restricted access Appendix A. Methodology
In lieu of an abstract, here is a brief excerpt of the content:

Appendix A  Methodology The Gun-Homicide Connection 1. Case-Control and Ecological Studies The gold standard among study designs is the experimental or interventional randomized control trial. The goal is typically to evaluate the effects of an exposure or a treatment for a disease or condition. Randomization of assignment of subjects to different exposure or treatment groups tends, on average, to balance the groups on other factors that could influence the outcome. All else being equal, the randomized control trial is more likely than any other to give the correct answer about whether the treatment works or the exposure is causally related to the outcome (Rothman 1986; Koepsell 2001). By contrast, in nonexperimental or observational studies, the investigator has no control over which subjects are exposed and which are not. In most circumstances, limitations imposed by society, ethics, and cost restrict epidemiologic research to nonexperimental studies (Rothman 1986; Koepsell 2001). Virtually all firearm studies have been nonexperimental studies. One type of nonexperimental study is the case-control study. “The sophisticated use and understanding of case-control studies is the most outstanding methodologic development of modern epidemiology” (Rothman 1986, 62). A major advantage of the case-control study design is that it is efficient for the study of rare outcomes (e.g., suicide or homicide), especially where exposure is common (e.g., to guns in the United States) (Cummings, Koepsell, and Roberts 2001). In case-control studies, the investigator begins by identifying a group of individuals known to have the outcome of interest (the cases) and compares them to a group of individuals known not to have the outcome (the controls). The controls must be people who would have been counted as cases if they 261 had the outcome of interest (Cummings, Koepsell, and Roberts 2001; Koepsell 2001). Epidemiologists have used case-control studies to examine risk factors for many injuries, including suicide and homicide (Cummings, Koepsell, and Roberts 2001). Like all study designs, case-control studies have various limitations . Three of the potential weaknesses in case-control studies of guns and violent death are measurement error (e.g., the possibility of recall bias), confounding , and reverse causation. In many but not all (e.g., Cummings, Koepsell, Grossman, Savarino, and Thompson 1997) of the case-control studies of homicide (or suicide) and guns, case and control families are asked about the presence of guns in the home. After a gun homicide (and particularly a gun suicide) occurs in the home, it is likely that all adults will know about the presence of a firearm. The relationship between guns and homicide (or suicide) could be overstated if guns in control households are underascertained while guns in case households are more accurately estimated (Kleck 1997b). The first case-control study of guns and homicide (Kellermann et al. 1993) and an early case-control study of guns and adolescent suicide (Brent et al. 1991) have been criticized in this regard (Kleck 1997b). The Kellermann study found that 36 percent of case households contained handguns, compared to 23 percent of control households. The Brent study found that 55 percent of case households contained handguns (72 percent contained firearms), compared to 20 percent of control households (37 percent contained firearms). These large differences suggest that the size of the underascertainment problem , at least by itself, is not important enough to eliminate the differences found in firearm availability between case and control households. (Kleck’s tables 8.2c and d make no sense.) A central problem for all epidemiologic studies is the possibility of confounding . The goal of the case-control study is to estimate the association between an exposure (e.g., guns) and an outcome (e.g., homicide). Confounding occurs when the estimate is erroneous because of failure to account for a third factor that is associated with both the exposure and the outcome (Cummings , Koepsell, and Roberts 2001). In case-control studies of guns and homicide , for example, one can imagine many potential confounders (e.g., aggressive behavior) and question the extent to which the distorting effect of potential confounders has been adequately accounted for in the reported associations. appendixes 262 The Kellermann et al. (1993) study has been criticized on the grounds of possible confounding. Gary Kleck (1997b, 244), for example, argues that A. L. Kellermann and coauthors “failed to control for whether subjects were drug dealers or members of street gangs, persons who are both much more likely to own guns and far more likely to become victims of homicide.” However, Kellermann did...