Which Gun Policies Actually Have Evidence Behind Them?

"Does gun control work" is not a question with an answer, because gun control is not a thing. It is eighteen different policies doing eighteen different jobs, and the central finding of the serious literature is that they do not perform alike.

The RAND Corporation has spent nearly a decade grading that literature. Its Gun Policy in America project screens every empirical study of American firearm policy against fixed methodological criteria, discards the ones that fail, and rates what survives. The fifth and final edition, published in January 2026, covers research through December 2024.

The headline is arithmetic. RAND set out to assess 18 policies against 8 outcomes, which is 144 pairings. After reviewing several thousand candidate studies it found 207 that met its criteria. Those 207 studies produced evidence on 68 of the 144 pairings. Of those, 18 showed an increase or decrease in an outcome, 50 were inconclusive, and for the remaining 76 no qualifying study existed at all.

So for more than half the questions people argue about, nobody has run a study capable of answering them.

This is the sentence the rest of the article depends on, so here it is in RAND's own words: where we conclude that evidence for a policy is weak or inconclusive, that does not mean that the policy is ineffective; the policy itself might well have important effects.

And from the report: they may or may not have the effects they were designed to produce; available scientific research cannot yet answer that question.

An inconclusive rating is a statement about the studies, not about the law. It usually means a policy has been adopted in too few places, or too recently, or that the outcome it targets is too rare for a study to detect a change in it. RAND makes the point directly about firearm stock: laws that govern who may buy a new gun act on a small fraction of the roughly 400 million firearms already in American hands, so even a policy working exactly as designed could produce an effect too small to measure.

Which cuts against both of the confident positions. "The evidence shows this works" is unavailable for most policies. So is "the evidence shows it doesn't."

RAND's five tiers, in ascending order of strength:

Out of 144 pairings, four reach RAND's top tier. Two of them are findings that a policy makes an outcome worse.

Child-access prevention is the strongest-supported policy in American gun law, and it is the one that gets the least political attention. RAND's own explanation for why the evidence is clearer here is mechanical: safe-storage laws act on how the existing stock is kept, rather than on who may buy something new, so there is more for a study to detect.

The two increase findings are worth sitting with, because both policies were adopted on the argument that they would reduce violence. On the concealed-carry finding RAND notes that permit holders are themselves rarely convicted of violent crime, so if the effect is real the mechanism is likely something else, such as more guns in circulation being stolen or diverted, rather than permit holders offending.

Below the top tier the picture gets specific, and several familiar claims do not survive it.

Background checks. RAND rates the evidence moderate that dealer and point-of-sale checks reduce firearm and total homicide. The evidence on private-seller checks specifically, which is what "universal background checks" means in practice, is inconclusive. That distinction rarely survives into the debate.

Assault weapons bans. The Congressionally mandated evaluation of the 1994 federal ban reached a conclusion that has not been improved on since. Assault weapons were used in around 2% of gun crimes before the ban. The evaluators wrote that should it be renewed, the ban's effects on gun violence are likely to be small at best and perhaps too small for reliable measurement. That is not a finding that the ban failed; it is a finding that the category of weapon was too small a share of gun crime for a ban on it to move the total. RAND folds assault weapons and high-capacity magazines into one category and rates only the magazine component as limited evidence of reducing mass shootings and their fatalities. Violent crime effects are inconclusive.

Waiting periods. Moderate evidence of reducing firearm suicide and moderate evidence of reducing total homicide. The most-cited study exploits every state waiting-period change between 1970 and 2014 and estimates roughly a 17% reduction in gun homicides. Its authors note their suicide estimate is not robust across specifications.

Red flag laws. Limited evidence of reducing firearm and total suicide. The underlying studies come from Connecticut and Indiana, where the populations subject to removal orders had suicide rates roughly 30 to 40 times the general public's, and estimate on the order of one suicide averted per ten to twenty removals. A competing analysis estimated far larger effects and was criticised on the grounds that it implies nearly every person subject to an order would otherwise have died. The case studies of orders used against threatened shootings cannot establish that any specific event was prevented.

Permit to purchase. Upgraded in the fifth edition from inconclusive to limited evidence of reducing total and firearm homicide, on the strength of new studies. It rests largely on two natural experiments, Connecticut adopting a licensing requirement in 1995 and Missouri repealing one in 2007, both with single treated states. RAND and the critics agree on the limitation: a single state cannot isolate one law's effect from everything else changing at the same time. Several of the authors are based at a centre that publicly advocates purchaser licensing, which is worth naming without treating it as disqualifying.

Gun-free zones. Inconclusive for violent crime, and no qualifying studies for anything else.

Armed staff in schools. No qualifying studies on any outcome.

Lost and stolen firearm reporting. No qualifying studies on any outcome.

This single fact reorganises the whole subject, and it is missing from most arguments about it.

Gun suicides have outnumbered gun homicides in every year since 1981, when the CDC’s published series begins, and in recent decades they have been a majority of all gun deaths. Any figure presented as "gun violence" that is actually the total is describing something roughly 60% of which is people killing themselves. That is not a semantic complaint. It changes which policies could possibly matter, because the policies with the clearest evidence behind them are the ones that act on suicide.

It also raises the objection that people who cannot reach a gun will use something else. The strongest evidence against complete substitution is a California cohort study following 26.3 million adults for up to twelve years. New handgun owners showed sharply elevated suicide risk driven entirely by firearm suicide, with no elevation at all in suicide by other means. Risk peaked shortly after acquisition and persisted for years afterwards.

What that supports is that reducing firearm access reduces firearm suicide, and in several designs total suicide. What it does not support is the stronger claim that method substitution never happens.

Download this conclusion: suicide

This section as one image: what American gun deaths are made of, and the cohort study on what happens when access changes.

Download the card (PNG, 1,200 × 630)

Estimates of how often Americans use a gun defensively run from about 55,000 to 116,000 a year, from the National Crime Victimization Survey, up to 2.5 million, from a 1993 telephone survey.

The gap is a survey-design problem rather than a mystery. The NCVS is large and government-administered, asks about victimisation first and never asks about defensive gun use directly, which critics say undercounts it. The 2.5 million figure comes from 66 respondents out of 4,977 reporting a defensive use, extrapolated to the national population. In a survey measuring something rare, a small share of false positives produces enormous overestimates, and internal cross-checks in that survey imply more defensive uses during burglaries than there were burglaries. RAND treats the 2.5 million figure as not plausible against better sources and treats the frequency question as unresolved.

RAND found five qualifying studies on any policy's effect on defensive gun use. All five were inconclusive. It is one of the least-studied outcomes in the entire field, which is notable given how central it is to the argument.

Four figures, two from each direction.

"40% of guns are acquired without a background check." Traces to a 1994 telephone survey in which about 251 people answered the acquisition question, roughly 36% reporting a non-dealer source. The survey largely predates the Brady Act. A 2017 study of 1,613 gun owners put the modern figure at 22%, rising to 57% in states that do not regulate private sales and falling to 26% in states that do.

"A gun in the home is 43 times more likely to kill a family member than an intruder." A raw count ratio from a 1986 study of three counties, not an adjusted risk estimate. The same author's later case-control study reported an adjusted odds ratio of 2.7. The 43 figure should not be quoted as a risk ratio.

"94% of mass shootings happen in gun-free zones." Produced using a narrow definition of mass shooting and one researcher's own database. A counter-analysis using a broader definition found around 10%. A fact-check applying the original method to a comparable window got about 86%. The disagreement is definitional, not factual, and neither number describes the same set of events.

"The US has 31% of the world's mass shooters." From a 2016 study. A rebuttal including many more foreign incidents put the figure between 1.4% and 3%. The original author declined to release the underlying data; the rebuttal's inclusion criteria are themselves disputed. Unresolved in both directions.

One more, because it explains why two people can quote wildly different mass shooting counts and both be accurate. Databases reporting several hundred mass shootings a year define one as four or more people shot, anywhere. Databases reporting a handful define it as four or more killed, in public. Neither is wrong. They are counting different things, and a figure from one cannot be set beside a figure from the other.

Gun control advocates must concede that the homicide-reduction claim attached to universal background checks and assault weapons bans outruns the evidence RAND found. That the federal assault weapons ban's own Congressionally mandated evaluation concluded any effect would likely be too small to measure. That for 76 of 144 policy-outcome pairings there is no qualifying study at all. And that defensive gun use, one of the main objections to several of these policies, has barely been studied.

Gun rights advocates must concede that "gun laws don't work" is not what inconclusive evidence means, and RAND says so in as many words. That four policy findings do reach the top evidence tier, and two of them find that policies adopted to reduce violence increased it. That child-access prevention has supportive evidence across three separate outcomes. And that "more guns, less crime" was reviewed by a National Research Council panel that could not confirm it, with fifteen of sixteen members stating the scientific evidence did not support the dissenting member's position.

The honest summary is narrower than either campaign wants. A small number of gun policies have real evidence behind them, mostly acting on suicide and on children. A larger number have been studied and produced nothing conclusive. The majority have not been studied adequately at all. And the two policies with the strongest evidence of making things worse were both passed on the argument that they would make things better.

1. RAND assessed 18 policies against 8 outcomes. Of 144 pairings, 18 showed an effect, 50 were inconclusive, and 76 had no qualifying study at all.

2. Four pairings reach the top evidence tier. Two of those are findings that a policy increases homicide.

3. Inconclusive means the studies cannot answer the question, not that the policy failed. RAND states this directly, and it defeats the strong claim in both directions.

4. Most American gun deaths are suicides, 58% in 2023 and 62% in 2024, and the policies with the clearest evidence are the ones acting on suicide and on children.

5. The assault weapons ban evaluation found the weapons covered were used in about 2% of gun crimes, so any effect on the total would likely be too small to detect.

6. Defensive gun use estimates differ by a factor of forty, and all five qualifying studies of any policy's effect on it were inconclusive.

Download the conclusion

The verdict above as one image, with RAND’s 144 pairings as its chart. It is 1,200 by 630 pixels, the size social networks use for link previews, and anyone may download it and post it.

Download the card (PNG, 1,200 × 630)