Full History Carryover — n=1,853, 5년 2개월 총 검증
2021-02-28 ~ 2026-04-11 · 1,853 nights · 84 tests · Year-stratified analysis
THE DEFINITIVE NULL
Within any single year, bedtime→HRV/RHR and total_sleep→HRV/RHR are ALL null (24 tests, all p>0.1 except total→HRV in 2024/2026 — within FP expectation).
Cross-year correlations (raw r up to +0.49) are ALL artifacts of a 5-year fitness transformation:
- RHR: 61 → 49 bpm (-12)
- HRV: 38 → 50 ms (+12)
- Sleep: 9.4h → 7.0h
The Only Real Signal
Late bedtime → fewer steps (β=-4.2 steps/min, r=-0.112, p<0.001). 1 hour later = ~250 fewer steps (~5% of daily average).
Year-by-Year Correlations
| Year | n | Onset→HRV | Onset→RHR | Total→HRV | Total→RHR |
|---|---|---|---|---|---|
| 2021 | 303 | +0.019 (—) | +0.051 (—) | +0.035 (—) | +0.090 (—) |
| 2022 | 360 | -0.064 (—) | +0.030 (—) | +0.080 (—) | +0.073 (—) |
| 2023 | 360 | -0.053 (—) | +0.059 (—) | +0.089 (—) | -0.063 (—) |
| 2024 | 364 | -0.076 (—) | +0.035 (—) | +0.141 (★) | +0.075 (—) |
| 2025 | 365 | +0.034 (—) | +0.045 (—) | +0.078 (—) | +0.017 (—) |
| 2026 | 101 | -0.125 (—) | +0.009 (—) | +0.270 (★) | +0.048 (—) |
113-day vs 5-year
| Finding | 113 days | 1,853 nights |
|---|---|---|
| Onset→HRV | null | null (confirmed) |
| Onset→RHR | null | null within-year |
| Onset→Steps | marginal | r=-0.11, p<0.001 |
| Cross-year r | N/A | Up to +0.49 (all confound) |
Why Cross-Year r=+0.49 Is Fake
Simpson's Paradox: pooling 5 years of improving fitness creates phantom correlations. In 2021: sleep 9.4h + RHR 61. In 2026: sleep 7.0h + RHR 49. Both trends are independent (fitness vs lifestyle change) but correlate when pooled.
Implications
The reason to sleep early is NOT about tomorrow's Apple Watch metrics. Your autonomic system resets overnight. The reasons are:
- That night's sleep architecture (REM, Core)
- Long-term accumulation beyond what 5 years of wearable data can detect
- Activity level: late bedtime = ~250 fewer steps
Full charts in full-history-carryover.html.