**Evidence for Individual Di**ff**erences in Behaviour and for Behavioural Syndromes in Adult Shelter Cats**

**Sandra Martínez-Byer 1, Andrea Urrutia 1,2, Péter Szenczi 3,\*, Robyn Hudson <sup>2</sup> and Oxána Bánszegi 2,\***


Received: 1 April 2020; Accepted: 29 May 2020; Published: 1 June 2020

**Simple Summary:** An important activity of modern animal shelters is the development of successful adoption programmes. In this regard, there is a need for reliable tests of individual differences in behaviour to help match the "personality" of potential adoptees with the lifestyle and needs of prospective owners; a companion animal for an elderly person remaining at home requires a different match than a pet for someone who will be away most of the day; a pet kept exclusively indoors in a small apartment requires a different match than an indoor/outdoor pet. In the present study, we repeatedly tested 31 mixed-breed adult cats of both sexes and a wide range of ages in five behavioural tests at a shelter in Mexico City, Mexico. The tests were designed to be easily implemented by shelter staff, and were short and low cost and intended to simulate common situations in a pet cat's everyday life. We found consistent (stable) individual differences in the cats' behaviour on all five tests, as well as correlations between their behaviour across tests. This suggests that such tests may contribute to reliably characterizing the "personality" of individual cats and so help increase the rate of successful adoptions.

**Abstract:** Consistent inter-individual differences in behaviour have been previously reported in adult shelter cats. In this study, we aimed to assess whether repeatable individual differences in behaviours exhibited by shelter cats in different situations were interrelated, forming behavioural syndromes. We tested 31 adult cats in five different behavioural tests, repeated three times each: a struggle test where an experimenter restrained the cat, a separation/confinement test where the cat spent 2 min in a pet carrier, a mouse test where the cat was presented with a live mouse in a jar, and two tests where the cat reacted to an unfamiliar human who remained either passive or actively approached the cat. Individual differences in behaviour were consistent (repeatable) across repeated trials for each of the tests. We also found associations between some of the behaviours shown in the different tests, several of which appeared to be due to differences in human-oriented behaviours. This study is the first to assess the presence of behavioural syndromes using repeated behavioural tests in different situations common in the daily life of a cat, and which may prove useful in improving the match between prospective owner and cat in shelter adoption programmes.

**Keywords:** individual differences; behavioural assays; behavioural syndromes; companion animal; *Felis silvestris catus*; shelter cats; human-cat relation

#### **1. Introduction**

For years, the domestic cat (*Felis silvestris catus*) has been among the most popular pets in the world [1,2]. Interest in cat behaviour, and particularly in inter-individual differences (animal personality), is reflected in recent reviews [3–6] and special issues in scientific journals treating such topics [7,8]. The cat is a good candidate for the study of individual differences as it is readily accessible and has a rich behavioural repertoire. It is also by far the most studied feline species in this respect [3]. As with other domestic animals (companion, farm and working animals), taking into account cats' personality differences when rehoming or selecting them for specific tasks can have implications for management, welfare and economy [3,9,10].

Broadly defined, animal personality refers to relatively stable inter-individual differences in behaviour [11–13]. When several of these behaviours correlate across contexts, they can be characterized as a behavioural syndrome [12,14,15]. The most common methods used to study individual differences in behaviour in the cat include observation [16,17], owner surveys [18,19] and behavioural tests [20,21]. The latter have the advantage that they can be used to evaluate and quantify the stability of individual differences across repeated standardised testing. Since an individual's behaviour is expected to be variable to some degree, some behaviours may be inconsistent and therefore less informative of the individual's behaviour at a later time. Therefore, when testing cats, reliable methods are needed, i.e., behavioural tests and measures that have been found to be highly repeatable.

Many studies of cat personality or temperament are based on behavioural observation ([3,4] see reviews), which provide important information about cats' behaviours in their daily environments. However, to explore cats' reactions to specific situations, behavioural tests are necessary. The two most commonly used tests in cat personality research are novel object tests, where the animal is presented with an unfamiliar object, and tests of reaction to either familiar or unfamiliar humans [3]. Novel object tests tend to use stimuli of unclear biological relevance (e.g., a fan with paper streamers, a remote-control car, a metal container with a spring, or a wooden box; [20,22,23]). While these tests have been reported to reveal individual differences, their meaning in daily situations of the life of the cat is unclear. Therefore, in the present study, we decided to test the behavioural responses of cats to situations corresponding to what they would likely encounter in real-life situations.

Given cats' popularity as companion animals, there has been a tendency to study their individuality in terms of their interaction with humans, for example, in their reaction to approach or handling by a familiar or unfamiliar person [20,24–27]. Other behaviours of interest for both companion and working cats (particularly mousers) include their reaction to everyday stressful situations or to prey, respectively. However, we are unaware of any studies that have experimentally addressed the inter-individual consistency of behavioural differences in these situations. Nevertheless, animal shelters have begun to implement personality testing as part of their adoption programmes, favouring a combination of surveys and behavioural testing, as in the Feline Temperament Profile [21] and the Meet Your Match Feline-ality assessment [28].

The present study is the first to incorporate repeated measurements using several behavioural tests and to take a behavioural syndrome approach by evaluating correlations among these measurements in a heterogeneous population of cats (wide age range, different backgrounds) housed in an animal shelter. Animal shelters have a continuing need for reliable personality tests, for example, to better match potential pets with prospective owners and households or to identify cats that may better fit a specific situation, such as working or therapy cats. We used five behavioural assays that we consider to be ethologically and ecologically relevant to the daily life of the domestic cat, repeated three times each (see details below). We previously reported an analysis of data which included a subset of the data presented in the present paper, gathered during the separation/confinement test [29], but here we include further behavioural tests with the aim of identifying a larger range of repeatable individual differences and behavioural syndromes.

#### **2. Materials and Methods**

#### *2.1. Study Site and Animals*

We collected data from 31 adult cats (14 males and 17 females) from a shelter in Mexico City, Mexico, aged between 1 and 11 years (mean 4.5, SD 2.6, Supplementary Material Table S1). In some cases, the cats' ages were not known with certainty and were estimated by veterinarians. Participants were chosen randomly from among the cats at the shelter, which were in good health and permitted handling. All the cats had been neutered and had received post-operative care by qualified veterinarians within three days of entering the shelter, and all cats participating in the study had been at the shelter for at least six weeks prior to the start of behavioural testing. The shelter was a four-storey house divided into sections; approximately 50 cats were housed in each section according to how well they tolerated each other. All sections consisted of at least two rooms (approx. 2.5 × 3.5 m each) with access to a fenced outdoor area (approx. 2 × 4 m). Each cat was free to roam within its section. The rooms were furnished with cat beds, boxes of assorted sizes with blankets, scratchers and toys. Water, commercial dry cat food and sand boxes were always available.

#### *2.2. Procedures*

Tests were performed weekly for 12 sequential weeks; each of the five tests was performed three times across three sequential weeks (the human approach tests were performed on the same days). One test was performed per day on all subjects, tested in randomized order between 13:00 and 18:00. Not all cats were available for all trials, therefore sample sizes differ slightly between the tests (see Supplementary Material Table S1 for information on which cats participated in each test). All tests were video recorded (GoPro© Hero3+, GoPro, Inc., San Mateo, CA, USA) for subsequent behavioural analysis.

#### *2.3. Behavioural Testing*

#### 2.3.1. Struggle Test

We proposed the struggle test as a proxy for the handling tests used in different mammalian [30–33] and bird species [34–36]. Since domestic cats are frequently handled by their owners, by other familiar and unfamiliar humans, and by veterinarians, we redesigned this test to evaluate the struggle response when they are picked up and restrained. We tested 30 adult cats (13 males and 17 females; mean age 4.5, SD 2.6 years, min = 1, max = 11). The test was performed in the section of the shelter where the cat normally resided. One of the experimenters (S.M.-B.) approached the cat and stroked it three times from the head to the base of the tail, then picked it up, holding it with both hands around the thorax, under its forelimbs. The test lasted until the cat began to struggle (see Table 1 for behavioural definition) or until 30 s elapsed after picking it up. When this happened, the cat was immediately set down. The experimenter wore gloves as a precaution against scratches.


**Table 1.** Behavioural variables recorded in each test.


**Table 1.** *Cont.*

#### 2.3.2. Separation/Confinement Test

Separation/confinement tests are used for personality testing in many animals, particularly in social species [37–41]. Despite the fact that cats are considered only facultatively a social species [42,43], in previous studies this type of test has been successfully used for evaluating individual differences in kittens of the domestic cat [44,45] and adult shelter cats [29]. Moreover, this test represents a common situation in a cat's daily life around humans, since cats are often confined in a carrier to take to other places outside their home.

The data from this test combined with other data from additional shelter cats have been previously reported in Urrutia et al. [29]. We tested 28 adult cats (12 males and 16 females; mean age 4.6, SD 2.7 years, min = 1, max = 11). Tests were performed in a small closed room unfamiliar to the cats; the room was 1.5 × 2 m, with flat-finished, unpainted concrete floor, walls and ceiling, and without furnishings. During the test, no other animals or humans were allowed to enter either the test room or the room adjacent to it to limit auditory and olfactory contact. One experimenter approached the cat (either S.M.-B. or A.U.), briefly stroked it and then carried it in her arms into the test room. With the help of a second experimenter, they placed the cat inside a standard commercial pet carrier (42 × 61 × 38 cm), which was a closed plastic box with a steel grill door at one end and ventilation holes along the sides. The carrier, with the cat inside, was then placed on the floor at a previously marked position and the experimenters left the room. The test lasted two minutes. Once this time had elapsed, the cat was removed from the carrier and returned by one of the experimenters to its home room. The video camera was set up 60 cm from the carrier. To improve visibility, a red light was mounted inside the carrier. The carrier was cleaned between trials with isopropyl alcohol. See Table 1 for definitions of the behaviours analysed in this test.

#### 2.3.3. Mouse Test

In our experience, neither kittens nor adult cats show sustained interest in interacting with the types of inanimate objects conventionally used in novel object tests. We therefore chose tame, laboratory-strain (BALB/c) mice as the "novel object" to more closely approximate a biologically relevant stimulus, since small rodents are the most common prey of the domestic cat [46–50] and because of the ease with which they can be maintained and handled (see below for details on how the mouse was presented; see also [51]). In a previous study by Yang et al. [52], the BALB/c mouse strain was found to show the least fearful reactions in response to a predator. In our tests, a total of five mice were used in rotation; three of them were taken to the shelter on test days. The mouse in the jar was switched every two trials (approx. 10 min) to minimize stress. The stimulus animals showed no obvious signs of fear in the presence of the cats; there were no signs of panic (e.g., freezing) or attempted escape or defence (e.g., jumping), they moved around in the jar in apparent calm, sometimes adopting

the stretch–attend posture—which according to previous research is indicative of risk assessment rather than a fearful reaction [52]—in apparent curiosity at the presence of the cats. At the end of the study, the mice were adopted by student participants. For more details on the housing of the mice outside the tests, see Supplementary Material File S2. Additionally, during pilot tests, thermal pictures of the mice were taken before and after being in the jar with a cat in the room. Analysis of these images showed that the stress experienced by the mice (as measured by the increase in eye temperature) was comparable to that experienced in routine laboratory tests [53,54].

We tested 23 adult cats (7 males and 16 females; mean age 4.4, SD 2.5 years, min = 1, max = 11). Cats were individually tested in an unfamiliar room (4 × 6 m) which was cleared of all other cats and any objects that could be distracting. Subjects were given a two-minute habituation period before introducing the mouse. During habituation, and throughout the test, an experimenter (S.M.-B.) remained in the room, standing motionless and silent in a corner.

At the end of the habituation period, the experimenter restrained the cat in the middle of the room while a second experimenter brought in a mouse inside a clear, thick glass jar (15 cm in diameter × 20 cm high) with a perforated lid and covered with a cardboard box. At a marked position approximately 1.5 m from the cat and against a wall, the second experimenter fixed the jar to the floor with double-sided tape, removed the cardboard box and left the room. The first experimenter then released the cat and returned to the corner. The cat could see and presumably hear and smell the mouse but could not access it. The cat was free to interact with the jar for two minutes, after which the test ended and the cat was returned to its section of the shelter. The video camera was mounted on the wall 2 m above the jar. See Table 1 for definitions of the behaviours analysed.

#### 2.3.4. Human Approach Tests

Human approach tests have been commonly used to evaluate cat behaviour [20,25,55–57], especially in shelters [27,58]. We modified the test from Adamec et al. [59] and tested the response of 28 adult cats (11 males and 17 females; mean age 4.6, SD 2.7 years, min = 1, max = 11) to an unfamiliar person. This person, a male volunteer, was the same person on a given test day but a different volunteer each week (age 21–25 years). To minimize unintentional odour cues, all were non-cat owners, were asked to wear fresh clothes and were unknown to the cats. Thus, the cats had the opportunity to interact with three different humans, one in each trial.

• Passive human approach test

Tests were performed in the same room as described for the mouse test. Before testing, two concentric circles, 1.5 and 3 m in diameter, were drawn on the floor with chalk to use as references of cat–human distance in the later video analysis, and the male volunteer was asked to sit cross-legged on the floor in the centre of the inner circle. When the volunteer was in position, the cat was carried in arms into the room by a familiar experimenter and placed in a shallow (20 cm deep) open wooden box against the wall next to the door. The experimenter then left the room. The test started when the door closed, leaving the cat alone with the unfamiliar person. The test consisted of two parts. For the first three minutes the unfamiliar volunteer sat cross-legged on the floor, looking at the wall and ignoring the cat however close it got. We used an approach score from 1 to 5 depending on whether the cat did the following: (1) remained outside the large circle; (2) entered at least its forepaws in the large circle; (3) entered at least its forepaws in the small circle; (4) established physical contact with the human (rub, sniff, touch with paw); (5) put at least its forepaws on top of the human. Then, in the second part, the volunteer continuously called the cat by its name for one minute while extending his arm and index finger as a greeting, pointing in the cat's direction, even if the cat had already made physical contact with him. See Table 1 for definitions of the behaviours analysed in this test.

## • Active human approach test

This test was performed immediately after the passive human approach test. The volunteer was instructed to slowly rise to his feet, approach the cat and attempt to stroke it six times from the head to the base of the tail. If the cat moved away before it could be stroked six times, the unfamiliar human walked after it and attempted to stroke it again. The test ended after the sixth stroking attempt (whether successful or unsuccessful) or after 1 min. The experimenter then entered the room and returned the cat to its home room.

#### *2.4. Ethical Considerations*

Throughout the study, animals were kept and treated according to the guidelines for the use of animals in research as published in Animal Behaviour (ABS, 2016), as well as the relevant legislation for Mexico (National Guide for the Production, Care and Use of Laboratory Animals, Norma Oficial Mexicana NOM-062-200-1999), and approved by the Institutional Committee for the Care and Use of Laboratory Animals (CICUAL, permission ID 6315) of the Institute of Biomedical Research, UNAM, Mexico City, Mexico.

#### *2.5. Video and Statistical Analysis*

All behavioural variables were coded using Solomon Coder software for video analysis [60]. Statistical analyses of the data were carried out using the programme R, version 3.6.1 (R Foundation for Statistical Computing, Vienna, Austria) [61]. Prior to fixed-effects and repeatability analyses, any non-normally distributed continuous variables were normalized using either a Box–Cox or log transformation with the R package MASS [62]. Effects of sex, age and trial number on behavioural variables were analysed using linear mixed-effects models (LMM) for continuous, and generalized linear mixed-effects models (GLMM) for count (i.e., Poisson distributed) or binary (binomially distributed) dependent variables with the R package lme4 [63]. As fixed effects, we included sex, trial number (1 to 3), age (as a covariate), the interaction of sex × age and the interaction of trial number × age. As a random factor, we included individual identity. We applied backwards stepwise reduction of the full models beginning with non-significant interactions followed by non-significant fixed effects when *p* > 0.05. Individual identity as a random factor was retained in all models to account for repeated measures of individuals. *p*-values were extracted by Wald chi-squared tests (type III).

We then analysed the repeatability of individuals' behaviour across the three trials by intra-class correlations calculated as the proportion of phenotypic variation that can be attributed to between-subject variation [64]. We used GLMM-based calculations for count (Poisson distributed) or binary (binomially distributed) data and LMM-based calculations (Gaussian distributed) for continuous data for testing the repeatability of individual differences using the R package rptR [65,66]. Individual identity was used as a random factor and the fixed effects found to have a significant effect on each behaviour in the previous analysis were included where applicable. For all intra-class correlations, we calculated 95% confidence intervals by 1000 bootstrap steps, and *p*-values were calculated by 1000 permutations.

To investigate the possible association of the behaviours between the different tests, we first performed principal component analyses (PCAs) independently on each of the following tests: separation/confinement, mouse and passive human approach using spectral decomposition assuming correlation matrices, to reduce the number of dimensions; no rotations were used. In the case of the struggle and active human approach tests, we used the raw behavioural data, since only one behaviour was coded in each of these two tests. Since phenotypic correlations between traits may originate from two sources, that is, (i) from individuals' average levels of two traits (between-individual correlation) or (ii) from individuals' change in behaviour (within-individual correlation) [67–70], we calculated between-individual and within-individual (residual) correlations by using multivariate linear mixed models with the R package sommer [71] to partition possible phenotypic correlations between the traits. *p*-values were corrected for multiple tests using the Benjamini–Hochberg method.

#### **3. Results**

#### *3.1. Repeatability of Individual Di*ff*erences within Tests*

#### 3.1.1. Struggle Test

No effects of age, sex or trial or of the interaction between these were found on the latency to struggle (Supplementary Material Table S3). All cats (*n* = 30) struggled within the 30-s limit, with only one cat still held at 30 s on one occasion. Individual differences in the latency to struggle were significantly repeatable across the three trials (Table 2).

**Table 2.** Repeatability of the variables analysed for each of the behavioural tests. Intra-class correlation coefficients (*R*), 95% confidence intervals (lower bound, upper bound) based on 1000 bootstrap steps and significance values (*p*) are given. Asterisks indicate significance levels at *p* < 0.05 \*, *p* < 0.01 \*\*, *p* < 0.001 \*\*\*.


#### 3.1.2. Separation/Confinement Test

Age and trial number (1–3) were found to have a small, significant effect on the number of vocalizations and the duration of motor activity; older cats vocalized less and moved less in the carrier, and both behaviours diminished in consecutive trials (Supplementary Material Table S3). In the case of latency to initiate motor activity, there was a significant but very small effect of sex, where males began motor activity slightly sooner. There was a small effect of the interaction between age and sex, where the latency to move was slightly higher in older males than in younger males. There was also a small effect of trial number, where latency to begin motor activity began slightly later in consecutive trials (Supplementary Material Table S3). Therefore, these significant fixed effects were included in the respective repeatability analyses. Individual differences in the latency to vocalize and the number of vocalizations emitted by the cats (*n* = 28) were highly repeatable. Duration of motor activity was also significantly repeatable, although the latency to locomote was not (Table 2).

#### 3.1.3. Mouse Test

The sequence of trials was found to have an effect on the duration of interactions (cats interacted less with the mouse on the third trial than during the first two trials) and was thus added as a fixed effect in the analysis (Supplementary Material Table S3). No other variable showed an effect of age, trial number or sex or the interaction between them. We found highly repeatable individual differences in the latency to approach and the time cats (*n* = 24) spent near the mouse across trials. Variables associated with proximity to the mouse were likewise repeatable, such as the time spent walking around the jar, the latency to interact and the duration of interaction (Table 2). Even tail swishing, which was coded from any area of the room, showed repeatable individual differences, a possible sign of interest or arousal of the animal even from afar.

#### 3.1.4. Human Approach Tests

None of the behavioural variables measured in these tests was significantly affected by age, trial number or sex or the interaction between them (Supplementary Material Table S3).

#### • Passive human approach test

We found repeatable individual differences (*n* = 28) for all behavioural measures in both phases of the test across trials, that is, the distance individual cats kept from the unfamiliar human was consistent even though each of the three trials used a different unfamiliar volunteer. We also found repeatable individual differences for the finger-nose contact measure of phase two. Moreover, individual differences in the latency to vocalize and in the number of vocalizations emitted during the entirety of trials were also highly repeatable (Table 2).

• Active human approach test

Individual differences in the latency for the unfamiliar person to be able to stroke the cat were consistent across trials and even though this involved three different people (Table 2).

#### *3.2. Correlations Between Tests*

For dimension reduction purposes, we performed three separate PCAs on the behavioural variables of the following tests: separation/confinement, mouse and passive human approach. For the full results of the PCAs, see Supplementary Material Table S4. In the separation/confinement test, two principal components were extracted. For factor 1 ("confinement/separation vocalization"), the behaviours with the highest loadings were those related to vocalization and, for factor 2 ("confinement/separation motor activity"), the highest loading was the duration of motor activity. In the mouse test, two principal components were extracted. For factor 1 ("interaction with the mouse"), the behaviours with the highest loadings were related to the cats' proximity to and interaction with the mouse jar and, for factor 2 ("tail swishing"), the highest loading was for the duration of tail swishing. In the passive human approach test, two principal components were also extracted. For factor 1 ("approaching the passive human"), the behaviours with the highest loadings involved the human approach score and finger–nose contact; for factor 2 ("passive human approach vocalization"), the behaviour with the highest loading was the number of vocalizations.

In each of the two remaining tests (struggle and active human approach), we measured only one behavioural variable (latency to struggle and latency to be stroked by the human, respectively), hence we did not perform a PCA for these tests. Using the raw data for these variables, along with the six previously described factors obtained from the PCAs, we calculated correlations using multivariate linear mixed models. From a total of 34 correlations (Supplementary Material Table S5), we found eight that were significant after adjusting *p*-values for multiple comparisons (Benjamini–Hochberg method; Figure 1).

**Figure 1.** Correlations between behavioural variable scores showing stable individual differences at the between-individual level. Asterisks indicate significance levels at *p* < 0.05 \*, *p* < 0.01 \*\*. Black lines correspond to positive correlations, red lines correspond to negative correlations. Line thickness corresponds to the strength of a correlation. Further details are available in Supplementary Material Table S5, including confidence intervals.

#### **4. Discussion**

#### *4.1. Consistency Across Time*

In this study, we first evaluated the consistency across time of individual differences in behavioural responses of adult shelter cats in five different tests, and for all tests we found measures that showed significant repeatability. Stable individual differences were evident even though the cats were a heterogeneous population that differed in age, sex and (largely unknown) background. Perhaps surprisingly, individual differences in behavioural responses in most of the tests were unrelated to age or sex, suggesting that the behaviours measured here may be useful for evaluating individual differences in adult cats in general. This is supported by previous studies reporting stable individual differences in cats and other mammals in tests similar to those used here, that is, struggle or restraint tests used in cats [26], mice [31], rabbits [32,72,73], North American red squirrels [74,75] and pigs [33,76]; social separation tests used in cats [29,44,45], horses [77] cows [39,78] and dogs [40]; mouse tests used in cats [51,79]; and human approach tests used in cats [21,25,27,56,57,59], dogs [80], pigs and cattle [81,82]. These tests in their various forms are all relevant to the daily life of most cats, and thus provide a promising basis for assessing cat personality across a wide range of populations and conditions, including in shelter cats.

#### *4.2. Behavioural Syndromes*

We found seven significant correlations between behavioural scores from the different tests (Figure 1). Most of these seemed to be connected with humans; for example, cats that readily approached the unfamiliar human in the passive human approach test also struggled sooner in the struggle test, which may suggest that these cats were more confident around humans. Cats that struggled sooner also tended to vocalize (meow) more during the confinement test when separated from humans and other cats, suggesting that these individuals may seek the company of humans more, since meowing is considered a human-oriented behaviour ([83,84] our observation). Such correlations may indicate the existence of behavioural syndromes as defined in the Introduction.

We can suppose that while the separation/confinement test was probably a negative experience for all the cats, the human approach test was a positive experience for at least some individuals. A more detailed acoustic analysis of the meows may help disentangle the emotional valence and motivation (e.g., stress, attention-seeking, greeting) underlying them in these tests, since meows emitted during distress have a distinct pattern (low mean fundamental frequency, longer duration; Schötz et al. [85]). Additionally, the cats for which the human approach test was a positive experience may have emitted

other vocalizations (e.g., purrs, which Fermo et al. [86] found are exclusively associated with positive experiences). However, we were not able to record them due to their low volume. It is also possible, as Guillette and Sturdy [87] have suggested, that the degree of arousal or readiness of the cat to act (due to activation of the sympathetic nervous system) may contribute to the pattern of vocal emissions in different contexts [88].

Consistent with previous findings, we did not find an association between the number of vocalizations and motor activity within the confinement/separation test, suggesting different underlying mechanisms (motivation) between these variables (see more details in [29]). However, there was a negative correlation between motor activity in the confinement/separation test and the number of vocalizations emitted during the passive human approach test. The only explanation we can presently offer is that the cats for which the passive human approach test was a positive experience may have "carried" this correlation, meaning that possibly only positive meows are correlated with motor activity. Further study into the relationship between meows and motor activity in positive and negative situations may help to disentangle this.

Additionally, interaction with the mouse was significantly correlated with three different variables. It was negatively correlated with vocalization in the human approach passive phase, which can be interpreted as cats that were more focused on the mouse were less demanding of human attention (vocalized less). The latter is supported by the positive correlation between interaction with the mouse and latency to be stroked in the active human approach test, i.e., cats that spent more time with the mouse took longer to allow themselves to be stroked. Taken together, these correlations suggest a syndrome where more prey-oriented individuals are also less human-oriented. Although cats' backgrounds in the present study were unknown, we speculate that such a syndrome may arise as a consequence of experiences prior to their arrival at the shelter, that is, cats that were more independent from humans may have relied more on hunting to obtain food, whereas cats that were more social with humans had relied on them for sustenance. Finally, there was also a positive correlation between interaction with the mouse and motor activity during the confinement/separation test, suggesting that some cats were more "excitable" than others, possibly due to differences in sympathetic nervous system arousal as discussed previously for vocalizations.

#### *4.3. Behavioural Testing in Animal Shelters*

All five tests implemented in this study are simple and fast (no more than five minutes each), and any materials used are inexpensive and easily procured. Because of this, they can be reproduced practically anywhere in the world with minimal instruction of shelter personnel. Together, this makes them a suitable option for shelters looking to evaluate personality as part of their adoption programme. While millions of cats enter animal shelters every year, in the United States, for example, only an estimated 11.5% of pet cats come from a shelter [4,28]. Furthermore, even if a cat is adopted, there is still a high chance that it will be returned due to not fulfilling the new owner's expectations, which risks euthanasia. Organizations like the American Society for the Prevention of Cruelty to Animals have managed to decrease the number of returned cats by applying questionnaires and personality tests [28].

However, these protocols are not applied worldwide, due to differences in owner expectations and the way shelters operate in different locations, among others. For example, animal shelter facilities in Mexico and throughout Latin America differ from those in the United States and Europe, something also noted by Fukimoto et al. [89] in their study of shelter cats in Brazil. Although our tests share some similarities with the ASPCA's Feline-ality behavioural assessment, we sought to develop tests that could be a better fit for the shelter conditions and owner expectations we are familiar with. For example, we chose to use the pet carrier as a test within itself to evaluate individual responses to isolation and confinement, as separation anxiety is a common concern for owners who work long hours away from home. We also included a novel test (mouse test) in which the cats are presented with a biologically relevant stimulus. Although we recognize that this test will not be relevant to all cats that are offered

for adoption as pets, nor is it feasible for all shelters to keep mice for this test, we would like to note that. In some shelters around the world. there are programmes to adopt out or loan "mouser" or "barn" cats ([c.f. [90] and also see the programs of the following organizations: Battersea Dogs & Cats Home (UK), Dereham Adoption Center (UK), Animal Humane Society (USA), Best Friends Animal Society (USA), Barn Cats Inc. (USA), among others). In recent years, there has been an increase in the demand for mousers by more environmentally friendly businesses and organic farms seeking to avoid rodenticides and to switch to biological pest control. This is an option for cats that are not sociable with people. Those individuals that show a strong interest in potential prey probably have a better chance of being successfully adopted into a working context.

Implementing repeated behavioural testing in the adoption process, whenever possible, could help match prospective owners with an animal that best suits the needs and lifestyle of both parties. For example, a family with small children needs a cat that tolerates handling; a calm person may want a calm cat; and someone who is not home most of the day would do better with a cat that is not stressed by separation.

#### **5. Conclusions**

Reliable, economic and easily implemented behavioural tests are needed by animal shelters to improve their adoption programmes by improving the match between the personality of the prospective pet, in this case the cat, and the context of its new home. This can be best achieved by using tests based on the natural, evolved behaviour of the cat relevant to its everyday life and using correlations between more than one behavioural measure to form a more reliable profile of each individual cat's personality. Results of the present study indicate that this is, indeed, feasible.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2076-2615/10/6/962/s1, Table S1: Details of the subjects and test assignment, File S2: Description of the housing of the mice during the experiments, Table S3: Results of the LMMs and GLMMs to evaluate the impact of age, sex and trial number on behaviour, Table S4: Full results of the PCAs performed on the confinement/separation, mouse and passive human approach tests, Table S5: Correlations between behavioural variable scores showing stable individual differences.

**Author Contributions:** Conceptualization, P.S., R.H. and O.B.; methodology, P.S., O.B., R.H. and S.M.-B.; data collection and video analysis, O.B., P.S., A.U. and S.M.-B.; statistical analysis, P.S. and A.U.; visualization, A.U.; writing—original draft preparation, S.M.-B. and A.U.; writing—review and editing, O.B. and R.H.; supervision, O.B. and P.S.; project administration, S.M.-B.; funding acquisition, R.H. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by the Dirección General de Asuntos del Personal Académico, Universidad Nacional Autónoma de México, Mexico City, Mexico, grant number DGAPA-IN212416, by a postdoctoral fellowship to O.B. by the Instituto de Investigaciones Biomédicas, Universidad Nacional Autónoma de México and by a Cátedra grant to P.S. from the Consejo Nacional de Ciencia y Tecnología, México, grant number 691. S.M.-B. and A.U. received a fellowship from CONACYT, number (número de becario) 948973 and 616266, respectively, and they thank the Posgrado en Ciencias Biológicas, UNAM, Mexico City, Mexico, for its support.

**Acknowledgments:** We are very grateful to Betty McGuire who invited us to participate in this Special Issue and made it possible. We thank Jimena Chacha for help with behavioural testing, and the staff of the Gatos Olvidados shelter for their support and for allowing us repeated access to their cats and facilities. This study was performed in partial fulfilment of the requirements for A.U. to obtain the PhD degree in the Posgrado en Ciencias Biológicas at the Universidad Nacional Autonóma de México, Mexico City, Mexico.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

*Article*

## **Characterizing Human–Dog Attachment Relationships in Foster and Shelter Environments as a Potential Mechanism for Achieving Mutual Wellbeing and Success**

## **Lauren E. Thielke \* and Monique A.R. Udell**

Department of Animal & Rangeland Sciences, Oregon State University, Corvallis, OR 97331, USA; Monique.Udell@oregonstate.edu

**\*** Correspondence: laurenthielke@gmail.com

Received: 25 November 2019; Accepted: 26 December 2019; Published: 30 December 2019

**Simple Summary:** The majority of research on attachment behavior in dogs has focused on the bonds between pet dogs and their owners. In this study, we examined attachment relationships between dogs living in animal shelters and foster homes and their temporary caregivers-shelter volunteers or foster volunteers, respectively. We also examined these results in relation to previously published data from pet dogs in order to contextualize our findings. Our findings indicate that the percentage of securely attached shelter dogs was significantly lower than that previously observed in scientific studies of the pet dog population. No differences were found between proportions of securely attached foster dogs and prior research with pets. We did not find significant differences between foster and shelter dogs in terms of attachment style proportions. We also found evidence of disinhibited attachment, which is associated with a lack of appropriate social responses with unfamiliar and familiar individuals in foster and shelter dogs. This is the first study to apply attachment theory to foster and shelter settings.

**Abstract:** This study aimed to characterize attachment relationships between humans and dogs living in animal shelters or foster homes, and to contextualize these relationships in the broader canine attachment literature. In this study, 21 pairs of foster dogs and foster volunteers and 31 pairs of shelter dogs and shelter volunteers participated. Each volunteer–dog dyad participated in a secure base test and a paired attachment test. All volunteers completed the Lexington Attachment to Pets Scale (LAPS), a survey designed to measure strength of attachment bonds as reported by humans. Although no significant differences were present in terms of proportions of insecure and secure attachments between foster and shelter populations, proportions in the shelter population were significantly lower (*p* < 0.05) than the proportions of attachment styles that would be expected in a population of pet dogs based on the published literature on pet dog attachment styles. Additionally, findings are presented in relation to data from a paired attachment test that demonstrate foster and shelter dogs spend more time in proximity to humans when the human is actively attending to the dog and encouraging interaction, as would be expected based on previous studies. We also present findings related to the presence of disinhibited attachment (previously reported in children who spent a significant portion of time living in institutionalized settings) which is characterized by a lack of preferential proximity seeking with a familiar caregiver and excessive friendliness towards strangers in foster and shelter dogs.

**Keywords:** attachment behavior; shelter dog; foster dog; disinhibited attachment; attachment style

#### **1. Introduction**

Although it is widely agreed that dogs and humans form attachment relationships with one another, the method of applying attachment styles to pet dog research is a fairly recent area of interest. Previous studies have explored attachment relationships in pet dogs [1–3] and in the ability of shelter dogs to form attachment relationships to an unfamiliar human in a shelter setting [4]. More recently, research has shown that dogs' behavior in attachment tests can be used to categorize dogs into attachment styles [5–7] originally described in literature focusing on infant–mother attachments [8]. While additional attachment styles have been described, the three primary attachment styles have commonly been defined in infant research as follows: secure (the infant shows signs of distress when separated from the mother and seeks proximity and contact when reunited), insecure-avoidant (the infant does not show much distress and does not seek proximity when reunited) and insecure-resistant (the infant is very distressed when the mother is absent but is not calmed when the mother returns and resists contact) [8].

While these attachment styles have since been applied to the pet dog–owner relationship, this component of attachment theory has not previously been applied to dogs in foster and shelter settings. Given that dogs living in animal shelters have been found to quickly form bonds to new humans [4] understanding the nature of these bonds, including the degree of attachment security that exists (which could have welfare implications) as well as similarities and differences with respect to the dog–'owner' bond seem especially relevant. Furthermore, although disinhibited attachment has been described among human children placed in homes after early life experiences in institutionalized settings [9], this topic has not previously been explored in dog attachment, and could be of particular interest for dogs in shelter settings. Disinhibited attachment is characterized by a lack of attenuating social responses to adults of varying familiarity, low levels of checking in with a familiar attachment figure in a stressful situation, and inclination to go off with an unfamiliar person. Disinhibited attachment can be mild or severe, and it can occur in individuals with any attachment style, although in humans, it is most pronounced in children with secure attachment styles.

Dogs that are housed in shelters for a prolonged period of time may be more likely to develop new behavior problems [10,11], or experience higher stress levels [12] and socio-cognitive declines [13]. For a review of sheltering's effects on dog behavior, welfare and physiology, see [14]. However, regular interactions with a person have been associated with improved behavioral outcomes [15,16] and decreased cortisol levels [17,18]. Another unexplored benefit of volunteer programs in which dogs are provided with opportunities to regularly interact with familiar people may be the opportunity for transitioning dogs to develop a secure attachment bond with these volunteers, which has been found to promote positive behavioral and cognitive outcomes in both pet dogs [19] and human children [20]. Orphaned children have been shown to have a greater likelihood of thriving and developing secure social bonds later in life if they develop a secure bond with a foster parent [21].

In this study, we used the Lexington Attachment to Pets Scale (LAPS) [22] to examine levels of attachment foster and shelter volunteers report feeling towards partnered dogs. This can allow us to gain a better understanding of how volunteers perceive their relationships to dogs in these settings. In many cases volunteers are responsible for carrying out the majority of enrichment and socialization activities that dogs experience while living in an animal shelter, directly impacting the welfare of animals housed there. However, shelter staff and volunteers—especially those with animal contact—often experience burnout. Conflicted feelings about bonding with dogs under their care is one potential source of stress for these individuals, due to concerns about how the dog will feel when separated from them at the time of adoption. Shelter and foster volunteers may also miss animals they have bonded strongly with, and thus may experience feelings of loss even if they are happy a dog has been adopted. However, potential benefits and costs of the bonding experiences shared between shelter and foster volunteers and dogs in their care are not well understood.

Given that dogs living in animal shelters have been found to quickly form bonds to new humans [4], foster dogs are likely primed to form some kind of attachment to their new caretaker quickly as

well. However, the style of attachment developed depends on both foster volunteer and foster dog behavior [3,23]. The existing body of prior research on children and pet dogs suggests that secure attachment formation in the foster home could be beneficial to foster dog welfare, improve behavior outcomes and increase the speed and likelihood of secure bond formation with their new owner in the adoptive home [19–21]. More information about how attachment bonds within the foster home are associated with foster volunteer perception, foster dog welfare and future adoption success could be used to promote optimal fostering practices that take into account both dog and volunteer wellbeing. There is also a great need to evaluate the relative benefits of fostering, including the potential for stable bonding opportunities for stray and relinquished dogs, compared with other in-shelter socialization opportunities. In many cases, it is not feasible for shelters to foster their entire canine population, and therefore there is a critical need for an empirical investigation into how regular interactions with a familiar volunteer affect shelter dog welfare and adoption outcomes. However, to date, the potential benefits of regular interactions with a familiar volunteer on dog welfare, including the potential to positively impact the formation of future bonds with adopters, has not been evaluated in shelter dogs.

The goals of this project included identifying different volunteer–dog attachment profiles using data from a behavioral test and from a scale measuring volunteer-reported attachment levels with shelter and foster dogs. We compared relative preference for an unfamiliar person in a paired attachment test, and also analyzed these data in conjunction with volunteer-reported attachment levels. In addition, we explored whether behaviors associated with disinhibited attachment in humans were also present among the foster and shelter dogs that participated in this study. Given the role of secure attachment formation in terms of positive behavioral and cognitive outcomes in human children, we wanted to explore attachment relationships in foster and shelter dogs. As shelter dogs have been shown to form attachments to unfamiliar people quickly [4], we wanted to discover whether attachment relationships between foster and shelter volunteers and dogs in these settings are secure, and to what extent they are similar to attachments seen in pet dogs living in homes. This is the first study looking at the quality of attachment using attachment styles in foster and shelter dogs.

#### **2. Materials and Methods**

#### *2.1. Animal Subjects*

Foster dog subjects included 21 dogs living in foster homes with volunteers of Willamette Humane Society in Salem, Oregon and other local rescue groups, including Senior Dog Rescue of Oregon and Greenhill Humane Society in Eugene, Oregon. Shelter dog subjects consisted of 31 shelter dogs at Willamette Humane Society. All dogs were spayed or neutered prior to participation in the study. Shelter dogs were selected by volunteers, and foster dogs were assigned to foster homes by animal shelter and foster staff. All dogs were eligible for adoption at the time of participation in the study. See Table A1 in Appendix A for a description of all subjects. All procedures were approved by Oregon State University's institutional ethical review boards, animal related procedures were covered under OSU ACUP #4837.

#### *2.2. Human Participants*

Foster participants included 20 foster parent volunteers. Although all foster parent volunteers were invited to participate in a second round of testing, only one volunteer participated in a second testing session. (In some cases, testing was not possible because foster dogs were returned to the shelter before sessions could be conducted.) Shelter participants included 20 shelter volunteers that interact with dogs regularly as part of their volunteer duties. Twenty shelter-dog pairs took part in the first round of testing, and all volunteers were invited to participate in a second round of testing with a different dog. A total of 11 volunteers from round one participated in round two of testing. All analyses focusing on dog behavioral testing have been pooled. However, survey measures do not include pooled data to avoid partial dependence within the data set. Each participant provided

informed written consent to participate in the study. All procedures were approved by Oregon State University's institutional ethical review boards, human related data collection was covered under IRB #7818.

#### *2.3. Behavioral Tests and Surveys*

For dogs residing in foster homes, all testing was conducted at least three days after the dog entered the foster home. Participating shelter volunteers were asked to select a dog that they had interacted with for at least three separate ten-minute sessions, as this has previously been established as a sufficient amount of time for shelter dogs to establish attachment relationships with an unfamiliar person [4]. Testing sessions consisted of a Secure Base Test, designed to assess attachment relationships between dogs and familiar humans, immediately followed by a Paired Attachment Test, which aims to assess preference for a familiar vs. unfamiliar human. Tests were always conducted in this order. While both the Strange Situation Test (SST) and Secure Base Test have been validated for use with dogs, we chose to use the secure base test methodology (modeled after the first tests [24] designed to measure secure base and social preferences of this type) because it has several methodological advantages noted in the prior literature including reduced testing time, a reunion phase by the 'caretaker' that directly follows the alone phase, the elimination of order effects and the focus on the alone and reunion phase which have been found to produce the most reliable results in the human literature [25,26]. Following these behavioral tests, volunteers were asked to fill out a series of surveys, including the Lexington Attachment to Pets Scale.

#### 2.3.1. Secure Base Test (SBT)

All SBT sessions were conducted in a testing room unfamiliar to the dog. All testing sessions took place in a location that was unfamiliar to the dogs. In some cases, foster dogs were tested in the Oregon State Human-Animal Interaction Lab's on-campus testing space, but most dogs were tested at the Willamette Humane Society or Greenhill Humane Society in a novel testing room. Two chairs were placed in the room, and a semi-circle of 1 m in radius was taped on the floor around the chair prior to the beginning of the testing session. Three toys of different types were placed on the floor (outside of the 1-m radius circle) before volunteers and dogs entered the testing room. Toys included a tennis ball with a squeaker, a rope toy and a stuffed toy with a squeaker.

**Phase one (Baseline, two minutes):** The familiar volunteer was asked to sit neutrally in a chair within the 1 m radius circle. The volunteers were permitted to interact freely (petting, talking, etc.) with the dog (without restraining it) each time the dog placed at least two paws in the circle, but were instructed to sit neutrally if the dog exited the circle. Dogs were able to freely explore the room. The volunteer could play with toys with the dog if the dog brought them to the volunteer inside the circle.

**Phase two (Alone, two minutes):** The familiar volunteer or adopter exited the testing room so that the dog was left alone.

**Phase three (Return, two minutes):** The familiar volunteer or adopter re-entered the testing room and the instructions were identical to phase 1 (baseline).

#### 2.3.2. Video Analysis of SBT

Two independent coders reviewed the return phase videos for each dog's SBT and categorize dogs' attachment styles based on patterns of behavior seen in the return phase. A holistic analysis was used for these categorizations (see [5–7]). Inter-rater reliability was assessed based on the percentage of independent agreement after this initial round of coding. After the two coders reviewed each video independently, they watched any videos for which they disagreed on attachment style categorization together and reached an agreement. A description of all attachment style classifications can be found in Table 1. Degree of independent agreement among coders for attachment style was 72%, and a consensus was reached for all dogs when coders reviewed videos together.



#### 2.3.3. Paired Attachment Test

The Paired Attachment test included the following phases:

#### **Phase 1 (two minutes):** A two-minute alone period immediately following the SBT.

**Phase 2 (Passive, two-minute phase)**: The dog's caretaker and a stranger sat neutrally for two minutes in chairs opposite each other surrounded by a 1m radius circle. Each individual was instructed to pet the dog twice each time it entered the circle with at least two paws, but were instructed to otherwise remain neutral.

**Phase 3 (Active, two minutes):** Both humans were asked to call the dog and provide continuous petting and attention if the dog entered their circle with at least two paws.

#### **Video analysis of Paired Attachment Test**

All videos were coded across phases for first person approached (unfamiliar or familiar), duration of proximity seeking with each person, and duration of contact with each person. An ethogram can be found in Table 2. Please see Figure 1 for a picture of the Paired Attachment test set-up.

**Table 2.** Paired Attachment ethogram.


**Figure 1.** Paired attachment test set-up.

## 2.3.4. Disinhibited Attachment Coding

Based on the methods used in [9], we developed a scale to assess disinhibited attachment using rankings of different measures that were combined into a composite score. We assigned rankings to each dog based on the proportion of the return phase of the SBT spent seeking proximity to the familiar person. Because severe disinhibited attachment is associated with a lack of proximity seeking with a familiar person, the highest proportion of time spent in proximity to the familiar person received the lowest rankings. Inter-rater reliability for this measure was 75%. In addition, we assigned rankings based on total amount of time spent in proximity (within the 1-m radius circle taped on the floor around the chair) to the unfamiliar person across both phases of the Paired Attachment Test. Inter-rater reliability for total proportion of time spent in proximity to the unfamiliar person was 93.8%. Low proximity seeking with the unfamiliar person received low rankings on the scale; higher proportions of time spent proximity seeking received higher rankings. Across all measures, in the case of a tie, those values were assigned the same rank. For instance, if two dogs spent 98.3% of the session seeking proximity to the familiar person, they would receive the same rank. Both scores were summed to create an overall disinhibition score.

#### *2.4. Statistical Methods*

All statistical analyses were conducted using R Studio (version 1.1.463). All statistics were two-tailed with an alpha level of *p* < 0.05.

#### 2.4.1. Attachment Analysis

A Fisher's Exact Test was used to compare proportions of securely and insecurely attached dogs within foster and shelter groups. A Chi Square test was used to compare the proportions of attachment styles seen in the foster and shelter populations to expected frequencies of proportions of pet dogs based on all published literature categorizing attachment styles in pet dogs [5–7]. A McNemar's Test was used to compare dogs' attachment styles for shelter volunteers who participated in two separate rounds of testing.

#### 2.4.2. Paired Attachment Analysis

Normality was assessed using the Shapiro–Wilk Test. For shelter and foster groups, all proximity and contact data were not normally distributed (*p* < 0.05). For both foster and shelter groups, a Kruskal–Wallis test was used to assess whether differences in proximity or contact seeking were present with respect to whether the humans were passively interacting with the dogs (two pets every time the dog enters each person's respective circle) or actively encouraging interaction from the dogs. Post hoc comparisons were made using Wilcoxon Signed-Rank Tests for within-subject comparisons across phases. Fisher's Exact tests were used to determine whether the first person approached (unfamiliar or familiar) varied according to dog source (shelter or foster) or attachment style (insecure or secure) for active and passive phases. Mann–Whitney *U*-tests were used to compare total amount of time spent with the familiar person vs. the unfamiliar person across both passive and active phases for both populations (shelter and foster) and attachment style (secure vs. insecure). A difference score was calculated to assess overall preference for a familiar human vs. an unfamiliar human by subtracting the total proportion of time spent with an unfamiliar person across active and passive phases from the total proportion of time spent with a familiar person across active and passive phases, and these data were not normal, *p* < 0.05. See Table 3 for inter-rater reliability scores for all measures. Percent agreement between two coders was assessed using an 8% tolerance.

**Table 3.** Inter-rater reliability for all Paired Attachment measures.


#### 2.4.3. Disinhibited Attachment Analysis

Normality of disinhibited attachment scores was assessed using the Shapiro–Wilk test, *p* > 0.05. *T*-tests were used to compare disinhibition scores across four categories: insecure shelter dogs, secure shelter dogs, insecure foster dogs, and secure foster dogs. We also used Mann–Whitney *U*-tests for

non-normally distributed data for proportion of time spent in proximity to familiar person for analyses by group and by attachment style.

#### 2.4.4. Lexington Attachment to Pets Scale (LAPS) Analysis

Normality was assessed using the Shapiro–Wilk test and LAPS data were normal, *p* > 0.05. Because assumptions of normality were not violated, parametric statistics were used to compare between groups. A two-way ANOVA with attachment category and environment (foster vs. shelter) and possible interactions was used to analyze LAPS data. A Pearson's Correlation was used to determine whether there was a relationship between LAPS score and overall preference score for a familiar vs. unfamiliar person. Only data from the first round of shelter and foster volunteer participation were used for these comparisons, in order to avoid any confounding effects of volunteers who participated in both rounds. A Pearson's Correlation was used to assess whether scores from shelter volunteers' first participation were related to LAPS scores from shelter volunteers' second participation.

#### **3. Results**

#### *3.1. Attachment*

Within the foster group, a total of twelve dogs (57.14%) were categorized as secure and nine dogs (42.86%) were categorized as insecure (eight dogs (38.10%) were categorized as insecure-ambivalent, and one dog (4.76%) was categorized as insecure-disorganized). Within the shelter group, a total of twelve dogs (38.71%) were categorized as secure and a total of nineteen dogs (61.29%) were categorized as insecure (sixteen shelter dogs were scored as insecure ambivalent (51.61%), and three insecure shelter dogs (9.68%) were scored as insecure avoidant). The Fisher's Exact test comparing the proportion of insecure and secure dogs in the shelter and foster groups was not significant (*p* = 0.26). See Figure 2 for a comparison of attachment styles in each population.

**Figure 2.** Proportion of dogs categorized into each attachment style for foster and shelter dog populations.

To obtain an overall picture of how each population of foster and shelter dogs, respectively, compared to pet dogs in terms of proportions of attachment styles, we summed data from all published literature involving categorization of attachment styles in pet dogs for each category of attachment styles (secure, insecure-ambivalent, insecure-avoidant, and insecure-disorganized) [5–7]. Only data from dogs in the saline condition were used for [6], a study which included a counterbalanced repeated measures design in which oxytocin was administered during one testing session and saline upon

another visit prior to participating in the secure base test, to avoid dependence (i.e., if we did not exclude the oxytocin sessions, we would be using data for the same dogs twice) and any effect of oxytocin on behavior. Across previously published studies, 68% of pet dogs had been categorized as having a secure attachment to their primary caretaker and 32% of pet dogs had been categorized as displaying a type of insecure attachment. The proportion of secure shelter dogs significantly differed from what was expected when compared against previously published attachment outcomes in pet dogs, χ<sup>2</sup> (1, *N* = 31) = 12.22, *p* = 0.0005 (Figure 3). No significant differences were found when proportions of attachment styles for the foster dog group were compared to pet dog attachment style proportions, *p* > 0.05 (Figure 3).

**Figure 3.** Proportions of observed foster and shelter dogs categorized as secure, compared to expected proportions of pet dogs with secure attachments based on published literature [5–7].

No significant differences were found in terms of comparisons of attachment styles among dog–volunteer dyads for shelter volunteers who participated in two rounds of testing with two different dogs, *p* = 0.62.

#### *3.2. Paired Attachment*

A trend was found for proportion of time shelter dogs spent seeking proximity to both the familiar and unfamiliar person when all conditions (familiar passive, familiar active, unfamiliar passive, unfamiliar active) were compared to each other, H (3) = 7.80, *p* = 0.05. A significant difference was found with respect to the proportion of time spent in contact across all conditions, H (3) = 23.103, *<sup>p</sup>* <sup>=</sup> 3.84 <sup>×</sup> <sup>10</sup><sup>−</sup>5. This difference was driven by the finding that shelter dogs spent significantly more time in contact with both the familiar person (W = 456, *p* < 0.001) and the unfamiliar person (W = 68, *p* = 0.001) in the active phase of the sociability test when compared to the passive phase. No significant differences were found in terms of proportion of time spent in contact with the unfamiliar person vs. the familiar person (*p* > 0.05 for both the passive and active phases).

For the foster group, a significant effect was found with respect to the proportion of time spent in proximity across all conditions, H (3) = 11.49, *p* = 0.01. A trend was present with respect to the proportion of time foster dogs spent in proximity to the familiar person compared to the unfamiliar person in the active phase, W = 165, *p* = 0.09. The median proportion of time foster dogs spent with the familiar person in the active phase was 0.52, and the median proportion of time spent in proximity to the unfamiliar person in the active phase was 0.28. A trend was also found within the foster group with respect to proportion of time spent in contact H (3) = 9.23, *p* = 0.03. After Bonferroni correction, the significance threshold for the Kruskal–Wallis test was 0.0257.

Based on the overall score for time spent with the familiar person vs. time spent with the unfamiliar person across phases, no significant differences were found with respect to group (foster vs. shelter) or attachment style (secure vs. insecure). In addition, we analyzed whether differences were present with respect to overall time spent in proximity to the familiar person compared to the unfamiliar person, and no significant differences were found with respect to group or attachment style, *p* > 0.05. No significant differences were found with respect to first person approached for active or passive phases when shelter and foster dogs were compared across groups, *p* > 0.05.

#### *3.3. Disinhibited Attachment*

Insecure foster dogs had the lowest disinhibited attachment rankings, with a mean of 22.0. Insecure shelter dogs had the second lowest scores on average with a mean score of 36.42. The mean score for the secure foster group was 47.17 and the mean score for the secure shelter group was 48.75. Insecure foster dogs displayed significantly lower mean disinhibition scores than secure foster dogs, *t*(18.99) = −3.5499, *p* = 0.002. Insecure foster dogs also displayed significantly lower mean scores of disinhibited attachment compared to secure shelter dogs, *t*(16.33) = −4.64, *p* = 0.0003. Furthermore, insecure foster dogs scored significantly lower on disinhibition than insecure shelter dogs, *t*(21.57) = −2.2835, *p* = 0.03. In addition, insecure shelter dogs displayed significantly lower disinhibited attachment scores than secure shelter dogs, *t*(28.97) = −2.184, *p* = 0.04. No significant differences were present with respect to insecure shelter dogs and secure foster dogs, *p* > 0.05. In addition, we did not find significant differences between secure foster dogs and secure shelter dogs on disinhibition, *p* > 0.05. See Figure 4 for mean scores on disinhibited attachment among foster and shelter dogs with secure and insecure attachments.

**Figure 4.** Mean disinhibited attachment scores for foster and shelter dogs with insecure and secure attachment styles. Error bars indicate standard error of the mean.

#### *3.4. Lexington Attachment to Pets Scale (LAPS)*

In terms of the results of the two-way ANOVA analysis of LAPS scores, no significant main effect or interaction effect was found, F (1, 36) = 0.20, *p* > 0.05 for attachment style categorization or group (shelter/foster). The average LAPS score for shelter volunteers was 15.45 and the average LAPS score for foster volunteers was 20.40. This difference was not statistically significant, *t*(37.96) = 1.11, *p* > 0.05. The average LAPS score for insecure dogs was 18.60 and the average LAPS score for secure dogs was 17.25. This difference was not statistically significant, *t*(37.16) = 1.11, *p* > 0.05. Based on comparisons between first participation and second participation for the 11 volunteers that participated twice, there was a positive correlation between LAPS scores on each round, r = 0.62, *p* = 0.04. With respect to the familiar person in the active phase, there was a significant positive correlation between the amount of time spent proximity seeking and the LAPS score, r = 0.35, *p* = 0.03. There was not a significant correlation between LAPS scores and the amount of time spent seeking proximity with the familiar person in the passive phase, *p* > 0.05.

There was a significant positive correlation between a dog's preference score (calculated by subtracting the total proportion of time spent with an unfamiliar person across active and passive phases from the total proportion of time spent with a familiar person across active and passive phases) for the familiar person compared to an unfamiliar person on the Paired Attachment Test and the strength of attachment reported by the caretaker/familiar person on the LAPS survey, r = 0.34, *p* = 0.03 (Figure 5).

**Figure 5.** Scatterplot for overall score for preference for a familiar human compared to an unfamiliar human compared to LAPS scores. Positive preference scores indicate a preference for the familiar human; negative preference scores indicate a preference for the unfamiliar human. Higher LAPS scores indicate a stronger degree of attachment; lower LAPS scores indicate a weaker degree of attachment, as reported by familiar volunteers. This suggests that there is a relationship between the strength of attachment human volunteers feel for a dog and the amount of preferential proximity seeking that dog displays in an attachment, however the direction and causality of the relationship is unknown.

#### **4. Discussion**

To our knowledge, this is the first application of attachment style categorization in human–dog relationships in foster and shelter settings. We explored the proportions of attachment relationships seen among foster and shelter volunteers and dogs in foster and shelter settings, and also explored disinhibited attachment in these populations. Our findings indicate that the proportion of secure attachment styles in shelter dogs included in this study were significantly lower than the proportion of secure attachment styles previously reported for pet dogs. Conversely, foster dogs formed secure attachments to their caretakers at rates more similar to those reported in dog–owner relationships, although no significant differences were found between proportions of attachment styles when comparing between foster and shelter dogs directly [5–7].

The finding that shelter dogs are significantly more likely than pet dogs (based on data from previously published literature) to form insecure attachments to their temporary caretakers is of note, especially given the high rate of insecure ambivalent attachments observed. This attachment style is associated with excessive proximity seeking upon return, even though this contact is less effective at reducing stress than in secure attachment relationships. Given that previous research has shown that lying in proximity to adopters positively influenced dog adopter decisions in a shelter setting [27], it is possible that attachment styles aligned with greater proximity seeking could have some benefit in this environment. For example, dogs behaving ambivalently might be perceived as more social and therefore be more attractive to potential adopters. However, more research should be done before conclusions can be drawn. Future research could explore whether attachment styles formed with familiar volunteers in the SBT corresponds to behavior in adopter-dog interactions at the shelter. Future studies could also explore attachment relationships between dogs that have been rehomed from foster and shelter settings and their adoptive owners, as this has not been previously explored.

We also found evidence of disinhibited attachment among dogs in foster homes and shelter settings. We expected to see higher levels of disinhibition among dogs displaying secure attachments within these settings (as that would be consistent with the human literature), and indeed, the highest levels of disinhibition were exhibited among secure shelter and foster dogs and the lowest levels among insecure shelter and foster dogs. Previous work has shown that disinhibited attachment was also present in human orphans from Romanian orphanages who had experienced institutional deprivation and children adopted in the UK who had no exposure to institutional deprivation [9]. Thus, disinhibited attachment could be a product of the temporary nature of relationships with caregivers, traumatic experiences early in life associated with becoming an orphan, or other factors beyond the quality of the environment. Shelter and foster dogs with secure attachments may show higher levels of disinhibition because it allows for greater social flexibility within these environments, which could provide advantages in terms of greater likelihood of being taken for walks, greater chances of interaction with potential adopters, or more social attention from volunteers. Experiences prior to placement in a foster home or a shelter could also contribute to the development of disinhibited attachment, and further research is needed in this area to better understand the development and implications of disinhibited attachment in foster and shelter dogs. It should also be noted that the dogs in the shelter environment receive a great deal of enrichment and social interaction, including play groups with conspecifics, training opportunities, and several walks each day. Future studies should assess disinhibited attachment in kenneled dogs that do not receive as many opportunities for enrichment and social exposure to both humans and conspecifics. We also found that insecure dogs spent a greater proportion of time in proximity to the familiar person during the SBT than secure dogs. Although disinhibited attachment has not previously been reported in canine attachment literature, it appears to be relevant for dogs in shelter and foster settings. To date no research has been done on disinhibition in pet dogs, so it is unknown to what degree these results may differ from pet dog populations, but such comparisons merit further investigation. Furthermore, additional aspects of disinhibition could be explored. For example, in the current study, we did not look at the dog's willingness to leave with the stranger. This would be an interesting additional condition for future studies, since it is one trait often noted in human children exhibiting disinhibited attachment disorders.

The results of the paired attachment test suggest that shelter dogs are more likely to seek contact depending on attentional state (i.e., whether the person is actively interacting with the dog/encouraging attention or sitting passively and petting the dog twice each time it enters the 1m radius circle surrounding the person's chair), while foster dogs are more likely to seek proximity regardless of whether the familiar volunteer or unfamiliar person are actively attending to the dog. We found that shelter dogs were attentive to the attentional state of the person to a greater degree than foster dogs. It is possible that although foster homes may allow for an easier transition to an adoptive home environment, that dogs living in foster homes may be less attuned to human attentional state. Shelter dogs, on the other hand, typically spend more time in social isolation than foster dogs, and paying close attention to the attentional state of humans may provide benefits, such as increased socialization, exercise, or interactions with adopters. Furthermore, no significant differences were

found with respect to contact or proximity seeking when comparisons were made based on attachment style. It should also be noted that no differences were found with respect to the first person approached by either shelter or foster dogs. Overall, no differences were found between foster and shelter dogs based on preference for an unfamiliar or a familiar person in the paired attachment test. We did not find any significant differences in terms of attachment style and preference for an unfamiliar person compared to a familiar person. This indicates that both shelter and foster dogs show flexibility in terms of social interaction.

We found a significant correlation between LAPS scores, which measured the attachment strength of shelter and foster human volunteers, and a dog's preference for that volunteer versus an unfamiliar human in the Paired Attachment test. This indicates that familiar volunteers whose dogs exhibited a stronger preference for them (i.e., dogs who spent more time in proximity to the familiar person across phases), also reported a stronger attachment to their chosen dog (or vice versa). While the causality of this relationship cannot be determined from the current data, this connection warrants further investigation in the future, although regardless of causality, this finding is interesting as it indicates reciprocity in the attachment relationship.

The reported strength of attachment by volunteers (LAPS scores) did not differ between shelter and foster volunteers. Although we expected that foster volunteers would report a greater degree of attachment to foster dogs than shelter volunteers to shelter dogs, this was not the case. Thus, it is possible that they were primed to think of themselves as having a stronger attachment to the dog that they chose for the study than they would have naturally. Future studies could examine LAPS ratings for shelter volunteers more generally, to determine if the level of attachment reported towards dogs in this study was unique (and possibly due to the experimental context) or if shelter volunteers commonly feel highly bonded to shelter dogs. Another possibility is that foster volunteers may try to avoid forming strong attachment bonds with foster dogs, to protect their own emotions or to protect the dog from perceived feelings of loss when eventually rehomed. Such strategies might also be used to reduce temptation to adopt the foster dog. Future research could measure LAPS scores in relation to several foster dogs, including foster volunteers who did and did not adopt their foster dogs, in order to determine whether strength of attachment, as reported by the person, plays a role in decision-making regarding adoption of foster dogs by their foster parent volunteers. It is also of note that the correlation between LAPS scores and proximity seeking towards volunteers appears to be driven by the active phase of the test, where the volunteer can call and interact with the dog. This suggests that the attachment strength and behavior of the familiar person may be influencing the dogs' attachment behavior, but more studies are needed to be sure. While future research should directly evaluate long term outcomes of foster and shelter dogs who formed secure and insecure attachment bonds to transitional caretakers, in the human literature, orphaned children who develop secure bonds to a foster parent are more likely to form secure social bonds later in life [20]. If this proves to be true of dogs as well, this would suggest that fostering may provide an additional benefit by increasing the likelihood of a dog establishing a secure attachment prior to final adoption.

It should be noted that differences were not found with respect to attachment styles of dogs for volunteers who participated in the first and second round of testing. However, only eleven shelter volunteers participated in the first and second round of testing, resulting in a relatively small sample size. For six of eleven participants, the dogs that they participated in the SBT with in the first round were insecurely attached. Based on these data, we cannot conclude whether the attachment styles dogs presented with in the SBT were due to the nature of their interactions with the volunteers, or due to personality traits of the dogs, or a combination of these factors. It would also be interesting to evaluate whether volunteer personality plays a role in dog selection. For instance, do volunteers with certain personality traits, such as higher anxiety levels, choose more anxious dogs? Research with owners of pet dogs has shown that attachment avoidance in people on the Adult Attachment Scale is associated with increased occurrence of separation anxiety in their dogs [28].

Another finding with respect to LAPS scores was that no differences were present with respect to attachment security. It is possible that volunteers are not attuned to attachment style-related behaviors, and therefore evaluate attachment to shelter and foster dogs based on other observations. It is also possible that some volunteers prefer dogs who are securely attached and others prefer dogs who are insecurely attached. Future research could evaluate the attachment styles of dogs owned by volunteers in conjunction with SBT results between volunteers and dogs in foster and shelter settings, in order to gain a better understanding of the role of the human in forming and maintaining attachment relationships in these contexts. Further research is needed to determine whether LAPS scores correspond to attachment style. Also, of note is that even with a relatively small sample size of 11 volunteers who participated in both rounds of testing in the shelter setting, there was a trend of a significant correlation between LAPS scores in each round. This suggests that volunteer personality and individual framework for attachment relationships may play a role in the formation and maintenance of social bonds between volunteers and dogs in animal shelters. It is also important to note that volunteers have many opportunities to interact with a variety of dogs, as dogs are typically adopted out relatively quickly from the shelter at which this study was conducted, and thus, it is remarkable that volunteers form strong attachment bonds to dogs over a relatively short time period. It is also possible that volunteers with lower LAPS scores may attempt to avoid developing attachment bonds to individual dogs at the shelter, and this could be a coping mechanism to help prevent compassion fatigue.

#### **5. Conclusions**

We have shown that shelter dogs differ significantly from a meta-analysis of pet dogs with respect to proportions of attachment styles, while no differences were found between attachment proportions of shelter and foster dogs. This provides evidence that social relationships formed between foster parent volunteers and foster dogs in the foster home may be more similar in nature to those formed in the typical home environment of pet dogs. Therefore, fostering dogs may provide an important additional benefit, by increasing the likelihood that a dog will experience the establishment of a secure attachment to a caretaker before final adoption. In studies with human children, this scenario not only allowed for better coping with stress in the short term, but also increased the likelihood that the child would develop a secure attachment to their caretaker in their final home upon being adopted. More research is needed to evaluate if the same is true for adopted foster dogs. Additionally, disinhibited attachment, characterized by excessive friendliness towards unfamiliar people, lack of discrimination among adults based on social familiarity, and lack of checking in with an attachment figure during a stressful situation in humans, is present in shelter and foster dogs. We also found preliminary evidence that would suggest a relationship between how attached a human volunteer feels towards a dog in their care and the dog's behavior towards that person, however more research is needed to determine if a causal relationship exists, and if so, the direction of that relationship.

**Author Contributions:** Conceptualization, L.E.T. and M.A.R.U.; methodology L.E.T. and M.A.R.U.; performed experiments and statistical analyses, L.E.T.; writing—original draft preparation, L.E.T.; writing—review and editing L.E.T. and M.A.R.U.; funding acquisition, L.E.T. and M.A.R.U. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by Maddie's Fund.

**Acknowledgments:** Thank you to Maddie's Fund for funding this study. We are grateful to the volunteers, shelter staff, and dogs who participated in this research. Thank you to Willamette Humane Society, Greenhill Humane Society, and Senior Dog Rescue of Oregon for assistance with participant recruitment and the use of facilities for data collection. Thank you in particular to Betsy Bode and Janine Catalino for their help with coordinating data collection. This project would not have been possible without the help of several research assistants in the Human-Animal Interaction Lab, particularly Holly Duvall and Hadley Schoderbeck. We are grateful to Giovanna Rosenlicht, Alexandra Protopopova and Frank Bernieri for their advice during study design and helpful comments on a previous version of this manuscript.

**Conflicts of Interest:** The authors declare no conflicts of interest.

#### **Appendix A**

**Table A1.** Subjects' demographic information, including names, groups, ages, modes of intake, length of stay (LOS), and breed (as listed by their associated shelter or rescue group). Several of the dogs in the study were returned; length of stay was calculated based on initial intake date and final adoption date. All dogs were spayed and neutered prior to participating in the study. Length of stay data were not available for some dogs that participated, and in a few cases (noted below), dogs were still available for adoption at the completion of the study. All breed determinations were based on visual inspection by the participating shelters and rescue groups, and it is important to note that reliability for this method is low [29].


### **References**


© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

## *Article* **Sex of Walker Influences Scent-marking Behavior of Shelter Dogs**

## **Betty McGuire 1,\*, Kentner Fry 1, Destiny Orantes 1, Logan Underkofler <sup>2</sup> and Stephen Parry <sup>3</sup>**


Received: 13 March 2020; Accepted: 3 April 2020; Published: 7 April 2020

**Simple Summary:** In diverse settings, human presence and handling influence the behavior and physiology of other animals, often causing increased vigilance and stress, especially if the human is unfamiliar. Domestic dogs are unusual in that human interaction often reduces stress and behavioral signs of stress. Nevertheless, there is some evidence that the sex of an unfamiliar person can influence canine behavior. To determine whether sex of an unfamiliar walker might influence the behavior of dogs at an animal shelter, we observed 100 dogs during leash walks and recorded all occurrences of scent-marking behaviors. Male dogs urinated at higher rates when walked by unfamiliar women than when walked by unfamiliar men. Female dogs urinated at similar rates when walked by unfamiliar men and unfamiliar women. Sex of walker also influenced urinary posture in male dogs. Both male and female dogs were more likely to defecate when walked by unfamiliar women than when walked by unfamiliar men. Based on our findings, and those of others, we suggest that the sex of all observers and handlers be reported in behavioral studies of dogs and considered in behavioral evaluations at animal shelters, where results can impact whether or not a dog is made available for adoption.

**Abstract:** Interactions with humans influence the behavior and physiology of other animals, and the response can vary with sex and familiarity. Dogs in animal shelters face challenging conditions and although contact with humans typically reduces stress and behaviors associated with stress, evidence indicates that shelter dogs react differently to unfamiliar men and women. Given that some aspects of canine scent-marking behavior change under fearful conditions, we examined whether sex of an unfamiliar walker would influence scent-marking behavior of 100 shelter dogs during leash walks. Male dogs urinated at higher rates when walked by unfamiliar women than when walked by unfamiliar men; female dogs urinated at similar rates when walked by unfamiliar women and unfamiliar men. Sex of walker influenced urinary posture in male dogs, but not in female dogs. Both male and female dogs were more likely to defecate when walked by unfamiliar women than by unfamiliar men. Based on our findings that shelter dogs behave differently in the presence of unfamiliar men and women, we suggest that researchers conducting behavioral studies of dogs record, consider in analyses, and report the sex of observers and handlers as standard practice. We also recommend recording the sex of shelter staff present at behavioral evaluations because the results of these evaluations can impact dog welfare.

**Keywords:** dog; scent marking; urination; urinary posture; defecation; ground scratching; animal shelter; human-animal interactions

#### **1. Introduction**

Human presence and handling can affect the behavior and physiology of other animals, including species living in the wild [1–4] and in captive settings, such as zoos [5,6], farms [7–9], and research laboratories [10]. Such effects often depend on the number of people present as well as their distance, behavior, and familiarity [11–15]. Additionally, human physical characteristics, including sex and age, can influence the behavior and physiology of other animals [10,16]. Laboratory rats and mice discriminate human sex using olfactory stimuli [10] and free-ranging elephants discriminate human sex and age using acoustic cues in voices [16]. In many interactions between humans and other animals, humans are perceived as either predators or at least as something to be feared [16,17], thus, human presence often causes increased vigilance, avoidance, and stress.

Domestic dogs present a somewhat special case in which contact with humans typically reduces both stress and the performance of behaviors associated with stress [18–26]. This has been shown in animal shelters where dogs experience challenging conditions, such as isolation, lack of control, and exposure to unfamiliar people, dogs, and surroundings. For shelter dogs, various forms of physical contact with humans (e.g., petting, massaging, and grooming) and different types of interactions with humans (e.g., walks, play sessions, training sessions, and simply having a person sit passively in the same enclosure), have been shown to reduce physiological measures of stress [18–20], produce favorable changes in behavior [21], or both [22–26].

Despite the general pattern that human contact has positive effects on shelter dogs, there is evidence that dogs respond differently to men and women. Shelter dogs enrolled in a human interaction program improved in their sociability toward unfamiliar women but not toward unfamiliar men [26]. When an unfamiliar man or woman stood in front of cages for a few minutes, shelter dogs decreased to a greater extent the time they spent barking and looking at the person when the unfamiliar person was female [27]. An initial report that shelter dogs petted by females had lower cortisol levels than those petted by males [28] was later found to reflect subtle differences in petting techniques of males and females: when men and women received specific training to standardize petting techniques, male and female petters reduced cortisol levels to similar degrees [20,29]. Differential responses to men and women also have been documented for dogs in settings other than animal shelters. For example, dogs in a guide dog training program made more frequent contact with unfamiliar women than with unfamiliar men [30] and during agility competitions, dogs with male handlers experienced greater increases in cortisol than did dogs with female handlers [31]. Finally, from a study in a commercial kennel, male dogs spent less time near an unfamiliar man than an unfamiliar woman, whereas female dogs spent equal amounts of time near an unfamiliar man and an unfamiliar woman; a similar pattern occurred for the frequency of direct body contact [32]. Although the responses of dogs to male and female humans have been studied in diverse settings and ways, we could find no information on how sex of an unfamiliar walker might influence the behavior of dogs during leash walks.

In the present study, we examined whether sex of an unfamiliar walker influenced scent-marking behavior of mature male and female shelter dogs during walks on a leash. Leash walking is commonly used by shelters to provide dogs with opportunities to exercise, socialize with humans, and perform species-typical behaviors, such as sniffing and urine-marking. At least one aspect of canine scent-marking behavior is sensitive to fearful or stressful conditions: adult male dogs that used the raised-leg urinary posture typical of mature males temporarily reverted in fearful situations to using the juvenile lean-forward posture in which all four feet remain on the ground [33,34]. Consistent with this finding, we reported that the percent of urinations in which adult male dogs used the raised-leg posture was lower in our study shelter (73%; [35]) than reported for mature male dogs living under other conditions (94%–97%; [36–39]). We found a similar effect for female dogs: 6% of urinations by adult females involved raising a hindlimb at our study shelter [35] compared with 19%–37% for adult female dogs living under other conditions [36–39]. These observations suggest that monitoring scent-marking behavior of dogs during walks might be a useful way to assess how shelter dogs respond to the sex of an unfamiliar walker. Given that dogs generally respond less favorably to unfamiliar men than unfamiliar women [26,27,30] and that this response can be stronger in male dogs [32], we predicted that the frequency of scent marking behaviors would be lower when dogs were walked by an unfamiliar male than by an unfamiliar female, and that such reductions would be more dramatic in male dogs than in female dogs. We predicted reductions in scent marking behaviors of mature dogs walked by unfamiliar men because mature male dogs reverted to using the juvenile urinary posture in fearful situations [33,34] and less frequent urination, defecation, and ground scratching represent the pattern of scent marking shown by juvenile shelter dogs during leash walks [40]. Finally, we predicted that dogs walked by an unfamiliar male would be more likely to use postures in which all feet remain on the ground (i.e., the lean-forward posture in males and the squat posture in females).

#### **2. Materials and Methods**

The data presented here were collected at the Tompkins County SPCA in Ithaca NY, USA, between September 2017 and December 2019, as part of a long-term research program on scent-marking behavior of shelter dogs. Tompkins County SPCA is a no-kill shelter with open-admission and scheduled intake. The shelter has very active volunteer programs for both cats and dogs. Dog volunteers must be at least 18 years old and can be either canine companions or dog walkers. Canine companions help socialize and train dogs in their cubicles, sit with them, and pet and groom them. Dog walkers take the dogs out for walks or to a large outdoor enclosure (Section 2.1). As time permits, volunteer dog walkers sometimes engage in canine companion activities as well. A one-time snapshot of dog volunteers at the end of our study showed that 71% (58/82) were women and 29% (24/82) were men. Although the numbers of male and female staff members are complicated by the variation in the extent of direct interaction with the dogs and the needs of individual dogs (e.g., some dogs may have extensive interactions with Medical Staff, whereas others have much less), staff was also female-biased in most positions over the course of our study (e.g., Animal Care Technicians, 10 females and one or two males; Medical Staff, all females except for one male intern in the past 6 months; Adoptions, Intake, and Behavior Program, approximately equal number of females and males).

#### *2.1. Dogs and Housing*

We observed 100 dogs (57 males and 43 females) that were at least one year old (Mean ± *SD*, 4.4 ± 3.4 years; range, 1–17 years). Housing and care of dogs have been described elsewhere [41], thus, we provide a brief description here. Most dogs in our study were mixed breeds. We did not have access to DNA analyses or pedigrees; thus, the number of purebred dogs is unknown. Dogs were either surrendered by owners (*n* = 44), transferred from other shelters (*n* = 24), picked up as strays (*n* = 18), or returned by adopters (*n* = 12); two dogs were seized by animal control officers. All dogs received veterinary care at intake (e.g., vaccinations, flea control, fecal exam and deworming, heartworm test, and any additional diagnostic tests deemed necessary). Dogs without a microchip received one. Screening blood work, including complete blood count/chemistry profile, was routinely run for older dogs. If owners provided information about urinary issues at the time of surrendering their dog to the shelter or if symptoms of disease were observed in the shelter (e.g., frequent urination), then urinalysis was performed for dogs of any age. We excluded from our study dogs with known medical issues. About 3 days after intake, dogs underwent behavioral evaluation by Behavior Program staff [42,43]. All dogs had received veterinary care, undergone behavioral evaluation, and were on the adoption floor by the time we walked them. Dogs on the adoption floor wore buckle or martingale collars and were individually housed in one of 13 cubicles (from 5.2 m<sup>2</sup> to 7.3 m2). Each dog had a water bowl, raised bed with blanket, and toys. Staff fed dogs between 08:00 and 09:00 h and between 15:00 and 16:00 h; additionally, a pre-measured bag of small treats was available for each dog each day. Shelter staff or volunteers either walked dogs or brought them to a large outdoor enclosure several times a day. Each day, the start time and end time of each walk or time in the outdoor enclosure were recorded on a dry erase board in the dog wing.

At the shelter, most dogs are spayed or neutered before placement on the adoption floor; all are spayed or neutered before adoption. In research previously conducted at this shelter, one of us (B.M.) found that rates of urination during walks decreased after castration in males but did not change after spaying in females (within-dog study; [44]). Similarly, intact males urinated at higher rates than castrated males, but intact and spayed females urinated at similar rates (between-dog study; [44]). Gonadectomy did not influence likelihood of defecation or ground scratching during walks in either males or females (between-dog study; [44]). Given the effect of reproductive condition on the rate of urination in male dogs, it was essential that we control for reproductive condition within each male dog when walked by male versus female walkers. Of the 57 male dogs that we observed, 56 were neutered for all of their walks and one was intact for all of his walks. Of the 43 female dogs that we observed, 36 were spayed for all of their observations, four were intact for all of their observations, and three were intact for some observations and spayed for others.

#### *2.2. Behavioral Observations*

Behavioral observations occurred during walks, which began on shelter grounds and continued into a large field across the street (16.6 ha; 42◦28'20"N, 76◦26'22"W). The field was bordered by a creek, forest, and other fields of very tall grass; the substrate where we walked was mostly grass, which was occasionally mowed in spring and summer. All procedures were carried out under protocol 2012-0150, which was approved by Cornell University's Institutional Animal Care and Use Committee.

Over the course of the study, five different walkers (two females, B.M. and D.O., and three males, K.F., L.U., and J.C.) conducted behavioral observations during individual first walks of dogs (i.e., only the walker was present with the dog and this was the first time that person had walked the dog). All walkers had extensive experience handling and observing dogs on walks, gained via research activities at the shelter, long-term dog-walking as a volunteer at the shelter, or independent employment as a dog walker. Individual walkers conducted observations between 12:00 and 17:00 h, typically once or twice a week, on days that were convenient for them. All dogs included in the data set were individually walked by at least one male and one female walker (one male walker and one female walker: 50 dogs; one male walker and two female walkers: 30 dogs; two male walkers and one female walker: 10 dogs; two male walkers and two female walkers: 10 dogs). Dogs were adopted throughout our study, which is why the number of times a given dog was walked varied from two (one male walker and one female walker) to four (two male walkers and two female walkers). Records of specific staff or volunteers who had walked each dog prior to our walks were not available.

On each walking day, a walker checked the dry erase board in the dog wing and selected dogs he or she had never walked before and that had not been outside for at least 2 h. Scheduled dog walking shifts at the shelter occur at 12:00, 14:30, and 17:00 h, thus, dogs included in our study were walked approximately 2–3 h after their previous walk. Once a team member had walked a specific dog for the first time, B.M. alerted other team members to prioritize that dog for walking. We used leashes and harnesses provided by the shelter; staff had previously fitted each dog with an appropriate harness (either a PetSafe Easy Walk Harness, Radio Systems Corporation, Knoxville, TN, USA or a Zack and Zoey Nylon Pet Harness, Pet Any Way LLC, model US2395 14 99) and placed the harness and a cloth lead (at least 1.8 m long) on a hook outside the dog's cubicle. Upon entering a dog's cubicle, each walker harnessed the dog, attached the lead and led the dog out of the shelter. Behavioral observations began once the dog was outside and lasted for 20 min, during which time we let dogs determine the pace of the walk (dogs were not kept in a heel position). Per shelter policy, dogs were not allowed to interact with other dogs during walks. We verbally recorded behavioral observations using our cell phones (e.g., the voice memo app on an iPhone 7, model MN9G2LL/A, Apple Inc., Cupertino, CA, USA). We recorded each urination, defecation, and occurrence of ground scratching (backward scraping of the ground with the front feet, hind feet, or both performed by some dogs after urination or defecation). For each urination, we also recorded the posture used (female postures: squat, used by juvenile females and most adult females, and squat raise, used by some adult females; male postures: lean forward, used by adult males under fearful conditions and juvenile males, and raised leg, typical posture for adult males; [39,45]). At the end of walks, we returned dogs to their cubicles and retrieved relevant information from shelter records (e.g., dog identification number, intake date, source, and age). We used the intake date to calculate the number of days each dog had been at the shelter at the time of each of its walks with us (= time at shelter); for dogs that were adopted and returned to shelter, time at shelter was left blank. (Note that this meant that the 12 returned dogs were dropped from analyses, which included time at shelter as a main effect and the interaction between time at shelter and walker sex; see Section 2.3). Each dog was photographed. We transferred data from verbal recordings to paper check sheets within hours of walks and scanned each check sheet as a .pdf file.

#### *2.3. Statistical Analyses*

A linear mixed model was used to model the rate of urination (total number of urinations/20 min) and a generalized linear mixed model with a binominal distribution and logit link was used to model defecation (yes/no) and ground scratching (yes/no) during walks. We used a generalized estimating equation (GEE) to model the predominant urinary posture. We defined the predominant posture as the posture used most frequently during a walk; ties were recorded as such. We coded males whose predominant posture was either the raised leg or a tie between the raised leg and the lean forward as one; those whose predominant posture was the lean forward as zero (i.e., not involving a raised hindlimb). Similarly, we coded females whose predominant posture was the squat raise or a tie between the squat raise and squat as one; females whose predominant posture was the squat were coded as zero (again, not involving a raised hindlimb). All models were initially fit with the fixed effects of the dog's sex and the walker's sex, and the interaction between the dog's sex and the walker's sex. Time at shelter was included as a main effect and interacted with walker's sex. In all of the mixed models, we included the dog's ID as a random effect; in the GEE model, we treated the dog's ID as a cluster effect with an unstructured covariance matrix. Reduced models were obtained by removing interactions that were not significant (except for the interaction between dog sex and walker sex, which was retained in models due to research interest), and then removing the main effects that were not significant. For the rate of urination, we used Cohen's d to calculate the effect size. Data were analyzed using either JMP Pro 12 (2015. SAS Institute, Cary, NC, USA) or R, version 3.6.2 (R Foundation for Statistical Computing, Vienna, Austria).

#### **3. Results**

Descriptive statistics for the three scent-marking behaviors and time at shelter are shown in Table 1. The statistics in Table 1 are meant to provide a general overview of the raw behavioral data collected and the length of time dogs had been at the shelter at the time of their walks.


**Table 1.** Descriptive statistics (Mean ± *SD*) for rate of urination by male and female dogs during a 20-min walk by either male or female walkers, along with time at shelter. Additionally, percentage of walks in which dogs defecated or ground scratched are shown.

<sup>1</sup> Total number of urinations/20 min.

The results that follow are from the reduced models. The results from the full models are provided as Supplementary Material.

#### *3.1. Urination Rate*

Of the six male dogs that did not urinate during their walks, five did not urinate when walked by a male walker but did urinate when walked by a female walker, and the remaining dog did not urinate when walked by either male or female walkers. One female dog did not urinate when walked by a male walker but did urinate when walked by a female walker. Dogs that did not urinate during walks were included in analyses. We found a significant interaction between dog sex and walker sex for rate of urination (total number of urinations/20 min; Table 2). Male dogs urinated at higher rates when walked by female walkers than when walked by male walkers (d = 0.87); in contrast, female dogs urinated at similar rates when walked by male walkers and female walkers (d = 0.36; Figure 1a). Additionally, sex differences in rates of urination (urination rates of male dogs > urination rates of female dogs) were apparent with female walkers but not with male walkers (Figure 1a). The main effect for time at shelter also was significant (Table 2).

**Table 2.** Effects of sex of dog, sex of walker, and time at shelter on rate of urination per min by dogs during a 20-min walk.


**Figure 1.** Scent-marking behaviors of shelter dogs in relation to sex of dog and sex of walker. (**a**) Predicted rates of urination by male and female dogs when walked by male or female walkers. (**b**) Predicted probabilities of defecation by male and female dogs when walked by male or female walkers. Walks were 20 min in duration.

#### *3.2. Likelihood of Defecation*

We found a significant main effect of walker sex on likelihood that a dog would defecate during a walk (Table 3). A dog had a 0.441 probability of defecating with a male walker and a 0.740 probability of defecating with a female walker. We did not find a significant interaction between dog sex and walker sex for the likelihood that a dog would defecate during a walk (Table 3; Figure 1b). The odds that a male dog will defecate with a female walker are 2.9 times larger than with a male walker (*p* = 0.013). The odds that a female dog will defecate with a female walker are 4.5 times larger than with a male walker (*p* = 0.004).


**Table 3.** Effects of sex of dog and sex of walker on likelihood of defecation by dogs during a 20-min walk.

#### *3.3. Likelihood of Ground Scratching*

There were no significant predictors of ground scratching during a walk (Table 4).

**Table 4.** Effects of sex of dog and sex of walker on likelihood of ground scratching by dogs during a 20-min walk.


#### *3.4. Urinary Postures*

For male dogs, the raw data revealed the following percentages of walks in which the lean forward was the predominant urinary posture (i.e., all limbs remain on the ground when urinating): when walked by male walkers, 20.6%; when walked by female walkers, 13.2%. For female dogs, the raw data revealed the following percentages of walks in which the squat was the predominant urinary posture (again, all limbs remain on the ground when urinating): when walked by male walkers, 92.0%; when walked by female walkers, 95.1%. We found a significant interaction between dog sex and walker sex (Table 5). Male dogs were more likely to use the lean-forward posture when walked by a male walker than when walked by a female walker; in contrast, the likelihood of female dogs using the squat posture did not differ when walked by male walkers or female walkers. The odds of a male dog using the lean-forward posture as its predominant posture were 1.9 times greater when walked by a male walker (predicted probability of 0.222) than when walked by a female walker (predicted probability of 0.127; *p* = 0.052). Finally, the predicted probability of a female dog using the squat posture as its predominant posture was 0.922 when walked by a male walker, which did not differ from the predicted probability of a female dog using the squat posture as its predominant posture when walked by a female walker (0.941; *p* = 0.33).


**Table 5.** Effects of sex of dog and sex of walker on the likelihood of dogs having a predominant urinary posture in which all limbs remain on the ground (i.e., lean-forward posture in males and squat posture in females).

#### **4. Discussion**

We found that two scent-marking behaviors of shelter dogs—urination (as measured by urination rate) and defecation (as measured by occurrence during a walk)—were influenced by the sex of an unfamiliar walker. Ground scratching, also measured by occurrence during a walk, was not affected by walker sex. The predominant urinary posture during a walk also was affected by walker sex. Only urination rate was affected by time spent at the shelter: rate of urination slightly declined with increasing time spent at the shelter.

In the case of urination rate, the effects of walker sex varied with sex of dog. Male dogs urinated at higher rates when walked by unfamiliar women than when walked by unfamiliar men, whereas female dogs urinated at similar rates when walked by unfamiliar women and unfamiliar men. In fact, the well-established pattern of higher rates of urination by mature male dogs than mature female dogs [37,39,40,46] was present with female walkers but disappeared with male walkers in our study (Figure 1a). For the predominant urinary posture, male dogs were more likely to use the lean-forward posture when walked by unfamiliar men than when walked by unfamiliar women. The urinary posture of female dogs did not differ when walked by unfamiliar men and unfamiliar women. These findings for urination rate and predominant urinary posture support our predictions and are similar to results reported for dogs responding to the presence of either an unfamiliar man or an unfamiliar woman in a commercial kennel setting. Lore and Eisenberg [32] found that male dogs spent less time near an unfamiliar man than an unfamiliar woman, whereas female dogs did not differ in this regard; the authors found a similar pattern for direct body contact with an unfamiliar man and an unfamiliar woman. For the likelihood that a dog would defecate during a walk, we found a main effect of sex of walker but no interaction between sex of walker and sex of dog: both male dogs and female dogs were more likely to defecate when walked by an unfamiliar woman than when walked by an unfamiliar man (Figure 1b). Wells and Hepper [27] found a similar pattern when either an unfamiliar man or an unfamiliar woman stood at the front of cages for a few minutes: both male and female shelter dogs decreased to a greater extent time spent barking and looking at the person when the unfamiliar person was female. In summary, depending on the particular category of behavior used to assess response of dogs to unfamiliar people, the presence of an unfamiliar man can either uniquely affect male dogs (time spent in proximity, time spent in direct contact, rate of urine marking, and predominant urinary posture), affect male and female dogs in a similar manner (time spent barking, time spent looking, and likelihood of defecation), or have no significant effect on either male or female dogs (ground scratching). Ground scratching is performed by a minority of shelter dogs (present study; [40]). Cafazzo et al. [37] studied members of a feral dog pack and found that high ranking individuals ground scratched more frequently than low ranking ones. Perhaps dogs that ground scratch are more confident, which might explain our failure to find an effect of sex of walker for this particular scent-marking behavior.

Our finding that rate of urination declined with increasing time spent at the shelter was unexpected; although the size of this effect was very small, it nonetheless was significant. In a previous study of scent-marking behavior during first walks of dogs at the same shelter, we found that time spent at the shelter did not influence urinary behavior (frequency of urination or percent of urinations directed at targets in the environment) or likelihood of defecation or ground scratching [40]. In a subsequent study in which many dogs were walked multiple times [47], we found that time spent at the shelter positively influenced rate of urination, percent of directed urinations, and likelihood of defecation (ground scratching was not studied). In light of our findings for first walks [40], we interpreted the positive influence of time at shelter in our second study [47] as having resulted from our inclusion of multiple walks on individual dogs, and suggested that the positive influence of time spent at the shelter on marking behavior could reflect dogs becoming more familiar with their surroundings and routine, as well as with us [47]. For these two earlier studies [40,47], there were no single male walkers or single female walkers; we always had two people on each walk, one student (male or female) and B.M. (one person walked the dog and the other recorded behavioral observations). These methodological differences make our current finding regarding urination rate and time at shelter challenging to interpret. One possible interpretation is that more timid dogs are characterized by longer stays at the shelter as well as lower rates of urination.

We did not determine the precise stimuli by which shelter dogs discriminated sex of an unfamiliar walker. Potential stimuli include olfactory, visual, auditory, and tactile/handling differences between male and female walkers. With respect to the latter, subtle differences in petting techniques of males and females appeared responsible for an initial report that shelter dogs petted by females had lower cortisol levels than those petted by males [28]. In two subsequent studies in which men and women received specific training to standardize their petting techniques, male and female petters reduced cortisol levels to the same degree [20,29]. In contrast, androgen-based olfactory cues are used by laboratory mice and rats to discriminate experimenter sex: olfactory cues from men caused a physiological stress response that induced an inability to feel pain in both rodent species [10]. Acoustic cues in voices are used by elephants to discriminate human sex and age: in response to playbacks of either adult male or adult female Maasai voices, members of elephant families were more likely to retreat and exhibit defensive bunching when hearing male voices [16]. It is important to determine which features of unfamiliar men shelter dogs attend to, so that the effectiveness of human interaction during walks, and perhaps other enrichment activities, can be maximized. Additionally, determining whether the observed reactions to unfamiliar male walkers disappear with familiarity would be useful.

Our study focused on scent-marking behaviors of dogs during walks by unfamiliar males and unfamiliar females; we did not measure the physiological responses of dogs to unfamiliar male and female walkers. However, Alberghina et al. [48,49] conducted two studies with shelter dogs to investigate the relationships between scent marking, cortisol, and supervised social exposures with another dog. Social exposures occurred in a fenced area, with both dogs initially on leashes and then eventually off leashes. In the first study, Alberghina et al. [48] found a significant positive relationship between frequency of urine-marking by dogs during social exposures and urinary cortisol-creatinine ratio (C/Cr) measured several hours later and a significant negative relationship between the frequency of defecation during social exposures and C/Cr. In the subsequent study, which differed from the first in some aspects of methodology (e.g., dogs were habituated to muzzles before social exposures in the second study but not in the first), Alberghina et al. [49] found the same patterns with respect to urination and C/Cr and defecation and C/Cr but the results did not reach statistical significance. As suggested by Alberghina et al. [48,49] and Protopopova [50], elevated levels of cortisol could indicate increased arousal and activity, rather than stress in dogs. Results from mammals studied under laboratory conditions suggest a complicated relationship between stress, scent marking, and cortisol. For example, when housed without access to a preferred outdoor cage, common marmosets exhibited elevations in cortisol and increases in scent marking behavior (rubbing scent glands on the substrate) [51]. In contrast, male Mongolian gerbils exhibited elevated cortisol levels but reduced scent

marking (rubbing the ventral gland on the substrate) when subjected to social defeat, a stress paradigm in which a male is repeatedly paired with a dominant male conspecific [52]. Thus, elevated cortisol has been associated with both increases and decreases in scent marking behavior in mammals, suggesting the relationship between scent marking, stress, and cortisol requires further study across species and stress paradigms.

A limitation of our study is that age also varied among walkers (female walkers: D.O., 21, B.M., 61; male walkers: J.C., 20, K.F., 22, L.U., 31). Few studies have examined the influence of human age on dog behavior, except in regard to dog bites (e.g., [53]). Although Koda and Shimoju [30] found that dogs enrolled in a guide dog program made more frequent contact with unfamiliar women than unfamiliar men, they found no difference in the frequency with which dogs contacted unfamiliar females who were either between 20 and 40 years old or between 8 and 13 years old. These findings suggest that, at least in the case of unfamiliar females, age might not matter to dogs; Koda and Shimoju [30] did not examine response of dogs to unfamiliar males from different age groups. Dog volunteers at the Tompkins County SPCA ranged from 18 years old to over 70 years old; thus, the dogs in our study likely had some experience interacting with humans of diverse ages before we walked them.

#### **5. Conclusions**

Given that sex of an unfamiliar human has been shown to affect both the in-kennel behavior of shelter dogs [27] and their behavior outside the kennel during leash walks (present study), we suggest that researchers conducting behavioral observations of shelter dogs (and perhaps dogs generally) record, consider in their analyses, and report the sex of observers/handlers as standard practice. Based on their findings that experimenter sex influenced the behavior and physiology of laboratory mice and rats, Sorge et al. [10] made a similar recommendation for researchers studying any phenomenon in laboratory rodents that could be affected by stress. Our findings might also have implications for canine behavioral evaluations at animal shelters. Such evaluations are usually conducted a few days after intake, when dogs are likely unfamiliar with at least some staff present at these tests. Additionally, behavioral evaluations often include a subtest in which an unfamiliar person knocks on the door and enters the room where testing is taking place (e.g., Stranger Test in the Modified Assess-A-Pet; [42,43]). Shelter dogs have been shown to differentiate sex of an unfamiliar person during behavioral evaluations: Bergamasco et al. [26] found that shelter dogs enrolled in a human interaction enrichment program and behaviorally evaluated several times over a period of weeks improved in their responses to unfamiliar females but not to unfamiliar males. Thus, sex of the unfamiliar person and perhaps sex of the evaluator/handler and scribe, could potentially influence results of canine behavioral evaluations, which might then affect dog welfare by influencing whether or not a dog is made available for adoption.

**Supplementary Materials:** The following are available online at http://www.mdpi.com/2076-2615/10/4/632/s1, Table S1: Effects of sex of dog, sex of walker, and time at shelter on the rate of urination per min by dogs during a 20-min walk. Results are from the full model, Table S2: Effects of sex of dog and sex of walker on the likelihood of defecation by dogs during a 20-min walk. Results are from the full model. Table S3: Effects of sex of dog and sex of walker on the likelihood of ground scratching by dogs during a 20-min walk. Results are from the full model. Table S4: Effects of sex of dog and sex of walker on the likelihood of dogs having a predominant urinary posture in which all limbs remain on the ground (i.e., lean-forward posture in males and squat posture in females). Results are from the full model.

**Author Contributions:** Author contributions were as follows: conceptualization, B.M.; supervision, B.M.; methodology, B.M. and S.P.; data collection, B.M., K.F., D.O., and L.U.; coordination of statistical analyses, B.M.; statistical analyses, S.P.; writing—original draft preparation, B.M. and K.F.; writing—review and editing, B.M., K.F., D.O., L.U., and S.P. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Acknowledgments:** We thank Jim Bouderau, Executive Director of the Tompkins County SPCA, for permission to walk dogs at the shelter. Emme Hones, Behavior Program Manager, Kat Pannill, Volunteer Coordinator, and Heather Marsella, Shelter Operations Manager, provided information on numbers of male and female staff and volunteers. Jordan Chan joined the walking team toward the end of this study and Emma Rosenbaum entered the data on urinary posture. Kate Bemis and William Bemis reviewed a draft of this manuscript.

**Conflicts of Interest:** The authors declare no conflicts of interest.

#### **References**


© 2020 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).

MDPI St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Animals* Editorial Office E-mail: animals@mdpi.com www.mdpi.com/journal/animals

MDPI St. Alban-Anlage 66 4052 Basel Switzerland

Tel: +41 61 683 77 34 Fax: +41 61 302 89 18