The Aim and the Mandated Evaluation

The federal assault weapons ban stated its purpose in functional terms. The prohibition aimed to reduce the criminal use of the semiautomatic firearms and the large capacity ammunition feeding devices that the statute covered. That aim differed from a broader ambition to reduce gun violence in general, and the difference matters for every judgment that follows in this article. A statute can succeed against the narrower target named in its own text and register little movement against the wider category, or it can miss the narrower target while wider trends move for unrelated reasons. The evidence assembled under the statute’s own evaluation mandate answers the narrower question and the wider question unevenly, and the unevenness is not a defect in the record.

The statute’s formal identity fixes the boundaries of that record. The ban was Subtitle A of Title XI of the Violent Crime Control and Law Enforcement Act of 1994, Public Law 103-322. It was in force from September 13, 1994 to September 13, 2004, a full decade bounded by dates written into the law at enactment. The prohibition ended by its own sunset. It was not repealed, and its expiration was not a later verdict delivered by a subsequent Congress on the evidence. The sunset operated on the schedule the enacting Congress had set, which means the ten year lifespan was part of the design from the first day.

The covered categories also need precise statement, because the evaluation literature measures the statute that Congress wrote rather than a generic idea of such a law. The statute barred manufacture, transfer, and possession of semiautomatic assault weapons as the statute defined them, through a list of named models and a two feature test. It barred transfer and possession of large capacity ammunition feeding devices, meaning devices holding more than ten rounds, made after enactment. Semiautomatic assault weapon is the statute’s own defined term, and the research literature uses the term large capacity magazines for the magazine category. Possession and transfer of covered firearms and devices lawfully possessed on September 13, 1994 remained lawful throughout the decade. That grandfather provision is treated in full in the next section, and it is restated wherever effects are discussed in this article, because no finding about the prohibition can be read accurately without it. Readers seeking the full definitional breakdown in one place can consult the statute profile of the ban.

What was the federal assault weapons ban meant to accomplish?

The statute aimed to reduce criminal use of semiautomatic assault weapons as defined in the law and of large capacity ammunition feeding devices holding more than ten rounds. It barred manufacture, transfer, and possession of newly covered items after September 13, 1994, while grandfathered items lawfully held on that date remained lawful.

That answer states the aim, and the rest of the article tests performance against it. Performance is a harder matter than purpose, because criminal use is not observed directly in a single national series. The evaluators had to work through proxies such as law enforcement trace requests, guns recovered by police in selected cities, homicide records, and market prices, and each proxy illuminates part of the aim while leaving other parts in shadow. A fall in trace requests for the covered firearms, for example, speaks to the mix of guns reaching police attention, and it speaks less directly to the number of crimes committed. The mandated evaluation is valuable precisely because the researchers named that measurement problem in print rather than treating any single series as a complete verdict.

The mandate itself came from section 110104 of the subtitle. That provision directed the Attorney General to evaluate the effects of the subtitle and to report to Congress within 30 months. Here, Congress ordered its own measurement as part of the statute. The reporting clock ran from enactment, so the first major accounting was due while the prohibition had been operating for only two and a half years. The Attorney General’s department housed the machinery for the work, the National Institute of Justice funded the evaluation, and researchers at the Urban Institute conducted it. The resulting interim report, written by Jeffrey A. Roth of the Urban Institute and Christopher S. Koper and published in March 1997 under the title Impact Evaluation of the Public Safety and Recreational Firearms Use Protection Act of 1994, arrived on the timetable the mandate had set.

A commissioned verdict of that kind is rare among federal statutes. Most laws produce administrative reports on spending, enforcement counts, and program operations, and those reports can describe activity without estimating effect. The section 110104 mandate asked for effects, which pushed the evaluators toward comparison: conditions before enactment against conditions after enactment, covered firearms against firearms the statute left untouched, and jurisdictions with their own bans against jurisdictions without them. The evaluators’ caution about attribution, visible in the interim report of 1997 and in the final assessment of 2004, followed directly from the question Congress had posed.

The evaluation did not end with the interim report. A summary of the early work appeared as a National Institute of Justice Research in Brief titled Impacts of the 1994 Assault Weapons Ban: 1994 to 1996, published in March 1999. The final evaluation arrived after the sunset approached: Christopher S. Koper, with Daniel J. Woods and Jeffrey A. Roth, published An Updated Assessment of the Federal Assault Weapons Ban: Impacts on Gun Markets and Gun Violence, 1994 to 2003 through the Jerry Lee Center of Criminology at the University of Pennsylvania in June 2004, as a report to the National Institute of Justice. The arc from a 1997 interim report to a 2004 final assessment means the statute was measured at the beginning of its life and again near the end of its life, under the same research lineage, with the later work able to revisit the earlier work’s provisional judgments.

That lineage matters because the two ends of the arc did not say identical things, and the differences are instructive rather than embarrassing. The interim evaluation warned that the mandated review period was too short and too close to enactment to judge effects on street violence, given the grandfathered stock in circulation. The final evaluation, covering the period from 1994 to 2003, could report that gun crimes involving the covered firearms had declined in the six study cities it examined, while also reporting that criminal use of non banned semiautomatic firearms fitted with large capacity magazines had held steady or risen. Neither report, read whole, supports a one sentence summary, and the history of quotation since 2004 demonstrates how readily one sentence summaries get manufactured from a record that resists them.

The method of this article follows from that resistance. Findings are reported question by question. Criminal use of the covered firearms is one question. Overall gun violence is another question, with a different evidence base and a different answer. Each question carries its own researchers, its own period, its own publication, and, where a synthesis exists, its own rating from the 2020 edition of The Science of Gun Policy, published by the RAND Corporation. Collapsing those questions into a single verdict would repeat the error that partisans on both sides commit when quoting the evaluation literature. The same final report supplies a sentence about crimes with the covered firearms falling and a sentence about the researchers being unable to clearly credit the prohibition with any of the national drop in gun violence during the period. Both sentences are accurate. Both come from Christopher S. Koper, Daniel J. Woods, and Jeffrey A. Roth writing for the National Institute of Justice in June 2004. An honest account keeps both sentences in view and explains the different questions each sentence answers.

Selective quotation works because the commissioned record is genuinely mixed, and mixed records offer material to every camp. One camp quotes the measured declines in criminal use of the banned models across the six cities in the final evaluation, and the quotation is accurate. The opposing camp quotes the same report’s judgment that the prohibition could not be clearly credited with the national decline in gun violence, and that quotation is accurate as well. A third quotation, also accurate, holds that any renewed prohibition’s effects on gun violence were likely to be small at best and perhaps too small for reliable measurement, in the evaluators’ own forward looking assessment in June 2004. The dispute between the camps is therefore not mainly a dispute about what the reports contain. The dispute concerns which question deserves primacy: the narrow question about the covered category, or the wide question about total gun violence. The statute’s own stated aim privileges the narrow question. Public argument usually privileges the wide one. This article reports both, question by question, and identifies which question each finding answers before stating the finding.

One further discipline governs the pages that follow. Every figure traces to the commissioned evaluations, the peer reviewed literature that grew from them, the later syntheses that rated the evidence, or the named studies of specific outcomes, and each figure is attached to the researchers and the period that produced it. No incident is narrated. No finding is extended to any later proposal in any jurisdiction. The vantage closes on July 15, 2021, and the evidence is presented as the record stood at that date. Within those limits, the record is rich enough to reward patience. The sections that follow walk that trail in the order the statute’s design dictates, beginning with the design constraint that shaped everything the evaluators found.

A Ban Built to Work Slowly: The Grandfather Design and the Baseline

The defining feature of the prohibition, for purposes of measuring effects, was what the statute chose not to do. The law barred new manufacture and new transfers of the covered categories on the terms described in the first section, and at the same time it left possession and transfer of already owned covered firearms and feeding devices entirely lawful. Items lawfully possessed on September 13, 1994, the grandfathered stock, could be kept, bought, and sold throughout the decade from 1994 to 2004. The evaluators returned to that provision in report after report, because it set the tempo of any possible effect. A prohibition that stops a flow while leaving a large stock untouched can change the composition of the market only as fast as the stock wears out, gets confiscated, or loses appeal, and none of those processes operates on a legislative timetable. The design guaranteed that effects, if they appeared, would arrive gradually.

How many covered firearms and magazines remained in private hands after enactment?

Christopher S. Koper, in a 2013 synthesis of the evaluations, carried estimates of roughly 1.5 million covered firearms and about 25 million covered magazines already in private hands at enactment. Possession and transfer of those grandfathered items remained lawful throughout the period from September 13, 1994 to September 13, 2004.

Those estimates come from Koper’s chapter America’s Experience with the Federal Assault Weapons Ban, 1994 to 2004: Key Findings and Implications, published in Reducing Gun Violence in America by Johns Hopkins University Press in 2013, in which Koper synthesized the evaluation program he had helped conduct. The scale of the two figures deserves separate attention. Roughly 1.5 million covered firearms is a substantial private stock for a category defined by named models and a feature test, and each unit in that stock remained a lawful article of commerce for the entire life of the statute. About 25 million covered magazines is a stock of a different order, and magazines are durable, interchangeable, and easy to store in quantity. The magazine stock was large enough to do the same for far longer, and, as the final evaluation documented, foreign magazines made before the ban continued to be imported in large numbers after enactment, enlarging the exempt stock rather than allowing it to shrink.

The events of 1994 enlarged the stock further before the prohibition had spent a single day in force. Manufacturers produced covered firearms at an accelerated pace during that year, ahead of the September 13 effective date, and the interim evaluation by Jeffrey A. Roth and Christopher S. Koper, published by the Urban Institute in March 1997 with National Institute of Justice funding, recorded the result: production during 1994 added more than an extra year’s normal supply of grandfathered weapons and lawful modified copies. The surge was a rational response to the statute’s terms. Every covered firearm completed and lawfully possessed before the deadline crossed into the protected class, where it could be held and transferred freely for the next ten years. The evaluators identified a related market consequence in the same body of work: after an initial speculative price rise, primary market prices for the covered firearms fell back toward pre ban levels as the oversupply worked through the market, a sequence examined in the market sections of this article.

The evaluators’ reasoning about timing followed directly from the stock figures. In the interim evaluation of 1997, Roth and Koper warned that the mandated review period was too short and too close to enactment to judge the effects on street violence, given the grandfathered stock in circulation. The warning followed from the design: if the lawful stock numbered in the millions for firearms and in the tens of millions for magazines, and if that stock could circulate freely, then criminal access during the first years of the prohibition would differ little from criminal access before enactment, because criminals drew from the same pool of existing guns as every other secondary market participant. The final evaluation in June 2004, written by Koper with Daniel J. Woods and Jeffrey A. Roth for the National Institute of Justice, restated the point as a structural conclusion: the grandfathering provision guaranteed that effects would occur only gradually. Between the 1997 warning and the 2004 conclusion, the research position never changed, because the underlying stock never stopped being the central fact of the statute’s operation.

Against that design, the baseline share sets the ceiling on what the prohibition could ever have accomplished, even under the most favorable assumptions. The interim evaluation reported that, before enactment, the covered firearms accounted for only 1 to 7 percent of guns used in violent crime across the data sources the evaluators examined. The final evaluation in 2004 stated the ceiling more bluntly: the covered firearms were used in no more than 8 percent of gun crimes even before the prohibition took effect. Handguns, which the statute did not cover, accounted for the large majority of gun crime throughout the period. The covered category, in other words, occupied a small corner of criminal gun use at the moment Congress acted, and the grandfather design then ensured that even that small corner would erode slowly rather than disappear.

The arithmetic consequence is severe, and the evaluators drew it explicitly. Even complete elimination of the covered firearms from crime, a result the grandfather design made unattainable in practice, could have moved total gun violence only within the small share those firearms represented. The interim evaluation stated the corresponding judgment about homicide in substance: at best, the prohibition could have only a limited effect on total gun murders, because the banned weapons and magazines were never involved in more than a modest fraction of all gun murders. That sentence, from Roth and Koper writing in March 1997, is among the most quoted in the entire evaluation literature, and it is quoted accurately by readers who treat the baseline as decisive. The baseline does not decide every question, since lethality per incident and magazine capacity raise separate issues examined later in this article, but the baseline does decide the maximum plausible effect on total gun violence, and the maximum was small.

Substitution tightens the arithmetic further. An offender denied one firearm category does not necessarily abandon an intended offense, because lawful firearms outside the covered definitions remained in production, in stores, and in the same grandfathered secondary market throughout the decade from 1994 to 2004. The evaluators treated substitution as an available pathway from the beginning of the research program, and the final evaluation documented the pathway in operation: while crimes involving the covered firearms fell by between 17 and 72 percent across the six study cities of Anchorage, Baltimore, Boston, Miami, Milwaukee, and St. Louis, measured through 2003, criminal use of non banned semiautomatic firearms fitted with large capacity magazines held steady or rose over the same period. Those non banned semiautomatics were used in crime far more often than the covered models, in Koper’s summary formulation in the June 2004 report. The design constraint explains the pattern without any paradox. The statute constricted one channel, grandfathered the existing stock in that channel, and left adjacent channels fully lawful, so movement between channels was the predictable response of the market and of offenders alike.

Restating the constraint at this point is not repetition for its own sake. The constraint determines how every later finding in this article should be read. A measured decline in criminal use of the covered firearms, against a grandfathered stock of roughly 1.5 million such firearms that remained lawful to possess and transfer, represents movement within a channel that Congress deliberately left partly open. Each outcome becomes legible once the design and the baseline are held in view together, and each outcome looks mysterious or contradictory when either element is dropped. The partisan quotations described in the first section generally drop one element or the other: celebratory readings drop the baseline, and dismissive readings drop the grandfather tempo.

The evaluators also left a forward looking implication embedded in the design analysis, and the final evaluation stated it plainly in June 2004. Because effects were guaranteed to occur only gradually, the ten year window bounded by the sunset of September 13, 2004 might close before the gradual effects had accumulated to a measurable size. Koper, Woods, and Roth wrote that effects were unfolding at the date of the final assessment and that definitive assessment was premature, particularly while foreign pre ban magazines continued to enter the country in large numbers. The same report identified the magazine provision as the element with the clearest potential for larger effects over a longer horizon, if the exempted stock were drawn down over time. The statute ended on its schedule in September 2004, with the exempt firearm stock and the far larger exempt magazine stock both substantially intact in lawful circulation, and with the evaluators’ central structural observation unrefuted by any party to the research record: a ban built around a grandfather clause is a ban built to work slowly, and slow work is what the commissioned evidence, question by question, went on to describe.

The Interim Verdict: The Roth and Koper Evaluation

The interim evaluation occupied a particular place in the history of the statute because it was not an outside commentary added after the fact. Section 110104 of the Violent Crime Control and Law Enforcement Act of 1994 directed the Attorney General to evaluate the effects of the subtitle and to report to Congress within thirty months. The subtitle at issue was Subtitle A of Title XI of Public Law 103-322, the Public Safety and Recreational Firearms Use Protection Act of 1994, in force from September 13, 1994. The commissioned work that answered that direction was conducted at the Urban Institute with funding from the National Institute of Justice. Its principal authors were Jeffrey A. Roth of the Urban Institute and Christopher S. Koper. Their report, titled “Impact Evaluation of the Public Safety and Recreational Firearms Use Protection Act of 1994,” appeared from the Urban Institute in March 1997. A summary followed in the National Institute of Justice Research in Brief titled “Impacts of the 1994 Assault Weapons Ban: 1994-96,” published in March 1999. The order of those dates matters for understanding the verdict. The March 1997 report was an interim evaluation in the strict sense of the term used by its authors and its sponsor. It examined the first years of the prohibition, a period defined in the Research in Brief title as the span from 1994 to 1996, and it did so under a timetable set by Congress before the law had operated for long. The evaluation therefore carried two qualities at once. It was the federal government’s commissioned account of what could be measured close to enactment, and it was, by its own repeated insistence, an incomplete account that refused to turn early movement in the data into a judgment about street violence that the window could not support.

That discipline shaped every part of the interim findings. Roth and Koper reported what had changed in selected indicators, placed those changes beside the size of the category the statute covered, and then tested whether the patterns that would be expected if the prohibition had driven a broader decline in gun murder were present in the early record. In each step the evaluators tied the interpretation back to the design of the law itself. The subtitle barred manufacture, transfer and possession of semiautomatic assault weapons as the statute defined them, a category that combined named models with a two feature test, and it barred transfer and possession of large capacity ammunition feeding devices of more than ten rounds when those devices were made after enactment. Possession and transfer of covered firearms and devices lawfully possessed on September 13, 1994 remained lawful throughout the period. The evaluators estimated that roughly 1.5 million covered firearms and about 25 million covered magazines were already in private hands at enactment and remained lawful to possess and transfer, a figure carried in Koper’s synthesis of the evaluations published in 2013. Production of covered firearms also surged in 1994 before the effective date, adding more than an extra year of normal supply of grandfathered weapons and lawful copies, a fact reported in the interim evaluation itself. Any reading of the interim verdict that omits that stock misstates what the researchers believed they were measuring. They were measuring a prohibition on new manufacture and on post enactment devices laid over a large exempted inventory that a person could lawfully keep, buy and sell during every month the interim study covered.

The first result to reach public attention came from firearm traces. Between 1994 and 1995, law enforcement trace requests for the covered firearms fell by 20 percent, against a fall of 11 percent for all guns, according to the interim evaluation by Roth and Koper published by the Urban Institute in March 1997. A Department of Justice announcement of March 28, 1997 reported the study and carried that comparison into public discussion. The comparison had an obvious attraction as an early indicator. It concerned the very category the statute named, it moved in the direction supporters of the prohibition would predict if criminal use of the covered firearms had begun to recede, and it moved by more than the corresponding figure for guns as a whole over the same interval from 1994 to 1995. A fall confined to the covered category would have meant one thing. A larger fall for the covered category than for all guns suggested, at the level of trace requests, a differential change worth recording. Roth and Koper recorded it as such. What the evaluators did not do was treat the differential as proof that the prohibition had reduced gun murder, reduced gun violence in general, or even reduced the total criminal presence of the covered firearms in a proportion that the trace figures alone could establish. The modesty of the claim attached to the trace result in the interim evaluation is part of the finding, and later quotation that strips away that modesty alters the report.

How reliable were firearm traces as a measure of early change?

Firearm traces provided a narrow and partial signal in the Roth and Koper evaluation because trace requests reflected only guns recovered and submitted by law enforcement agencies, not every gun used in crime, and changes in requests could reflect enforcement practice as well as changes in criminal use itself.

A fuller account of the trace measure explains why the evaluators confined it to that role. A trace request begins when a law enforcement agency recovers a firearm and asks for the firearm to be traced. The resulting series therefore reflects a chain of events at a distance from the underlying act of violence. A gun must come into police hands, must be selected for submission, and must generate a request that enters the recorded system. Movement in requests can arise at more than one point in that chain. Criminal use of a category can fall. Recovery patterns can shift between categories. Submission practice can change in particular places or for particular kinds of firearms. None of those possibilities means the series is worthless. The contrast between the 20 percent fall for covered firearms and the 11 percent fall for all guns between 1994 and 1995 gave Roth and Koper a comparison internal to the measure itself, one that controlled in a rough way for a general fall in requests that would have affected every category. The limit lay in the distance between that differential and the outcomes the statute had been adopted to affect. Trace requests do not count murders, wounds, shots fired, victims in an incident, or the share of gun crime committed with any category. A fall in requests for covered firearms can accompany a fall in criminal use of those firearms without measuring gun violence outcomes. The interim evaluation therefore used traces as an indicator of presence in the recovered gun stream rather than as a substitute for homicide and injury data, and the evaluators turned to those outcome data for the questions that traces could not answer.

The ceiling that governed those outcome questions was stated by Roth and Koper with unusual directness, and it formed the central conclusion of the interim evaluation as summarized in the National Institute of Justice Research in Brief of March 1999. In substance, the evaluators concluded that at best the prohibition could have only a limited effect on total gun murders, because the banned weapons and magazines were never involved in more than a modest fraction of all gun murders. Every word of that formulation does analytical work. The phrase “at best” marks the statement as a ceiling rather than a prediction of achieved results. The phrase “limited effect” measures the possible gain against total gun murders rather than against the smaller universe of murders involving the covered category. The reason given for the ceiling rests on prevalence before and apart from any change the statute might induce. If a category accounts for a modest fraction of all gun murders, then even a large proportional reduction within that category translates into a small proportional change in the total, and even complete removal of that category from murder, an outcome no evaluator proposed as realistic, would leave the large majority of the total untouched. The interim evaluation placed that arithmetic beside the baseline shares it had examined. Before enactment, the covered firearms accounted for only 1 to 7 percent of guns used in violent crime across the data sources the evaluators examined, according to the interim evaluation. The final evaluation later carried a ceiling statement in a related form, reporting that the covered firearms were used in no more than 8 percent of gun crimes even before the prohibition. Handguns, which the subtitle did not cover, accounted for the large majority of gun crime in the baseline the evaluators described. Those shares explain why Roth and Koper framed the question about total gun murders as a question about a modest fraction. They also explain why a finding of reduced trace requests for covered firearms and a finding of limited possible effect on total gun murders could appear in the same report without contradiction. The first finding concerned movement within a small category. The second concerned the weight of that category in the total to which any movement would have to be applied.

Substitution reinforced the ceiling and belonged to the reasoning the evaluators drew from the design of the law. The subtitle defined the covered firearms by named models and by a two feature test, and it left lawful a wide range of other firearms, including non covered semiautomatic firearms. Possession and transfer of the grandfathered stock also remained lawful throughout. A person intent on acquiring a semiautomatic firearm during the period the interim evaluation covered therefore faced a restriction on particular models and features and on newly made large capacity devices, not an absence of lawful semiautomatic alternatives. The evaluators drew the arithmetic consequence explicitly in the body of reasoning summarized on the fact record for the evaluations. Even complete elimination of the covered firearms from crime could have moved total gun violence only within that small share, and substitution to lawful firearms was available. That sentence matters because it distinguishes two kinds of success that public discussion often blends. Success in reducing the presence of the exact covered models in crime is one outcome. Success in reducing total gun murders is a different outcome that depends on whether activity displaced from the covered models disappears or moves to firearms outside the definition. The interim evaluation’s limited effect conclusion addressed the second outcome. It did not deny the first outcome as a possibility. It held that the first outcome, even if achieved in full, had a bounded value for the second because the share was modest and the lawful alternatives stood ready.

The outcome tests that spoke most directly to the purpose attributed to the subtitle returned null results in the interim evaluation. Roth and Koper could detect no reduction in the two categories of gun murder thought most associated with the covered weapons. The first category was murders with multiple victims in a single incident. The second category was murders producing multiple bullet wounds per victim. The selection of those two categories was not incidental. If the distinctive concern about the covered firearms and large capacity magazines concerned the capacity to fire many rounds without reloading and to injure or kill more than one person or inflict more than one wound in a single attack, then an early effect of restricting those firearms and devices, if an effect were visible in the short window, would be expected to appear in those categories before it appeared in the undifferentiated total of gun murders. A general fall in gun murders during a period of broad decline would prove little on its own. A disproportionate fall in multiple victim murders or in multiple wound murders would have aligned with the mechanism attributed to the covered category. The interim evaluation reported that no such reduction could be detected. The finding was a finding about detection in a defined period under stated conditions, not a claim that the researchers had observed an increase and not a claim about a later period the report did not cover. Koper and Roth carried the same result into the journal literature, where the 2001 article in the Journal of Quantitative Criminology reported no evidence that the prohibition had reduced multiple victim gun homicides or multiple gunshot wound victimizations in the early period. Magazine related wound analyses elsewhere in the evaluation record ran in a mixed direction. A Milwaukee analysis found that victims in cases involving large capacity magazines were shot about three times on average, while a multi site analysis found no post enactment reduction in the average number of wounds per case or in the share of cases involving multiple wounds. Roth and Koper’s interim verdict rested on change over time, and on that measure the multiple victim and multiple wound categories showed no detectable reduction during the years the commissioned report examined.

The state comparison supplied a second reason the interim evaluation refused early attribution, and its logic ran against a simple reading of the national decline. If states that entered the federal period with assault weapon bans of their own had experienced the federal law on top of a state regime, while other states experienced the federal law alone, a comparison between those groups could offer a rough check on whether places with bans showed larger homicide declines. Roth and Koper reported that homicide declines were not larger in states that already had bans. Over the study window, homicides fell by 10.3 percent in states with no ban in place when the federal law took effect and by 0.1 percent in states with bans, according to the interim evaluation published in March 1997 and summarized in the National Institute of Justice Research in Brief of March 1999. The direction of that contrast is essential. The group without a preexisting state prohibition recorded the larger decline. The group with a preexisting state prohibition recorded a decline close to zero over the window. A reader who expected the comparison to corroborate an early effect attributable to prohibitions of this kind would find the figures pointing the other way. Roth and Koper treated the comparison as one reason among others to refuse early attribution of the broader homicide movement to the federal subtitle. The caution followed from the structure of the comparison as well as from its direction. States differ in baseline homicide levels, in the composition of gun crime, and in enforcement practice, and a state group comparison over a short window cannot isolate the federal subtitle from every other force moving homicide in the same years. Even on its own terms, however, the comparison failed to show the pattern that early attribution would require. The states with bans did not fall faster. They fell by 0.1 percent against 10.3 percent for the states without bans. That failure removed one piece of corroboration that might have strengthened a claim that the federal law had driven the decline visible in national figures during the interim window, and the evaluators left the claim unmade.

The journal versions of the work fixed the interim findings in the peer reviewed literature and separated the two strands of the evaluation for specialist readers. Christopher S. Koper and Jeffrey A. Roth published the outcomes strand as an article in the Journal of Quantitative Criminology, volume 17, issue 1, in 2001. That article addressed gun violence outcomes and reported no short term effect detected, a formulation consistent with the interim evaluation’s null results on multiple victim murders and multiple wound murders and with its refusal to credit the prohibition with the homicide movement of the first years. Koper and Roth published the markets strand in the same journal, the Journal of Quantitative Criminology, volume 18, issue 3, in 2002. That article addressed gun markets, the arena in which prices, production and availability responded to enactment and to the grandfathered supply. The separation of the two articles clarifies a feature of the interim verdict that compressed summaries often lose. The evaluators reported distinct records for outcomes and for markets, gathered over the same early period but capable of moving in different directions. For the purposes of the interim verdict on gun violence, the governing text remained the outcomes record. That record was the March 1997 Urban Institute report, the March 1999 Research in Brief summary, and the 2001 Journal of Quantitative Criminology article. Each of those publications carried the same essential position. No reduction could be detected in the categories of gun murder most associated with the covered weapons during the period examined, and the possible effect on total gun murders was bounded by the modest fraction of murders in which the banned weapons and magazines had ever been involved.

The evaluators closed the interim account with a warning about time, and that warning was methodological rather than rhetorical. Roth and Koper warned that the mandated review period was too short and too close to enactment to judge effects on street violence, given the grandfathered stock in circulation. The statute had directed a report within thirty months. A report filed in March 1997 could observe the interval from September 1994 through 1996 and little beyond that interval. Street violence effects, if the design of the prohibition produced them at all, would depend on how the stock of covered firearms and magazines in circulation changed as time passed and how often items in that stock entered crime. None of that process could complete within the span covered by a congressionally mandated interim report. The grandfather clause made the timing problem structural. The roughly 1.5 million covered firearms and about 25 million covered magazines already in private hands at enactment remained lawful to possess and transfer, and the production surge of 1994 added more than an extra year of normal supply of grandfathered weapons and lawful copies before the effective date arrived. An evaluator assigned to report in March 1997 was therefore evaluating a restriction whose exempted supply stood near its largest point. Under those conditions, a failure to detect an effect on street violence admitted two readings that the data of the period could not separate. The restriction might have no effect of the kind sought, or it might have an effect that required the exempted stock to recede over a longer span than the mandated window allowed. Roth and Koper declined to choose between those readings on the evidence available to the interim evaluation. The interim verdict was therefore neither an endorsement nor a dismissal. It was a bounded measurement taken at the point Congress had specified, stating a ceiling on what the statute could do to total gun murders, recording no detectable gain in the two murder categories most closely tied to the covered weapons, and reserving the larger question for a period the report could not reach.

The Final Evaluation and the Crime Decline It Would Not Claim

The final commissioned evaluation of the federal assault weapons ban arrived in June 2004, three months before the statute reached the end date written into its own sunset provision. Christopher S. Koper, with Daniel J. Woods and Jeffrey A. Roth, prepared “An Updated Assessment of the Federal Assault Weapons Ban: Impacts on Gun Markets and Gun Violence, 1994 to 2003” at the Jerry Lee Center of Criminology at the University of Pennsylvania. The report was submitted to the National Institute of Justice, the body that had funded the commissioned evaluation program. Its period ran from 1994 to 2003, which meant that it covered nearly the whole operative life of the prohibition enacted as Subtitle A of Title XI of the Violent Crime Control and Law Enforcement Act of 1994, Public Law 103-322. The statute had taken effect on September 13, 1994, and its terms barred manufacture, transfer, and possession of semiautomatic assault weapons as the statute defined them, together with transfer and possession of large capacity ammunition feeding devices holding more than ten rounds when those devices were made after enactment. The final evaluation therefore had a longer record to examine than the interim evaluation completed during the first years of the prohibition, and that longer record is one reason its findings resist compression into a single sentence. Crimes involving the covered firearms fell. Substitution toward lawful firearms fitted with large capacity magazines complicated that fall. Measures of lethality and injuriousness did not move in the manner a stronger effect would have produced. Attribution for the national decline in gun violence was refused. The forecast for a renewed prohibition was restrained. The same report then qualified its restraint with a counterweight about time, grandfathered stock, and the magazine provision. Each of those findings has to be held apart from the others before any of them can be understood, because the report’s meaning lies in the pattern formed by the parts rather than in any part quoted alone.

The design constraint that governed every one of those questions has to be restated before the findings themselves, because the researchers returned to it repeatedly and because no effect estimate in the report makes sense without it. The prohibition did not remove the covered firearms or the covered magazines from private hands. Items lawfully possessed on September 13, 1994, remained lawful to possess and to transfer throughout the life of the statute, and the evaluations estimated that roughly 1.5 million covered firearms and about 25 million covered magazines were already in private hands at enactment under that grandfathering provision. The figure for the exempted stock came from estimates carried in Koper’s synthesis of the evaluations published in 2013, and its scale explains why the final report treated time as a variable rather than as a neutral backdrop. Production of covered firearms had also surged during 1994 before the effective date, adding more than an extra year of normal supply in grandfathered weapons and lawful copies, according to the interim evaluation. The baseline share of crime attributable to the covered firearms imposed a second constraint that was arithmetic rather than legal. Before enactment, the covered firearms accounted for only 1 to 7 percent of guns used in violent crime across the data sources the evaluators examined, and the final report stated the ceiling in stronger form by reporting that the covered firearms were used in no more than 8 percent of gun crimes even before the prohibition took effect. Handguns, which the statute did not cover, accounted for the large majority of gun crime. The evaluators drew the consequence directly from those figures. Even complete elimination of the covered firearms from crime could have moved total gun violence only within that small share, and substitution to lawful firearms remained available to anyone displaced from a covered model. The final evaluation’s findings were produced inside those two limits, the grandfathered stock on the supply side and the small baseline share on the demand side, and Koper, Woods, and Roth framed their conclusions with both limits in view.

The first question the final evaluation answered concerned the covered firearms themselves, and on that question the measured movement was downward. Gun crimes involving assault weapons declined by between 17 and 72 percent across the six study cities examined in the report, measured through 2003. The six cities were Anchorage, Baltimore, Boston, Miami, Milwaukee, and St. Louis. The range itself carries information that a single national figure would obscure. The decline was not uniform across places, and the distance between the lowest reduction of 17 percent and the highest reduction of 72 percent indicated that local conditions, local enforcement practice, local market structure, and local patterns of criminal preference all shaped what the federal prohibition coincided with in each city. Koper, Woods, and Roth presented the finding as a finding about criminal use of the banned models, and stated in those terms it was favorable to the statute. The covered firearms became less prominent in the gun crimes recorded in the study cities during the period from 1994 to 2003. Selective quotation usually begins at precisely this point in the document, because the city figures are the report’s clearest favorable numbers and they can be detached from the findings that followed them on the next questions. Detached, the figures appear to describe the whole effect of the statute. Attached to the rest of the report, they describe one measured change inside a larger record that the researchers characterized with considerably more caution.

The second question concerned substitution, and the answer limited what the first answer could support. Use of non-banned semiautomatic firearms fitted with large capacity magazines stayed steady or rose during the period the final evaluation examined, and those firearms were used in crime far more often than the covered models. That pairing of findings matters because the statute drew its line by model definition and by features, while criminal demand for semiautomatic firepower was not confined to the models on the prohibited side of the line. Koper summarized the combined record on criminal use of the banned guns and magazines by stating that success on that measure was mixed, a formulation that the 2004 report applied to the whole pattern rather than to either half of it. The mixed characterization holds both halves in place at once. Criminal use of the specifically covered firearms fell across the six study cities. Criminal use of the broader class of semiautomatic firearms fitted with large capacity magazines did not fall in parallel, because use of the non-banned members of that class held steady or increased. The grandfather design constraint runs through this finding as well, since the magazines that gave the substitute firearms their capacity circulated lawfully in large numbers under the exemption for items possessed at enactment, and foreign magazines manufactured before the ban continued to enter the country in large numbers after enactment. A decline inside the named category paired with stability or growth outside it is not a contradiction in the report. It is the report’s description of a market and a pattern of criminal use adjusting around a statutory boundary, and it is the reason the researchers declined to treat the city declines as a measure of the statute’s total effect.

The third question concerned lethality, and on that question the final evaluation detected no movement of the kind the statute’s stated aim would have predicted. Koper, Woods, and Roth reported that there had been no discernible reduction in the lethality and injuriousness of gun violence during the period from 1994 to 2003. The researchers measured lethality and injuriousness through two indicators. The first was the share of gun crimes resulting in death. The second was the share of gunfire incidents resulting in injury. Neither indicator fell in a manner the researchers could discern. The distinction between this finding and the city findings on the covered firearms is fundamental to the report’s logic. A statute can coincide with reduced criminal use of the particular firearms it names while failing, within the same period, to coincide with any measured change in how often gun crimes kill or how often gunfire wounds. The absence of discernible movement on death shares and injury shares therefore constrained the strongest claim the city figures could carry. Whatever changed in the mix of firearms recovered or traced in Anchorage, Baltimore, Boston, Miami, Milwaukee, and St. Louis, the aggregate character of gun violence in terms of death and injury did not change detectably alongside it during the years the final evaluation covered. The report stated that result without converting it into a claim that no effect of any size existed anywhere in the system. It stated that the measured indicators showed no discernible reduction, which is a statement about what the evidence of the period could show, framed by researchers who elsewhere in the same document warned that the exempted stock made gradual effects the only kind the statute’s design could produce.

Could the final evaluation separate the effect of the ban from the wider fall in crime?

No. Christopher S. Koper, Daniel J. Woods, and Jeffrey A. Roth reported in June 2004 that crimes involving the covered firearms had fallen in the six study cities, but the researchers could not clearly credit the statute with the broader national drop in gun violence during the same period.

The attribution question was the fourth question the final evaluation faced, and it was the question on which the report’s language grew most careful, because the background against which the statute operated was a broad national decline in crime during the period from 1994 to 2003. Gun violence fell at the national level during those years. Homicide and other violent crime categories fell as well, and a decline of that breadth presents an identification problem for any evaluation of a single federal statute aimed at a narrow category of firearms. When the outcome moves in the desired direction across the whole country at the same time as many other conditions change, assigning a share of the movement to one law requires a basis for separating that law’s contribution from the contributions of everything else moving alongside it. Koper, Woods, and Roth reported that they could not clearly credit the ban with any of the nation’s drop in gun violence. The phrasing of that refusal deserves attention, because it is neither an affirmation nor a denial of an effect. The researchers did not report that the prohibition had contributed nothing. They reported that the evidence available to them did not permit a clear crediting of the statute with any portion of the national decline. The grandfather design constraint bears directly on that refusal. A statute that left roughly 1.5 million covered firearms and about 25 million covered magazines in lawful circulation, with lawful transfer preserved, was not expected by its own evaluators to produce an abrupt national signal large enough to stand apart from a general crime decline, and the baseline share of the covered firearms in gun crime, no more than 8 percent even before enactment on the final report’s ceiling statement, capped the size of any signal the statute could plausibly have generated. The refusal to credit the statute with the national decline is therefore best read as the report’s central act of discipline. It separated a favorable local finding about a named category of firearms from an unsupported national claim about total gun violence, and it declined to let the first finding borrow authority from a trend the researchers could not assign. The difficulty of separating one statute from a broad simultaneous trend is part of the wider crime decline attribution problem.

The fifth question the final evaluation answered looked forward rather than backward, and the researchers answered it in the conditional form the question required. The statute was scheduled to end on September 13, 2004, by its own sunset, and the report appeared in June 2004, when renewal was a live legislative question in the plain chronological sense that Congress faced the date with the report in hand. Koper, Woods, and Roth stated that, should the ban be renewed, its effects on gun violence were likely to be small at best and perhaps too small for reliable measurement. The fragment “small at best” captures the ceiling the researchers placed on the expected effect, while the second half of the sentence placed a floor under detectability rather than under the effect itself. An effect too small for reliable measurement is not necessarily an effect of zero. It is an effect that the instruments available to evaluators, applied to crime statistics shaped by many forces at once, could not reliably distinguish from background variation. The renewal forecast rested on the same two constraints that structured the backward looking findings. The covered firearms had accounted for a small share of gun crime before enactment, between 1 and 7 percent across the data sources the evaluators examined, so even a complete and permanent removal of those firearms from criminal use addressed only that share before substitution was considered. Substitution had been observed rather than merely hypothesized, in the steady or rising criminal use of non-banned semiautomatic firearms fitted with large capacity magazines. A renewed prohibition built on the same definitions and the same grandfathering structure would face the same arithmetic, and the researchers’ forecast followed from the arithmetic they had spent two evaluations documenting.

The sixth element of the final evaluation was a counterweight that the same report placed against its own restrained conclusions, and any account that omits the counterweight misrepresents the document as surely as an account that omits the restraint. Koper, Woods, and Roth emphasized that the grandfathering provision guaranteed that the effects of the prohibition would occur only gradually. That guarantee followed from the design itself. With an exempted stock estimated at roughly 1.5 million covered firearms and about 25 million covered magazines lawfully possessed at enactment and lawful to transfer afterward, the supply conditions the statute sought to change could change only as that stock slowly left circulation. The researchers stated that the effects of the prohibition were unfolding during the period they examined and that a definitive assessment was premature, particularly while foreign magazines manufactured before the ban continued to be imported in large numbers. Imports of that kind enlarged the exempted stock rather than allowing it to shrink, which meant that one of the mechanisms through which a gradual effect would have accumulated, the drawing down of grandfathered magazines, was being offset even as the evaluation proceeded. Within that counterweight, the report identified the magazine provision as the element of the statute with the clearest potential for larger effects over a longer horizon, if the exempted stock were drawn down. The reasoning assigned the magazine provision that position because magazine capacity travels across firearm models. A restriction that eventually reduced the availability of large capacity magazines would bear on the non-banned semiautomatic firearms that offenders had substituted toward, as well as on the covered models, whereas the model based prohibition bore mainly on the named category whose criminal use had already fallen from a small baseline. The counterweight therefore did not contradict the renewal forecast so much as it specified the conditions under which the forecast could change. Effects likely to be small at best under the structure and time horizon the researchers examined could become larger over a longer horizon if the exempted magazine stock declined, and the researchers said plainly that the ten years the statute was given did not test that longer horizon. The prohibition ended on September 13, 2004, before the gradual process its evaluators described had run the course their analysis contemplated.

The report paired its counterweight with an observation about the mechanics of gun attacks that cut in the opposite direction, and the pairing illustrates why the document sustains two selective readings at once. Koper, Woods, and Roth observed that relatively few gun attacks involve more than ten shots fired. They also observed that how often attack outcomes depend on firing more than ten shots without reloading was unclear. The ten round threshold was the statute’s own line for large capacity ammunition feeding devices, so the observation addressed the precise point at which the magazine provision would change the conduct of an attack. If most attacks end, for whatever reason, before a shooter reaches the point at which a larger magazine would matter, then a capacity limit changes the mechanical possibilities of only a minority of attacks, and within that minority the researchers could not establish how often the outcome, in deaths or injuries, turned on the ability to continue firing past ten rounds without reloading. That uncertainty connects directly to the lethality finding. The two indicators the report used, the share of gun crimes resulting in death and the share of gunfire incidents resulting in injury, showed no discernible reduction, and the attack mechanics observation offered one candidate explanation grounded in the distribution of shots fired. Most attacks did not reach the threshold at which the provision operated. For the attacks that did reach it, the evidence did not establish how much difference the threshold made. The researchers left both parts of that observation in place without resolving them into a verdict on the magazine provision, which they had elsewhere identified as the element with the clearest long term potential. Readers who quote only the ten shots observation present a report more dismissive than the one Koper, Woods, and Roth wrote. Readers who quote only the magazine potential present a report more promising than the one they wrote. The document contains both passages because the evidence, as the researchers assessed it in June 2004, supported both statements at once.

Read question by question, the final evaluation of June 2004 forms a single coherent account whose coherence depends on keeping the questions apart. Koper, Woods, and Roth found that gun crimes involving the covered firearms declined by between 17 and 72 percent across Anchorage, Baltimore, Boston, Miami, Milwaukee, and St. Louis through 2003. They forecast that a renewed prohibition would produce effects on gun violence likely to be small at best and perhaps too small for reliable measurement. They insisted, in the same report, that grandfathering guaranteed a gradual pace of effect, that effects were unfolding, that definitive assessment was premature while foreign magazines manufactured before the ban kept entering the country, that the magazine provision held the clearest potential for larger effects over a longer horizon, and that the frequency with which attack outcomes turn on firing more than ten shots without reloading remained unclear. The grandfather design constraint, restated once more because the report never evaluates an effect without it, is the thread connecting those findings. A prohibition that preserved lawful possession and transfer of roughly 1.5 million covered firearms and about 25 million covered magazines, applied to firearms that had accounted for no more than 8 percent of gun crimes before enactment, produced the record the researchers described. A measured decline in use of the named firearms, substitution around the statutory boundary, unchanged lethality indicators, refused attribution, a restrained forecast, and an open long term question about magazines are not rival summaries of the 2004 report. Together, in that order and with those qualifications, they are the report.

Markets First: Speculation, Oversupply and Prices

The market record occupied a central place in the government evaluations because the statute had been written around a grandfathered stock, and prices registered the consequences of that design before any measure of criminal use could change in a durable way. Section 110104 of the Violent Crime Control and Law Enforcement Act of 1994 directed the Attorney General to evaluate the effects of Subtitle A of Title XI and to report to Congress within thirty months, and the commissioned work that followed treated primary market prices, production volumes and the supply of lawful substitutes as evidence in their own right. Jeffrey A. Roth of the Urban Institute and Christopher S. Koper conducted the interim evaluation, published by the Urban Institute in March 1997 with funding from the National Institute of Justice under the title Impact Evaluation of the Public Safety and Recreational Firearms Use Protection Act of 1994, and Koper and Roth returned to the market questions in the Journal of Quantitative Criminology in 2002. Christopher S. Koper, with Daniel J. Woods and Jeffrey A. Roth, then extended the account in An Updated Assessment of the Federal Assault Weapons Ban: Impacts on Gun Markets and Gun Violence, 1994 to 2003, issued by the Jerry Lee Center of Criminology at the University of Pennsylvania in June 2004 as a report to the National Institute of Justice. Across those publications, the sequence remained consistent, and the sequence mattered more than any single price observation within it. A speculative rise came first, a production surge followed during 1994, a reversal carried prices back toward earlier levels after enactment, and a large exempt stock, enlarged for magazines by continued imports, explained why the reversal persisted through the period the interim evaluation could observe.

The design constraint framed every stage of that sequence. The prohibition, in force from September 13, 1994 to September 13, 2004, barred manufacture, transfer and possession of semiautomatic assault weapons as the statute defined them and barred transfer and possession of large capacity ammunition feeding devices made after enactment, while possession and transfer of covered firearms and devices lawfully possessed on September 13, 1994 remained lawful throughout. Koper, in the 2013 synthesis titled America’s Experience with the Federal Assault Weapons Ban, 1994 to 2004: Key Findings and Implications, published in Reducing Gun Violence in America by Johns Hopkins University Press, carried the evaluations’ estimates that roughly 1.5 million covered firearms and about 25 million covered magazines were already in private hands at enactment and remained lawful to possess and transfer. A market in which an exempt stock of that scale stayed in lawful circulation differed in kind from a market in which the covered items disappeared at the effective date, and the evaluators returned to that difference whenever prices, production or availability entered the analysis. The grandfathered stock set the ceiling on scarcity, the lawful copies and substitutes set the terms of competition against that stock, and the importation of foreign pre-ban magazines kept adding to the exempt supply of feeding devices after the domestic manufacture of new covered devices had stopped under the law.

The speculative run-up preceded enactment and belonged to the period when the terms of grandfathering were visible but the effective date had not arrived. Koper and Roth, in the market analyses carried in the interim evaluation of March 1997 and in the Journal of Quantitative Criminology in 2002, reported that primary market prices of the covered firearms rose by upwards of 50 percent before enactment. The evaluators attributed that rise to speculation by distributors, dealers and collectors who expected grandfathered weapons to become valuable collector items. The expectation rested on the structure of the subtitle itself. If manufacture of the covered models for the civilian primary market ended while examples lawfully possessed on the effective date retained lawful status, a buyer who treated those examples as a fixed and shrinking class of objects could reason that scarcity would support a premium, and distributors, dealers and collectors acted on that reasoning in advance of the date itself. The 50 percent figure therefore measured anticipation as much as it measured any change in the underlying supply during the run-up period, because the production surge of 1994 and the stock already in private hands both stood against the scarcity premise even as the speculation proceeded. Koper and Roth presented the rise as a pre-enactment phenomenon documented in primary market prices, and the later reversal gained much of its evidentiary weight from the contrast with that elevated level.

The identity of the speculators, as the evaluators described the market, also mattered to the interpretation. Distributors and dealers occupied positions in the primary market through which new and newly transferred stock moved, and collectors formed a class of buyers prepared to pay for anticipated rarity rather than for immediate use alone. A price increase driven by those participants signaled a bet on future value attached to grandfathered status, and the bet depended on grandfathered examples remaining distinctly more desirable than the lawful alternatives that manufacturers could offer after the effective date. The interim evaluation and the 2002 journal version treated that bet as the opening fact of the market record, and neither publication treated the higher prices of the run-up as a settled condition of the prohibition years. The rise was temporary by the evaluators’ own account, bounded by enactment on one side and by the production decisions of 1994 on the other, and the account of prices after enactment in the same publications reversed the direction entirely.

Manufacturers answered the approaching effective date with volume. The interim evaluation reported that production surged during 1994 before the effective date, adding more than an extra year’s normal supply of grandfathered weapons and lawful modified copies. The scale of that addition deserves emphasis in the evaluators’ terms, because a quantity described as more than an extra year’s normal supply entered a market that already held the grandfathered stock estimated in Koper’s 2013 synthesis at roughly 1.5 million covered firearms. The surge had two components, and the evaluations kept both in view. Grandfathered weapons, lawfully possessed on September 13, 1994, acquired their exempt status by that possession date, and production completed before the date enlarged the pool that could claim it. Lawful modified copies, altered to fall outside the statute’s definitions while remaining available for lawful manufacture and sale, enlarged a different pool, the pool of substitutes that would compete with grandfathered examples after enactment. Together, the two components undercut the scarcity premise from both directions. The exempt originals became more numerous than a normal year of production would have left them, and the modified copies ensured that a buyer turned away from a grandfathered example by price had a lawful alternative produced by the same industry during the same year.

Why did prices fall after the initial surge passed?

Prices fell because the market held more grandfathered firearms than buyers required at elevated levels, and lawful substitutes remained available in variety. Koper and Roth found in the interim evaluation that primary market prices for the covered firearms and most large capacity magazines returned to nearly pre-ban levels and remained there through mid-1996.

The reversal after enactment formed the center of the interim market account. Koper and Roth reported that primary market prices of the banned guns and most large capacity magazines fell to nearly pre-ban levels after enactment and remained at those levels at least through mid-1996. The wording carried two qualifications that the evaluators preserved and that any faithful summary has to preserve with equal care. The fall applied to most large capacity magazines, in the publications’ phrasing, rather than to every magazine category without exception, and the period of observation for the sustained lower level ran at least through mid-1996, the window the interim evaluation, published in March 1997, could document from the market record available to Roth and Koper at the Urban Institute. Within those qualifications, the direction was unambiguous. Prices that had risen by upwards of 50 percent in speculation before enactment gave back that elevation after the law took effect, and the primary market settled near the levels that had prevailed before the speculative episode began. The Journal of Quantitative Criminology version of the market analysis, published by Koper and Roth in 2002, carried the same sequence forward as the considered journal statement of the interim findings, so the reversal rested on both the commissioned report of 1997 and the peer reviewed restatement of 2002 rather than on a single unpublished observation.

The mechanisms the evaluators gave for the reversal were supply mechanisms, stated plainly and tied directly to the design of the subtitle. The first mechanism was oversupply of grandfathered guns. Production during 1994 had added more than an extra year’s normal supply before the effective date, and that addition joined the exempt stock already in private hands, so the primary market after enactment faced a quantity of lawful grandfathered firearms large enough to defeat the premium that speculation had attached to them. A collector premium depends on buyers competing for a limited pool, and the evaluators described a pool enlarged at exactly the moment when the premium had been bid highest. The second mechanism was the availability of lawful substitutes. Manufacturers had produced lawful modified copies during 1994, and the variety of lawful substitutes available after enactment gave buyers alternatives to the covered models, so a seller of a grandfathered example competed not only against other grandfathered examples but against modified firearms that remained lawful to manufacture and sell. Koper and Roth connected the price path to those two conditions together, oversupply in the exempt class and variety in the substitute class, and the connection explained both the speed of the fall and its persistence through mid-1996. Either condition alone might have softened prices, but the evaluations presented a market pressed from both sides, with abundant exempt originals above and lawful alternatives beside them.

From those mechanisms the evaluators drew an unintended consequence that reversed the expected direction of a prohibition on manufacture. Koper and Roth identified that, by 1996, the statute had made the covered firearms more accessible in the primary market than before. Accessibility in that finding followed from the price record and the supply record combined. Prices for the banned guns had returned to nearly pre-ban levels, the grandfathered stock stood enlarged by the production surge of 1994, and lawful modified copies broadened the range of firearms a primary market buyer could choose among, so the practical barrier that elevated prices might have placed between a buyer and a covered firearm had dissolved within the first portion of the prohibition period. The finding was bounded by its date, and the evaluators located it by 1996 rather than across the full decade from 1994 to 2004, but within the interim window the judgment was direct. The law, through the interaction of grandfathering, anticipatory production and lawful substitution, had produced a primary market in which the covered firearms were more accessible than they had been before enactment, even though manufacture of new covered firearms for that market had ceased under the subtitle. The final evaluation of June 2004, prepared by Koper with Woods and Roth for the National Institute of Justice, operated against that interim baseline, and the market chapters of the later report extended the supply account rather than overturning the price sequence the interim work had established.

The magazine side of the market carried the same grandfathering logic further, because the exempt stock of large capacity magazines behaved differently from the exempt stock of firearms after enactment. The final evaluation of 2004 reported that foreign pre-ban magazines continued to be imported in large numbers after enactment, enlarging the exempt stock. Koper’s 2013 synthesis carried the estimate that about 25 million covered magazines were already in private hands at enactment, and the final evaluation added that imports of foreign magazines made before the ban kept adding to that total during the prohibition years instead of allowing the stock to draw down through loss, wear or confiscation. The distinction between the two categories of covered items mattered for the timetable the evaluators described. A grandfathered firearm stock enlarged once, by the production surge of 1994, could at least be described as fixed after the effective date passed, while a grandfathered magazine stock replenished by large scale imports lacked even that fixed character. Koper, Woods and Roth, in the final evaluation, connected that replenishment to the pace of any effects, observing in the same report that the grandfathering provision guaranteed that effects would occur only gradually and that definitive assessment remained premature particularly while foreign pre-ban magazines continued to arrive in large numbers. The magazine provision, in the final evaluation’s assessment, held the clearest potential for larger effects over a longer horizon if the exempted stock were drawn down, and the import record explained why that condition had not been met during the years the report covered, from 1994 to 2003.

A late signal from the magazine market arrived after the prohibition had ended, and the evaluations’ successors presented that signal at a deliberately hedged strength that a summary has to match. A Washington Post analysis of guns recovered in Virginia, reported in January 2013, found that the share of recovered guns fitted with large capacity magazines was generally high through the ban years and lower by 2011. Koper commented on that pattern, and the comment, as reported with the analysis, held that the pattern suggested the prohibition might have been starting to make a dent in the magazine market by the time the prohibition ended. Every element of that sentence carries an attribution or a hedge, and each one is necessary. The underlying observation belonged to the Washington Post analysis of Virginia recoveries rather than to the commissioned evaluations of 1997 or 2004, the geographic base was a single state recovery series rather than a national market measure, the characterization of the ban years was a generally high share rather than a precise figure, and Koper’s interpretation used the language of suggestion and possibility, a dent that might have been starting as the law reached its sunset on September 13, 2004. Placed beside the interim and final evaluations, the Virginia pattern pointed in a different direction from the early price record, toward gradual tightening in the magazine supply over a long horizon of the kind the final evaluation had described as the condition for larger effects, but the analysis and the comment together established a suggestion about timing rather than a measured national effect, and the report of January 2013 left the suggestion at exactly that strength.

The systematic review literature closed the market account by rating the strength of the evidence behind the price and sales findings rather than by restating the findings alone. The 2020 edition of The Science of Gun Policy, the second edition of the RAND Corporation synthesis titled The Science of Gun Policy: A Critical Synthesis of Research Evidence on the Effects of Gun Policies in the United States, found limited evidence that bans of this kind increase prices or sales of banned firearms in the short term. The rating fit the record assembled by Koper and Roth in a specific way. Limited evidence, on the scale used in the 2020 edition of The Science of Gun Policy, stood above an inconclusive rating and below the moderate and supportive ratings, and the outcome rated was the effect of bans on the sale of assault weapons and high capacity magazines on prices and sales in the short term, the outcome category the RAND synthesis labeled as gun industry outcomes. The pre-enactment rise of upwards of 50 percent in primary market prices, documented in the interim evaluation of 1997 and the Journal of Quantitative Criminology in 2002, supplied the principal price movement the rating summarized, and the production surge of 1994 supplied the sales and supply movement beside it, while the reversal to nearly pre-ban levels through mid-1996 and the greater accessibility the evaluators identified by 1996 bounded how far that short term increase extended. The 2020 edition of The Science of Gun Policy therefore left the market record where the commissioned evaluations had placed it, as a short term speculative and production response followed by an oversupplied primary market, supported by evidence the RAND reviewers graded as limited for the increase itself.

The Magazine Question: Capacity, Casualties and the Stock That Stayed

The magazine question occupied a distinct place inside the evaluations of the federal assault weapons ban. The statute, Subtitle A of Title XI of the Violent Crime Control and Law Enforcement Act of 1994, Public Law 103-322, in force from September 13, 1994 to September 13, 2004, barred transfer and possession of large capacity ammunition feeding devices, meaning devices holding more than ten rounds, made after enactment. The research literature called that category large capacity magazines. A firearm could fall outside the defined category of semiautomatic assault weapons and yet be fitted with such a magazine, so the magazine provision reached a wider population of objects than the named models and the two-feature test reached. Possession and transfer of devices lawfully possessed on September 13, 1994 remained lawful throughout. The evaluations estimated that about 25 million covered magazines were already in private hands at enactment, alongside roughly 1.5 million covered firearms, and none of that stock was removed from lawful circulation by the subtitle.

That grandfather design governed every magazine finding. A prohibition on newly made devices, applied while tens of millions of exempt devices remained lawful to possess and transfer, operated on the margin of supply rather than on the stock itself. The final evaluation, Christopher S. Koper with Daniel J. Woods and Jeffrey A. Roth, An Updated Assessment of the Federal Assault Weapons Ban: Impacts on Gun Markets and Gun Violence, 1994 to 2003, prepared at the Jerry Lee Center of Criminology at the University of Pennsylvania for the National Institute of Justice and issued in June 2004, drew the corresponding inference. The grandfathering provision guaranteed that effects would occur only gradually. Definitive assessment was premature, particularly while foreign pre-ban magazines continued to be imported in large numbers, enlarging the exempt stock rather than allowing the stock to contract.

The mechanism researchers examined began with a physical relationship. Magazine capacity governs how many rounds can be fired before reloading. A device holding more than ten rounds permits a longer uninterrupted sequence of shots than a device holding ten rounds or fewer. The hypothesis built on that relationship concerned a connected chain of outcomes within an attack. Capacity, on the hypothesis, affects the number of shots fired during an incident, the number of victims struck during an incident, and the number of wounds inflicted per incident. Each link in that chain is separate, and the evaluations treated the links as separate questions. Shots fired and persons struck are different measurements, and wounds per case, wounds per victim and cases involving multiple wounds describe different distributions of injury, so no one measure served as a substitute for another in the literature.

The scope condition attached to the hypothesis was equally specific. The hypothesized effect concerned the small subset of attacks involving many rounds. The final evaluation stated the limiting fact directly, in the June 2004 report by Koper, Woods and Roth for the National Institute of Justice. Relatively few gun attacks involve more than ten shots fired. The same report observed that how often attack outcomes depend on firing more than ten shots without reloading is not clear. The ten round threshold in the statute corresponded to the point at which reloading would become necessary under a compliant device, yet the frequency with which attacks crossed that threshold, and the frequency with which crossing that threshold changed an outcome, were both uncertain quantities. A capacity effect, if present, would concentrate in a minority of incidents, and an average across all gun attacks would dilute such an effect among incidents in which capacity beyond ten rounds played no part.

Do large capacity magazines increase the number of wounds in a shooting?

The Milwaukee analysis in the interim evaluation found that victims in cases involving large capacity magazines were shot about three times on average, while the multi-site analysis found no post-ban reduction in wounds per case or in cases with multiple wounds, a mixed record that later syntheses preserved.

The interim evaluation requires introduction before its wound findings can be weighed. Jeffrey A. Roth of the Urban Institute and Christopher S. Koper prepared Impact Evaluation of the Public Safety and Recreational Firearms Use Protection Act of 1994 for the Urban Institute in March 1997, with funding from the National Institute of Justice, under the mandate in section 110104 directing the Attorney General to evaluate the subtitle and report to Congress within 30 months. The National Institute of Justice summarized that work in the Research in Brief titled Impacts of the 1994 Assault Weapons Ban: 1994 to 96, issued in March 1999. The authors warned that the mandated review period was too short and too close to enactment to judge effects on street violence, given the grandfathered stock in circulation. For magazines, the exempt stock of about 25 million devices had had only a brief interval in which any drawdown could occur, and imports were adding to that stock during the same interval.

The Milwaukee analysis reported a level of injury associated with magazine involvement. Victims in cases involving large capacity magazines were shot about three times on average in that city analysis. Read on its own terms, the figure describes the cases in which large capacity magazines appeared, and the figure speaks to the association between the presence of a high capacity device and the number of times a victim was shot. A victim shot about three times on average represents a heavier wound burden than a victim shot once, and the figure gave the capacity hypothesis a concrete early observation consistent with the mechanism. The figure was a city analysis rather than a national estimate, however, and the figure did not by itself measure a change produced by the statute. A change measurement requires a comparison across periods and an account of the exempt stock circulating in both periods, and about 25 million covered magazines remained lawful throughout the period the interim evaluation measured.

The multi-site analysis in the same interim evaluation pointed in a different direction on the change question. Across the sites examined, the analysis found no post-ban reduction in the average number of wounds per case, and no post-ban reduction in the share of cases involving multiple wounds. Those measures correspond closely to the final links of the hypothesized chain. Average wounds per case addresses how much injury an incident produced, and the share of cases involving multiple wounds addresses how often injury crossed from a single wound into repeated wounding within a case. If capacity constraints had begun to reduce casualties per incident during the early ban years, movement in one or both measures would have been a plausible place for change to appear, subject to the limits of the short window and the undiminished exempt stock. No such movement appeared. One city analysis had documented heavier wounding in cases involving large capacity magazines, and a broader analysis documented no reduction after enactment, and the two findings address different questions, one about the profile of magazine-involved cases and one about change over time under the statute.

The journal literature tested the early period against two further casualty categories. Christopher S. Koper and Jeffrey A. Roth, in the Journal of Quantitative Criminology, volume 17, issue 1, in 2001, found no evidence that the ban reduced multiple-victim gun homicides in the early period, and no evidence that the ban reduced multiple-gunshot-wound victimizations in the early period. Multiple-victim gun homicides capture incidents producing more than one homicide victim, the pattern in which a longer uninterrupted sequence of fire would be expected to matter most under the hypothesis. Multiple-gunshot-wound victimizations capture victims who sustained more than one gunshot wound, the pattern closest to the wounds per victim measure in the Milwaukee analysis. A reduction in either category would have supported an early casualty effect, and Koper and Roth reported no evidence of a reduction in either category. The interim evaluation had drawn a related conclusion for gun murder as a whole, stating in substance that at best the prohibition could have only a limited effect on total gun murders, because the banned weapons and magazines were never involved in more than a modest fraction of all gun murders.

The final evaluation returned to the casualty question with a longer window, through 2003. Koper, Woods and Roth reported in June 2004 that there had been no discernible reduction in the lethality and injuriousness of gun violence from 1994 to 2003, measured through the share of gun crimes resulting in death and the share of gunfire incidents resulting in injury. That conclusion operates at the level of gun violence overall rather than magazine-involved cases alone. An overall measure aggregates incidents involving covered firearms, incidents involving other firearms fitted with large capacity magazines, and the large majority involving handguns and other firearms outside the subtitle. The evaluators had established that covered firearms accounted for only 1 to 7 percent of guns used in violent crime across the sources examined before enactment, and for no more than 8 percent of gun crimes even before the prohibition, so a genuine change inside a small category could disappear inside the aggregate. The final evaluation nevertheless recorded no discernible reduction in that aggregate across the ban decade.

Attention then shifts from wounds to prevalence, because the casualty tests presuppose an exposure that prevalence series could check. If the share of crime guns fitted with large capacity magazines had fallen during the ban years, a casualty effect would have a clear pathway through reduced presence of the devices in crime. Koper assembled the prevalence evidence in America’s Experience with the Federal Assault Weapons Ban, 1994 to 2004: Key Findings and Implications, a chapter in Reducing Gun Violence in America, published by Johns Hopkins University Press in 2013. That synthesis of the commissioned evaluations could array city series across the ban decade through 2003, rather than resting on the early window that constrained the interim evaluation of 1997 and the journal article of 2001. The grandfather condition applies here with the same force, since about 25 million exempt magazines remained lawful to possess and transfer throughout the decade.

The Baltimore series in Koper’s 2013 synthesis showed no decline. About 14.0 percent of recovered crime guns in Baltimore were fitted with large capacity magazines in 1993, the year before enactment. The share stood at 15.5 percent in 1998 and at 15.7 percent in 2003, near the end of the ban decade. Movement across the decade was therefore upward by a small margin rather than downward, in a series that began before the statute and ended before expiration. Recovered crime guns form a particular evidentiary base, reflecting firearms recovered by law enforcement and recorded in the systems the evaluators used, rather than a census of every firearm used in crime, and the synthesis presented the city series as indicators of criminal use drawn from that base. Within those limits, the Baltimore figures answer the exposure question across three points spaced over ten years. The share did not fall in 1998 relative to 1993, and the share did not fall in 2003 relative to 1998.

The Milwaukee series moved in the same direction over a shorter span and by a larger margin. In Milwaukee, 22.4 percent of recovered crime guns were fitted with large capacity magazines in 1993, and the share reached 36.4 percent in 1998, a rise rather than the decline the magazine provision would have predicted if the exempt stock had been contracting. Milwaukee holds a particular place in this literature because the same city supplied the wound analysis in the interim evaluation, in which victims in cases involving large capacity magazines were shot about three times on average. Read together, the prevalence series and the wound analysis describe a city in which cases involving the devices carried heavy wounding and in which the devices appeared in a growing share of recovered crime guns during the early ban years. A handgun-based series for Anchorage in the same synthesis ran higher than both shares. Koper’s summary judgment in the final evaluation had been that success in reducing criminal use of the banned guns and magazines was mixed, and the city magazine series supplies the prevalence half of that judgment in numerical form.

Koper attributed the failure to reduce magazine use in crime during the first several years to a specific cause. The cause, in the explanation given in the 2013 synthesis chapter in Reducing Gun Violence in America, was the immense exempted stock, enlarged by imports. The stock figure behind that explanation was the estimate of about 25 million covered magazines already in private hands at enactment, devices lawfully possessed on September 13, 1994 that remained lawful to possess and transfer for the entire decade. Around that core stock, other flows added volume. Manufacturers produced more than an extra year’s normal supply of grandfathered weapons and lawful copies during 1994, and foreign pre-ban magazines continued to be imported in large numbers after enactment, a flow the final evaluation of 2004 identified as a reason definitive assessment was premature. The design constraint, restated here because the prevalence conclusion depends on the constraint, is that the subtitle barred devices made after enactment while leaving the pre-enactment stock in lawful circulation, and a stock measured in the tens of millions does not draw down quickly within the first several years of a ten year term.

The final evaluation added a counterweight that the synthesis retained. Koper, Woods and Roth judged in June 2004 that the magazine provision was the element of the subtitle with the clearest potential for larger effects over a longer horizon, if the exempted stock were drawn down. Capacity connects directly to rounds fired before reloading, and magazines are consumable objects in a way that could, over a sufficiently long period without replenishment, reduce the available pool. The condition embedded in that judgment was drawdown, and the evaluators observed that foreign pre-ban magazines continued to arrive in large numbers, working against drawdown throughout the measured period. The statute ended by its own sunset on September 13, 2004, so the longer horizon the final evaluation contemplated extended beyond the life of the law. The magazine record through 2003 therefore holds a tension internal to the government evaluations. Measured prevalence did not fall in the first several years, measured casualty aggregates did not fall across the decade, and the same evaluators identified the magazine provision as the element most likely to produce larger effects under conditions the decade as enacted did not supply.

The prevalence picture after expiration came from a study published in 2018. Christopher S. Koper, William S. Johnson, J. L. Nichols, A. Ayers and N. Mullins, in Criminal Use of Assault Weapons and High-Capacity Semiautomatic Firearms: an Updated Examination of Local and National Sources, published in the Journal of Urban Health in 2018, examined local and national sources on criminal use in the period after the federal prohibition had expired. The study updated the prevalence side of the earlier evaluations, asking how often the covered categories appeared among guns used in crime. The findings require separation of two categories. The first was assault weapons as such. The second and larger category combined assault weapons with other high-capacity semiautomatic firearms, capturing firearms outside the statutory definition that were nevertheless capable of accepting large capacity magazines or fitted with such devices. The final evaluation had already flagged that structure, reporting that use of non-banned semiautomatic firearms fitted with large capacity magazines had stayed steady or risen during the ban years, and that those firearms were used in crime far more often than the covered models.

For assault weapons as such, Koper, Johnson, Nichols, Ayers and Mullins reported in the Journal of Urban Health in 2018 that assault weapons accounted for 2 to 12 percent of guns used in crime across the sources examined, with most estimates under 7 percent. That range sits close to the pre-enactment baseline, when the interim evaluation found covered firearms accounted for 1 to 7 percent of guns used in violent crime across the sources examined, and the final evaluation put the ceiling at no more than 8 percent of gun crimes even before the prohibition. A narrower and more serious category produced higher figures in the 2018 study. Assault weapons accounted for 13 to 16 percent of guns used in murders of police in the sources examined, a share well above the share for guns used in crime generally in the same study.

The combined category produced the largest shares in the 2018 study. Assault weapons and other high-capacity semiautomatic firearms together accounted for 22 to 36 percent of crime guns across the sources examined, and the authors reported that the combined share ran higher in serious violence than in crime generally. The combined range reframes the magazine question. A policy addressed to a defined list of weapons reaches the smaller category, in which most estimates sat under 7 percent in the 2018 sources. A policy addressed to capacity reaches toward the larger combined category, because the common property across that category is the ability to fire many rounds before reloading, whether the firearm fell inside the statutory definition or outside the definition. The 2018 study measured a period in which the federal prohibition no longer operated and the exempt stock from the ban era had never been removed from circulation. Within that setting, the authors found that the share of crime guns accounted for by assault weapons and other high-capacity semiautomatics had grown since the federal ban expired, a direction that aligns with the exposure logic of the earlier evaluations, in which prevalence of the devices in crime was the precondition through which any casualty effect would have to travel.

The question the literature leaves open follows from that sequence, and the literature leaves the question open rather than resolved in either direction. Whether constraining capacity changes outcomes in the rare high-round attacks, the small subset of incidents in which more than ten shots are fired and in which the difference between reloading after ten rounds and continuing without reloading could alter the number of victims struck and the number of wounds inflicted, remains the unresolved core of the magazine inquiry. The aggregate measures could not isolate that subset with the precision the hypothesis requires, the prevalence series showed the devices present in crime throughout the measured years, and the final evaluation cautioned that how often attack outcomes depend on firing more than ten shots without reloading is not clear. The mass shooting literature, which examines the extreme end of that same subset under definitions and methods of its own, takes up the unresolved question in the section that follows, and the magazine findings assembled here form the background against which those mass shooting findings, and the disagreements among those findings, have to be read.

Mass Shootings: A Real Finding and Its Methodological Shadow

The mass shooting literature on the federal prohibition occupies a distinctive place in the evidence record assembled for the statute that operated from September 13, 1994 to September 13, 2004. For most outcomes examined in the government evaluations, the measured effects were small, mixed, or undetectable within the ten year period. For mass shootings, by contrast, several research teams reported changes during the prohibition years that were large in percentage terms, and those reported changes attracted sustained attention from readers who treated the category as a central test of the law. The same literature also attracted sustained methodological criticism, directed less at the arithmetic inside any single study than at the definitions, thresholds, endpoints, and comparison periods on which that arithmetic depended. A faithful account of the outcome therefore has to hold two elements together in every instance. The first element is the finding as its authors reported it, with the researchers, the publication, the definition, and the period named. The second element is the criticism or qualification that accompanied the finding in the published record, reported with the same specificity. Neither element cancels the other on the page, and neither element is adopted as the position of the account. The synthesis rating assigned by the review literature then closes the sequence, because that rating expresses how a systematic review weighed the whole body of work rather than how any single team characterized its own results.

The difficulty begins with scale. Mass shootings, under every definition used in the studies covered in this section, were rare events counted in single digits or low double digits per decade or per year, and rarity shapes everything that follows. When an incident count moves from eight to six, the absolute movement is two incidents, and the proportional movement is 25 percent. When a death count moves from 81 to 49, the absolute movement is 32 deaths, and the proportional movement is 40 percent. Both descriptions of the same movement are arithmetically correct, and the choice between an absolute and a proportional presentation changes the impression a reader receives without changing the underlying record. Researchers who emphasized proportional change described declines or increases that sounded substantial. Commentators who emphasized absolute counts described movements that sounded modest. The disagreement between those presentations is not a disagreement about counting. The disagreement is about which scale honestly represents a rare event category, and that question has no answer inside the numbers themselves. The review literature treated the smallness of the counts as a standing problem for the entire outcome category, separate from the merits of any individual study, because small counts allow a single year or a single incident to move a decade level comparison by a large proportional amount.

How does the choice of definition change the mass shooting record?

The choice changes both the numerator and the comparison. A threshold of six or more killed counts fewer incidents than a threshold of four or more killed, and a public settings restriction counts fewer than a measure covering all settings. Each definition therefore produces a different ban period change from the same underlying years.

Definition is the second standing difficulty, and the studies covered in this section illustrate the range of choices researchers made. A threshold of six or more killed, used by Louis Klarevas and by John J. Donohue and Theodora Boulouta, defines a narrower category that the literature calls gun massacres. A threshold of four or more killed, used by Charles DiMaggio and co authors and embedded in the Mother Jones database discussed in the criticism of the Donohue and Boulouta analysis, defines a broader category that includes incidents killing four or five people. A restriction to mass public shootings, used by Grant Duwe, narrows the setting rather than the death count, and excludes incidents in private settings from the category regardless of the number killed. Each choice is defensible as a research decision, and each choice answers a slightly different question. A six death threshold asks whether the rarest and deadliest incidents changed. A four death threshold asks whether a wider class of multiple victim killings changed. A public settings restriction asks whether violence in public places changed. Because the questions differ, the answers can differ without any team having miscounted, and a reader who compares percentages across studies without comparing definitions first compares answers to different questions as though the questions were identical. The review literature identified the absence of an agreed definition as a separate standing problem, with thresholds of four, five, or six deaths and with settings defined as public only or as all settings, and that absence frames every finding reported in the remainder of this section.

The first study in the sequence is the book length treatment by Louis Klarevas, Rampage Nation, published in 2016. Klarevas counted gun massacres, defined by a threshold of six or more killed, and organized the count around the life of the federal statute. The comparison had three parts. The first part was the decade before the prohibition. The second part was the ban decade itself, the ten years during which the statute barred manufacture, transfer, and possession of the covered firearms and feeding devices as the statute defined them, subject to the grandfathering of items lawfully possessed at enactment. The third part was the decade after expiry, following the end of the statute by its own sunset on September 13, 2004. On that three part comparison, Klarevas reported that incidents fell by 37 percent in the ban decade against the decade before, and that deaths fell by 43 percent across the same comparison. The reported movement after expiry ran in the opposite direction and was larger in proportional terms. Incidents rose by 183 percent in the decade after expiry against the ban decade, and deaths rose by 239 percent across that comparison. Presented as a sequence, the figures describe a decline during the prohibition years and a sharper rise after the prohibition ended, and that sequence is the finding for which Rampage Nation became widely cited in discussion of the statute.

The criticism carried in the literature against the Klarevas series addressed the construction of that sequence rather than the existence of the counts. The series, the criticism held, is heavily affected by single years at its endpoints, and the counts are small. Endpoint sensitivity matters in a three decade comparison because the first and last years of each decade carry the same weight in the total as every other year, while a single unusual year at a boundary can raise or lower the total for an entire decade. If a decade begins or ends with an atypical year, the comparison between decades reflects that atypical year in part, and a different choice of boundary years would produce a different proportional change from substantially the same underlying record. Small counts compound the concern, because the proportional changes Klarevas reported rest on a base of rare events in which a small absolute movement registers as a large percentage. A 37 percent decline and a 43 percent decline describe real counted movements within the definition the author chose, and the criticism does not allege that the incidents or deaths were invented. The criticism alleges something narrower and harder to answer, namely that a series built from small counts and sensitive endpoints cannot by itself show how much of the movement belonged to the statute and how much belonged to the placement of the boundaries and the distribution of rare events across particular years. Rampage Nation therefore enters the record with both elements attached. The finding is a reported decline during the ban decade and a reported rise after expiry, measured on a six or more killed definition. The shadow is a documented concern about endpoints and scale that follows the finding wherever the percentages are quoted without the construction that produced them.

The second study in the sequence moved from a book to a peer reviewed medical journal and from a six death threshold to a four death threshold. Charles DiMaggio and co authors published “Changes in US mass shooting deaths associated with the 1994 to 2004 federal assault weapons ban: Analysis of open source data” in the Journal of Trauma and Acute Care Surgery, volume 86, issue 1, 2019, pages 11 to 19. The definition of a mass shooting in the analysis was four or more killed, excluding the perpetrator. The data construction was restrictive in a different way from the Klarevas threshold. An incident entered the DiMaggio data only if the incident appeared in all three open source databases the team used, namely the Los Angeles Times database, the Stanford University database, and the Mother Jones database. A triple listing requirement of that kind trades completeness for verification, because an incident recorded in one or two sources and absent from the third fell outside the count even if the incident met the death threshold. On that definition and that construction, the team identified 44 incidents and 501 deaths across the full study period from 1981 to 2017, a span covering years before the prohibition, the prohibition years themselves, and years after expiry.

The findings DiMaggio and co authors reported were expressed in several distinct measures, and the measures answer different questions. The first measure was relative. Mass shooting fatalities were 70 percent less likely during the ban period when measured against total firearm homicides, expressed as a relative rate of 0.30. A relative rate of that form compares the share of firearm homicide mortality accounted for by mass shootings across periods, rather than comparing raw death counts alone, so the measure adjusts in part for the rise and fall of firearm homicide generally across a long study period. The second measure translated the relative result into an absolute rate. The team reported nine fewer mass shooting deaths per 10,000 firearm homicides during the ban period. The third measure returned to raw annual averages. Average deaths per year were 7.2 before the prohibition, 5.3 during the prohibition, and 25 from 2004 to 2017. The fourth element of the paper was a modeled estimate rather than an observed count. The authors estimated that a continuous ban might have prevented 314 of the 448 deaths that occurred in the non ban years of the study period. That figure describes the output of a model that extended the ban period association across years in which the statute was not in force. The figure does not describe deaths observed to have been prevented in any particular year, and the distinction between a modeled counterfactual and an observed reduction is part of the finding as the authors presented the analysis.

The qualification attached to the DiMaggio findings came, in substantial part, from the authors themselves. DiMaggio and co authors stated that the observational analysis cannot definitively establish that the ban caused the reduction or that expiry caused the later growth. That statement reflects the design of the study. An observational analysis of open source data across a single national timeline has no comparison population that lived under a different law during the same years, and the association between the ban period and lower measured rates therefore competes with every other change that occurred across the same years as an explanation for the movement. The authors reported the association, quantified the association in relative and absolute terms, modeled the size of the effect if the association were causal and continuous, and then declined to claim that the design could establish causation definitively. A reader who quotes the 70 percent figure or the estimate of 314 deaths without the authors own causation statement quotes the paper selectively, because the paper contains both the estimate and the limit on the estimate in the same analysis. The finding and the qualification stand together in the published record, and the sequence required for this outcome reports both together for that reason.

The third study in the sequence returned to the six death threshold and to a direct decade comparison, and the published response to the study turned the definition question into the center of the dispute. John J. Donohue and Theodora Boulouta presented their analysis in 2019, in the New York Times. Using a threshold of six or more killed, Donohue and Boulouta reported that gun massacres fell from eight to six between the pre ban decade and the ban decade, a decline of 25 percent in incidents, and that fatalities fell from 81 to 49 between the same decades, a decline of 40 percent in deaths. The absolute movements behind those percentages were two fewer incidents and 32 fewer deaths across a decade comparison. The authors presented the proportional declines as evidence consistent with an effect of the prohibition on the deadliest category of incidents, defined at the six death line the authors shared with the Klarevas analysis.

The published criticism of the Donohue and Boulouta analysis came from Jacob Sullum, writing in Reason in September 2019. Sullum argued that the six victim threshold excluded shootings killing four or five people and magnified the change the authors reported. The mechanism of the alleged magnification follows from the arithmetic of small categories. A narrow threshold produces a small base, and a small base converts a movement of two incidents into a 25 percent change and a movement of 32 deaths into a 40 percent change. A broader threshold produces a larger base, against which a movement of similar absolute size registers as a smaller proportional change. Sullum supplied the broader comparison from the Mother Jones definition of four or more killed. On that definition, the same decades show 16 incidents and 125 deaths before the prohibition against 15 incidents and 96 deaths during the prohibition. The incident movement on the broader definition was a decline from 16 to 15, and the death movement was a decline from 125 to 96. Sullum characterized that broader record as a slight decline, in contrast to the larger proportional declines produced by the six death threshold. The dispute between the analysis and the criticism is therefore unusually transparent. Both sides worked from counted incidents in the same decades. Neither side alleged that the other side had fabricated incidents or deaths. The difference lies entirely in where the threshold line was drawn, which incidents the line admitted, and how large a proportional change the admitted incidents produced. A threshold at six deaths answers a question about the deadliest incidents only. A threshold at four deaths answers a question about a wider class that includes four and five fatality incidents. The Donohue and Boulouta finding and the Sullum criticism, read together, show that the size of the reported ban period decline in this literature is partly a function of a definitional choice made before any counting began.

The fourth study in the sequence narrowed the setting instead of raising the death threshold, and reached a different result on incidence while reporting a similar direction of movement after expiry. Grant Duwe published “Patterns and Prevalence of Lethal Mass Violence” in Criminology and Public Policy in 2019. Duwe examined mass public shootings, meaning that the setting restriction rather than a six death threshold defined the category under study, and the analysis controlled for population growth across the study period. Controlling for population growth changes the measure from a raw count to a rate sensitive comparison, because a growing population can produce a rising count of rare events without any rise in the underlying incidence per person. On that construction, Duwe reported that the federal ban period was not associated with the incidence of mass public shootings. The null association on incidence during the prohibition years stands in contrast to the declines reported by Klarevas, by DiMaggio and co authors, and by Donohue and Boulouta under their respective definitions and constructions. Duwe also reported that the incidence and severity of mass public shootings rose in the decade after the ban expired. That second element of the Duwe findings runs in the same direction as the post expiry increases reported by Klarevas on the six death definition and the higher annual average deaths reported by DiMaggio and co authors for the years from 2004 to 2017. The finding is therefore a finding about mass public shootings as Duwe defined the category, reported with the population control the author applied, and the finding neither confirms nor refutes the differently defined findings reported by the other teams.

The review literature organized those divergences into a set of standing methodological problems that apply to the outcome as a whole, and the problems deserve statement in their own right because the problems explain why agreement among some studies did not produce a settled rating. The first problem is the very small number of events. Mass shootings under any of the definitions in use occurred rarely enough that a decade of national data contains a limited number of incidents, and statistical analysis of rare events has limited power to separate a policy association from random clustering of events in particular years. The second problem is the absence of an agreed definition of a mass shooting. Thresholds of four, five, or six deaths all appear in the research record, and a study may restrict the category to public settings only or extend the category to all settings, so the literature does not measure one consistently defined outcome across studies. The third problem is sensitivity to start and end years. A comparison organized around a statute that operated from 1994 to 2004 assigns every event to a before, during, or after period according to boundary dates, and rare events near a boundary can change a period total enough to change the reported proportional movement. The fourth problem is the impossibility of a control group for a single national law. The prohibition applied across the United States during the same years, so no comparable population lived without the prohibition during the ban decade and supplied the counterfactual record against which the observed record could be measured directly. Each problem is distinct, and the problems compound one another, because small counts make endpoint choices more consequential and definitional variety makes cross study comparison less direct.

The synthesis rating for the outcome reflects that combined record. The RAND Corporation addressed bans on the sale of assault weapons and high capacity magazines in the 2020 edition of The Science of Gun Policy, the second edition of that review, published in 2020. For the effect of assault weapon bans on mass shootings, the 2020 edition of The Science of Gun Policy rated the evidence as inconclusive. An inconclusive rating in the RAND scale, which runs from supportive through moderate and limited to inconclusive, is a statement about the body of evidence rather than a verdict that an effect did or did not occur. The rating records that the research reviewed in the 2020 edition did not permit the reviewers to determine whether bans of that kind affected mass shootings, given the definitions, counts, designs, and divergences present in the literature. The rating sits alongside the individual findings rather than replacing the findings. The 2020 edition of The Science of Gun Policy weighed that record and rated the evidence inconclusive, and that rating, attached to the findings and to the criticisms in the order the record produced them, is the standing synthesis for the mass shooting outcome in the evidence on the federal assault weapons ban.

State Statutes as Comparisons and the Syntheses That Followed

The federal prohibition presented evaluators with a basic identification problem. A single national statute took effect on September 13, 1994, and ended on September 13, 2004, and no untreated part of the country remained against which outcomes under the statute could be measured directly. The review literature and the synthesis discussion in the 2020 edition of The Science of Gun Policy, published by the RAND Corporation, identified the impossibility of a control group for a single national law as a standing methodological problem for the entire evaluation record. State statutes offered the one form of comparison the federal record lacked, because states adopted, retained, or declined comparable restrictions at different times, and researchers could place those differing regimes beside one another over the same calendar years.

That promise carried an equally plain limit. A state comparison was never a federal experiment run at smaller scale. State definitions of the covered category differed, the firearms and magazines at issue moved easily across state lines, and the grandfathered stock created by the federal statute circulated in every state regardless of state law. Christopher S. Koper, in the synthesis published as part of Reducing Gun Violence in America by Johns Hopkins University Press in 2013, carried the evaluations estimate that roughly 1.5 million covered firearms and about 25 million covered magazines were already in private hands at enactment and remained lawful to possess and transfer. Any state finding therefore measured a state rule operating on top of that exempted national stock, and no state finding, taken alone, answered the federal question. A side by side presentation of the state statutes compared shows how far the state definitions diverged from the federal definition and from one another.

What do state comparisons add to the federal record?

State comparisons add variation that the single federal statute lacked. Researchers could compare states with and without bans over the same years. That variation produced one suggestive homicide finding and one magazine finding on fatal mass shootings, but differing definitions, porous borders, and the grandfathered stock limit what either result can prove.

The first named state study in the record belongs to Mark Gius, whose article, An Examination of the Effects of Concealed Weapons Laws and Assault Weapons Bans on State-Level Murder Rates, appeared in Applied Economics Letters, volume 21, number 4, in 2014. The treatment of that study in the 2020 edition of The Science of Gun Policy, published by the RAND Corporation, supplies the careful characterization that the evidence discipline of this article requires. In RAND characterization, the Gius study reported a suggestive effect consistent with state assault weapon bans decreasing firearm homicides. The adjective matters. A suggestive effect, in the language of that synthesis, described a direction of association reported by one study, and it did not describe a body of replicated findings pointing the same way.

The same RAND treatment drew a second distinction that readers of the Gius study often miss. RAND held that the federal ban analysis inside the Gius study did not meet the inclusion criteria that RAND applied to its evidence review. The state level result and the federal level result inside one published article therefore carried different standing in the synthesis record. The state result entered the review as a suggestive finding. The federal result did not enter the review at all. A reader who cited Gius for a proposition about the federal statute would be citing the part of the study that the RAND reviewers excluded, and a reader who cited Gius on state statutes would be citing one study, with the characterization that RAND itself supplied.

That restraint follows a pattern set by the commissioned federal evaluations themselves. Jeffrey A. Roth of the Urban Institute and Christopher S. Koper, in Impact Evaluation of the Public Safety and Recreational Firearms Use Protection Act of 1994, published by the Urban Institute in March 1997 with funding from the National Institute of Justice, used state comparisons as one check on early attribution. Over the window that the interim evaluation examined, homicide declines were not larger in states that already had bans, and the evaluators reported declines of 10.3 percent in states with no ban when the federal law took effect against 0.1 percent in states with bans. The interim evaluators treated that pattern as one reason to refuse early attribution of national homicide movement to the federal subtitle. State comparisons disciplined claims across this literature. State comparisons rarely, on this record, supplied the claim itself.

The second named state study addressed a narrower outcome and a different provision, and its findings divided by provision in a way that resists summary by slogan. Daniel W. Webster, Alexander D. McCourt, Cassandra K. Crifasi, Marisa D. Booty, and Elizabeth A. Stuart published their analysis in Criminology and Public Policy in February 2020. The Webster team examined fatal mass shootings in 45 states from 1984 to 2017, with a fatal mass shooting defined by a threshold of four or more killed. A design spanning 45 states and more than three decades gave the team variation that the federal record lacked, and the team estimated the associations of several distinct state policies inside a common framework.

Within that framework, the provision that carried the finding was the magazine provision. Webster, McCourt, Crifasi, Booty, and Stuart found that state bans on large capacity magazines were associated with significant reductions in the rate of fatal mass shootings, defined at the threshold of four or more killed, and with significant reductions in deaths from those events. The authors reported that effect sizes varied across their analyses, a qualification that belongs inside any faithful account of the result, because a range of estimates across model specifications describes a less settled quantity than a single point estimate repeated without its companions. The direction of the magazine association was consistent in the authors reporting, and the magnitude depended on the analysis chosen.

Assault weapon bans by themselves carried no comparable association in the same models. Webster, McCourt, Crifasi, Booty, and Stuart reported that assault weapon bans as a category were not associated with fatal mass shootings in their analyses of the 45 states from 1984 to 2017. The contrast between the magazine result and the assault weapon result, inside one study, one dataset, and one set of models, forms the most instructive feature of the Webster record. A summary that merged the two provisions into a single verdict on state bans would erase the distinction the authors themselves estimated. A summary that quoted the magazine association and omitted the assault weapon result would misrepresent the study just as surely, in the opposite direction.

One further estimate from the Webster models belongs in this account strictly as context, and the context concerns a different policy. Handgun purchaser licensing was associated in the Webster analyses with an estimated 56 percent fewer fatal mass shootings. That figure is stated here once, and for one reason only. The licensing estimate shows the company the magazine estimate kept inside the Webster framework, alongside another provision with a larger estimated association, estimated by the same team over the same states and years. Licensing is a distinct policy aimed at purchasers rather than at firearm features or magazine capacity, and no part of the licensing estimate transfers to the magazine bans, to the assault weapon bans, or to the expired federal subtitle. The figure locates the magazine finding inside a wider model. The figure supports no other use.

The structural limits of state comparison qualify both the Gius result and the Webster result, and the limits are features of the subject matter rather than defects peculiar to either team. The first limit is definition. The federal statute defined a semiautomatic assault weapon through named models plus a two feature test, and state statutes defined their covered categories through their own lists and tests, which did not coincide with one another or with the federal definition in every particular. A study pooling state bans therefore pooled related but nonidentical legal categories, and an estimated average association across those categories described the pooled category, not any single statutory text an individual state enacted.

The second limit is geography. A firearm or a large capacity magazine is a portable good, and a purchase restricted in one state could be completed in a neighboring state without a comparable restriction and carried across a state line that no inspection regime policed. Porous borders of that kind dilute the contrast between a state coded as treated and a state coded as untreated, because residents of the treated state retained practical access to the restricted items through ordinary travel, complicating the reading of both null and positive results.

The third limit returns to the design constraint that runs through the entire federal evaluation record, and the constraint applies to state regimes with full force. Possession and transfer of covered firearms and covered devices lawfully possessed on September 13, 1994, remained lawful throughout the federal period, and the estimates carried in Koper 2013 synthesis placed that grandfathered stock at roughly 1.5 million covered firearms and about 25 million covered magazines, enlarged for magazines by continued importation of foreign pre ban magazines in large numbers, as Christopher S. Koper, Daniel J. Woods, and Jeffrey A. Roth reported in An Updated Assessment of the Federal Assault Weapons Ban, published in June 2004 as a report to the National Institute of Justice. That stock circulated in states with bans and in states without bans alike during the years the Webster team and the Gius study examined. A state ban adopted against that background regulated new transactions at the margin of a large lawful installed base, and any measured association, positive or null, was an association produced under that condition.

The syntheses that followed the primary studies performed a different task. The synthesis bodies weighed whole literatures against stated standards, asked whether the accumulated evidence determined an answer, and, where a rating system applied, assigned a grade to the evidence for each outcome separately. Three synthesis efforts, published in 2004, 2005, and 2020, together with two scholarly characterizations from 2020 and 2021, make up the record.

The National Research Council addressed the ban in Firearms and Violence, A Critical Review, published by the National Academies Press in 2004. The Council committee concluded that the evidence was insufficient to determine the effectiveness of the ban. That formulation deserves exact reading. Insufficient evidence to determine effectiveness is a statement about the evidence base available to the committee in 2004, and it is neither a determination that the statute worked nor a determination that the statute failed. The committee reached that position with the commissioned evaluations directly in view. The summary of the Council report carried in the 2020 edition of The Science of Gun Policy, published by the RAND Corporation, records that the Council described the studies by Christopher S. Koper and Jeffrey A. Roth as finding no short term effect on gun violence outcomes alongside a temporary increase in prices.

Robert A. Hahn and co-authors reached a parallel formulation through a different institutional route one year later. Hahn and co-authors published Firearms Laws and the Reduction of Violence, A Systematic Review, in the American Journal of Preventive Medicine, volume 28, number 2, in 2005, as the review conducted for the Task Force on Community Preventive Services. For bans on specified firearms or ammunition, the Hahn review found insufficient evidence to determine effectiveness on violent crime outcomes. Two independent synthesis efforts, applying systematic review methods in 2004 and 2005, landed on the same judgment. Neither body graded the literature as demonstrating ineffectiveness.

The RAND Corporation synthesis supplied the most granular version of that judgment, because RAND graded the evidence outcome by outcome rather than issuing one verdict on bans as a class. The Science of Gun Policy, second edition, published by the RAND Corporation in 2020, carries the full title The Science of Gun Policy, A Critical Synthesis of Research Evidence on the Effects of Gun Policies in the United States. The edition and the year are part of the citation, because RAND maintained the review as a living synthesis and ratings in the 2020 edition describe the evidence as that edition assessed it. RAND rating scale ran through five grades. Supportive evidence stood at the top of the scale, followed by moderate evidence, then limited evidence, then inconclusive evidence, with a separate designation for outcomes where no qualifying studies existed. The scale graded the strength and consistency of the underlying research, and a grade of inconclusive recorded that the qualifying research did not support a confident determination of an effect in either direction.

Against that scale, the 2020 edition of The Science of Gun Policy rated the evidence on bans on the sale of assault weapons and high capacity magazines as inconclusive for mass shootings. The rating addressed the effect of assault weapon bans on mass shootings in particular, and the April 2021 account by FactCheck of the 2020 update stated that rating in those terms. An inconclusive grade on this outcome sat alongside a primary literature, reviewed elsewhere in this article, whose individual studies reported divergent results under divergent definitions and thresholds. The grade summarized that divergence. The grade did not resolve the divergence in favor of any single study within it.

For homicide, the 2020 edition of The Science of Gun Policy returned the same grade on both outcome measures and for both ban types. RAND rated the evidence inconclusive for the effect of assault weapon bans on total homicides, inconclusive for the effect of assault weapon bans on firearm homicides, inconclusive for the effect of bans on high capacity magazines on total homicides, and inconclusive for the effect of bans on high capacity magazines on firearm homicides. Four outcome and policy combinations, graded separately, produced one repeated grade, and the repetition is itself the finding worth reporting. The Gius state level result, characterized by RAND as suggestive of a decrease in firearm homicides under state assault weapon bans, formed part of the literature behind the firearm homicide grade without displacing it, because one suggestive study inside a wider and less consistent literature does not convert an inconclusive body of evidence into a supportive one under the RAND standard.

For violent crime generally, the 2020 edition of The Science of Gun Policy likewise rated the evidence inconclusive for bans on the sale of assault weapons and high capacity magazines. The general violent crime grade aligned with the position Hahn and co-authors had taken in the American Journal of Preventive Medicine in 2005, when the systematic review for the Task Force on Community Preventive Services found insufficient evidence to determine the effectiveness of bans on specified firearms or ammunition on violent crime outcomes.

One outcome received a stronger grade, and the outcome concerned the gun industry rather than crime. In the 2020 edition of The Science of Gun Policy, RAND found limited evidence that bans on the sale of assault weapons and high capacity magazines increase prices or sales of banned firearms in the short term. Limited evidence, on the RAND scale, sits above inconclusive and below moderate. The grade also harmonized with the market record in the commissioned evaluations, where Roth and Koper documented speculative price increases for covered firearms before enactment, followed by a reversal toward pre ban levels, and with the National Research Council description of the Koper and Roth studies as finding a temporary increase in prices.

Andrew Morral, project director for the RAND review, characterized the assault weapon ban evidence in March 2021 in terms that fix the meaning of the inconclusive grades with unusual precision. Morral stated that the research reviewed does not provide compelling evidence one way or the other on assault weapons bans. Morral paired that characterization with an explicit caution against the most common misreading of it. An absence of compelling evidence in either direction, in Morral formulation, is not a finding that the bans are ineffective. The distinction separates a claim about a literature from a claim about a policy. The literature, as graded in the 2020 edition of The Science of Gun Policy, failed to determine the effects of the bans on mass shootings, on total homicides and firearm homicides, and on violent crime generally. A determination of ineffectiveness would have required affirmative evidence of no effect, and the RAND review reported no such determination.

Philip J. Cook and Kristin M. Goss, professors at Duke University, offered a scholarly assessment in the same period that reached a compatible position from the evaluation literature rather than from a formal rating scale. In The Gun Debate, second edition, published in 2020, Cook and Goss summarized the evaluation literature on the federal ban and concluded that there is no compelling evidence that the ban saved lives. The Cook and Goss formulation and the Morral formulation converge on the word compelling. Neither characterization claimed proof that the statute accomplished nothing.

The record of what the statute plainly did not accomplish requires no recourse to any of those contested grades, because the final evaluation and the statutory design establish the points directly. The statute did not remove the covered items from circulation. Grandfathering preserved lawful possession and transfer of every covered firearm and every covered magazine lawfully possessed on September 13, 1994, the estimates carried in Koper 2013 synthesis placed that stock at roughly 1.5 million firearms and about 25 million magazines, production surged during 1994 before the effective date, and foreign pre ban magazines continued to enter the country in large numbers after enactment, enlarging the exempted magazine stock while the prohibition on new manufacture and transfer operated. Circulation of the covered category continued, lawfully and at scale, for the entire decade.

The statute did not produce a measurable reduction in the lethality or injuriousness of gun violence within its ten years. Christopher S. Koper, Daniel J. Woods, and Jeffrey A. Roth stated that conclusion in An Updated Assessment of the Federal Assault Weapons Ban, the June 2004 report to the National Institute of Justice covering the period from 1994 to 2003, reporting no discernible reduction in the share of gun crimes resulting in death or in the share of gunfire incidents resulting in injury. The statute did not reach the firearms used in the large majority of gun crime. The interim evaluation placed the covered firearms at 1 to 7 percent of guns used in violent crime across the data sources examined, the final evaluation carried a ceiling of no more than 8 percent of gun crimes even before the ban, and handguns, which the statute did not cover, accounted for the large majority of gun crime throughout, with substitution to lawful firearms available. Finally, the statute did not outlive its own schedule. The subtitle ended on September 13, 2004, by its sunset, at the moment when the same final evaluation judged definitive assessment premature, described the effects of the ban as unfolding only gradually because of grandfathering, and identified the magazine provision as the element with the clearest potential for larger effects over a longer horizon if the exempted stock were drawn down. The ten year term closed before the gradual effects the evaluators described could accumulate, and the synthesis grades recorded in the 2020 edition of The Science of Gun Policy were grades on a record that the sunset had fixed in 2004.

The five-finding evidence table

The table compresses the question by question record into five outcomes, with each row carrying the direction, period, principal research, identification problem, and synthesis rating that belong to that outcome alone.

Outcome	Direction of the measured effect	Period measured	Principal research	Identification problem	RAND 2020 edition rating
Criminal use of the covered firearms	Declined (17 to 72 percent across six study cities)	1994 to 2003	Koper, Woods and Roth final evaluation for the National Institute of Justice, 2004	Trace and recovery data measure enforcement contact as well as use, and the exempted stock remained in lawful circulation throughout	no RAND rating for this exact outcome; closest rated outcome: inconclusive (violent crime under bans of this type, RAND 2020 edition)
Overall gun violence and the national crime decline	No effect creditable to the statute	1994 to 2003	Koper, Woods and Roth final evaluation, 2004, with the interim evaluation of 1997	A single national law with no control group during a broad crime decline with many candidate causes	Inconclusive (total and firearm homicides, RAND 2020 edition)
Prices and availability of the covered items	Sharp pre-enactment rise, then a fall to near pre-ban levels	Pre-enactment through mid-1996	Roth and Koper interim evaluation and Koper and Roth, Journal of Quantitative Criminology, 2002	Speculative anticipation and a production surge confound the enactment date as a treatment point	Limited evidence of short-term price or sales increases (gun industry outcomes, RAND 2020 edition)
Magazine use in crime and casualties per incident	No early decline in magazine use in crime; no measured reduction in wounds per incident	1994 to 2003, with later prevalence data to 2018	The interim and final evaluations, Koper’s 2013 synthesis, Koper and co-authors, Journal of Urban Health, 2018	The exempted stock was never drawn down and imports replenished it, so exposure barely changed	Inconclusive (violent crime, RAND 2020 edition)
Mass shooting deaths	Lower during the ban period in the DiMaggio analysis, not associated in the Duwe analysis	1981 to 2017 across the studies	DiMaggio and co-authors, 2019, and Duwe, 2019, with Klarevas, 2016	Very small event counts, competing definitions and endpoint sensitivity	Inconclusive (mass shootings, RAND 2020 edition)

What a Careful Reader Can Say

The namable claim about the federal assault weapons ban is a claim about quotation. Both camps in the long argument over Subtitle A of Title XI of the Violent Crime Control and Law Enforcement Act of 1994 quote the same document, the final evaluation written by Christopher S. Koper, with Daniel J. Woods and Jeffrey A. Roth, for the National Institute of Justice in June 2004. One camp quotes the finding that gun crimes involving the covered firearms declined by between 17 and 72 percent across Anchorage, Baltimore, Boston, Miami, Milwaukee, and St. Louis, measured through 2003. The opposing camp quotes the judgment in the same report that the researchers could not clearly credit the prohibition with any of the national drop in gun violence. Both quotations are accurate. A third passage, also accurate, forecast that a renewed prohibition would produce effects likely to be small at best and perhaps too small for reliable measurement. The honest unit of account is therefore the question by question finding, because each passage answers a different question about a different outcome over a defined period.

The record supports several statements without qualification beyond the researchers and periods attached to each statement. Criminal use of the covered firearms fell in the six study cities in the final evaluation. Primary market prices rose by upwards of 50 percent before enactment and then fell to nearly pre ban levels through the middle of 1996, while manufacturers added more than an extra year of normal supply during 1994. Possession and transfer of the grandfathered stock remained lawful throughout, with estimates carried in the 2013 synthesis by Koper placing that stock at roughly 1.5 million covered firearms and about 25 million covered magazines at enactment, and foreign pre ban magazines continued to enter in large numbers. City series in the same synthesis showed no early decline in the share of recovered crime guns fitted with large capacity magazines. The statute also plainly ended on September 13, 2004, by its own sunset, before the gradual effects the evaluators described could accumulate over a longer horizon.

The record does not support several larger statements that quotation sometimes attempts to build from those findings. The final evaluation did not credit the statute with the national crime decline, and the interim evaluation had already reported that homicide declines were not larger in states with bans in place. Neither evaluation found a discernible reduction in the lethality and injuriousness of gun violence, measured through deaths among gun crimes and injuries among gunfire incidents. The magazine record does not support a claim of early success, because prevalence in crime did not fall in the city series and casualties per incident did not fall in the wound analyses. The mass shooting record does not support a settled conclusion in either direction, because DiMaggio and co authors reported lower deaths during the prohibition period under one definition and construction, Duwe reported no association for mass public shootings under another, and the 2020 edition of The Science of Gun Policy rated the evidence inconclusive for that outcome. Claims that move past those limits enter the territory mapped by claims built on these findings, where accurate fragments are assembled into verdicts the underlying reports did not issue.

The synthesis record reinforces that restraint. The National Research Council in 2004 and Hahn and co authors for the Task Force on Community Preventive Services in 2005 each found the evidence insufficient to determine effectiveness. The 2020 edition of The Science of Gun Policy rated the evidence inconclusive for mass shootings, for total and firearm homicides, and for violent crime generally, with limited evidence only for a short term increase in prices or sales. Andrew Morral, the RAND project director, stated in March 2021 that an absence of compelling evidence in either direction was not a finding of ineffectiveness, and Philip J. Cook and Kristin M. Goss wrote in 2020 that no compelling evidence showed the prohibition saved lives. Those judgments concern measured effect. Questions of legal authority belong to the constitutional context, a separate record that neither supplies nor removes the identification problems in the evaluation literature.

The design constraint explains why the measured record took that shape. Grandfathering guaranteed gradual effects, the baseline share of the covered firearms was no more than 8 percent of gun crimes even before enactment, substitution to lawful semiautomatic firearms remained available and was observed, and the magazine provision held the clearest potential for larger effects only if the exempted stock were drawn down. That drawdown did not occur during the ten year term. A careful reader can therefore state the findings question by question, attach each finding to its researchers and period, carry the synthesis rating assigned to that outcome, and decline both a verdict of success and a verdict of failure where the commissioned record itself declined each verdict.

Readers assembling a research file from these evaluations, studies, and ratings can keep your statute notes, citations, and case chronologies together free on VaultBook.

Frequently Asked Questions

Q: Did the assault weapons ban reduce gun crime?

The answer depends on which category of gun crime the question addresses. Christopher S. Koper, with Daniel J. Woods and Jeffrey A. Roth, reported in the final evaluation for the National Institute of Justice in June 2004 that gun crimes involving the covered firearms declined by between 17 and 72 percent across the six study cities measured through 2003. The same researchers could not clearly credit the statute with the national drop in gun violence, and criminal use of non banned semiautomatic firearms fitted with large capacity magazines held steady or rose. The 2020 edition of The Science of Gun Policy rated the evidence inconclusive for total homicides, firearm homicides, and violent crime generally.

Q: What did the government study of the assault weapons ban conclude?

The commissioned evaluations reached a mixed set of conclusions rather than a single verdict. Jeffrey A. Roth and Christopher S. Koper concluded in the interim evaluation of March 1997 that the prohibition could have only a limited effect on total gun murders at best, because the covered weapons and magazines were involved in a modest fraction of gun murders. Koper, Woods, and Roth concluded in June 2004 that criminal use of the covered firearms had fallen in the study cities, that no discernible reduction in lethality or injuriousness had appeared, that the national crime decline could not be clearly credited to the statute, and that grandfathering guaranteed effects would occur only gradually.

Q: Why would grandfathering weaken the assault weapons ban effect?

Grandfathering preserved lawful possession and transfer of every covered firearm and magazine lawfully held on September 13, 1994. Estimates carried in the 2013 synthesis by Christopher S. Koper placed that stock at roughly 1.5 million covered firearms and about 25 million covered magazines at enactment. Production during 1994 added more than an extra year of normal supply before the effective date, and foreign magazines made before the ban continued to enter the country in large numbers afterward. The final evaluation therefore held that effects were guaranteed to occur only gradually. A stock of that scale, circulating lawfully for the entire decade, limited how quickly criminal access could change before the sunset arrived in September 2004.

Q: Did mass shooting deaths fall during the assault weapons ban?

Some studies reported lower figures during the prohibition period, and other studies reported no association, so the literature does not support a single unqualified answer. Charles DiMaggio and co authors, writing in the Journal of Trauma and Acute Care Surgery in 2019, found that mass shooting fatalities were 70 percent less likely during the ban period when measured against total firearm homicides, with average deaths per year of 5.3 during the ban against 7.2 before the ban. Grant Duwe, writing in Criminology and Public Policy in 2019, found that the ban period was not associated with the incidence of mass public shootings. The 2020 edition of The Science of Gun Policy rated the evidence inconclusive for mass shootings.

Q: How often were banned weapons used in crime before the assault weapons ban?

The covered firearms accounted for a small share of criminal gun use before enactment. Jeffrey A. Roth and Christopher S. Koper reported in the interim evaluation of March 1997 that the covered firearms accounted for only 1 to 7 percent of guns used in violent crime across the data sources the evaluators examined. Christopher S. Koper, Daniel J. Woods, and Jeffrey A. Roth stated the ceiling in the final evaluation of June 2004 as no more than 8 percent of gun crimes even before the prohibition took effect. Handguns, which the statute did not cover, accounted for the large majority of gun crime throughout the period the evaluations examined.

Q: What does RAND say about assault weapons ban evidence?

The RAND Corporation addressed bans on the sale of assault weapons and high capacity magazines in the 2020 edition of The Science of Gun Policy. That edition rated the evidence inconclusive for the effect of such bans on mass shootings, inconclusive for total homicides and firearm homicides, and inconclusive for violent crime generally. The same edition found limited evidence that such bans increase prices or sales of banned firearms in the short term, a rating for gun industry outcomes. Andrew Morral, the project director for the RAND review, stated in March 2021 that the research reviewed provided no compelling evidence one way or the other, and that the absence of compelling evidence was not a finding of ineffectiveness.

Q: Did prices rise for pre ban firearms under the assault weapons ban?

Prices rose sharply before enactment and then reversed. Christopher S. Koper and Jeffrey A. Roth reported in the interim evaluation of March 1997 and in the Journal of Quantitative Criminology in 2002 that primary market prices of the covered firearms rose by upwards of 50 percent before enactment, as distributors, dealers, and collectors speculated that grandfathered weapons would become valuable. After enactment, prices of the banned guns and most large capacity magazines fell to nearly pre ban levels and remained there at least through the middle of 1996, reflecting an oversupply of grandfathered guns and the availability of lawful substitutes. The evaluators identified the result that, by 1996, the statute had made the covered firearms more accessible in the primary market.

Q: Do state assault weapons ban laws show different results?

The named state research points in more than one direction. Mark Gius reported a result that the 2020 edition of The Science of Gun Policy characterized as a suggestive effect consistent with state assault weapon bans decreasing firearm homicides, from a single study published in Applied Economics Letters in 2014. Daniel W. Webster and co authors reported in Criminology and Public Policy in February 2020 that state bans on large capacity magazines were associated with significant reductions in fatal mass shootings, while assault weapon bans as such were not associated with fatal mass shootings. The interim federal evaluation had earlier reported that homicide declines were not larger in states that already had bans, with declines of 10.3 percent in states without bans against 0.1 percent in states with bans.

Q: What did the interim evaluation find in the ban’s first years?

Jeffrey A. Roth and Christopher S. Koper, writing for the Urban Institute in March 1997 with funding from the National Institute of Justice, reported an early fall in trace requests, a ceiling on possible effects, and no detectable gain in the murder categories most associated with the covered weapons. Trace requests for the covered firearms fell by 20 percent between 1994 and 1995, against a fall of 11 percent for all guns. The evaluators could detect no reduction in murders with multiple victims in a single incident or in murders producing multiple bullet wounds per victim. The evaluators also warned that the mandated review period was too short and too close to enactment to judge effects on street violence, given the grandfathered stock in circulation.

Q: Did crimes committed with the banned firearms themselves decline?

Yes, on the measure and in the places the final evaluation examined. Christopher S. Koper, with Daniel J. Woods and Jeffrey A. Roth, reported in June 2004 that gun crimes involving assault weapons declined by between 17 and 72 percent across the six study cities of Anchorage, Baltimore, Boston, Miami, Milwaukee, and St. Louis, measured through 2003. The same report qualified that finding in two ways. Criminal use of non banned semiautomatic firearms fitted with large capacity magazines held steady or rose during the same period, and those firearms were used in crime far more often than the covered models. Koper summarized success in reducing criminal use of the banned guns and magazines as mixed.

Q: Did the ban reduce the lethality of gun violence?

The final evaluation detected no such reduction. Christopher S. Koper, Daniel J. Woods, and Jeffrey A. Roth reported in June 2004 that there had been no discernible reduction in the lethality and injuriousness of gun violence during the period from 1994 to 2003, measured through the share of gun crimes resulting in death and the share of gunfire incidents resulting in injury. Earlier work pointed the same way on related measures. Koper and Roth reported in the Journal of Quantitative Criminology in 2001 no evidence that the prohibition had reduced multiple victim gun homicides or multiple gunshot wound victimizations in the early period. The final evaluation also observed that relatively few gun attacks involve more than ten shots fired.

Q: What happened to the criminal use of large capacity magazines during the ban?

City prevalence series assembled by Christopher S. Koper in the 2013 synthesis of the evaluations showed no early decline in the share of recovered crime guns fitted with large capacity magazines. In Baltimore, that share stood at about 14.0 percent in 1993, 15.5 percent in 1998, and 15.7 percent in 2003. In Milwaukee, the share stood at 22.4 percent in 1993 and reached 36.4 percent in 1998. A handgun based series for Anchorage ran higher than both shares. Koper attributed the failure to reduce magazine use in crime during the first several years to the immense exempted stock of about 25 million magazines, enlarged by imports of foreign magazines made before the ban.

Q: How did substitution to lawful firearms affect what the ban could achieve?

Substitution placed a practical ceiling on the effect the statute could produce. The covered firearms had accounted for no more than 8 percent of gun crimes even before enactment, on the ceiling statement in the final evaluation of June 2004, and lawful firearms outside the covered definitions remained in production, in stores, and in the grandfathered secondary market throughout the decade. The evaluators drew the arithmetic consequence directly. Even complete elimination of the covered firearms from crime could have moved total gun violence only within that small share, and substitution to lawful firearms was available. The final evaluation documented the pathway in operation, reporting that criminal use of non banned semiautomatic firearms fitted with large capacity magazines held steady or rose.

Q: How did the National Research Council assess the evidence on the ban?

The National Research Council addressed the prohibition in Firearms and Violence, a critical review published by the National Academies Press in 2004. The committee concluded that the evidence was insufficient to determine the effectiveness of the ban. That formulation was a statement about the evidence base available in 2004, and the committee presented the conclusion as neither a determination that the statute worked nor a determination that the statute failed. The summary of the Council report carried in the 2020 edition of The Science of Gun Policy records that the Council described the studies by Christopher S. Koper and Jeffrey A. Roth as finding no short term effect on gun violence outcomes alongside a temporary increase in prices.

Q: What did later research find about assault weapons in crime after the ban expired?

Christopher S. Koper, William S. Johnson, J. L. Nichols, A. Ayers, and N. Mullins examined local and national sources in the Journal of Urban Health in 2018. The authors reported that assault weapons accounted for 2 to 12 percent of guns used in crime across the sources examined, with most estimates under 7 percent, and for 13 to 16 percent of guns used in murders of police. Assault weapons and other high capacity semiautomatic firearms together accounted for 22 to 36 percent of crime guns across the sources, with a higher share in serious violence than in crime generally. The authors also reported that the share of crime guns accounted for by that combined category had grown since the federal ban expired.

Q: Why do studies of mass shootings during the ban period reach different conclusions?

The review literature and the synthesis discussion in the 2020 edition of The Science of Gun Policy identified four standing problems that push studies toward different answers. Event counts are very small, so a single year or a single incident can move a decade comparison by a large proportional amount. No agreed definition exists, and studies use thresholds of four, five, or six deaths, with some studies restricted to public settings and others covering all settings. Results are sensitive to the start and end years chosen for the comparison periods. A single national law also leaves no control group, because no comparable population lived under a different law during the same years. Charles DiMaggio and co authors and Grant Duwe reached different findings in 2019 under different definitions and constructions.

Q: What did the Johns Hopkins researchers find about state magazine bans?

Daniel W. Webster, Alexander D. McCourt, Cassandra K. Crifasi, Marisa D. Booty, and Elizabeth A. Stuart published an analysis of fatal mass shootings in 45 states from 1984 to 2017 in Criminology and Public Policy in February 2020. The team found that state bans on large capacity magazines were associated with significant reductions in the rate of fatal mass shootings, defined by a threshold of four or more killed, and with significant reductions in deaths from those events. The authors reported that effect sizes varied across the analyses. Within the same models, assault weapon bans as such were not associated with fatal mass shootings, a contrast between the magazine provision and the weapon provision inside one study and one dataset.

Q: Did the ban change the number of wounds inflicted in gun assaults?

The wound analyses in the evaluation record were mixed on association and negative on change over time. A Milwaukee analysis in the interim evaluation found that victims in cases involving large capacity magazines were shot about three times on average. A multi site analysis in the same evaluation found no post ban reduction in the average number of wounds per case and no reduction in the share of cases involving multiple wounds. Christopher S. Koper and Jeffrey A. Roth reported in the Journal of Quantitative Criminology in 2001 no evidence that the prohibition reduced multiple gunshot wound victimizations in the early period. The final evaluation of June 2004 likewise reported no discernible reduction in the lethality and injuriousness of gun violence through 2003.

Q: What evidence suggests the ban was beginning to affect magazine supply by the time it expired?

The evidence is a single state recovery series presented at hedged strength. A Washington Post analysis of guns recovered in Virginia, reported in January 2013, found that the share of recovered guns fitted with large capacity magazines was generally high through the ban years and lower by 2011. Christopher S. Koper commented on that pattern and held that the pattern suggested the prohibition might have been starting to make a dent in the magazine market by the time the prohibition ended on September 13, 2004. The observation came from one state series rather than a national market measure, and the interpretation used the language of suggestion and possibility. The analysis and the comment together established a suggestion about timing rather than a measured national effect.

Q: Which findings about the assault weapons ban are most often quoted by each side?

Readers favoring the statute quote the findings on the covered category itself, principally the decline of between 17 and 72 percent in gun crimes involving assault weapons across the six study cities in the final evaluation of June 2004, and the reported declines in mass shooting deaths in studies by Louis Klarevas and by Charles DiMaggio and co authors. Readers skeptical of the statute quote passages from the same final evaluation, principally the judgment that the researchers could not clearly credit the prohibition with the national drop in gun violence, the absence of a discernible reduction in lethality, and the forecast that renewed effects were likely to be small at best. Both sets of quotations are accurate, and the dispute concerns which question deserves primacy.