Using judgement bias to measure positive affective state in dogs

Applied Animal Behaviour Science 132 (2011) 160–168 Contents lists available at ScienceDirect Applied Animal Behaviour Science journal homepage: www.elsevier.com/locate/applanim Using judgement bias to measure positive affective state in dogs Oliver Burman a,b,∗ , Ragen McGowan a , Michael Mendl b , Yezica Norling a , Elizabeth Paul a,b , Therese Rehn a , Linda Keeling a a b Department of Animal Environment and Health, Swedish University of Agricultural Sciences, Box 7068, SE-750 07, Uppsala, Sweden Department of Clinical Veterinary Science, University of Bristol, Langford, BS40 5DU, UK a r t i c l e i n f o Article history: Accepted 3 April 2011 Available online 7 May 2011 Keywords: Positive affect Dogs Cognition Animal welfare a b s t r a c t Interest in the induction and measurement of positive affective states in non-human animals is increasing. Here, we used a test of cognitive (judgement) bias, based on the finding that individuals experiencing different affective states judge ambiguous stimuli differently, to measure whether a positive low arousal affective state (e.g. ‘satisfaction’/‘contentment’) could be induced in domestic dogs as a result of their experiencing a food-based rewarding event. In this rewarding event, subjects (1 year old female Beagles) had to search for small amounts of food randomly placed within a maze arena. Using a balanced withinsubjects design, the dogs (N = 12) received a cognitive bias test either without experiencing the rewarding event (the ‘Neutral’ treatment), or directly after experiencing the rewarding event (the ‘Post-consumption’ treatment). In the test, dogs were trained that one visual cue (e.g. dark grey card) predicted a positive event (food in a bowl) whilst a different cue (e.g. light grey card) predicted a relatively ‘negative’ event (empty bowl). We hypothesised that dogs tested after experiencing the rewarding event, and in a presumed post-consummatory positive affective state, would be more likely to judge visually ambiguous stimuli (intermediate grey cards) positively, compared to dogs in the ‘Neutral’ treatment. In contrast, we found that they took significantly longer to approach an intermediate ambiguous stimulus, suggesting that they were less likely to anticipate food (a negative judgement) compared to dogs in the ‘Neutral’ treatment group. Various explanations for the observed results are discussed, in particular how reward acquisition and consumption may influence positive affective state induction in animals. © 2011 Elsevier B.V. All rights reserved. 1. Introduction There is an increasing acceptance that the study of affective states in non-human animals (henceforth animals) is a critical component in our understanding of animal welfare (e.g. Dawkins, 1990, 2000). Historically, there has been an emphasis on the study of negative, rather than positive, affective states (Reefmann et al., 2009), and in ∗ Corresponding author at: Department of Biological Sciences, University of Lincoln, Riseholme Park, Riseholme, Lincoln LN2 2LG, UK. Tel.: +44 01522895453. E-mail address: [email protected] (O. Burman). 0168-1591/$ – see front matter © 2011 Elsevier B.V. All rights reserved. doi:10.1016/j.applanim.2011.04.001 animal welfare science this has resulted in a focus on indicators of poor welfare, such as abnormal behaviour (e.g. Mason, 1991; Garner et al., 2004; Moinard et al., 2003; McAdie and Keeling, 2000) and physiological ‘stress’ responses (e.g. Burman et al., 2008a; Rooney et al., 2007a; Dawkins et al., 2004). Such an emphasis is understandable, in that the first responsibility of animal care and management should be to minimize negative affective states such as suffering, fear, pain, and distress (e.g. the ‘Five Freedoms’: Farm Animal Welfare Council (FAWC), 1992). Whilst there has been a recent increase in interest in the assessment of positive affect (e.g. Boissy et al., 2007; Yeates and Main, 2008), and this change in focus acknowledges that animal welfare encompasses both positive and negative O. Burman et al. / Applied Animal Behaviour Science 132 (2011) 160–168 affective states/experiences (Farm Animal Welfare Council (FAWC), 2009), there remains a relative paucity of research not only concerning under what circumstances such positive affective states might occur, but also how they can be measured. If affective states are states elicited by rewards and punishers, and rewards are “anything for which an animal will work” (Rolls, 2005), then different phases of reward acquisition and consumption can be identified, i.e. appetitive/anticipation of reward (e.g. food, drink, sex), consumption of reward, and post-consumption. Because these phases are all linked to acquiring rewards they are likely to induce positive affective states (e.g. Rolls, 2005) except when an anticipated reward is unexpectedly absent, in which case negative affective states of ‘disappointment’ and/or frustration may be generated (e.g. Burgdorf and Panksepp, 2006; Burman et al., 2008b). In the current study we decided to investigate whether or not a positive affective state associated with the post-consummatory phase of reward acquisition could be induced in dogs. Unlike the anticipatory and consummatory phases of reward acquisition that have both been the focus of some attention (e.g. Bos et al., 2003; Moe et al., 2006; Burgdorf and Panksepp, 2006), comparatively little research has been directed towards the positive affective states that might be experienced ‘post-consumption’. Such research is of particular interest because indicators of positive affective state induced ‘post-consumption’ may have more cross-species similarities compared to other phases of reward acquisition, thereby providing more convenient indicators of positive affect than those that are less applicable to a range of different species. For example, the behavioural expression of anticipation can differ between species (e.g. rats and cats (Bos et al., 2003)). Measuring affect in animals is challenging and, as mentioned above, traditional measures tend to focus on negative affective states such as fear and anxiety, and some of these measures (e.g. heart rate, cortisol/corticosterone) may be better at detecting arousal rather than valence (positivity or negativity). Cognitive measures, in contrast, are particularly suited to detecting the valence of affective states (e.g. Paul et al., 2005), and hence are used in the current study. The particular cognitive measure selected for use, cognitive affective bias (henceforth cognitive bias), is based on empirical findings from humans, and theoretical predictions, that affective state influences cognitive processes including judgement, memory and/or attention. Specifically, individuals in a negative affective state pay more attention to threatening stimuli, retrieve more negative memories, and judge ambiguous stimuli more negatively (are relatively ‘pessimistic’) than happier individuals (e.g. Paul et al., 2005; Mendl et al., 2009, 2010a,b). Cognitive bias studies based on the judgement of ambiguity have successfully been used to inform researchers about affective valence (generally negative) in a range of animal species, including rats (e.g. Harding et al., 2004; Burman et al., 2009), starlings (e.g. Matheson et al., 2007; Brilot et al., 2010), dogs (e.g. Mendl et al., 2010a) and sheep (e.g. Doyle et al., 2010). Here, we utilise a judgement bias technique to assess a putative positive affective 161 state, and use dogs as our subject species because they appear sensitive to the induction of positive affective states, e.g. via play interaction with humans (e.g. Rooney et al., 2001), are able to learn visual discrimination tasks (e.g. Pretterer et al., 2004) and are common companion animals (e.g. c. 10 million owned dogs in the UK in 2006 (Murray et al., 2010)). Our prediction is that, following the experience of a rewarding event (i.e. ‘post-consumption’), dogs tested in a cognitive bias task will be more likely to judge ambiguous stimuli positively, i.e. be more ‘optimistic’, due to experiencing a post-consummatory positive affective state (e.g. satisfaction/contentment), compared to when tested without experiencing the rewarding event. 2. Methods 2.1. Subjects, housing and husbandry The subjects were 12 young female Beagles of between 11 and 12 months of age at testing (average body weight of 10.5 kg) housed at the Swedish University of Agricultural Sciences for use in behavioural observation studies of positive affective state (e.g. Rehn and Keeling, 2010; McGowan et al., 2010). Of the 12 dogs, there were four pairs of full sisters, two pairs of which were half-sisters with one other dog. The dogs were housed in indoor enclosures (9 m × 2.7 m) between 1600 and 0800 h in groups of three, and in outdoor enclosures (5.8 m × 25 m) between 0800 and 1600 in groups of between 6 and 12 animals. Food (Hill’s Adult Advanced Fitness) was provided at 0800, immediately before the dogs were taken outside, and at 1600, when the dogs returned indoors. Water bowls were available indoors and outdoors. Both indoor and outdoor enclosures were also provisioned with ‘enrichment’ items, including rawhide chews, footballs and soft toys. The outdoor enclosures also contained wooden chalets for shelter (these were also used as vantage points/resting places) and a water bath for thermoregulation (the study was carried out in the summer). Dogs were taken for walks on leads once a day in groups. This study was reviewed and approved by the Regional (Uppsala County) Ethical Committee on Animal Experiments. 2.2. Familiarisation period Prior to the start of the study, the dogs were familiarised with the two test rooms, two ‘maze’ arenas (see Section 2.4), the researchers, and two different types of food reward (standard food pellets and FrolicTM ). This familiarisation consisted of five days during which each dog was allowed to explore individually the rooms and ‘maze’ arenas in the presence of the researchers for 5 min, followed by a 5 min period of training (e.g. to sit on command), once per day. This also allowed the dogs to become used to brief periods of separation from conspecifics. The dogs were familiar with both food types, receiving standard food pellets as their regular diet and being rewarded with Frolic during training. 162 O. Burman et al. / Applied Animal Behaviour Science 132 (2011) 160–168 2.3. Treatments We labelled the two treatments to be investigated ‘Neutral’ and ‘post-consumption’. The ‘post-consumption’ treatment consisted of bringing each dog from their outdoor pen to the test room and exposing them to a rewarding event (see below). Once the reward task had been completed, the dogs were tested on the cognitive bias task. In contrast, the ‘Neutral’ treatment consisted of bringing each dog from their outdoor pen to the test room and immediately testing on the cognitive bias task (i.e. without experiencing the rewarding event). The two treatments therefore differed in all aspects of the rewarding event, including: gaining food rewards; searching activity; and contact time with humans, and this whole experience was therefore considered to be part of the affect-inducing ‘postconsumption’ treatment. Following cognitive bias testing, the dogs were returned to the outdoor pens. All dogs received both treatments, balanced for order, with a one week interval between treatments. Both treatments consisted of three consecutive days of testing (see later). The treatments took place between 1100 and 1500, balanced for order. 2.4. Rewarding event The rewarding event was a ‘search and forage’ task in which small amounts of food were randomly placed within a maze arena. We decided to use a foraging task rather than simply providing food directly because working to obtain food (i.e. searching and finding) may be particularly rewarding for animals (e.g. de Jonge et al., 2008). The dogs had to use visual and olfactory senses to locate the food as well as learning and memory abilities to navigate their way around the maze, e.g. to avoid re-visiting areas that they had already depleted of food within the same trial. The task was selected in order to stimulate a period of arousing appetitive behaviour in the dogs, incorporating elements known to be rewarding for dogs (e.g. food (Schipper et al., 2008); searching (Rooney et al., 2007b; Kaminski et al., 2008)). We used two identical areas to construct mazes. Half the dogs were tested in one side, and half in the other. Both mazes had the same layout (see Fig. 1), with long and short corridors as well as dead ends. During testing (post-consumption treatment only), we collected the dog and took it to the release point located 10 m from the entrance to the maze. One researcher held the dog whilst another scattered the food reward (standard food pellets (30) and quarters of FrolicTM (4)) randomly throughout the maze. On the sounding of an artificial tone (i.e. an electronic beep), the dog was released and allowed to enter the maze. We recorded the time taken for the dog to reach the maze entrance from the release point as an estimate of general motivation for the task. The dog was given 5 min to explore the maze and find the food. Once the 5 min was completed, the dog was collected, returned to the release point and the procedure repeated a further two times. The total maximum food intake was 90 standard food pellets (40 g) and 3 FrolicTM , well below the normal intake of the dogs at feeding time (160 g, twice daily). Once the dog completed its final search of the Fig. 1. A diagram showing the maze arena for the rewarding event. The dog started 10 m away from the maze. We randomly scattered food (diet pellets: black circles; quarters of FrolicTM : white squares) throughout the maze. Solid barriers/walls are indicated by solid lines, and mesh barriers (through which the dog could see/smell but not gain direct access to the food) are indicated by dotted lines. maze, the cognitive bias testing was begun. Thus, during the ‘post-consumption’ treatment only, each dog received three consecutive days exposure to the reward task, with three separate searches on each test day. 2.5. Cognitive bias task We adapted a previously developed methodology (see Bateson and Matheson, 2007) training the dogs to discriminate between five different shades of grey. The chosen grey shades were selected using the ‘Red Green Blue colour model’ (RGB: e.g. Gevers and Smeulders, 1999) ranging from 0 to 255 (the equivalent to a greyscale of 0–100%), including: (1) RGB 240 (greyscale 6% – very light grey); (2) RGB 185 (greyscale 27.5%); (3) RGB 130 (greyscale 49%); (4) RGB 75 (greyscale 70.5%); (5) RGB 20 (greyscale 92% – very dark grey). There was a gap of 55 RGB (21.5% greyscale) between each of the five grey shades, a similar difference to that used previously (e.g. 20% greyscale (Bateson and Matheson, 2007)). Dogs have been shown to discriminate successfully between shades of grey much closer in hue than those used here (e.g. Pretterer et al., 2004). In order to present the grey shades to the dogs, we constructed goal boxes (30 cm × 22 cm × 22 cm) made of cardboard. Each box had a high quality laser-printed A4 piece of card of the appropriate grey shade glued to the front, the bottom and the back of the inside of the box, with one goal box constructed for each shade. All pieces of card were printed in the same batch to ensure consistent quality. Before the start of each trial, a clear plastic bowl was placed inside the goal box and a food reward placed into the bowl if the trial was to be rewarded, but no food if the trial was to be unrewarded. The dogs therefore could O. Burman et al. / Applied Animal Behaviour Science 132 (2011) 160–168 163 Fig. 2. The experimental set up in the testing room. The goal box was placed in the same location 6 m from the start position. It could be one of five shades of grey, with the darkest and lightest shades being the unrewarded and rewarded “reference” shades. The three ambiguous grey shades were intermediate (but equally spaced) between the reference shades. The “handler” remained at the start position, the “timer” remained out of sight of the dog. Sheets of the appropriate shade of grey were attached to the front, bottom and back of the goal box. The food dish upon which the food reward was placed within the goal box was transparent. not see the plastic bowl, or the food within it, because it was placed inside the goal box that provided the visual cue for the dogs. The same plastic bowl was used for each trial (i.e. switched between different goal boxes) so that it smelt of food even when unrewarded. Prior to the start of training we ensured that each dog ate from the goal box, and all dogs quickly learned to approach the box. During training the dogs were exposed to two boxes, but only ever one box at a time, either ‘240RGB’ (the lightest grey) or ‘20RGB’ (the darkest grey) (see Fig. 2), one of which was rewarded (half a piece of FrolicTM ), the other unrewarded (no food). The greyscale shade of these ‘reference’ boxes were balanced between dogs so that for half the dogs the ‘240RGB’ box was rewarded and the ‘20RGB’ box unrewarded, and vice versa and this remained the same for each dog for the duration of the experiment. At the start of training, each dog received two consecutive rewarded trials in order to confirm the presence of food in the bowl for the dog before introducing any unrewarded trials that could potentially extinguish the response. The dog then received two consecutive unrewarded trials to allow the dog to experience that food may not always present, depending on the grey shade of the bowl. The dogs were subsequently presented with either rewarded or unrewarded trials in a pseudorandom order, with no more than two rewarded or unrewarded trials presented consecutively. The goal box was placed in the same location in the test room every time, with the dogs released from a starting point 6 m away (see Fig. 2). One researcher (the ‘handler’) sat at the start position in order to release the dog at the beginning of each trial, and a second researcher (the ‘timer’) was positioned out of sight of the dog adjacent to the goal box in order to record the time taken for the dog to reach the goal box in each trial with a stopwatch. The ‘timer’ was also responsible for changing the goal box between trials, and determining the start of each trial. The criterion for learning this visual discrimination task was that each dog must run faster to the rewarded than to the unrewarded goal box for six consecutive trials (half rewarded, half unrewarded), with at least half a second difference between the slowest run to the rewarded box and the fastest run to the unrewarded box. 2.6. Trial procedure Baiting of the goal boxes was done out of sight of the dog and each trial was kept as similar as possible in terms of preparation time and activity – even if the same trial type was to be repeated. The start of each trial was signified by an artificial tone, upon which the ‘handler’ released the dog and the latency for the dog to reach the bowl, defined as putting head in line with the edge of the box, was recorded. Dogs were held in the same way throughout training/testing and the handler did not look at the goal box during the setting up of each trial, being effectively ‘blind’ to the trial, so that they could have no influence on the dog’s response. The dogs were given a maximum of 30 s to visit the box during a trial. If, after this time, they had not visited the box, a time of 30 s was noted and the next trial begun as though the dog had reached the bowl. 2.7. Training stages Two weeks before the start of the study, dogs were given up to 104 trials in one session (lasting about 1 h) in which to reach criterion. If they failed to reach criterion in this session, then they were given a further session of 104 trials every day until criterion was achieved (see Table 1), with 164 O. Burman et al. / Applied Animal Behaviour Science 132 (2011) 160–168 Table 1 The number of trials that each dog required to reach criterion during (1) initial training; (2) “refresher” training 1; (3) “refresher” training 2. Italics are used to highlight when criterion was achieved within ≤15 trials. Dog Trials to criterion Initial training Trials to criterion “Refresher” training 1 Trials to criterion “Refresher” training 2 Strimma Doppa 155 trials 136 trials 9 trials 7 trials Vixi Daisy 200 trials 45 trials Saga 36 trials Ginger 42 trials Arista 108 trials Belle 30 trials Trolla 108 trials Chaos Tingeling 69 trials 99 trials Jasmin 90 trials 10 trials 104 trials 10 trials 11 trials 29 trials 13 trials 21 trials 24 trials 23 trials 18 trials 9 trials 36 trials 13 trials 54 trials 10 trials 30 trials 15 trials 13 trials 21 trials 35 trials 13 trials 10 trials 9 trials 7 trials 9 trials 18 trials 7 trials 10 trials 13 trials 15 trials 10 trials 7 trials 13 trials sessions terminated as soon as criterion was reached. Two additional periods of ‘refresher’ training were carried out to ensure that all dogs were able to carry out the discrimination after a two week interval, but also that they were able to reach criterion within 15 trials, as this was the number that they would receive during testing (see later). This additional training occurred no longer than three days prior to the start of the testing for both treatments. Each dog was trained until it reached criterion, if this was achieved in 15 trials or less then the dog was considered ready for the experiment and no further training was given. If it failed to reach criterion in 15 trials or less, then it was re-trained on further days until criterion was achieved in 15 trials or less. We found that 11/12 dogs were able to do this (see Table 1) during the first period of ‘refresher’ training, and all 12 animals during the second period. Fig. 2). The probe boxes were therefore intended to be ambiguous to the dogs in terms of their contingency outcome. 2.8. Testing 3.1. Rewarding event During the experiment each dog was tested with a sequence of 15 trials incorporating one exposure to each of the three probes (Near Rewarded (NR), Middle (M), Near Unrewarded (NU)) interspersed between six rewarded (+) and six unrewarded (−) trials, e.g. + − − + M + − + −NR− + +−NU. In order to increase the number of probe exposures, but attempting to minimise the chance of the dogs learning that probe boxes were unrewarded, dogs were tested three times (on three separate days), achieving three exposures to each of the three probes (i.e. nine probe exposures in total). The position of the probes within the trial sequence was balanced for order. Testing involved presenting the dog with ‘probe’ boxes (always unrewarded) in one of three grey shades intermediate (e.g. ‘185RGB’, ‘130RGB’ and ‘75RGB’) between the two ‘reference’ shades (e.g. ‘240RGB’ and ‘20RGB’) (see We looked to see if there was any change in ‘motivation’ (i.e. latency to run to the maze entrance) for dogs receiving the post-consumption treatment, either across the three test days, or across the three separate searches that took place on each test day. Data were analysed using a repeated measures GLM with Day (1–3) and Search trial (1–3) as within-subject factors. No significant effects were observed (all P > 0.1). Average latency for all dogs across Day and Search trial was 3.5 s. 2.9. Data analysis All data were tested to see if they satisfied requirements for parametric testing. For example, Shapiro–Wilk test of normality on the residuals. If they did not satisfy those requirements, data were log-transformed where indicated in the text and analysed using repeated measures General Linear Models (GLMs) or paired t-tests. The specific statistical models used are described in Section 3. The statistical package used was SPSS v.16. 3. Results 3.2. Cognitive bias training The dogs took from between 30 and 200 trials to reach criterion in the initial stage of training, with an average of 93.2 trials. Using a repeated measures GLM with ‘Stage of training’ (initial, refresher training 1, refresher training 2) O. Burman et al. / Applied Animal Behaviour Science 132 (2011) 160–168 165 Latency (s) to reach the goalbox (unadjusted) 25 20 15 'Postconsumption' 'Neutral' 10 5 0 Rewarded Near Rewarded Middle Near Unrewarded Unrewarded Goalbox Fig. 3. A graph showing the latency (s) for the dogs to reach the Rewarded and Unrewarded goal boxes and the three intermediate goal boxes (Near Rewarded, Middle, and Near Unrewarded) when tested after experiencing the rewarding event (“post-consumption”) and without experiencing the rewarding event (“Neutral”). Data are means ± standard error of the mean (unadjusted raw data), comprising, for all 12 dogs, the average latency for each goal box. as a within-subjects factor, we observed a clear improvement, in terms of the number of trials required to achieve criterion, from the initial stage of training to the first and second ‘refresher’ stages (F2,22 = 14.5, P < 0.001). Fewer trials were necessary over time (means ± standard error of the mean – initial training: 93.2 ± 15.2; first refresher: 43.5 ± 9.2; second refresher: 11.2 ± 1.5). 3.3. Cognitive bias testing In our initial analysis we compared the dogs’ response to the reference stimuli (i.e. rewarded and unrewarded grey shades) only. The data were combined for the three test days and the reference stimuli analysed separately to the probe data, due to the difference in the number of trials contributing to the data (see Burman et al., 2008c). In total, each dog received 18 rewarded and 18 unrewarded trials during each treatment. We used a repeated measures GLM with Outcome (rewarded vs. unrewarded) and Treatment (Neutral vs. post-consumption) as within subject factors. There was a strong effect of Outcome (F1,11 = 58.2, P < 0.001) with significantly longer latencies shown for the unrewarded compared to the rewarded reference stimuli (see Fig. 3). There was no significant effect of Treatment (F1,11 = 0.18, P = 0.68) or an interaction (F1,11 = 0.01, P = 0.98). In our next analysis we investigated the response to the probe stimuli. In total, each dog received 3 exposures to each of the three different probe types during each treatment. In order to control for the possibility that differences in latencies to probe test stimuli could be due to intrinsic differences between the dogs in their running speeds, we adjusted the data using the following equation, using values from the test period only: Adjusted score = mean latency to probe − mean latency to rewarded stimulus mean latency to unrewarded stimulus − mean latency to rewarded stimulus The adjusted score expresses all probe test latencies as a proportion of the difference between each dog’s baseline mean latencies to the rewarded and unrewarded goal boxes (see Mendl et al., 2010a,b). Previously used adjustment techniques (e.g. Burman et al., 2008c, 2009) based on the inclusion of covariates in the statistical model were considered unsuitable due to the repeated measures design of the current study. We used log-transformed data in a repeated measures GLM with Treatment (Neutral vs. postconsumption) and Probe (Nearest Rewarded (NR), Middle (M), Nearest Unrewarded (NU)) as within subject factors. We observed a significant effect of Probe (F2,10 = 25.9, P < 0.001), a trend towards a Treatment × Probe interaction (F2,10 = 3.6, P = 0.065) but no overall Treatment effect (F1,11 = 0.2, P = 0.7). Further investigation of the interaction using post-hoc analysis was carried out because, even with a small sample size, the interaction approached statistical significance and showed a large effect size (Partial eta Squared = 0.421). This further analysis (paired t-tests) revealed that, whilst there was no significant difference between treatments for either the NR or NU probe, there was a statistically significant difference between the treatments for the M probe (t11 = −2.66, P = 0.022), with dogs taking longer to approach the middle probe in the post-consumption treatment compared to the Neutral treatment (see Fig. 3). 4. Discussion 4.1. Rewarding event Our initial question was to ask if the dogs appeared equally ‘motivated’ (in terms of their latency to approach the entrance of the maze arena) for the rewarding event, both across the three test days, and across the three separate searches that took place on each test day. Our results 166 O. Burman et al. / Applied Animal Behaviour Science 132 (2011) 160–168 indicated that the dogs remained consistent in their motivation for the rewarding event, suggesting that they neither got more nor less enthusiastic about the task during the three searches on the same day, or over the three test days. That the dogs continued to ‘work’ for access to, and during, the task suggests that they found it rewarding (e.g. Rolls, 2005). 4.2. Cognitive bias training There was a wide range (30–200 trials) between dogs in the number of trials required to reach criterion in the visual discrimination task during initial training. Only 2 out of the 12 dogs took less than 40 trials, which compares unfavourably to the number of trials dogs required to reach criterion during a spatially based cognitive bias task (a range of 21–61 trials (Mendl et al., 2010a)). Indeed, in the current study 5 out of the 12 dogs required over 100 trials to reach criterion, suggesting that this visual task, at least in the format used in the current study, was more difficult for the dogs to learn (mean no. trials to reach criterion: 29.4 (spatial task); 93.2 (visual task)). Our dogs appeared to find it particularly difficult to learn not to approach the unrewarded goal box, and this could be explained by the lack of any negative consequences and/or the drive to gather information about potential future food sources – as appears to underpin the phenomenon of contrafreeloading (e.g. Inglis et al., 1997). The dogs took significantly fewer trials to reach criterion at each stage of training, suggesting that, even once they had reached criterion, the dogs’ performance at the discrimination task was still capable of being improved with further experience/training. 4.3. Cognitive bias task testing That the dogs ran significantly faster to the rewarded compared to the unrewarded reference stimuli confirms that the dogs were able to discriminate between the reference stimuli during testing, as we would expect given that they achieved criterion during training. The absence of any significant treatment effect in response to the reference stimuli provides some indirect evidence that there were no overt differences between the treatments in terms of motivation, as demonstrated in similar studies (see Burman et al., 2008c, 2009). For instance, the dogs were unlikely to have been either fully satiated or excessively tired following experience of the rewarding event, because, if they had been, then we would have expected the dogs to run slower for the rewarded stimulus (if fully satiated) and/or slower for the unrewarded stimulus (if excessively tired) following completion of the post-consumption rewarding event. As in previous cognitive bias studies that have used unrewarded vs. rewarded outcomes for the reference stimuli, we observed a general ‘optimistic’ tendency for the subjects to approach the probes quickly–presumably because there was little perceived cost attributed to receiving a ‘negative’ (i.e. unrewarded) outcome (see Mendl et al., 2009). When considering the response of the dogs to the ambiguous probe stimuli, there was a suggestion that, when tested after experiencing the rewarding event, the dogs had a more negative (less positive) judgement of ambiguity, at least for the middle probe, than when tested without having experienced the rewarding event. This result was in the reverse direction to our initial prediction, that dogs would show a more positive judgement of the ambiguous probe stimuli during post-consumption. When considering this unexpected result we should first consider possible non-affective explanations such as motivation, learning, and/or activity (see Mendl et al., 2009) – particularly given that we used food as a reward for both the cognitive bias task and the treatment. Even though there was no evidence of any treatment difference in the dogs’ response to the reference stimuli (see earlier), we cannot rule out the influence of non-affective explanations entirely. For instance, one possibility is that dogs tested post-consumption were satiated, but not completely so. As such, although they showed no decline in performance when a reward was guaranteed/certain (i.e. when the rewarded stimulus was presented), they responded more slowly when there was some uncertainty (i.e. when the ambiguous stimuli were presented). This could relate to the relationship between motivational state and the evaluation of reward certainty with, for example, a lower food motivation associated with being risk-averse (see Caraco et al., 1980). Further studies in which different types of rewarding event are compared, that are either non-food based and/or less physically demanding (e.g. social contact) might shed light on this issue. We should also consider the possibility that the rewarding event was not actually rewarding for the dogs and so did not elicit a positive affective state. Although this cannot be ruled out for certain, the finding (see earlier) that the dogs remained consistent in the speed with which they ran towards the maze for the ‘search and forage task’ over the three trials for each session suggests that the task was rewarding, with no apparent decline in interest over the three trials. On the other hand, this same finding might lead us to suggest that, by terminating their experience of the rewarding event, we could have unintentionally elicited a negative affective state (e.g. Rolls, 2005) in the dogs tested post-consumption. Affect-based explanations for why the ‘Neutral’ treatment appeared to have a more positive effect on judgement include the possibility that, when tested without experiencing the rewarding event, rather than being affectively ‘Neutral’ as we had intended, it could be that we had inadvertently induced the anticipatory phase of reward acquisition (see Section 1) and a positive high arousal affective state of ‘excitement’ (e.g. Mendl et al., 2010a,b). Many different individual aspects of the methodology (e.g. the removal of each dog from their social group for testing, putting the dog on a lead, the test room environment, etc.) could have acted as clear signals cueing anticipatory behaviour. The dogs could therefore have learned simple associations between the occurrence of signals and the subsequent delivery of valued outcomes (e.g. Bos et al., 2003; van der Harst et al., 2005; Moe et al., 2006). Performance in the cognitive bias task itself, the experience of the rewarding event and/or human contact could all, either individually or cumulatively, have acted as potential valued outcomes for the dogs in this study. So, rather than being affectively ‘Neutral’, the dogs tested without experi- O. Burman et al. / Applied Animal Behaviour Science 132 (2011) 160–168 encing the rewarding event could have been in a positive affective state associated with anticipation, resulting in a more positive judgement of ambiguity compared to when tested post-consumption. In other words, the ‘positiveness’ of anticipating the rewarding event was greater than the ‘positiveness’ of just having experienced it – and this explanation may be particularly appropriate for the dogs that took part in this study because they were young and, compared to dogs kept as pets, were housed in a relatively homogeneous environment. 5. Conclusion Our aim was to investigate whether or not a positive affective state could be elicited in dogs during the post-consumption phase of reward acquisition as measured using a cognitive bias test. However, our results unexpectedly suggested that the dogs had a negative (or less positive) judgement of ambiguity when tested post-consumption. Such an apparent discrepancy between predicted and observed outcomes in cognitive bias studies has occurred previously, and appears to be more likely to occur when a short-term induction of affect is stopped just prior to the cognitive bias test (e.g. Doyle et al., 2010). This suggests that additional research is necessary, not only in order to elucidate further the relationship between affective state induction and cognitive bias, but also to investigate the elicitation of positive affective states in dogs, particularly in relation to the timing of reward acquisition and consumption and, more generally, the effectiveness of different methods used to induce putative changes in affective state. Whether the findings of this study generalise to the wider pet dog population also remains to be determined. Acknowledgements Our thanks to the Swedish Research Council for Environment, Agricultural Sciences and Spatial Planning for funding OB and EP during this study, and to the BBSRC (Biotechnology and Biological Sciences Research Council) for funding OB and EP outside of this period. We also thank two anonymous referees for their constructive comments. References Bateson, M., Matheson, S.M., 2007. Performance on a categorisation task suggests that removal of environmental enrichment induces “pessimism” in captive European starlings (Sturnus vulgaris). Anim. Welfare 16, 33–36. Boissy, A., Manteuffel, G., Jensen, M.B., Moe, R.O., Spruijt, B., Keeling, L.J., Winckler, C., Forkman, B., Dimitrov, I., Langbein, J., Bakken, M., Veissier, I., Aubert, A., 2007. Assessment of positive emotions in animals to improve their welfare. Physiol. Behav. 92, 375–397. Bos, R.v.d., Meijer, M.K., van Renselaar, J.P., van der Harst, J.E., Spruijt, B.M., 2003. Anticipation is differently expressed in rats (Rattus norvegicus) and domestic cats (Felis silvestris catus) in the same Pavlovian conditioning paradigm. Behav. Brain Res. 141, 83–89. Brilot, B.O., Asher, L., Bateson, M., 2010. Stereotyping starlings are more “pessimistic”. Anim. Cogn. 13 (5), 721–731. Burgdorf, J., Panksepp, J., 2006. The neurobiology of positive emotions. Neurosci. Biobehav. Rev. 30, 173–187. Burman, O., Owen, D., Abouismail, U., Mendl, M., 2008a. Removing individual rats affects indicators of welfare in the remaining group members. Physiol. Behav. 93, 89–96. 167 Burman, O., Parker, R., Paul, E., Mendl, M., 2008b. Sensitivity to reward loss as an indicator of animal affect and welfare. Biol. Lett. 4, 330–333. Burman, O., Parker, R., Paul, E., Mendl, M., 2008c. A spatial judgement task to determine background emotional state in laboratory rats (Rattus norvegicus). Anim. Behav. 76, 801–809. Burman, O.H., Parker, R.M., Paul, E.S., Mendl, M.T., 2009. Anxiety-induced cognitive bias in non-human animals. Physiol. Behav. 98, 345–350. Caraco, T., Martindale, S., Whittam, T.S., 1980. An empirical demonstration of risk-sensitive foraging preferences. Anim. Behav. 28, 820–830. Dawkins, M.S., 2000. Animal minds and animal emotions. Amer. Zool. 40, 883–888. Dawkins, M.S., 1990. From an animals point of view – motivation, fitness, and animal-welfare. Behav. Brain Sci. 13, 1–61. Dawkins, M.S., Donnelly, C.A., Jones, T.A., 2004. Chicken welfare is influenced more by housing conditions than by stocking density. Nature 427, 342–344. de Jonge, F.H., Tilly, S., Baars, A.M., Spruijt, B.M., 2008. On the rewarding nature of appetitive feeding behaviour in pigs (Sus scrofa): do domesticated pigs contrafreeload? Appl. Anim. Behav. Sci. 114, 359–372. Doyle, R.E., Fisher, A.D., Hinch, G.N., Boissy, A., Lee, C., 2010. Release from restraint generates a positive judgement bias in sheep. Appl. Anim. Behav. Sci. 122, 28–34. Farm Animal Welfare Council (FAWC), 2009. Farm Animal Welfare in Great Britain: Past, Present and Future. Farm Animal Welfare Council (FAWC), 1992. FAWC updates the five freedoms. Vet. Rec. 17, 357. Garner, J.P., Dufour, B., Gregg, L.E., Weisker, S.M., Mench, J.A., 2004. Social and husbandry factors affecting the prevalence and severity of barbering (“whisker trimming”) by laboratory mice. Appl. Anim. Behav. Sci. 89, 263–282. Gevers, T., Smeulders, W.M., 1999. Color-based object recognition. Pattern Recogn. 32, 453–464. Harding, E.J., Paul, E.S., Mendl, M., 2004. Animal behavior – cognitive bias and affective state. Nature 427, 312–1312. Inglis, I.R., Forkman, B., Lazarus, J., 1997. Free food or earned food? A review and fuzzy model of contrafreeloading. Anim. Behav. 53, 1171–1191. Kaminski, J., Fischer, J., Call, J., 2008. Prospective object search in dogs: mixed evidence for knowledge of what and where. Anim. Cogn. 11, 367–371. Mason, G.J., 1991. Stereotypies: a critical review. Anim. Behav. 41, 1015–1037. Matheson, S.M., Asher, L., Bateson, M., 2007. Larger, enriched cages are associated with “optimistic” response biases in captive European starlings (Sturnus vulgaris). Appl. Anim. Behav. Sci. 109, 374–383. McAdie, T.M., Keeling, L.J., 2000. Effect of manipulating feathers of laying hens on the incidence of feather pecking and cannibalism. Appl. Anim. Behav. Sci. 68, 215–229. McGowan, R.T.S., Rehn, T., Norling, Y., Keeling, L.J., 2010. Positive affect and learning: exploring the “Eureka Effect” in dogs. In: Proceedings of the 44th Congress of the International Society for Applied Ethology (ISAE). Mendl, M., Burman, O., Parker, R., Paul, E., 2009. Cognitive bias as an indicator of animal emotion and welfare: emerging evidence and underlying mechanisms. Appl. Anim. Behav. Sci. 118, 161–181. Mendl, M., Brooks, J., Basse, C., Burman, O., Paul, E., Blackwell, E., Casey, R., 2010a. Dogs showing separation-related behaviour exhibit a “pessimistic” cognitive bias. Curr. Biol. 20, R839–R840. Mendl, M., Burman, O.H.P., Paul, E., 2010b. An integrative and functional framework for the study of animal emotion and mood. Proc. R. Soc. B: Biol. Sci. 277, 2895–2904. Moe, R.O., Bakken, M., Kittilsen, S., Kingsley-Smith, H., Spruijt, B.M., 2006. A note on reward-related behaviour and emotional expressions in farmed silver foxes (Vulpes vulpes)—basis for a novel tool to study animal welfare. Appl. Anim. Behav. Sci. 101, 362–368. Moinard, C., Mendl, M., Nicol, C.J., Green, L.E., 2003. A case control study of on-farm risk factors for tail biting in pigs. Appl. Anim. Behav. Sci. 81, 333–355. Murray, J.K., Browne, W.J., Roberts, M.A., Whitmarsh, A., Gruffydd-Jones, T.J., 2010. Number and ownership profiles of cats and dogs in the UK. Vet. Rec. 166, 163–168. Paul, E.S., Harding, E.J., Mendl, M., 2005. Measuring emotional processes in animals: the utility of a cognitive approach. Neurosci. Biobehav. Rev. 29, 469–491. Pretterer, G., Bubna-Littitz, H., Windischbauer, G., Gabler, C., Griebel, U., 2004. Brightness discrimination in the dog. J. Vis. 4, 241–249. Reefmann, N., Bütikofer Kaszàs, F., Wechsler, B., Gygax, L., 2009. Ear and tail postures as indicators of emotional valence in sheep. Appl. Anim. Behav. Sci. 118, 199–207. 168 O. Burman et al. / Applied Animal Behaviour Science 132 (2011) 160–168 Rehn, T., Keeling, L.J., 2010. Investigating greeting behaviour in dogs reunited with a familiar person. In: Proceedings of the 44th Congress of the International Society for Applied Ethology (ISAE). Rolls, E.T., 2005. Emotion Explained. Oxford University Press, Oxford. Rooney, N.J., Bradshaw, J.W.S., Robinson, I.H., 2001. Do dogs respond to play signals given by humans? Anim. Behav. 61, 715–722. Rooney, N.J., Gaines, S.A., Bradshaw, J.W.S., 2007a. Behavioural and glucocorticoid responses of dogs (Canis familiaris) to kennelling: Investigating mitigation of stress by prior habituation. Physiol. Behav. 92, 847–854. Rooney, N.J., Gaines, S.A., Bradshaw, J.W.S., Penman, S., 2007b. Validation of a method for assessing the ability of trainee specialist search dogs. Appl. Anim. Behav. Sci. 103, 90–104. Schipper, L.L., Vinke, C.M., Schilder, M.B.H., Spruijt, B.M., 2008. The effect of feeding enrichment toys on the behaviour of kennelled dogs (Canis familiaris). Appl. Anim. Behav. Sci. 114, 182–195. van der Harst, J.E., Baars, A., Spruijt, B.M., 2005. Announced rewards counteract the impairment of anticipatory behaviour in socially stressed rats. Behav. Brain Res. 161, 183–189. Yeates, J.W., Main, D.C., 2008. Assessment of positive welfare: a review. Vet. J. 175, 293–300.