Infant sleep doctrine, scored by the odds it’s true

The question "which sleep practice is best for her long-term development" has a scale problem before it has an answer. Measured In school-age children, sleep duration correlates r = .08 with cognitive performance and r = .09 with behaviour problems across 86 studies and 35,936 children, and not at all with intelligence (Astill 2012); with objectively measured sleep the figure is r = .06 (Short 2018). That is under 1% of variance, observational, before confounding is removed.

For infancy specifically, the best systematic review (22 studies of sleep at 0–18 months against Bayley, WPPSI and similar tests) found 53.5% of about 101 tested associations null, 39.6% mixed and 5.9% consistently significant, two of those six in the wrong direction. Study power ranged from 0.06 to 0.88; one study reached 0.80 (Butler 2024). Sleeping through at 6 or 12 months had no association with Bayley scores at 6, 12 or 36 months in 388 infants (Pennestri 2018); three or more wakings a night at 8 months made no Bayley-III difference at 8 or 24 months (Mäkelä 2018).

The positive signals that do exist share a signature. Outcomes measured by direct testing are mostly unrelated to infant sleep. Outcomes rated by the same parent who rated the sleep show small associations: 0.14–0.18 SD higher mother-rated internalising after sleep problems at 6 months in 86,724 Norwegian infants (Valla 2021). Teacher-rated outcomes shrink toward zero: effect size −0.12 for infant-limited sleep problems at age 10–11 in 5,107 Australian children, with no difference in school achievement (Williamson 2020). That is what shared-reporter variance plus modest confounding produces.

The associations also sit in the tails. The shortest-sleeping 18% across 3–24 months scored 5.2 Bayley points, about 0.35 SD, lower at age 2 (Smithson 2018); combined sleep, crying and feeding problems carried d = 0.41 for later behaviour problems across 22 studies (Hemmi 2011). She is in none of those groups. Several widely quoted results are exposure-window mismatches: the Quebec short-sleep finding used sleep from age 2.5 to 6 (Touchette 2007), the Norwegian result is from 18 months (Sivertsen 2015), and the bedtime-regularity findings are from ages 3–7 (Kelly 2013).

Two designs test causation, and neither supports a large effect. The only genetically informed analysis attributes the link between day/night sleep at 6 months and later language mainly to shared genes (1,029 twins, Dionne 2011). Randomised trials that changed sleep parenting produced no child difference in behaviour, sleep, stress or relationship measures at ages 2 and 6 (Hiscock 2008; Price 2012); BMI reductions in two trials faded by ages 9–11 (Paul 2025; Taylor 2025). No trial has measured cognition or language after an infant sleep intervention.

Modelled Replacing the entire rearing family raised IQ at 18–20 by 4.4 points in a Swedish sibling design (Kendler 2015), and mean heritability across traits is 49% (Polderman 2015). One practice difference inside an already favourable home (45- versus 60-minute naps, rocked versus self-settled onset, one night feed versus none) therefore plausibly moves long-term cognition by well under one IQ point. That bound is an inference from behaviour genetics, not a measurement.

Measured US sudden unexpected infant death (SIDS, plus unknown cause, plus accidental suffocation and strangulation in bed) ran at 93.6 per 100,000 live births in 2024 and has not improved since about 2000 (CDC). In the CDC case registry, at least 76% of deaths involved multiple unsafe sleep factors and 0.9% had none documented (Erck Lambert 2024).

The cause mix shifts with age. Among 250 explained suffocations, overlay had a median age of 2 months with none after 7 months; soft-bedding deaths a median of 3 months; wedging a median of 6 months, most often between mattress and wall or mattress and headboard. In infants aged 5–11 months, blankets obstructed the airway in 55% of soft-bedding deaths (Erck Lambert 2019). A comparison of younger and older victims concludes that "rolling into objects in the sleep area is the predominant risk factor for older infants" (Colvin 2014). That is her age band and her new skill.

Four caveats bound the bed-sharing row. The two estimates for older infants rest on few exposed cases (Blair's on one), though a third analysis also found none (1.07, Tappin 2005), and US surface-sharing carries an adjusted OR of 2.1–2.5 without adjustment for alcohol or drugs (Parks 2023). All hazard-stratified data pre-date 2007. SIDS case-control studies under-capture accidental suffocation and entrapment, which is exactly the mechanism that remains in adult beds at 5–8 months; unadjusted 1990s product-safety data put suffocation risk 20–40 times higher in adult beds than in cribs, with no adjustment for alcohol, smoking or sofas (Scheers 2003). And no study has examined bed-sharing by mothers who feed pumped milk by bottle: the protective behaviours seen on video (infant at breast height below the pillows, mother curled around her, synchronised arousals) were observed in mothers nursing at the breast, while 71% of formula-fed infants lay at face height on or between pillows (Bartick 2022; Ball 2006).

Bed-sharing, applied to this family

At about 4 months she is past the 15-week window in which Carpenter finds significant hazard-free risk, though not by a wide margin. Here the US absolute and the UK risk-stratified position diverge, and the UK position fits the hazard-stratified data better. Finishing the night in the parents' bed is the pattern English investigators coded as "taken to and found in parental bed": elevated before adjustment, not significant after adjusting for alcohol, heavy duvets, parental tiredness and crowding, and not significant beyond 14 weeks or for non-smokers (Blair 1999). The exposure that matters is the unplanned 4 am transfer, made when tiredness peaks, into a bed nobody prepared. Room-sharers have four times the odds of exactly that (Paul 2017).

Scored the same way as every other claim: P 0.38 that a prepared bed, with a sober, non-smoking, breastfeeding mother, raises her all-cause sudden-death risk at least 1.5-fold over a crib in your room. If it does, the absolute addition is a few per 100,000. P 0.93 that bed-sharing after alcohol or sedatives sharply raises it.

Judgement Decide each evening. Either the bed is set up and both adults are sober and rested, in which case bringing her in after the night feed is a defensible choice on current evidence, or it is not, in which case she goes back to the crib without exception. The mother returns her to the crib if she herself feels unusually depleted. A mother who nurses at the breast at night and has bed-shared before plausibly keeps the protective behaviours; the parent who gives a bottle should not be the one beside her.

The federal bassinet rule incorporates ASTM F2194, whose scope is sleep "for an infant up to approximately 5 months in age or when the child begins to push up on hands and knees, whichever comes first", with a required fall-hazard warning at that point or at the maker's weight limit. Weight limits are typically 15–20 lb. Some manufacturers set the stop earlier: Chicco says to stop once the baby can roll over or push up "even if they have not yet reached the maximum weight limit", and an Evenflo bassinet insert says stop when the baby can "push up on hands and knees or can roll over".

The mechanism is physical. Shallow sides and a high centre of gravity make falls and tip-overs possible once she pushes up, and a small soft-sided space lets a rolled infant end up face-against-side or in a corner, worsened by any tilt. CPSC logged more than 200 bassinet incidents, including 7 deaths, from 2017 to 2022 (Consumer Reports 2024). Limits of this evidence: no count of deaths specific to rolling in an over-age bassinet was found, no study compares bassinets with cribs by age, and the wording of the 2025 standard update could not be verified.

Judgement P 0.73 that staying in the bassinet past rolling raises her risk: no death series exists for over-stage use, but the mechanisms are concrete and the cost of acting is a crib. The most permissive reading of her model's label buys a few weeks at most. Read the label tonight; if it names rolling, she is already past it.

The compliant replacements that keep her in your room are a full-size crib (16 CFR 1219), a mini crib (16 CFR 1220) or a play yard (16 CFR 1221), each with the manufacturer's flat mattress or pad, a fitted sheet and nothing else. Mini cribs and play yards are rated until the child reaches 35 inches or can climb out, which covers the whole first year (ASTM F406). A bedside sleeper does not solve the problem, because its standard carries the same 5-month, hands-and-knees limit (ASTM F2906); neither does a SNOO, in-bed sleeper, pod or nest.

Sitting-device deaths were 2.8 times as likely (95% CI 1.5–5.2) under a child-care provider's supervision, and 2.0 times under a babysitter's, as under a parent's (Liaw 2019). A new or unusual sleep location was 6.5 times as common in sofa deaths (Rechtman 2014). A hired caregiver holding a sleeping infant for two to three hours a day in a comfortable chair is a foreseeable drowsiness scenario. Hand this card over, and run it yourselves.

Her pattern (an 11-hour night with one feed and immediate resettling, 45-minute contact naps) is at or above the reference mean on every available measure, and every sleep-intervention efficacy trial enrolled infants with a parent-reported problem she does not have. The settings below are mostly "keep", each with its reason and evidence class.

Rank-order stability of infant sleep is weak beyond a few months: night sleep at 3 months correlates 0.53 with 6 months and 0.24 with 12 months (Schoch 2020). A good 4-month sleeper is likely to stay in the low-problem majority: in one birth cohort, 6% of infants without an early sleep problem had one in the third year, against 21% of those with one (Byars 2012). Modelled Prevalence shifts between 3 and 9 months imply that between one in five and one in three settled infants has a noticeable stretch of more waking in the second half-year; no cohort reports that fraction directly. Lasting relapse after 6 months was rare in the one cohort that tracked it (1 of 29, Henderson 2020).

Video and actigraphy agree that essentially all infants wake two to four times a night throughout the first year; 78–90% of those awakenings are vocalised, and the commonest single pattern is "vocalised, no intervention" (Goodlin-Jones 2001). What separates "good" from "poor" sleepers is whether a parent arrives.

Four observational findings point the same way. In 107 families filmed at 1 and 3 months, more responses to infant distress went with a steeper decline in night waking to 9 months, while in solitary sleepers more interventions when the infant was not distressed, or was asleep, went with a slower decline (Voltaire & Teti 2018). In 80 infants filmed across the year, longer parental response time at 3 months (mean 3.3 minutes) was one of three predictors of unaided resettling at 12 months (Burnham 2002). Bedtime involvement at 4 months predicted rising night wakings to 8 months in 207 families (Matzliach 2025). In camera data from 3,158 infants, each additional parental crib visit raised the odds of the parent perceiving a sleep problem by 18% (Nanit-funded, cross-sectional, Kahn 2025).

Scored: P 0.70 that the pause increases unaided resettling. All of it is observational, part of it runs backwards (poorly sleeping infants draw parents in, Simard 2008), and the one trial that advised waiting before attending found no objective effect (Santos 2019). But four independent studies agree, a responsive-parenting trial that taught “give a few minutes” gained 40 reported minutes (Lavner 2023), and operant learning in young infants is well established. The outcome at stake is signalled waking, the parents' experience, with no demonstrated bearing on her long-term development. Impose the pause deliberately for stirring, grunting and brief calls, and respond without delay to real crying. This is not extinction, and no source shows it harms.

The same error has other forms in a high-resource household, each buying nothing measurable: rocking through every 45-minute arousal by default; continuous tracking and physiological wearables (5% of 1,128 coded product reviews described adult stress, fear or lost sleep from false alarms, preprint, not peer-reviewed); iron drops without a measured deficiency; overnight pumping to match milk; keeping her awake to hit a schedule; and rebuilding the routine in response to a two-week motor-milestone disruption. Effort has far better-evidenced targets in her waking hours.

The unknowns sit exactly where this family's discretionary choices sit.