Does the Vibe Weather lead anything real? A test written before the data

Published 2026-10-01 · every number links to the public page it was read from · corrections welcome

1. The series

The index. One reading per borough per engine run, from local news tone, 311 counts and public-post counts (counts only, never words), archived unrevised in the daily archive from 2026-10-01. The daily value is the mean of the day's runs.

The outside series, archived beside it in the signal archive, each from a public, official source:

Nothing about any person enters any series.

2. The hypotheses, fixed now

1. H1 (same day): a borough's index is negatively associated with its non-noise 311 volume on the same day, after removing day-of-week and weather.
2. H2 (one day ahead): a fall in the citywide index of at least 0.15 is followed, the next day, by lower subway ridership than the seasonal-naive forecast (the same weekday of the previous week) more often than chance.
3. H3 (nights out): a borough reading of Lifted or Warm on a Friday is followed by higher Saturday bikes-out than the previous Saturday more often than chance.
4. H0 for all three: the index carries no information beyond weekday and weather. The test is designed to be able to say so.

3. The method, fixed now

• Lead/lag correlations at 0 to 7 days, on residuals after weekday and weather are removed.
• Granger-style tests: does yesterday's index improve today's forecast of each series beyond the series' own past and the weather?
• Out-of-sample skill: a rolling forecast, trained on the first half, scored on the second, against the seasonal-naive baseline. The index earns the word "useful" only if it beats that baseline out of sample.
• Multiple comparisons corrected (Benjamini-Hochberg, 5 boroughs by 3 hypotheses by 8 lags).
• Minimum history before any result: 180 days. Before "predicts": 365 days, and a second out-of-sample period.

4. What will be published

The full table, every lag, every borough, including the nulls, with the code that produced it, on this page, on the first weekday after the minimum history is met. A result against the index is published with the same prominence as one for it. Until then the Vibe Weather is a reading, not a forecast, and is labeled as one everywhere it appears.

5. A paper predictor, no money

From day 180, a public ledger of weekly questions written from the index ("Will Brooklyn's non-noise 311 count this weekend exceed last weekend's?"), each with the index's call, the baseline's call, and the answer when it arrives. Scored in public. No money is placed on anything.

Pre-registered 2026-10-01. Changes to this page after that date are listed here with their reason; there are none yet.

Written to the studies standard. If a figure here does not match the page it links to, the page is right and this study is wrong: write to us and it gets corrected in place, with a note.