Behavioural differences and similarities between dog breeds: proposing an ecologically valid approach for canine behavioural research

ABSTRACT

The behaviour of dogs holds great relevance for not only scientists from fundamental and applied research areas, but also due to the widespread roles of dogs in our societies as companions and working animals; their behaviour is also an important factor in animal and human welfare. A large proportion of dogs currently under human supervision belong to one of roughly 400 recognised breeds. Dog breeds can be characterised by distinctive, predictable and reproducible features, including some of their behavioural traits. To the scientist, the comparative analysis of the behaviour of dog breeds provides an opportunity for investigating an array of intriguing phenomena within an easily accessible model organism created from natural and human‐driven evolutionary processes. There are many ways to design and conduct breed‐related behavioural investigations, but such endeavours should always be based around biologically relevant research questions and lead to ecologically valid conclusions. In this review, we surveyed recent research efforts that included dog behaviour‐related comparisons and applied a critical evaluation according to their methods of breed choice and the subsequent research design. Our aim was to assess whether these two fundamentally important components of experimental design provide a solid basis to reach valid conclusions. Based on 97 publications that fulfilled our selection criteria, we identified three primary methods used by researchers to select breeds for their investigations: (i) convenience sampling; (ii) hypothesis‐driven, ancestry‐based sampling; and (iii) hypothesis‐driven, functional sampling. By using the SWOT (Strengths, Weaknesses, Opportunities, Threats) evaluation system, we highlight each of these techniques' merits and shortcomings. We identify when particular methods may be inherently unable to produce biologically meaningful results due to a mismatch between breed choice and the initial research goals. We hope that our evaluation will help researchers adopt best practices in experimental design regarding future dog breed comparisons.

I. INTRODUCTION

The dog (Canis familiaris) is the oldest domesticated species (e.g. Lee et al., ²⁰¹⁵), cohabits with humans in every culture, and according to some estimations, has reached nearly a staggering one billion individuals worldwide (Coppinger & Coppinger, 2021). Especially in countries of the global North, people consider dogs as primarily fulfilling the role of companions for humans (Meyer et al., ²⁰²²; Pongrácz & Dobos, ²⁰²³), and to a lesser extent, as a working animal (military and police dogs, assistance dogs, herding and hunting dogs, etc.). However, the largest proportion of the global dog population exists as unowned individuals living in the cities and villages of Southeast Asia, Africa and Central and South America (Coppinger & Coppinger, 2021), where they are mainly known as pariah (Oppenheimer & Oppenheimer, 1975), village (Pendleton et al., ²⁰¹⁸) or street dogs (Reece, Chawla & Hiby, 2013). The common shared feature of these unowned dogs is their reliance on human resources and unrestricted reproduction (Boitani, Ciucci & Ortolani, 2007). Additionally, these dogs show remarkable similarity across populations, implying that their appearance and behaviour has been shaped by natural selection (Coppinger & Coppinger, 2021).

Compared to unowned canines, dogs that live as companions or working animals show considerable variability, to the extent that the dog can be considered as the most variable mammalian species (Creevy et al., ²⁰²²). For example, there is a roughly 50‐fold difference in body mass between the smallest and largest dog breeds. While there is a relatively restricted set of genes responsible for the main features differentiating the appearance of dogs [i.e. size, limb and tail length, ear shape and position, fur length and texture, colour (Ostrander, 2007; Wayne & VonHoldt, ²⁰¹²)], alongside the countless variations of mongrels, there are at least 400 different dog breeds that are recognised by various organisations (“kennel clubs”) across the world. Although dog breed descriptions (“standards”) mainly emphasise physical appearance, they usually also enumerate the original (“working”) function and typical behaviour of the given breed. Behaviour (working purpose) also serves as the main organising factor behind the “breed groups” of the various breeding organisations [e.g. FCI (Fédération Cynologique Internationale) – https://www.fci.be/en/Nomenclature/; AKC (American Kennel Club) – https://www.akc.org/dog-breeds/].

The behaviour of dogs has received increasing interest from ethologists and comparative psychologists since the 1990s (Aria et al., ²⁰²¹). A species that formerly was largely ignored by ethologists because of doubts regarding its “natural” form (i.e. free from the effects of domestication) became of interest when it was realised that the natural habitat for dogs was the anthropogenic niche (Miklósi & Topál, 2013). The evolutionary processes and the resulting socio‐cognitive features that helped dogs adapt so well to the human social environment thus attracted both behavioural scientists and comparative psychologists (Topál et al., ²⁰⁰⁹). In parallel, the complex system of interactions between companion animals and their owners provided research opportunities for applied ethologists (Pongrácz & Dobos, 2023), particularly as dog–human interactions are considered important aspects of welfare for both parties (e.g. Glenk, ²⁰¹⁷; Thielke & Udell, ²⁰¹⁹). Dog behaviour phenotypes proved to be important in studies of evolution/domestication (e.g. in comparative investigations of the behaviour of tame wolves and dogs; Gácsi et al., ²⁰⁰⁵), dog–human interactions (McGreevy et al., ²⁰¹²), cognitive ethology (e.g. understanding of projected images; Péter, Miklósi & Pongrácz, ²⁰¹³), and various aspects of applied behavioural sciences, such as training/selecting dogs for working tasks (Bray et al., ²⁰²¹), animal welfare issues (e.g. behavioural signs of separation‐related problems; Pongrácz et al., ²⁰¹⁷), and human safety (e.g. behavioural assessments at dog shelters; Mornement et al., ²⁰¹⁰).

Breed‐related differences in dog behaviour were investigated even before dogs became a focus of scientists studying human–animal interactions and domestication [e.g. the early experiments of Scott & Fuller (1974) regarding the sensitive period for early socialisation; or the studies of Feddersen‐Petersen (1990) on breed‐specific environmental needs and ontogenetic changes]. However, even today, studies on between‐breed similarities and differences often lack the necessary systematic approach based on ecological validity in behavioural sciences. Apart from some recent endeavours that applied evolutionary theory (e.g. based on dog clades' genetic distances from a hypothetical ancestor; e.g. Parker, ²⁰¹²), to establish a molecular genetic background for breed‐related behavioural features (e.g. Morrill et al., ²⁰²²), dog breeds mostly appear in behavioural studies as either a targeted independent factor (Serpell & Duffy, 2014), or in ad‐hoc comparisons based on the most popular breeds (e.g. Pongrácz et al., ²⁰⁰⁵; Junttila et al., ²⁰²²). While researchers often aim to answer questions related to the assumed different selection past of various dog breeds (e.g. Udell et al., ²⁰¹⁴), they rarely include more than a few “handpicked” breeds in their investigation, with the results inevitably impacted by numerous additional differences among the chosen breeds (e.g. Pogány et al., ²⁰¹⁸; Ujfalussy et al., ²⁰²³).

In this review, we focus on studies from the last few decades where behavioural researchers tested multiple dog breeds and presented breed‐related results. We provide a brief enumeration of the typical aims of these papers, as well as their focal breeds (or breed groups). Then, through examples, we critically analyse whether their methodologies were suitable for achieving their declared goals. Finally, based on the principles of the SWOT (“Strengths, Weaknesses, Opportunities and Threats”) method (e.g. Van Belle et al., ²⁰²⁴), we synthetise recommended future research strategies that reflect the fundamental criterion of ecological validity (biological relevance) for investigations of breed‐related behavioural similarities/differences of dogs.

II. METHODS: LITERATURE SURVEY

We performed a literature search on two separate occasions: October 16, 2023, using Google Scholar; and June 3, 2024 using Science Direct, to identify relevant literature that involved comparative behavioural analysis of dog breeds. On both occasions, we ran the same three consecutive searches using the search terms: “dog breeds AND behaviour”, “dog breeds AND learning”, and “dog breeds AND communication”. For each search, the first 100 results were processed, thus a total of 600 publications were analysed in detail (i.e. reading the titles and abstracts), to select suitable articles for our review.

Our primary focus was on original research papers and book chapters. Review articles, monographs, conference abstracts, and dissertations were not included. We excluded duplicate hits and publications based on topical mismatch (see Table 1). Most of the excluded papers either reported purely molecular genetic results with no highlighted behavioural relevance, did not include behavioural, or behaviour‐related analysis, lacked any breed‐related comparison, or did not include more than one dog breed.

Table 1.

The type and number of papers identified by the literature search that were excluded from this review because the topic of the paper did not meet our selection criteria (article should provide behavioural or behaviour‐related results that involve dog breed comparisons).

Reasons for exclusion and example of excluded study	Number of papers
Veterinary breed identification (Simpson et al., 2012)	1
Paper focused on human behaviour (Wells & Hepper, 2012)	1
Anatomy‐related paper (e.g. Vedat et al., 2023)	2
Not dog‐related paper (e.g. Lloyd et al., 2008)	6
Paper analysed the subjective opinion of respondents (e.g. Packer et al., 2017)	8
Review paper (e.g. Mehrkam & Wynne, 2014)	14
Paper focused on human societal issues (e.g. Clarke et al., 2013)	18
Veterinary report on health/disease issues (e.g. Berg et al., 2024)	19
Purely molecular genetic results, with no direct behavioural links (e.g. Bionda et al., 2023)	37
No behavioural data presented (non‐genetic papers) (e.g. O'Neill et al., 2023)	83
No breed comparisons (e.g. McGreevy & Nicholas, 1999)	148

Our final list included 97 publications with data/results on behavioural phenotypes of dogs that reported breed‐related comparisons.

III. RESULTS OF THE LITERATURE ANALYSIS

We assessed the 97 identified publications according to three, hierarchically arranged characteristics:

• (1) The typical reasons why authors compared the behaviour of dog breeds (proximal goal), and conclusions (ultimate goal) of the study. We separated these according to two main categories: (i) the authors targeted fundamental research questions while comparing particular breeds or breed groups; or (ii) the goals/conclusions were chosen according to an applied question with regard to breed or breed group comparisons (e.g. training, safety, companion animal related).
• (2) Breed choice. We separated these according to three main methods: (i) hypothesis‐driven/functional breed choice; where authors analysed the results of either a few, or a larger number of dog breeds (or breed groups), based on a priori functional hypothesis. These hypotheses were linked with artificial selection for particular working tasks, or anatomical features, or physiological attributes of the dog breeds (breed groups); (ii) hypothesis‐driven/ancestry‐based breed choice (e.g. wolf‐like versus more modern breeds, etc.). Here the authors investigated the associations of behavioural traits with the genetic “distance” between breeds or breed groups; (iii) convenience breed choice, where the authors chose their breeds based on their availability, often opting for the most common breeds in a particular location/country. These studies commonly used the clustering systems of the large kennel clubs (FCI or AKC breed groups).
• (3) The number of compared breeds. We sorted papers into two clusters: (i) papers that compared no more than four breeds were considered as using “few breeds”; and (ii) publications that used breed groups or numerous dog breeds for comparisons were considered as using “several breeds”. We opted for four breeds as cut‐off between “few” and “several” breeds, because if authors intended to compare “breed groups” in a study, they used at least two groups with a minimum of two breeds in each.

The results of this process are shown in Table 2.

Table 2.

Results of the literature survey. Publications (N = 97) included in this table both (i) reported the results of behavioural tests, behaviour‐related analyses, or surveys on dog behaviour, and (ii) compared more than one dog breed, or groups of breeds.

Research aim/scope of conclusions	Method of breed (breed group) choice	Number of breeds (breed groups)	References
Association between behaviour and various breeds (breed groups)	Hypothesis‐driven, functional	Several breed groups, several breeds in each	Bognár et al. (2021); Dobos & Pongrácz (2023); Gácsi et al. (2009); Hecht et al. (2019); Hecht et al. (2021); Helton & Helton (2010); Lenkei et al. (2021a ); Clarke et al. (2019); Svartberg (2006)
		Few (2–4) breeds	Kovács et al. (2016); Eretová et al. (2024); McGreevy et al. (2010); Ujfalussy et al. (2023)
	Hypothesis‐driven, ancestry‐based	Several breed groups, several breeds in each	Hansen Wheat et al. (2019); Kerswell et al. (2009); Konno et al. (2016); MacLean et al. (2019); Morrill et al. (2022); Smith et al. (2017); Tonoike et al. (2015); Vaysse et al. (2011); Dutrow et al. (2022)
		Few (2–4) breeds	Nagasawa et al. (2017); Nagasawa et al. (2024)
	Convenience (“popular” breeds)	Several breeds	Dorey et al. (2009); Feddersen‐Petersen (2000); Gnanadesikan et al. (2020); Hart & Hart (1985); Hart & Miller (1985); Horschler et al. (2019); Lit et al. (2010); Pongrácz et al. (2001); Pongrácz et al. (2005); Serpell & Duffy (2014); Riedel et al. (2008); Turcsán et al. (2011); Salonen et al. (2023); Eo et al. (2016)
		Few (2–4) breeds	van den Berg et al. (2010); Niimi et al. (1999); Maglieri et al. (2019); Persson et al. (2018); Jakovcevic et al. (2010); Burza et al. (2022)
Applied aspects of breed–behaviour associations	Hypothesis‐driven, functional	Several breeds	Caddiell et al. (2023); Christiansen et al. (2001); Hammond et al. (2022); Helton (2009); Polgár et al. (2016); Pongrácz et al. (2020); Van Poucke et al. (2022)
		Few (2–4) breeds	Fadel et al. (2016); Kinka & Young (2018); Strychalski et al. (2015); Overall et al. (2016); Sundman et al. (2016); Vas et al. (2005)
	Hypothesis‐driven, ancestry‐based	Several breeds	Marshall‐Pescini et al. (2016); Wobber et al. (2009); Wójcik & Powierża (2021)
		Few (2–4) breeds	–
	Convenience (“popular” breeds)	Several breeds	Anderson et al. (2022); Asp et al. (2015); Bellamy et al. (2018); Chapagain et al. (2017); Col et al. (2016); Cornelissen & Hopster (2010); Creedon & Ó Súilleabháin (2017); Donohue et al. (2024); Ekiz et al. (2023); Essig et al. (2019); Ghirlanda et al. (2013); Helton (2010); Kogan et al. (2019); McGreevy et al. (2013); Notari & Goodwin (2007); Salvin et al. (2012); Schalamon et al. (2006); Starling et al. (2013); Storengen & Lingaas (2015); Sundman et al. (2020); Vajányi et al. (2024); Westgarth et al. (2010); Zapata et al. (2022)
		Few breeds (2–4)	Andelt (1999); Baranyiová et al. (2007); Bloom et al. (2021); Ott et al. (2008); Ott et al. (2009); Sandøe et al. (2017); Wright et al. (2007); Rice & Velasco (2023); Wilsson & Sundgren (1997); Morrow et al. (2015); Jezierski et al. (2014); Santariová et al. (2023); Bowden et al. (2018)

IV. CRITICAL EVALUATION OF STUDY DESIGN IN DOG BREED‐RELATED BEHAVIOURAL RESEARCH

(1). Convenience sampling of breeds – caveats and advantages

Most articles that we included had a proximal goal focusing on breed‐related (behavioural) differences. However, methodologies differed greatly. Many papers used dog breeds as a grouping (independent) variable without an a priori hypothesis, resulting in an exploratory study instead of stating why the authors expected particular dog breeds would behave differently from others (e.g. Hart & Hart, ¹⁹⁸⁵; Turcsán, Kubinyi & Miklósi, ²⁰¹¹; Ghirlanda et al., ²⁰¹³). In other cases, authors targeted various behavioural (e.g. Hecht et al., ²⁰²¹), socio‐cognitive (e.g. Gnanadesikan et al., ²⁰²⁰), or temperament‐based (e.g. Zapata et al., ²⁰²²) phenotypes, and assumed that dog breeds (or sometimes breed groups; e.g. Pongrácz et al., ²⁰⁰⁵) would show consistent variability along these scales. The way dog breeds were included in these investigations was most commonly based on convenience sampling, which in practice mirrored the most abundantly encountered (popular) breeds through large‐scale behavioural assessment (Pongrácz et al., ²⁰⁰⁵), questionnaire (Turcsán et al., ²⁰¹¹; McGreevy et al., ²⁰¹³), and veterinary databases (Anderson et al., ²⁰²²). The potential strength of this approach is access to very high numbers of individual records (up to 10,000; e.g. Serpell & Duffy, ²⁰¹⁴), usually from a wide array of dog breeds. The results therefore can be regarded as representative, at least for breeds that contributed large numbers to the sample. However, these surveys face an inherent problem of being explorative [e.g. the pioneering study of Feddersen‐Petersen (2000) on between‐breed differences in vocalisations], which limits the possibility of drawing ultimate conclusions regarding, for example, selection processes resulting in breed (or breed‐group)‐related behavioural differences. For example, Pongrácz et al. (2005) investigated breed‐related associations in how efficiently dogs could master a detour task on their own, as well as after observing a human demonstrator making the detour. The method of breed choice limited their analysis to the 10 most common dog breeds at dog schools, which inevitably led to skewed selection. The sample consisted of mostly cooperative breeds [sheepdogs, gundogs and utility (police) dog breeds], with only two breeds that were selected for independent working tasks (a livestock guardian and a terrier breed). The authors did not find differences in breed performance in either the trial‐and‐error or social learning scenario. However, their unstructured convenience sampling of breeds would make it difficult to identify major effects regarding capacities of dogs that seem quite generally present in the included species [e.g. the capacity to pay attention to humans (Topál, Kis & Oláh, 2014) and performance in detour‐like spatial tasks (Pongrácz et al., ²⁰⁰¹)].

The limitations of convenience breed choice are not always obvious at first sight: many of the surveyed articles contained conclusions that referred to assumed evolutionary changes since domestication (e.g. Konno et al., ²⁰¹⁶; Hansen Wheat et al., ²⁰¹⁹), or post‐domestication selection [e.g. working function (Pongrácz et al., ²⁰⁰⁵); dog–human interactions (Kogan et al., ²⁰¹⁹)]. However, in the absence of carefully planned breed sampling, the resulting “random” (i.e. popularity/availability based) inclusion/exclusion of breed representatives will result in numerous alternative explanations for the results. Thus, availability‐based breed comparisons are best restricted to explorative detection of consistent behavioural differences (e.g. Serpell & Duffy, ²⁰¹⁴; Hansen Wheat et al., ²⁰¹⁹). For investigating causative relationships (e.g. the effect of selection on dog behaviour in general, or on particular behavioural phenotypes), a planned, systematic choice of breeds (or breed groups) is recommended.

(2). Planned breed choice

Within the framework of our survey, we encountered two main types of planned breed choice. Their common feature was that the authors usually targeted groups of breeds (or breed groups, breed types) as a planned independent factor, however, they included either (i) only one breed (or very few breeds) per group as a “representative” (e.g. Vas et al., ²⁰⁰⁵), or (ii) multiple breeds from each group (e.g. Dobos & Pongrácz, ²⁰²³). Using only one, or very few breeds per group unavoidably limits the scope of conclusions that can be drawn from the results. No matter how well the given dog breed represents a group, there will be additional phenotypes that confound the results. For example, in the study of Ujfalussy et al. (2023), the authors found that brachycephalic English and French Bulldogs showed more pronounced gazing behaviour towards nearby humans than the mesocephalic Mudi dogs in the so‐called “unsolvable task”. The authors argued that this result showed that breeding dogs for extremely short heads could result in paedomorphic (i.e. strongly dependent) behavioural characteristics compared to normocephalic dog breeds. However, with this very limited choice of breeds the study could not convincingly reject alternative hypotheses, such as the potential effects of functions of these breeds: Mudis are herding dogs, while French and English Bulldogs have long acted as companion animals. Another study, also using the “unsolvable task”, compared only three breeds (Czechoslovakian Wolfdogs, German Shepherd Dogs and Labrador Retrievers), to see whether an assumed “wolf‐likeness” or “dog‐likeness” was associated with willingness to look at humans during the task (Maglieri et al., ²⁰¹⁹). Unfortunately, as the Czechoslovakian Wolfdog was originally created by hybridising wolves with German Shepherd Dogs to produce a high‐endurance border patrol breed (Moravčiková et al., ²⁰²¹), any comparison with one of its originators (the German Shepherd Dog) and Labrador Retrievers, a gundog breed, cannot provide meaningful information about whether differences can be attributed to distance from wolf‐like ancestors of all dog breeds. Finally, Kovács et al. (2016) treated two dog breeds (Siberian Huskies and Border Collies) with intranasally administered oxytocin, to investigate whether their reactions were different in tasks where they had to interact with people. The authors found that the “social responsiveness” of Border Collies was correlated with a stronger effect of oxytocin, and argued that selection for cooperativeness in this herding dog breed could explain this effect, compared to the independently working sled dog Huskies. However, as only one representative breed from each breed group (cooperative versus independent working dogs) was chosen, it is not possible to exclude confounding effects such as a different evolutionary past (Huskies belong to the “ancient” breeds, while Border Collies are a more “derived” or “younger” dog breed; Parker et al., ²⁰¹⁷).

To avoid the limitations inherent in representing entire breed groups with a single or very limited number of breeds, an alternative is to test/survey as many breeds as possible within each cluster (e.g. Starling et al., ²⁰¹³). If the experimental design can ensure that none of the breeds become over‐represented within their clusters (Dobos & Pongrácz, 2023), this method offers an advantage: with many included breeds it is less likely that non‐targeted confounding factors will bias the results. However, the choice of grouping variable is still crucial. Here we stress the importance of biological relevance and ecological validity in ethological research. In some papers, breed groups were defined according to criteria from a purebred dog‐breeding organisation such as the AKC or FCI (e.g. Turcsán et al., ²⁰¹¹). Although such systems go some way towards defining groups based on the origin and function of the breeds, non‐scientific traditions of purebred dog breeding, and more recent breed introductions, have created a rather confusing structure across these categorisations. Although the FCI breed groups (Table 3) seem to have a functional structure, and these groups show considerable overlap with the genetic clusters established by Parker & Ostrander (2005), some of their groups are comprised of a bewildering array of breeds (e.g. Group 1, which contains almost all the herding and livestock‐guarding breeds), while others focus on a handful of very closely related breeds (e.g. Group 4, Dachshunds). The AKC breed groups (Table 4), which seem to be used most often in the surveyed literature, offer almost no biological relevance/ecological validity for a behavioural researcher. The AKC breed groups represent an odd mixture of straightforward functionality/origin (e.g. terriers, herding dogs); however, some of their breed groups are comprised of a difficult‐to‐categorise array of breeds. The AKC itself even admits their lack of clarity (https://www.akc.org/expert‐advice/lifestyle/7‐akc‐dog‐breed‐groups‐explained/): “The breeds of the Non‐Sporting Group have two things in common: wet noses and four legs”. It is therefore not surprising that it is difficult to attribute a cause to behavioural differences between breeds in the AKC's groups, apart from some trivial differences (e.g. terriers are “bolder” than herding dogs while herding dogs are more “trainable” than hounds; Turcsán et al., ²⁰¹¹). Indeed, it is hard to predict results for such complex groupings as “Non‐Sporting”, or “Sporting” dogs. In such studies, the group‐level results could easily be biased by a few similarly behaving and well‐represented breeds in an otherwise heterogeneous group. By contrast, the heterogeneous nature of clusters defined by dog breeding organisations could mask likely existing behavioural differences. Dorey, Udell & Wynne (2009) found no association between point‐following capacity and AKC breed group. Relying on the visual (pointing) signals of humans is a widespread and well‐documented capacity of dogs (Pongrácz et al., ²⁰¹³), and one might expect that the function of a dog breed could affect this behaviour. However, the breed groups of the breeding organisations often lump very different breeds together, which could blur potential differences. To summarise, it is likely that breed clusters from the large dog breeding organisations do not provide a biologically relevant grouping variable for ethological investigations.

Table 3.

The 10 breed groups of the Fédération Cynologique Internationale (FCI), which is the world's largest kennel club organisation based on the number of member countries. Most FCI groups are defined according to the dog breeds' original function, but some, for example Group 5, can also be considered as ancestry‐based cluster. For each group, we provide a “conventional” example that most people could easily assign to that group, and when possible, an “unconventional” breed that might be more surprising.

FCI group	Conventional breed example	Unconventional breed example
Group 1: Sheepdogs and Cattledogs (except Swiss Cattledogs)	Australian Shepherd	Czechoslovakian Wolfdog
Group 2: Pinscher and Schnauzer – Molossoid and Swiss Mountain and Cattledogs	Dobermann	Russian Black Terrier
Group 3: Terriers	Airedale Terrier	American Staffordshire Terrier
Group 4: Dachshunds	Dachshund	–
Group 5: Spitz and primitive types	Alaskan Malamute	Icelandic Sheepdog
Group 6: Scent hounds and related breeds	English Foxhound	Dalmatian
Group 7: Pointing dogs	Vizsla	Brittany Spaniel
Group 8: Retrievers – Flushing Dogs – Water Dogs	Labrador Retriever	–
Group 9: Companion and Toy Dogs	Bolognese	Standard Poodle
Group 10: Sighthounds	Greyhound	Irish Wolfhound

Table 4.

The seven breed groups of the American Kennel Club (AKC), which is the most significant dog breeding organisation in the USA. The AKC groups were primarily formed on the basis of breed purpose/function. For each group, we provide a “conventional” example that most people could easily assign to that group, and when possible, an “unconventional” breed that might be more surprising.

AKC group	Conventional breed example	Unconventional breed example
Sporting	Golden Retriever	–
Hound	Beagle	Basenji
Working	Rottweiler	Akita
Terrier	Border Terrier	Miniature Schnauzer
Toy	Pug	Italian Greyhound
Non‐sporting	Bulldog	Standard Poodle
Herding	Border Collie	Spanish Water Dog

(3). Is there any way to create ecologically valid multi‐breed comparisons?

The fact that hundreds of dog breeds are currently recognised is undoubtedly the result of human attempts to standardise and preserve distinct landraces of working dogs, mostly in the second half of the 19th century (Serpell & Duffy, 2014). Landraces originated as behaviourally useful working companions, aiding their owners in a multitude of tasks (Lord, Coppinger & Coppinger, 2014). Capitalising on the realisation that artificial selection could potentially result in fundamentally different socio‐cognitive capacities among working dogs, depending on their intended deployment, some researchers have investigated behavioural differences between the main clusters of working dogs. Comparison between “independent” and “cooperative” working dogs (Gácsi et al., ²⁰⁰⁹) has been an especially promising and widely applicable paradigm. Many independent breeds have been selected for tasks in which they have to operate either out of sight of their handlers, or without being regularly instructed. Cooperative breeds on the other hand, have been selected for tasks where they perform under close human supervision with regular instructions. The specific advantage of using this grouping is that both the “independent” and the “cooperative” groups are comprised of a wide variety of breeds, originating from often very distant genetic clades (Parker et al., ²⁰¹⁷; Dobos & Pongrácz, ²⁰²³). Thus, when the test groups are assembled from evenly represented and widely distributed breeds, any significant behavioural differences between the two groups are less likely to be confounded by within‐group genetic relationships among the breeds. Obviously, to ensure a reliable breed‐group effect, other factors such as the dogs' training level (Serpell & Duffy, 2014) and housing conditions (Lenkei, Pogány & Fugazza, 2019) should also be balanced across the groups. From a more theoretical point of view, an owners' choice of a particular breed is likely to be biased by their preference for not only a particular appearance, but also for the assumed behavioural and personality traits of that breed. In turn, owners will likely engage their dogs with joint activities that fit expectations for that particular breed. For example, cooperative breeds such as herding dogs and utility breeds (e.g. German Shepherd Dogs, Belgian Malinois) are preferentially used for activities that require high levels of attention for commands and signals (e.g. agility, obedience); independently working dog breeds are likely trained for activities where determination and stamina are needed (e.g. sled dogs in canicross and bikejöring; Staffordshire and Pitbull terriers in weight pulling). “Breed‐appropriate” training and activities may further strengthen particular behavioural phenotypes in the dogs as well as breed‐related stereotypes among people (e.g. Clarke, Cooper & Mills, ²⁰¹³).

Although the “independent” versus “cooperative” working breeds comparison offers an ecologically valid approach to the comparative analysis of behaviours that are connected to dog–human interactions, it is crucial that there is representative sampling of the dog breeds for a given study. If researchers only test a handful of breeds, the problems discussed above again arise where an over‐represented breed could bias the results. For example, Kovács et al. (2016) drew conclusions about differences in human‐directed gazing for Siberian Huskies and Border Collies based on the alleged effects of selection for cooperativity or independent work. However, using only a single breed per cluster cannot provide a generalisable effect for the whole group of breeds, as the randomly chosen two breeds differ in many other features (e.g. genetic distance from the common ancestor; typical housing conditions, training level), and not only in their work‐related ancestry. Similarly, the study of Pogány et al. (2018) employed only four dog breeds, three cooperative (Labrador Retrievers, Border Collies and Mudis) and an independent‐working breed (Beagles). The authors found differences between the breeds' reward‐maximising behaviour in a cognitive bias task, and concluded that this may be the consequence of the working style of the breeds. However, again, the unevenly distributed and very low number of breeds would preclude such ambitious explanations.

Nevertheless, some studies used high numbers of breeds for both functional clusters, allowing well‐supported conclusions about the typical behavioural profiles of independent and cooperative breeds. Gácsi et al. (2009) showed that cooperative breeds follow human pointing signals with higher success rates than independent working breeds. Lenkei et al. (2021a ) found that the capacity to form an attachment bond with the owner is invariably present in both breed clusters, supporting the idea that dog–owner attachment is a fundamental, species‐level attribute of dogs. Bognár et al. (2021) tested the time taken to establish eye contact with humans in breeds varying in work function, cephalic index and age. The authors found that cooperative dogs and mongrels established eye contact with humans more rapidly than did independent working dogs. Shorter headed dogs also established eye contact faster. This well‐designed study posed biologically valid questions, however it was somewhat limited by the relatively low number of breeds and purebred subjects, especially from the independent breeds. Recently, Dobos & Pongrácz (2023) showed that although both independent and cooperative breeds show similar performance in the V‐shape fence detour task, they behave differently when the task is demonstrated by a human experimenter. While cooperative dogs learned how to detour more efficiently, the independent dogs did not utilise the opportunity for observational learning. All four of these studies employed dozens of breeds from both clusters and the researchers took care not to over‐represent any of the included breeds. As a result, their findings can be considered as biologically relevant features of the investigated breed groups, resulting from selection for working companions with various levels of cooperativity.

(4). Can ancestry provide a biologically relevant basis for breed‐related behavioural comparisons?

In our survey we found several papers that based between‐breed comparisons on ancestry‐related features. While this method may suffer from the aforementioned problems if it is applied to a handful of breeds only (e.g. Udell et al., ²⁰¹⁴; Nagasawa et al., ²⁰²⁴), where genetic relationship‐based behavioural analysis is carried out on a large number of breeds (e.g. Morrill et al., ²⁰²²), or if sufficiently sampled breed clusters are compared (e.g. Hansen Wheat et al., ²⁰¹⁹), it can be a promising method for biologically relevant investigations.

Some large‐scale molecular‐genetic surveys targeted behavioural phenotypes and provided useful information about the genetic background of such behaviours, or socio‐cognitive capacities with high ecological relevance in comparisons of dog breeds. For example, based on the genetic analysis of more than 2000 dogs (with roughly half being purebreds), Morrill et al. (2022) found that behaviours showing the strongest between‐breed segregation belonged to clusters that characterise either the cooperative or the independent working dog types. The cooperative breeds scored highly on such genetically determined features as “biddability” and “easy to train”. Remarkably, in this study, single individual breeds could not be characterised with “breed‐typical” behaviours, suggesting that functional clusters of breeds could be key to understanding more recent tendencies in behavioural selection among dog breeds. Additionally, the increasingly prevalent segregation of behaviourally different “working” and “show” lines in many dog breeds may be responsible for “masking” previously established breed‐typical behaviour (Sundman et al., ²⁰¹⁶). Vaysee et al. (2011) performed genome‐wide mapping with a genotyping array of more than 170,000 single nucleotide polymorphisms (SNPs), looking for regions that show between‐breed differences in their associations with morphological and behavioural traits. Their sample consisted of more than 500 dogs from 46 breeds. The behavioural trait “boldness” (where the binary classification of “bold” versus “non‐bold” breeds was taken from Jones et al., ²⁰⁰⁸) showed a significant association with the breeds' body size and ear position: small breeds with pricked ears tended to belong to the “bold” category, while all but one of the sampled drop‐ear breeds were classified as “non‐bold”. Obviously, these results cannot establish reliable causative connections between the breed's behavioural phenotype and their selection past, for which researchers would need to control for the genetic relatedness of the breeds and also, for example, their traits of functional selection. Regarding morphological traits such as ear position, direct surgical alterations further complicate interpretation of the results of Vaysee et al. (2011), as the Boxer and the Dobermann breeds were classified as “bold”, but these dogs only become “prick eared” if their ears are cropped, as both breeds naturally have drop‐ears.

In some instances, the experimental design may prevent extrapolation of conclusions. Konno et al. (2016) compared representatives of various dog breeds in visual contact‐seeking tasks with humans and concluded that “ancient breeds” perform with lower intensity in these tasks than dogs they sorted into various other groups (“herding”, “working”, “retriever‐mastiff” and “hunting”). The authors explained their results in an evolutionary/ancestry‐based framework, arguing that looking at humans is weakly represented in wolves (Miklósi et al., ²⁰⁰³), thus it would likely also be weak in “ancient” dog breeds. However, their group of ancient breeds also differed from most of their other groups on the basis of cooperativity – another factor associated with dog–human visual interactions (Gácsi et al., ²⁰⁰⁹). In another study, Nagasawa et al. (2017) tested Japanese dog breeds belonging to the “ancient” cluster of breeds for gazing and contact‐seeking behaviour towards their owner after intranasal treatment with oxytocin. They hypothesised that breeds closer to their wolf‐like ancestors would show weaker human‐directed gazing. Although the authors did record human‐directed gazing after oxytocin treatment in the ancient Japanese breeds, as their study lacked any subjects from “Western” breeds as a comparison, it is difficult to draw firm conclusions. Ancestry‐based genetic clustering may not necessarily result in an explanation for variance in behavioural traits that involve complex social understanding and interactivity with humans. Dorey et al. (2009) performed a meta‐analysis of previous publications, and found that the success rate of dogs' point‐following behaviour was not associated with the four ancestry‐based clusters defined by Parker & Ostrander (2005). Recall that point‐following capacity was associated with purebred dogs' assignment to either the independent or the cooperative breed clusters (Gácsi et al., ²⁰⁰⁹). This perhaps explains why the genetic clusters did not provide similar results, because even closely related breeds have been selected for very different working tasks. For example, many sighthounds and herding dogs are closely related (Parker et al., ²⁰¹⁷), but the former breeds were selected for ability to work independently, whereas the latter work cooperatively with their handler. The potential role of functional (work‐related) selection can be inferred from the investigation of Gnanadesikan et al. (2020), who estimated the heritability of cognitive features across breeds. They used the citizen science behavioural database “Dognition.com”, from which they extracted records of more than 1,500 purebred dogs from 36 commonly occurring breeds. Using genetic relatedness data from Parker et al. (2017), they found the highest heritability indices for the behavioural traits “inhibitory control” and “communication”. Gnanadesikan et al. (2020) speculated that the strong heritability of these traits might be the result of their crucial role involvement in domestication. It is our opinion that an alternative hypothesis could be that these traits have been strongly affected by more recent directional selection for particular working tasks. Posing an ecologically valid research question is of crucial importance when researchers apply ancestry‐based (genetic) methods for behavioural breed comparisons.

A clear and biologically relevant hypothesis was presented by Smith, Browne & Serpell (2017), who compared the problematic behaviours of dingoes kept as companions with modern and ancient dog breeds. They used thousands of records from the validated “Canine Behavioural Assessment and Research Questionnaire” (C‐BARQ) instrument and found that dingoes (“primitive dogs”) represented an outlier, even when compared to the ancient breeds. Dingoes were less trainable, showed more human‐directed fear, and a higher propensity to roam and escape, than ancient and modern dogs. The authors concluded that the behaviour of dingoes may reflect a very early domestication effect, followed by natural, rather than artificial selection. Dingoes are only recently and still rarely kept as companion dogs, with environmental factors such as socialisation, housing and training conditions also potentially relevant to their observed behavioural traits. In a similarly well‐designed study, Hansen Wheat et al. (2019) used an immense database of the Swedish Kennel Club containing results from almost 80,000 purebred dogs tested using the “Dog Mentality Assessment” (DMA) procedure. The DMA consists of a series of behaviour tests and has a validated profile of the evaluated behavioural traits (Svartberg, 2005). Hansen Wheat et al. (2019) hypothesised that there would be a difference between the “ancient” and “modern” dog breeds' behavioural traits typical for the domestication syndrome. According to this syndrome (Kaiser, Hennessy & Sachser, 2015), domesticated animals show a general increase in sociability, and a decreased reactivity profile (e.g. lower aggression and fear levels). In a comparison of the DMA results from seven ancient and 71 modern dog breeds, Hansen Wheat et al. (2019) found that ancient dog breeds show a consistent relationship between sociability and reactivity as expected according to the domestication syndrome. However, the two main traits were decoupled across the modern breeds (Hansen Wheat et al., ²⁰¹⁹). As ancient breeds are thought to be relatively unchanged across longer time periods, together with a considerable admixture with wolves (VonHoldt et al., ²⁰¹⁰), these results fit with the hypothesis that the earliest (“ancient”) dog population showed more uniform effects of domestication. Modern breeds are the products of more recent functional and morphological selection (Parker et al., ²⁰¹⁷). Consequently, this study convincingly drew attention to the more specific needs of humans for specialised working companions, whereby sociability and reactivity were sometimes given different importance in modern breeds.

(5). Dog breed comparisons in light of applied ethology

A large proportion of publications in this survey focused on behavioural traits with applied relevance, that is they are important from the aspects of choosing, training or co‐existing with dogs as working or companion animals. Breed comparisons have great relevance from this point of view. Although the applied investigations are often constructed around a well‐defined research question and hypothesis, we still found issues regarding study design/breed choice. Below we summarise the main tendencies and typical approaches used.

(a). Problematic behaviours

When surveying and comparing the behavioural characteristics of dog breeds, researchers often try to find a solution to behavioural problems. This includes helping owners choose proper breeds for their needs (e.g. Notari & Goodwin, ²⁰⁰⁷), or surveying populations for the potential cause of problematic behaviours (e.g. “dangerous dog breeds”; Hammond et al., ²⁰²²). This is obviously a complex situation, as membership of a particular breed cannot be a fully reliable predictor of (problematic) behaviours. Environmental factors [e.g. early‐life environment (Lenkei et al., ²⁰¹⁹); socialisation (Martínez et al., ²⁰¹¹); traumatic events (Wallis, Szabó & Kubinyi, 2020); training (Burghardt, 2003)]; as well as within‐breed artificial selection for distinct lines of “working” and “show line” types (Fadel et al., ²⁰¹⁶) can generate considerable within‐breed behavioural variability. Behavioural issues often influence the relinquishment of dogs to shelters (Patronek, Bradley & Arps, 2022) and their chances of adoption also depend on the potential adopting owners' knowledge about their behaviour. Wright et al. (2007) found that exposure to a video clip of an aggressive (misbehaving) dog, enhances the formation of negative breed stereotypes for the breed in the video (and similar dogs). The limitation of this study design was that they used only four dog breeds (German Shepherd Dog, Bloodhound, Border Collie, Pointer), and only the German Shepherd Dog was used for the video clip.

Random (convenience) breed sampling can lead to problems with interpretation of results in some studies. Bellamy et al. (2018) investigated the genetics of fearful behaviour in Bichon Havanese dogs (a breed with considerable variation in fearful/outgoing behaviours). They found that associations of two SNPs at the DRD2 gene increased social fear in the Havanese. When they tried to link associations of these SNPs with noise reactivity in an additional four breeds (Standard Poodle, Nova Scotia Duck Tolling Retriever, Irish Soft Coated Wheaten Terrier, Collie), they found contradictory results (positive versus negative associations in two versus three breeds). The low generalisability of their results could be connected to a seemingly random choice of breeds.

Large sample sizes often provide a more representative picture; however, convenience sampling remains problematic from the aspect of uncontrolled confounding factors. In a questionnaire study, Storengen & Lingaas (2015) analysed the responses of more than 5000 owners regarding the noise‐sensitivity of their dogs. Breed choice was arbitrary, and was based on the willingness of 17 breed clubs' to participate. Although the authors reported a significant breed effect of noise sensitivity, the dogs' age, sex and reproductive status, as well as sensitivity to separation, acted as confounding factors. Thus, the identified breed differences could not be explained on the basis of ancestry or breed function.

As problematic behaviours usually have a complex aetiology and causative background (e.g. separation‐related problems; Lenkei et al., ^2021b ), it is potentially hard to identify biologically relevant associations between them and individual dog breeds. McGreevy et al. (2013) used the C‐BARQ instrument (with more than 8000 responses concerning the 49 most popular breeds in Australia) to compare behavioural problems with breed body mass, height and cephalic index. They found that unlike trainability, problematic behaviours increased with decreasing body size. Also, shorter headed dogs exhibited more allogrooming and a lower propensity to chase. However, as both dog size and cephalic index can be associated with multiple other factors (e.g. breed function, housing conditions, longevity; Turcsán & Kubinyi, ²⁰²⁴), their findings remain difficult to interpret without specific follow‐up studies. Regarding their results, the authors aptly concluded in their paper “The biological basis for, and significance of, these associations remain to be determined.” (McGreevy et al., ²⁰¹³). However, hypothesis‐driven approaches can provide biologically relevant results even for proneness to particular behavioural problems. Pongrácz, Gómez & Lenkei (2020) tested whether dogs from cooperative or independent working breeds show more signs of stress during a short separation period from their owner. Cooperative working dogs reacted more quickly and with more frequent distress vocalisations, which may be the consequence of their more pronounced preference for remaining in the vicinity of their owner.

(b). Working dogs

During the last century, even dog breeds that originally existed only as utility (working) animals, have become companions. However, there is still a need for working dogs in various fields, and some traits can be important for maintaining the efficiency of these populations. In our survey we encountered mostly well‐designed evaluations of working dogs and their relevant behavioural phenotypes. Kinka & Young (2018) compared three Old World livestock guarding breeds with American Whitedogs (mixed‐breed livestock‐guarding dogs, developed in North America, from Eurasian livestock guarding breed founders) in their working style and efficiency with sheep, and their responses to simulated wolf attacks. The authors found that all breeds were suitable for the task, with only small between‐breed differences.

As the capacity to detect odour is an important factor for scent‐work, Polgár et al. (2016) compared the olfactory performance of three groups of dog breeds (scent hounds, other working dogs, brachycephalic breeds) and socialised wolves in a natural scent‐detection task. With a wide selection of breeds included, they found that scent hounds and wolves outperformed the other breeds, showing the effects of natural (in the case of the wolf) and artificial selection (in the case of the scent hounds) on olfactory performance. Another important feature in working dogs is their trainability. Knowing the association of trainability with various inherited factors could provide important insights for selecting the best candidates for costly training programs. Helton (2009) investigated the relationship between dogs' head shape and their performance in various tasks requiring training. Trainability of the breeds was categorised by using the Coren (1994) ranking scale. The results showed that breeds with high trainability were mostly mesocephalic, while both brachy‐ and dolichocephalic breeds fell into the “less trainable” category. Helton (2009) concluded that selecting for fighting (shorter heads) or running fast during hunting (longer heads) are specialised functions that may require less training. This study utilised an interesting functional approach with a well‐outlined biological relevance. Among its limitations are the questionable validity of the scoring system and the using of cephalic index as a categorical factor instead of as a continuous variable (see Georgevsky et al., ²⁰¹⁴).

(c). Interacting with dogs

Breed comparisons can provide insights into behavioural and personality characteristics of importance to dog owners, veterinarians, and even legislators (Kogan et al., ²⁰¹⁹). For example, distinct “working” and “show/companion” lines started to appear in several dog breeds during the last century, and it may be important to understand whether they exhibit consistently different behaviours. Fadel et al. (2016) used the results of the “Dog Impulsivity Assessment Scale” (DIAS) test battery on over 1000 Border Collies and Labrador Retrievers to investigate whether impulsivity scores differed between show and working lines in these breeds. They found that Border Collies were more impulsive than Labradors, but only for the working lines. We can conclude that recent breeding efforts for show/companion types in some typical working breeds may have relaxed breed‐typical behavioural stereotypes (like impulsivity characteristics). Sundman et al. (2016) also evaluated thousands of Golden and Labrador Retrievers, finding different curiosity‐score distributions, between their recently diverged “hunting” and “show” lines.

Companion dogs with “extreme features” are very popular, despite many health and behavioural issues that negatively affect the welfare of both the dogs and their owners. Sandøe et al. (2017) used a large‐scale questionnaire survey (nearly 1000 respondents), in which they compared owners' opinions about four small, popular dog breeds in Denmark. In this well‐designed study they included two breeds with extreme features (Chihuahua, French Bulldog), the Cavalier King Charles Spaniel that suffers from many health issues, and Cairn Terriers that are a generally healthy breed. Intriguingly, they found that owners of problematic breeds cared less about health problems and would choose the same breed again. By contrast, owners of Cairn Terriers considered health as being more important. The study highlights that a preference for popular breeds with serious health conditions often reflects active and conscious decisions, rather than lack of knowledge from the owner.

It is often thought that particular features could help to sort dog breeds into particular types, which would be advantageous for prospective dog owners to choose their best‐fitting companion. A frequently encountered personality/behavioural dimension in ethology and behavioural ecology is the shyness–boldness continuum (e.g. Wilson et al., ¹⁹⁹⁴). Starling et al. (2013) used the C‐BARQ instrument to investigate the “boldness supertrait” in dogs owned by over 1000 respondents in Australia. They used the breed group structure of the UKC (United Kennel Club, USA), which is similar in its functionality to the FCI system but has fewer groups (more suitable when researchers have fewer subjects). Principal Components Analysis established one main trait that had positive loadings on items describing playfulness with humans and other dogs, and negative loadings on items about fear. By including more breeds than is typical in other surveys (e.g. ancient dogs, sighthounds), the study added new details to the characteristic dimensions of the breeds, for example, playful interactions with other dogs. Between‐breed, and between‐group differences were established on the shy–bold continuum. This study exemplifies the usefulness of considering a wide selection of breeds, and the advantages of functional groupings. At the same time, “supertraits”, such as boldness, might contain multiple items, making it hard to interpret their meaning in everyday practice.

Breed‐clustering is an often‐used practice that allows researchers to investigate questions with practical applications. While inherited characteristics are usually the focus, environmental effects, such as the training history of dogs, can represent a decisive confounding factor. Marshall‐Pescini, Frazzi & Valsecchi (2016) investigated the effect of training and breed group on dogs' problem‐solving capacity in the V‐shaped detour and a puzzle box condition. While breed group (ancestry‐based; VonHoldt et al., ²⁰¹⁰) did not affect performance in the detour test, trained dogs were more successful than untrained ones. For the puzzle box task, training enhanced dogs' success, and “working dogs” performed better than retrievers and herding dogs. Although this study was based on a biologically relevant question and methodology, the breed choice showed signs of convenience sampling, as all four groups consisted of only 2–3 breeds, and almost all breeds in each group represented the “cooperative” functional cluster.

V. DISCUSSION AND SWOT ANALYSIS

We found a large number of research papers where authors performed comparative analysis of behavioural phenotypes between dog breeds. The range of targeted breeds numbered from a handful to many dozens, and the targeted behaviours were relevant from aspects of both applied and fundamental research. We identified three main strategies of research design regarding the chosen methods the researchers used to select dog breeds. Two variants of hypothesis‐driven approaches (based on either the function of the dog breeds, or their genetic relatedness) and convenience sampling (based on the availability/popularity of the dog breeds) were equally common. A closer look at the combination of the initial aims of the reviewed papers with the methods of breed choice and the conclusions drawn from their results, allowed us to identify both positive aspects and weaknesses of these methods. Below, we present a detailed SWOT analysis, in which we highlight the characteristics of “best practice” alongside caveats connected to the three approaches for the choice of dog breeds (Table 5).

Table 5.

Overview of our findings regarding the methods used to choose dog breeds in the surveyed publications. The three main experimental designs were then evaluated according to the SWOT (Strengths, Weaknesses, Opportunities, Threats) principle (Van Belle et al., ²⁰²⁴).

Experimental design	Assessment according to the SWOT principle
Hypothesis‐driven, functional breed sampling	Strengths: based on biologically relevant hypotheses, which, in theory, should result in findings of ecological or practical (applied) validity.
	Weaknesses: some dog breeds will inevitably be omitted if they cannot be sorted clearly into a functional group.
	Opportunities: a wide range of hypotheses can be tested, for both fundamental and applied themes. The method is equally applicable for wider (breed‐group based), or narrower (targeted to a few breeds only) investigations – although, in the latter, the question to be investigated should be specifically tailored to the involved breeds.
	Threats: using only a few breeds, or allowing some breeds to be over‐represented in breed‐group comparisons can lead to confounding breed‐specific factors other than the targeted functional variable(s). Unless the sample is assembled with careful balancing, confounding factors such as training level or housing conditions can cause breed‐ or breed group‐related bias.
Hypothesis‐driven, ancestry‐based breed sampling	Strengths: molecular‐genetic studies provide a firm evolutionary (relatedness‐based) background, which can be effectively utilised as a predictor variable for a wide array of behavioural phenotypes.
	Weaknesses: ancestry‐based approaches are usually not readily applicable when the behavioural phenotype has emerged more recently (e.g. during the development of modern dog breeds). This method is less suitable for analysing behavioural differences between closely related dog breeds.
	Opportunities: the method is especially suitable for meta‐analyses, where already existing, large data sets from robust behavioural assessments can be compared with ancestry‐based, between‐breed relationships (i.e. genetic distances between breeds; distinct “clades” of breeds).
	Threats: the behavioural phenotype should be carefully selected, as in some cases, recent functional breed selection can represent a serious confounding factor. Dog breeds with similar inclination for dog–human interactivity levels (e.g. cooperative versus independent working dog breeds) can exist in distant genetic clades. Over‐representation of particular breeds in the sample, or when the sample consists of only a few dog breeds, can bias the results as this magnifies the impact of potential confounding (non‐ancestry‐related) factors.
Convenience sampling (“popular” breeds)	Strengths: this method is the easiest to apply because no dog breeds have to be excluded from the analysis. The sample usually represents the dog population well in a given area and time.
	Weaknesses: differences are difficult (and often at best tentative) to associate with explanatory variables. The ad‐hoc breed assortment in the sample may even conceal breed‐ or breed group‐related differences that would be possible to detect with a hypothesis‐driven approach.
	Opportunities: this is a good method for exploratory investigations, where, for example, the “presence–absence” status of particular behaviours is of interest. The method is also suitable for assessing the preferences, experiences, or opinions of respondents about dog breed‐related behavioural phenomena.
	Threats: even with inclusion of numerous breeds in the sample, there is a strong chance that various confounding variables (e.g. functional breed selection, genetic relatedness/distance, housing conditions, training level, etc.) will impact the results. Convenience sampling should not be used to draw conclusions which would only be possible after hypothesis‐driven breed choice.

From this detailed review of the sampled publications, and within the framework of SWOT analysis, it is clear that comparative research on the behaviour of dog breeds includes multiple purposes and distinct research designs. However, to find biologically meaningful results, it is important to identify the potentials and limitations of each method, because this will fundamentally affect the scope and validity of any conclusions drawn. Therefore, during the selection of breeds for study, certain considerations should be taken into account.

The most relevant threats to each of the reviewed research designs were confounding factors that can hamper the applicability and relevance of the results. These can be demographic and environmental features of the dogs, such as their age, sex, housing conditions, and training levels. Unless these potential confounding variables are the target of research (e.g. age of the dog in Chapagin et al., ²⁰¹⁷), their effects should minimised, for example by balanced recruitment of subjects. Serious issues can arise when the targeted behavioural phenotype can be affected by mechanisms other than those under consideration. Convenience sampling is inherently sensitive to both ancestry‐ or breed function‐based effects, while ancestry‐ and function‐based investigations can mutually confound the other's effect. These issues can be avoided with a well‐designed breed‐choice plan where a sufficient number of breeds are included and there is no over‐representation of particular breeds. Additionally, a common weakness noted in many of the reviewed methods was that the choice of breeds did not serve the intended goals of the research.

Convenience breed sampling – choosing breeds randomly, based on their availability – is an explorative method, and therefore suitable for research questions aiming to identify the presence or absence of certain behaviours in the population (e.g. prosocial and reactive behaviours; Hansen Wheat et al., ²⁰¹⁹). As this method is easy to apply, a wide variety of breeds can be included, large sample sizes achieved (e.g. Morrill et al., ²⁰²²), and the sample is likely to be representative of the dog population in a particular location, or time frame. However, the conclusions that can be reached are limited, because even with a large number of breeds, various confounding variables can bias the results (e.g. original breed function, differences between individuals of the same breed; Feddersen‐Petersen, ²⁰⁰⁰). Many breed‐related differences would be better investigated using a hypothesis‐driven approach.

Hypothesis‐driven sampling is the recommended method to examine the effect of domestication and/or subsequent artificial selection on dogs' behaviour. Ancestry‐based molecular‐genetic studies can provide ecologically valid information about the genetic background of behavioural phenotypes (Vaysse et al., ²⁰¹¹). For these questions, a combination of convenience sampling with analysis of genetic relatedness could allow for investigation of a large number of breeds with some biological relevance (Serpell & Duffy, 2014), but the use of hypothesis‐driven sampling of targeted breed clusters would be optimal to derive more precise and evolutionary valid conclusions (Smith et al., ²⁰¹⁷). An ancestry‐based method is especially suitable for meta‐analyses of between‐breed differences, and has gained popularity in recent empirical research (e.g. dogs following human pointing gestures; Dorey et al., ²⁰⁰⁹). However, when drawing conclusions based on genetic relatedness, besides the effects of domestication and early genetic‐clade formation based on certain behavioural traits, it is also important to consider the potential effects of more recent artificial selection (Gnanadesikan et al., ²⁰²⁰), as breeds with a similar inclination for dog–human interactivity levels (e.g. cooperative versus independent breeds) can be found in distant genetic clades (Parker et al., ²⁰¹⁷; Morrill et al., ²⁰²²; Dobos & Pongrácz, ²⁰²³). This method is less suitable for analysing behavioural differences between closely related breeds (Dorey et al., ²⁰⁰⁹).

Functional sampling is based on biologically relevant hypotheses, and should result in valid conclusions if a wide variety of breeds is tested (Bognár et al., ²⁰²¹). As “independent” and “cooperative” groups often include breeds originating from distant genetic clades (Parker et al., ²⁰¹⁷; Dobos & Pongrácz, ²⁰²³), any between‐group differences should not be confounded by within‐group genetic relationships among breeds. However, some breeds will be omitted if they cannot be sorted clearly into a functional group. Future researchers should attempt to identify variables that provide a suitable theoretical framework for the selection that resulted in dog breeds without well‐defined work functions. A wide variety of breeds are allocated to the “toy” or “companion” categories, but even traditional working dog breeds are often divided into “show” and “working” lines (e.g. Fadel et al., ²⁰¹⁶). It would be interesting to investigate whether there is any resemblance between the behaviour of the “toy/companion” dog breeds and the “show‐line” working dogs, because this might indicate that recent selection for “being a good companion” in many of traditional working dogs, could result in behavioural features that were present long ago in “toy” breeds.

Functional working breed selection was not the only common feature that we observed among publications using clearly outlined (non‐ancestry‐based) hypotheses. Applied behavioural research can target an almost infinite array of research questions, but the same principles of best practice still apply. Based on our SWOT analysis (Table 5), a concise outline for best practice in breed‐comparison experimental designs is: (i) use hypothesis‐driven (planned) methods when selecting breeds (or groups of breeds) for a study; (ii) unless the goal is a comparison of a few specific breeds, use well‐balanced and not over‐represented breed groups with several breeds in each; (iii) pay attention to potential confounding variables, such as training level, housing methods, age, sex, and neuter status of the dog; and (vi) in the case of ancestry‐based and functional selection‐based approaches, take steps to avoid these factors being mutually confounding. If the breed choice fits the chosen goals/hypotheses well, even where a small number of breeds is used, biologically relevant results can be obtained (e.g. curiosity‐related behaviours in show and working lines of Labrador and Golden Retrievers; Sundman et al., ²⁰¹⁶). However, higher sample sizes and representatively chosen breeds will usually be a better solution for hypothesis‐driven applied research (e.g. when comparing pain sensitivity of dog breeds based on empirical methods versus the opinion of veterinarians; Caddiell et al., ²⁰²³). Opting for a non‐representative sample, or a narrow breed choice, can lead to potentially confounding effects (e.g. clicker trainability in three dog breeds of “different purpose”; Strychalski, Gugołek & Konstantynowicz, ²⁰¹⁵).

For any of the sampling methods, choosing only a few breeds can result in bias as the characteristics of the over‐represented breeds may lead to an influence of potential confounding factors (e.g. the markedly different ancestry of Huskies and Border Collies; Kovács et al., ²⁰¹⁶). However, even with large numbers of breeds, the sample should be carefully balanced for possible effects of demographic and environmental factors (e.g. housing conditions, training levels). Low numbers of breeds in a study may only be appropriate when there are high sample sizes or for specific research questions focused only on those particular breeds (e.g. work‐line and show‐line comparisons; Fadel et al., ²⁰¹⁶). Additionally, the use of FCI or AKC groupings is biologically problematic from both a functional and evolutionary point of view (Hecht et al., ²⁰²¹), as these grouping methods may have little ecological validity.

VI. CONCLUSIONS

• (1)Convenience sampling of dog breeds is recommended only for broad, explorative assessments of the presence/absence of particular behavioural phenotypes.
• (2)Researchers should attempt to reach ecologically valid conclusions of biological relevance, with regard to comparisons between breeds or breed groups, which can be achieved by hypothesis‐driven study design, either via ancestry, or a function‐based approach.
• (3)The main weakness of comparative breed research can be the inappropriate (e.g. non‐representative, too narrow, skewed) inclusion of breeds.
• (4)For all reviewed methods, significant threats to the validity of the results can be posed by confounding variables, such as environmental factors and demographics, or by alternative hypotheses.
• (5)The most important aspect when planning breed‐comparison experiments is that the collection, quantity, and classification of the breeds, should be consistent with the aims of the research and provide a biologically relevant background for the confirmation of the hypotheses.
• (6)Unless specific research aims require the inclusion of particular breeds, researchers should include as many suitable breeds as possible, without over‐representation of any, in both applied and fundamental research.
• (7)We propose the use of representative, hypothesis‐driven research designs in dog breed behavioural research, preferably with a balance between ancestry and breed function‐based considerations.