It does say
Searching 1,026 overlapping annual relationships produced a selected ten-year line that weakened sharply in the next fixed decade.
1,026 searches. One line stood out.
The best-looking line
nearly disappeared.
Among 1,026 ten-year searches—many sharing the same years—the strongest absolute correlation was 0.835. In the immediately following decade, the same comparison measured 0.094.
This is about how we search—not about whether climate matters to tourism.
THE DECADE THE SEARCH COULD SEE
0.835
Palau tourism change vs land-temperature anomaly · 1998–2007
THE NEXT DECADE
0.094
Same comparison · 2008–2017
The search trail
Watch the record climb.
The display order ignores the correlation values: territory code, rainfall → sea/EEZ temperature → land temperature, lag, then start year. Move the slider to open more searches.
Use Left and Right Arrow for one comparison, Page Up and Page Down for ten.
BEST SO FAR
|r| 0.257
Cook Islands · rainfall · lag 0 · 1996–2005
1 record found
| Opened | Case | Training |r| | Next decade |r| |
|---|
The time curtain
Then the next ten years arrived.
The search used 1998–2007 to select this line. The immediate 2008–2017 decade played no part in selection. Reveal it one year at a time.
Read the twenty plotted observations as a table
| Period | Year | Temperature anomaly (°C) | Tourism change (log %) |
|---|
The selected line did not travel well
r −0.835 became −0.094.
The selected correlation weakened sharply. This is not evidence that climate is irrelevant, and it is not a claim about Palau or tourism impacts.
The wider audit
The top ten all weakened.
The median absolute correlation fell from 0.787 in the selected decade to 0.156 in the immediate holdout.
Read the ten values as a table
| Rank | Case | Training |r| | Holdout |r| |
|---|
Shuffled calendars
Most shuffled calendars found a maximum at least as large.
Tourism-change years were reassigned within each territory and the same fixed search was rerun 199 times. In 187 of 199 runs, the largest correlation was at least 0.835.
This is a descriptive stress test, not a p-value. The windows and specifications overlap and are not independent.
Read the result narrowly
What this says—and what it does not.
It does not say
That climate is irrelevant to tourism, that the selected relationship is “fake”, that correlation is causal, that Palau experienced a climate-driven tourism effect, or that this is a forecast.
Coverage
15 territories, total tourism arrivals, three annual climate signals, lags 0–2 and complete 20-year windows. It is not every Pacific place or every mechanism.
Method boundary
Frozen SPC-derived inputs; 15 territories; total tourism arrivals; three annual climate signals; lags 0–2; complete 20-year windows. Exact sources and limitations are documented with the work.