The Definition of 'What Works'
'What works' is defined as a statistically significant reduction in the occurrence the crime. Statistically significant means that the intervention or program is responsible for the reduction in the occurrence, and the reduction is not due to chance or part of a trend.
For example, there have been studies on getting guns out of the hands of criminals. The Kansas City Gun Experiment is one such excellent example. Firearm offenses were reduced by nearly 50 percent. This was a very effective way to decrease gun crimes without displacement. This is an empirical approach to finding out what works. Success reduces the prevalence of the violence but any reasonable person knows we are unlikely to ever eliminate crime.
- Pre-test / Post-test
Here is why I consider these junk science.
Anecdote is not evidence. It is a story. It may or may not be true. Even if it were true, it may not be representative; any changes may be caused by third variables; and any changes may only be temporary.
For example, claiming that crime decreased because Virginia passed a concealed weapons law is not evidence. The decrease in crime could have been caused by some other variable; it could have been part of a trend; it could have been part of other less salient crime prevention efforts. It could be anything. This isn't evidence.
Correlation is not causation. Correlation is not evidence. My age correlates with the population growth in America. Does the number of my days on the Earth have anything to do with population rates? Clearly not. We can correlate an unlimited number of variables that have nothing to do with each other. Correlation is not causation. People who suggest causality with correlation are not using evidence.
For example, the states with the strictest gun laws are among the states that have the lowest gun violence. Also, states with higher gun ownership and weak gun laws lead nation in gun deaths. These are true findings, but these do not prove gun bans work. It could be than laws or it could be the culture in those states; it could be police funding; it could be an excellent mental health or an excellent school system. It could be any number of third variables.
Pre-test / post-test isn't evidence. It doesn't account for third variables; it doesn't account for trends; and it doesn't always have to have a logical relationship. This so-called evidence is most frequently used in the gun ban debate. This is not evidence. It is junk science.
For example, pro-gun advocate and economist John Lott often points out that countries that started a gun ban saw an increase in gun crimes after the ban. A line on a graph is not evidence. The examples he provides are not evidence. It could be part of a trend; it could be due to some third variable; it could be due to a random fluctuation. It could be any number of reasons.
What is Evidence (and Why?)
For something to be considered 'evidence' we need rigorous scientific evaluation. This is more than a line on a graph. It is more than a single state or a single country's experience, or even the collection of a group of states or countries.
Evidence is something that shows causality. It rules out trends, it rules out third variables, and it rules out bias. Evidence is very hard to come by. 'Little evidence' doesn't mean something does or doesn't work; it means we need more research to make a conclusion.
The best form of an experiment is a randomized-controlled trial (RTC); an experiment. These are very hard to come by but this is the highest level of evidence. We could also have a matched pair comparison of two groups that are essentially identical except that one group gets an intervention, such as a gun ban, and the other group does not. This is not as rigorous as a RTC, but it is far better than pre-test / post-test, correlation, and isolated examples. Evidence must make use of a comparable group that did not get the intervention; without a comparison group, we really can't get an effect size to determine the effect of the intervention. Pre-test / post-test, correlation, and isolated examples don't show effect sizes.
Researchers often used various types of regression analyses to do research. Regression has its limitations; it should not be used to claim a causal relationship unless all potential third variables are ruled out, which is extremely difficult. This is why good researchers don't claim 'cause' with regression and why the experimental design is the preferred research method.
There have not been, to my knowledge, any experimental empirical studies (using the definition of evidence above) on the effectiveness of gun bans. For various reasons, we don't have very good evidence (read: experimental or quasi-experimental studies) on gun bans' effect on gun crimes.
The 1994 assault weapons ban has yielded "mixed" results. One reason is that the law sought to reduce the prevalence of assault weapons that were not being used in gun crimes. When we don't target the type of crime we are interested in decreasing and we focus on high profile items that are very emotional and fear inducing, we are going to miss what we really need to be addressing.
There are also studies on the availability of guns, which shows that with increased access, increased caliber and increased round capacity, so too do we see an increase in homicides. It would stand to reason that if we can decrease access, caliber, and round capacity that we would decrease gun violence. But we need to have good evidence to make this claim.
In addition to this, we have a phenomena known as the 'etiology-maintenance-treatment confound', which means that which caused a problem isn't necessarily that which maintains the problem, and the solution to the problem may be unrelated to the cause or the maintenance of the problem. In other words, the cause of crimes with guns (whatever that may be) may or may not be related to the solution.
In short, maybe gun bans reduce gun violence, maybe they don't. We really don't yet know.
What Should We Do?
Any good law will seek to identify the type of gun violence we are trying to reduce. Non-specific laws will produce non-specific results. Specific laws will produce specific results. If we are trying to reduce homicides in an inner city neighborhood, our law should be very different than if we are trying to reduce mass shootings. One could argue that if we just ban the guns then we won't have the gun crimes. Well, as noted above, we don't have good evidence for that -- maybe we would, maybe we wouldn't.
If a state legislature or the federal government is going to pass a law that is intended to restrict some aspect of guns in one form or another, it would behoove citizens to have performance measures (experimental or quasi-experimental design) attached at the onset of such a ban. It is also important to have these measures designed by a team of renowned researchers with experience in criminal justice evaluations; these are usually the folks who are not in the news all the time pushing one agenda or another, but are hard at work doing good research. We need good research to make claims about what works and what doesn't.
When you hear the pundits or politicians on TV talking about the only or the best way to reduce gun violence, ask yourself if you have heard the person talk about the different types of gun violence. Not all gun violence is the same. Going into the different ways to reduce the different types of gun violence is well beyond the scope of this article. If someone is proposing a one-size-fits-all approach to reducing gun violence, you can change the channel -- there is no such thing.
As awful as high profile examples of mass shootings are, it is important to remember that high profile events are high profile precisely because they are unusual and unlikely. Making policy based on high profile events is a surefire way to overreact and make inefficient and, worse, ineffective policy. A high profile event is good time find out where a shortcoming of a policy or a failure of a policy might reside, but a high profile event is not necessarily what policy should target. Doing so would result in the majority of cases being marginalized and a strategy designed around an unlikely event.