Folk causation

Does the Full Moon Make People Act Up? The Results

October 26, 2026

The short answer

No. The full moon does not predict mood, violence, police calls or ER visits.

Around the full moon, Twitter was no grumpier, assaults were no more common, police got no more calls and ERs saw no more injuries. Every difference was under half a percent, smaller than chance produces, and it flipped direction from year to year. The full moon calendar did nothing to help forecast any of them.

That’s what I predicted before looking.^[1]

There was one surprise, and it runs the opposite way to the folklore: police got slightly more calls around the new moon, the darkest nights.

What I tested

The claim: around the full moon, people get moodier and more violent, and police and ERs get busier.

• Mood: daily happiness of English-language Twitter, 5,367 days from 2008 to 2023.^[2]
• Violence: 3.7 million assaults and batteries reported in Chicago (2001–2026)^[3] and New York (2006–2026),^[4] counted by the day they happened.
• Police calls: calls from the public to police in Seattle (2010–2026)^[5] and New Orleans (2014–2026).^[6] Calls officers start themselves are left out.
• ER visits: 7.4 million injury visits from a federal sample of about 100 emergency rooms, 2006–2025.^[7]
• Births (secondary): every US birth by date, 1994–2014.^[8]
• The full moon: every one since 1994, from NASA’s planetary data.^[9] The “full-moon window” is the full-moon date plus the day before and after, in each city’s own time zone.
• The fair comparison: each day was compared with the same city, month and weekday, with holidays accounted for. A Saturday in July should be compared with other Saturdays in July, not with a Tuesday in February.
• The chance check: in every lunar cycle I moved the window to a random spot, 10,000 times, and asked how often a fake calendar showed a gap as big as the real one.

What I found

To count as unusual after running four tests, a gap needed to show up in fewer than about 1 in 80 fake calendars. None came close. Three of the four leaned the opposite way to the claim, and every gap is tiny. On a night that would normally see 100 assaults, the full moon window averages about 99.6.

Can it forecast? For each measure I trained a model on the earlier years and asked it to predict the later ones, with and without the full moon calendar. The real calendar did no better than random calendars: it beat 51% of them for mood, 38% for violence, 45% for police calls and 47% for the ER. The rule I set in advance needed 95%.

The direction flips from year to year

If the full moon really did something, it should lean the same way most years. Here’s how many years leaned the way believers expect:

Each of these is what you’d get from a coin toss. Violence is the cleanest example: assaults ran up to 3% higher around the full moon in some years and up to 3% lower in others, and over 26 years they split exactly 13 to 13. Anyone who worked through a run of bad full moons in, say, 2001–2003 would have seen a pattern that then disappeared.

The refinements

I also tested the details believers add:

• The full-moon night alone, without the days on either side: nothing, for any of the four.
• Nights only (6 p.m. to 6 a.m.), when moonlight could matter: assaults 0.3% fewer, police calls 0.4% fewer. Both are ordinary.
• Births: 0.16% more, a gap that random calendars match one time in three.
• Moon brightness and full moon against new moon: nothing for mood, violence or the ER. Police calls are the exception, below.

The surprise: more police calls on the darkest nights

One result held up after correcting for the 15 extra tests: police calls were about 1.5% lower around the full moon than around the new moon, and they fell as the moon got brighter. Random calendars produce a gap that size about 1 time in 1,250.

That’s the opposite of the folklore. The full moon itself wasn’t busier; the new moon was.

I then split it by city. This step wasn’t in the plan, so treat it as exploration:

• New Orleans: about 1.9% more calls around the new moon. It’s the bulk of the effect.
• Seattle: about 0.5% more, within chance.

I don’t know why. Darker nights could change what people do or what they report, or it could be something particular to New Orleans. One city isn’t enough to believe it. A 1–2% shift is also far too small for anyone working a dispatch shift to notice. It would need testing in other cities before it means anything.

Why the belief survives anyway

None of this makes the nurses and officers foolish.

• Busy shifts happen constantly. These are noisy numbers. A city’s daily assault count routinely moves 10% or more from one day to the next for reasons that have nothing to do with the sky.
• The full moon is visible and memorable. A chaotic shift under a full moon gets remembered and retold. A chaotic shift under a half moon is just a bad Tuesday.
• Runs happen. A few years of the gap leaning one way, as Chicago’s did in the early 2000s, are enough to feel like experience.

What this does and doesn’t show

• It does show: across up to 25 years and five cities, mood, violence, police calls and ER visits don’t rise at the full moon beyond chance, and the full moon can’t forecast them.
• It doesn’t cover sleep, suicides or psychiatric admissions. There’s no public daily data for any of them in recent years.
• The ER data is injuries involving consumer products (falls, bikes, sports, tools), not every ER visit.
• Twitter is not everyone, and the mood data stops in 2023.
• It doesn’t test “the full moon affects me.” That would need people tracking themselves day by day.
• It shows lining up, not cause. Including the new moon result.

Where I filled in details the plan didn’t specify

• Random windows: placed anywhere in each cycle except the last two days before the next full moon, so windows never overlap.
• Holidays: 16 of them, including Easter, which is itself set by the moon.
• The forecast baseline: the previous 28 days’ average, plus weekday and holiday effects learned on the training years.
• Training years: through 2018 for mood, violence and the ER, and through 2021 for police calls (New Orleans starts in 2014).
• Data gaps: two New Orleans days in November 2023 look like a recording gap. They were kept, since the plan didn’t allow dropping data.
• ER files: the federal site blocks automated downloads, so I pulled them through a web browser and counted visits per treatment date.
• Blinding: while pulling the data, a few raw daily counts were briefly displayed (three Chicago days in 2001, and partial ER years). None were compared to the moon before the analysis ran.

The verdict

Hypothesis: the full moon is predictive of mood, violence, police calls or ER visits. Not supported.

The one result that held up points the other way: slightly more police calls on the darkest nights, mostly in one city. It’s worth checking elsewhere, not believing yet.

The code, the data and every number behind this page are public,^[10] so anyone who thinks I got it wrong can check.

A note on how this was written

Based on the way this is written, you can probably tell it was written heavily by AI. The structure of everything is built by me in conjunction with AI, and nothing is published without my final approval.