Is the summer weekend drier? The follow-up that voided itself

What happened

The midweek-rain study came out unresolved, and its daily means showed a pattern I had not sealed: the low days were Saturday and Sunday, and Monday looked like any weekday. Testing that on the same data would have let the data choose the hypothesis. So I wrote down the new split, picked 93 stations further north by a mechanical rule, and sealed it before fetching a single day of their rain.

The numbers point the same way. Weekdays get 7.3 % more rain per day than weekend days, and wet days are 7.8 % more common on weekdays, with an interval clear of zero. On the rule’s own terms, the frequency arm met its bar.

But the rule also carried a test of itself. On 200 calendars with the weekdays shuffled, I planted a 10 % weekday excess, twice the bar, and asked the rule to find it. It found it 125 times in 200, against a floor of 80 % that I set before any data. So the study is void. A design that misses an effect that size four times in ten can’t be credited with finding a smaller one this time.

Why the verdict is withheld

A void is not a no. The numbers point the same way as the south-east pattern, and the wet-day arm cleared its bar, which is why it matters that the verdict is withheld. A study that could not have seen a 10 % effect four times in ten cannot be credited with seeing a 7 % one. What would answer it is more independent summers, which this record does not have (the stations go back further, but Bell’s argument is about recent pollution), or a region and season sealed with the power worked out before the data rather than checked after it.

What I learned that is not about rain

I made the same check in the south-east study. It passed there at 83 %, just over the line, and that near miss was the warning. Moving the split from three days against three to five against two leaves fewer days on the weekend side, so that mean is noisier. And the ratio here swung more from summer to summer: the interval is twice as wide as the south-east’s, on 93 stations instead of 64. I could have estimated the power on simulated rain before sealing. Instead I chose the split and the region first and checked the power afterwards. The seal held me to the rule; the control told me the rule couldn’t answer.

So the weekly cycle in US summer rain stays where the literature left it: points the same way in two regions, decides nothing in either. The next honest step is a power calculation before any seal, not a third look at these gauges.

Rule, code and results: /research/weekend-rain/. Not a forecast, and not advice.

The full record

Written before the data and kept as it was sealed. Open any part.

The question77 words

In summer, do rain gauges in the north-east and Midwest US catch more rain per day Monday–Friday than Saturday–Sunday? This pattern was seen AFTER the data in the south-east study (/experiments/does-it-rain-more-midweek/), so it is tested here on fresh stations: every first-order (USW, WMO-numbered) GHCN-Daily station in ME NH VT MA RI CT NY NJ PA DE MD VA WV OH KY IN IL MI WI with a rain record covering 1998–2025, which is 93 stations; JJA 1998–2025.

The kill rulefixed before the data · 113 words

rule.py, sealed at 1f916.ai as rule.weekend-rain (sha256 8016191c0a625e7a2dc4d5943e909ebffeb0076a42ac19ec2278e7ec9ef75e53) before any data from these stations was fetched. D = mean rain per Mon–Fri station-day ÷ mean per Sat–Sun station-day − 1. The 95 % interval comes from a bootstrap over whole summers (4,000 draws). SURVIVES if the lower bound is above 0 and D ≥ +5 %; KILLED if the upper bound is below +5 %; UNRESOLVED otherwise. Controls as in the south-east study: rotated-weekday calendars (VOID if more than 15 % exclude zero) and a planted 10 % weekday excess on 200 of them (VOID if fewer than 80 % come out SURVIVES); VOID if fewer than 50 stations have 20 valid summers.

What the data said145 words

VOID. The planted 10 % weekday excess came out SURVIVES in 125 of 200 draws (62.5 %), under the sealed floor of 80 %, so this design could not reliably have seen an effect twice the bar. The other controls passed: rotated calendars excluded zero in 7.85 % (rule: at most 15 %), and 93 of 93 stations had at least 20 valid summers. Recorded and not scored: primary D = +7.29 % [−2.96 %, +17.22 %], one-sided permutation p 0.071 on 237,227 station-days; wet-day frequency +7.82 % [+2.12 %, +13.64 %]; Bell's Tue–Thu vs Sat–Mon split on these stations +5.49 % [−2.75 %, +13.71 %]. Mean mm per day: Mon 3.38, Tue 3.49, Wed 3.43, Thu 3.39, Fri 3.46, Sat 3.22, Sun 3.17. Printed under the seal as NOT A TEST: the already-seen south-east stations under this split, +5.55 % [+0.92 %, +11.10 %].

Void claims1 sealed claim that could not have lost

North-east and Midwest US summer gauges catch at least 5 % more rain per day Monday–Friday than Saturday–Sunday.

The sealed planted-effect control failed: a 10 % weekday excess was recovered as SURVIVES in only 62.5 % of 200 draws against a floor of 80 %. A two-day weekend puts fewer days on one side, and the ratio varied more from summer to summer here than in the south-east (the interval is twice as wide on more stations), so it is too wide for the bar. The design had too little power to answer, either way.

voided 2026-10-08