Crunching the numbers to make an educated guess
Technology has revolutionized almost every aspect of daily life. Your handheld device is more powerful than a 1970s supercomputer, AI can write odes that would knock Chaucer’s socks off, and even weather forecasting has gotten fantastically more accurate. Despite these advances, crime forecasting remains a frontier land — wild and uncivilized.
That said, the potential of crime forecasting to more effectively guide policy and practice is manifest, offering objective evidence about how to allocate scarce policing resources, anticipate where more prevention is needed, plan budgets and provide useful guidance to the public. Particularly for a big-city government like New York City’s, which wrestles with hugely competing ideas and the effect of those choices is clearly evident, it seems inevitable that the role of these forecasts will only grow.
In that spirit, we offer a numerical forecast of serious crime in New York City in 2024. The statistical model we rely on is called an ARIMA statistical model (it stands for “auto-regressive integrated moving average,” but that’s more than you need to know), which uses underlying trends in outcomes — in this case, serious crimes — to predict future crimes. Of course, outside forces can nudge, or shove, that trend in unexpected directions. As we noted last year, one of the key outside forces is the behavior of the police: Larger numbers of arrests tend to reduce crime as more people who are motivated to offend are incapacitated and the threat of deterrence increases. But in addition to arrests, there are many forces that could affect serious crime this year.
National trends and local realities
National trends have a gravity all their own, and no matter what else is going on in New York, there is some pull toward broader trends. The question is how much force these trends exert — is it a nudge or a shove? Here, we chart the relationship between changes in New York City crime rates and overall national trends, using homicide as an example. The chart compares the percentage change in homicide from the past year.
It is clear that there is a relationship, but it is not an all-powerful one. New York City is about 2.5% of the nation’s population, and short-term trends in the city tend to be more volatile. In periods with substantial changes in the national trajectory, say 10% or more, New York City follows that trend and indeed tends to exaggerate it. In particular, during the great crime decline of the 1990s, homicides in the city fell at a far faster rate than in the U.S. as a whole. In the pandemic period, as the figure above shows, homicides rose higher and fell further than in the nation as a whole, but followed the same temporal pattern.
But what is equally notable in the graphic is that during periods when homicide was not changing in a substantial way, New York City seems relatively unaffected by national trends. In the 2000-2010 period, for instance, the movements up and down in NYC and the US appear uncorrelated.
So the question becomes, which type of period are we in or entering now — a period of rapid decline or a steadier state? Homicides declined by more than 10% in both 2022 to 2023 in both the U.S. and New York City. We can use a little calculus to note that the first derivative, the rate of change in homicide is negative, meaning the yearly change is negative. But the second derivative, the rate of change of the rate of change, is approximately zero — the homicide decline is not speeding up or reversing; it’s in neutral, or at least it was last year. As a result, our model does not include any push or pull from national trends. We think they will be relatively neutral in 2024.
Although we know a decent amount about the effects of these policies and practices one at a time, the research on their additive and/or interactive effects at scale is vanishingly thin.
Local forces
What else might push or pull New York City’s serious crime numbers? Here is where being on the frontier dramatically limits our forecast. The list of possible gravitational forces is long and distinguished. On the macroeconomic side, unemployment, GDP, interest rates and inflation are likely to shift in 2024 — but the existing body of crime research suggests these have surprisingly little effect on short-term crime rates in specific locales.
There are many more direct forces at work in the five boroughs. For example, the city government appears to have largely recovered from pandemic losses in workers. Will fully staffed support systems reduce serious crimes? Will new policing initiatives meet their goals? Will rapidly maturing community-led violence interruption programs improve public safety?
We don’t know, and the reason we don’t know is that, although we know a decent amount about the effects of these policies and practices one at a time, the research on their additive and/or interactive effects at scale is vanishingly thin. New York City is on pace for more than 100,000 serious crimes again this year. Will inside-the-city forces reduce serious crime not just here and there, but by the thousands? That’s what it will take to change the serious crime rate in a city of almost 9 million. There is simply no comprehensive research on which to base a prediction, and thus we do not include these forces in our model.
The forecast model
We conducted several ARIMA regression models to predict the number of complaints (which is shorthand for the number of crimes reported to police) based on past complaints and different combinations of past arrest data. In simple terms, we’re using data from 2021-2023 arrests (in two different ways) to make predictions about 2024 complaints. After testing various models, we found that the most accurate one includes the number of arrests in the current month, the number of arrests from one month ago and the number of arrests from two months ago. To make our predictions more accurate, we’re taking into account the natural fluctuations in the number of complaints over time.
Overall, we expect serious crimes to decline modestly in New York City in 2024. Our confidence in the forecast is medium. We are fairly confident that there will be no extreme changes in serious crime here or across the country, and thus we are fairly confident that the trends of the last three years have predictive value — and they point to a reduction in serious crimes. The lack of predictive mechanisms deters us from greater confidence.
Conclusion
There is every reason to believe that crime forecasting models will improve in the coming years and that these will be assets in policymaking in New York City. But critical elements must improve. A necessary, but insufficient, condition for better forecasts is more policy-modeling — research that asks the question: How much of an impact will this change have at the city (rather than program) level? To be sufficient, this research must also investigate causality. Ideological and political arguments have little value in numerical forecasting. What is needed instead is to say, for instance, that a government that is fully back to work reduces crime because it supports specific activities that are shown to cause declines in crime and improvements in public safety. And here is how much of that activity it supported and the effect size.
There is much work to be done to achieve either of these objectives.