From: [j--l] at [upchrch.UUCP] (Joel Upchurch) Newsgroups: talk.politics.guns Subject: NEJM Study on Homicide: I Actually Read It. Date: Fri, 15 Oct 93 21:51:34 EDT After all the discussion of this study, I decided to go down to the public library and read it. It is in the 10/7/93 issue of NEJM and is only eight pages long, so I copied it and took it home and read it. I wasn't expecting much, since I'm anti-gun-control and I'm generally underwhelmed by what passes as scientific research in the social sciences in any case. I was surprised: This appears to be a fairly well done piece of research; much better than most of the stuff I've read. The kicker is that I'd be hard put to prove that whoever wrote the conclusions section ever read the body of the article. This paper has ten coauthors, which causes me to suspect that what we have here is some peons who did all the work and a couple of senior people who put their imprimatur on the whole thing by writing the conclusions and the abstract. I think that reading the abstract by itself might tend to give a misleading impression of the content of the study. The study consisted of analyzing all the homicides taking place in Kings County, WA and Shelby County TN for five years and in Cuyahoga County, OH for two and a half. They only analyzed homicides where the victim was in or near their home. They then recruited someone close to the victim who could answer personal questions about then. They then recruited controls of the same race, sex and age from that neighborhood. They were able to establish 316 'matched pairs' for whom they were able to obtain complete data. They interviewed the subjects on various topics concerning lifestyle. They then did a statistical analysis of the data. The technique they used was, "Mantel-Haenzel chi-square analysis for matched pairs was used to calculate the crude odds ratio associated with each variable. Multivariate analyses used conditional logistic regression, the appropriate technique for a matched-pairs design." I haven't a clue what that means; the main Orlando library didn't have a word on logistic regression in the statistics section. If someone could post an explanation of what this means, I'd appreciate it. Anyway they came up with this table: Table 3. Univariate Analysis of Hypothesized Risk on Protection Factors Derived from Data on 388 Matched Pairs Of Case Subjects and Controls. --------------------------------------------------------------------------- Variable Case Controls CRUDE ODDS Subjects RATIO (95% CI)* Behavioral factors no @ Any household member drank 277 (73.3) 217 (55.9) 2.4 (1.7-3.3) alcoholic beverage Case subject or control drank 238 (62.8) 162 (41.9) 2.6 (1.9-3.5) Alcoholic beverages Drinking caused pnoblems in 92 (24.8) 22 (5.7) 7.0 (4.2-11.8) the household Any household member had trouble 32 (9.0) 3 (0.8) 10.7 (4.1-27.5) at work because of drinking Case subject or control had trouble 20 (5.7) 1 (0.3) 20.0 (4.9-82.4) at work because of drinking Any household member hospitalized 41 (11.4) 9 (2.3) 9.8 (4.2-22.5) because of drinking Case subject or control hospitalized 28 (7.6) 2 (0.5) 14.0 (4.7-41.6) because of drinking Any household member used 111 (31.3) 23 (6.0) 9.0 (5.4-15.0) illicit drugs Case subject or contro used 74 (20.3) 16 (4.2) 6.8 (3.8-12.0) illicit dnugs Any physical fights in the home 92 (25.3) 13 (3.4) 8.9 (5.2-15.3) during drinking Any household member hit or hurt 117 (31.8) 22 (5.7) 7.9 (5.0-12.7) in a fight in the home Any Family member required medical 62 (17.3) 8 (2.1) 10.2 (5.2-20.0) attention because of a fight in the home Any adult household member 103 (29.9) 70 (18.8) 2.1 (1.4-3.0) involved in a physical fight outside the home Any household member arresed 193 (52.7) 90 (23.4) 4.2 (3.0-6 0) Case subject or control arrested 132 (36.0) 60 (15.7) 3.5 (2.4-5.2) Environmental factors Home rented 271 (70.4) 183 (47.6) 5.9 (3.8-9.2) Public housing 41 (11.1) 38 (9.8) 1.5 (0.7-3.3) Case subject or control lived alone 103 (26.8) 46 (11.9) 3.4 (2.2-5.1) deadbolt locks 243 (68.8) 292 (75.3) 0.8 (0.5-1 0) Window bars 71 (19.2) 81 (20.9) 0.8 (0.5-1 3) Metal security door 95 (25.4) 104 (26.8) 0.9 (0.6-1.3)+ Burglar alarm 26 (7.1) 43 (11.1) 0.6 (0.4-1.0) Controlled security access 52 (13.9) 38 (9.8) 2.3 (1.2-4 4) to residence dog or dogs in home 94 (24.2) 87 (22.4) 1.1 (0.8-1.6) Gun or guns in home 174 (45.4) 139 (35.8) 1.6 (1.2-2 2) Handgun 135 (35.7) 90 (23.3) 1.9 (1.4-2 7) Shotgun 50 (13.6) 65 (16.8) 0.7 (0.5-1.1) Rifle 45 (12.2) 54 (13.9) 0.8 (0.5-1.3) Any gun kept unlocked 105 (29.6) 69 (17.8) 2.1 (1.4-3.0) Any gun kept loaded 93 (26.7) 48 (12.5) 2.7 (1.8-4.0) Guns kept primarily for 125 (32.6) 86 (22.2) 1.7 (1.2-2.4) self-defense -------------------------------------------------------------------------- * Results were caculated with the Mantel-Haenszel chi-square analysis for matched pairs. CI denotes confidence interval. @ Percentages reflect the proportion of subjects who responded yes among all subjects who gave a response. + The value is statistically significant; the upper bound of the 95 percent confidence interval is 1.0 because of rounding. I thing I found a little odd was that victims were less likely to have a shotgun or rifle in the home, but the abstract equates the problem with all guns, not just handguns. Anyway they go through their multivariant analysis and produce this table: Table 4. Variables included in the Final Conditional Logistic-Regression Model Derived from Data on 316 Matched Pairs of Case Subjects and Controls.* -------------------------------------------------------------- Variable Adjusted Odds Ratio (95% CI) Home rented 4.4 (2.3-8.2) Case subject or control lived alone 3.7 (2.1-6.6) Any houshold member hit or hurt 4.4 (2.2--8.8) in a fight in the home Any household member arrested 2.5 (1.6-4.1) Any household member used illicit drugs 5.7 (2.6-12.6) Gun or guns kept in the home 2.7 (1.6-4.4) -------------------------------------------------------------------- *Conditonal logistic-regression analysis requires that data on all the variables of interest be availlable tor both case subjects and their matched controls. Therefore. 72 pairs with missing data on any of the six variables of interest were excluded from this analysis. CI denotes confidence interval. Anyway by this point looking at the numbers I'm starting to think that they are using the term 'odds' in some weird technical sense that doesn't have much relation to what we mean when we are talking to our bookie. Note that while the AOR for guns was 2.7, the AOR for rented was 4.4. Imagine if they had used that as a headline instead. The National Board of Realtors would have loved them. "Buying a Home: It's not just a good investment, it may save your life.". The study also reasonably pointed out that reverse causation may account for the guns. The victims may have had guns in their homes because of a greater danger of violence against them. It's is also possible that both are caused by some other third factor. One thing that bothers me is that there is no way to verify the claims of the control group about firearms in their homes. If 35 out of 388 controls had lied or were mistaken about not having guns in their home, then the correlation disappears. I recall Kleck saying something about people in low income groups tending to be less than forthcoming when nosy strangers ask questions about their guns. If some other member of the household had a gun, the control might not even know about it. The proxies for the case subjects would have less opportunity to lie about it because the gun may have used in the homicide or discovered by the police in the subsequent search of the premises, Since this study was done in cooperation with local police departments, I think it would have been good if the study had made more effort to verify the proxies statements by checking police records. A lot of the information could have been checked against records for domestic dispute call to that location, DWI and drug records, as well as criminal records Joel Upchurch @ Upchurch Computer Consulting uunet!aaahq01!upchrch!joel 718 Galsworthy Ave. Orlando, FL 32809-6429 phone (407) 859-0982