Overnight HRV

Overnight HRV#

What your device delivers#

Most wearables report one number per night: the average heart rate variability during sleep, in milliseconds. The measure is called RMSSD (the typical size of the change between neighbouring heartbeats), and the app labels it "HRV (RMSSD)". We accept a night between 5 and 250 ms and ignore anything outside, as a sensor error. One night is kept per date.

Some devices also report the best five minutes of the night (used by Nervous System) and a usual range of their own (Garmin does). The device's own weekly figure is not used for your week: we compute the week from your nights.

Different devices measure HRV on different scales, so 40 ms from one is not 40 ms from another, and nobody's number is compared with yours. The only comparison that carries information is you against your own history. For the same reason, if two of your devices both deliver HRV, see two devices.

Only the overnight HRV is scored. A reading taken after waking, such as a watch's two-minute morning measurement, is not used in any score.

The week, not the night#

A single night is noisy: a late meal, alcohol or a poor position of the sensor can move it by a lot, and in research the weekly mean tracks changes in running performance far better than one day does (Plews and colleagues, 2013). So the number we score is the week.

1. For each night we take the natural logarithm of the HRV. HRV is not spread evenly, and the logarithm makes averaging fair.
2. We average the nights that fall in the last 7 calendar days, counting the day being scored.
3. We convert the average back to milliseconds. This is the "7-night average" the card shows as the large number, labeled "mean of 5 nights this week".

The week needs at least 3 nights. With fewer, the card shows a counter such as "2/3 nights this week" and says "2 of 3 nights this week, the mean shows from the third". Three nights is the least a week can rest on in the studies we follow (Plews and colleagues, 2014). Days count, not rows: a night from nine days ago is not in the week, however few nights there are.

The ladder and your own shift#

Exact calculationThe ladder and your own shift, step by step

To turn the weekly milliseconds into a 0 to 100 score we use two steps.

Step 1, a fixed ladder. It is our own curve, not a population norm: 100 / (1 + e^(-0.06 × (ms - 40))).

Weekly HRV	Ladder score
20 ms	23
30 ms	35
40 ms	50
50 ms	65
60 ms	77
80 ms	92
100 ms	97

Step 2, a shift from your own history. We take where your week sits among your own weekly averages of the last 60 days (see your baseline). At your own mean nothing moves. At two standard deviations away (a standard deviation is how widely your numbers usually scatter) the score moves by 20 points, up or down, and never further. Until you have at least 7 weekly averages the shift is skipped and the ladder alone stands.

The shift exists because the same milliseconds mean different things on different devices: a Fitbit wearer and a Garmin wearer in the same shape sit tens of milliseconds apart.

Your baseline#

Exact calculationHow the baseline is built

The baseline is built from your nights of the last 60 days:

• for every day that has a night, we compute the 7-night average ending on that day, but only when the week holds at least 3 nights;
• the baseline is the mean of these rolling weekly averages, and its spread is their standard deviation, how widely they scatter;
• it exists once there are 7 such weekly averages.

Nine nights in a row are enough to give you 7 weekly averages (they overlap), so the shift above can start early. The band, which is a stricter promise, waits longer.

Your band#

The band is the range inside which a week counts as "your usual". Its edges are half a standard deviation either side of the baseline, where the deviation is that of your single nights (on the logarithmic scale). Half a standard deviation is the smallest change that research treats as worth noticing (Plews and colleagues, 2012; Buchheit, 2014).

The band appears only when your nights span at least 28 days from the first to the last in the 60-day window, and the baseline exists. Until then the card says "Learning your band" and counts down: "12 more days at the earliest", or "It needs more nights" when the span is long enough but the nights are too sparse.

Where the week sits against the band is written as one of these, with the rule behind it:

Exact calculationWhy the score and the band use different spreads, with an example

The score and the band use two different spreads on purpose. The score uses how much your weekly averages vary, which is the right ruler for a week. The band uses how much single nights vary, which is the published smallest worthwhile change, and is about twice as wide.

Example

Your baseline is 50 ms, your single nights scatter by 0.20 on the logarithmic scale, and your weekly averages by 0.08. This week's nights are 44, 47, 41, 45 and 43 ms.

• Week: the average of the logarithms, converted back, is 44 ms.
• Band: 50 × e^(-0.10) = 45.2 ms to 50 × e^(+0.10) = 55.3 ms.
• Position: ln(44 / 50) / 0.20 = -0.64. That is below -0.5, so "below your band", and since it is within 1, "Slightly below your band". The card says it is 1 ms below the edge (the distance between the printed 44 and the printed 45).
• Score: the ladder gives 56. Your own shift is ln(44 / 50) / 0.08 = -1.60 weekly spreads, so -1.60 / 2 × 20 = -16. The week scores 56 - 16 = 40.

Last night, and how it fades#

Exact calculationHow last night counts and fades

The newest night is scored on the same ladder and shifted by where it sits among your single nights. It then counts by how old it is, in days from the scored day:

Age of the newest night	Share it keeps
0 or 1 day	100 %
2 days	60 %
3 days	30 %
4 days or more	0 %

Recovery uses it as a smaller term next to the week (Recovery). The week is never replaced by it: a very bad night is real news on the morning it happens, but it is one night. Nervous System uses the newest night in place of the week only when the week has fewer than 3 nights and the newest night is less than 7 days old.

In the example above, last night of 41 ms scores on the ladder at 51 and is shifted by ln(41 / 50) / 0.20 = -0.99 single-night spreads, that is -9.9 points: 41.

Two devices, the phone, mornings#

• One device at a time. If your nights come from two devices, the scores read the nights of the device that wrote your newest night, each device against its own band. A device on another scale never joins your window, so changing devices does not look like a collapse of your HRV. If the second device's own week sits on the same side of its own band, the card adds "[Device] agrees: its week is below its own band too", otherwise "[Device] does not confirm it ... It may be one device's swing; watch whether the next days confirm it." We never print the second device's milliseconds beside yours.
• The phone yields to a wearable. Overnight HRV that the phone averages from Health Connect is on its own scale. While a wearable has written a night within 30 days before your newest night, the phone's nights leave the window. Past that, you have moved to the phone, and the wearable's old nights leave instead.
• Garmin's own band. A Garmin wearer sees a band in Garmin Connect too. Garmin does not describe how it draws it, and ours comes from your nights over the last 60 days, so the two can differ. The card says so.

The card and what it says#

• Large number: the week in ms, in the color of its state, with "mean of 5 nights this week" beside it.
• Verdict box: one sentence, for example "Your nights are slightly below your band this week.", then a line on what it means ("It is just past your usual variation." or "It is more than your usual variation."), and, for a low week on a card without the coach's plan line, the advice "An easier day makes sense; what counts is whether it holds over the next days." A week above the band reads "Above the band means more than usual; on its own it is not a warning."
• Resting heart rate line: see the resting heart rate card. A resting rate that is up beside an HRV under its band deepens the color one step.
• Rail: a ruler with a pin for the week against the band, and "weekly variation 4.2 %", which says how widely the week's nights scatter.
• Charts: "14 days" (the week on screen beside the week before it), "30 days", and "HRV and HR", a small map that trails the week ending 12, 8, 4 and 0 days back against your resting heart rate.
• Plan line on the Health tab: "Today's plan: easy", the coach's verb, never advice of the card's own. In the Overview pair the plan is left to the coach.
• Empty states: "Too few nights" when the week has under 3 nights but the band exists, "Learning your band" while the band is being built. On a Garmin account with no night in the last 60 days the card stays, empty, with the way to get the data. On any other account the card disappears when the newest night is more than 14 days old.

What the band does to today's plan#

A week below the band caps the coach's verb for the day at EASY, however green the rest looks. If the week has been below the band for at least 14 days in a row, with no hard session yesterday and a weekly load no higher than your month, the cap becomes MODERATE instead, because easy days answer training fatigue and cannot answer poor sleep, stress or an infection; the banner then says so. A week within or above the band changes nothing here. See the daily coach. Body Status also reads the band: a week below it vetoes a Peak Day (Body Status, Peak Day).

What moves your HRV#

Under the card sits a folded panel, "What moves your overnight HRV". For each factor it compares your nights after the factor with your other nights over the last 90 days, and prints the difference in per cent.

Exact calculationHow a difference is judged

The difference is the average of the logarithms of the first group against the second, converted to per cent. A figure appears only with at least 3 nights with the factor and 5 without it. It is called "clearly" when it is big against the nights' own scatter (a t value of 2 or more) and at least 3 %, "a hint" from a t value of 1, and "within the noise" below that. A clear drop is drawn in warm color. Alcohol and illness compare only against nights that have a check-in, because a night with no check-in is not a night without a drink. The panel shows when you have at least 8 nights in the window, and the folded line shows the strongest clear difference ("Most: After alcohol -12 %").

It reports what came before a low or high night, not what caused it: the words say "after", never "because of".

Frequently asked questions#

Why is my HRV different from my watch's own app?#

Our week is the average of your last 7 nights on a logarithmic scale, and our band is built from your nights of the last 60 days. Your watch may show last night alone, or its own weekly figure, and may draw its own range by a method it does not describe. The nights are the same, so the direction should agree; the numbers and the range need not.

Why does it say "Learning your band"?#

The band needs nights that cover at least 28 days, from your first night in the last 60 to your newest, and a baseline of at least 7 weekly averages that each hold 3 nights or more. The card counts down the days it needs at the earliest. The Nervous System score starts earlier, because its personal shift needs only the 7 weekly averages.

Why is my HRV week "below my band" when it is only 1 ms lower?#

The band's edges are fixed by your own spread, so a week very close to an edge can fall just outside. The card then says "Slightly below your band", and when the printed numbers are level it says "At the lower edge of your band". A week just outside is not an alarm: what counts is whether it holds.

Does a low HRV mean I am ill?#

No. A low week can come with a hard block of training, poor sleep, alcohol, stress or an infection, and HRV cannot tell which. When it stays under your band for two weeks without more training, the coach says that training is unlikely to be the cause and points you to sleep, stress or a starting illness. If you feel unwell, rest and see a doctor if it does not ease.