Comment on "The Effects of Gun Prevalence on Burglary: Deterrence vs Inducement" by Philip J. Cook & Jens Ludwig

By David B. Kopel

This is a comment which appears in the book Evaluating Gun Policy: Effects on Crime and Violence (Brookings Institution, 2003). Philip J. Cook & Jens Ludwig, editors. For more on this subject see David B. Kopel, Lawyers, Guns, and Burglars, 43 Arizona Law Review 345 (2001).

The Cook-Ludwig article is a valuable contribution to the social science of firearms, but its conclusion is somewhat overstated. Indeed, given the constraints of what was examined, the article could not possibly prove that, overall, the burglary-inducing effect of guns in America exceeds the burglary-deterring effect.

International versus Local Effects

To begin with, the article's new research examines only data from America. The authors rather brusquely dismiss the comparative international data. While the authors are certainly right that there is very ample room for more detailed comparative research, the authors do not acknowledge the possibility that their findings are consistent with the comparative data suggesting that gun density is one important reason why the U.S. home invasion burglary rate is lower than in other countries.

To see why this is so, let us step back for a moment, and consider two highly-regarded books about firearms. In Crime is not the Problem, Franklin Zimring and Gordon Hawkins (1999) suggest that high rates of firearms ownership help explain why the U.S. homicide rate is higher than the rate in many other nations--principally because American assaults are more likely to turn into homicides, because Americans perpetrating an assault are more likely to have a firearm handy. Of course the Zimring-Hawkins international comparison is subject to many of the same data-comparability objections which Cook and Ludwig raise about international burglary comparisons, but let us, for purposes of argument, accept that the Zimring-Hawkins thesis is true.

Now consider John Lott's famous More Guns, Less Crime(Lott, 2000). Like Cook-Ludwig, Lott looks mainly at county-level American data. He finds that when adults who pass a background check and (in most states) a gun safety class are allowed to obtain permits to carry handguns for lawful protection, the violent crime rate, including homicide, falls. The more people who carry, the more the crime rate falls. Although Lott's work has been subjected to intensive and not always persuasive data-torturing by people determined to disbelieve his conclusions, let for the sake of argument assume that Lott's study is a valid.

It is entirely possible that the Zimring-Hawkins international conclusion and the Lott county-level conclusion are both true. That is, as Zimring-Hawkins suggests, America's much higher rates of gun density do result in a much higher American homicide rate. (Perhaps in part because high gun density creates a large pool of stealable guns which are readily and casually obtained by the kind of thugs and other miscreants who go off half-cocked and all-crocked, starting fights with each other on street corners where drugs are sold or in the parking lots of bars after closing time. This would be a population which carries guns regardless of whether it is legal to do so.) At the same time, it is possible that allowing non-criminal adults to carry guns pursuant to a government permit might deter or thwart a certain amount of crime, an amount far larger than the tiny amount of crime which these permit holders would perpetrate.

Thus, greater American gun density on the macro scale (Zimring-Hawkins, international) might cause more homicide, while greater gun carrying at the micro scale (Lott, county-level) might reduce homicide. It would hardly be surprising that increased gun availability among criminals (the kind of people who perpetrate assaults) would increase unlawful homicide, while increased gun availability among the law-abiding (people who obtain government-issued permits) would reduce unlawful homicide. At least as a logical matter, there is no reason why both effects cannot be true.

Again, I am not necessarily arguing that Zimring-Hawkins or Lott are correct; Kleck's much larger cross-national data set finds no statistically significant relation between gun availability and homicide, and Kleck argues that "shall-issue" concealed handgun laws result in too small an increase in actual gun carrying for Lott's crime-reduction results to be plausible. (Kleck, 1997, pp. 253-54, 372). Rather, my point is that as a logical matter, the Zimring-Hawkins findings and the Lott findings could both be true.

What Cook-Ludwig attempt to do is use county-level data to show that more guns leads to more burglary. It is entirely possible for this micro-level hypothesis to be true and for the Kopel (2001) and Kleck (1997) hypothesis that, on the macro/international scale, America's much higher gun density helps produce a lower burglary rate. Indeed, it is easy to see how Cook-Ludwig and Kopel/Kleck could be simultaneously true.

Compared to foreign burglars, a higher percentage of American burglars are apt to avoid occupied residences, for fear of getting shot, and this is why American home invasion burglaries are rarer than in other countries (Kopel, 2001; Kleck, 1997). Nevertheless, an especially reckless or foolhardy cohort of American burglars are not deterred, and these burglars actually prefer to burglarize homes with guns, since guns are such uniquely valuable loot; thus counties with more guns have relatively higher burglary rates (Cook-Ludwig). Thus, at the most, the Cook-Ludwig article tells us about the relative effects of gun density within America, but the article does not disprove the evidence from international burglary data.

What Cook and Ludwig Measured

A second reason why the Cook-Ludwig conclusion should not be so strongly stated is that the study itself looks at only a subset of counties. Low population counties are excluded. The effect of this exclusion may not be neutral. Because low-population counties are especially likely to have a small, scattered population that is far away from law enforcement assistance, such counties may be precisely the places where people are most likely to rely on guns to resist burglars. Because of the relative impotence of sheriffs and police (due to long distances) and because of the paucity of neighbors to report suspicious persons, the marginal burglary-deterrent effects of firearms in such counties might be especially high. Notably, smaller counties tend to be more rural, and usually tend to have higher rates of gun ownership and lower rates of burglary than more heavily-populated counties within a given state.

In short, the Cook-Ludwig article proves, at most, that within medium to large American counties, the burglary-inducing effects of marginal increases in gun density outweigh the burglary-deterring effects of marginal increases in gun density.

Even for the medium and large counties, the findings are not nearly as clear as Cook-Ludwig urge. The correlation coefficient is a tiny 0.01--meaning that different counties yielded extremely disparate results. If the Cook-Ludwig article included a scatter graph, we would not see points bunched in clusters. To the contrary, the points would more likely appear like a vague, dissipated cloud. To suggest that this cloud of weakly-correlated results can be turned into a generally-applicable principle of cause-and-effect is an overstatement.

Of course the study does not attempt to quantify the possible secondary costs and benefits of home firearms and burglary. For example, we know that about 98% of burglary alarm activations are "false positives" and that responding to false alarms wastes a great deal of police resources. (Blackstone et al., 2002). To the extent that gun owners are less likely to rely on alarm systems, because of their confidence in their ability to protect their families and homes personally, gun ownership would create a positive social externality, by reducing the false alarm drain on police resources.

Treating Guns Like Cars

Even if, as Cook and Ludwig conclude, the collective decisions of many families to purchase firearms does, in the aggregate, very slightly raise the burglary rate, the families are not necessarily acting irrationally or anti-socially. After all, the collective decisions of many families to purchase automobiles (or to purchase second or third cars for the family) very likely does, in the aggregate, raise the community automobile accident rate (including accidents involving people who don't own cars, such as when cars hit bicyclists or pedestrians). More cars, more driving. More driving, more accidents. Also, more autos, more auto theft -- and thus more autos in criminal hands.

Yet before telling families not to buy cars, we would consider what the families gain, in exchange for the incremental accident rate increase and the incremental increase in the supply of cars to criminals which is inflicted on the community. The family with the car gains mobility and independence, the option of coming and going according to the family's needs. Many people also derive pleasure from driving.

Similarly, whatever the effects of firearms on aggregate burglary rates, the family with firearms gains the opportunity to participate in shooting sports, as well as the ability to use a firearm to protect the family in various situations which might have nothing to do with home invasion burglary.

Prior Scholarship

In summarizing prior literature on the burglary/gun issue, Cook and Ludwig go overboard in their claims that evidence is conflicting and inconclusive. To the contrary, well-designed studies of prisoners (J. Wright & Rossi, 1994), of active burglars (R. Wright & Decker, 1994), and of ordinary citizens (Ikeda, et al., 1997) consistently find that active armed defense against burglars by victims is frequent and is feared by most burglars. Against this evidence, Cook-Ludwig merely cite studies which do not directly ask about use of firearms against burglars, but simply allow for that information to be volunteered. (Cook, 1991; Kellermann et al., 1995).

Moreover, the brush-off of studies which do ask about armed defense is unnecessary for Cook-Ludwig's thesis. It is possible that armed home defense is frequent and is widely-feared by most burglars and that reckless/foolhardy burglars--attracted by guns as loot, or excited by the thought of a gun battle--commit extra burglaries when guns are thought likely to be in the victims' home. Thus, Cook-Ludwig's conclusion doesn't depend on armed home defense being rare, and their article does nothing to disprove the research suggesting that armed home defense is common and widely feared by most but not all burglars.

But is the conclusion technically valid? Perhaps not. First, the article uses the percentage of suicides which are committed with guns (PSG) as a proxy for changes in gun ownership over time. However, the creator of the PSG proxy, Gary Kleck, has demonstrated that the PSG is not valid measure of gun ownership changes over time, although the PSG can measure differences between areas (or groups) at a common point in time (Kleck, 2002).

Indeed, the PSG correlates very well to surveys of household gun ownership--but this is simply correlation to another proxy, not correlation to the thing itself (actual gun ownership). Household surveys of American gun ownership rather consistently yield disparate results (as low as a third of homes, to as high as half of homes) which are well outside the margin of error (Kleck, 1997)--so we know that household surveys are not a particularly accurate reflection of reality.

The accuracy of household gun surveys is seriously impaired by the "dark figure" created by gun owners who refuse to disclose their gun ownership to a pollster (Kleck, 1997). This fact becomes especially important in the burglary context. Cook-Ludwig plausibly suggest that in many cases for which stealing a gun was the sine qua non of the burglary, the burglary may be an inside job--perpetrated by an acquaintance who personally knows the victim and knows that the victim owns a gun. Obviously the kind of gun owner who is at greatest risk of such targeted burglary is the gun owner who freely tells people that he owns guns--in other words, the kind of gun owner who is likely to answer a household gun survey truthfully. Conversely, the secretive gun owner--who won't talk to a pollster--may be more likely to keep her gun ownership a secret from her acquaintances as well. Thus, the secretive gun owner would be unlikely to be the victim of a targeted gun-seeking burglary.

Now since the PSG and the household surveys correlate so well, what Cook-Ludwig may have ended up measuring is not the actual availability of guns in various counties, but the actual availability of loose-lipped gun owners. The proper conclusion for the article might be changed from "Guns cause burglary" to "Talking about guns with the wrong people causes burglary."

Cook and Ludwig forthrightly acknowledge the problem of reverse causality: Do guns cause burglary (by inducement), or does burglary cause guns (as a citizen response to rising crime). One of Cook-Ludwig's approaches to the causality problem is to use an "instrumental variable" (IV)--namely the percentage of a state's population which was rural in 1950. This choice is based on the common-sense and explicitly-stated assumption that rurality (or rural heritage) is correlated with participation in the American gun culture. The choice of rurality as an IV also requires the assumption that rurality does not affect participation in burglary. If this latter assumption is incorrect, then we have not escaped from the problem of reverse causality. Cook and Ludwig do not offer evidence that rurality does not affect burglary.

In the panel study, the causation problem recurs. Cook and Ludwig address this by lagging burglary rates and gun ownership; thus, they account for someone who buys a gun in 1998 in response to burglary conditions in 1996. But what about people who buy guns in 1996 because of burglary conditions in 1996? The Cook-Ludwig model appears to assume that gun-buying in response to burglary is always lagged by a year or more from actual burglary conditions. The model does not account for a current increase in the burglary rate causing a current increase in gun ownership.

The Cook-Ludwig article makes a passing reference to a study by Duggan (2001) in which the circulation of Guns & Ammo magazine was used a proxy variable for gun ownership. As Moody and Marvell (2002) detail, Duggan uses the coefficient from the proxy variable as if the proxy were the actual variable itself. Moody and Marvell demonstrate how this error undermines Duggan's assertion that "guns cause crime" more than "crime causes guns." Moody and Marvell suggest that the same problematic use of the coefficient for the proxy variable affects the Cook-Ludwig study as well, undermining Cook and Ludwig's claim to have solved the causality problem. If the elasticity of the proxy (PSG) with respect to actual gun ownership is less than one, then Cook and Ludwig's causation conclusion would be reversed: burglary would actually cause guns to a much greater degree than guns would cause burglary.

Cook and Ludwig offer an additional attempt to address causality; they find a relationship between burglary victimization (as reported by NCVS respondents) and gun prevalence in the surrounding area (as measured by the PSG). But Cook and Ludwig fail to control for crime rates in the surrounding area. Cook-Ludwig's gun prevalence measure thus serves as a surrogate for rates of crime (and, therefore, the number of criminals) in the surrounding areas. Hence, the coefficient for the gun prevalence variable at least partly represents a response to the presence of so many criminals in the respondent's neighborhood, rather than the burglary-inducing effect of gun availability. More criminals, more burglary. The data do show a relationship between burglary and gun prevalence, but not necessarily, as Cook and Ludwig argue, because guns induce burglary, rather than vice versa.

Even so, Cook and Ludwig have advanced the debate. At least in the scholarly world, the gun control discussion continues to evolve constructively, as scholars abandon arguments about whether guns, in general, are good or bad, and instead focus on the effect of firearms in particular situations. The Cook-Ludwig article provides an important contribution to the small but growing body of research on gun and burglary.

I would like to thank Paul Blackman, Gary Kleck, Carlisle E. Moody, and Iain Murray for their helpful analysis.

Citations for items not cited in Cook-Ludwig:

Blackstone, Erwin A., Simon Hakim, and Uriel Spiegel (2002) "Not Calling the Police (First)" Regulation, Vol. 25 (No. 1, Spring): 16-19.

Kleck, Gary (2002) "Measures of Gun Ownership Levels for Macro-Level Crime and Violence Research" (manuscript).

Moody, Carlisle E. and Thomas B. Marvell (2002) "Pitfalls of Proxy Variables: Comment on Mark Duggan, "More Guns, More Crime'" (manuscript).

Zimring, Franklin E. and Gordon Hawkins (1999) Crime Is Not the Problem: Lethal Violence in America. N.Y.: Oxford Univ. Pr.