From Inconsistent identification of pit bull-type dogs by shelter staff

A key finding of this study was that agreement among different shelter staff members evaluating the breeds of the same shelter dogs at the same time was only moderate. Lack of consistency among shelter staff in breed assignment confirmed that visual identification of pit bull-type dogs was unreliable.

There is no standardized breed signature for the mixed breed dog known as the ‘pit bull,’ and the surrogate DNA breed signatures used in this study were for the American Staffordshire terrier and the Staffordshire bull terrier. One in five dogs genetically identified with pit bull heritage breeds were missed by all shelter staff at the time of the study. One in three dogs lacking DNA evidence for pit bull heritage breeds were labeled pit bull-type dogs by at least one shelter staff member.

These findings are consistent with previous reports of poor inter-observer agreement among individuals attempting to identify the predominant breeds of dogs. In a large Internet survey, a national sample of 5922 self-identified ‘dog-experts,’ including breeders, exhibitors, trainers, groomers, behaviorists, rescuers, shelter staff, veterinarians, and veterinary technicians, was recruited to complete an anonymous Internet survey in which they selected the most likely breed for dogs depicted in photographs.1 One hundred dogs were included in the Internet survey, and each respondent was randomly shown photographs (front facial and lateral whole body) of 20 of these dogs. Based on the photographs and information about the height, weight, sex, and age of each dog, respondents selected from a drop-down menu of 181 breed options, including ‘no predominate breed.’ An average of 53 different breeds was selected for each dog, ranging from a low of 11 breeds selected for a purebred Beagle to a high of 84 different breeds for a single mixed-breed dog. In another study, 923 survey takers involved in dog-related professions and activities watched 1 minute color videos of 20 different dogs, and based on the images and information about age, weight, and sex, recorded one or two predominant breeds or ‘mix’ (Voith et al., 2013). Agreement among survey participants was poor, with at least half of respondents agreeing on the breed for only 7/20 dogs.

Our findings are also consistent with previous reports comparing visual breed identification with results of DNA breed profiles. In a study of 20 dogs adopted from 17 different agencies, the agency's breed designation matched DNA breed profiles in only four dogs (Voith et al., 2009). In the subsequent study using videos of the same dogs, visual breed identifications matched DNA results less than half of the time in 14/20 dogs in the study (Voith et al., 2013). Dogs were selected for the Internet survey if they were reported to have at least one breed that comprised at least 25% of their DNA profile.1 Visual identifications were considered correct if at least one named breed matched at least one breed in the DNA profile. On average, visual breed identifications matched DNA breed signatures for only 27% of dogs; 6% of dogs were never correctly identified. Although these previous studies included dogs with pit bull-type DNA breed signatures in 10% (Voith et al, 2009, Voith et al, 2013) and 23%1 of the dogs tested, respectively, the topic of identification of pit bull-type dogs was not specifically discussed.

Participants in two of the studies overestimated their ability to correctly identify breeds visually. In the Internet survey, 68% of respondents predicted they would correctly identify breeds at least half of the time, but only 4% actually did.1 In the study using videos, after the survey was completed, respondents attended an educational session in which the heredity of phenotypic attributes was discussed and images of breed crosses that looked nothing like their parents were displayed (Voith et al., 2013). Despite being presented with evidence of the poor correlation of physical appearance with breed composition in mixed breed dogs, some respondents clung to their opinions that the DNA results must be wrong; authors of the study called for the completion of similar studies to confirm the findings in additional dogs (Voith et al., 2013).

From Defaming Rover: Error-Based Latent Rhetoric in the Medical Literature on Dog Bites

Dog-breed blaming occurs when authors generalize without providing suitable supporting evidence that members of certain breeds are responsible for a disproportionate number of “severe” dog bites or “attacks” (e.g., Ching, Adelgais, & Daane, 2004; Rohrich & Reagan, 1999). Authors describe certain breeds as “bite prone” (Gurunluoglu, Glasgow, Arton, & Bronsert, 2014), “offending” (Horswell & Chahine, 2011), or “high risk” (Schalamon et al., 2006), but do so with flawed support.

“Pit bulls,” “pit bull terriers,” or “pit bull mixes” were frequently blamed “breeds” for dog bites in the United States (e.g., Prendes, Jian-Amandi, Chang, & Shaftel, 2016) and United Kingdom (e.g., Kasbekar et al., 2013). Elsewhere, other breeds were blamed for biting more often and more severely, such as German Shepherd dogs (called Alsatians in some British papers) in Australia and Scotland (Greenhalgh, Cockington, & Raftos, 1991; Klaassen, Buckley, & Esmail, 1996; Thompson, 1997). Other articles on bites in the United States point to breeds such as German Shepherd dogs, Rottweilers, Chow Chows, Poodles, or simply mixed breeds as the major offenders (e.g., Lauer, White, & Lauer, 1982; Morton, 1973; Pinckney & Kennedy, 1982; Steele et al., 2007).

Such generalizations about “pit bulls” are particularly problematic (e.g., Viegas, Calhoun, & Mader, 1988) since this is not a breed or even a term with a single definition. This unofficial descriptor refers to a group that may include dogs of multiple recognized breeds, dogs believed to have those breeds to some degree in their ancestry based on someone’s idea of a physical resemblance, and mixed-breed dogs whose ancestry cannot be reliably identified (Patronek, Sacks, Delise, Cleary, & Marder, 2013). Nevertheless, articles use the expression “pit bull” or “pit bull type” in the same context in which they discuss recognized breeds of dogs (e.g., Daniels, Ritzi, O’Neil, & Scherer, 2009).

The claim that certain breeds are more likely to bite also assumes the reliability of calculating “relative risk,” a statistic that conveys the likelihood of an incident (e.g., a dog bite) being caused by a member of one group (e.g., a breed of dog) versus a member of another group (e.g., Gurunluoglu et al., 2014). The American Veterinary Medical Association Task Force on Canine Aggression and Human–Canine Interactions (2001) reported that relative risk calculations with respect to dog bite–related injury are impossible due to the lack of accurate information about the breed of the biting dog in question (numerator) or that of the comparator group (denominator). The pedigreed portion of the U.S. dog population has been estimated at only 54% (American Veterinary Medical Association, 2012) and dogs of mixed heritage are not members of breeds, nor can they be expected to significantly resemble their ancestral breeds in appearance (Parker et al., 2004; Scott & Fuller, 1965).

We have also found examples where HHCPs made generalizations about breed from data they collected, but the data precluded generalizing. For example, Dwyer, Douglas, and van As (2007) claim that “pit bull terriers” and German Shepherd dogs—presumed to be correctly identified—were the most common breeds in their study to “attack” children, but they note that only 1% of their sample reported the presumed breed of the dog, a percentage far too low to permit generalizations about the other 99% of the sample. Similarly, Kasbekar et al.’s (2013) study of dog bites to children blamed Staffordshire Bull Terriers as being the most frequent breed to bite, but analysis of the study’s data shows that only 30% of their sample included a report of breed: again, a percentage too low from which to reliably generalize. Bini et al. (2011) acknowledged that a low proportion of cases with any information about breed was a limitation of their study but nevertheless made unwarranted claims about the role of breed.

Another source of inaccuracy about breed is the breed identification itself, such as when presumed breed is self-reported by bite victims or their family, friends, or others. For example, Kasbekar et al. (2013) identified breeds by relying on the child’s guardian or reports from the child’s doctor, and Lang and Klassen (2005) could not report the source of breed identifications but thought they “probably” came from victims’ families except in cases where the dog was unknown to the victim. One study compiled breed data not based on the dog that inflicted the bite but by showing a series of pictures of dogs to bite victims and asking which picture resembled the dog they believed had bitten them (Jarrett, 1991).

Data relating to presumed breed may also be unreliable when obtained either from news accounts or hospital records, where the authors assume that whatever is written is reliable evidence (e.g., Chiam, Solanki, Lodge, Higgins, & Sparnon, 2014; Sacks, Sinclair, Gilchrist, Golab, & Lockwood, 2000) and fail to integrate into their data analysis the difficulty of accurate breed identification, a problem long recognized and demonstrated even now among dog professionals (Olson et al., 2015; Voith, Ingram, Mitsouras, & Irizarry, 2009; Voith et al., 2013). Nor did these studies attempt to address the problem through pedigree documentation.