Scientific Studies

Some food for thought before we delve into this complicated subject, stated elegantly by an expert on dog attack statistics:

There are many studies and data related to dog bites and dog attacks. However, dog attacks on humans occur in the course of complex interactions between two sentient beings and occur in the most uncontrolled and unscientific settings, involving dozens of variables and circumstances which cannot be measured accurately.

For these reasons, there is no “science” behind any of the studies conducted on cases of dog attacks.—Karen Delise (NCRC website, Dog Bite Statistics: Science or Junk Science?)

Dog Attack Statistics: A Primer

by J. Thomas

“There are three kinds of lies: lies, damned lies, and statistics.”—Benjamin Disraeli

Journalists who wish to use dog bite statistics in their articles must be familiar with proper interpretation of those statistics and must also have a fundamental understanding of the flaws that are inherent in the most popular dog attack studies available today. Unawareness of these flaws almost certainly guarantees misinterpretation of the numbers. Many journalists have already fallen into the traps presented by these studies, either by drawing their own incorrect conclusions or by perpetuating another journalist’s mistakes.

Where do the numbers come from?

There is no uniform dog bite reporting procedure, nor is there a national agency charged with collecting such data. Dog bite data is collected and reported haphazardly. Animal control departments, hospitals, law enforcement agencies, and state health agencies may all collect different types of data, or none at all.

For example, when this author contacted Jim Schuermann, Staff Epidemiologist (Zoonotic and Vectorborne Diseases) of the Infectious Disease Epidemiology and Surveillance Group of the Texas Department of State Health Services, in March 2005 to inquire why the state of Texas no longer tracked dog bite statistics, he replied:

I’m sorry to report that this program has been discontinued. There was never a legislative mandate which required animal control agencies to submit reports on Severe Animal Bites. Although the Zoonosis Control Division highly encouraged all animal control agencies to voluntarily submit these reports, we would receive only 500 to 600 reports a year, and none of the major metropolitan areas (Amarillo, Brownsville, Dallas, El Paso, Fort Worth, Houston, Laredo, Lubbock, or San Antonio) participated. As I’m sure you can imagine, without data from these cities the information we did manage to collect was severely limited in scope. We completed a program review and realized that the information collected could not be proven to be scientifically representative of the entire state and that analysis of the small data set we did collect did not reliably reflect statewide trends.

Because fatal dog attacks are so rare, they are slightly easier to track, and the CDC attempted one major study using fatal dog attack statistics (see below for analysis of the study). However, because fatal attacks represent an infinitely small fraction of all attacks, they cannot be considered representative of all dog attacks.

Several studies have attempted to gather dog bite information from news articles and other news media; in fact, Merritt Clifton’s often-referenced dog attack “study” relies entirely on news media. This is a particularly unreliable source of information about dog attacks, as shall be explained next.

First, not all dog attacks are covered by the media, and it is unclear what criteria is being used by the news media to decide whether or not to report a particular attack. If all dog bites were reported, approximately 915 dog bites would be featured in the news every day. To date, no one is tracking all of these dog bites.

And second, the news media frequently misidentifies breeds and types of dogs. Breed identification is next to impossible, especially in the case of mixed breed dogs (the majority of dogs in the U.S.). It often consists of a wild guess based on appearance, not DNA testing or any truly scientific method of identification. Journalists may record a dog’s breed based on statements from a victim, a neighbor, an animal control officer, a police officer, or a dog owner—none of whom may be experienced with breed identification. Journalists may also make a wild guess based on their own visual assessment of the dog. News media also tends to identify dogs as “pit bulls” even when they are not pit bulls. Corrections, if they are made, are usually obscure. Thus, breed-specific data from studies that rely on media sources to identify breeds must be considered non-scientific and unreliable.

Risk Assessment and the Population Problem

One of the most common and enduring myths surrounding the oft-abused CDC report on fatal dog attacks (see below) is that it somehow demonstrates that certain breeds of dogs are “more likely” to attack or kill than other breeds.

The CDC has made it clear, both in the preface and the conclusion of their study, as well as on their website below the link to the study, that the study’s statistics cannot be interpreted in that manner.

In order to determine whether a breed of dog is “riskier” than another breed, a standard risk calculation must be performed. The easiest way to understand this calculation is through an example.

If you record one bite by a green dog and ten bites by purple dogs, which is more likely to bite—a green dog or a purple dog? If you look at the numbers alone, you might think that purple dogs are more dangerous than green dogs, because there are more bites by purple dogs.

However, it turns out that there are five green dogs total, one of which bit. And there are one hundred purple dogs total, ten of which bit. Now which type of dog is more likely to bite? Based on the data, one out of five green dogs have bitten, or 20%, while only one out of ten purple dogs have bitten, or 10%.

Once we know what the total population of green and purple dogs is, we are able to calculate risk and, as the CDC has determined, it is not possible to accurately provide total population data for all breeds or types of dogs. Additionally, there are a number of complicating factors, including how to categorize mixed breed dogs.

To add to the difficulty, while risk assessment might work if all dogs were genetically identical and were raised and kept in identical environments, this does not reflect reality. Individual dogs have widely varying temperaments and are raised and trained by different owners in different environments, so there are a number of factors beside breed that play into whether a dog is likely to bite or not. In fact, there are a handful of environmental factors (such as the way the dog is kept) that are far more predictive of aggressive behavior than a dog’s breed or type. (Delise, 2002, 2007)

Correlation Versus Causation

Another logical fallacy that has been made when looking at dog bite statistics is the assumption that correlation equals causation: that is, when events are shown to be related, it is because one event causes the other.

For instance, it has been shown that senior citizens are more likely to vote than middle-aged individuals. There is a correlation between age and voting record, but it would be a fallacy to assume that old age somehow causes people to vote. Rather, other factors—such as how much time an individual can devote to participating in the election process, or how concerned a voter is about the issues to be voted on—must be considered. It is not necessarily the case that older people are somehow genetically driven to vote, but that they are mostly retired (giving them time to engage in voting activities) and also have concerns about health care and retirement benefits (common election issues).

Similarly, apparent correlations between dog breeds and numbers of fatal attacks can not be interpreted as proof that the dog’s breed is the reason why the dog attacks.  Dog behavior, particularly aggression, is extremely complicated and involves a number of environmental and situational factors. Failure to recognize and analyze all of these factors results in misinterpretation and misapplication of statistics, with potentially dangerous results (such as the mythology of the dangerous/safe breed dichotomy).

A Closer Look At The Popular Studies

by Jackie Fitzgerald

Many studies have attempted to analyze the frequency and contributing factors of dog bites. Unfortunately, these studies are biased and use inaccurate counts of dog bites. There are numerous factors in a dog attack, many of which are not even considered in the major studies.

As dog attack expert Karen Delise says:

All dog bites/attacks are situational. Dogs bite in reaction to certain threats or stimuli—statistics about dog attacks purport to represent “canine aggression” however [they] do not take into account ANY of the situations under which a bite occurred. (E-mail interview with Delise, 4/22/08)

Centers for Disease Control and Prevention
Breeds of dogs involved in fatal human attacks in the United States between 1979 and 1998

This is perhaps the most misused and misunderstood dog bite report. Politicians and the media often quote this report inaccurately.

The main flaw in the CDC study is that it attempts to characterize dog attacks by breed, while ignoring all other possible factors.

Media as a source of data

The CDC study uses media accounts in their findings. The media is known to misreport and skew articles on dog attacks and misidentify breeds (see Difficulty of Breed Identification).

Missing data

The report also admits that it does not cover twenty-eight percent of fatal dog attacks. It is not clear what the study results would have been if all fatal dog attacks were included.

Miscategorization and misidentification

In the study, on the chart showing the number of dog bite-related deaths, the CDC has divided the attacks into sections titled Purebred and Crossbred. The CDC has listed Pit Bull-type and Husky-type under both the Purebred and Crossbred divisions. A “type” is not a breed.

Pit Bull-types are often categorized as dogs with short fur and a boxy head. There are over twenty breeds of dogs that fit this description, including the American Bulldog, Boxer, Olde English Bulldogge, Alapaha Blue Blood Bulldog, American Staffordshire Terrier, Bull Terrier, Bull Mastiff, American Pit Bull Terrier, Dogo Argentino, Tosa Inu, Staffordshire Bull Terrier, Dogue de Bordeaux, Fila Brasileiro, Presa Canario, Catahoula Leopard Dog, Cane Corso, Black Mouth Cur, and the Shar Pei.

Husky-types are often identified as medium sized dogs with long fur. This is extremely vague. Dogs that meet this description include the Akita Inu, Shiba Inu, Alaskan Malamute, Samoyed, Elkhound, Hokkaido Inu, Laika, Siberian Husky, Chow Chow, Alaskan Husky, and the Greenland Dog.

It is inaccurate to list all such breeds under one title; such groupings distort the study’s findings. Yet few people could recognize all these breeds correctly.

The CDC study uses inconclusive sources, fails to account for breed misidentification, and erroneously groups breeds.

Study conclusion

Despite the study’s flaws, the study authors conclude that breed-specific legislation is inefficient; BSL fails to recognize that any dog of any breed can exhibit aggressive behaviors.

Merritt Clifton
Dog Attack Deaths and Maimings, U.S. and Canada, 1982 through 2007 (updated yearly)

Merritt Clifton’s study is a medley of newspaper articles that present a very biased and inaccurate overview of dog bites. It is more of an incomplete tally of severe bites than a study.

Media as only source of data

Clifton’s only source for his findings is the media, and he focuses on cases that required “extensive hospitalization.” This term is never defined in his article. It might mean stitches, or it might mean amputation.

Missing data

In the beginning of the study, Clifton states that attacks by police dogs, guard dogs, dogs trained to fight, and dogs whose breed may be uncertain are excluded. Therefore, it is reasonable to assume a good number of attacks are not included. This might leave the reader with the assumption that Clifton has included all other dog attacks.

The CDC reports in their Morbidity and Mortality Weekly Report that of the “333,700 patients treated for dog bites in emergency departments in 1994, approximately 6,000 were hospitalized.” (July 4, 2003 article at

However, Clifton lists only 2,363 bites total—and that is over the 25 years that he has tallied media reports of attacks.

If approximately 6,000 people require hospitalization each year because of a dog attack, then over 25 years, there would have been 150,000 people hospitalized. Yet Clifton has apparently only found media reports for 1.6% of all these attacks.

Clifton’s report therefore implies that the remaining 98.4% of bites that required “extensive hospitalization” according to the CDC were by non-identifiable types of dogs or police, guard, or fighting animals. This is highly unlikely. Clifton’s data is so incomplete as to make it virtually useless for analyzing patterns related to severe dog attacks.

Miscategorization and misidentification

On Clifton’s list of all dog attacks and the dogs’ breed, he makes several mistakes.

He lists the Australian Blue Heeler, the Australian Cattle Dog, the Blue Heeler, and the Queensland Heeler as separate breeds. These are all different names for the same breed. Listing these attacks under separate breed names skewed the results of the study.

It should be noted that Clifton does not attempt to divide pit bull attacks into separate breed names. If he were to do so, it is not clear what his study results would show; “pit bull” is a generic term for at least three different breeds of dogs, and dozens of other breeds are often lumped into the “pit bull” category based on their similar appearance.

There are also 33 attacks that were supposedly done by “Bull Mastiff (Presa Canario).” Bull Mastiffs and Presa Canarios are distinctly different breeds, and if there is question about which breed the dog is, this attack should not be listed as a “clearly identified breed.”

The report also attempts to identify the predominant breed in dogs. Clifton gives no reason as to why he listed an attack as being done by an Akita/Chow mix instead of a Chow/Akita mix. How did he determine that Beagle was the predominant breed in the attack done by a Beagle/German Shepherd Dog?

Clifton makes several spelling mistakes throughout his report. Misidentified breeds listed as a “Chox mix,” “Dauschund,” “Doge De Bordeaux,” “Fila Brasiero,” “Buff Mastiff,” “Great Pyranees,” and “Weimaeaner” compromise Clifton’s credibility.

Inability to determine risk scientifically

In Clifton’s analysis, he attempts to evaluate dog behavior based on breed, bite frequency, and “degree of relative risk.”

Yet Clifton has shown numerous times in his report that he cannot identify a breed properly, or even spell breed names correctly.

Both bite frequency and degree of relative risk are impossible to calculate. No one knows how often breeds bite since hundreds of bites go unreported. And to attempt to determine a “degree of relative risk,” Clifton would have to know every factor that contributed to every dog bite.

Even the CDC concluded at the end of their own flawed study (see above) that there is no way to determine relative risk:

There is currently no accurate way to identify the number of dogs of a particular breed, and consequently no measure to determine which breeds are more likely to bite or kill.

Merritt Clifton apparently does not understand the many factors that go into a reliable calculation of relative risk, nor does he wish to acknowledge that trained researchers realize that many, if not most, of those factors can never be known or calculated.

Misapplied and misinterpreted data

Clifton’s analysis section is full of faults and absurd assumptions.

Of the breeds most often involved in incidents of sufficient severity to be listed, pit bull terriers are noteworthy for attacking adults almost as frequently as children. This is a very rare pattern . . . Pit bulls seem to differ behaviorally from other dogs in having far less inhibition about attacking people who are larger than they are.

As discussed, Clifton has tallied less than two percent of all severe dog attacks. He clearly has no idea how frequently pit bulls—or any other type of dog, for that matter—bite.

Furthermore, without knowing all bite factors, including the dog’s health, condition, sexual state, training, environment, and the behavior of the victim, there is no way Clifton could possibly conceive any possible pattern or difference as to who pit bulls attack.

Since Clifton is tallying media articles, his conclusion seems to be more telling of media coverage of dog bites. If one was to assume that the media is more likely to publish a pit bull attack than an attack by another type of dog, and more likely to publish an attack on a child than an attack on an adult, it stands to reason that while media-reported pit bull attacks include both adults and children, media reports about other types of dogs’ attacks may only be considered newsworthy when a child is involved. Thus, it may appear that pit bulls are overrepresented in attacks on adults.

Misunderstanding of dog behavior and ignorance about breed standards

They [pit bulls] are also notorious for attacking seemingly without warning, a tendency exacerbated by the custom of docking pit bulls’ tails so that warning signals are not easily recognized. Thus the adult victim of a pit bull attack may have had little or no opportunity to read the warning signals that would avert an attack from any other dog.

All dogs exhibit warning signs. Pit bull expert Diane Jessup, a retired animal control officer and police dog trainer, stated in her book The Working Pit Bull, “all Pit Bulls do give some warning that they are going to attack.”

Studies have indicated that, generally, people do not understand dog body language. A person may not recognize that a dog standing very still, legs apart, tail waving slowly, is indicating an impending attack. When one cannot identify all possible threat behaviors, it might appear that a dog is attacking without warning. Clifton provides no evidence to show that victims are oblivious to impending attacks by pit bulls at a greater rate than impending attacks by other dogs.

Clifton’s statement that pit bulls’ tails are customarily docked demonstrates his lack of familiarity with the breed-type. A list of traditionally docked breeds can be found on the Council of Docked Breeds website ( None of the pit bull breeds, to include the American Pit Bull Terrier, the American Staffordshire Terrier, and the Staffordshire Bull Terrer, can be found on this list. Nor can any of the breeds that are occasionally mistaken to be “pit bulls,” such as the American Bulldog, Bull Mastiff, and Bull Terrier. Tail docking has never been common or customary with any of the pit bull types. Docking the tail of an American Pit Bull Terrier, American Staffordshire Terrier or Staffordshire Bull Terrier means immediate disqualification from the show ring.

To substantiate his assertions that 1) pit bulls customarily have their tails docked, and 2) tail docking results in an inability for people to read canine body language, Clifton would need to provide evidence that a disproportionate number of pit bulls or attacking dogs have had their tails docked, and further, that a dog’s tail is the primary predictor of an impending attack. He provides no such evidence.

There are over 50 different breeds of dogs, including the Cocker Spaniel, Airedale Terrier, German pointer, Jack Russell Terrier, Poodle, and Corgi, whose tails are traditionally docked. (Council of Docked Breeds) If tail docking inhibits the communication of impending aggression, why are tail-less breeds not disproportionately represented in any list of severe and fatal attacks?

Excuses for some breeds’ behavior

Rottweilers . . . seem to show up disproportionately often in the mauling, killing, and maiming statistics simply because they are both quite popular and very powerful . . .

Clifton excuses Rottweilers’ attacks due to the fact that they are both popular and powerful. Yet pit bulls, who are also popular and strong, are not given this same excuse.

In the German shepherd mauling, killing, and maiming cases I have recorded, there have almost always been circumstances of duress: the dog was deranged from being kept alone on a chain for prolonged periods without human contract, was starving, was otherwise severely abused, was protecting puppies, or was part of a pack including other dangerous dogs. None of the German shepherd attacks have involved predatory behavior on the part of an otherwise healthy dog. [sic]

Here Clifton excuses German Shepherd attacks due to outside factors. This implies that no other type of dog in his study attacked because it was left neglected, abused, chained or left untrained and unsocialized. Yet he offers no proof to substantiate the idea that all other cases he recorded involved trained, socialized, beloved family pets.

[I]t is sheer foolishness to encourage people to regard pit bull terriers and Rottweilers as just dogs like any other, no matter how much they may behave like other dogs under ordinary circumstances.

Clifton implies that pit bulls and Rottweilers no longer behave like dogs under extra-ordinary circumstances. What those extra-ordinary circumstances are is unstated, and how pit bulls and Rottweilers suddenly become behaviorally different under those circumstances is not demonstrated in the report.

To imply that pit bulls and Rottweilers are not to be regarded as dogs even though they act like ordinary canines is absurd. Clifton’s agenda is quite clear—he badly wishes to portray pit bulls and Rottweilers as somehow unique—but his “study” is so flawed that he cannot prove any of his sweeping generalizations.

Temperament is not the issue, nor is it even relevant. What is relevant is actuarial risk.

Here Clifton returns to the idea that, somehow, we can calculate the “riskiness” or “relative danger” of particular breeds or types of dogs. As demonstrated earlier in this article, it is not possible to do this.

Furthermore, it is totally bizarre to say that temperament is not an issue. Temperament plays a huge part in dog attacks, as any canine behaviorist or dog bite researcher would agree. A very large dog may be able to do a lot of damage if it bites someone, but if the dog is extremely placid by nature (temperament), there’s very little danger to the public. On the contrary, a smaller dog may do less damage if it attacks, but if it is extremely aggressive, it could maul or kill someone. To suggest that temperament isn’t even relevant is ridiculous.

If almost any other dog has a bad moment, someone may get bitten, but will not be maimed for life or killed, and the actuarial risk is accordingly reasonable. If a pit bull terrier or a Rottweiler has a bad moment, often someone is maimed or killed—and that has now created off-the-chart actuarial risk, for which the dogs as well as their victims are paying the price.

Clifton’s own “study” disproves his assertions. His own tally of severe and fatal dog attacks includes over 50 different types and breeds of dogs. It seems clear that dogs of all types can have “bad moments” that result in severe injury.

Study conclusion

Clifton concludes in his article that he “[does] not know how an effective, fair, enforceable, humane dangerous dog law could be constructed.” He goes on to propose breed-specific legislation as the solution to dangerous dogs, yet provides no scientific evidence that BSL actually works.

Jaclyn Barnes
Ownership of High-Risk (“Vicious”) Dogs as a Marker for Deviant Behaviors: Implications for Risk Assessment

Jaclyn Barnes’s study is an attempt to find parallels between choice of dog breed and criminal activity. The study attempts to help law enforcement identify “risk factors” that would lead a person to cause harm to either themselves or others by dog breed.

The study used 355 owners of “licensed or cited dogs that represented high or low-risk breeds.” Some of the owners had a “criminal background,” which included minor traffic citations.

Barnes used the Ohio Revised Code definition for “vicious dog” to classify high-risk dogs for the purposes of the study. However, at the time of this study, Ohio state law classified all pit bulls to be prima facie vicious dogs (this law was repealed in 2012). For this reason, all pit bulls were automatically placed into the “high-risk” category regardless of the individual dogs’ behavior.

[S]ome breeds, namely Pit Bulls, may qualify as “vicious dogs” simply by reputation, not because a specific dog has behaved in a harmful manner.

The study therefore defines a “high-risk” dog as a dog that without provocation has killed or seriously injured a person, killed another dog, or is a pit bull.

Potential for skewed population due to breed misidentification

The study author does not explain how breeds are identified, but the reader supposes that the breed is taken off either license or citation paperwork. This means that, in the case of a license, the owner decides what a dog’s breed is. In the case of a citation, an animal control officer probably decides what a dog’s breed is.

This naturally leads to a serious question about identification accuracy, especially since most dogs are not purebred. For instance, animal control officers may have been inclined to over-identify troublesome dogs as “pit bulls” because the category is broad and vaguely defined, and because Ohio’s state law at the time gave animal control more tools to deal with problematic “pit bulls” than with other types of problematic dogs, thus encouraging them to declare dogs “pit bulls.”

Barnes also observes that “some owners license a HR [high risk] dog such as a Pit Bull as another breed, such as Boxer” to avoid the automatic designation of “vicious” that Ohio placed on pit bulls. Obviously, this suggests that Barnes’s population may be skewed due to the effects of BSL; some dog owners were intentionally misidentifying their dog’s breed, and Barnes has no ability to correct for this problem. This means that data for the other breeds tallied by Barnes may actually have been data for pit bull mixes that were intentionally recorded by the owners as a different breed.

Barnes also includes two “breeds” that aren’t recognized by any reputable kennel club—the “Ahra” and the “Terripoo.” It is not clear what an Ahra is, but Terripoo might be a mix of poodle and terrier, so the latter, at least, should have been included as a “mixed breed.”

Concerns about population selection

Furthermore, examination of the base population studied by Barnes raises some serious concerns. Barnes collected the study population by choosing from citations issued by the Cincinnati SPCA and by choosing from licenses issued by Hamilton County in Ohio.

Presumably these choices were random, yet a look at the breed tally raises eyebrows immediately. Only one Labrador Retriever is included. It is hard to understand how random selection of 355 dogs would only produce a single Labrador Retriever. Similarly, only eight mixed breeds are listed. By contrast, 153 Pit Bulls were included in the 355-dog total.

It is unclear how supposedly random selection produced such a bizarre population makeup. Barnes does not explain why this occurs.

Breed bias

The study also shows apparent breed bias. As mentioned previously, there is one Labrador Retriever included, under “high-risk cited,” and it has a notation next to the listing. The note declares that the dog was only listed because it attacked and killed another dog.

What was it about this attack that required a notation? Dogs that kill other dogs are already included as “high-risk” as mentioned in the study. Was it the condition of the attack that prompted Barnes to include the notation? Or was it the fact that no one expects a Labrador to be capable of attacking?


Despite so many serious issues with Barnes’s study setup and population selection, Barnes goes on to conclude that there is a correlation between ownership of a “high-risk” dog and criminal convictions. She correctly cautions that this cannot be translated into a cause-and-effect relationship.

It is the opinion of this author that the mere fact that the study was performed in an area with BSL immediately brings about narrow results that cannot be applied to the rest of the U.S. Had this study been done in a place with no BSL, this author feels that the results would have been distinctly different.

Sources and Resources

Barnes, Jaclyn. “Ownership of High-Risk (“Vicious”) Dogs as a Marker for Deviant Behaviors: Implications for Risk Assessment.” Journal of Interpersonal Violence 2006.

Bradley, Janis. Dogs Bite: But Balloons and Slippers Are More Dangerous. James & Kenneth Publishers, 2005.

Bradley, Janis. “Dog bites: Problems and solutions.” Animals and Society Institute,2006.

Centers for Disease Control and Prevention. “Breeds of dogs involved in fatal human attacks in the United States between 1979 and 1998.”

Clifton, Merritt. “Dog Attack Deaths and Maimings, U.S. and Canada, 1982 through 2007.” Self published.

Delise, Karen. Fatal Dog Attacks: The Facts Behind the Statistics. Anubis Press, 2002.

Delise, Karen. The Pit Bull Placebo: The Media, Myths, and Politics of Canine Aggression. Anubis Press, 2007.

National Canine Research Council.

Statistics Help For Journalists.