Keeping with Tradition: Preference for the Longstanding Definition of Mass Shooting

James Alan Fox Email the Corresponding Author1 and Emma E. Fridel2 

1 School of Criminology and Criminal Justice, Northeastern University
2 College of Criminology and Criminal Justice, Florida State University  |  Full Citation Volume: 1, Issue: 2, Page(s): 17-26

Article History: Received June 27, 2022 | Accepted August 11, 2022 | Published Online September 12, 2022

As defined back in the 1980s, the term “mass shooting” has long been understood to mean the intentional killing of four or more victims with gunfire in a single incident. However, recent efforts to examine this rare and tragic crime have employed alternate definitional criteria. In order to facilitate cross-study comparisons and curb rampant public fear, it is imperative that scholars, politicians, and the media avoid using the same terminology to describe very different phenomena. In this article, we advocate for the traditional definition in view of a variety of theoretical and methodological considerations.

mass shooting, mass killing, definitions

Virtually all criminal offense types are clearly and incontrovertibly defined. Indeed, no one debates the meaning of homicide, robbery, auto theft, or kidnapping. In contrast, the definition of mass shooting has been—and continues to be—very much a matter of debate. For decades, the term “mass shooting” was simply an extension of the Federal Bureau of Investigation’s (FBI) seminal definition for mass murder: four or more victims killed in a single event, typically within 24 hours (Ressler et al., 1988). A mass shooting was essentially just a mass murder committed with firearms. Though somewhat arbitrary, this definition was designed to distinguish among types of multiple homicide and has been used by mass violence scholars for decades with minimal controversy, perhaps due to the paucity of research in this area.

Both public and academic interest in mass shootings, however, surged in the wake of several high-profile incidents in 2012, including the massacres in Aurora, CO, and Newtown, CT. With the “discovery” of mass shooting as a worthy focus of criminological research, the need for reliable data became clear. Given that there was no official resource for such incidents (other than the rather limited and sometimes erroneous cases drawn from the FBI’s Supplementary Homicide Reports), several news organizations, academic centers, and advocacy groups undertook efforts to fill the void by building their own databases from scratch.

(Click table to enlarge)

Unfortunately, there was no consensus as to the parameters of such an endeavor. As shown in Table 1, definitions of mass shooting vary by victim count threshold, extent of victim injury, and incident type, all of which fundamentally alter the nature and prevalence of the phenomenon itself. Based on our evaluation of the extant literature, the three most common definitions of mass shooting are:

  1. Four or more victims shot (but not necessarily killed), used, for example, by the Gun Violence Archive (GVA) beginning in 2013. There are approximately 600 mass shootings per year in the United States using this definition, averaging one fatality per incident.
  2. Four or more victims killed by gunfire, used, for example, by the Associated Press/USA TODAY/Northeastern University Mass Killing Database. This definition has been the standard since the 1980s. There are approximately two dozen mass shootings annually according to this definition, with an average of five victim deaths per incident.
  3. Four or more victims killed by gunfire in a public setting not involving ongoing criminal activity (such as gang conflict and drug trafficking), used, for example, by The Violence Project. This is a subset of the second definition, which is often referred to as “public mass shootings.” There are about a half dozen such incidents per year, claiming an average of seven victim fatalities per incident.

While all three definitions have their utility in different contexts, it is crucial that scholars, politicians, and the media avoid using the same terminology to describe these very different phenomena. Aside from limiting cross-study comparisons (e.g., datasets using differing definitions yielding contradictory results), the lack of a consensus on the definition for mass shootings has fomented public fear massively disproportionate to the risk of victimization. As such, we advocate for the traditional approach (Definition #2) for both theoretical and methodological reasons. Specifically, we contend that:

  1. Injury and death are qualitatively distinct and should not be conflated.
  2. Lowering the victim fatality threshold increases the risk of missing data bias, especially for cases occurring in earlier decades.
  3. Perpetrator deaths are not equivalent to those of victims and should not be included in fatality counts.
  4. All types of incidents—regardless of motivation, location, or weapon—are worthy of study.
  5. A concrete time frame establishing the length of “a single incident” is necessary to distinguish these events from serial and spree killings.
  6. Definitions must be applied consistently across cases and over time.

In the sections to follow, we discuss each of these six points in turn. It should be noted that these points are not equal in terms of salience nor addressed in such a way as to suggest order of importance.

Death is Different from Injury

A shooting with a large number of injuries is hardly equivalent to one that results in a large number of fatalities. A high death toll brings with it a series of funerals and memorials, which in certain cases attract national attention. We do not mean to ignore the awful suffering that comes from nonfatal gunshot wounds, but death is inherently different. Homicide leaves an indelible mark on the families, friends, and communities of the victims in a way that injury never can, no matter how severe. 

Is it reasonable to equate suffering a life-threatening wound with being grazed by a bullet? Is it logical that injuring four victims counts as a mass shooting, but fatally killing three does not, despite the fact that the latter is much more serious? Conflating fatalities with injuries, some of which may be minor, trivializes the severity and permanence of death. Expanding to include nonfatal injuries in the victim threshold fundamentally alters the nature of the crime—it is called mass murder, after all, not mass attempted-murder.

Even further, mass confusion arises when figures associated with the broadest notion of mass shooting (Definition #1) are referenced by the media in their reporting on an incident of much greater severity (Fox & Levin, 2015). After the May 2022 Robb Elementary School shooting in Uvalde, TX, that claimed the lives of 19 school children and two teachers, for example, countless media outlets noted that there already had been more than 200 mass shootings up to that point of the year, according to GVA statistics. But most of these were not like Uvalde: close to half of the GVA’s mass shooting incidents since 2013 resulted in no fatalities, and less than one-quarter involved multiple deaths (some of which were of the assailant). Beginning in 2019, the GVA began tallying mass murders (four or more victims killed by gunfire) in addition to mass shootings (four or more victims killed or wounded by gunfire). Of the 1,711 mass shootings from 2019 to 2021, only 80 (fewer than five percent) reached the four-victim fatality threshold for mass killing.

Equating fatalities with injuries can be terribly misleading and tends to elevate the level of fear, even if it was hardly the intention of the GVA to do so. Claims that mass shootings are “the new normal” (Holt & Gosk, 2018) or that there are more incidents than days (e.g., Silverstein, 2020) unfortunately lead many Americans to conclude incorrectly that massacres with double-digit death tolls are happening every time they turn around. 

As a prime example of blurring the distinction between injury and death, in May 2021, The New York Times (see Victor & Taylor, 2022) published what was described as a “partial list” of the 13 mass shootings that had occurred up to that point in the year, adding that there were “many more” not included. However, the “partial list” of mass shootings was the entire list of mass killings (with four or more victim fatalities), and the incidents not listed were of much lesser severity. In effect, the “partial list” characterization misleadingly implied that the omitted incidents were like the 13 deadliest.

Focusing on fatalities also has benefits for data collection and analysis. Death is unambiguous and thus easy to measure; injuries are a matter of degree, which can lead to confusing and somewhat subjective inclusion criteria. How severe does the wound have to be to count? And how accurate are casualty counts, if some victims do not report minor injuries that do not require medical treatment? In the same vein, it is generally much more challenging to find detailed information on victims who did not perish during the incident.

A key limitation of relying upon fatalities, however, is that the difference between life and death is often a matter of luck. Victim mortality depends on the number and location of wounds, emergency response times, and distance to medical facilities, and may not be a reflection of offender intent. In other words, many perpetrators may seek to kill multiple victims, but fail to do so as a result of factors outside of their control. Nevertheless, this raises the thorny issue of operationalizing motive, an impossible task when many offenders do not survive the incident. Therefore, while we acknowledge that a definition based exclusively on fatalities is inherently imperfect, we argue that it remains the most viable option.

Victim Threshold and Missing Data Bias

Another reason to exclude nonfatal injuries in establishing a victim count threshold—and to maintain a minimum of four fatalities—relates to the perils of retrospective collection of historical data. As mentioned earlier, for decades most researchers adopted the traditional definition of four or more victims shot to death, gradually collecting data on cases over the course of their careers. As a result, we have relatively reliable data on such incidents over a long period of time, providing crucial information for contextualizing whether mass shootings are indeed on the rise. In contrast, definitions involving fewer victims and/or including injuries are relatively new, and they require identifying cases from media reports and public records that are often decades old. Even further, such incidents are much more difficult to find due to the lower level of public attention and press coverage they receive—a task that becomes almost impossible for cases occurring prior to the internet era.

Existing databases have dealt with this issue in two ways: simply restrict the time period covered to more recent years (as does the GVA) or accept a high level of missing cases in earlier years (as does the Stanford University Geospatial Center, which has ceased data collection). Either way, such databases cannot provide accurate prevalence estimates of mass shootings over time and attempts by the media to do so have produced the myth that these massacres are becoming more and more commonplace. 

A case in point is the FBI’s effort, starting in 2014, to collect data on so-called “active shooter events,” defined as episodes in which “one or more individuals is actively engaged in killing or attempting to kill people in a populated area” (FBI, 2022, p. 2). Through a series of statistical reports, the FBI, in collaboration with Texas State University, has documented a sharp rise in such attacks since 2000. Although the FBI notes that active shooter events do not necessarily result in any deaths, the distinction is often lost in media coverage, with active shooter events incorrectly treated as synonymous with mass shootings (e.g., Schmidt, 2014). Aside from the confusion in terminology, the reported manifold increase in active shooter events is suspect because of apparent missing cases in the early years of the database (Lott, 2015). Although “active shooter event” became part of the law enforcement lexicon following the 1999 Columbine massacre, it did not start appearing in news stories until 2006, creating a challenge for identifying cases, particularly those of lesser severity, from the distant past. For example, the initial FBI report on active shooter events (Blair & Schweit, 2014) showing a nearly three-fold increase in cases—a finding that was covered widely by the media—was modified years later with additional incidents that had been overlooked. In fact, for the first three years (2000-2002), nine cases were added to the original 11, an update that would substantially lower the trendline presented in the original report.

Evidence of missingness bias is clear when comparing the FBI’s active shooter data from the early and later years. In the first five years of the dataset (2000-2004), where cases were identified retrospectively, 7.7% of the gunmen failed to kill anyone. By contrast, in the first five years of the project (2014-2018) where cases were identified concurrently, 37.0% of assailants failed to kill anyone. Although some of this increase in the share of nonfatal incidents may be linked to improved police response, most likely it reflects an inability to capture less serious cases of the earlier years that may not have been covered by major newspapers. Moreover, even if a nonfatal shooting did get reported in the local press, it could have been easily overlooked in searches of news archives for years when the term “active shooter” was relatively obscure.

It is important when assessing the ongoing frequency of mass shootings to have an accurate historical benchmark. For example, many observers comment on the prevalence of recent mass shootings in the GVA as an epidemic with no idea of how frequent shootings with four or more victims—but few, if any, fatalities—were in the early 1990s, when the nation’s murder rate was nearly twice the level it is now. Such early points of comparison can be hard to measure.

Perpetrators are not Victims

The idea that death is a singular phenomenon does not mean that all deaths are equally weighty. Specifically, an assailant’s death—be it by suicide, or by the intervention of police or a bystander—should not be included in a count of fatalities (as is the practice of the GVA and some other databases). Unlike the victims, the assailant is a willing participant in a crime that may lead to his or her demise. Including the death of an offender in the fatality count along with those of the victims only serves to inflate the severity of the event in the public’s mind.

The number of fatalities reported in the GVA incident listings, for example, can easily be misinterpreted as a victim fatality count because there is no indication of which incidents include an assailant’s death. One would need to open and inspect each incident report individually (a daunting task indeed) to identify which mass shooting fatality counts include an offender and which do not. Of course, this could very easily be resolved by noting with an asterisk which include an assailant’s death or by using separate columns for victim deaths and offender deaths. Moreover, offender fatalities could also be designated as a suicide versus killed by first responder or bystander.

All Types of Incidents are Important

While we consider death an important element in defining mass shooting, we do not concur with those who limit cases based on location, motive, or even weapon. Definition #3, for example, narrows the focus to indiscriminate slaughter of strangers as well as to targeted shootings in public places that may also put the lives of innocent bystanders at risk. These are the kinds of incidents that are reflected in the common stereotype of a mass shooting and that are most frightening to the public—as they can happen to anyone, at any time, and anywhere. In reality, public mass shootings account for only a small portion of incidents; familicides or family annihilations, for example, represent nearly half of all mass shootings annually, yet rarely inspire the same kind of public fear (Fox & Levin, 2022). Although they occur approximately as often as incidents in public, felony-related massacres involving gang violence and drug trafficking are all but ignored by the majority of Americans (Fridel, 2021). Focusing exclusively on public mass shootings essentially allows the rarest and most sensational type to define the phenomenon as a whole. Ignoring incidents in private spaces also does a disservice to the victims, who are just as dead no matter where or by whom they were killed. Some scholars, for example, list a victim count of 26 for the 2012 Sandy Hook Elementary School shooting, ignoring the fact that the perpetrator killed his mother at their joint home immediately prior for a total of 27 victims, each equally dead.

Besides and independent of our preference for including all types of motivations, relationships, and locations, we also advocate for expanding the focus of mass violence to encompass all types of weapons, not just firearms. Scholarship on other forms of lethal violence like homicide and suicide may focus on the role of guns, but certainly does not ignore deaths caused by other means. Although shootings account for nearly 80 percent of all mass killings in the United States (Carroll, 2021), ignoring victims who were slain with alternate weapons marginalizes their deaths. To them, did it really matter what weapon was used? In fact, the level of suffering experienced by victims who are stabbed, bludgeoned, or burnt to death is often more excruciating than that of gunshot victims. In the same vein, many of the deadliest massacres in U.S. history involved weapons other than guns, including the 1995 Oklahoma City bombing and the 1990 Happy Land Social Club arson fire. Including all mass murders also facilitates international comparisons, as massacres abroad disproportionately involve other weapons, such as vehicles.

Including all types of mass murders has its methodological advantages as well. Many cases defy categorization, as private violence sometimes spills into public spaces or vice versa. In a 2017 incident in Marathon County, WI, a 45-year-old man arrived at his wife’s workplace, threatening to kill her if she did not sign their divorce papers. As she fled to a nearby restaurant, the assailant retrieved his gun from his car and proceeded to shoot and kill two of his wife’s coworkers before gunning down his wife’s lawyer at her office and a responding officer at his own apartment. Similarly, in 2006, a 28-year-old man shot and killed six ravers at an open afterparty inside a private residence in the Capitol Hill neighborhood of Seattle, WA. Both cases illustrate the difficulty of defining what “public” really means. The first clearly was motivated by a domestic dispute, but just so happened to involve public locations; the second was a quintessential indiscriminate rampage shooting that took place in a private home. Moreover, many so-called “public” mass shootings occur in places that are not accessible to outsiders (e.g., military bases), undermining the “it could happen to anyone” logic often used to justify the focus on these incidents.

Including deaths from all weapons avoids similar issues for cases where multiple weapons are used, or the deaths are caused indirectly (e.g., jumping out of a window or being trampled while attempting to escape a mass shooting). In 2021, for example, a 30-year-old man in North Bend, OR, fatally shot two victims (including his father), stole his dad’s truck, and then drove over and killed two pedestrians.

While we recommend including all types of incidents, we acknowledge the unique offender, victim, and incident traits that characterize distinct types of mass shootings. Accordingly, we recommend that researchers disaggregate by type (e.g., familicides vs. felony killings vs. public massacres; shootings vs. other methods) rather than focusing exclusively on one subset of cases (e.g., public mass shootings).

Establishing a Time Frame

When FBI profilers originally distinguished among mass, serial, and spree murders, they identified two primary differentiating criteria: the number of victims and length of time between kills (Ressler et al., 1988). Although every definition of mass shooting includes some sort of victim threshold, many fail to establish a concrete time frame within which the incident must occur. Generally speaking, mass murders occur in “a single incident,” while serial and spree cases typically take place in multiple locations (the former separated by long “cooling off” periods, and the latter occurring in relatively quick succession over a few days). In the spirit of the original, we recommend defining mass shootings as incidents that occur within a 24-hour period, regardless of the number of physical locations. This would include, for example, the March 2021 killing of eight victims at three Metro Atlanta day spas during a time span of just over two hours. Also included would be rare cases in which an offender began or ended a multi-day killing spree with a mass shooting. In such instances, we recommend counting all of the victims (as long as four were killed within twenty-four hours). Although somewhat arbitrary, this choice of time frame is unambiguous, easy to apply, and avoids confusion with other forms of multiple homicide.

Consistency is Key

Regardless of which definition is chosen, it should be applied consistently to all cases over time to ensure the accuracy of prevalence estimates. Unfortunately, some databases include cases that violate their own inclusion criteria, especially if the incident is deemed particularly infamous. Mother Jones, for example, explicitly limits cases to lone shooters (Follman et al., 2022), yet includes the 2015 San Bernardino, CA, massacre, the 1999 Columbine High School shooting, and the 1998 Westside Middle School shooting, all of which involved two perpetrators. More problematic, Mother Jones changed its minimum victim threshold from four to three fatalities in accordance with the U.S. Congress’ Assistance for Violent Crimes Act of 2012 (which somewhat arbitrarily defined mass shootings as three or more victims killed) for all cases occurring in 2013 or later. They did not, however, retroactively update the incident counts for earlier years, leading to a dramatic and artificial increase in the number of mass shootings in recent years. Although Mother Jones makes note of the modified definition, individuals who download the data have sometimes failed to recognize the change and derive erroneous conclusions about the trend in mass shootings (see, for example, Reynolds, 2018).

Our Recommendation

 As indicated, we adopt the conventional definition of four or more victims killed. The broader notion of four victims injured or killed unjustifiably conflates injury with death, the most serious and permanent of outcomes. The narrower notion of four or more killed in a public setting unrelated to an ongoing criminal enterprise often by a stranger unfairly demeans the deaths of victims who are killed in private settings, who are slain by assailants known to them, or who are targeted by rivals in gang warfare or the illicit drug trade. Given the above arguments, we crafted this definition used in the Associated Press/USA TODAY/Northeastern University Mass Killing Database:

Mass murder is defined as the intentional killing of four or more victims— excluding the deaths of unborn children and the offender(s)—by any means within a 24-hour period. This definition includes cases involving all weapons (e.g., shooting, blunt force, stabbing, explosives, etc.), types (e.g., public, felony-related, and familicides), motivations (e.g., domestic dispute, profit, revenge, terrorism, hate, etc.), victim-offender relationships (e.g., stranger, family, acquaintance, co-worker, etc.), and number of locations. The time frame of 24 hours was chosen to eliminate conflation with spree killers who kill multiple victims over several days in different locations, and to satisfy the traditional requirement of occurring in a “single incident,” even if that incident involves an offender targeting multiple locations in an extended assault but within a relatively short time span. However, offenders who kill four or more victims during any 24-hour period of time as part of a multi-day spree are included, as are all their victims within seven days of the mass killing. Negligent homicides related to driving under the influence or accidental fires are excluded due to the lack of intent. Finally, consistent with the traditional definition, mass shootings are those mass killings (four or victim fatalities) in which most or all the victims are killed by gunfire.

It is important to acknowledge, however, that any definition is somewhat arbitrary and, thus, imperfect. In some cases, whether an incident qualifies as a mass killing is simply a matter of timing, opportunity, or the assailant’s skill. Public shooters try to gun down as many people as possible, but their aim may be poor, or first responders and ambulances may arrive in time to intervene and save lives. Also, some family annihilators are not mass murderers because they did not have enough children to kill. These caveats notwithstanding, we contend that this definition is at least unambiguous, easy to apply, and avoids subjective judgement calls. Regardless of whether one agrees with our approach, it is important to be aware of which definition is being used when interpreting study results and other statistics.

Based on our definition articulated above, Figures 1 and 2 display trends in the number of mass shootings and mass shooting victim fatalities from 2006 through 2021 by type of incident. Contrary to the widely held public perception of a growing epidemic, when it comes to the more serious shootings in which at least four victims are killed, the incident counts have not changed much over the past 15 years, except for a modest increase in public mass shootings. Even then, the increase is rather small in terms of absolute number—from an average of 4.6 incidents per year for 2006-2015 up to 7.0 for 2016-2021. There also has not been much change in the total number killed, except for an increase in public mass shooting victimization, which is largely the result of a few cases with unusually large death tolls: 49 killed in Orlando, FL, in 2016; 25 killed in Sutherland Springs, TX, and 60 killed in Las Vegas, NV, both in 2017; 17 killed in Parkland, FL, in 2018; and 23 killed in El Paso, TX, in 2019. 

(Click table to enlarge)

(Click table to enlarge)


Regardless of definition, mass shooting deaths constitute a small share of the annual number of gun homicides in the U.S. (five percent using Definition #1, one percent with Definition #2, and less than half of one percent by Definition #3). Despite these small shares, as many as one-quarter of Americans believe that mass shootings are responsible for the most gun fatalities—more than suicide, accidental shootings, and homicides other than mass shootings (APM Research Lab, 2019). 

Survey after survey has found disturbingly high levels of fear associated with mass shootings. Nearly half of Americans report being worried that they or a family member will fall victim to a mass shooting (Brenan, 2019), and one-third say they avoid certain public places because of the threat of a mass shooting (American Psychological Association, 2019). These exaggerated perceptions may in part be the result of frequent media reporting on broad definitions of mass shootings, such as the intent to kill multiple victims (as in the FBI’s active shooter data) or counting all victims regardless of the extent of injury (as in the Gun Violence Archive). Many politicians and pundits have characterized mass shootings as an epidemic. That certainly may be true with regard to public fear, but in terms of the actual incidence, such suggestions do not appear to be supported by the facts—regardless of how those facts are measured.

No potential conflicts of interest were reported by the authors.


American Psychological Association. (2019, August 15). One-third of US adults say fear of mass shootings prevents them from going to certain places or events.

APM Research Lab (2019, October 2). Knowledge of gun-related deaths.

Blair, J. P., & Schweit, K. W. (2014). A study of active shooter incidents in the United States Between 2000 and 2013. Texas State University and Federal Bureau of Investigation, U.S. Department of Justice, Washington D.C.

Brenan, M. (2019, September 10). Nearly half in U.S. fear being the victim of a mass shooting. Gallup.

Carroll, N (2021, May 28). The backstory: We’ve been tracking mass killings since 2006. USA TODAY.

Federal Bureau of Investigation. (2022). Active shooter incidents in the United States in 2021. U.S. Department of Justice, Washington, D.C.

Follman, M., Aronsen, G., & Pan, D. (2022, June 17). A guide to mass shootings in America. Mother Jones.

Fox, J. A., & Levin, J. (2015). Mass confusion concerning mass murder. The Criminologist, 40(1), 8-11.

Fox, J. A., & Levin, J. (2022). Mass murder in America: Trends, characteristics, explanations, and policy response. Homicide Studies,26(1), 27-46.

Fridel, E. E. (2021). A multivariate comparison of family, felony, and public mass murders in the United States. Journal of Interpersonal Violence, 36(3-4), 1092-1118.

Holt, L., & Gosk, S. (2018, November 8). Mass shootings becoming painful new normal in America. NBC Nightly News.

Lott, J. R. (2015). The FBI’s misrepresentation of the change in mass public shootings. Academy of Criminal Justice Sciences Today, 40(2), 18-29.

Ressler, R. K., Burgess, A. W., & Douglas, J. E. (1988). Sexual homicide: Patterns and motives. Lexington Books.

Reynolds, A. (2018, February 15). Are mass shootings becoming more frequent? The Cato Institute.

Schmidt, M. S. (2014, September 24). F.B.I. confirms a sharp rise in mass shootings since 2000. The New York Times.

Silverstein, J. (2020, January 2). There were more mass shootings than days in 2019. CBS News.

Victor, D., & Taylor, D. B. (2022, May 17). A partial list of mass shootings in the United States in 2021. The New York Times.

About the Authors

James Alan Fox is the Lipman Family Professor of Criminology, Law, and Public Policy at Northeastern University. His four decades of research on mass killing has been presented in several books, including Extreme Killing: Understanding Serial and Mass Murder, and dozens of articles in both academic journals and popular outlets. Finally, he is one of the principals in developing the Associated Press/USA TODAY/Northeastern University Mass Killing Database.

Emma E. Fridel is an assistant professor in the College of Criminology and Criminal Justice at Florida State University. She primarily studies violence and aggression with a focus on homicide, including homicide–suicide, serial and mass murder, gun violence, and police use of lethal violence. Her work has been published in Criminology, Social Forces, the Journal of Quantitative Criminology, and Justice Quarterly. She is also co-author of Extreme Killing: Understanding Serial and Mass Murder.

CITATION (APA 7th Edition) Fox, J. A., & Fridel, E. E. (2022). Keeping with tradition: Preference for the longstanding definition of mass shooting. Journal of Mass Violence Research, 1(2), 17-26.

Download an infographic summarizing the main takeaways of this article.
Available Formats: PDF  |  PNG