same
time, one in each training arena. Each dog was accompa-
nied by a handler and they could hear what was happening
in the other arena, but they could not see into it.
The dogs were trained to perform three operant tasks
(learning one at a time) by means of shaping with food
rewards and acoustic signaling. Different acoustic signals
(click or beep) were used for each dog of the pair. Shaping
was conducted in three stages where dogs were rewarded
for (1) approaching the operant device, (2) making contact
with the operant device, and then (3) correctly manipulat-
ing the operant device. In each phase, appropriate action by
the dog was immediately rewarded with a click/beep and a
treat. By having two operant devices present in each
training arena and training each dog of the pair on a dif-
ferent device of the pair, and with a different acoustic
signal (Table 1) to confirm that the dog was taking
appropriate action, each dog in the pair could be trained as
an experimental and control animal simultaneously. That is
to say, while dogs were learning to manipulate one device,
they were not receiving any training on their matched
experimental partner’s operant task even though that same
device was in their own training arena. Any manipulation
of the ‘‘control’’ device was ignored by the handler. This
experimental design meant that while learning to be
controls, dogs were exposed to identical circumstances as
their matched experimental dogs (i.e., they heard the same
number of these two acoustic cues, were exposed to the
same two identical operant devices, and received the same
length of ‘‘training’’ session). Once a dog approached and
successfully manipulated the device three times on its own
accord, training ceased. Due to the differences in learning
ability between dogs, the total length of training sessions
for each matched pair varied, being determined by the
slowest learning dog of each pair. The pair of dogs then
proceeded to learn the second and then the third operant
task, in a similar way.
Dogs were removed from their groups just prior to each
training session and returned immediately afterward. The
handlers, operant devices, location of the training arena,
and location of the devices within the arena were balanced
between dogs.
Testing
One week after completing their training sessions, the dogs
were tested in a new context where the performance of the
learned task was necessary to gain access to one of the
three rewards. This one-week time gap along with a change
in location between training and testing created a new
situation in which the dogs were required to put together
the pieces of their training to solve a novel problem.
Fig. 1 Diagram of the training arenas. Two dogs of a matched pair
were trained simultaneously: dog I trained as the experimental dog on
device A and exposed to device D as a control; dog II trained as the
experimental dog on device D and exposed to device A as a control
Table 1 List of the operant tasks and corresponding signal signifying
the task was completed successfully
Task Signal Description of device
Press wood
lever
Click Wooden lever attached to a wheel with
spokes that clicked when the lever
was pressed
Press paddle
lever
Bell A canoe paddle attached to a bicycle
bell that rang when the paddle was
pressed
Push box off
stack
Plastic
hitting
floor
A stack of four plastic boxes where the
lower three boxes were attached to the
floor and the top box could be pushed
off the stack
Pushover dog
obelisk
Plastic
hitting
floor
A tall plastic container that was
weighted at the top and hinged so that,
with some force, it could be tipped
over
Push ball of
stand
Ball
hitting
floor
A child’s rubber ball placed on top of a
heavy cardboard tube affixed to the
floor so that, with some force, the ball
could be pushed off
Press key on
dog piano
Metal ding A child’s piano with plastic keys that
when manipulated struck a metal
xylophone making a tone
Each dog trained to complete three tasks (one per reward type)
Anim Cogn (2014) 17:577–587 579
123
Testing took place in a room to which all dogs were
habituated prior to the initiation of the first test session. The
test room was divided into four parts such as (1) start arena,
(2) hide for experimenter, (3) runway, and (4) reward area
(Fig. 2). The start arena had an entrance door, through
which the dogs were led into the test room, and a door into
the runway, through which they could access the reward.
Dogs were accompanied to the test room by a handler who
remained in the start arena with the dog for the duration of
each test run. The handler wore headphones, and after
releasing the dog into the start arena, they stood in the
corner with their back turned toward the dog so as to
prevent providing any cues that might aid the dog in
solving the task at hand (Kupa
´
n et al. 2011; Lit et al. 2011;
Buttelmann and Tomasello 2012). This procedure was
necessary because if the handler exited the start arena, most
dogs focused on the door the handler left from, instead of
the operant device. An experimenter was present during the
entire test session, but was visually isolated from the dog in
a hide located outside the start arena. From the hide, the
experimenter could manipulate the door to the runway via a
pulley system. Cameras were mounted on both the start
arena and the runway, which allowed the experimenter to
view the dogs’ actions via a monitor located in the hide.
As with the training sessions, dogs were removed from
their groups just prior to each test session and were
returned immediately afterward. Each test session con-
sisted of eight consecutive runs conducted on Day 1.
During the first two runs, the dog was led into an empty
start arena, and upon release of the dog by the handler, the
door to the runway was immediately opened by the
experimenter. This demonstrated to the dog that there was
a way out of the start arena and that there was a reward at
the end of the runway. For these initial two runs, the dogs
were allowed 30 s to interact with the reward once they
reached the end of the runway.
Before the beginning of the third run, the experimenter
placed one of the operant devices into the start arena. This
device was familiar to both dogs in the matched pair;
however, only the dog acting as the experimental dog was
trained to manipulate this device. Although the experi-
mental dogs were trained on how to manipulate the device
in general, here the device was presented in a new context
where appropriate manipulation led to the door to the
runway being opened. This provided the opportunity for
experimental dogs to learn the causal relationship between
performing the task and the door opening. If successful, the
door to the runway was immediately opened providing
access to the runway and the dogs were allowed 30 s to
interact with the reward once they reached the end of the
runway. The door to the start arena was closed after the dog
exited into the runway, preventing the reentrance into the
start arena. Experimental dogs were allowed 5 min per test
run to explore the start arena and perform the operant task.
If a dog did not complete the task during the 5-min
exploration period in the start arena, she was collected by
the handler and led out of the room for a 5-min rest period
before being brought in for the next test run. If a dog failed
to perform the task during 5 min on two consecutive runs,
she was retested the following day. After each successful
run, the dog was collected by the handler, led out of the
room through an exit at the end of the runway, and then
was returned to the start arena. This cycle was repeated 5
times for a total of 6 test runs.
The latency for each experimental dog to successfully
perform the operant task was recorded. Control dogs were
then placed in the start arena for a time period matched to
the time it took the experimental dog in the pair to com-
plete the operant task. For this reason, experimental dogs
were tested first followed by their matched control. When
the dogs were acting as controls, the experimenter opened
the door to the runway at the same time point as when their
matched experimental dog had successfully completed the
operant task regardless of what action the dog acting as the
Fig. 2 Diagram of the test arena
Fig. 3 Yoked control design. Dogs acting as experimental animals
had the opportunity to learn a causal relationship between performing
a specific task and gaining access to a specific reward. Dogs acting as
controls did not have the opportunity to learn this relationship but did
gain access to the same reward
580 Anim Cogn (2014) 17:577–587
123
control made. When acting as controls, the dogs were,
therefore, granted access to the same reward independent
of their actions (Fig. 3).
This yoked control design ensured that when dogs were
acting as experimental, but not control animals, they had
the opportunity to learn a causal relationship between
performing a specific task and gaining access to a specific
reward. When acting as controls, the dogs did not have the
opportunity to learn this relationship but did gain access to
the same reward. Differences between the dogs’ responses
to both the experimental and control sessions could then be
used to differentiate emotional reactions to learning from
emotional reactions to the reward, as well as changes in
how this was experienced (changes in heart rate and
behavior) over the six runs. For dogs in the experimental
situation, access to the reward should have become pre-
dictable as they learned the causal relationship between
their actions and the door opening, while for the control
dogs the situation would remain unpredictable.
Training and testing sessions were repeated with each
matched pair (Fig. 4) to compare emotional reactions to
rewards catering to different motivational systems: (1) food
(40 small bits of highlypalatable kibblespread at the end of the
runway); (2) social contact with a human (a familiar human
sitting at the end of the runway, calmly petting the dog once
the dog made contact); and (3) social contact with other dogs
(two group members of the dog on leashes at the end of the
runway, allowing the opportunity for social interaction and
play). Dogs were trained to complete three different operant
tasks in order to gain access to the three different rewards. The
operant tasks, rewards, and their order were balanced between
dogs and across treatments to control for differences in the
difficulty level of tasks and any carryover effects of one
reward to another. Each matched pair was subject to three
training days and six testing days (three acting as an experi-
mental and three acting as a control dog) with a one-week time
gap between each training and testing session.
Measures
All test sessions were video recorded for the later analysis.
Each dog’s demeanor when entering the start arena was
noted, namely whether the dog led the handler into the start
arena or whether the handler had to coax the dog into the
start arena. Latency to perform the operant task, latency to
exit the start arena, and latency to reach the reward after the
runway door was opened were measured. In addition, each
dog’s behavior in the start arena was evaluated to assess the
frequency of steps (activity level) and tail wags (excite-
ment level) that occurred during the problem-solving event
when acting as experimental dogs and the corresponding
waiting period when acting as controls. Steps from all four
paws were counted when the foot was lifted completely off
the floor and placed down again. A tail wag was counted
each time the tail moved from a central position (in line
with the spine) to either the right or left and back again.
Wags were counted in this manner regardless of tail car-
riage position (e.g., high or low); however, high tail car-
riage was most common with low tail carriage being
observed only rarely in the context of this study. The
manner in which the dogs manipulated the operant devices
was also noted. Specifically, whether the dogs manipulated
the device appropriately (pawing or nosing) or inappro-
priately (chewing on the device). All videos were scored by
a single observer in a random order and a test of intra-
observer concordance (on six videos chosen at random
between experimental and control sessions across matched
pairs for the three reward types) indicated that the reli-
ability of data scoring was C95.6 %.
Mean heart rate (recorded every 5 s) for each dog
throughout the entire test session was also measured using
Polar S810 (Polar Electro Oy, Kempele, Finland) heart rate
monitors along with Polar Precision Performance Soft-
ware
TM
. The monitors were strapped around the chest of
the dog before the start of each test session and recording
started immediately upon each dog’s entry into the start
area for the first of the eight runs. All dogs were habituated
to the heart rate monitors prior to the beginning of the test
sessions.
Statistical analysis
We used a mixed linear model (Proc Mixed) of the SAS
Ò
Institute (version 9.1) with run as a repeated measure; dog
as a random effect; and pair, task, device, treatment, and
reward as fixed effects in the model. We tested for overall
effects of treatment, task, reward, and their interactions on
the latencies and mean heart rate throughout the test ses-
sion as well as the number of steps, tail wags, and mean
heart rate when the dog was in the start arena. Analyses
conducted on both raw and ranked data yielded similar
Fig. 4 Dogs were assigned to matched pairs and served as both
experimental and control animals at different stages of the exper-
iment. We trained all dogs to perform distinct operant tasks
(experimental) and exposed them to additional operant devices to
which they were not trained (control)
Anim Cogn (2014) 17:577–587 581
123
results, suggesting that valid assessment could be con-
ducted with parametric statistics. Results from the analyses
on ranked data are reported. Separate analyses were con-
ducted for the latencies, heart rate, steps, and tail wags.
Comparisons between variables were made based on the
differences in least-squares means, with p values adjusted
for multiple comparisons using the Tukey’s option.
In addition, we took a retroactive approach looking at
the data collected as part of another experiment where
these same dogs were trained on a cognitive bias task
(Burman et al. 2011) paired with the heart rate data col-
lected in this study to look at how the physiological
responses of the dogs in this study compared with their
cognitive abilities during a previous study. These ad hoc
comparisons were made using paired t tests in Minitab
Ò
(version 16).
Results
From the onset of the test sessions, it was noted by the
animal handlers that when acting as experimental animals,
the dogs were eager to enter the test room. They were
pulling the handler toward the door leading into the test
room and in most cases would enter the start arena ahead of
the handler. When acting as controls, on the other hand, the
dogs were initially eager to enter the test room during their
first two or three test runs, but soon grew reluctant to enter
the test room. By the end of the test sessions, these dogs
would enter the start arena only after some coaxing from
the handler. These observations are in line with the idea
that apathy will develop when the animal has no way to
control their environment, whereas hyperreactivity may
occur when an animal is under the impression that it is able
to control such events (Boissy et al. 2007).
Latencies
All dogs acting as experimental animals were successful in
manipulating the operant device within the time frame
provided in the start arena (mean ± SE; 50.53 ± 7.48 s)
on all test days. As expected, there were differences
between individuals with some dogs learning to manipulate
the device more quickly than others (Fig. 5). Nine out of
twelve of the dogs manipulated the operant device cor-
rectly on at least one occasion when acting as a control
dog, and this occurred most frequently during the last test
session after the control dogs had served as experimental
dogs on three occasions. However, as explained, this did
not have the effect of opening the door, but shows that the
dogs were transferring their experience of being an
experimental dog and that they had learnt the association
between manipulating a familiar operant device and the
opening of the door to the novel operant device.
There was a treatment effect on the time it took dogs to
exit the start arena after the door to the runway was opened,
with dogs acting as experimental animals taking longer
(mean ± SE; 5.81 ± 1.40 s) than when they were acting
as controls (4.17 ± 0.47 s) (t
22
= 2.76, p = 0.03). The
latencies to exit the start arena were also influenced by
reward type (Fig. 6), with experimental dogs taking longer
than when they were acting as control dogs to leave the
start arena when there was a food reward (t
22
= 3.71,
p = 0.014) or a social opportunity with other dogs
(t
22
= 2.41, p = 0.041). However, there were no differ-
ences between acting as an experimental and a control dog
when the reward was a social opportunity with a human
(t
22
=-0.24, p = 0.817).
The dogs remained consistent in their approach speed
for the reward at the end of the runway over the three tasks
suggesting that the event was rewarding, with no apparent
decline in interest over the three tasks. There was no
treatment effect on the time it took dogs to reach the
reward at the end of the runway once they had exited the
start arena irrespective of the reward type (t
22
B 0.82,
p [ 0.05).
Fig. 5 Means ± SE latencies to complete the operant task across all
reward types for dogs acting as experimental animals
Fig. 6 Means ± SE latencies to exit the start area expressed by
reward type
582 Anim Cogn (2014) 17:577–587
123
Activity level
There was a treatment effect on the activity level in the
start arena, with dogs acting as experimental animals
(mean ± SE; 2.63 ± 0.08 steps/s) being more active
(taking more steps) than when acting as controls
(2.29 ± 0.16 steps/s) (F
1,11
= 37.36, p \ 0.0001). Overall,
irrespective of whether they were acting as experimental
dogs or controls, dogs were less active during the third
round of testing (task 3) than they were during the first
(t
22
= 4.29, p \ 0.001) or second (t
22
= 5.26, p \ 0.0001)
rounds of testing.
The expected reward influenced the activity levels of the
dogs in both treatments (Fig. 7). When dogs were expect-
ing a food reward, they were more active than when
expecting access to other dogs (t
22
= 3.41, p = 0.0067).
Comparing activity in dogs expecting access to a human
and those expecting access to other dogs, t
22
= 2.48,
p = 0.053.
Chewing
When acting as control animals, the dogs were observed to
bite and chew the operant devices in a manner not appro-
priate for manipulating the device. No dog acting as an
experimental animal chewed a device in this manner.
Chewing of the devices by control dogs occurred
throughout the experiment, but most frequently during the
third round of testing (Fig. 8). This behavior was not
influenced by the reward the dogs were expecting at the
end of the runway.
Tail wags
There was a treatment effect on tail wagging in the start
arena with the dogs acting as experimental animals
(mean ± SE; 2.70 ± 0.01 wags/s) wagging their tails more
than when acting as controls (1.21 ± 0.08 wags/s)
(F
1,11
= 205, p \ 0.0001). Tail-wagging frequency did not
change over time (i.e., no difference across tasks,
F
2,22
= 2.04, p = 0.15).
The expected reward influenced tail-wagging frequency
for dogs acting as experimental animals (Fig. 9) with dogs
expecting a food reward wagging their tails more than
when expecting access to other dogs (t
22
= 3.02,
p = 0.006) and a tendency for dogs expecting access to a
human to wag their tails more than when expecting access
to other dogs (t
22
= 1.81, p = 0.083).
Heart rate
There was no overall difference in mean heart rate between
dogs when dogs were acting as experimental animals
(mean ± SE; 126.1 ± 3.4 bpm) and acting as controls
(127.5 ± 2.2 bpm) when considering the entire test ses-
sion, nor there was a difference in mean heart rate for these
dogs in the start arena during the problem-solving oppor-
tunity for the dogs acting as experimental animals
(132.1 ± 2.9 bpm) or during the corresponding waiting
period for the dogs acting as controls (131.0 ± 2.0 bpm).
Burman et al. (2011) found that it took the dogs
93.2 ± 15.2 trials to reach criterion during initial training
on a cognitive task. In this study, dogs that had higher heart
rates in the start arena when acting as experimental animals
than when acting as controls averaged 137.8 ± 19.2 trials
Fig. 7 Means ± SE number of steps/second in the start arena during
the problem-solving opportunity for experimental dogs or the
corresponding waiting period for control dogs, expressed by reward
type
Fig. 8 Means ± SE frequency of chewing the operant device for
control dogs in the start arena over time
Fig. 9 Means ± SE number of wags/second in the start area during
the problem-solving opportunity for experimental dogs or the
corresponding waiting period for control dogs, expressed by reward
type
Anim Cogn (2014) 17:577–587 583
123
to reach criterion in the Burman et al. (2011) study. Dogs
that had higher heart rates in the start arena when acting as
controls than when acting as experimental animals aver-
aged 61.3 ± 11.9 trials to reach criterion in the Burman
et al. (2011) study (t
7
= 2.36, p = 0.011).
Discussion
As predicted, we found differences in behavior between the
dogs when they were acting as experimental and control
animals in our study. In general, the dogs showed more
signs of positive excitement (e.g., more activity and more
tail wagging) when acting as experimental animals than
when acting as controls. Hagen and Broom (2004) sug-
gested that differences in emotional response to learning
might occur either specifically during the process of
understanding, or only occur once a task has been acquired.
They found evidence for the latter as behavioral differences
in cattle were detected only in the runway after the animals
had learned that they could control access to the reward.
Because they found these behavioral differences after the
cattle had learned the causal relationship between their
actions and gaining access to the reward, Hagen and Broom
(2004) speculated that the cattle were reacting to their own
learning success and thus, in a sense, to their own
achievement. In our study, we found behavioral differences
during the actual process of understanding (e.g., in the start
arena), providing evidence that it was success in the
problem-solving that elicited a positive affective state in
the experimental animals.
Differences in demeanor were observed in that dogs
were quick to enter the test arena initially, but control dogs
became increasingly reluctant. Behavioral differences were
also reflected in the fact that dogs acting as controls, but
not when acting as experimental animals, were observed to
chew the operant device on several occasions. Chewing of
the devices by control dogs occurred throughout the
experiment, but most frequently toward the end of the
study during the last matched pair of testing after they had
already served as experimental animals on three occasions.
The dogs may have developed a set of expectations about
the test environment, with the understanding that they
could open the door to the runway through the operation of
the operant device. When faced with being in the control
situation where they no longer had the ability to open the
door to the runway by manipulating the operant device,
they may have suffered from a mismatch between their
expectations and the situation at hand. This mismatch most
likely would lead to frustration (Cuenya et al. 2012).
Because manipulation of the operant device did not lead to
the door opening, their frustration escalated into biting, and
chewing at the operant device.
Although some degree of negative stress and frustration
will probably occur in the initial stage of learning (Lang-
bein et al. 2004; Meehan and Mench 2007), successful
actions with a positive outcome are, as explained in the
introduction, likely to induce positive feelings (Kalbe and
Puppe 2010; Zebunke et al. 2011) and thus motivate the
animal to continue the behavior. Hagen and Broom (2004)
noted more agitation in their experimental group when the
cattle were starting to learn the causal relationship. We did
not observe any clear signs of agitation when our dogs
were acting as experimental animals. However, this may
have been due to the fact that our dogs were pre-trained on
manipulating the operant devices that they were later
exposed to during the problem-solving opportunity. Hagen
and Broom (2004) shaped their experimental subjects to
use the operant device during the test situation. This
entailed providing food in the start arena near the operant
device. The cattle, therefore, may have learned to associate
both the start arena and the runway with a food reward. To
avoid this, we pre-trained our dogs on how to manipulate
the operant devices during sessions in a special training
room, then after a delay of 1 week tested them in a new
context where correct manipulation of the device led to the
door to the runway being open and access to the reward.
Through this process, the dogs never associated the start
arena, but only the runway, with a reward.
Experimental dogs were more active in the start arena
than control dogs despite the fact that when acting in both
roles, the dogs had equal knowledge of the reward at the
end of the runway and thus should have demonstrated
similar levels of anticipatory excitement in the start arena
(Spruijt et al. 2001). The dogs were clearly more excited
when acting as experimental animals than as controls, not
generally (e.g., in the runway), but in relation to the only
difference between the treatments (e.g., in the start arena).
That is to say, the excitement was when the dogs were
experiencing a learning process that they did not experi-
ence when they were acting as controls. Upon entrance into
the start arena and recognition of the operant device on
which they had been trained, the experimental dogs
immediately went to work on attempting to manipulate the
device. Control dogs were often standing directly in front
of the runway door sniffing or pawing at the door as they
waited for it to open. This meant that they were quick to
exit as soon as the door opened. Experimental dogs, on the
other hand, were standing near to the operant device when
the door opened and had to run over to the door before
exiting. This behavior may have contributed to the treat-
ment differences we detected in the time it took the dogs to
exit the start arena; however, it would not explain the
differences according to the reward type.
Tail wagging is an indicator of arousal (Beerda et al.
1999; Prescott et al. 2004), and confident dogs express
584 Anim Cogn (2014) 17:577–587
123
more tail wagging than unsure dogs (Goddard and Beilharz
1985; Svartberg and Forkman 2002). There is also behav-
ioral and neurochemical evidence to suggest that tail
wagging is tied into emotion. Dogs in playful or positive
situations (McLeod 1996; Broom 1988) and dogs shown
positive stimuli (Horva
´
th et al. 2007; Quaranta et al. 2007;
Norling et al. 2012) may increase tail-wagging behavior. It
has also been shown that opioid agonists decrease (Pank-
sepp et al. 1983; Knowles et al. 1989) while opioid
antagonists increase tail wagging (Knowles et al. 1989).
Our findings that having the opportunity to solve a problem
and control access to a reward increased tail wagging
suggests that tail wagging can be used as an indicator of a
positive affective state in dogs.
There are two plausible explanations for the different
levels of tail wagging we observed when dogs were
expecting different reward types. Tail wagging could (1)
directly reflect the ‘‘positiveness’’ of the reward or (2)
serve as a better indicator for some motivational systems
over others. Dogs wagged their tails more in the test arena
when they were expecting a food reward or contact with a
human and less when they were expecting contact with
conspecifics. It may be that contact with conspecifics was
experienced as the least positive reward since they were
never deprived of contact with other dogs. The dogs were
removed from their groups immediately before testing and
then returned to their groups immediately after testing.
This did not give them a chance to miss being with their
group mates. The dogs were, however, very much food-
motivated and did not receive treat items on a regular basis
so they were most likely motivated to reach this reward. By
the same light, the dogs received less human contact than
they did dog contact, so were probably more motivated to
reach a friendly human for extra attention than they were to
reach their group mates. Plus reunion with a familiar per-
son is suggested to be positive (Rehn and Keeling 2010).
On the other hand, there is evidence to support the view
that tail wagging could be a better indicator for some
motivational systems than others. Motivation for food is
distinct from motivation for social contact, and dog–dog
and dog–human interaction cater to different motivational
systems as is supported by the fact that the performance of
dog–dog play does not suppress dogs’ motivation to play
with their owners as would be predicted if they were
motivationally interchangeable (Rooney et al. 2000). But,
given the evidence that tail wagging evolved as a mecha-
nism for social signaling and other studies report more tail
wagging toward familiar people than treats (e.g., Norling
et al. 2012), neither of these explanations is entirely
satisfactory.
It is difficult to determine the exact ‘‘Eureka’’ moment,
i.e., when the dogs acting as experimental animals made
the connection between their action and the door to the
runway opening. During the first two or three test runs,
most dogs would remain in the start arena investigating the
operant device for a few seconds after the door to the
runway was opened. After these initial test runs, the dogs
would immediately orient themselves toward the door to
the runway upon successful completion of the operant task,
suggesting that their expectation was for the door to open.
We took an extensive look at the heart rate data collected
for each session the dogs spent in the start arena as both an
experimental and control animal in an attempt to identify
this moment of realization. We analyzed the data collected
immediately before, immediately after, and during each
session in the arena as well as compared the data between
each of the six test runs in attempt to find any association
of changes in heart rate with learning. Given that there are
very few published studies looking explicitly at how heart
rate changes in relation to emotional situations, and given
the overall ambiguity of the studies that are available, it
was difficult to predict what specific reactions we might
have expected to see when dogs are experiencing frustra-
tion or reward during a problem-solving opportunity. We
had however predicted that changes in heart rate would
depend on the personality of the dogs (Carere and Locurto
2011) and the way in which they experienced the problem-
solving event (i.e., with excitement or frustration (Black-
well et al. 2010; Zebunke et al. 2011; Gygax et al. 2013)).
Unfortunately, we were unable to detect any such differ-
ences. It may be that such a change occurs rapidly, ‘‘in the
Eureka moment,’’ and is lost when collecting data with a
sampling rate of every 5 s. Given the known connection
between heart rate and physical activity, it is surprising that
we did not find an increase in heart rate in dogs acting as
experimental animals as they were more active (taking
more steps and wagging their tails more) than when acting
as controls. We did however find an interesting connection
between the physiological response of the dogs in this
study and their cognitive performance in a previous study.
Dogs that had higher heart rates when acting as experi-
mental animals required significantly more training to
reach criterion in a cognitive task (see Burman et al. 2011)
than the dogs that had higher heart rates when acting as
controls. This may provide evidence that success or failure
in problem-solving tasks affects dogs with different per-
sonalities differently (Carere and Locurto 2011). It might
also support the work of Groothuis and Carere (2005) who
classified individuals into those that were good at learning
new tasks (in our case, solving the learning task) and those
who are good at sensing environmental change within a
familiar task (in our case, the cognitive bias task). Indi-
viduals that are skilled in one aspect tend to not be good at
the other, in which case the argument might be that dogs
with different personalities find problem-solving tasks
differently rewarding.
Anim Cogn (2014) 17:577–587 585
123
In conclusion, we found differences between the dogs
acting as experimental and control animals when the only
difference in treatment between these two situations was
the opportunity to learn the causal relationship between
their action and a reward. The yoked control design,
therefore, was effective in separating the emotional effects
of the learning process from those associated with the
expectation of reward. The experimental animals in our
study were excited not only by the expectation of a reward,
but also about realizing that they themselves could control
their access to the reward. These results support the idea
that opportunities to solve problems, make decisions, and
exercise cognitive skills are important to an animal’s
emotional experiences and ultimately, its welfare. From an
evolutionary standpoint, it makes sense that animals should
react emotionally to their own achievements during prob-
lem-solving tasks as, to some degree, heightened states of
emotion can facilitate learning and memory (Hu et al.
2007; Broom 2010; Blackwell et al. 2010) as long as they
are not too intense (i.e., too much excitement or fear can
interfere with the learning process). Positive affective
feelings help animals to better identify behaviors that are
biologically useful and to encourage animals to carry out
these behaviors to their benefit in the long term (Panksepp
1998; Mendl et al. 2010). By providing dogs with a bio-
logically based circumstance under which positive emo-
tional states could develop, we were able to identify tail
wagging as an indicator for measuring such positive states.
The fact that we found clear differences in tail wagging
between treatments and across rewards catering to different
motivational systems (e.g., not just in social situations)
suggests its robustness as an indicator of a positive affec-
tive state in dogs.
Acknowledgments This project was funded by the Swedish
Research Council for Environment, Agricultural Sciences and Spatial
Planning (FORMAS). The final stages of this work were carried out in
collaboration within the Centre of Excellence in Animal Welfare
Science, a Swedish collaborative research environment.
Conflict of interest None.
Ethical standards This study was reviewed and approved by the
Swedish Ethical Committee on Animal Research in Uppsala, Sweden.
All aspects of the study comply with the current laws of Sweden
where the study was performed.
References
Beerda B, Schilder BH, Van Hooff JARAM, DeVries HW, Mol JA
(1999) Chronic stress in dogs subjected to social and spatial
restriction. I. Behavioral responses. Physiol Behav 66:233–242
Bethus I, Tse D, Morris RG (2010) Dopamine and memory: modulation
of the persistence of memory for novel hippocampal NMDA
receptor-dependent paired associates. J Neurosci 30:1610–1618
Blackwell E-J, Bodnariu A, Tyson J, Bradshaw JWS, Casey RA
(2010) Rapid shaping of behaviour associated with high urinary
cortisol in domestic dogs. Appl Anim Behav Sci 124:113–120
Boissy A, Arnould C, Chaillou E, De
´
sire
´
L, Duvaux-Ponter C,
Greiveldinger L, Leterrier C, Richard S, Roussel S, Saint-Dizier
H, Meunier-Salau
¨
n MC, Valance D, Veissier I (2007) Emotions
and cognition: a new approach to animal welfare. Anim Welf
16:37–43
Broom DM (1988) The scientific assessment of animal welfare. Appl
Anim Behav Sci 20:5–19
Broom DM (2010) Cognitive ability and awareness in domestic
animals and decisions about obligations to animals. Appl Anim
Behav Sci 126:1–11
Burman OHP, Parker RMA, Paul ES, Mendl M (2008) Sensitivity to
reward loss as an indicator of animal emotion and welfare. Biol
Lett 4:330–333
Burman O, McGowan R, Mendl M, Norling Y, Paul E, Rehn T,
Keeling L (2011) Using judgement bias to measure positive
affective states in dogs. Appl Anim Behav Sci 132:160–168
Buttelmann D, Tomasello M (2012) Can domestic dogs (Canis
familiaris) use referential emotional expressions to locate hidden
food? Anim Cogn. doi:10.1007/s10071-012-0560-4
Carere C, Locurto C (2011) Interaction between animal personality
and animal cognition. Curr Zool 57:491–498
Chowdhury R, Guitart-Masip M, Bunzeck N, Dolan RJ, Du
¨
zel E
(2012) Dopamine modulates episodic memory persistence in old
age. J Neurosci 32:14193–14204
Cuenya L, Fosacheca S, Mustaca A, Kamenetzky G (2012) Effects of
isolation in adulthood on frustration and anxiety. Behav Process
90:155–160
Fraser D, Duncan IJH (1998) ‘Pleasures’, ‘pains’ and animal welfare:
toward a natural history of affect. Anim Welfare 7:383–396
Goddard ME, Beilharz RG (1985) Individual variation in agonistic
behaviour in dogs. Anim Behav 33:1338–1342
Green TC, Mellor F (2011) Extending ideas about animal welfare
assessment to include ‘quality of life’ and related concepts. N Z
Vet J 59:263–271
Groothuis TGG, Carere C (2005) Avian personalities: characteriza-
tion and epigenesis. Neuro Biobehav Rev 29:137–150
Gygax L, Reefmann N, Wolf M, Langbein J (2013) Prefrontal cortex
activity, sympatho-vagal reaction and behaviour distinguish
between situations of feed reward and frustration in dwarf goats.
Behav Brain Res 239:104–114
Hagen K, Broom DM (2004) Emotional reactions to learning in cattle.
Appl Anim Behav Sci 85:203–213
Harding EJ, Paul ES, Mendl M (2004) Animal behavior: cognitive
bias and affective state. Nature 427:312
Horva
´
th Z, Igya
´
rto
´
BZ, Magyar A, Miklo
´
si A
´
(2007) Three different
coping styles in police dogs exposed to a short-term challenge.
Horm Behav 52:621–630
Hu H, Real E, Takamiya K, Kang M-G, Ledoux J, Huganir RL,
Malinow R (2007) Emotion enhances learning via norepineph-
rine regulation of AMPA-receptor trafficking. Cell
131:160–173
Kalbe C, Puppe B (2010) Long-term cognitive enrichment affects
opioid receptor expression in the amygdala of domestic pigs.
Genes Brain Behav 9:75–83
Knowles PA, Conner RL, Panksepp J (1989) Opiate effects on social
behavior of juvenile dogs as a function of social deprivation.
Pharmocol Biochem Behav 33:533–537
Kupa
´
n K, Miklo
´
si A
´
, Gergely G, Topa
´
l J (2011) Why do dogs (Canis
familiaris) select the empty container in an observational
learning task? Anim Cogn 14:259–268
Langbein J, Nu
¨
rnberg G, Manteuffel G (2004) Visual discrimination
learning in dwarf goats and associated changes in heart rate and
heart rate variability. Physiol Behav 82:601–609
586 Anim Cogn (2014) 17:577–587
123
Langbein J, Siebert K, Nu
¨
rnberg G (2009) On the use of an automated
learning device by group-housed dwarf goats: do goats seek
cognitive challenges? Appl Anim Behav Sci 120:150–158
Lit L, Schweitzer JB, Oberbauer AM (2011) Handler beliefs affect
scent detection dog outcomes. Anim Cogn 14:387–394
McGowan RTS, Robbins CT, Alldredge JR, Newberry RC (2010)
Contrafreeloading in grizzly bears: implications for captive
foraging enrichment. Zoo Biol 29:484–502
McLeod PJ (1996) Developmental changes in associations among
timber wolf (Canis lupos) postures. Behav Process 38:105–118
Meehan CL, Mench JA (2007) The challenge of challenge: can
problem solving opportunities enhance animal welfare? Appl
Anim Behav Sci 102:246–261
Mendl M, Burman OHP, Paul ES (2010) An integrative and
functional framework for the study of animal emotion and
mood. Proc R Soc B 277:2895–2904
Norling Y, Wiss V, Gorjanc G, Keeling LJ (2012) Body language of
dogs responding to different types of stimuli. In: Proceedings of
the 46th congress of the international society for applied
ethology, July 31st–August 4th, Vienna, p 199
Panksepp J (1998) Affective neuroscience: the foundations of human
and animal emotions. Oxford University Press, New York
Panksepp J (2011) The basic emotional circuits of mammalian brains:
do animals have affective lives? Neurosci Biobehav Rev
35:1791–1804
Panksepp J, Conner R, Forster PK, Bishop P, Scott JP (1983) Opiod
effects on social behavior of kennel dogs. J Appl Ethol 10:63–74
Paul ES, Harding EJ, Mendl M (2005) Measuring emotional
processes in animals: the utility of a cognitive approach.
Neurosci Biobehav Rev 29:469–491
Prescott MJ, Morton DB, Anderson D, Buckwell A, Heath S,
Hubrecht R, Jennings M, Robb D, Ruane B, Swallow J,
Thompson P (2004) Refining dog husbandry and care.
BVAAWF/FRAME/RSPCA/UFAW joint working group on
refinement. Lab Anim 38(S1):11–24
Quaranta A, Siniscalchi M, Vallortigara G (2007) Asymmetric tail-
wagging response by dogs to different emotive stimuli. Curr Biol
17:199–201
Rehn T, Keeling LJ (2010) Investigating greeting behaviour in dogs
reunited with a familiar person. In: Proceedings of the 44th
congress of the international society for applied ethology,
August 4–7, Uppsala, p 54
Rooney NJ, Bradshaw JWS, Robinson IH (2000) A comparison of
dog–dog and dog-human play behaviour. Appl Anim Behav Sci
66:235–248
Sakaki M, Niki K, Mather M (2012) Beyond arousal and valence: the
importance of the biological versus social relevance of emotional
stimuli. Cogn Affect Behav Neurosci 12:115–139
Spruijt BM, Van den Bos R, Pijlman FTA (2001) A concept of
welfare based on reward evaluating mechanisms in the brain:
anticipatory behaviour as an indicator for the state of reward
systems. Appl Anim Behav Sci 72:145–171
Svartberg K, Forkman B (2002) Personality traits in the domestic dog
(Canis familiaris). Appl Anim Behav Sci 79:133–155
Wemelsfelder F (2007) How animals communicate quality of life: the
qualitative assessment of behaviour. Anim Welf 16(S):25–31
Zebunke M, Langbein J, Manteuffel G, Puppe B (2011) Autonomic
reactions indicating positive affect during acoustic reward
learning in domestic pigs. Anim Behav 81:481–489
Anim Cogn (2014) 17:577–587 587
123
