Hagen, K. and Broom, D.M. 2004 Emotional reactions to learning in cattle. Appl. Anim.
Behav. Sci. 85, 203-213.
It'has'been'suggested 'that'd urin g'ins trum e nta l'learn ing ,'anim als 'are'lik ely'to're act'e m otion ally'
to'the'reinforcer.'They'may'in'addition 'r ea c t'emotionally'to'their'ow n 'a ch ie ve ments.'These'reaction s'
to'open'a'gate'for'access'to'a'food'reward.'For'he ifers'in'the'contro l'group ,'the'gate'op ene d'after'
race'from'the'gate'to'the'food'trough.'The'hea rt'rate'of'the'heifers,'and'their'beha viou r'wh en'm oving'
improvem e n ts'in 'le ar n in g,'th e y'were'more'likely'th an 'o n 'ot he r 'oc ca sio n s'to 'h av e 'h igh e r'h e ar t'r ate s'a n d'
tended'to'move'more'vigoro usly 'along 'the'race'in'com p arison 'w ith'their'contro ls.'This'exp erim ent'
found'some ,'alb eit 'inc on clu s ive ,'ind ic at ion 'th a t'ca tt le'may'react'emotiona lly 'to 'th e ir'o w n 'le a rn in g'
improvem e n t.'
Emotional responses to stress exposure are greater when the stressful stimuli are not
controllable, than when the subject can learn to control them (Drugan et al., 1997). In the'
204 K. Hagen, D.M. Broom/Applied Animal Behaviour Science 85 (2004) 203–213
absence of fear or pain, is it possible that the ability to control something might in itself be
rewarding? Dogs that are trained to assist people with severe disabilities in everyday tasks,
have been noticed to perform at high levels of excitement, reliability and versatility when
they have learned to experience task solving as intrinsically rewarding (N. Bondarenko,
personal communication). During a previously conducted learning experiment with cattle
(Hagen and Broom, 2003).
Thus, animals might not only get excited about, for instance, the expectation of a reward,
butalso aboutrealising thatthey themselvesto some extent control the deliveryof a reward.
In other words, if they develop an understanding of a causal relationship in which they are
the agents, this might be exciting to them. If this were the case, their emotional reactions in
a situation where they learned a causal relationship should be different from their reactions
in a situation where they just learned to expect something. The differencemight occur either
speciﬁcally during the process of understanding, or it might be retained after a task has been
acquired. To investigate the occurrences of such differences, we designed a yoked control
2.1. Animals and apparatus
Twelve Holstein–Friesian yearling heifers were kept together in a 0.5 ha paddock
from about 3 months prior to the experiment. In addition to grazing, they were fed
a total of 80kg of hay and 1–2kg each per day of concentrates (rolled barley and
The heifers were assigned to six matched pairs on the basis of weight, age, and sire (when
known). One heifer from each pair was randomly assigned to the experimental or control
group. One of the heifers became difﬁcult to handle, resulting in the exclusion of both her
and the heifer matched to her from the experiment, reducing the sample size to ﬁve pairs.
The experimental heifer had to learn a task, whereas her matched control was yoked, that
is, the control heifers obtained the same conditions as their respective matched partners,
irrespective of what they themselves did.
An experimental apparatus was built within the paddock and consisted of a start area
and a race, partly covered with tarpaulin to visually isolate the heifers from the experi-
menter and from the other heifers (Fig. 1). The start area had an entrance gate, through
which the heifers were let in, and an exit gate, which was opened to let them go down
the race. A small panel, mounted on a wooden plate in the start area, 50cm away from
the exit gate and at a height of 1m, was used as an operant for experimental heifers.
The race to which the exit gate gave access was 15m long, and at the end of it there
was a food trough into which the reward was delivered. A control panel for the experi-
menter was located just outside the entrance to the start area. It had two light emission
diodes (LEDs) that signalled when an operant was manipulated, or when latencies had
passed. In addition, it had a switch with which the onset and offset of a trial could be
K. Hagen, D.M. Broom/Applied Animal Behaviour Science 85 (2004) 203–213 205
Fig. 1. Map of apparatus: 1, corner of grazing area; 2, holding area; 3, experimental and feeding area; 4, place for
putting heart rate equipment on; 5, entrance gate to 6, start area; 7, operant; 8, exit gate; 9, race; 10, food trough;
11, building with computer and video equipment. Cameras, c1–c8.
2.2. Structure of the experiment and learning procedure
One session was carried out per day on consecutive days. All heifers had one trial per
session. Trials were carried out with an experimental animal ﬁrst, followed by its control.
At the start of each trial, a heifer was taken out of the holding pen, led to the experimental
area, and fed some concentrates while being ﬁtted with the heart rate measurement device.
She was then led intothe start area of the maze and the entrance gate was closed behind her.
If the heifer was in the experimental group, an LED on the control panel, visible only to
the experimenter, would light up as soon as the heifer had pressed the panel once. The exit
gate would then immediately be opened, allowing access to the race and the food reward.
A computer recorded the time at which the panel was pressed. If the panel was not pressed
within 3 min, some concentrate was fed to the heifer near the panel and the gate was opened.
For control heifers, the computer recorded their panel presses, but gave the light signal for
the experimenter to open the door independently of presses, matched to the latency of the
experimental heifer previously recorded.
Once the gate had opened, the heifer was free to go down the race to the trough and eat
the food. Having eaten, the heifer could exit the race or move back into it. However, it was
not possible for the heifer to get into the start area. Three minutes after she had ﬁnished
eating, the heifer was rejoined with the group outside the experimental area.
2.3. Data collection and analysis
Latencies from when a heifer’s head was inside the start area until she made the exit
gate open were categorised into four groups with equal counts. The latency categorisation
and the proportion of behaviour that the heifer in the start area directed towards the operant
were used to assign a performance index (Table 1) to the experimental heifers for each trial.
206 K. Hagen, D.M. Broom/Applied Animal Behaviour Science 85 (2004) 203–213
Scheme used to assign a performance index
Panel pressed? Latency before
Proportion of time
directed towards operant
No – – 0
Yes >60s <half 2
21–60s <half 3
9–20s – 5
<9s – 6
For further analysis, the difference between the performance index of an animal on the trial
in question and its index on the day before was calculated. This change was categorised
into a binary variable where 0 denotes no change from previous day or lower performance
(difference values from −5 to 1) and 1 denotes that performance is clearly better than on
the day before (difference values >1). This binary performance change index was used for
further analysis and is referred to as ‘learning index’. Notes were also kept on behaviours
such as repeated butting of the operant panel after the exit gate had opened. In addition,
whether the heifers chose to go back into the race after they had eaten their reward, and
whether they tried to get back into the start area or tried to reach the panel was recorded.
The heifers’ behaviour from the time that they entered the experimental area and had the
heart rate monitor ﬁtted until they were led out, was recorded on video with eight cameras
(Fig.1).The time taken to go from theexitgatetothe food trough and behavioursduring this
locomotion (Table 2) were coded from the video records by two observers. Before coding,
all video clips were edited out of context and compiledon new tapes in random order. Inter-
andintra-observeragreementwasensuredby randomly interspersedre-observationofclips.
For further analysis, an index for the gait was derived from a combination of the ‘main gait’
and ‘other gait’ classiﬁcation as outlined in Table 3.
Polar Sport Testers storing 5s interval averages (Polar Electro Oy, Finland) were used
to record heart rate. For each heifer, the mode of her heart rate across all measurements
Behaviour during locomotion
Variable Levels Description
Main gait Walk, trot, gallop/canter The gait that dominates the clip
Other gait None, walk, trot, gallop/canter Additional gait that occurs during the clip (if several
additional gaits occur, the one that appears most)
Jump Yes, no The front legs are lifted together, top line descends
sharply from back to front and all feet are then in the air
Buck Yes, no Both hind legs are lifted simultaneously and the top line
ascends sharply from front to back
Kick Yes, no One of the hind legs is extended back- and sideward in a
K. Hagen, D.M. Broom/Applied Animal Behaviour Science 85 (2004) 203–213 207
Derivation of gait index from raw scores of ‘main gait’ and ‘other gait’
Possible combinations Gait index
Main gait Other gait
Walk None 1
Trot Walk 3
Gallop Walk 5
was calculated as an individual baseline. For further analysis, the mean deviation from an
individual’s mode was calculated for functional phases of the experiment: before a heifer
entered the start area (a); during the last 15s before the exit gate was opened (b); during the
ﬁrst 10s after gate has been opened (c); 10–20s after gate had been opened (d); from 20 s
after gate had been opened until the heifer stopped eating (e); after the heifer had stopped
eating (f). The deviation of the heart rate in each phase from an individual’s mode was
expressed as a percentage higher or lower than the mode, i.e. deviation = ((mean/mode) −
1) × 100.
As a consequence of the experimental design, pairwise differences between experimen-
tal subjects and their matched controls were of particular interest as dependent variables.
Pairwise differences were always calculated as experimental-control, i.e. the pairwise heart
rate differences were the differences, in each functional phase of each trial, between an
experimental heifer’s heart rate deviation from mode, and her matched control’s heart
rate deviation from mode. Only cases where both were available were included in the
Prior to non-parametric tests following Siegel and Castellan (1988), proportions and
numerical values were averaged to one value per subject per treatment level to ensure
independence of the data. Effectson heart rate were investigated with a general linear mixed
model calculated in R (Ihaka and Gentleman, 1996). The repeated measures design and
individualdifferences were taken into account by including the individual pairs as a random
factor. Normality of the residuals’ distributions was tested with the Kolmogorov–Smirnov
test and homogeneity of variances was tested with the Bartlett test.
Thelatencytopressthe panel variedfrom upto 3min (at thebeginningofthe experiment)
to 3 s when the heifers had learned the task well. Fig. 2 showsthe performance index and the
derived learning index for the experimental heifers. The ﬁgure also shows the occurrence
of panel investigation or panel-directed play behaviour after the gate has opened, and of
208 K. Hagen, D.M. Broom/Applied Animal Behaviour Science 85 (2004) 203–213
Fig. 2. Performanceindexon each day foreach of theheifers. Numbers within squares indicate where the resulting
learning indexis 1. Asterices indicate that the heifer showed investigativeor play-like behaviour towards the panel
after the gate had opened. Plus signs indicate that the heifer went back into the race and investigated the gate after
K. Hagen, D.M. Broom/Applied Animal Behaviour Science 85 (2004) 203–213 209
gate investigation after the heifer has eaten the food. Only one of the control group heifers
showed such behaviour on one occasion.
The time taken to move down the race, from when a heifer’s head was through the
exit gate until it was in the food trough, ranged from 4 to 71s. After a decrease from the
ﬁrst day, the values remained close to a median of 8 s. There was no difference between
experimental and control animals (Wilcoxon signed-ranks test: T = 9, N = 5, P = 0.81),
and the differences in speed between experimental and controlheifers within matchedpairs
were not inﬂuenced by the binary learning index (T = 12, N = 5, P = 0.32).
The main gait when moving down the race was walk in 88 cases (74%), trot in 26 cases
(22%), and gallop in 5 cases (4%). The distribution of the gait index was thus strongly
skewed towards lower values. It was not correlated with the speed when moving down the
race (Spearman rank correlation: r
=−0.05, N = 112, P = 0.59). Gait scores increased
over the experiment (Page test: L = 1232, k = 7, N = 10, P<0.001), but they did
not differ between experimental and control groups (Wilcoxon signed-ranks test: T = 10,
N = 5, P = 0.63). There was a trend for a relationship (T = 15, N = 5, P = 0.062)
between the binary learning index and the pairwise differences in gait scores between the
experimental and control heifers: when the learning index was 1, the experimental heifers
tended to be more likely to score higher than their matched controls, than when it was
0. Jumping occurred in ﬁve cases: heifer 4 on day 13, heifer 12 on days 6 and 12, heifer
8 on day 2 and heifer 5 on day 5; bucking in three cases: heifers 12 and 5 on the same
occasions as when they jumped; and kicking in one case: heifer 4 on day 6 (see Fig. 2 for
comparison with their learning curves). In the control group, neither jumping, bucking or
kicking occurred on any occasion.
Individuals’ median heart rate values differed but were not correlated with body weight
or age. There were no overall treatment group differences (Wilcoxon signed-ranks test:
T = 8, N = 5, P = 0.89). Heart rate varied across functional phases (Friedman two-way
Fig. 3. Deviation of heart rate from mode (mean percentage ± S.E.) pooled over days for experimental (䉱, solid
line) and control (
䊉, dashed line) groups, in the functional phases of the experiment: a, up to 15 s before gate
opens; b, last 15 s before gate opens; c, ﬁrst 10 s after gate has opened; d, 10–20 s after gate has opened; e, while
eating reward; f, after eating.
210 K. Hagen, D.M. Broom/Applied Animal Behaviour Science 85 (2004) 203–213
Fig. 4. Deviationofheart ratefrom mode(mean percentage± S.E.) pooledoverfunctional phases for experimental
䉱, solid line) and control (䊉, dashed line) groups, on each day of the experiment.
analysis of variance: F
= 33.7, d.f. = 5, N = 10, P<0.001; Fig. 3) and increased with
time (days of the experiment) in the control group (Spearman rank correlation: r
N = 60, P<0.001), with a trend for increase in the experimental group (r
N = 60, P = 0.055). Inspection of Fig. 4 shows that the groups only differed on the ﬁrst 2
days, and that the relationship between heart rate and time was relatively linear from day 3
to day 11.
Heart rate while, and shortly after, going down the race was not correlated with the
number of seconds taken to go down the race (Spearman rank correlation: r
N = 54, P = 0.46 for control group in phase c; r
= 0.04, N = 48, P = 0.80 for
experimental group in phase c; r
= 0.22, N = 55, P = 0.12 for control group in phase d;
Fig. 5. Pairwise differences in deviation of heart rate from mode (mean percentage ± S.E.) in relation to the
learning index for experimental heifers.
K. Hagen, D.M. Broom/Applied Animal Behaviour Science 85 (2004) 203–213 211
=−0.08, N = 49, P = 0.59 for experimental group in phase d). It did correlate weakly
with the gait index in the control group (r
= 0.31, N = 52, P = 0.026 for control group
in phase c; r
= 0.07, N = 52, P = 0.23 for experimental group in phase c; r
N = 52, P = 0.019 for control group in phase d; r
= 0.17, N = 53, P = 0.23 for
experimental group in phase d). A stronger correlation was found between gait index and
heart rate for both groups in phase b, i.e. just before the gait opened (r
= 0.46, N = 52,
P = 0.001 for control group; r
= 0.31, N = 52, P = 0.024 for experimental group).
Nine extreme values (heart rate deviation of more than 50%) were excluded to achieve
a normal distribution of residuals in the general mixed model for effects on the differences
in heart rate deviation between experimental group and control group. The only signiﬁcant
factor in the ﬁtted model was the learning index (ANOVA: F
= 6.98, P = 0.0087,
Five of the six experimental heifers acquired the operant tasks and reached levels of
reliable and quick performance within 12 trials. This ease of trainingcattle to operanttasks,
provided that they are not scared, is in line with previous ﬁndings (Kiley-Worthington and
As the heifers had no prior experience of the start box and maze, it is not surprising
that on their ﬁrst trials they took longer to go down the race than on subsequent trials. In
the later course of the experiment, speed remained stable and was not correlated with the
gait index, which increased in both treatment groups. On days when learning performance
increased, heifers tended to have higher gait scores than their matched controls. Jumps,
bucks and kicks only occurred in the experimental group. Taken together, the behaviours
while moving down the race indicate more agitation in the experimental group when the
learning curve was steep.
Heart rate dropped in the two groups on days 1 and 2, respectively, and then rose in both
groups until day 11. This may reﬂect the stress of habituation to the learning apparatus, in
similar way as has been observed in the context of visual discrimination learning in dwarf
goats (Langbein et al., 2003), although in their case, a similar tendency was observed not
during learning itself, but in periods of rest between learning sessions. In our study, heart
rate also varied with functional phases, with peaks around the time of the main locomotor
activity and a low during eating, reﬂecting the amount of movement. However, heart rate
was only weakly correlated with the gait index. Metabolic activity could therefore not
explain the variation in heart rate. Pairwise differences between experimental heifers and
theirmatched controls with regardto heart rate deviationfrom individualmode were greater
when the learning index was 1, than when it was 0. This indicates that there was a treatment
effect on heart rate which corresponds to the effect on movement.
The experimental animals got more excited than the control animals, not generally, but in
a temporal relation to the only difference between the groups: that the experimental heifers
experienced an operant learning process whereas the control heifers did not. There are two
experimental heifers were reacting to their own learning process and thus in a sense to their
212 K. Hagen, D.M. Broom/Applied Animal Behaviour Science 85 (2004) 203–213
own achievement. Secondly, it could also be argued that increased arousal or motivational
levels occurred randomly, but led to better performance. We speculate that the arousal
leading to improved performance may well be the result of a process of understanding after
the previous trial, rather than occurring randomly.
The investigation of non-verbal ‘self’ and of consciousness in animals and in infants have
been focused on cognitive abilities and self-recognition in mirrors (Marten and Psarakos,
1995; Mitchell, 1997; De Veer and Van den Bos, 1999; Shillito et al., 1999; Swartz et al.,
1999). In the light of theories of self-consciousness that make attempts at explaining its
evolution and ontogeny in terms of emotional and bodily responses rather than cognition
only (Neisser, 1991, 1997; Bermúdez et al., 1995; Bermúdez, 1998; Damasio, 2000), there
is a challenge to devise alternative empirical approaches (Rochat and Hespos, 1997; Rochat
andGoubet,2000).The present study represents anattemptatestablishing such an approach
to investigating ways in which there may be non-verbal self-referral, as there would be in
the emotional response to one’s own understanding.
One assumption underlying the experimental design was that the operant acquisition
task involved such a process of understanding. However, the yoked control design is
not in itself sufﬁcient to demonstrate instrumental learning (Church, 1988; Church et al.,
1989). The idea of a process agency, and of understanding the causal connection between
a response and a reward, can be accommodated in theories of goal-directed action and
incentive learning (Dickinson and Balleine, 1994). Further investigation of emotional re-
actions to learning would involve validation of the instrumental nature of the learning
To our knowledge, our experimental approach to investigate whether animals might
respond emotionally to a process of understanding, has not been used before. The yoked
control design was ﬁrst used to show that rats with control over the termination of electric
shocks developed fewer stomach ulcers than their yoked controls (reviewed by Drugan
et al., 1997). In the present experiment, the purpose was to control for all variables that
might inﬂuence emotional responses other than the learning process.
In conclusion, the yoked control design was effective to some extent in separating the
effectsof the operant learningprocess from other variableslike habituation and expectation
of reward. The study indicated that cattle might be more agitated when they are just about
to acquire a task, i.e. understand, and thus that they may have an emotional perspective on
their own agency. However, because of the novelty of the approach and the small number
of animals, this study should be seen as a ﬁrst step towards further investigation of the
EricAllen and GavinHughesatCambridge UniversityFarm for lettingus work with their
Touma and Sophie Prowse for practical help. Richard Kirkden for help with programming
and electronics, and Irene Rochlitz for advice and for reading an earlier draft of this paper.
KH was supported by the Overseas Research Students’ Awards Scheme, the Cambridge
Overseas Trust and a St. John’s College Benefactors’ Scholarship.
K. Hagen, D.M. Broom/Applied Animal Behaviour Science 85 (2004) 203–213 213
Bermúdez, J.L., 1995. Ecological perception and the notion of a nonconceptual point of view. In: Bermúdez, J.L.,
Marcel, A., Eilan, N. (Eds.), The Body and the Self. MIT Press, London, pp. 153–174.
Bermúdez, J.L., 1998. The Paradox of Self-consciousness. MIT Press, London.
Church, R.M., 1988. Yoked control designs for assessment of contingency. Behav. Brain Sci. 11, 451–452.
Church, R.M., 1989. The yoked control design. In: Trevor, A., Nilsson, L.G. (Eds.), Aversion, Avoidance,
and Anxiety: Perspectives on Aversively Motivated Behaviour. Lawrence Erlbaum Associates, Hillsdale,
Damasio, A., 2000. The Feeling of What Happens: Body, Emotion and the Making of Consciousness. Vintage,
De Veer, M.W., Van den Bos, R., 1999. A critical review of methodology and interpretation of mirror
self-recognition research in nonhuman primates. Anim. Behav. 58, 459–468.
Dickinson, A., Balleine, B., 1994. Motivational control of goal-directed action. Anim. Learn. Behav. 22, 1–18.
Drugan, R.C., Basile, A.S., Ha, J.H., Healy, D., Ferland, R.J., 1997. Analysis of the importance of controllable
versus uncontrollable stress on subsequent behavioral and physiological functioning. Brain Res. Protoc. 2,
Hagen, K., Broom, D.M., 2003. Cattle discriminate between individual familiar herd members in a learning
experiment. Appl. Anim. Behav. Sci. 82, 13–28.
Ihaka, R., Gentleman, R., 1996. R: a language for data analysis and graphics. J. Comput. Graph. Stat. 5, 299–314.
Kiley-Worthington, M., Randle, H., 1998. Animal educational psychology: a comparative study of teaching four
mammals of different species. Anim. Cogn. Training 013b.
Langbein, J., Nürnberg, G., Manteuffel, G., 2003. Telemetric measurement of heart rate and heart rate variability
for evaluating psychological stress induced by visual discrimination learning in dwarf goats. In: Cox, S. (Ed.),
Precision Livestock Farming. Wageningen Academic Publishers, Wageningen, pp. 105–111.
Marten, K., Psarakos, S., 1995. Using self-view television to distinguish between self-examination and social
behavior in the bottlenose dolphin (Tursiops truncatus). Conscious. Cogn. 4, 205–225.
Mitchell, R.W., 1997. Kinesthetic-visual matchingand the self-concept as explanations of mirror-self-recognition.
J. Theor. Soc. Behav. 27, 17–39.
Neisser, U., 1991. Two perceptually given aspects of the self and their development. Dev. Rev. 11, 197–209.
Neisser, U., 1997. The roots of self-knowledge: perceiving self, it, and thou. In: Snodgrass, J.G., Thompson, R.L.
(Eds.), The Self Across Psychology: Self-recognition, Self-awareness, and the Self Concept. Ann. NY Acad.
Sci., 19–33. New York Academy of Sciences, New York.
Rochat, P., Hespos, S.J., 1997. Differential rooting response by neonates: evidence for an early sense of self. Early
Dev. Parenting 6, 105–112.
Rochat, P., Goubet, N., 2000. Connaissance implicite du corps au debut de la vie. Enfance 52, 275–285.
Shillito, D.J., Gallup, G.G., Beck, B.B., 1999. Factors affecting mirror behaviour in western lowland gorillas,
Gorilla gorilla. Anim. Behav. 57, 999–1004.
Siegel, S., Castellan, N.J., 1988. Nonparametric Statistics for the Behavioural Sciences. McGraw-Hill,Singapore.
Swartz,K.B., Sarauw, D., Evans, S., 1999. Comparativeaspects of mirror self-recognition in great apes.In:Parker,
S.T., Mitchell,R.W. (Eds.), The MentalitiesofGorillas and Orangutans:Comparative Perspectives. Cambridge
University Press, New York, pp. 283–294.