Chicago police recorded 818,888 traffic crashes between January 1, 2019 and July 18 of this year. 989 people died in 913 of them. The hour that produces the most crashes and the hour when a crash is most likely to kill someone are not the same hour. They sit almost exactly twelve hours apart.
Both faces run the same 24 hours, noon at the top and midnight at the bottom, each spoke reaching further the higher its number. Notice which half of each face is full.
Crashes pile up in the afternoon, peaking at 3 p.m. and staying heavy through the evening rush. The share of them that kills somebody does the opposite. At 4 p.m., the hour when a crash was least likely to turn fatal, one reported crash in 2,508 killed somebody. At 3 a.m. it was one in 224 — about 11 times more lethal. Take the whole blocks and the same gap holds: between 9 a.m. and 5 p.m., one crash in 1,762 was fatal; between 10 p.m. and 4 a.m., one in 312.
This is a rate, not a body count: far more crashes happen in the afternoon. The question the right-hand dial answers is narrower: of the crashes that get reported to Chicago police, what share kill somebody? That number barely moves all afternoon and then climbs all night.
Is it darkness?
| Hour | Dark, crashes | Fatal per 10k | Daylight, crashes | Fatal per 10k |
|---|---|---|---|---|
| 5 p.m. | 13,147 | 5.3 | 38,613 | 5.2 |
| 6 p.m. | 16,973 | 7.1 | 23,999 | 6.7 |
| 7 p.m. | 19,637 | 15.8 | 9,829 | 20.3 |
Chicago’s evening hours fall in darkness for part of the year and in daylight for the rest, which gives us the comparison for free. Hold the clock hour still and split the crashes by the lighting condition on the police report, and the two rates sit almost on top of each other. At 5 p.m. they differ by a tenth of a point and at 6 p.m. by four tenths, both times with the dark crashes marginally worse. At 7 p.m. the daylight crashes were the deadlier ones, by a wider margin than either.
The limits of that test run in every direction and are worth stating plainly. It covers three evening hours, not the small hours where the gap actually opens, and nobody can drive at 3 a.m. in daylight to give us that comparison. Daylight at 7 p.m. means summer and darkness at 7 p.m. means winter, so season travels with light and the two cannot be separated here. The fatal-crash counts inside each cell are small, so each rate carries real uncertainty.
What survives is narrow. In the only hours where Chicago hands us both conditions, the dark and daylight rates came out close together. That is a reason not to assume visibility explains the night, and it is not a finding about visibility, or evidence for anything that might replace it. For that, the report forms are the next place to look.
The same mistakes, worse outcomes
- OPERATING VEHICLE IN ERRATIC, RECKLESS, CARELESS, NEGLIGENT OR AGGRESSIVE MANNER9033
- DISREGARDING TRAFFIC SIGNALS8926
- FAILING TO REDUCE SPEED TO AVOID CRASH599
- IMPROPER OVERTAKING/PASSING282
- IMPROPER LANE USAGE241
- FAILING TO YIELD RIGHT-OF-WAY226
These are the causes the responding officer wrote down, in the language of the form, counted as fatal crashes per 10,000 crashes — 10 p.m. to 4 a.m. in the first column, 9 a.m. to 5 p.m. in the second. Every one of them is deadlier at night, and none of them is a different kind of error than the daytime version. Running a red light is running a red light. At night it kills roughly four times as often. Following too fast to stop in time kills six times as often.
Speed is the candidate that fits most naturally. A driver who misjudges a gap on an empty road at 3 a.m. carries more energy into the mistake than the same driver in stop-and-go traffic at 5 p.m., and the crash has more of it to spend. But we cannot test that here. This file records what an officer concluded at the scene rather than a reconstruction, and it carries no measured impact speed at all. Nor have we ruled out the other things that change after midnight: who is driving, whether they had been drinking, whether they were belted, how long a wrecked car sits before someone finds it. The cause codes show the same errors producing worse outcomes at night. They are consistent with speed. That is the whole of what they establish.
How we counted
- Source: the city’s Traffic Crashes record (85ca-t3if), which is the crash reports Chicago police file, published by the city. We used every crash from January 1, 2019 through July 18, 2026: 818,888 crashes, 913 of which killed somebody, 989 people in all.
- Why 2019 and not earlier: the dataset starts in 2015 but was still filling up — 44,000 crashes in 2016 against about 110,000 a year since. Earlier years would understate crashes without necessarily understating deaths, which would bend the rate. 2019 is the first year we treat as stable.
- The measure: crashes that killed at least one person, per 10,000 crashes, computed inside each hour. Note that this counts fatal crashes rather than people killed: 989 people died in 913 crashes, so the two are not interchangeable. We use the crash figure throughout, because the question is what happens when a crash happens. This is deliberately a rate. A chart of fatal crashes by hour would mostly redraw the traffic-volume curve and hide the whole effect.
- Small numbers, handled carefully: deaths are rare, so any single hour is noisy — the hourly fatal counts run from 14 to 80, and we would not lean on any one of them. The block comparison is where the sample is large enough to be confident: 373 fatal crashes in the night block against 257 in the day block, a rate ratio of 5.65 with a 95 percent confidence interval running from 4.8 to 6.6. The gap is solid. Its exact size is the part that moves.
- The darkness test: we held the clock hour fixed and split on the report’s own lighting field, using 5, 6 and 7 p.m. because those hours carry thousands of crashes in both darkness and daylight across the seasons — from 9,829 in the smallest cell to 38,613 in the largest. The test cannot reach the small hours, where no daylight comparison exists, and it cannot separate light from season, since daylight at 7 p.m. means summer. Treat it as evidence against the simplest visibility story rather than as a finding about darkness.
- What we are not showing: who was driving, whether anyone had been drinking, or how fast they were going. The crash file carries a driver condition field, but it is recorded by an officer at the scene, is unknown or blank in a large share of cases, and is not a toxicology result. We left it out rather than dress it up.
Computed by KCM Desk from Chicago police crash reports through July 18, 2026; published July 19, 2026. If you spot an error, corrections come first.