What the data found

Eight findings, in order of how much they change the picture. Per-year comparisons are Mann-Whitney U against 30-day windows before and after each Ramadan. Cross-year pooling uses a sign test on year direction and a mixed-effects model with a random intercept per year. Multiple comparisons were not formally corrected; year-over-year agreement is the working standard instead.

1. Resting heart rate drops during Ramadan, on both devices, in both measurement windows.Apple −7.1 bpm · Oura −2.6 bpm · p<0.0001→ 01

   Eight of eight Apple years and three of three Oura years agree in direction. The drop is larger in the daytime proxy (−8.8 bpm) than in the sleep window (−2.0 bpm), which is also the biggest open threat to the finding: if Apple's resting-HR algorithm leans on low-activity daytime periods, less daytime activity could lower the number without the heart changing.

2. The HRV "increase" was a measurement-window artifact. The real effect is daytime-only and peaks before iftar.all-day +4.3 ms p=0.003 → sleep window −0.3 ms p=0.91 → daytime +7.4 ms p<0.0001→ 02

   Apple's HRV is an all-day average and Ramadan moves the day later, so the first pooled result was answering a different question. Restricted to the sleep window it vanished and matched Oura's null. Daytime HRV is genuinely higher, +17.2 ms in the 16:00 to 20:00 bin, and active energy is lower in the same bins, which favours "less afternoon activity" over a fasting-specific autonomic story.

3. Sleep shifts later. Oura says there is more of it; Apple says there is not.bedtime +0.64 h · wake +0.61 h · Oura +0.58 h asleep, +0.79 h in bed→ 03

   Both timing shifts are pooled-significant on Apple (p=0.0025, 0.0047). Oura shows more total sleep in all three of its years and more REM (+1.9 points of the night, 3/3 years). Apple's total-sleep model is flat (p=0.91). Unresolved; Apple's Watch-only sleep coverage in those years was 17 to 18 nights per window, so it may simply lack power.

4. Structured exercise drops by about two thirds.workouts ~3.2/wk → ~1.1/wk · 8 of 9 years→ 04

   Both count and minutes, mixed model p<0.0001. Oura's active energy is consistently lower (3/3 years); Apple's is pooled-significant but the sign test is not (7 of 9 years lower, p=0.18), most likely third-party workout apps contaminating Apple's activity record rather than the behaviour varying.

5. App-logged meditation goes down, the opposite of what I would have said.−4.1 min/day · p<0.0001 · 2024 −54%, 2026 −94%→ 05

   Headspace minutes fall during Ramadan in the two cleanest data years. A 2025 sync bug that produced one-minute fragments was checked and is confined to that year. The likely explanation is a shift to practices the app does not log, not that the self-report is wrong.

6. The phone's GPS says Ramadan is more home-bound, not more social. The iftar-and-taraweeh hypothesis fails.2025 · 1 vs 3 places/day p=0.001 · late nights out 7% vs 33% of days→ 06

   Fewer distinct places per day, fewer late-night hours away, earlier last return home (18:27 vs 00:49 median). Total away time is not significantly different. One recurring Friday-evening outing survived; every other weekday's evening time fell. Two Ramadans of GPS, one with a proper baseline, so this is a single controlled year.

7. Waking later is a recent, deliberate habit, and it takes about three weeks to wear off.2025 +0.7 to +0.9 h · 2026 +1.3 to +1.6 h · 2021, 2023, 2024 no shift→ 07

   Three sources (Watch, ring, phone accelerometer) agree on the 2025 shift, so it is not a scoring artifact. Pooled over the shifted year-source pairs, the post-Eid offset is +1.0 h in week one, +0.5 in week two, +0.3 in week three, then about zero. Per pair it is messier: two of five never returned to their pre-Ramadan band within 75 days.

8. Nothing here is season-controlled, and that is the largest limitation of the whole study.Ramadan moved from mid-May to mid-February over the nine years→ caveats

   A matched same-calendar-date non-fasting control was attempted and found infeasible with nine years of dates. Every finding above is potentially confounded with season. Also single subject, observational, no diet or mood data by design.

Resting heart rate WatchOura

The headline result and the most replicated one. Two devices, two sensors, two algorithms, and on Apple two separate measurement windows. Each chart below lays every year on the same axis, with day 0 the first day of fasting and the month shaded. The line is the median of all years at each day. The column on the right gives each year's change against the 30 days before; hover a year there to see its own trace.

• Every Apple year with data is lower during Ramadan than in its pooled baseline. Against the month before alone, seven of eight years are lower, from −1.5 bpm (2026, n.s.) to −14.5 bpm (2023); 2020 is +2 bpm and not significant. Against the month after, the gap is often larger: post-Ramadan resting HR runs high in several years (2022 post-window mean 79 bpm), so the data cannot say whether Ramadan lowers it or Eid raises it.
• Oura, which measures only during sleep, sees a smaller but unanimous drop. The Apple sleep-window proxy, built from raw heart-rate samples inside the scored sleep window, lands in the same range (−1.95 bpm). That agreement is the best evidence the effect is not just an algorithm quirk.
• The daytime-only proxy drops four times as much. That is consistent with a real cardiovascular change and equally consistent with less daytime movement making "resting" easier to find. Section 04 shows active energy is lower in the same hours. This was not resolved.

HRV, and a correction WatchOura

The first pooled pass said HRV rises during Ramadan (+4.25 ms, p=0.003). That number was computed correctly and was still wrong, because Apple's HRV is an all-day average and Ramadan moves bedtime and wake time about 40 minutes later. Change when the readings land and the average moves without anything physiological changing. Rebuilt inside the scored sleep window, the effect disappeared.

• The bins tell one story twice. HRV is higher in every daytime bin from 08:00 onward and the effect grows through the afternoon, peaking pre-iftar. Active energy is lower in exactly the 08:00 to 20:00 bins (p ≤ 0.002 each) and unchanged overnight. Higher HRV where there is less movement is what you would expect from sitting still, before invoking fasting physiology.
• This matches the self-report that the later part of a fasting day is the lowest-energy part, spent reading or resting. It does not rule out an autonomic effect of fasting. It means the simpler explanation was tested and not eliminated.
• The correction was only caught because the sibling Apple-vs-Oura comparison had already established that the two devices define HRV over different windows. Re-applying a known confound from one project to another is what found it.

Sleep WatchOura

Timing shifts later on both ends, which the self-report confirms as deliberate. Whether there is more sleep depends on which device you ask, and the two have not been reconciled.

• Oura's REM share rises 1.9 points in all three of its years (p=0.003). Nothing else in stage composition moves on either device.
• The Apple-vs-Oura duration disagreement is the one finding left genuinely open. The self-report (deliberately protecting sleep, cutting other obligations to do so) tips it toward Oura being right and Apple underpowered. Section 07 adds a third source, the phone, on the timing half of this.

Exercise and activity WatchOura

The self-report says "considerably less, if at all." The workout log agrees precisely, and the two devices' activity records agree in direction while disagreeing about how cleanly.

• Baseline sits around 3.2 workouts a week; Ramadan around 1.1. Minutes fall the same way. 2019 is the one exception in direction.
• Apple's active energy is the cautionary tale for pooling. The mixed model gives −119 kcal/day at p<0.0001, but seven of nine years are lower and two are higher, and the sign test is not significant (p=0.18). Apple's activity record includes third-party workout apps that had to be filtered; Oura's does not, and Oura shows −222 kcal/day with every year agreeing. The pooled p-value alone would have overstated the Apple result.

Meditation minutes Watch

I would have said I meditate more during Ramadan. The app log says less, by a lot, in the two years with clean data.

Where I was Phone GPS

Added after the main study. The phone runs Beiwe, the research platform I work on, which samples GPS passively. The question was whether Ramadan evenings are busier (iftar, taraweeh, later returns home) and how that sits next to the sleep shift. GPS starts in March 2024, so 2024 is Ramadan-only and descriptive; 2025 has a full 30-day window either side and is the real comparison. No coordinates appear anywhere in the outputs; "home" is a 150 m disc around a monthly home cell.

• The hypothesis was more evening and late-night movement. The record shows the opposite. The Ramadan curve sits below both baselines from 17:00 through the small hours, and overnight away time roughly halves. The only thing that rose is a 15:00 to 17:00 sliver.
• One outing survived: Friday evenings, 202 away-minutes in Ramadan versus 84 before and 141 after. Every other weekday's evening time fell.
• This sits consistently next to the companion communication study, which found more unique contacts reached and more message volume during Ramadan. More checking in by phone, less going out.
• 2024, uncontrolled, shows a quiet month with evening and late-night time concentrated in the last third (days 21 to 30: 96 evening and 29 late minutes per day, versus about 7 and 0 earlier). 2025 shows the same within-month gradient from a lower base. A third GPS year would make that testable.

Waking later, and for how long WatchOuraPhone

Section 03 found wake time later on Apple, pooled over 2021 to 2026. This follow-up asks two narrower questions. Is the shift the same every year? And after Eid, how long until mornings go back to normal?

• Apple 2021 and 2023 show no wake shift (+0.15 h, +0.02 h). 2024 shows none on either device. 2025 is +0.9 (Apple), +0.7 (Oura), +0.9 (phone). 2026 is +1.6 and +1.3. That matches the account that sleeping in during Ramadan is a recent, deliberate habit, not a long-standing one.
• The phone estimator knows nothing about sleep stages and sees the same 2025 shift, so the ring is not inventing it. Bedtime did not move in 2025 on either the ring or the phone (all p>0.24). The extra sleep Oura reports is on the morning side.
• Pooled over the five shifted year-source pairs, the post-Eid offset is +1.0 h in week one, +0.5 in week two, +0.3 in week three, about zero from week four. Treat three weeks as the central estimate, not a per-year fact: Apple and Oura 2025 are back within 8 to 19 days, while Oura 2026 and the phone in 2025 stay +0.6 to +1.0 h for the whole 75-day window. Two sources disagreeing on the same nights means the baselines themselves differ between devices, and a spring drift toward later waking that has nothing to do with Ramadan would look identical.
• 2024's post window reads as six hours early on both devices. That is an Eid trip abroad with wake times still expressed in Eastern time, not a real early-waking spell.

Where each number comes from