Editing 43 times

[[File:BS.jpg|right|]]
==The myth==
'''"A gun owner is 43 times more likely to kill a family member than an intruder."'''  

This infamous and persistent factoid<ref>'''Factoid''' ''n.'':<br> '''1.''' Something which becomes accepted as fact, although it may not be true (OED).<br> '''2.''' A piece of unverified or inaccurate information that is presented in the press as factual, often as part of a publicity effort, and that is then accepted as true because of frequent repetition ([http://www.thefreedictionary.com/factoid The Free Dictionary] online).</ref> is one of the most widely quoted anti-gun statistics in North America.  This allegedly damning statistic is meant to both emphasize the inherent danger imagined in merely having a [[gun]] in the house, while simultaneously denying that guns are in any way effective for [[self-defence]].

It is essentially a variation on the old [[gun-grabber]] saw which says that a gun makes someone into an unstoppable killing machine, unless they are intent on using that gun for reasonable and lawful self-defense, in which case it is useless and will likely be taken away by the unarmed bad guy, at which point it will suddenly become an unstoppable killing machine again and be used against you and everyone else in the area.  Especially women and children (preferably from some socially or economically disadvantaged class).

==The facts==
This particular bit of misinformation originally comes from a 1986 article in the New England Journal of Medicine titled ''"Protection or Peril: An analysis of firearm-related deaths in the home,"'' by A.L. Kellermann, and D.T. Reay<ref> AL Kellermann and DT Reay, "[http://content.nejm.org/cgi/content/abstract/314/24/1557 Protection or peril? An analysis of firearm-related deaths in the home]," ''The New England Journal of Medicine,'' Volume 314:1557-1560, June 12, 1986.</ref>.

This article &mdash; based on a single study carried out in King County, Washington, from 1978-1983 &mdash; is quite possibly one of the most influential works in the entire North American gun-control dogma and is presented as absolute proof that any positive self-defence benefit that guns may have is overwhelmed by the damage they do, by the damning ratio of 43 to 1. In the study, there were 389 gun deaths in a dwelling where the gun was kept; only 9 were self-defence homicides, thereby supposedly giving the 43:1 ratio.

The study, however, is also a perfect textbook example of pseudo-science in action. This page examines how that particular pseudo-science gets dressed up in drag to look like real science.

===Context===
Kellermann begins his article with an impressive set of numbers &mdash; 120 million guns in private hands, half of all homes contain one, and three quarters of owners keep them "at least partially for [[self defence]]." He notes that during the study period, there were 3,273 violent deaths of one sort or another, which were not traffic-related (car accidents were excluded) . Out of these deaths:
* 743 were from shooting,
* 473 took place in a dwelling and
* 389 took place in a dwelling where the firearm involved was kept.
He continues to set the scene by making the peculiar claim that he is studying the "epidemiology of fire-arms related deaths in the home," a statement that seems to imply the existence of an "epidemic" of some sort.  It may be interesting to point out that Kellermann's little opus is also one of the earliest (if not ''the'' earliest) attempts to re-frame gun control as somehow being a medical issue (it isn't) instead of a criminological issue (it is).

A closer peek at those bothersome numbers puts the "epidemic" in context. The first, we have the fact that firearm-related deaths make up only 22.7% of all violent deaths in the study area. Accordingly, even if wander of into la-la land and pretend that anti-gun policies are going to be 100% effective against gun violence (and while we're at it, let's also assume that simply using some different weapon is impossible), the world's most perfect anti-gun effort still won't affect 77.3% of all violent deaths in any way. Any anti-violence policy that focuses exclusively gun deaths or gun violence can't possibly address anything more than a small part of the much bigger problem.

Another way to determining the "peril" of gun ownership [insert ominous organ music here] is to examine just how common these fatal shootings actually are. To do this we have to unpack that alarming number of firearms deaths in King County: 743. Using ''Kellermann’s own data,'' we discover that this figure is in fact ''a six-year total'', running from 1978 through 1983, with an average of 124 deaths per year. To properly assess the level of real danger this number represents, it needs to be seen in relation to the County’s total population of roughly 1,280,000 at the time of the study. If we assume that a) the average household has four people living in it and b) half of them own guns, we find something in the neighbourhood of 160,000 gun-owning households in King County at that time.

The figures 124 out of 160,000 indicate that gun deaths only occur in a minuscule percentage (about 0.0775%, actually) of the households with guns, which doesn't exactly qualify as some sort of "epidemic," like Kellermann implies he's studying. It also presents an inconvenient statistical problem for Kellermann and co.:  If owning a gun really is 43 times more dangerous than not owning one, how is it that only a tiny fraction of gun owning households are affected? Kellermann also found that, excluding suicides, there were '''75''' gun deaths in households without guns, but only '''65''' gun deaths in households where the involved gun was kept. Since half of all households have guns and the other half doesn’t, why would there be '''15% more deaths''' in "gun-free" households? 

To put it a little more bluntly: if the half of all households that are gun owning are 43 times more dangerous, why are there '''more''' gun deaths in non-owning households?  According to Kellerman's magic "43" equation, 75 gun deaths in non-gun households should be balanced with 3225 deaths in gun-owning households.  Instead, there are only 65.  Where the hell are the other 3160??  To look at the reverse, 65 deaths in gun-owning households should add up to about 1.5 deaths in non-gun homes.

Is there something fishy here?  Absolutely.  The table below points out what little relevant data can be ascertained from Kellerman's study:
{| class="wikitable collapsible collapsed" align="center" style="border: 1px solid #aaa; font-size:85%" width="100%" cellspacing="0" cellpadding="3"
|- style="font-size:120%" 
! colspan="15" | <big>Table of violent deaths over the 6 year period of Kellerman's study<big>
|- style="font-size:120%" 
! rowspan="2"|Manner of death !!colspan="2"|Whole numbers !!colspan="11"|Percentages based on 1.28M population; 320K households (avg. 4);<br>50% gun ownership (160K households, 640K individuals)!!rowspan="2"|Manner of death
|- style="font-size:120%" 
! #!!per 100K<br>pop.||% of deaths||Non gun related||Gun related||in dwelling||in gun owning home||expected<br>(x43)||suicide||accident||shot w/ house gun||domestic||Non-gun home
|-
! style="font-size:120%" |All violent deaths<ref>Not including accidents.</ref>
|3273||'''256'''||'''100'''||||||||||||||||||||
! style="font-size:120%" |Total violent
|-
! style="font-size:120%" |Non firearm related
|2530||'''198'''||'''77.30'''||'''100'''||3.5x||||6.5x||||7.5x||211x||39x||60x||
! style="font-size:120%" |Non gun related
|-
! style="font-size:120%" |Firearm related
|743||'''58'''||'''22.70'''||29.368||'''100'''||||||||||||||||
! style="font-size:120%" |Firearm related
|-
! style="font-size:120%" |In a dwelling<ref>Includes all deaths in dwellings, including non firearm related and suicides.</ref>
|473||'''37'''||'''14.45'''||||||'''100'''||||||||||||||
! style="font-size:120%" |In Dwelling
|-
! style="font-size:120%" |In dwelling where gun is kept<ref>Includes all violent deaths in gun owning households, including non firearm related, suicides and shootings not with a gun kept in the home.</ref>
|389||'''30'''||'''11.89'''||15.375||||'''82.241'''||'''100'''||12.062||||||||||
! style="font-size:120%" |Dwelling w/ gun
|-
! style="font-size:120%" |x43 prediction<ref>The numbers expected to die by a gun kept in the house, calculated using the alleged 43:1 ratio by multiplying the number of people shot in non gun owning households by 43.</ref>
|3225||252||98.53||127||434||||829||100||||||50x||||43x
! style="font-size:120%" |x43 prediction
|-
! style="font-size:120%" |Firearm suicide in home
|333||'''26'''||'''10.17'''||||'''44.818'''||'''70.402'''||'''85.604'''||||'''100'''||28x||5x||8x||
! style="font-size:120%" |shoot self in home
|-
! style="font-size:120%" |Firearm accidents<ref>Includes ALL firearm accidents: negligent discharge, hunting accidents, unintended discharge, ricochets etc.</ref>
|12||'''1'''||'''0.37'''||0.47||'''1.615'''||||||||||'''100'''||||||16
! style="font-size:120%" |Gun accidents
|-
! style="font-size:120%" |With gun that was kept in house<ref>Persons killed in their own home with a firearm that was normally kept in the home.</ref>
|65||'''5'''||'''1.99'''||2.57||'''8.748'''||1'''3.742'''||'''16.710'''||2.016||||||'''100'''||||86.667
! style="font-size:120%" |Shot w/ house gun
|-
! style="font-size:120%" |As above, following altercation<ref>Persons killed, following an "altercation," in their own home with a firearm that was normally kept in the home.</ref>
|42||'''3'''||'''1.29'''||1.66||'''5.653'''||'''8.879'''||'''10.797'''||||||||'''64.615'''||'''100'''||56.0
! style="font-size:120%" |Domestic shooting
|-
! style="font-size:120%" |Shot in "gun free" home<ref>Persons killed with firearm in home where no firearms were owned.</ref>
|75||'''6'''||'''2.29'''||||'''10.09'''||'''15.856'''||||||||625||115.385||178.571||'''100'''
! style="font-size:120%" |Shot in non gun home
|-
! style="font-size:120%" |÷43 prediction<ref>The numbers expected to die by gun in a non gun owning home, calculated using the alleged 43:1 ratio by dividing the number of people shot with a household gun by 43.</ref>
|1.5||<1||0.05||||0.203||||||100||||12.597||2.326||||2.016
! style="font-size:120%" |÷43 prediction
|-
|colspan="15"| 
* Information that is irrelevant or cannot be proven with available figures has been removed from this table.  
* Proportions over 100% have been replaced with multipliers rounded to the nearest half whole number for simplicity.  
* Non-bold values represent calculations that do not necessarily include the same individuals.  Eg.: A person who dies "in a dwelling where a gun is kept" is automatically included in the "in a dwelling" group, but not all people who died in a dwelling did so in one where a firearm was kept.
|}

===Cherry-picking the sample===
The very low (0.0775) percentage of gun-owning households with a gun death throws a monkey wrench at the presumption of just how typical the households in Kellermann’s study really are. If they're not representative of all households, his study loses its predictive power. This means that instead of producing conclusions that can be extrapolated to all households, the study would only apply to households of a particular type. Kellermann completely fails to explicitly address this critical question. However, his study does contain some intriguing hints as to what type of household is really being examined. He states that after excluding suicides there were 65 deaths involved firearms kept in the home and that 42 (84%) of them involved an "altercation." It doesn't take a rocket scientist (or a cumbersome academic study, for that matter) to figure out that "altercations" leading to a fatal shooting are not typical of gun-owning households. Instead, the very low percentage of gun owning household involved in these events confirms that Kellermann’s sample is anything but typical of gun-owning households. This rhetorical maneuver &mdash; seriously misrepresenting what is "typical" in order to reach a predetermined conclusion &mdash; is a common sign that you are seeing pseudo-science in action.

Another rather large problem with Kellermann’s study is that it's based on a "gun death" metric that mixes three very different types of events: suicides, self-defence and accidents. Now, in all honesty, this can be justified to a certain extent because it is, after all, a "gun death" study.  Can't have a gun death study without counting gun deaths, right?  The problem here is that these are wildly different types of events, each with very different dynamics. Kellermann’s attempt to lump them in one pile and link them all to a simple common cause, gun ownership, is problematic (to put it politely) because these three occurrences '''do not''' share the same causation. Trying to peg gun ownership as a causation factor also raises another problem: Kellermann simply assumes, with no supporting evidence whatsoever, that gun ownership somehow, in some mysterious and mystical fashion, leads to violence. The possibility that some people can be just plain violent, long before they ever get their hands on any guns, is simply thrown out with the bath water.

So, out of this cocktail of three very different types of events, we quickly notice that suicide seriously outranks all the others, accounting for 333 (70%) of the 473 violent deaths in homes. Apparently, Kellermann's rear end doesn't contain any evidence for him to pull out to demonstrate that gun owners are either a) more likely to commit suicide than other people or b) that this is somehow a characteristic typical of gun owners as a group. Without that missing link, it's hard to see why suicide should even be associated with gun ownership at all, let alone flagged as a specific "risk" of some sort. Kellermann even recognizes this problem himself (in a roundabout way) when he admits, "the precise nature of the relation between gun availability and suicide is unclear."  Then, without even breaking dogmatic stride, he promptly justifies this admission with pure speculation that:
* most gun suicides are "impulsive" and that
* gun suicides have a "high (implying ''high'''er''''') case-fatality rate," and
* these two unfounded assumptions somehow justify the inclusion of suicide as a particular danger of gun ownership. 
The inclusion of suicide in a study whose alleged underlying concern is addressing the utility guns for self-defence is problematic, to say the least.  It gets even sloppier when Kellermann fails to point to any real connection between owning a gun and even attempting suicide in the first place.  What the inclusion of suicide ''does'' do quite handily, is that it allows Kellermann to cook the books and inflate his numbers and make a (superficially) much stronger looking case for gun control.  And that, ladies and gentlemen, makes it one very handy assumption.
{{quote|When you make an assumption, you make an ass out of you and Umption.|Mitch Henessey (Samuel L. Jackson), [http://www.imdb.com/title/tt0116908/ The Long Kiss Goodnight] (1996)}}
===Cherry-picking the Indicator===
One of the more critical decisions in any research design is the choice of what indicator or metric the benefits or lack of benefits of any one thing are going to be determined on. In this study, Kellermann chooses "shot dead" not just as a metric for suicide or accidental death where they apply but for defensive gun use as well. Kellermann explicitly states, "mortality studies such as ours do not include cases which burglars or intruders are wounded or frightened away by the use of, or display of, a firearm;" obviously one of the things "a complete determination of firearm risks versus benefits would require." Undaunted, he dolefully goes on to assess "protection or peril" with any self-defence considerations (i.e., the "protection" portion of the equation) conveniently excluded.

The greatest problem with using Kellermann’s "shot dead" indicator as a measure of self-defence is how much it fails to line up with the realities of gun use in self-defence. How well it lines up with the anti-gun rhetoric about gun use is, naturally, another matter altogether. Criminologist [[Gary Kleck]] has found that only about one in every thousand (0.1%) instances of self-defence with a gun actually result in the death of the criminal involved<ref>Gary Kleck, "Point Blank: Guns and Violence in America." New York: Aldine de Gruyter. (1991)</ref>. Hence, Kellermann has seriously distorted his findings by chosen a metric that underestimates the value of self-defence firearms by a factor of a 1,000<ref>Edgar Suter, "Guns in the Medical Literature-A Failure of Peer Review," ''Journal of the Medical Association of Georgia'' 83(3) 1994, pp. 136-37.</ref>

===Summary===
In brief: This statistic is based on a three-county study comparing households in which a homicide occurred to demographically similar households in which a homicide did not occur. After controlling for several variables, the study found that gun ownership was associated with a 2.7 times increase in the odds of homicide.<ref>[http://content.nejm.org/cgi/content/full/329/15/1084 "Gun Ownership as a Risk Factor for Homicide in the Home."] By Arthur L. Kellermann and others. ''New England Journal of Medicine'', October 7, 1993. {{quote|After controlling for these characteristics, we found that keeping a gun in the home was strongly and independently associated with an increased risk of homicide (adjusted odds ratio, 2.7; 95 percent confidence interval, 1.6 to 4.4).}}  Note that the "odds ratio" is not the same as the "relative risk" (which is the measure of probability that [http://www.ncbi.nlm.nih.gov/pubmed/18238982 most people understand]).</ref> This study does not meet standards of credibility because:

# The study blurs cause and effect. As explained in a comprehensive analysis of firearm research conducted by the National Research Council, gun control studies such as this (known as "case-control" studies) "fail to address the primary inferential problems that arise because ownership is not a random decision. ... Homicide victims may possess firearms precisely because they are likely to be victimized."<ref>''Firearms and Violence: A Critical Review''. By the Committee to Improve Research and Data on Firearms and the Committee on Law and Justice, National Research Council of the National Academies. Edited by Charles F. Wellford, John V. Pepper, and Carol V. Petrie. National Academies Press, 2005. Page 5: {{quote|Because of current data limitations, researchers have relied primarily on two different methodologies. First, some studies have used case-control methods, which match a sample of cases, namely victims of homicide or suicide, to a sample of controls with similar characteristics but who were not affected by violence. ... 

Case control studies show that violence is positively associated with firearms ownership, but they have not determined whether these associations reflect casual mechanisms. Two main problems hinder inference on these questions. First and foremost, these studies fail to address the primary inferential problems that arise because ownership is not a random decision. For example, suicidal persons may, in the absence of a firearm, use other means of committing suicide. Homicide victims may possess firearms precisely because they are likely to be victimized.}}</ref>
# The study's results are highly sensitive to uncertainties in the underlying data. For example, minor variations in firearm ownership rates (which are determined by interview and are thus dependent upon interviewees' honesty) can negate the results.<ref>''Firearms and Violence: A Critical Review''. By the Committee to Improve Research and Data on Firearms and the Committee on Law and Justice, National Research Council of the National Academies. Edited by Charles F. Wellford, John V. Pepper, and Carol V. Petrie. National Academies Press, 2005. 

Page 119 [regarding case-control studies such as Kellermann et al., 1993]: {{quote|[E]ven small degrees of misreporting on ownership by either the cases or the controls can create substantial biases in the estimated risk factors (see Kleck, 1997, for an illustration of these biases).}}

Page 35: {{quote|While surveys of firearms acquisitions, possession, and use are of varying quality and scope, they all share common methodological and survey sampling-related problems. The most fundamental of these is the potential for response errors to survey questionnaires. Critics argue that asking people whether they own a firearm, what kind it is, and how it is used may lead to invalid responses because ownership is a controversial matter for one or more reasons: some people may own a firearm illegally, some may own it legally but worry that they may use it illegally, and some may react to the intense public controversy about firearm ownership by becoming less (or even more) likely to admit to ownership (Blackman, 2003). 

While in most surveys respondents are provided confidentiality, the concern is still expressed that violations of confidentiality directly or through data mining could lead to the identification of specific respondents in a way that might allow the identification of firearms owners.}}</ref><ref>Armed: New Perspectives on Gun Control. By [[Gary Kleck]] & Don B. Kates. Prometheus Books, 2001. Chapter 2: "Guns and Public Health: Epidemic of Violence or Pandemic of Propaganda?" By Don B. Kates. Page 82: {{quote|To reiterate, NEJM-1993's conclusions depend entirely on there having been no substantial underestimation of the control group's gun ownership. It would take only 35 of the 388 controls falsely denying gun possession to make the control ownership percentage exactly equal to that of the homicide case households. If indeed the controls actually had gun ownership equal to that of the homicide case households (45.4%), then a false denial rate of only 20.1 percent among the gun-owning controls would produce the 35 false denials and thereby equalize ownership. Such a 20.1 false denial rate is smaller than either of the "refused consent for interview" category of the pilot study, or the "inaccurate registration data" category. Therefore the results of the pilot study are consistent with a false denial rate sufficiently high to bring the control group gun ownership rate up to a level equal to, or even higher than, the homicide case household rate, although the authors cite the pilot study to the reverse effect. Neglect of the false denial rate can produce a bias large enough, by itself, to account for the entire association between gun ownership and homicide claimed in this study.}}</ref>
# The results are arrived at by subjecting the raw data to statistical analyses instead of letting the data speak for itself. (For reference, the raw data of this study shows that households in which a homicide occurred had a firearm ownership rate of 45% as compared to 36% for non-homicide households. Also, households in which a homicide occurred were twice as likely have a household member who was previously arrested (53% vs. 23%), five times more likely to have a household member who used illicit drugs (31% vs. 6%), and five times more likely to have a household member who was previously hit or hurt during a fight in the home (32% vs. 6%).<ref>[http://content.nejm.org/cgi/content/full/329/15/1084 "Gun Ownership as a Risk Factor for Homicide in the Home."] By Arthur L. Kellermann and others. New England Journal of Medicine, October 7, 1993. {{quote|"Multivariate analyses used conditional logistic regression, the appropriate technique for a matched-pairs design.}}</ref>)

==Signs of pseudo-science==
Kellermann’s study is a classic case of pseudo-scientific research in action.  It is designed to create a pre-determined conclusion, and NOT to seriously expose subject in question to critical analysis.  Indeed, it actively shields the subject from any sort of accurate measurement of risk vs. benefit.  Some (but by no means all) of the characteristics typical of this approach, which you can keep an eye out for in the future, are:

# The use of large and impressive numbers without placing them in their proper context
# Conflating and expanding categories, often tossing in irrelevant data, to produce the desired numbers
# Mis-defining what is or isn't typical, in order to misrepresent the situation in favour of the study's initial presumption
# Carefully choosing indicators that will bias or predetermine the results

==Notes==
<references/>

[[category:Myths]]