Judging a Dog by Its Cover: Morphology but Not Training Influences Visitor Behavior toward Kenneled Dogs at Animal Shelters

Anthrozoös A multidisciplinary journal of the interactions of people and animals ISSN: 0892-7936 (Print) 1753-0377 (Online) Journal homepage: http://www.tandfonline.com/loi/rfan20 Judging a Dog by Its Cover: Morphology but Not Training Influences Visitor Behavior toward Kenneled Dogs at Animal Shelters Alexandra Protopopova & Clive D. L. Wynne To cite this article: Alexandra Protopopova & Clive D. L. Wynne (2016) Judging a Dog by Its Cover: Morphology but Not Training Influences Visitor Behavior toward Kenneled Dogs at Animal Shelters, Anthrozoös, 29:3, 469-487, DOI: 10.1080/08927936.2016.1181381 To link to this article: http://dx.doi.org/10.1080/08927936.2016.1181381 Published online: 17 Aug 2016. Submit your article to this journal View related articles View Crossmark data Full Terms & Conditions of access and use can be found at http://www.tandfonline.com/action/journalinformation?journalcode=rfan20 Download by: [Alexandra Protopopova] Date: 17 August 2016, At: 07:09

ANTHROZOÖS VOLUME 29, ISSUE 3 REPRINTS AVAILABLE PHOTOCOPYING ISAZ 2016 PP. 469 487 DIRECTLY FROM PERMITTED PRINTED IN THE UK THE PUBLISHERS BY LICENSE ONLY Judging a Dog by Its Cover: Morphology but Not Training Influences Visitor Behavior toward Kenneled Dogs at Animal Shelters Alexandra Protopopova * and Clive D. L. Wynne * Department of Animal and Food Sciences, Texas Tech University, USA Department of Psychology, Arizona State University, USA Address for correspondence: Alexandra Protopopova, Department of Animal and Food Sciences, Texas Tech University, PO Box 42141, Lubbock, TX 79409, USA. E-mail: a.protopopova@ttu.edu ABSTRACT Currently, visitor behavior in companion animal shelters is not adequately explored. A sequence of experiments investigated how visitors interacted with kenneled dogs at an animal shelter and whether training dogs to not engage in undesirable behavior in their kennels would evoke more interest from shelter visitors. In a set of two experiments, two sides of an animal shelter were differentially entered into training phases in a multiple baseline design. During the training phase, dogs were trained to not engage in undesirable in-kennel behavior (i.e., to not be in the back of the kennel, to not face backwards, to not lean on kennel walls, and to not bark) through pairing of the visual presentation of the experimenters (experiment 1) and shelter visitors (experiment 2) with treats. Across both experiments, visitors attended to approximately 35% of kenneled dogs and only spent an average of 15 s looking at individual dogs. We found that whereas training was effective in decreasing undesirable behavior in dogs (experiment 1: U = 4.83, p < 0.0001; experiment 2: U = 4.01, p = 0.0001), only morphology influenced visitor behavior. In experiment 1, morphologically preferred dogs (i.e., puppies, longcoated dogs, small dogs, and certain breeds) had a 1.3 times higher frequency of visits to their kennel (F (1, 248) = 5.93, p < 0.05), and in experiment 2, these dogs had a 9 times higher frequency of being taken out of their kennel for further inspection (F (1, 69) = 4.66, p < 0.05), compared to other dogs. One reason for a lack of effect of training may be the relatively small number of visitors observed (n = 115 across both experiments). An alternative reason may be that shelter visitors pay more attention to the morphology rather than the lack of undesirable behavior of kenneled dogs. Keywords: adoption, dog, dog training, overpopulation 469 Anthrozoös DOI: 10.1080/08927936.2016.1181381

Judging a Dog by Its Cover: Morphology but Not Training Influences Visitor Behavior 470 Anthrozoös Fewer dogs live in animal shelters than in previous years; however, the American Society for the Prevention of Cruelty to Animals (2016) estimates that approximately 3.9 million dogs enter shelters every year and 31% are subsequently euthanized. Morphology has been found to predict adoption rates in shelter dogs; puppies, smaller dogs, and certain breed types are consistently found to have higher rates of adoption (see discussion by Brown, Davidson and Zuefle 2013). Other variables such as how the dog entered the shelter, color, and coat length have been more variable in predicting higher adoption (e.g., Wells and Hepper 1992; Lepper, Kass and Hart 2002; Clevenger and Kass 2003; Normando et al. 2006; Protopopova et al. 2012; Brown, Davidson and Zuefle 2013; Siettou, Fraser and Fraser 2014). Unfortunately, previous attempts to improve adoption rates through training or enrichment in the kennel have not been successful (Protopopova et al. 2012; Herron, Kirby-Madden and Lord 2014; but see Luescher and Medlock 2009). One potential reason for the limited effectiveness of the evaluated training programs is that these programs were based on assumptions about what adopters are looking for when selecting their new dogs. However, these assumptions may not be accurate. For example, the evaluated training programs consisted of focusing on behaviors such as sitting or lying down at the front of the kennel and not barking or jumping (Luescher and Medlock 2009; Herron, Kirby-Madden and Lord 2014) and gazing (Protopopova et al. 2012; Herron, Kirby-Madden and Lord 2014). These behaviors were targeted because previous data showed that people preferred photos of dogs depicted as not barking and being at the front of the kennel (Wells and Hepper 1992). Furthermore, professionals in the animal sheltering industry often assume that jumping is an undesirable behavior and sitting or lying down in the kennel is desirable to adopters. However, contrary to previous research and opinion, a recent observational study found that only a few behaviors in the kennel predicted a longer length of stay at the shelter (Protopopova et al. 2014). These behaviors included back and forth motion in the kennel, facing backward, and leaning or rubbing on the kennel walls. Furthermore, Weiss et al. (2012) found that new owners often reported that dogs jumped on them prior to their adoption decision, suggesting that jumping is not necessarily an undesirable behavior as previously assumed. It is possible that previous research on interventions focused on training of behaviors that were potentially irrelevant to adopters and thus any effects of training on adoption were substantially weakened. To overcome this limitation, the present study set out to assess whether focusing training on behaviors that have been previously found to predict length of stay, through observational research, may be beneficial in achieving adoption for dogs housed at shelters. We recently demonstrated that simple pairing of a person followed by food reduced these previously found undesirable in-kennel behaviors in dogs at the shelter (Protopopova and Wynne 2015). In fact, when a behavioral intervention focused on behaviors which were found to be predictive of adoption in out-of-kennel interactions with adopters, the likelihood of adoption was subsequently increased (Protopopova, Brandifino and Wynne 2016). However, it remains to be seen whether such an approach would prove useful for training appropriate in-kennel behavior and subsequently increasing adopter interest in the dogs. Another potential reason for the limited success of previous programs is that the efficacy of dog training programs has been assessed using adoption rate or length of stay as the dependent variable. However, there may be several levels of human selection that the dog has to go through before adoption. A shelter visitor may view all of the dogs available at the shelter and indicate an interest to take a dog out of the kennel for further inspection, as a first level of selection leading to potential adoption. The second selection level may then occur outside of

Protopopova and Wynne the kennel where the visitor decides whether or not to adopt. At this level of selection, dogs have approximately a one in three chance of being adopted (Protopopova and Wynne 2014). However, not much research has been conducted on the primary level of selection. It is possible that in-kennel dog training may only influence this first level of selection, but not the second, and thus, not substantially influence adoption rates. It is possible that previous research did not find effects of in-kennel dog training because the dependent variable, adoption rate or length of stay, was too far removed from the variable that was actually influenced through training (i.e., taking the dog out for further inspection). Therefore, the current study assessed visitor behavior toward the kenneled dogs, and not adoption rate, as the dependent variable. The behavior of visitors in animal shelters has not been adequately studied. Currently, only one observational study explored how visitors behave at the animal shelter. Wells and Hepper (2001) found that visitors stopped to look at approximately a third of the dogs and spent an average of 70 s looking at individual dogs once stopped. Furthermore, out of the visitors who looked at the dogs, 31% further interacted with a dog for an average of 20 s through the kennel. An intriguing finding was that single visitors stopped to look at more dogs and stayed longer in the shelter than visitors who came in groups. Also, only three visitors (out of 76) adopted a dog during their visit. However, we currently do not know how the behavior and/or morphology of kenneled dogs influences the time visitors spend looking at them and the various interactions that the visitors have with them. The first aim of the present study was to assess whether training dogs to reduce their undesirable behavior inside the kennel, using a validated response-independent pairing procedure, has a positive impact on shelter visitor behavior. The second aim of the study was to provide additional general information on how shelter visitors interact with kenneled dogs and what morphological variables, if any, influence visitor behavior. Experiment 1 We aimed to experimentally investigate whether training dogs at a shelter to not exhibit undesirable behavior would evoke more interest from potential adopters. We predicted that training dogs to not engage in undesirable behavior in the presence of visitors would increase the amount of time visitors spent looking at and interacting with them, and increase the number of requests to take the dogs out of the kennel. An additional aim of the study was to characterize the behavior of visitors at an animal shelter. Methods Animals and Housing: The study used dogs that were housed at Alachua County Animal Services (ACAS), an open-admission municipal animal shelter. The dogs were housed mostly singly in cement and chain-link kennels, with half of the kennel outdoors and the other half indoors. At the time of data collection, the approximate average length of stay prior to adoption was 3 weeks. The visitors were only allowed to see the dogs from the outside portion of the kennel. The shelter was split into two sides: A (kennels 1 through 35) and B (kennels 36 through 70). The two sides of the shelter were identical except side A faced the entrance of the shelter, while side B faced the back of the shelter. The dogs had visual access to the dogs in the opposite kennels. The staff unsystematically placed dogs into any available kennels as the dogs entered the shelter, resulting in a random assignment of dogs into each side. Table 1 lists the primary breed, age, coat length, and size of the dogs. Detailed information about the housing of the dogs and husbandry procedures may be accessed in Protopopova et al. (2014). The volunteers, staff, and 471 Anthrozoös

Judging a Dog by Its Cover: Morphology but Not Training Influences Visitor Behavior Table 1. The number of dogs by reported primary breed, age, coat length, and size are listed by the side of the shelter in each phase of experiment 1. The asterisk denotes the levels of each variable that comprised the morphologically preferred category. The numbers of dogs in the morphologically preferred category by side of the shelter in each phase are listed below. Phase 1 Phase 2 Phase 3 Side A Side B Side A Side B Side A Side B Baseline Baseline Trained Baseline Baseline Trained Primary Breed American Bulldog 1 2 3 3 2 3 American Pit Bull Terrier 20 22 16 16 9 19 Australian Cattle Dog* 0 0 0 0 1 1 Australian Shepherd* 0 0 1 0 1 1 Basset Hound 0 1 0 0 0 1 Beagle 0 0 1 0 2 1 Bloodhound 0 0 0 1 0 0 Blackmouth Cur 2 1 1 0 0 0 Black and Tan Coonhound 1 0 0 0 0 0 Border Collie* 2 0 0 0 0 0 Boxer 2 0 1 0 1 2 Bull Terrier 0 0 0 0 0 1 Carolina Dog 0 0 0 0 0 1 Catahoula Leopard Dog 0 0 0 0 2 1 Chihuahua* 3 3 0 2 0 0 Chow Chow 0 0 0 1 1 1 Dachshund* 2 0 1 1 0 0 Doberman Pinscher 0 0 0 0 0 1 English Bulldog 1 1 2 0 2 1 German Shepherd Dog* 2 1 0 0 1 2 Great Dane 0 0 1 0 1 0 Harrier 0 1 0 0 0 0 Hound (unspecified) 0 0 0 0 2 0 Jack Russel Terrier* 0 1 1 0 0 0 Labrador Retriever 5 8 2 6 7 9 Plott Hound 0 2 1 2 2 0 Pointer 0 1 0 1 2 0 Pomeranian* 0 0 0 0 1 0 Rat Terrier* 0 1 1 1 0 0 Redbone Coonhound 0 0 0 1 0 0 Rottweiler 1 0 1 0 1 0 Treeing Walker Coonhound 0 0 1 0 0 0 Shetland Sheepdog* 0 0 0 0 1 0 Siberian Husky 0 1 0 1 1 0 Vizsla 1 0 1 0 1 1 Weimaraner 0 1 0 1 0 1 Age 4 month* 2 3 1 2 1 4 > 4 months 41 44 34 35 40 43 Average Age (months) (SD) 19.0 (16.1) 18.2 (14.9) 20.8 (17.1) 20.3 (18.6) 24.7 (21.2) 16.4 (13.9) 472 Anthrozoös Coat Length Long* 2 1 2 1 1 1 Short 41 46 33 36 40 46 Size Small* 7 8 5 4 1 0 Medium/Large 36 39 30 33 40 47 Morphology Type Preferred 11 10 6 6 5 9 Not Preferred 32 37 29 31 36 38

Protopopova and Wynne Table 2. Ethogram of visitor behavior while looking at kenneled shelter dogs. Behavior Frequency Duration Reaching Bending Down/Kneeling Speaking Reading Kennel Card Asking Questions Taking out Operational Definition Total number of visits to the dog s kennel. A visit encompassed a continued gaze directed at the dog s kennel. Gaze diversions of > 2 s ended a visit, and a new visit was counted if gaze was resumed and kept for > 2 s thereafter. Total seconds spent in front of the dog s kennel across all visits. Timer began when the visitor s gaze was directed at the dog s kennel for > 2 s and ended when gaze was diverted. Visitor s arm is visibly extended towards the kennel (> approximately 0.2 m away from body). Visitor must be in duration for this variable to be counted. Visitor s torso is visibly bent towards the kennel and/or visitor is in a crouching or kneeling position. Visitor is vocally addressing the dog with words and/or noises (e.g., whistling, clicking the tongue, lip smacking, etc.). Visitor must be in duration for this variable to be counted. Visitor s gaze is directed at the kennel card located on the kennel door. Visitor must be in duration for this variable to be counted. The visitor asks the experimenter a question about the dog. The visitor requests to take the dog out of the kennel in order to interact further with the dog in an off-leash area. shelter visitors were not aware that different sides of the shelter may have been receiving different experimental treatment. Data on the morphology of the dogs were taken from the dog s kennel card and grouped into two categories: morphologically preferred and not preferred. As described in Protopopova et al. (2014), morphologically preferred dogs consisted of puppies ( 4 months), small dogs, long coated dogs, and preferred breed types (herders, lap dog breeds, and ratters). The remaining dogs were categorized as morphologically not preferred. For an explanation of breed types, see Protopopova et al. (2012, 2014). Visitor Behavior Coding: An ethogram was developed based on preliminary observations of shelter visitor behavior as directed toward dogs in their kennels. The ethogram included the duration of time spent in front of each kennel, the presence of affiliative behavior toward each dog (attempting to pet through the cage; kneeling down, speaking with the dog), the frequency of the adopter gathering information about the dog (i.e., reading the kennel card and asking questions about the dog), and noting which, if any, dogs were taken out of their kennel for further inspection. This full list of all behaviors and their operational definitions are available in Table 2. All variables except frequency and duration were coded as presence or absence of the behavior directed toward that dog. For example, even if the visitor spoke to the dog five times across three visits to the dog s kennel, the speaking variable was recorded as a 1 during that session. Dog Training: Experiment 1 used a variation of a multiple baseline across sides of the shelter design. In the first 28 calendar days, both sides of the shelter received the baseline condition (phase 1). For the next 28 calendar days, side B continued to receive the baseline condition, while side A received the experimental condition (phase 2). For the next 28 calendar days, side B entered the experimental condition and side A was reversed back to baseline (phase 3). The sides of the shelter were thus treated as subjects, resulting in the possibility that a single dog could have experienced both the experimental and the control conditions if it stayed 473 Anthrozoös

Judging a Dog by Its Cover: Morphology but Not Training Influences Visitor Behavior Table 3. The operational definitions of all behaviors that comprised the measure of undesirable behavior of kenneled dogs. Behavior Back of Kennel or Out of Sight Facing Backward Contact with Kennel Walls Barking Operational Definition Dog is located behind the midpoint of the outdoor portion of the kennel or in the inside portion of the kennel. No eyes are visible to the camera. Physical contact with the kennel wall, such as leaning or jumping on the wall (excludes contact with the cage door). Dog is emitting a bark, howl, or yelp. for longer than 28 days at the animal shelter and/or if it was moved by the shelter staff from one side of the shelter to another, for reasons extraneous to the study. Each morning that the shelter was open (Tuesday through Saturday), an experimenter carried out the following procedures, dependent on the condition. During the experimental condition, dogs received 10 daily trials of pairing a 2 s sound of a hand bell, which the trainer held, with the delivery of food. During the first trial, the experimenter rang the bell while simultaneously video recording the dog with a small inconspicuous hand-held video camera (Kodak PlaySport Zx3; Kodak Company, Rochester, NY, USA). These videos were coded by noting the presence or absence of undesirable behavior for each dog in their kennel during the 2 s period of time. If the dog was in the experimental condition, a second experimenter threw a treat (Canine Carry Outs Chewy Treats for Dogs) as soon as the 2 s bell had rung with no regard to the behavior of the dog. For the remaining nine trials of pairing of the bell with treats, the dogs behavior was not videotaped. To promote generalization of the training across people, trainers varied daily for a total of six trainers. This simple pairing of the experimenter with a bell and treats has previously been shown to reduce undesirable in-kennel behavior of shelter dogs (Protopopova and Wynne 2015). Undesirable in-kennel behavior was an aggregate measure of the behaviors that were previously found to predict a longer length of stay at the shelter; these behaviors included staying at the back of the kennel, facing backwards, and rubbing or leaning on the kennel walls (Protopopova et al. 2014). Barking, even if not a correlate of length of stay, was nevertheless included due to concerns for the wellbeing of both dogs and staff, as noise levels can reach damaging levels at animal shelters (Sales et al. 1997). Table 3 lists all of the behaviors included in this aggregate measure and their operational definitions. During the baseline condition, dogs heard the 2 s bell ringing, were videotaped, but the second experimenter did not throw treats; thus, in the control condition, the dogs did not receive the food pairing. Each video was coded by noting the presence or absence of undesirable behavior for each dog in their kennel during the 2 s period of time. The videos were used as probes to assess the effect of response-independent food delivery on the behavior of dogs and to calculate inter-observer agreement (IOA). 474 Anthrozoös Data Collection on Visitor Behavior: Different experimenters, who were blind to which side of the shelter was receiving which condition, observed shelter visitors, who spontaneously entered the shelter during a daily 2-hr block of time. The times at which observations were made were varied, in order to capture various kinds of visitors. When a visitor walked in and indicated an interest in looking at the adoptable dogs, an experimenter led the person to the dog area and rang a bell for 2 s before the visitor reached every dog kennel. If asked, the experimenter

Protopopova and Wynne told the visitor that the bell alerted the dogs to his or her presence. A 2-hr block of time was chosen to ensure that the dogs behavior would not undergo extinction, as the experimenter never paired food with the bell when following visitors. If the same visitor came to the shelter more than once, only the data collected from their first visit was included. An additional observer, also blind to the condition of the dogs, followed behind the visitor and collected data on visitor behavior (with pencil and paper using the ethogram; Table 2). For 36.5% of observations, an additional observer was present to collect IOA for the behavior of visitors. Data Analysis: All statistical analyses were performed in Stata/IC 13.1 (Stata LP, College Station, TX). To verify that training had an effect on behavior, a Mann-Whitney U Test was used to evaluate the effect of condition (baseline versus training) on undesirable dog behavior. Because of a lack of previously published research on visitor behavior at the shelter, we first conducted a principal component analysis (PCA) on all of the visitor behaviors (Table 2) that were collected during the study. To assess whether the training had an effect on visitor behavior, the different sides of the shelter were compared for the two components resulting from the PCA using multivariate general linear regression models, with side of shelter (A or B) as the independent variable, dog identification number and morphology as factors, and the components from the PCA as dependent variables for each phase of the study. To further explore the effect of undesirable in-kennel behavior on visitor behavior, all dogs were divided into categories of low (< 0.2) or high ( 0.2) proportions of undesirable behavior, and an additional multivariate regression model compared the effects of undesirable dog behavior (high or low) on human behavior. To assess whether the morphology of the dogs influenced adopter behavior, the data were aggregated across phases and entered into a multivariate general linear regression model, with morphological type (preferred or not preferred) and dog identification number as factors, and the behavior of visitors (two components from the PCA) as dependent variables. IOA was assessed for both behavior of the dogs as well as the behavior of the visitors. A second experimenter coded videos of the dogs in-kennel behavior for 33.3% of the days, and the IOA for the dogs behavior was calculated by the sum of agreements divided by the sum of agreements and disagreements and multiplying by 100. The final percentage was derived by averaging across days. An agreement was scored by both of the observers scoring a dog as behaving undesirably or not. The average IOA for the dogs behavior across days was 82.3% (range: 53.8 97.9%). For 36.5% of observations, an additional observer was present to collect IOA for the behavior of visitors. IOA for each individual behavior across visitors was calculated by the sum of agreements divided by the sum of agreements and disagreements and multiplied by 100. The final percentage agreement for each behavior was calculated by taking the average across all of the visitors. Agreement for frequency consisted of agreeing on the exact number of times the visitor stopped to look at each dog. Agreement for duration consisted of the observers scoring a maximum of 2 s difference in the duration of time that the visitor spent in front of a dog s kennel. Agreements for asking, bending down, reaching, reading, speaking, and taking out consisted of both observers scoring that these behaviors either occurred or did not occur for each dog. The average IOA for visitor behavior across all of the behaviors was 97.85% (range: 95.45 99.92%). Ethical Note: The procedures were approved by the University of Florida Institutional Review Board, the University of Florida Institutional Animal Care and Use Committee, and Alachua County Animal Services. 475 Anthrozoös

Judging a Dog by Its Cover: Morphology but Not Training Influences Visitor Behavior Figure 1. Proportion of undesirable in-kennel behavior in the experimental condition in experiment 1. Whiskers show 10th and 90th percentiles, boxes show 25th and 75th percentiles and the medians, and the dots are outliers. 476 Anthrozoös Results A total of 52 visitors were observed across all phases of the experiment: 19 visitors in phase 1, 13 visitors in phase 2, and 20 visitors in phase 3. A total of 250 dog observations were conducted, with 228 unique individual dogs enrolled in the study across the three phases: 90 dog observations (43 in side A and 47 in side B; 83 unique dogs) in phase 1, 72 dog observations (35 in side A and 37 in side B; 72 unique dogs) in phase 2, and 88 dog observations (41 in side A and 47 in side B; 79 unique dogs) in phase 3. The Shapiro-Wilk test for normality showed that the undesirable behavior of the dogs was not normally distributed (W = 0.96, p < 0.05), therefore supporting the use of a nonparametric test to compare the differences in undesirable behaviors across the two conditions. Dogs in the training condition (aggregated data across all phases) spent less time engaging in undesirable in-kennel behavior than in the baseline condition (U = 4.83, p < 0.0001). The median proportion of undesirable behavior during training was zero (interquartile range [IQR] = 0.25) and during baseline was 0.33 (IQR = 0.56; Figure 1). On average across all phases, visitors attended (by visiting the dog s kennel at least once) to 35.0% of the dogs available at the shelter (SD = 24.5%). When stopped to look at a dog, visitors spent an average of 14.83 s observing each dog (SD = 10.68 s). When counting only the kennels actually visited, people visited a dog s kennel on average 1.22 times (SD = 0.30). On average, visitors read the kennel cards of 27.0% of the dogs (SD = 23.6%). Visitors bent down to interact with on average 11.7% (SD = 16.0%), reached for 13.9% (SD = 17.2%), and spoke to 14.6% (SD = 14.6%) of all of the dogs. Visitors asked for more information about 3.3% of the dogs (SD = 4.5%) and requested to take only 1.0% of the dogs out of their kennels for further inspection (SD = 1.9%). An initial PCA was conducted to find patterns of visitor responding toward kenneled dogs. Two components with eigenvalues greater than 1 were obtained; therefore, the PCA was

Protopopova and Wynne Table 4. Principal component analysis of visitor behavior towards kenneled dogs in experiment 1. Component loadings 0.30 are in bold. Visitor Behavior Component 1 Component 2 Reaching 0.90 0.14 Bending Down/Kneeling 0.81 0.08 Speaking 0.80 0.08 Reading Kennel Card 0.66 0.10 Duration 0.33 0.75 Asking Questions 0.25 0.89 Taking Out 0.09 0.62 Frequency 0.47 0.53 repeated restricting the analysis to two components with a direct oblimin rotation. The correlation among the two components was moderate (0.46). Component loadings are shown in Table 4. Speaking, bending down/kneeling, reading the kennel card, and reaching for the dog loaded onto component 1. The frequency of taking the dog out of the kennel and asking about the dog loaded onto component 2. Duration of time spent in front of kennel and the frequency of observing the same dog in the kennel loaded more heavily on component 2. Component 1 may be thought of as exhibiting affiliative behavior toward the dog, whereas component 2 may be thought of as showing interest in assessing the dog further in an out-of-kennel environment. A multivariate general linear regression model showed that the two components did not differ in phase 1, in which both sides of the shelter were in the baseline condition (F (2, 87) = 0.68, p > 0.05; 2 = 0.02, 95% CI: 0 0.08). The mean score for component 1 on side A was 0.17 (SD = 0.73, n = 43) and on side B was 0.20 (SD = 0.89, n = 47). The mean score for component 2 on side A was 0.25 (SD = 1.02, n = 43) and on side B was 0.11 (SD = 0.81, n = 47). A multivariate general linear regression model was repeated for phase 2, in which side B continued to receive the baseline condition, while side A received the experimental condition. The sides did not differ in either component (F (2, 69) = 0.38, p > 0.05; 2 = 0.01, 95% CI: 0 0.08). The mean score for component 1 on side A was 0.30 (SD = 1.27, n = 35) and on side B was 0.10 (SD = 0.87, n = 37). The mean score for component 2 on side A was 0.05 (SD = 0.81, n = 35) and on side B was 0.14 (SD = 0.39, n = 37). Finally, the model was once again repeated for phase 3, in which side B received the experimental condition, while side A received the baseline condition. The sides, once again, did not differ in either component (F (2, 85) = 0.56, p > 0.05; 2 = 0.01, 95% CI: 0 0.08). The mean score for component 1 on side A was 0.42 (SD = 0.78, n = 41) and on side B was 0.29 (SD = 1.10, n = 47). The mean score for component 2 on side A was 0.03 (SD = 1.54, n = 41) and on side B was 0.17 (SD = 1.02, n = 47). To further examine the effect of undesirable behavior, regardless of experimental condition, on visitor behavior, all dogs were divided into categories of low (< 0.2) or high ( 0.2) proportions of undesirable behavior. The model found no effect of undesirable behavior on the two components (F (2, 247) = 1.02, p > 0.05; 2 = 0.01, 95% CI: 0 0.04). The mean score for component 1 for the low category of dogs was 0.04 (SD = 1.14, n = 113) and for the high category was 0.04 (SD = 0.87, n = 137). The mean score for component 2 for the low category of dogs was 0.10 (SD = 0.89, n = 113) and for the high category was 0.08 (SD = 1.08, n = 137). 477 Anthrozoös

Judging a Dog by Its Cover: Morphology but Not Training Influences Visitor Behavior Figure 2. The mean number of times the visitors looked at morphologically preferred dogs (i.e., puppies, long-coated dogs, small dogs, and certain breed types) compared with morphologically non-preferred dogs in experiment 1. Standard errors are shown. A reduced multivariate general linear regression model revealed that component 1 (affiliation) was predicted by the morphology of the dogs (F (1, 248) = 3.91, p < 0.05; 2 = 0.02, 95% CI: 0 0.06). The mean score for component 1 for the morphologically non-preferred dogs was 0.06 (SD = 0.07, n = 203) and for preferred dogs was 0.26 (SD = 0.96, n = 47). To further explore the effect of morphology on visitor behavior, a multivariate general model including all visitor behaviors (log-transformed) found that only the frequency of looking at the kenneled dog was predicted by its morphology (F (1, 248) = 5.93, p < 0.05; 2 = 0.02, 95% CI: 0.001 0.07). Morphologically non-preferred dogs were looked at an average of 0.44 times (SD = 0.43, n = 203) and morphologically preferred dogs were looked at an average of 0.58 times (SD = 0.45, n = 47; Figure 2). 478 Anthrozoös Experiment 2 Experiment 1 showed that the morphology of the dogs, but not training, influenced visitor behavior. One explanation for a lack of effect may have been due to the experimental procedure. Because dogs were not trained during the time when visitors were actually observing them, it is possible that the dogs learned to discriminate training sessions (when food was present) from testing sessions (when visitors but not food were present). If so, the dogs would not engage in undesirable behavior during the training sessions, but possibly engage in undesirable behavior during the sessions in which visitors observed them. Therefore, it is possible that a lack of undesirable in-kennel behavior does influence visitor behavior, but that the dogs were not, in fact, behaving in such a way when the visitors arrived. To overcome this limitation, in experiment 2 we combined the training sessions together with the sessions in which visitors observed the dogs at the shelter. Therefore, we ensured that the dogs did not behave undesirably during the visits. Methods Animals and Housing: In experiment 2, 71 dog observations were conducted, with 47 unique dogs enrolled in the study and housed at ACAS. The dogs were housed in the same

Protopopova and Wynne Table 5. The number of dogs by reported primary breed, age, coat length, and size are listed by the side of the shelter in each phase of experiment 2. The asterisk denotes the levels of each variable that comprised the morphologically preferred category. The numbers of dogs in the morphologically preferred category by side of the shelter in each phase are listed below. Phase 1 Phase 2 Side A Trained Side B Baseline Side A Baseline Side B Trained Primary Breed American Pit Bull Terrier 9 2 7 1 Australian Shepherd* 0 1 0 1 Beagle 2 0 1 0 Black and Tan Coonhound 0 1 0 0 Cairn Terrier* 0 0 0 1 Catahoula Leopard Dog 1 1 1 0 Dachshund* 1 0 1 0 English Bulldog 0 1 0 1 German Shepherd Dog* 1 0 1 0 Labrador Retriever 5 4 5 3 Mastiff (unspecified) 1 0 1 0 Rottweiler 1 0 0 1 Treeing Walker Coonhound 3 1 2 1 Siberian Husky 1 0 1 0 Smooth-Coated Collie 0 1 0 0 Vizsla 1 0 1 0 Weimaraner 0 1 0 1 Whippet 0 0 0 1 Age 4 month* 0 2 0 0 > 4 months 26 11 21 11 Average age in months (SD) 21.5 (19.6) 26.6 (29.2) 28.2 (26.4) 36.2 (29.4) Coat Length Long* 1 1 1 1 Short 25 12 20 10 Size Small* 1 2 1 1 Medium/Large 25 11 20 10 Morphology Type Preferred 3 3 3 3 Not Preferred 23 10 18 8 environment as described in experiment 1. Table 5 lists the breed, age, coat length, and size of the dogs enrolled in the study. Dog Training: The procedure for experiment 2 was similar to experiment 1, except the experimenters, who showed the visitors around, also gave treats to the dogs in the experimental condition after ringing the bell, thereby continuing training throughout the day. For the first three days (phase 1), side A received the experimental condition, while side B received the baseline condition. In the next three days (phase 2), side A received the baseline condition, while side B received the experimental condition. Because the dogs were trained throughout the day, the morning training trials were replaced with the following procedure. During the experimental condition, dogs received three daily sessions of a single training trial (at 9:00h, 12:00h, and 15:00h) of pairing a 2 s sound of 479 Anthrozoös

Judging a Dog by Its Cover: Morphology but Not Training Influences Visitor Behavior a hand bell, which the trainer carried, with the delivery of food, while video recording the dog at each trial. During the baseline condition, dogs were videotaped but did not hear the bell or receive the food. The videos were used as probes to assess the effect of responseindependent pairing on the behavior of dogs themselves as well as to calculate IOA. Data Collection on Visitor Behavior: Experimenters observed shelter visitors, who spontaneously entered the shelter. Because the dogs were now trained throughout the day by the observers themselves, the observers stayed at the shelter for the duration of the day to maximize data collection and were no longer blind to which side of the shelter was receiving which condition. When a shelter visitor walked in and indicated an interest in looking at the dogs, an experimenter led the visitor to the dog area. An additional experimenter rang a bell for 2 s and delivered a treat before the visitor reached every dog kennel. If asked, the experimenter told the visitor that the bell alerted the dogs to their presence. An additional observer, also not blind to the condition of the dogs, followed behind the visitor and collected data, using paper and pencil, on visitor behavior. Dogs were categorized into morphologically preferred and not preferred as described above. Data Analysis: As in experiment 1, a Mann-Whitney U Test was used to evaluate the effect of condition (baseline versus training) on undesirable dog behavior. To verify our previously established pattern of visitor responding, a PCA was repeated for the data obtained in experiment 2. Similarly to the previous experiment, two multivariate general linear regression models (one per each phase) were used to assess the effects of sides of the shelter on visitor behavior. We again explored the effect of undesirable in-kennel behavior on visitor behavior using a multivariate general linear regression model. Finally, we assessed whether the morphology of the dogs influenced adopter behavior by aggregating the data across the two phases; we used a multivariate general linear regression model with morphological type (preferred or not preferred) as factors and the behavior of visitors as dependent variables. To establish IOA, a third (33.3%) of the probes of dog behavior were coded by an additional observer. IOA was calculated in the same manner as in experiment 1. The average IOA across days was 94.4% (range: 90.3 100.0%). We did not collect data on visitor behavior IOA in experiment 2 due to concerns of behavioral reactivity of the visitors to the experimental sessions; the visitors were already accompanied by three experimenters during their visit. 480 Anthrozoös Ethical Note: The procedures were approved by the University of Florida Institutional Review Board, the University of Florida Institutional Animal Care and Use Committee, and the Alachua County Animal Services. Results A total of 63 visitors were observed across the two phases of the experiment: 29 visitors in phase 1 and 34 visitors in phase 2. A total of 71 dog observations (47 unique dogs) were conducted in the study across the two phases: 39 dog observations (26 in side A and 13 in side B; 39 unique dogs) in phase 1 and 32 dog observations (21 in side A and 11 in side B; 31 unique dogs) in phase 2. Undesirable in-kennel behavior of the dogs was, again, not normally distributed (W = 0.96, p < 0.05), therefore supporting the use of a nonparametric test to compare the differences in undesirable behavior across the two conditions. Dogs in the training condition (aggregated data across all phases) spent less time in engaging in undesirable in-kennel behavior than in the baseline condition (U = 4.01, p = 0.0001). The median proportion of undesirable behavior during training was zero (IQR = 0.25) and during baseline was 0.44 (IQR = 0.28; Figure 3).

Protopopova and Wynne Figure 3. Proportion of undesirable in-kennel behavior in the experimental condition in experiment 2. Whiskers show 10th and 90th percentiles, boxes show 25th and 75th percentiles and the medians, and the dots are outliers. The pattern of visitor behavior was highly similar to experiment 1. Visitors attended to 33.8% of the dogs available at the shelter (SD = 20.5%). When stopped to look at a dog, visitors spent an average of 14.37 s observing each dog (SD = 7.11s). When counting only the kennels actually visited, people visited a dog s kennel on average 1.28 times (SD = 0.59). On average, visitors read the kennel cards of 21.2% of the dogs (SD = 15.1%). Visitors bent down to interact with on average 12.6% (SD = 15.9%), reached for 12.8% (SD = 15.9%), and spoke to 15.4% (SD = 14.7%) of all of the dogs. Visitors asked for more information about 2.9% of the dogs (SD = 4.9%) and requested to take only 1.0% of the dogs out of their kennels for further inspection (SD = 2.6%). To confirm the pattern found in experiment 1, a PCA, restricted to two components with a direct oblimin rotation was conducted. The correlation among the two components was moderate (0.24). Component loadings were similar to what was found in experiment 1 and are shown in Table 6. Component 1 included reaching for the dog, frequency of visits, bending down/kneeling, duration of time spent in front of kennel, speaking, and reading the kennel card. This component was similar to component 1 from experiment 1, with the exception that the frequency of visits and duration variables loaded wholly on this component. Component 2 consisted of the frequency of taking the dog out of the kennel and asking about the dog, which was similar to component 2 from experiment 1. However, reading the kennel card also loaded onto component 2, but less than onto component 1. A multivariate general linear regression model showed that the two components did not differ in phase 1, in which side A received the experimental condition, while side B received the baseline condition (F (2, 36) = 1.49, p > 0.05; 2 = 0.08, 95% CI: 0 0.24). The mean score for component 1 on side A was 0.20 (SD = 0.80, n = 26) and on side B was 0.35 (SD = 1.33, n = 13). The mean score for component 2 on side A was 0.09 (SD = 1.50, n = 26) and on side B was 0.12 (SD = 0.60, n = 13). 481 Anthrozoös

Judging a Dog by Its Cover: Morphology but Not Training Influences Visitor Behavior Table 6. Principal component analysis of visitor behavior towards kenneled dogs in experiment 2. Component loadings 0.30 are in bold. Visitor Behavior Component 1 Component 2 Reaching 0.90 0.17 Frequency 0.87 0.06 Bending Down/Kneeling 0.85 0.15 Duration 0.84 0.14 Speaking 0.78 0.21 Reading Kennel Card 0.62 0.37 Taking Out 0.29 0.88 Asking Questions 0.26 0.80 Figure 4. The mean number of times the visitors asked to take morphologically preferred dogs (i.e., puppies, long-coated dogs, small dogs, and certain breed types) out of the kennel compared with morphologically non-preferred dogs in experiment 2. Standard errors are shown. 482 Anthrozoös A multivariate general linear regression model was repeated for phase 2, in which side A received the baseline condition, while side B received the experimental condition. The sides did not differ in either components (F (2, 26) = 1.56, p > 0.05; 2 = 0.11, 95% CI: 0 0.31). The mean score for component 1 on side A was 0.16 (SD = 0.97, n = 18) and on side B was 0.31 (SD = 1.05, n = 11). The mean score for component 2 on side A was 0.05 (SD = 0.34, n = 18) and on side B was 0.06 (SD = 0.80, n = 11). As before, for descriptive purposes, all dogs were divided into categories of low (< 0.2) or high ( 0.2) proportions of undesirable behavior, but no effect was found on the two components (F (1, 68) = 2.40, p > 0.05; 2 = 0.03, 95% CI: 0 0.15). The mean score for component 1 for the low category of dogs was 0.18 (SD = 0.91, n = 32) and for the high category was 0.15

Protopopova and Wynne (SD = 1.05, n = 39). The mean score for component 2 for the low category of dogs was 0.05 (SD = 1.38, n = 32) and for the high category was 0.04 (SD = 0.54, n = 39). Finally, a reduced multivariate general linear regression model found that morphology influenced component 2 only (F (1, 69) = 4.47, p < 0.05; 2 = 0.06, 95% CI: 0 0.19). The mean score for component 2 (interest) for the non-preferred dogs was 0.11 (SD = 0.58, n = 59) and for preferred dogs was 0.54 (SD = 2.05, n = 12). As in experiment 1, to further explore the effect of morphology on visitor behavior, a multivariate general model including all visitor behaviors (log-transformed) found that only the frequency of taking the dog out of the kennel was predicted by its morphology (F (1, 69) = 4.66, p < 0.05; 2 = 0.06, 95% CI: 0 0.19). Morphologically non-preferred dogs were taken out of the kennel an average of 0.01 times (SD = 0.02, n = 59) and morphologically preferred dogs were taken out of the kennel an average of 0.09 times (SD = 0.29, n = 12; Figure 4). General Discussion Training reduced inappropriate behavior in the kennel, as shown in previous studies (Protopopova and Wynne 2015); in both experiments, dogs in the response-independent pairing condition spent less time engaging in undesirable in-kennel behavior. In fact, this training essentially eliminated undesirable behavior in kenneled dogs in both experiments. We did not find an effect of training on visitor behavior in either experiment. Dogs in the experimental condition, even when verified to not behave in an undesirable manner during visitor observation times (experiment 2), did not induce different behavior from visitors than the dogs in the baseline condition. Furthermore, a lack of undesirable behavior of the dogs, regardless of condition, did not affect visitor behavior when aggregated across the different phases of the study. In experiment 2, we eliminated the possibility that the dogs behaved undesirably during the times when visitors were observing. Therefore, our data suggest that in-kennel training, although effective at decreasing inappropriate behavior, might not affect visitor behavior. Only a few other studies explored the use of training dogs to behave a certain way in their kennels in order to affect the behavior of visitors or adoption likelihood. The most thorough study to date, by Herron, Kirby-Madden and Lord (2014), found that whereas in-kennel training that involved delivering food reinforcers contingent on desirable in-kennel behaviors was effective at increasing these in-kennel behaviors (i.e., staying quiet, sitting or lying down in the kennel, and not jumping on the kennel doors), there were no differences in adoption rates among the trained and the control groups of dogs. Luescher and Medlock (2009) provided exciting findings that out-of-kennel dog obedience training combined with in-kennel training to not bark and to stay in the front of the kennel increased adoption rates compared with a control group of dogs. However, because the training combined both out-of-kennel and in-kennel components, judging the relative efficacy of each is difficult. In fact, current evidence suggests that behavior outside of the kennel has a large impact on adoption (Protopopova and Wynne 2014) and that training the dogs to behave appropriately outside of their kennels can significantly improve adoption rates (Protopopova, Brandifino and Wynne 2016). Our findings, together with previous research, suggest that visitors might be influenced by the dogs behavior outside of their kennels, but potentially not by their in-kennel behaviors. However, a more thorough investigation of all in-kennel behavior, and not just undesirable behavior as defined in this study, is warranted prior to reaching this final conclusion. An extensive literature exists on the effects of morphology on length of stay and/or adoption rate. Adopters consistently prefer young and small-sized dogs as well as certain breed types 483 Anthrozoös

Judging a Dog by Its Cover: Morphology but Not Training Influences Visitor Behavior 484 Anthrozoös (Brown, Davidson and Zuefle 2013). Other morphological variables that may improve adoption likelihood are long coat length (e.g., Wells and Hepper 1992; Siettou, Fraser and Fraser 2014; Protopopova et al. 2014) and light coat color (e.g., Lepper, Kass and Hart 2002; but see Woodward, Milliken and Humy 2012; Goleman, Drozd and Czyzȯwski 2014). Therefore, our finding that morphology affects visitor behavior is not surprising. Our data showed that the morphology of the dogs influenced visitor behavior but not in a consistent manner. Specifically, we found that morphologically preferred dogs (puppies, small dogs, long-coated dogs, and herding, ratter, and toy breed type dogs) had a higher number of visits to their kennels in experiment 1. In experiment 2, morphologically preferred dogs were taken out of their kennels more often. It is currently unclear why the two experiments found that morphology affected different visitor behaviors. One possibility is that the two behaviors may not be fundamentally different and that morphology may influence both of these behaviors. An alternative possibility is that the low sample size did not permit us to reach clear conclusions. Experiment 1 found that visitor behavior may be divided into two categories: interest in assessing the dog further, and affiliative behavior toward the dog. Interest in assessing the dog further in an out-of-kennel environment included the frequency of taking the dog out of the kennel and asking about the dog. Affiliative behavior toward the dog included speaking, bending down/kneeling, reading the kennel card, and reaching for the dog. Frequency of visits to the dog s kennel and the duration of time spent in front of the kennel loaded on both components. To confirm the pattern of visitor behavior that emerged in experiment 1, a PCA was conducted on the new set of data in experiment 2. The pattern was quite similar with two exceptions. Duration of time spent observing the kenneled dog and the frequency of visits switched from loading onto both components in experiment 1 to loading entirely on the affiliation component in experiment 2. Reading of the kennel card switched from loading onto the affiliation component in experiment1 to loading onto both components in experiment 2. Experiment 1 found that morphology affected the affiliation component, but experiment 2 found that morphology affected the interest component. This inconsistency across two experiments may be due to the relatively small sample size of the visitors included in the study. Also, the correlation of the components was moderate to high in both experiments, suggesting that the two components may not be independent of each other. A replication of this study with a larger sample size of visitors is warranted to clarify the patterns of shelter visitor behavior. Wells and Hepper (2001) conducted the only other study to date, to the best of our knowledge, assessing the behavior of visitors toward kenneled dogs at an animal shelter. Similarly to our data, they found that visitors only observed approximately 30% of the available dogs. In the present study, visitors observed approximately 35% of all of the available dogs. We found that when accounting for only those dogs that visitors looked at, visitors spent an average of just 15 s looking at them. Surprisingly, this is substantially less time than was found previously. Wells and Hepper (2001) reported that visitors spent 70 s on average observing and 20 s on average interacting with, kenneled dogs. It is possible that by following visitors, we have truncated the time they would have spent observing the dogs. However, it is also possible that the location of the shelters in different countries resulted in the sampling of different populations of visitors and are thus not readily comparable. Our results, together with previous research (Protopopova et al. 2014) suggest that the in-kennel selection of dogs may primarily involve morphology. Potential adopters may be