Snopes Got Mother's Day Crime Wrong: 372,128 Records Show the Pattern

What Snopes tested

Snopes evaluated the strong-form viral claim that Mother's Day is the lowest crime day of the year. They used the tools any small-team fact-check newsroom uses: one example city, one or two years, total crime as a single bucket. On that question, with that sample, the FALSE rating is defensible. Mother's Day is not the literal lowest day on the calendar. Weather, day-of-week effects, COVID re-opening dynamics, and the occasional single-day spike all push the title around year to year. A meme claiming a specific record-holder day deserves a skeptical read, and theirs was skeptical and honest given what they could see in front of them.

What their sample could not see was the gap between two opposite signals inside that "total crime" bucket. When the metric you test is everything-summed, two crime types moving in opposite directions cancel each other out before the t-test ever runs. The viral claim was about street crime, in spirit; the methodology rated the literal text against total crime. Both can be defensible. Both can also miss the underlying pattern that lives one cut deeper.

Why their methodology missed it

The arithmetic is the simple part. Domestic violence on Mother's Day runs about +4% above the May Sunday baseline in Chicago and NYC. Everything else (theft, robbery, aggravated assault, motor vehicle theft, simple assault, burglary) runs about 10% to 11% below. When you combine the two into one "total crime" bucket and run a test on the sum, the small positive from DV partly counteracts the larger negative from the rest. The signal is still there in the all-crime cut (Chicago + NYC paired t = -4.023, p = 0.003), but it is much weaker than the non-DV cut that the meme's actual mechanism implies.

Tiny samples make this worse. With one or two years of total-crime data from a single city, the variance is so large that the underlying effect lives well below the detection threshold. Adding DV into the same bucket pushes the point estimate toward zero. The combination of small n and a mixed metric is a methodology trap that catches even careful fact-checkers. This is not a "Snopes was wrong" framing. It is a "their sample could not detect" framing, and it is the same kind of trap any small-sample analysis falls into when the metric mixes signals.

The data we extended their work with

The panel pulls every May Sunday from three cities for five years. Per-city volume: Chicago at 103,888 rows from the open ijzp-q8t2 endpoint, New York City at 233,112 rows from qgea-i56i, Seattle at 35,128 rows from tazs-3rd5. Mother's Day is always a Sunday, so the only fair control is other Sundays in the same city and the same year. Each (city, year) cell contributes one matched difference to the paired test.

The headline cut is non-DV crime in the two large cities. Chicago + NYC, 2021 through 2025: paired t = -5.00 on the ten year-matched differences, p = 0.0007. Mean drop 10.19%. Sign streak 10 of 10 city-years below baseline, with a one-sided sign-test p of 0.002 directly from the binomial (no distributional assumption). Theft alone moves the same direction at p = 0.006. The all-crime cut, which is the metric closest to what Snopes tested, still drops at p = 0.003 despite combining DV in. The cleanest single statement is the sign streak.

When we add Seattle into the three-city panel, the headline holds (p = 0.006 on the paired non-DV test) but the point estimate moves toward zero because Seattle's volume is much smaller and its DV proxy is much narrower than Chicago's clean domestic flag. The full per-category numbers and the Welch comparisons live in the methodology drawer on the evergreen.

What this means for the fact-check ecosystem

Two findings sit comfortably together. Snopes rated the literal viral claim correctly: Mother's Day is not the calendar's lowest crime day, and a small-sample test was the right tool for that literal text. We rated the meme's underlying mechanism honestly: non-DV crime does drop when families are together, in the two cities with enough volume to detect it, every year from 2021 through 2025.

This is what extending fact-checks looks like, not opposing them. The viral form of any meme overstates the mechanism, and a fact-check graded against the strong form rightly grades it FALSE. When a larger panel and a different cut surface the underlying pattern, the right framing is "their sample could not detect" and "here is the metric the meme was actually about." Both pieces of work end up on the same page, and the readers who care about whether the meme is reassuring or alarmist get a fuller answer than either piece could give alone.

Read the full analysis at The Mother's Day Crime Analysis, including the per-category breakdown, the methodology drawer, the year-matched panel grid, and the dataset download.

Sources and provenance

All numbers come from raw incident-level open data via the three Socrata endpoints listed below. No published studies, news articles, or pre-computed summary statistics were used. The full analytical source-of-truth, including the paired t and Welch comparisons, the sensitivity check dropping 2021 for pandemic re-opening dynamics, and the limitations section, is the analysis report behind the evergreen at /mothers-day-crime. The raw daily panel for all three cities is available at /data/mothers-day-crime.csv under CC-BY-4.0; replications are welcome.

Sources: City of Chicago Crimes 2001 to Present (ijzp-q8t2, data.cityofchicago.org); NYC Open Data NYPD Complaint Data Historic (qgea-i56i, data.cityofnewyork.us); Seattle Open Data SPD Crime Data (tazs-3rd5, data.seattle.gov).