Data collection
An online owner-filled questionnaire was designed to collect information on canine behaviour in a pet dog population17. The questionnaire was divided into seven main sections based on different behavioural traits, including fear of dogs, strangers, and novel situations (labelled as ‘fear’); aggression toward human family members and strangers (labelled as ‘aggression’); fear of thunder, fireworks, and gunshot (labelled as ‘noise sensitivity’); fear of surfaces and heights; inattention and hyperactivity/impulsivity; separation anxiety; and compulsive behaviour. In addition, the questionnaire included several questions concerning the background and the current living environment of the dog, such as socialisation experiences during the age of 7–16 weeks, activities that the dog possibly participates in (e.g. dog shows, agility, or herding), amount of daily exercise, and the number of other dogs in the family. All the questions included in the behaviour questionnaire can be found as Supplementary material in the paper of Salonen et al.17.
In the questionnaire section concerning fear, the potential fearful reactions towards strangers and other dogs (labelled as ‘social fearfulness’) as well as novel situations (labelled as ‘non-social fearfulness’) were asked. As this paper focuses on studying the demographic and environmental factors associated with social fear in dogs, only the structure of the social fear subdivision is described in more detail here. If the owner answered that his/her dog reacts fearfully when meeting a stranger and/or a strange dog, we required him/her to report how often these reactions occur (rarely, 0–20% of occasions; sometimes, 20–40% of occasions; often, 40–60% of occasions; almost always, 60–100% of occasions; always, 100% of occasions). In addition, we asked him/her to more specifically indicate how the dog behaved in these situations (e.g. the dog withdraws when meeting a stranger). Moreover, if the owner responded that his/her dog does not react fearfully in those situations, he/she was asked to describe the dog’s behaviour more specifically to give us possibility to evaluate the behavioural reactions of the dog. The fear section of the questionnaire was previously shown to have high reliability and validity56.
As our aim was to reveal demographic and environmental factors associated with both fear of dogs and fear of strangers, two separate binomial (event/non-event) variables (‘fear of dogs’ and ‘fear of strangers’) describing these behaviours were derived from the questionnaire data. According to the owners, non-fearful dogs are reported to indicate no fear of other dogs (‘fear of dogs’) or strangers (‘fear of strangers’) whereas fearful dogs are reported to show fear of strange dogs (‘fear of dogs’) or strangers (‘fear of strangers’) in more than 40% of the occasions (Supplementary Table S6).
The questionnaire was advertised to dog owners on Facebook and via breed organisations. We obtained an informed consent from all participants and informed them that the questionnaire answers would be used for research. We emphasised that all data will remain confidential and that individual dogs and owners cannot be identified from the published results. Questionnaire replies were collected during 2015–2018.
Demographic and environmental variables
Before statistical analyses, some demographic and environmental variables derived from the questionnaire were edited and some new variables were created. First, only breeds with enough individuals in both non-fearful and fearful groups (>10 individuals/group) were selected for the analyses. Individuals in other breeds were grouped together under ‘other’ breed group. The selected breeds included Border Collie, Cairn Terrier, Chihuahua, Chinese Crested Dog, Coton de Tuléar, Finnish Lapponian Dog, German Shepherd Dog, Golden Retriever, Irish Soft Coated Wheaten Terrier (labelled as ‘Wheaten Terrier’), Jack Russell Terrier, Labrador Retriever, Lagotto Romagnolo, Lapponian Herder, Medium size Spitz, Miniature Poodle (including Toy, Miniature, and Medium Poodle), Miniature Schnauzer, Pembroke Welsh Corgi, Rough Collie, Shetland Sheepdog, Smooth Collie, Spanish Water Dog, and Staffordshire Bull Terrier (labelled as ‘Staff. Bull Terrier’) (Supplementary Table S1). Mixed breed dogs were also included in the data. In the ‘fear of strangers’ analysis, Cairn Terrier, Pembroke Welsh Corgi, and Staff. Bull Terrier were included in the group ‘other’ as they had too few individuals for the logistic regression analysis.
Second, we created a new categorical variable, body size, based on the average heights of breeds. The heights were drawn from FCI and AKC standards, when available. For other breeds, with no FCI or AKC standards, the average heights were determined based on heights reported by breed clubs. If different heights were reported for females and males, the mean was calculated and used as the average height. Individuals with breed average heights of ≤35 cm, 36–49 cm, or ≥50 cm belonged to small, medium, or large categories, respectively. When body size was included in the analyses, mixed breed dogs were excluded as their body size could not be determined.
Third, a continuous variable labelled as socialisation score was calculated based on the frequency of socialisation events when the dog was 7–16 weeks old. The score is a sum of the frequencies (0 = never; 1 = 1–2 times during the puppyhood; 2 = 1–2 times during the puppyhood to 2 times per month; 3 = twice a month to twice a week; 4 = twice a week to once a day; 5 = several times a day) the dog met unfamiliar men, women and children, unfamiliar adult dogs, visited city or other place with traffic and many people, and travelled by car or by bus.
Fourth, a continuous variable labelled as urban environment score describing the environmental land-use around the dog’s home was created. We derived the geographical coordinates for each home from addresses provided by the dog owners, and calculated the coverage of three land-use types, artificial surfaces, agricultural areas, and forests and semi-natural areas within a three-kilometre range around the homes using the land-use database CORINE2012. Finally, we simplified the coverages into one continuous variable with principal component analysis (PCA). The higher the urban environment score, the more urban was the environment.
Variables derived from the questionnaire data and included in the analyses are described in more detail in the Supplementary Table S6.
Statistical analyses
Logistic regression was used to find demographic and environmental factors associated with two subtraits of social fearfulness, fear of dogs and fear of strangers. Initially, the data consisted of 13,715 dogs in 264 breeds. After including only fearful and non-fearful dogs and excluding individuals with missing or incomplete responses, the data consisted of 5,343 dogs in ‘fear of dogs’ and 5,858 in ‘fear of strangers’. The subtraits were used as binary response variables in the analyses: fearful and non-fearful dogs constituted the event and the non-event, respectively. Based on previous literature, we included several explanatory variables in the analyses. Demographic explanatory variables included age, sex, sterilisation, breed, and body size. Environmental explanatory variables included socialisation score, weaning age, urban environment score, daily exercise, activities/training, owner’s dog experience, dogs in the family, family size, and daily time spent alone. To find the models with the best fit, a forward stepwise model selection by Akaike Information Criterion (AIC) values was used and was initiated with a starter model of sex and age in both ‘fear of dogs’ and ‘fear of strangers’ analyses. The AIC model selection process and the final models are presented in the Supplementary Table S7. In the ‘fear of dogs’ model, the interaction between sex and sterilisation was significant and improved the AIC value of the model. Therefore, the interaction term (sex*sterilisation) was included in the final model. To maximise the sample sizes, new subsets of the initial data were created to include only those variables included in the final models. This resulted in datasets of 5,973 and 5,932 individuals for ‘fear of dogs’ and ‘fear of strangers’, respectively.
After model selection, generalised additive models were fitted with the package ‘gam’57 in R to test the linearity assumption of continuous variables. If the linearity assumption of a variable was not met, the variable squared was also included in the model (e.g. age^2). Packages ‘broom’58 and ‘dplyr’59 in R were used to inspect possible outliers in the datasets. Standardised residuals were plotted using the package ‘ggplot2’60 in R. Generalised variance inflation factor (gVIF) was utilised to test multicollinearity with the package ‘car’61, and the area under the receiver operator characteristic curve (AUC) was calculated to estimate how well the model predicted the event and the non-event using the package ‘pROC’62 in R. The predictive abilities of both models were reasonable (AUC(fear of dogs) = 0.728 and AUC(fear of strangers) = 0.688).
Estimated marginal means were calculated for categorical explanatory variables with the package ‘emmeans’63 in R. The effects of continuous explanatory variables (adjusting for other variables in the models) were obtained with the package ‘effects’64 in R. To obtain the overall effects of the explanatory variables, analysis of variance (ANOVA) was conducted with the package ‘car’61 in R. Based on previous literature, we determined several contrasts between levels of explanatory variables a priori. First, we requested a contrast between potentially more fearful (Chihuahua, Jack Russell Terrier, Lagotto Romagnolo, and Shetland Sheepdog) and less fearful (German Shepherd Dog, Golden Retriever, Labrador Retriever, and Staff. Bull Terrier) breeds36,37,38. Second, we hypothesised that female dogs would be more fearful than male dogs33,36,37,52,53. Third, we hypothesised that large dogs would differ from small dogs in their behaviour38,39,40,41. Fourth, contrast between early weaning (weaned <7 weeks of age) and normal weaning age (7–8 weeks of age) as well as contrast between normal weaning age and late weaning age (>8 weeks of age) were requested22,33. Fifth, we hypothesised that younger dogs would be more fearful22. Finally, we made a hypothesis that dogs with less socialisation (i.e. lower socialisation scores) would show more fearful behaviour22. The hypotheses and contrasts were the same in both ‘fear of dogs’ and ‘fear of strangers’ analyses, except for weaning age which was only included in the ‘fear of strangers’ model. Furthermore, in the ‘fear of strangers’ analysis, the breed contrast was as follows: Chihuahua, Jack Russell Terrier, Lagotto Romagnolo, and Shetland Sheepdog versus German Shepherd Dog, Golden Retriever, and Labrador Retriever (Staff. Bull Terrier was left out due to a low sample size).
A priori contrasts and all pairwise comparisons between levels of the included categorical variables were examined with the package ‘emmeans’63 in R. As we had several categorical variables, the number of pairwise comparisons was high and therefore all obtained p-values, except contrasts chosen a priori, were controlled for false discovery rate (FRD). The significance cut-off was set at p-value <0.05.
All statistical analyses were conducted in R version 3.6.165.
Source: Ecology - nature.com
