Murder Is Likely Falling At The Fastest Pace Ever Recorded In 2024

Some thoughts on uncertainty and analyzing a historic trend.

Richards Heuer’s Psychology of Intelligence Analysis was required reading when I was a baby analyst at CIA and I like to think that its teachings have permeated throughout my analysis and writing. If you haven’t read it and are looking for a page turner then I would highly recommend it!

One of my favorite graphs is an experiment that was conducted to evaluate how people perceive various terms frequently used in analysis. Heuer writes that this graph

“Shows the results of an experiment with NATO military officers accustomed to reading intelligence reports. They were given a number of sentences such as: “It is highly unlikely that....” All the sentences were the same except that the verbal expressions of probability changed. The officers were asked what percentage probability they would attribute to each statement if they read it in an intelligence report. Each dot in the table represents one officer’s probability assignment. While there was broad consensus about the meaning of “better than even,” there was a wide disparity in interpretation of other probability expressions. The shaded areas in the table show the ranges proposed by (legendary CIA analyst) Sherman Kent.

Coming back to crime analysis, the word “Likely” in the title of this piece is still doing a fair amount of lifting on the analytic bottom line given the remaining uncertainty with respect to 4 more months and an incomplete dataset. I would put the word “Likely” to mean 65 to 75 percent, but as the above figure shows that impression can change from person to person (as an aside, the officer who said “Likely” means 35 percent probably shouldn’t have been working in intelligence).

There is a lot of evidence that murder is falling faster nationally than it has ever before — with the caveat that official murder data only exists through 1960.

We won’t know just how fast murder fell in 2023, but all of the available data points to a decline that was at or near the fastest pace ever recorded last year. Murder was down 11.7 percent in our sample of 214 cities (see the link to 2023 data at the bottom) with available data, it was down 13.2 percent in the FBI’s quarterly data through Q4 2023, and it was down 12.2 percent in the 31 states that had published data as of about a month ago. Other sources such as CDC’s WONDER and the Gun Violence Archive also point to large declines in homicides and fatal shootings respectively.

One reason to suspect that last year was at or near the largest ever recorded is that murder usually doesn’t decline all that fast from one year to the next. The largest decline ever recorded came in 1996 and that was just 9.1 percent (by contrast, there had been 7 one-year increases that were larger than 9.1 percent before 2020’s monumental increase), so even a double-digit decline in murder would be the largest ever recorded.

That said, there’s still some uncertainty, and none of the available data sources tell us what the FBI is going to estimate with a ton of precision (though all of them point to a very large drop). So let’s arbitrarily add a 3 percent margin of error on either side of the big city sample and say that murder was down somewhere most likely down between 8.7 and 14.7 percent nationally in 2023. We’re probably correct in saying that 2023’s decline was the largest ever, but we also shouldn’t be shocked if it doesn’t quite eclipse the 1996 decline.

All of which brings us to 2024.

Hilary Clinton was favored over Donald Trump in 2016 with about 70 to 30 odds according to Nate Silver’s 538 forecast for that year. Joe Biden and Barack Obama were even heavier favorites, roughly 90 to 10 odds, according to the 2020 and 2012 538 forecasts.

The reason that Clinton was a smaller favorite while Biden and Obama were much closer to sure things was because a Clinton loss (which you may not remember eventually happened) only required a normal polling error while a Biden or Obama loss would’ve required polls to have been historically wrong (they were off by a reasonably normal amount in 2020). It's the difference between being up 4 percent on election day (Clinton) versus 8 percent (Biden).

The 2023 and 2024 declines in murder are similarly situated.

Murder is currently down 17.8 percent in 277 cities with available 2024 data. It’s down 17.7 percent in 171 cities with data available through July 2024.

Murder is down 19.9 percent in 90 cities of under 100,000 people. It’s down 17.8 percent in all 10 cities with 1 million+ people¹.

There are 36 cities with a double-digit decline in the number of murders this year compared to the same timeframe last year. There are three cities (Savannah, Baton Rouge, and Charlotte) with a double-digit increase.

The decline in murder in 2023 would require the city sample to be off by a large — but not inconceivably large — amount to be remembered as not quite the largest decline ever.

The decline in murder in 2024 would require the city sample — which has more cities than last year — to be off by a historically abnormal amount to not be the largest ever recorded.

Here’s the percent change in murder every year since 1961² highlighting just how large the 2023 and 2024 murder declines are (assuming a 10 percent drop in 2023 and a 17 percent decline this year). There is a ton of wiggle room for this year’s city sample to not quite represent the national trend and for murder to fall at a record pace this year.

Of course there are caveats.

Something could happen in the last 4ish months of the year that makes this analysis moot. The percent change typically doesn’t change around all that much this late in the year, but it’s possible that this year is an aberration given how many of those we’ve been having lately.

I would reasonably expect the percent decline to shift a percentage point or two (probably towards zero) by the end of the year, but it’s certainly possible that it shifts more. The percent change hasn’t budged in a month and a half though — it was -17.7 percent on July 3rd and it’s -17.8 percent now. It’s also plausible that last year’s decline was even bigger than the city sample suggests and this year’s decline doesn’t quite match it.

Crime data is imperfect and our collection of crime data is imperfect. Moreover, we don’t fully understand the drivers of these trends — we can barely accurately describe them. It’s really hard to analyze and forecast a trend when you don’t understand its drivers.

All that said, the available evidence points to the largest decline ever recorded likely occurring in 2024. Could that assessment be off? Of course! But all we can do is offer an informed guess in the absence of good, solid, fast national data like they have in other fields.

1. I updated these figures after publication to match what we had on the day of publication rather than what they were when I wrote this last week.

2. I’m using the 2022 Table 1 for murder counts for 2003 through 2022 and FBI UCR estimates for prior to that. More info on this issue in a previous post here.