A list of tips for sleeping better always gives the same impression: ten habits lined up, every one delivered with the same confidence. Cut the screens, eat early, go to bed early, air the room. Nothing tells you which one carries the most weight, nor which rests on a trial and which on a writing habit.
This article ranks them. Some are backed by cohorts of tens of thousands of people with measured sleep, exactly one by a randomised trial designed for it, and the most repeated of them all aims beside its target.
| The habit | What supports it | What it is worth |
|---|---|---|
| Regular schedules | Cohort of 60,977 people, sleep measured | Best supported, and most neglected |
| Lowering evening light | Laboratory trials, dose effect measured | Solid, but the threshold is individual |
| Reading in bed, on paper | One dedicated randomised trial | Modest effect, clean method |
| Removing flicker | Double-blind trial on headaches | Demonstrated, but not on sleep |
| Dinner three hours earlier | Case-control study on gastric reflux | Good for the stomach, poorly demonstrated for sleep |
| A weighted blanket | Randomised trial, negative on objective measures | Real preference, measured gain absent |
Regularity carries more weight than duration
A team computed a sleep regularity index from more than ten million hours of accelerometry across 60,977 UK Biobank participants, then followed mortality for an average of six years. Compared with the least regular fifth, the other four fifths show an all-cause mortality risk 20% to 48% lower.
The result that counts is not that figure. The authors put the two models in competition: regularity predicts mortality better than duration does. That is an inversion of priorities, since all the public conversation is about the number of hours.
What the index measures changes everything: day-to-day consistency of bed and wake times, not the hour on the clock. Going to bed at 3 a.m. every day scores better than an 11 p.m. bedtime that slips by three hours at weekends. I have a late chronotype, and that consistency is exactly the variable I hold steady. When I have to be up at 7, I do not shave the sleep window, I move it.
The result that troubles this reasoning most is a large one. Across 433,268 adults from the same UK Biobank, people describing themselves as definite evening types had a 10% higher all-cause mortality risk than definite morning types, after adjustment for sleep duration, smoking, body mass and socioeconomic status.
What it invalidates: the idea that being an evening type is inconsequential. What it leaves standing: the hour itself is not named as the cause. The authors attribute the excess risk to chronic misalignment between internal timing and the schedules imposed by work, and close by calling for more scheduling flexibility for evening types. Being late and regular is not the situation measured here, since these late participants were getting up early.
Evening light is measured, not guessed
Fifty-five adults spent entire evenings under controlled illuminance, from 10 to 2,000 lux. At 10, 30 and 50 lux, ordinary indoor levels, the onset of melatonin secretion was delayed by 22, 77 and 109 minutes.
The second result is the more useful one: sensitivity varies enormously between people. The most sensitive participant responded at 6 lux, the least sensitive needed 350. A factor of fifty-eight. No single instruction can be right for everyone, which is an argument for measuring at home rather than applying a number found online.
A consensus bringing together the field's main laboratories translated that into three thresholds, expressed in melanopic equivalent daylight illuminance, the unit that weights light by the sensitivity of the retinal cells that inform the internal clock: at least 250 lux during the day, at most 10 lux in the three hours before bed, at most 1 lux during sleep.
Warm, yes, but dim first
The French health agency recommends warm white, below 3,000 K, for home lighting, and notes that measured screens sit between 4,100 and 7,000 K. The logic holds: at equal illuminance, warm light contains less blue and therefore sends less signal to the clock.
It is incomplete, and that is the most common error on the subject. Colour temperature alone does not fix the dose received: a 2,200 K bulb aimed at a white ceiling can deliver more signal than a warm screen held far from the eyes. When researchers tested an iPad's Night Shift mode, melatonin suppression did not differ between two colour-temperature settings. On filtering glasses, a review of 17 randomised trials concludes the effect on sleep quality is indeterminate.
Flicker, the variable nobody discusses
A cheap LED lamp does not produce continuous light. It follows the variations of its supply current and blinks, too fast to be seen. This is temporal light modulation, and the opinion issued in 2019 is explicit on the point that matters: effects can appear without any conscious perception of modulation, among them visual fatigue, headaches and migraines.
The cleanest demonstration dates from 1989. Offices were fitted either with conventional fluorescent tubes, whose illuminance pulsed at 100 Hz with a modulation depth of 43% to 49%, or with electronic ballasts running at 32 kHz, which brought that modulation below 7%. Double-blind, headaches and eyestrain were more than halved under the high-frequency lighting. The staff had no idea when the lamps had been changed.
The agency adds the comparison that matters to anyone choosing a bulb: a large number of domestic LED lamps perform worse on this count than halogen technology. The explanation is physical. A heated filament keeps its thermal inertia from one cycle to the next, so it cannot switch off a hundred times a second.
| Technology | Colour temperature | Flicker | The rest |
|---|---|---|---|
| Entry-level LED | 2,700 to 6,500 K | Often pronounced, invisible to the eye | The cheapest |
| High-end filament LED | 2,200 to 2,700 K | Low, but almost never documented | The common compromise |
| Incandescent or halogen | 2,700 K, warmer still as you dim it | Near zero, through filament inertia | Runs hot, and uses more power |
I have both in my bedroom: a high-end filament LED sold as very warm, and an incandescent bulb, still findable if you look. The incandescent clearly wins on the urge to sleep it creates. I file that as an impression rather than an observation, because I did not alternate blind and three variables move together: colour temperature, power and modulation. What I can say is that the direction of the impression is the one all three variables predict separately. The drawback is measurable: it runs hot.
Reading, and why on paper
This is the best-tested habit on the whole list, and nobody expects it. An online randomised trial allocated 991 people between reading a book in bed and not reading. Among those who completed, 42% reported better sleep against 28% in the control group, a 14-point gap. The trial is pragmatic: no blinding possible, self-reported outcomes, plenty of dropout. What it does have is the thing lists of tips never have, a randomly allocated control group.
The medium, meanwhile, was tested in the lab. Compared with reading a printed book, readers on a backlit e-reader took longer to fall asleep, secreted less melatonin, shifted their clock later and were less alert the next morning.
My personal rule goes further than what has been tested: a novel in the evening, an essay or a self-improvement book in the morning. A text prompting reflection on your own life brings back exactly the thoughts you are trying to set aside, whereas fiction sends the imagination elsewhere. No trial has compared the two genres, so it is a bet, and I am naming it as one.
The bet does have indirect support. Forty-one people with insomnia were allocated between an imagery instruction, which means occupying the mind with a detailed imagined scene, a general distraction instruction, and no instruction at all. Only imagery shortened sleep-onset latency, because a specific mental task takes up enough room to keep worries from returning.
Dinner, a solid rule for the wrong organ
The instruction to stop eating three hours before bed is everywhere. Its most cited source opens by acknowledging the problem: it was being given despite a remarkable lack of supporting clinical evidence. The study compared 147 patients with gastro-oesophageal reflux disease against 294 matched controls. Below three hours between dinner and bed, the reflux odds figure reaches 7.45 compared with four hours or more.
The number is clear, and it is about reflux. Eating late, lying down, letting acid come back up, waking: the link with sleep quality is mechanical and credible, but this study does not measure it. With sleep taken as the direct outcome, food-diary data remain cross-sectional and contradict each other. The rule holds, for a different reason than the one attached to it.
What I do with little to lean on
I sleep tightly wrapped. The bed is a bunker, the duvet is rolled around me, only breathing room is left. That is deep pressure, and it is a common sensory need among autistic people, which I am.
The most rigorous trial is disappointing, and it is better said plainly. Seventy-three autistic children with severe sleep problems alternated between a weighted blanket and an identical usual-weight blanket, two weeks each, measured by actigraphy. No gain in total sleep time, in sleep-onset latency, or in sleep efficiency. Children and parents, however, clearly preferred the weighted blanket.
What that invalidates: that a weighted blanket makes you sleep longer. What it leaves standing: a strong, documented preference, and preference is what makes an evening ritual get repeated rather than dropped after a week. A follow-up of 85 people with ADHD or autism does report benefits on falling asleep, but it is retrospective and conducted by telephone interview, the weakest design in this entire article.
The second habit is stranger: I tell myself a story, not memories, not the day ahead, a fiction. That is the imagery instruction from the trial cited above, arrived at by accident.
Molecules come last
The order is not a moral one, it is practical: a capsule offsets neither a bedroom at 200 lux nor a schedule that slips by three hours. Two hold up for me, and each has its own article: glycine, 3 g forty-five minutes before bed, and magnesium, where the salt matters less than the elemental magnesium it delivers. On bisglycinate specifically, no trial has compared it with other salts on a sleep outcome: it is a digestive comfort choice, not a demonstrated result.
What this article does not say
- That it treats insomnia. When the difficulty has lasted months, first-line care is cognitive behavioural therapy, not a light bulb.
- That every habit here was tested on sleep. Flicker is demonstrated on headaches, late dinner on reflux.
- That the effects add up. No study has tested them together, and a stack is not the sum of effects measured separately.
- That my two rituals count as data. Deep pressure even has a negative randomised trial against it.
« The most repeated piece of advice is almost never the best demonstrated. It is the easiest to write. »
What Helix does
Every habit has a different strength of evidence, a different measured outcome, and an effect that depends on a personal threshold no article can know. That is what the engine computes: what was measured, on which outcome, and what your own run of nights says about your own threshold. A laboratory number points in a direction, it does not replace the measurement made at home.