You've probably heard the claim that gun in the home is 43 times more likely to be used in shooting a fellow occupant than a criminal.  Read this and find out why this claim is a fraud.

Dave Markowitz

BBS: AirPower BBS

Date: Thu, 01-01-98 (10:46) Number: 88 Refer: 0



Subj: Kellerman analyzed

Conf: PR_NET (103) Read: No Status: Public

This from a local fellow who works in the health care field:

Last week I covered some of the more glaring and easily understood errors of the two and one-half page 1986 New England Journal of Medicine article by ER doctor Arthur Kellerman which claimed that "a gun in the home is 43 times more likely..."

Kellerman used many statistical tricks such as limiting any defensive gun use to that where the trigger was actually pulled and excluding any gun use which ocurreding any gun use which ocurred outside the home (such as the front yard). Using multiple categories and then condensingg the data from there tended to obscure the more obvious factors such as, most suicides in his study did not use a gun. And he also refused to allow criminologists such as Kates and Kleck to examine his "data." Too many to mention here.

Despite the 1986 "study" being trashed by every major criminologist, the "43 times" soundbite that emerged lives on in news stories and HCI literature. The 1986 "study" is actually kind of boring and tacky when compared to Kellerman"s 1993 NEJM submission Again, without a hint of peer-review (since the peers didn't get a chance to look at the data until May of 1997), Kellerman this time has not only one co-author, but nine!! Five MDs, three PhDs and one BA in total, not a recognizable name from the world of criminology among them.

Oh, and Kellerman was now one of the major recipients of over two million in grants from the CDC. The CDC head that gave Kellerman the money and helped him hide his data from the probing eyes of scholarly criminologists, thus violating federal law, is now President Clinton's pick for Surgeon General. Anyway, the soundbite that comes from the 1993 "study"? "A gun in the home is 2.7 times..."

On purpose. Kellerman"s first "mistake" is building his is building his paper around a "case-control model." A CCM does well when used to from a general hypothesis about, say, what is causing a particular disease outbreak. But it has a problem with being very sensitive to sample errors and it played right into Kellerman"s hands.

Kellerman took homicide data from three different counties from August 1987 to August 1992. These counties surrounded Seattle (co-author and medical examiner, Donald Reay"s jurisdiction), Memphis and Cleveland. He then investigated each homicide case to see what lifestyle factors it had (drug-use, renting/owning). The proper thing to do would have been to use a random sample of people who may, or may not have been involved in a homicide. Having found the "case", Kellerman now finds a "control" nearby who supposedly shared the demographics but did not have a homicide in the household.

Here is where everything goes to hell in a handbasket. First, Kellerman was only able to get 388 valid cases out of 1860 official reports but only 316 matched controls. Such a small sample is prone to wide statistical variations absent an agenda such as Kellerman et al. It would be interesting to see why, exactly, Kellerman decided to drop so many cases from the study--but none of the analyses that I have read were writtenread were written after 1996. Again, homicides whether lawful or not are included along with suicides. And even "police" was listed as a category in the "Offender" column.

What Kellerman found was that in the homicide homes, 71% of the victims had high rates of criminal activity which agrees with other studies which find that 75% of murderers and 75% of cop killers are adults with long felony histories.. Hardly something that would compare to the rest of America. No, the gun didn't do it. The nature of the criminal did. Criminals killing criminals.

The study never made an effort to decide whether the person being killed was an intruder --OR--whether the gun which may or may not have been present was actually used. And since Kellerman's study shows that 71% of the homicide victims were killed by people whose relationship to the victim indicates that the killer did not live in the victim's house--and presumably brought his own weapon, if any, with him. All Kellerman asked was, "In this household where a homicide was committed, was there a gun, any gun, in the house?" Nothing more.

It turns out that the cases did not quite match up with the controls. While the cases had an overwhelming violent history of crime, assault, drug abuse and, drug abuse and domestic violence (real stuff, notthe pushy- shovey type) the controls background check consisted of, "Were you ever arrested?" Well, I have been arrested--failure to appear in court for not licensing my dog. Not the same stuff. Also, nothing was asked of convictions or seriousness of the crime.

Kellerman, goes to the control house which must be at least "one block" away from the case house, but no maximum is listed. They could have been way far away from the nasty murder neighborhood. Anyone who has traveled through Phoenix can pass by $200,000 homes and stumble right into run-down apartments and nasty trailer parks in a heartbeat.

So Kellerman comes up with what he calls a "risk ratio" or "relative risk" which is actually just a statistical "odds ratio". Kellerman uses the term "risk" to make his pronouncement more terrifying. In logic, this is called an "equivocation." But was does a 2.7 odds ratio mean? An odds ratio of 1.0 denotes a positive risk. Negative 1.0 denotes an opposite association. An odds ratio from 1.1 to 3.0 is shown to mean a "weak or nonexistent" association. 3.1 to 8.0 is considered moderate. 8.0-16 is considered strong. And over 16.0 is extremely strong. Kellerman's odds ratiorman's odds ratio of 2.7 then, is meaningless. And his 95% confidence interval measurement of 1.7 to 2.7 shows that his measurement is actually quite good. Tight CI spreads are better. If a CI shifts in one direction, it suggests that the correct number is actually toward that shift. Recall that Kellerman's odds ratio of 43 in his 1986 study was not matched with any CI. Now you know why. :)

So, Kellerman measured odds ratios for other behavioral and environmental factors. I wonder how they aligned with the chance of our criminal buddies getting themselves killed? Let's have a look. The tables might take some adjusting to view properly.

Behavior Odds Ratio Confidence Interval

1) Victim Drank Alcohol 2.6 1.9-3.5

2) Drinking Problems (house) 7.0 4.2-11.8

3) Drinking Problems (work) 10.7 4.1-27.5

4) Victim Drink Prob (work) 20.0 4.9-82.4

5) Housemember drugs 9.0 5.4-15.0

6) Victim uses drugs 6.8 3.8-12.0

7) Physical fights (drinking) 8.9 5.2-15.3

8) Medical attention (fight) 10.2 5.2-20.0

9) Any household arrested 4.2 3.0-6.0

10 Victim arrested 3.5 2.4-5.2

ed 3.5 2.4-5.2

Environmental Factors

1) Home Rented 5.9 3.8-9.2

2) Victim lived along 3.4 2.2-5.1

3) Security access 2.3 1.2-4.4

And now, finally, we get to guns...

4) Gun(s) in home 1.6 1.2-2.2

Handgun 1.9 1.4-2.7

Shotgun 0.7 0.5-1.1

Rifle 0.8 0.5-1.3

Did you guys catch the significance of that? Handguns are insignificant and shotguns and rifles are positively good!!!

Oh, joy.

5) Any gun unlocked 2.1 1.4-3.0

6) Any gun loaded 2.7 1.8-4.0

Ahhh! there's that 2.7 number buried in there. I knew we'd find it.

7) Guns kept mainly for self-defense 1.7 1.2-2.4

So as we can see, with the odds ratios alone, we can see that the entire soundbite that came from this study is bogus. But wait, there's more. Presumably, the cops knew if there was a gun in the homicide house since they searched it (even if it was not used in the homicide.) How was the presence of a gun deduced for the control households? Why, Kellerman just asked them. "Hi, I'm doing a study on homicides with guns (yes,cides with guns (yes, they did tell them the intent of the study) you haven't had anyone killed in your family, do you have any guns there?"

Well, if you were a felon, how would you answer that question? "Hell no I ain't got no guns!!" What if you were unsure whether it was legal to keep a gun? What if the homicide victims were involved in crime and felt (rightly) that they had a good chance of getting killed? Perhaps they have more guns than usual.

Well, Kellerman thought of that. In a separate validation study, he proceeded to get a list of gun permit holders and called them up and asked them if they had a gun. If they said "yes" they were telling the truth. "No" meant the opposite. Well, if one had a permit, that would mean that they not only went through the trouble of government hoop jumping, but they were likely not felons. So not only were they more likely to be more truthful than the control households, they were also more likely to be cooperative with "authorities."

So, what were the gun ownership differences between the case households and the control households that lead to this magic 2.7 "relative risk" figure? It turns out that of the case homicide households, 45.4% of them had a gun somewhere ingun somewhere in the house. Loaded, unloaded, locked, unlocked. And the non homicide control households? 35.8% of them did likewise. A spread of just 9.6 points difference and this guy justifies a two-million dollar CDC grant??!!

I can't take it anymore. For a more thorough discussion on the 1986 study, contact Jews For the Preservation of Firearms Ownership and ask them for their Winter 1997 issue. The 1993 study is handled nicely in the Spring 1995 Tennessee Law Review in an article co-written by Don Kates and others.




"Millions of people armed in the holy cause of liberty, and in such a country as that which we possess, are invincible by any force which our enemy can send against us." Patrick Henry

--- FMail 1.22

* Origin: CyberSupport Hq/Co.A PRN/SURV/FIDO+ (602)231-9377 (176:700/500)

Dave Markowitz's Home Page | Shooting Tech | RKBA | Send me email