Monthly crime trends: global patterns and city departures

Andrew P. Wheeler

Abstract

Crime trends are commonly summarized by national aggregates or short-term percentage changes, both of which can obscure sustained local departures and overstate ordinary variation. This paper decomposes monthly reported crime from the Real-Time Crime Index into a global nonlinear trend, recurring annual seasonality, calendar-month shocks shared across cities, city-specific trend and seasonal departures, and city-month residual variation. Explore the results at https://apwheele.github.io/CrimeDecomp/app/ and reproduce the analysis at https://github.com/apwheele/CrimeDecomp.

Introduction

Crime trends are often summarized with two numbers and a percent change. That is convenient, but it makes normal volatility look important and hides whether a change is temporary or sustained. Prior work has shown how time-series graphs, prediction intervals, and fan charts give a more honest picture of crime trends than binary before-and-after comparisons (Wheeler 2016; Wheeler and Kovandzic 2018; Yim, Riddell, and Wheeler 2020). Those tools answer whether a recent observation is unusual. Here I address a related question: what part of a city’s change is shared with other cities, and what part is local?

There are at least three patterns worth separating. Crime can follow a gradual trend shared across the country, a city can have a persistently different trajectory or seasonal pattern, and an individual month can depart sharply from both. Lumping those patterns into one residual term makes a short-lived spike look like a new trend (or smooths away a local trend that is actually different).

I use monthly reported crime from the Real-Time Crime Index (RTCI) (AH Datalytics 2026) and fit a hierarchical binomial model separately for 7 offenses. The model has a smooth trend and seasonal pattern shared across the sample, but it also gives every city its own partially pooled trend and season. Two additional terms separate monthly movement shared by all cities from the remaining city-month variation.

The goal is descriptive. The model provides a practical way to compare the national pattern, sustained city departures, and unusual months. It is not a test of why crime changed, and it is not a forecasting model.

Data and rate construction

I use 458,392 city-month-offense observations from 586 RTCI cities, covering 7 offenses from January 2017 through May 2026. I do not impose a population threshold. For the figures, a monthly count