The Rhythm of a City

Read this first: every number here counts reported incidents, not crime. Willingness to call the police varies by neighborhood, offense, and decade; classifications get revised after investigation. That caveat is inherited by every chart below.

Chicago's crime data is one of the most complete municipal records in America: every reported incident since 2001, updated daily, with type, time, arrest flag, and block-level location. Read as a single object, it has three registers — a daily rhythm, a weekly one, and a generational one. All three are worth looking at, and the generational one does not say what most people assume.

1 Three eras

Reported incidents fell almost by half between 2001 (486,000) and 2019 (262,000) — a long, quiet, monotone decline that matches the national story. Then three years of disruption: pandemic lockdowns crushed street crime while homicides spiked to 813 in 2021, their worst since the 1990s. And then the part that rarely makes headlines: by 2025 homicides had fallen to 434 — below the pre-pandemic level — and total incidents settled near their 2019 floor.

A quarter-century of reported incidents

Monthly counts (thin) with 12-month average (bold) · left: all incidents · right: homicides

Data table (annual)

Figure 1. Both panels share the x-axis but have their own y-scales (homicides are ~0.3% of incidents). Shaded band = the 2020–22 disruption. The sawtooth in the left panel is seasonality — Chicago crime peaks every summer. 2026 is a partial year and its final months are still being backfilled.

2 The clock

Averaged over years, each offense keeps its own office hours. Theft peaks weekday afternoons — it lives where wallets and parked cars do. Battery peaks at 3 p.m. and again near midnight. Robbery is a creature of 9–11 p.m. Narcotics enforcement (which tracks police activity as much as drug activity) runs on the evening shift. The heatmap below is the city's timetable; switch offense and era to watch it deform.

When each crime happens

Share of the offense's weekly total in each hour × weekday cell · darker = busier

Offense Era

Data table (hourly totals)

Figure 2. Cells are the percent of that offense-era's total falling in each hour × day slot (a uniform clock would put ~0.6% in every cell). The midnight and noon columns are inflated by a recording artifact — incidents with unknown times default to 00:00 or 12:00 — so read them with suspicion. Hover any cell for values.

3 The TikTok car theft wave

The sharpest single feature of the post-pandemic record is not violence — it's motor vehicle theft. A long decline had brought it from ~2,300 a month in 2001 down to 700–900 by the late 2010s. Then it tripled in the second half of 2022, peaking above 2,900 a month, after a viral video demonstrated that Kia and Hyundai models sold without engine immobilizers could be started with a USB cable. Software patches and insurance pressure have pulled it back down since — a natural experiment in how much "crime" is really opportunity.

Motor vehicle theft: a software vulnerability, measured monthly

Monthly reported incidents · burglary shown for contrast

Data table (annual)

Figure 3. Burglary (gray) is the natural control: same property-crime family, same neighborhoods, same police — no exploit video. It declined smoothly through the entire period. The car-theft surge begins within weeks of the "Kia Boys" videos spreading in mid-2022.

4 Geography is the strongest variable

Nothing in this record — not hour, not season, not era — separates experiences like place. Averaging 2023–25, the annual violent-incident rate ranges from under 5 per 1,000 residents in Forest Glen and Edison Park to 94 in West Garfield Park — a nineteen-fold gap. (Tiny Fuller Park, population ~2,300, posts 118, but small denominators make its rate jumpy.) Plotting rates against neighborhood affluence gives the classic curve: steep, nonlinear, and with essentially no high-income, high-violence quadrant.

Violence and affluence across 77 community areas

x: share of households earning $100k+ (2024) · y: violent incidents per 1,000 residents per year (2023–25) · circle area = population

Data table (all 77 areas)

Figure 4. Violent = homicide, robbery, assault, battery, criminal sexual assault. Rates use residential population — the Loop's rate is inflated by its huge daytime and visitor population. Reporting propensity differs across neighborhoods, which most likely understates the true gradient. Hover for names.

5 The quiet collapse of the arrest

In 2001, 29% of reported incidents ended in an arrest. By 2023 that figure was 12%, recovering only to 16% by 2025. Some of this is composition — the near-disappearance of drug arrests, which are police-initiated and nearly always end in arrest, mechanically drags the average down. But the slide appears within offense categories too. Whatever it measures — enforcement capacity, priorities, clearance difficulty — it is one of the largest institutional shifts in the record.

Share of reported incidents ending in arrest

Annual, 2001–2025

Data table

Figure 5. Narcotics incidents are arrest-definitional (~99%) — their volume collapse after 2015 explains part, not all, of the overall slide. Homicide "arrest rate" here is arrests recorded on the incident, a rough proxy for clearance, and reads low because clearances often come later. Hover for all four series.

The rhythm, the wave, the gradient, the slide: four different clocks running in one dataset. What they share is that none of them is well described by the phrase "crime is up" or "crime is down" — the city's pulse is legible only when you let each register keep its own time.

Data & methods

Source. City of Chicago data portal ("Crimes — 2001 to Present," CPD CLEAR system), via the BigQuery public mirror, extracted 2026-07-09; 8,589,450 incidents through 2026-07-01. Chicago open data license; the city provides data "as is."
Reported ≠ actual. Reporting propensity varies by neighborhood, offense, and era. Victimization surveys suggest under half of violent crime is reported. Classifications are revised after investigation; recent months backfill.
Time fields. Incident times are as recorded; unknown times heap at 00:00 and 12:00 (flagged in Figure 2). Reported time ≠ occurrence time, worst for burglary (discovered mornings).
Geography. Community-area population and income-bracket shares: CMAP Community Data Snapshots 2024 (ACS-based). The per-capita income figures quoted for extremes are 2008–12 ACS vintage. Locations are anonymized to block level by the city.
Eras. 2001–19 / 2020–22 / 2023–26, chosen for the pandemic discontinuity; Figure 2's era selector uses the same cuts. 2026 is a partial year (through July 1) and never appears in annual totals.