Violence in America (2022 Edition)

Spread the love

TLDRUpFront: The 2022 update on the historical patterns and current dynamics of violence in the United States. No policy arguments. Only data: 22 charts on 13 topics from credible sources. Includes notes on methods and sources. Bookmark the post and share the post, or individual charts from it, as may be useful in the coming year.

Criminal Homicides per Capita 1960-2020. Source FBI UCR Data.


In 2016 I thought I’d be doing this once a year.  The goal was to create a semi-regular update people could bookmark and refer back to, with charts to share to improve media literacy on the dynamics of violence in the United States. Six years, a completed Ph.D. and one global pandemic later I’m getting around to that next update of 13 trends in violence.

1 Criminal Homicides 1960-2020

2 Historical Homicide Trends 1900-2020

3 Murder by Means: Distribution & Trends in Weapon Used in Homicides (1996-2020)

4 Justifiable Homicides vs. WAPO Fatal Police Killings 2015-2020 (NEW)

5 Fatalities of Students at Schools: Homicides & Suicides 1992-2019 (NEW)

6 Mass Shootings (4+ Killed) 1966-2021 (NEW)                  

7 Mass Shooting by Means: Distribution & Trends in Weapons Used in Mass Shootings (NEW)

8 All Related Deaths: Firearms, Drug Overdoses, & Alcohol Use 1999-2020 (NEW)

9 Aggravated Assaults 1960-2020

10 Rapes (Legacy & Revised Definitions) 1960-2020

11 Robberies

12 Arsons 1995-2020 (NEW)

13 Terrorism (1995-2021) (NEW – pending)





1- Criminal Homicides 1960-2020

Overall criminal violence last peaked in the United States in 1991/1992 and across all forms: homicides, aggravated assaults, rapes, violent robberies, and arsons has been declining gradually since. The per capita number is used as the population of the US has grown from 265M to 332M over the period of this report. I’ve added economic recessionary periods, from the NBER.  Without a doubt 2020 showed a large uptick in violence, meaning I blew my prediction at the start of the pandemic that violence would remain at or below the last five years.

Criminal Homicides per Capita 1960-2020. Source FBI UCR Data.

As the chart shows the increase in homicides began around 2015. But homicides in 2020 showed a large spike up and roared past the last five years reaching back to levels of violence not seen since 1998. We see this same decline-reversal leading into an increase in some other forms of crime too, such as aggravated assaults,  but not in others such as violent robberies, rapes, or arsons (discussed below.) Note that the criminal homicide data excludes the deaths of the 9/11 and Oklahoma City Bombing terrorist attacks. FBI UCR data excludes by definition suicides from criminal homicides.


Sources: [1], [2], [3]



2- Historical Homicide Trends 1900-2020

A new addition this year is something I’ve always wanted to do. I’ve seen long term charts floating around for years, but never had access to the tabular data or time to reconstruct one myself. This chart combines two sources of data, the FBI UCR for 1933 forward[1]. But for 1900 to 1933 I’m relying on econometric estimates using US Census Data put together by a researcher in 1985[2].

Sources: [4], [5]

3- Murder by Means: Distribution & Trends in Weapon Used in Homicides (1996-2020)

The UCR includes data on what was used to commit criminal homicides and this represented below in three ways. First, two charts show the breakdown of homicides in 2020 by the top contributors and then everything else. This is done for clarity as the boxes of “all other” would be too small to read on the first chart. First here are the major contributors.

If you add these up you’ll get to 5.4 Homicides per 100,000, which is 83% of the 6.5 2020 Homicide Rate. If this sounds like an inequality distribution or patero principle, don’t forget to click on the YouTube video below on the Coastline of Violence in America where I go into detail on these kind of inequality fractal patterns. What’s left is everything else that fits into that 17% remaining.


Kudos to defenestration for making the list this year, as it was not a method in 2016.

What’s clear from these 2020 charts is that the handun, whether a pistol or revolver, is doing the yeoman’s work of bloodletting in the United States, and it has for a very long time.  In the next chart the top seven means of murder are charted over time, 1996-2020 and handguns have been undisputably the highest means of murder throughout this time.


This is surprising to some as the evidence appears counterintuitive to what we might expect. People think the overall lethality or effectiveness of the weapon would be paramount inc ommitting a crime. But they, not being criminals, are not thinking like criminals. When it comes to crime the prevalence of a particular weapon in a criminal homicide has more to do with other factors including accessibility, concealability, cost, and how easy is it to dispose of. In this regard the lowly handgun, small, cheap, easily concealed or disappeared checks all the boxes.

Also I want to repeat and remphasize a note I first made 2016 about the unexplained growth of “Other Firearms” over the years. In 2016 this category was split between “Unstated”  and “Other” but since the modernization the data has now been merged. I’m not sure what kind of “Other” firearm there is that isn’t a handgun, rifle, or shotgun (blunderbuss perhaps?) If we’re being charitable, this field of “Other Firearms” is because an officer reporting a crime either doesn’t state the type because they weren’t told or they don’t know.  But why then would that field be growing over time? To a data analyst, seeing a very clear growth rate over 20 years speaks to a different conclusion. We can see from the data that use of handguns is declining over the same period of time, but that decline isn’t represented by an uptick in any other weapon. My concern here – of which I admit I have no proof – is that meaningful data on the type of firearm used in a crime is not being entered correctly for ideological reasons. The hypothesis there is that  police officers, who tend to be pro 2nd Amendment, are not reporting assault rifles used in crimes under the rifle category because they do not wish to see them banned. That’s quite the stretch however and one immediate problem with the hypothesis is that although that may explain local discrepancies, it doesn’t really explain a nationwide problem. Other hypotheses are less flashy, but more plausible: for lack of sufficient resources invested in training how to report correctly the quality of the data is eroding.  Of course the problem is almost no one thinks to investigate the “Other” category of a pareto chart – so I haven’t seen a lot of research on what might be causing this and I personally haven’t had the time to dig into state by state, or even city by city, reporting data to see if this effect is truly nationwide, or being generated by a few departments. If a Mulleteer has particular insights on this category from a reporting standpoint that might explain the growth I’m all ears. For now though it’s just “interesting” to see that increase in what is essentially a “we don’t know” category that matches the decline of the dominant means of murder.

Sources: [6]

4- Justifiable Homicides vs. WAPO Fatal Police Killings 2015-2020 (NEW)

In 2016 the national concern over unjustified police killings was already strong given both protests and national media coverage of two Blacks killed by police: Michael Brown of Ferguson, MO and Eric Garner of New York City, NY. However national-level credible data was limited. Today however we have the beginnings of a track record of credible national data. The Washington Post began an effort in 2015 to track via open-source every police-involved killing in 2015.  Its methodology is robust, transparent, and detailed. In 2016 the FBI UCR began reporting on justified homicides; both by police officers and private citizens. However, given the caveats of FBI UCR and the inherent conflict of interest in reporting, I’ve included both FBI UCR data and WAPO in the below charts.  However I’m just including “fatal police killings” from the WAPO data. Even though the WAPO database goes further to identify criteria, such as whether the person killed had or didn’t have a weapon, was combatitive at the scene or fleeing, and make a determination of “justified” and “unjustified” I’m not including that data here. Combat circumstances are too murky to know whether someone constitutes a threat or not simply on whether they are armed with a firearm or not. Likewise, given the paucity of detailed reporting in many cases – WAPO may be understating the total of unjustified police homicides. So I simply present the total here.


Sources: [7] & [8]

5- Fatalities of Students at Schools: Homicides & Suicides 1992-2019 (NEW)

With a recent mass-shooting in Uvalde, TX concerns about the safety of children at school grades K-12 is running high.  But this is an area where the cognitive biases (including affect heuristic & prevalence induced concept change) can give us a very different perception of violence.  The affect heuristic means we over-emphasize risks that are scary versus risks that may be more likely, but aren’t scary. And a prevalence-induced concept change means that as a risk begins declining, we look harder for it. Combine this with the mass media attention received anytime a mass shooting occurs at a school – and its understandable why parents are concerned.

However, as the data shows – the overall trend in fatalities from crime per capita of students at school is declining 1992-2019.

Students were far more likely to be killed in the 1980’s and 1990’s prior to Columbine. The Columbine-cliff resulted from nation wide attention and introduction of many security procedures that were already common in inner-city schools (where violence had historically been higher). Fatalities at school from crime remains low, generally declining over the long term trend. Where it spikes, is largely the result of specific ASE mass-shootings that occurred at schools:   Red Lake Highschool in 2005-2006; Sandy Hook in 2011-2012; and Stoneman Marjory Douglas in the 2017-2018 school years respectively.  (Disclosure the VA Tech mass shooting with 32 fatalities in 2007 is not included in this data because it only looks at K-12 Schools.)

But a per capita value expressed in the 2nd decimal to the right of zero isn’t easy to understand. Below I also show the actual fatalities chart below.



The average trend over the last 10 years of the data is to decline from ~40 to ~30 students killed nationwide on school grounds in grades K-12. Suicides committed at school by comparison have remained relatively flat over this period. A key caveat of course is that this data is only showing the risk to school-aged children while they are at school. If there is interest in future updates I can add further breakdown of overall homicide and suicide risk for children ages 18 and under.

Putting this data together wasn’t easy, I stitched together National Council of Educational Statistics (NCES) Data from Tables 201.10, 203.10, 205.60, 228.10 to determine population of students (public and private) K-12 nationwide, aged 5-18.  These charts focus only on student fatalities so exclude instructional, administrative, and facilities staff from both population and fatalities. Student Population Data for 1992-1995 is interpolated. The current data for fatalities on student grounds ends at the 2018-2019 school year (Table 228.10)[14].

Sources: [9]

Mass Shootings (4+ Killed) 1966-2021 (NEW)    

Going hand-in-hand with a concern of the safety of children at school is an ongoing concern of mass shootings. This actually falls in my area of research and I have written several articles on this recently. However, although many websites purport to have data on mass shootings, few have very good data from a researcher’s perspective. That changed with the Violence Project’s release, and regular update of a mass-shooting dataset that extends historically back to 1966. This is truly a wonderful resource that gets the InfoMullet gold star, and my appreciation, for the work they have put into this. Unlike most data sets which are just trying to record that a shooting happened, and provide minimal useful information, the Violence Project goes into dozens of parameters and variables for each mass shooting. The kind of data-rich incident analysis we’ve come to expect out of similar high-quality projects like the University of Maryland’s Global Terrorism Datasest. Advanced techniques such as fractal segmentation or hyperdimensional analysis (see below) is impossible without this kind of data. So hats off to the team at the Violence Project and you should check them out.

A few caveats before the charts. What constitutes a mass-shooting is a big debate and I’ve seen sources defined as any shooting that kills or wounds 3 or more people besides the perpetrator, any shooting that kills 3 or more people, and any shooting that kills 4 or more people. There’s also different kinds of mass shooting including those that arise from criminal gang violence, domestic violence, “public” Active Shooter Events (ASE), and workplace violence. Although it appears that the Violence Project’s data set may be able to be sorted into these categories, I didn’t have the time to dig in and create that segmentation, so instead I’m just presenting the high level data.

NOTE: Due to their statistical rarity mass shooting per capita are measured on a scale of 1 per 1,000,0000 rather than 1 per 100,000 as most other crime is and even then the actual measure usually is in tenths of a decimal.

In the first chart, I plot the incident frequency of mass shootings where four or more people, not including the perpetrator were killed from 1966-2021.



The frequency chart shows a gradually increasing trend over time in mass shootings from the 1970’s. The data is also very noisy as frequency increases and decreases. I wouldn’t read too much into the spikes on this chart as being indicative of anything other than that noise. This chart, frequency, simply shows how many incidents occured where 4+ were killed. The next chart plots the fatalities per capita over time, but remember the scale is 1 per 1,000,000.


As we’d expect, as frequency of incident rates rise so to do fatalities given a minimum threshold of 4+ fatalities to make this list. rises fatalities also rise. Unliked incident frequency where spikes may be noise, the spikes in this graph are causally associated with high fatality mass shootings: San Ysidrio McDonald’s Massacare in 1984, Lubby, TX in 1991, Columbine  in 1999 and the 2017 Las Vegas Casino shooting, which remains the deadliest shooting on record in the United States  and within the top ten of the world, killing 59 and wounding several hundred.  These individual events drive the spikes up and influence the trendline over time. However, the directional data indicates growing fatalities over time, even those fatalities are unequally distributed and cluster around periodic major events.  One question about mass shootings is to what extent they are associated with underlying rise and fall in criminal homicide patterns. These incidents are too infrequent to show up in the per capita incident rates of homicides, but we can chart the percentage of all criminal homicides attributable to FBI UCR homicides for a given year, which is the next chart.


Contrary to the rise and fall of general crime cycles in homicides, mass shootings show a clearly increasing contribution as a percentage of all homicides. But it’s worth keeping in mind what an overall small percentage this is. Again, spikes in the data correlate to specific incidents, but even with 2017’s fatalities the overall amount mass shootings contributed to overall homicidal violence was less than 1%.

Up to this point, the focus has been on fatalities, which is definitional to the criteria the Violence Project used. One of the complaints against using 4 fatalities as the threshold is it may be obscuring injuries that might have been fatal in the past but due to advances in emergency response and trauma healthcare, the patient survives. It may also obscure shooter intents who wished to kill more people but failed too. I find this particular argument unconvincing as the typology of the kind of mass shooting matters more to determining fatality and casualty counts than just assuming anything with more than 4 people killed or injured is the kind of mass shooting that most people worry about. The four main types of mass shootings that generate most casualties and fatalities: criminal gang violence, domestic violence, workplace violence, and ASE’s all have a typology and ratio between fatalities and wounded based on the circumstance the attack is conducted in.  For example criminal gang violence usually has fewer fatalities but higher wounded; domestic violence higher fatalities with fewer wounded, ASE’s high fatalities and high wounded.  Within ASE’s the specific template method employed, and the learning curve by which later mass shooters learn from previous mass shooters has a telling effect as well, clearly shown below charting total casualties (killed & wounded) using data from the Violence Project.


The giant spike on the right correlates with the 2017 Las Vegas Casino. In this incident a perpetrator studied previous mass shootings but also spent a million or more dollars acquiring equipment, traveling and scouting locations, and planning his template method. As a result, its unlikely to see that template method as replicated as the Columbine style school shooting or Great Replacement theory white nationalist template method; both involve mass shooting, and both are cheaper and easier to replicate than Las Vegas.

Sources: [10]

Mass Shooting by Means: Distribution & Trends in Weapons Used in Mass Shootings (NEW)

Speaking of templates, a key part of my research into the terror contagion hypothesis, is focused on studying the modus operandi, or method of crime, for a mass violent terrorist event. Part of that is studying the equipment in weapons that attackers bring on incidents. Most people probably imagine a mass shooting occuring with a single, so-called assault rifle. But just like typological differences show up in fatalities, so to do they show up in weapons. Most criminal gang violence and domestic violence may rely on a single weapon. Whereas in an ASE the attackers often bring multiple weapons. The first chart looks at the average number of weapons used in a mass-shooting.


Here is where the work of the Violence Project pays off, and why I prefer them over something like GVA or other, less rigorous data sets. In many mass shootings a perpetrator may bring more weapons to an incident than they use. Often some are left in the vehicle to assist in a get away that never happens, or they are carried for contingencies that don’t arise. One contignency frequently arising, especially in ASE’s, is the suicide of the perpetrator.  The Violence Project tracked down individual police reports to determine which firearms were used rather than being left behind and then also marked which firearms were only used for the suicide of the perpetrator. The result of this work, which I’ll be incorporating into future research, is a much clearer view of the load-out is for shooters who conduct these kind of mass casualty attacks. Also again, the outlier of 2017 Las Vegas Casino attacks is skewing the trendlines, as that perpetrator brought 24 firearms into his hotel suite over a series of days.


But what kinds of firearms are mass shooters using? I use a similar approach to Murder by Means above in the next two charts, but this time exclusively to mass shootings and over the entire time horizon studied by the Violence Project, 1966-2021.


Once again, the handgun is dominant, but astute readers will see Assault Weapons come in second and may ask by what criteria the Violence Project identified these. From their own codebook the Violence Project describes the breakdown of above categories: “A handgun has a short barrel; a shotgun has a long barrel and usually has a smooth bore; a rifle has a long barrel with rifling, which puts spin on the bullet, increasing accuracy and distance; an assault weapon is any semi-automatic gun that can accept a detachable ammunition magazine that has one or more additional features considered useful in military and criminal applications but unnecessary for sports or self-defense, such as a folding, telescoping or thumbhole rifle stock. This is consistent with the Federal Assault Weapons Ban of 1994.”

But the percentage of which firearms is used in mass shootings is not the complete picture.  Remember that the incident frequency of mass shootings is a tenth of a decimal of a one per million per capita measure. For clarity, in the chart below, I list the actual number by weapon type used in mass shootings by 1966 – 2021.

In other words, only 263 firearms out of the hundreds of millions in the US have been used to perpetuate the lion’s share of mass shooting incidents, fatalities, and casualties.  Specific to commonly debated weapons, such as the so-called assault weapon (see criteria of inclusion above) that number drops to only 64.

Sources: [2] & [10]

All Related Deaths: Firearms, Drug Overdoses, & Alcohol Use 1999-2020 (NEW)

Shifting away from criminal homicides, by everyday violence or mass shootings, this year I’ve added an “All Related Deaths” charted across three different types of ‘items.’  An “item” in this sense is a broad category of a physical thing that fits a similar criteria: their use splits the population on sides of for and against, they cause a lot of deaths, and they have varying degrees of legislation seeking to prohibit their misuse.  The three items are firearms, drugs, and alcohol, and all related deaths are inclusive of crime, suicide, accidents etc.

As a sourcing note, this data comes from the CDC WONDER and NIH NIDA datasets. This is because not all of these deaths are captured in UCR Criminal Data. For example, all-related deaths per capita of firearms is close to double in this chart than the Criminal Homicides (the majority of which are committed with firearms) above. That is largely because of suicides using firearms. It also includes Accidental Death & Discharge (ADD) but those only amount to a few hundred a year and are a small contributor.

In the chart below I plot the actual related deaths as a number, rather than per capita, over the same time frame.

These three also present a useful continuum for potential policy analysis, as the item in question is regulated to different degrees. At one extreme are illegal drugs, which are almost universally banned to possess in any context.  On the other extreme is alcohol which, other than an age requirement, are generally not legislated to prohibit possession outside of risky-behavior specific contexts (e.g. driving under the influence or carrying open containers in a car.) In the middle are firearms which constitute of a mix of laws, varying greatly by state.

Sources: [11], [12]

Aggravated Assaults 1960-2020

Next up is aggravated assaults, criminal violence that is significant but doesn’t result in death.  Like criminal homicides this has been increasing gradually since 2015 but saw a large spike upwards in 2020.

Sources: [2]

Rapes (Legacy & Revised Definitions) 1960-2020

The FBI revised it’s reporting procedures on how rape was calculated in 2013.  This was a clear case of the NCVS data diverging from UCR given the known reluctance of victims of sexual assault and rape to wish to interact with police. And a culture of police departments in the past not taking these claims seriously. Both the legacy data, extending back to 1960, and the revised methods are presented below by way of comparison, with the FBI ceasing reporting of legacy data in 2016.


Unlike homicides and aggravated assaults, rapes sharply decreased in 2020 despite a similar increasing rise from 2014 onwards. Obviously the specific role pandemic lockdowns played in this effect is something many are going to be looking into – but its still too early to know for sure. Also of interest will be whether this downward trend continues into 2021 and beyond, or revrses as the pandemic lockdowns eased moving into 2022.

Sources: [2]


The chart below shows a forty-year behavior pattern of violent robbery. This is the final component of the “big four” that constitute violent crime including homicides, aggravated assaults, and rapes. The chart clearly shows the peaking of “big four” violence in 1991/1992 followed by a long decline into the 21st Century.

What’s interesting is how that decline continues for violent robberies whereas the other three: homicides, aggravated assaults, and rapes showed increases beginning in 2014/2015 through 2019. Homicides and aggravated assaults showed sharp increases in 2020 whereas rapes, potentially pandemic related, declined. In contrast to all this oscillation, violent robberies have continued a steady persistent decline in frequency, continuing to decline even in the pandemic.

Sources: [2]

Arsons 1995-2020 (NEW)

With increased attention to the role of arsons in 2020 I thought it might be useful to add a long-term chart of arsons per capita from 1995-2020 using the FBI UCR data.

What’s interesting is that although 2020 did show a spike in arsons, and a reversal of the gradual decline, it was a much less steep spike than seen in homicides or aggravated assaults. Where we’d have to go back into the 1990’s to find similar levels of violence in homicides, for arsons the spike only takes us back to, well, about 2015-2016. Which doesn’t really seem all that bad in retrospect, especially in comparison to previous time periods, such as 2000 where the arson rate was more than double what it is now.

Sources: [2] & [13]

Terrorism (1995-2021) (NEW – pending August)



State Level Segmentation of Homicides (REMOVED & REPLACED BY COASTLINE OF VIOLENCE)

The charts above are very simple “time series” data of a single or few dimensions plotted on the y-axis over time on the x-axis. The dimensions are of a crude fidelity, presenting a national number. This is a consequence of FBI UCR reporting data, because as we had dimensions and fidelity to our analysis, the shape and nature of violence begins to change before our eyes. In the 2016 edition, I demonstrated this with a very crude state-level fractal segmentation. But since then I’ve put up ~60minute special presentation on the “Coastline of American Violence” that demonstrates many dimensions sometimes up to the fidelity of a street-level view.



Notes on Methods & Sources

FBI Uniform Crime Reports (UCR)

The People are represented by two separate, yet equally important government criminal statistics reporting entities. The FBI which compiles Uniform Crime Report (UCR) and the Bureau of Justice Statistics whom conducts the National Crime Victims Survey (NCVS). These are the stories – dong dong.

Well – really only one story. I rely on the FBI UCR over the NCVS in these updates. The FBI UCR has been around since 1929 and publishing since 1930. The UCR also reports actual incident counts as reported by police departments, whereas NCVS relies on inferential statistics derived from a sample of ~75,000 surveys made across the population. In this way the UCR represents an institutional count of reported crimes, while NCVS tries to capture victim experience in a broader way. Of these two I prefer incident counts – but with the acknowledgement the data is only as good as what is reported to police and what the police reports to the FBI. This voluntary reporting regime and reliance on police reports, rather than victim experiences, are the biggest weaknesses of the UCR. Victims may be reluctant or unable to report incidents to police, nor is it always assured that police are providing all the cases that they are involved in.  For example, for years police under-reported their own violent encounters with citizens and what data they did report could be considered compromised because of the conflict of interest in how they present themselves. That’s one reason this year’s update on officer-involved Justifiable Homicides reported through the UCR is compared against the Washington Post Fatal Police Killings data sets.  Generally the UCR is a good data source, but one has to be careful to understand these nuances and they will be called out where appropriate.

One reason I held off on making this update is the UCR itself underwent a major modernization effort over the last few years. One part of the modernization was updating the front-end user interface to make it more “accessible.” This is a trend in websites that reminds me of all the workplaces shifting to “open spaces” over the last decade – it looks pretty until you actually try to use it then it’s just a pain in the proverbial. Where the data is located changed during the update, and it’s now harder to find unless you really know what you’re looking for and where to look.[1]

While that’s annoying – the second part of the modernization represents a real risk to the data integrity. Undernath the hood of the UCR is a computer system police departments use to report their incidents on. In the past this was the “Summary Reporting System” (SRS) but modernization rolled out a newer National Incident Based Reporting System (NBIRS). The NBIRS has more reportable data fields for incidents, which is a good thing. But like all IT ‘enterprise’-wide updates, roll-out has been less than smooth. The new systems have new fields, which means new trainings, and not all agencies are using the NBIRS system. This creates a risk of inconsistent and under-reporting. Worse, different parts of the FBI’s own site point to SRS and NBIRS reported data, and this can create big confusion.  For example when I put this together the pages Crime-> Homicides doesn’t match the data on Crime -> Expanded Homicide data.


Parameter Page: Crime -> Homicides Page: Crime -> Expanded Homicide Difference (SRS-NBIRS)
Per Capita Murder Rate 2020: 6.5 2020: 5.4 1.1 Murders per 100,000
Actual Murders 2020: ~21,450 2020: 17,815 ~3,635


It also means whomever programmed their interface goofed. When you click on Crime –> Homicides the chart generator is pulling from SRS and if you click on Crime –> Expanded Homicides the chart generator is pulling from NBIRS. Looking at the wrong source could give an undercounting of homicides by over 3,500.

What is shown here is only the tip of the iceberg in UCR information. With both historical SRS and the new NBIRS there’s many additional crimes, and details about crimes, I don’t cover here including race or gender of the victims or the criminal, and where the crime occurred for example.


Per Capita vs. Actuals 

Most of the charts data in both per capita and actuals. We use per capita because it normalizes for changing populations.  Long-term violence reporting that doesn’t utilize per capita is at risk of misstating the relative frequency and severity of violence. If 10,000 people experience 1 crime and 100,000 people experience 10 crimes, that simply means the population has grown while the relative frequency of that particular crime has stayed constant. In most charts the per capita amount is 1 per 100,000 which is standard for criminal reporting. However, when it comes to exceptionally rare acts, such as mass shootings, the conventions are 1 per 1,000,000.

Fractal Segmentation/Hyperdimensional Analysis

Violence in mathematical terms is a “highly differentiated system” which means that it’s occurrence is not evenly spread throughout the population. One method to display this high differentiation is with fractal segmentation or hyperdimensional statistics. In 2016 I made a crude effort at this by breaking down murder rates by states. But since then I’ve released the “Coastline of Violence in America.” This special presentation is over an hour long and goes through a full description of fractal segmentation on homicides. I walk through fractal segmentation from country descending to state, county, city, neighborhood and street levels to show how differentiation changes the nature and the scope of the problem. The presentation doesn’t just cover geography – but other factors often used in analysis of homicides including day, circumstance of relationship between perpetrator and victims. Rather than try and reproduce that here I’ll point you to the link and encourage you to go check it out.


Comparison to Economic Recessions or Other Periods

This year I’ve added economic recessions in light blue to many of the charts.  I did this for a couple of reasons. First, there is a persistent misunderstanding among Americans that connects the level of economic prosperity to levels of violence.  The belief goes that as the economy tanks, crime goes up. And as the economy does better, crime goes down. I suspect this is rooted in America’s puritanical cultural traditions, which equates ethics with economics. If one is prosperous, according to this theory, one does not consider crimes. When one is poor and desperate, they’re more likely to commit crime. But this turns out not to be the case.  Over the long arc significant research has established that there’s very little correlation between economic recessions and rising criminal violence. This is most notable during both the Great Depression (which can be seen in the historical homicides chart) and the Great Recession, in most other charts. Although it is not universal, violence seems to increase as economic activity increases, then reverse and begin declining just prior to a recession. The decline often then extends through the official recession into the longer recovery period. I talk more about the potential causal mechanisms and research supporting that in Ashes to Ashes, one of a three part series I wrote write as we entered the initial periods of pandemic and uncertainty was high.  Of course past behavior is not always indicative of future behavior, in that forecast I suggested an economic slowdown associated with the pandemic would mean a limited increase of violence, and of course as we can see that was really wrong. However, I did also predict that specific lockdown conditions of the pandemic and/or unjustified police killings that sparked dictator dillemma street protests may alter the normal course of behavior.  To the extent any of those causes, or something else we haven’t detected yet, is impacting overall violence levels is beyond the scope of this udpate. But by way of caution – it took us decades after the initial rise to unusually high levels of violence in the late 1970s, 80’s, and early 1990’s to identify a likely contributing cause in the great lead hypothesis[15]. We have to be open that these effects we’re seeing might be local causes, or some other systemic effect we haven’t detected yet, or a combination of the two. Time, and research, will tell.


Footnote Sources:

[1] SRS to NBIRS update for FBI UCR:

[2] For most FBI UCR data obtained through this page including SRS Estimated Crimes 1979-2020, Master File Download 2020 etc.

[3] All economic cycles were taken from NBER official dating here:

[4] For homicides, 1900-1932 see Eckberg’s 1995 work

[5] For homicides, 1933-current see FBI UCR data available in source [2].

[6] Murder by Means data assembled from multiple FBI UCR data as follows:

1996-2000 Data: Table 2.10
2001-2002 Table 2.10
2003 Table 2.12
2007-2011 Table 8

[7]  For justifiable homicides by LEO and private citizens see UCR(SRS) 2020 Expanded Homicide Tables 14 & 15

[8] For Washington Post Fatal Police Killings dataset see:

[9] See NCES Tables 201.10, 203.10, 205.60, 228.10

Excludes instructional, administrative, and facilities staff from both population and fatalities. Student Population Data is interpolated 1970-1980 for public schools and 1970-1995 for private schools.

[10] The Violence Project Database is the largest, most comprehensive source of information on the psychosocial histories of mass public shooters. Responsibility for the contents of the database lies solely with the Principal Investigators, professors Jillian Peterson (Hamline University) and James Densley (Metro State University).

Peterson, J., & Densley, J. (2022). The Violence Project database of mass shootings in the United States (Version 5).

[11] Drug overdoses:

[12] Alcohol related deaths:

Alcohol 1999-2017 DOI: 10.1111/acer.14239 &
Alcohol 2019-2020 &

[13] Arson Data comes from “Table 15 UCR” and “Table 2.31” in earlier (<2004?) versions






No Responses