# Unpacking the Decline in Firearm Violence

### What trends we should be looking at and what they imply for the future

Homicide is declining rapidly in the US. Beginning in 2022, homicides have declined from the highest levels in two decades to near the lowest level of the last 60 years. There is a lot of interest in uncovering why homicide has declined so precipitously and I’ve written quite a bit about that, here and here. In this essay, I want to talk about an adjacent issue: how firearms use has changed. What I’m interested in here is looking into whether firearms use in homicides has changed, and what it means for predicting future violence.

In this essay, I will argue that understanding changes in the proportion of homicides that are committed with a firearm is key to understanding both why homicide has declined and whether that decline is sustainable. There is a lot of skepticism about the current homicide decline. There is skepticism both about whether the decline is real, and whether it is sustainable.

I think much of this skepticism comes from a belief that homicide was trending up throughout the period from 2015 to 2019. As a result, it has widely been argued that the most reasonable expectation for the post-pandemic period is for homicide rates to return to that specific longer-term path. I certainly believe that long-term trends are critical to understanding what is happening today and predicting what will happen tomorrow. My argument is not that pre-pandemic trends are misleading in understanding current events.

Rather, we have been looking at the wrong trends for guidance. Here, I will argue that understanding the proportion of homicides that are gun homicides is *more* important than understanding trends in total homicides. I will argue that virtually all of the rise and fall in homicide between 2020 and 2024 is due to changes in that proportion.

Now, to make this argument, I would have to demonstrate that homicide committed in other ways—with knives, feet and fists, blunt instruments, etc.—does not have a material effect on homicide rates. At least not in the US. Further, by arguing that overall homicide rates are driven almost exclusively by firearm homicides I am implicitly arguing that there is little substitution of some other deadly instrument in place of a gun. I would like to make that argument explicit—the data suggest that if the firearm homicide rate declines, overall homicide will decline in the same proportion, meaning there is almost no substitution.

Before diving into the data to support these claims, I would like to point out that the number of firearm homicides is also important in its own right. It is often said that easy access to firearms can quickly change a fistfight into a gunfight and as a result, an assault becomes a homicide. Now, suppose that something in the world changed and firearms became less available, or people became less inclined to use them, all else being equal. What I’d like to show below is what happens to homicide overall when triggers are pulled less often.

There is some evidence already available. We know that nationally for the last 15 years or so, about three in four homicides are committed with a firearm. But it is also true that firearms are easier to obtain (legally and illegally) in some cities compared to other cities. As a result, the proportion of homicides committed with a firearm varies substantially across cities. And, there is an important correlation between homicide rates and the proportion of homicides with a fiream: in places like New York City, where the percentage of homicides with a firearm is much lower than in the average city, the overall rate of homicide is also much lower than in peer cities with a higher percentage of firearms homicides.

So, what happens when the rate of homicides with a firearm suddenly increases, or suddenly declines? The period around the pandemic provides data that lays out the implications very clearly.

It’s worth skimming the next few slides, there’s a pretty big payoff at the end!

## Recent Trends in Homicide

**Table 1. All Firearms Homicides, US, Monthly, 2018-2023**

First, a note on the data I used in this analysis. The data in **Table 1** through **Table 5** rely on monthly homicide data from the Centers for Disease Control. The data are drawn from death certificates, which show one (primary) underlying cause of death (UCD). Under injury intent, I selected ‘Homicide’ and under UCD Injury mechanism I selected ‘Firearm’. I made this data request August 25, 2024, and the data cover the period 2018 through February 2024. Data from early 2024 are labeled as (provisional and partial) so I don’t include them. The data for 2023 are labeled provisional, but since more than eight months has passed to allow any lagged death certificates to be included, I include these data. For** Table 6, **I use the CDC data series on homicides from 2009-2020, applying the same selection criteria.

What the data show is a rapid increase in the number of firearm homicides in the spring of 2020, peaking in the summer of 2020 and returning again to the same high levels in the summer of 2021. But the peaks are not as high in the summer of 2022, suggesting a modest decline in firearms homicides. By 2023 it is clear that there is a material decline in firearms homicides. The annual data support this conclusion.

**Table 2. All Firearms Homicides, US, Annual, 2018-2023**

In percentage terms, this shows a 25.4% increase in firearms homicides between 2019 and 2020, the largest one-year increase since 1960, and a 9.6% decrease from 2022 to 2023 which is as large a decline as has been recorded.

But hidden in these data is a thread I want to pull on. If you follow these numbers closely, you will note that the firearms homicide increase from 2019 to 2020 (25.4%) is a little smaller than the overall increase reported by the FBI (about 28%). The decline from 2022 to 2023 (9.6%) is also a little smaller than reported by the FBI (13.6%). Now, this may just be the kind of mild discrepancy that shows up when you compare two different datasets from different sources, but let’s pull on that thread a little and see where it takes us.

Next, let’s compare all homicides to firearms homicides over the 2018-2023 period.

**Table 3. All Homicides and all Firearms Homicides, US, Monthly, 2018-2023**

Here is where it gets really interesting. If you just use PROC EYEBALL (that’s a little SAS programmer joke) you might say, wow those two trends are identical! The ‘all’ homicide and the firearm homicide trends appear to move in lockstep, in every period, whether up or down. But we know that the percentages do not quite match up, so it must be that these two trends are not quite as identical as they appear.

And they are not. The key to understanding these trajectories is to understand what is happening in the gap between the two trend lines. The clue is this: if the two trends are identical, then the gap between the two trends should get a little bigger when the two trends are at higher levels, and a little smaller when the two trends are at lower levels. For the two lines to move identically, it would have to be true that firearms always make up exactly the same percentage of all homicides.

Here is what I mean. Earlier I told you that typically three out of four homicides (75%) are firearms homicide. So, what would Table 3 look like if 75% of homicides were committed with a firearm? In July, 2020, the peak of homicide in the series, there were about 2,500 total homicides and you would expect to see about 2,000 firearms homicides (75%) and 500 non-firearms homicides (25%). At a lower point in the series, November 2018, there were about 1,500 homicides, like there were in November 2018 and with a fixed proportion of homicides with a firearm, you would expect to see 1,125 firearms homicides (75%) and 375 non-firearm homicides (25%). If the two trends moved in lockstep, this fixed proportion would make the lines go up and down identically.

But you need only to glance at the data in **Table 3** to see that is not the case. The gap between the trend lines appears to be almost identical whether the trend is up or down. So let’s pull on the thread a little more and look specifically at the gap, which is filled with all of the non-firearm homicides.

**Table 4. All Non-Firearms Homicides, US, Monthly, 2018-2023**

Throughout this period, non-firearms homicides are virtually unchanged. There is no increase in non-firearm homicides when all homicides spike, and no decrease when all homicides decline. The number of non-firearm homicides stays within a fixed interval, regardless of the overall trajectory. The number of non-firearm homicides is extremely stable over time.

What is extremely unstable is the number of firearm homicides.

**Table 5. Percentage, Firearms Homicides, US, Monthly, 2018-2023**

The percentage of homicides with a firearm was relatively stable in the first part of the series and then increased by more than eight percentage points in the Spring and Summer of 2020. This is followed by a slight downward slope beginning in the middle of 2022, which matches the slight downward trend in all homicides at the end of this period.

The point of all of this is that 1) non-firearm homicide trends are flat and stable and 2) firearm homicide trends are volatile. Thus, changes in overall homicide between 2018 and 2023 are almost entirely explained by changes in firearm homicide.

**Solving for the Firearm Homicide Rate**

All of the above is about looking out the back of the boat to see where we have been. Now, let’s use that information here to figure out where we are and where we are going.

Saying this information is important to understanding where we are is probably a little bit of a jolt. Crime and justice data is lagging in the US. For example, official FBI national crime statistics for January 2023 have not yet been released. The CDC data I showed above is also lagged by 8-12 months—medical examiner and coroner data it takes about six months to get into the CDC WONDER data, and then another few months may pass before the data be considered complete, if provisional. That suggests we will not have death certificate data for homicides in August 2024 until the Spring of 2025.

But if we are in the midst of a historic crime decline, it is crucial that we stay ahead of the trend. We need to figure, as quickly as possible, what we are doing well so we can repeat and sustain it, and critically, not replace effective policies, programs, and practices with ineffective ones. As the old hiker says, the top priority when you are lost is not to get more lost.

To understand the trend and where we are on that trajectory, what we need to know right now is the current rate of firearm homicides. Since we have great historical data, stretching way back before 2018, figuring out that statistic is key to figuring out whether this trend is sustainable.

Fortunately, with a little mathematical ju-jitsu, we can figure that out right now, rather than waiting until next Spring.

I’m going to do a teeny bit of math here. My daughter had a summer math packet, consider this to be your summer math packet. You can complain about it together while you watch Love Island. Just please do not make me watch Love Island.

So, we know that:

Number of gun homicides (G) = Total Number of Homicides (H) X Proportion of Gun Homicides (P). And,

Total Number of Homicides (H) = Number of gun homicides (G) + Number of non-firearms homicides (N)

This is very straightforward. If we want to know how many gun homicides there are, we can multiply the total number of homicides by the proportion of gun-related homicides. Easy. If instead, we want to know the total number of homicides, we can just sum the number of the two types of homicides. Also very easy. This will be really helpful, because in August 2024, we do not know how many gun homicides there are or the total number of homicides.

To make this simpler, let’s just say:

G= H X P

H= G + N

Now, from the trend data above, we know that the number of non-firearms homicides (N) is extremely stable. It’s so stable that we can just say it is a constant. So, rather than having four unknown variables, we have three (G, H, and P) and one constant, which is known. And because N is constant, we can just look at the whole time series (2018-2023) and calculate it. N is the average of the monthly non-firearms homicides from 2018-2023, which turns out to be 416.17. So, N=416.17.

Now, we can do a bit of substitution to rearrange these two equations and reduce them to a single unknown variable, which we can then solve for. If we substitute G +N into the formula for G, you get

G= (G+N) X P

If we solve for G we get:

G= (P X N)/(1 - P)

This is helpful because p/1-p is the odds of an event. So, in this equation, the number of firearm homicides is just the odds that a homicide is with a firearm, multiplied by the constant.

The same logic allows us to simplify the equation for all homicides. If we put the gun homicide odds equation into the formula for Homicides (H) we get:

H=(P X N)/(1 - P) + N

We know N= 416.17 so we have our formula:

H = 416.17P/(1 – P) + 416.17

Again, the reason to do this is to figure out how many homicides there are right now and the number of gun homicides there are right now (which we won’t know from CDC until next Spring) and use that to calculate the proportion of all homicides that were committed with a homicide. We can then look at the trend in this proportion and see where we are along that trend. Which should help us think about whether this decline is sustainable or not.

But first, let’s look at some earlier data to see how well our model works! We know the real values of each of these variables from existing data, so let’s put the observed firearm proportions into the formula above and see if it produces values that are consistent with what we actually observed. I made a little table of the month with the highest and lowest proportion of gun homicides between January 2018 and December 2023 to check the math.

To further validate this model, I ran this calculation for all 72 months and the error rate is 6.5% for any given month. The annual error rate, the error rate for a full year of data, is smaller, just 2.5%. So, this is a pretty good estimator.

**Estimate for 2024**

So now, we can calculate how many homicides we should expect in 2024 and use that to calculate the proportion of homicides with a firearm. And see how it compares to long-term trends.

As mentioned above, there is a broad consensus that homicide has declined substantially in 2024.

Looking at the first published estimates to date, I estimate the average estimated 2024 decline to be about 18% (I am discounting the FBI numbers as there is a consensus that at least the last month in that series (March 2024) includes only partial data). Adding that to the long-term trend yields this:

Now we want to solve for the percentage of firearm homicides. There are about 345 million Americans and a homicide rate of 4.5 per 100K would translate to 15,525 homicides for 2024.

Assuming that the non-firearms homicide rate is still constant at 416 per month, that translates to a percentage of firearms homicides of 67.84%, which is lower than it has been in the last six years.

Opening the aperture, CDC data from 1999-2017 suggests that this percentage is much more plausible. Before 2015, a firearms proportion of 68% was not at all unusual.

**Implications**

The key idea here is that there is a constant number of non-firearms homicides in a month or year and that changes in firearms homicides are the key explanation for any crime spike or decline. The crime spike of 2020 seems wholly explained by increases in firearm homicide. The decline of the past 18 months seems wholly explained by it as well, though the data are not yet available to demonstrate this empirically.

The key question is whether there have been forces at work that would explain the decline in firearm harm. Three theories seem to hold the most promise in explaining the decline, but they suggest some caution is warranted in thinking about the sustainability of the decline.

**1. Government Investment**. There has been substantial investment by the federal government in police-led anti-gun violence initiatives.

Of the $350 billion in ARPA investments, the Marshall Project—which is skeptical of these investments—found more than $50 billion went to law enforcement-related local investment spending. That level of investment is unprecedented—the Clinton-era program to hire 100,000 new police officers cost about $9 billion and even adjusted for inflation, it was a much smaller program.

The Bipartisan Safer Communities Act included a range of provisions supporting anti-gun initiatives or restricting access to firearms for high-risk individuals.

There has been a dramatic expansion of community-based violence strategies, including

Community violence intervention/interruption (CVI) initiatives.

Programs to reduce environmental factors that increase the risk of violence.

Cognitive-behavioral Interventions with wrap-around support

**2. A renewed focus on gun violence by cities that had lost focus**. That the last two years have seen a focus on reducing gun violence by all actors (police, courts, community) seems uncontroversial. More is being done than before, and more evidence-based practices are being applied. This seems to be particularly true in cities that have lagged in their approach to violence. Council on Criminal Justice data shows that the biggest homicide declines are focused on cities that have historically been unusually unsuccessful in curbing serious gun violence. A lessening of political pressure against aggressive prosecution of serious gun offenders and the urgency of responding to the dramatic violence spike likely contributed to the focus on arresting and prosecuting shooters.

**3. A reduction in the number of willing shooters.** A common narrative from prosecutors and police as well as researchers is that even in a violent gang or crew, there is usually only one highly willing shooter (others may be ordered to shoot or be cornered and shoot). The criminal justice system basically ground to a halt in many places, and a huge backlog of cases and warrants piled up from 2020 into 2022 and 2023. When the system began to unclog, cases with shooters, contract hitters, and other high-volume homicide perpetrators were at the top of the list. There is also ample evidence that the sudden violence spike also claimed many of these shooters as victims. And with the most willing shooters off the streets, the number of retaliatory and self-defense shootings declined precipitously.

Still, there are countervailing forces that must be accounted for. In particular, the pandemic gun buying spree left more guns on the street than ever before, ten million or more guns than would have been sold absent the pandemic. That increase in the availability of firearms should have put upward pressure on the firearm homicide percentage. The bottom line is that there are more guns on the street, there just seem to be fewer people willing to pull the trigger.

**A Quick Forecast**

The bottom line is that we should be careful about patting ourselves on the back for this decline. There is a fiscal cliff coming since governments must spend their anti-crime dollars soon. Shooters will start returning from prison. And the political pressure to fight gun violence will ease and cities that have traditionally not had their eye on the ball, may once again shift their gaze elsewhere. That leaves only a two- or three-year period where the wind is at our backs. And we are well into year one. We should move quickly to shore up these gains.

The bottom line is that this decline mostly seems to have happened because of things that we, collectively, did. We should no longer submit to the tired trope that violence is unavoidable, that the best we can hope for is a high clearance rate and dangerous people locked away. There is so much more that can be done. Much more than just solving crimes.

John, love your work. I have a few questions that are likely wording issues. (1) When you say "non-homicide rate" do you mean "non-firearm homicide rate"? (2) When you say "nonfatal homicides" do you mean something like "nonfatal shootings" or (more likely) "non-firearm homicides"? And when you say "proportion of firearm homicides" do you mean "proportion of homicides involving a firearm"? Again, love that you are doing this work and just want to check to make sure I understand what you are saying. -Don

I love this guys work. Careful and thoughtful.