How Many Nights Does Melatonin Need Before You Can Judge It?

/// AUDIO TRANSMISSION · ES

Listen to this transmission · 2-host podcast version

0:00 / 3:52

28 nights. The published effect of melatonin on sleep efficiency is 3.7%. Our own threshold at 28 nights is 2.43%. The published effect sits above that threshold. So a window of 28 recorded nights is enough to judge whether melatonin moves your sleep efficiency. You do not need a year of data. You need 28 nights.

That is the short answer. The rest of this file shows where each number comes from, how the threshold was computed, and what this data does NOT say. Every figure on this page traces back to a published source row. None of them is rounded «toward the round number». If a figure is not in the row, it is not in the text.

Where does the 3.7% come from?

From a randomized, double-blind, placebo-controlled trial: 59 healthy adults with poor sleep took sustained-release melatonin at a dose of 2 mg. The outcome was measured with polysomnography, the lab instrument that records sleep signal by signal. The result: sleep efficiency rose +3.49 pp against placebo. In relative terms, a change of 3.7%. The reported time to effect was 28 days. The citation is Thanawala 2026, Clocks & Sleep 8(2):31, DOI 10.3390/clockssleep8020031. The source row in our vault is catalogo.generado.json#evidencia[12].

It helps to define the metric. Sleep efficiency is the share of the night you spend asleep, out of the total time you spend in bed. It is a demanding metric. In our records, the median is 93%.

What is the 2.43% threshold?

It is the half of the math almost nobody publishes. An effect can only be judged if it beats the noise of the metric where you look for it. Our archive holds 650 nights of sleep efficiency from one single body, inside a vault that follows 31 variables across up to 712 nights. Those 650 nights yield the noise numbers: a median of 93%, a coefficient of variation of 4.38%, and a night-to-night autocorrelation of 0.058. From those, we compute how much the average must move, and across how many nights, for the change to stand out from the normal swing.

The result is a threshold per window: the smallest detectable change for the number of nights you record. With more nights, the threshold drops, because the average of many nights is steadier than the average of a few. The table comes straight from the source row; it was not written by hand:

Window (nights)	Detectable threshold on sleep efficiency
28	2.43%
56	1.72%
84	1.4%
112	1.21%
140	1.09%
168	0.99%

The same row holds one more useful number: the rcv, 12.13%. That is the reference swing between single nights. Comparing two isolated nights tells you nothing: the normal swing from one night to the next is far larger than the effect you are hunting. Only the average of a full window can see a change of this size.

Why 28 nights and not fewer?

Because two things happen at 28 nights. First, our own threshold drops to 2.43%, and the published effect, 3.7%, sits above it. Second, the trial itself reports 28 days as the time to effect. Both routes — our measured noise and the published literature — point at the same window.

What about fewer nights? The threshold rises, and the 3.7% effect stops standing out from the noise. One week of records is not enough: that is too few nights for the average to separate from the normal swing.

What about more? More nights never hurt. At 56 nights the threshold drops to 1.72%; at 84, to 1.4%. If you want more resolution, keep recording. But the question in this file is when you can make a first honest call, and that answer is 28 nights.

The practical use is simple. Record your baseline without the compound. Then record the window with melatonin, night by night, without skipping the bad ones. Average the sleep efficiency of the full window and set it against your baseline. If the change beats 2.43%, it moved more than your noise. If it does not, then for your body and on this metric, there is no signal to defend.

The limit

What this data does NOT say. This section is part of the house format, not a footnote.

• The 2.43% threshold comes from 650 nights of one single body. It is not a universal constant. It is the noise of this archive. Your noise may differ.
• The 3.7% comes from 59 healthy adults with poor sleep, on sustained-release melatonin at 2 mg. It says nothing about other doses, other release forms, or other groups of people.
• The source row also records what the compound did not do: it did not move HRV or recovery. An effect on sleep efficiency does not spread to the rest of the body on its own.
• 28 nights judge sleep efficiency. They do not judge "sleep" as a whole, or how you feel at waking, or anything outside the metric.
• Detectable does not mean meaningful for you. The data says whether the change can be seen. Whether that change is worth it is your call.

And the verdict can be null: you run the 28 nights, you average, and the change does not beat 2.43%. That is also an answer, and it also gets published. A null verdict is not a failure of the method. It is the method working.

Follow the file: Mental Work — all protocols · related file: magnesium-for-sleep · open science — the dataset