Your baby was sleeping — and then stopped. The word "regression" explains it beautifully: an age, a cause, a duration. We opened the papers this story usually rests on, and most of that beautiful picture was not in them. Here is what is known, what is not, and what to do while the cause is unknown.
It names any sudden worsening of sleep in a child who used to sleep fine. Convenient — but there is no diagnosis and no measurable entity behind it. The same word covers a routine that is reorganising, a house move, the first day of an illness, teething, and simply a week with a lot of visitors.
You can check this yourself. Searching the exact phrase "sleep regression" in PubMed titles and abstracts returns zero publications (checked 16 Aug 2026; for comparison, "sleep consolidation" returns 225 and "night waking" 394, so the phrase index is working). The term appears in no clinical document of the American Academy of Pediatrics or the American Academy of Sleep Medicine.
We went looking for what the best-known part of the calendar rests on in the peer-reviewed literature. Two things turned up, and both are worth knowing.
First. In a recent systematic review of infant sleep architecture (Lenehan et al., "The Architecture of Early Childhood Sleep Over the First Two Years", Maternal and Child Health Journal, 2022/2023; 93 studies, about 90,000 children) the phrase "sleep regression" appears exactly once — and the reference attached to it points not to research but to a consumer sleep website. Even where the term made it into a scientific text, what stands behind it is a popular page, not a measurement.
Second. The one sentence in that review which gives an age window near four months reads:
17–20 weeks is exactly "four months". We opened the source that sentence cites. The paper by Giganti and colleagues (Physiology & Behavior, 2001) is titled "Polygraphic investigation of 24-h waking distribution in infants", and its abstract says:
So: twelve infants aged from 2 weeks to 11 months 3 weeks, and "between 17 and 20" refers to the hours of the day — the stretch from 5 to 8 in the evening. The paper is not about age at all: it is about the "evening forbidden zone for sleep", the well-known adult phenomenon of finding it hardest to fall asleep in the early evening. The reviewers read clock hours as weeks of life, and evening wakefulness as night waking.
There is independent support that this is a misreading rather than a typo: in the review's own table of included studies, Giganti's sample is listed as "2 weeks to 11 weeks & 3 months" — weeks and months swapped again. And a third distortion: Giganti measured wakefulness in the evening, while the review calls it night waking. Three distortions in one citation.
Sleep structure really does mature in infancy, and the same review describes it: across the first two years the share of NREM sleep increases and REM decreases, "in tandem with development". But that is gradual maturation over two years, not an event in one week, and it cannot be pinned to a date.
And where sleep at this age has been measured directly, the picture is the opposite of the calendar:
| What was measured | Result |
|---|---|
| Longest self-regulated sleep period, diaries of 75 infants, monthly across the first year (Henderson 2010) | The largest increase occurred from 1 to 4 months — a mean of 504 minutes. "The most rapid consolidation in infant sleep regulation occurs in the first 4 months" |
| Actigraphy in 57 families at 3 and 6 months (Tikotzky 2015) | Between 3 and 6 months, moderate improvement in infant sleep |
| Regular night waking, 118 infants, surveyed at 3 / 6 / 9 / 12 months (Scher 1991) | 46% → 39% → 58% → 55%. A decline by six months, and a rise towards nine months, not four |
| Diaries over 13 consecutive nights at 6 months (Pennestri 2020) | In 72.7% of infants the longest sleep stretch varied widely from night to night. Variability is normal, not a fault |
An honest caveat: none of these studies was designed to hunt for a dip precisely at four months, and a short dip falling between measurement points would be invisible to them. This is absence of evidence for a worsening, not evidence of its absence. But there is also no basis for promising parents a worsening by the calendar.
The practical difference is large. "A regression starts at 4 months" promises a start date and an end date. "Sleep structure matures gradually, and night-to-night variability is wide" promises neither — but it is true.
Of all the explanations for sudden bad nights, one has survived testing: the acquisition of new motor skills. Two studies by the same group, and they are worth telling apart.
| Study | Method | Finding |
|---|---|---|
| Scher & Cohen, 2005 n = 107, 5–8 months |
Maternal questionnaire + motor checklist | Infants who were already crawling had more bedtime difficulties and night waking than pre-crawlers. Among 7- and 8-month-olds, locomotion predicted about 17% of night-waking variance |
| Scher & Cohen, 2015 n = 28, followed from 5 to 11 months |
Actigraphy every 2–3 weeks — a device, not a questionnaire | A rise in long wake episodes was temporally linked to crawling onset. Meanwhile sleep consolidation over the period kept improving |
The key word is "onset". The link is not with age but with the moment a particular child masters the skill. One baby crawls at six months, another at nine — and their "windows" fall at different times. That is exactly why a calendar of regressions fails even where the phenomenon itself is real.
And about effect size, honestly: 17% of the variance is substantial, but it is not "this explains everything". The other four fifths belong to everything else, and sleep records cannot separate them.
The week-by-week calendar of "leaps" comes from the book by the Dutch researchers van de Rijt-Plooij and Plooij. The fairest thing to say about it is this: it is a contested model, not an established fact. The original 1993 study rests on 15 mothers; the 2002 Catalan replication on 18 infants, using the model authors' own instruments, and it found 8 of the claimed 10 periods. An independent group in Groningen (van Geert & de Weerth) contested the model back in 1998, a full scientific exchange followed, and that group's own weekly observations produced broad bands of behavioural variability rather than ten tightly timed "leaps".
So both popular claims are wrong: "leaps are proven" and "leaps have been debunked". The accurate statement is that the model exists, its support is small and comes from its own authors, and independent testing did not back it.
Teething is the most popular explanation for bad nights, and it has been tested directly. In the prospective study by Macknin and colleagues (Pediatrics, 2000), 125 infants were followed daily from four months to one year: 19,422 child-days and 475 tooth eruptions. Symptoms clustered in an eight-day window around each tooth. However, sleep disturbance was not among the symptoms significantly associated with teething (restless wakefulness was), and no symptom occurred in more than 35% of teething infants. A second cohort (Wake 2000) also failed to confirm the expected strong associations between tooth eruption and a range of symptoms including sleep — even though, retrospectively, every parent in it reported that their own child had suffered teething symptoms.
For accuracy: "teeth do not affect sleep" is too strong. In the same study, restless wakefulness was associated with teething while sleep disturbance was not; those were two separately recorded items. The honest wording is that the effect, if any, is small and the findings differ across sleep measures.
Here is what Baby Sleep Planner actually does — so that the analysis in the app and this article do not contradict each other.
The app only sees sleep records. From them it can notice a settled change: for example, that over two weeks daytime sleep has become lower, and not for one or two days but for a run of days in a row. That is an observation, and the wording in the app says exactly that: "daytime sleep has settled at a lower level".
The app does not name a cause and will not. A cause cannot be established from sleep records: a reorganising routine, a changed daily schedule, the first days of an illness and a newly learned skill all look the same. Saying "your baby is having a sleep regression" would mean presenting a guess as an observation, on data in which it is not visible.
The good news: almost everything useful here is independent of the cause.
You cannot, and that is the honest answer. Sleep records show that sleep has changed, but not why. Look at two things instead of the label: whether your child is getting enough sleep across 24 hours, and whether there are any signs of illness.
No. There is no peer-reviewed study showing a spike of sleep problems specifically at four months, and direct measurements point the other way: sleep consolidates fastest from 1 to 4 months, while the measured rise in night waking sits closer to nine months.
That link has been measured and found, but it tracks the skill in a particular child, not the calendar age. In Scher & Cohen 2005 (n = 107, 5–8 months) infants who were already crawling had more night waking; among 7- and 8-month-olds locomotion predicted about 17% of the variance. In the same authors' 2015 actigraphy study (n = 28) the rise coincided with crawling onset, while sleep overall kept improving.
Less than people assume. In Macknin 2000 (125 infants, 19,422 child-days, 475 tooth eruptions) sleep disturbance showed no significant association with tooth emergence, and no symptom occurred in more than 35% of infants. Careful with strong conclusions either way: restless wakefulness did show an association. The authors' practical advice is to rule out other causes before blaming the teeth.
Figures like "2–6 weeks" circulate without a source. We could not find a study they could come from, so we do not quote a duration. Track the 24-hour sleep total and your child's wellbeing instead of the calendar.
Because it does not know. The app only sees sleep records. It can tell you that daytime sleep has settled at a lower level — that is an observation, not a cause. A cause cannot be established from sleep records, so the app does not name one.
What we do not know, and therefore do not write: how long "it" lasts; whether every child goes through it; whether a reorganising routine can be told apart from a developmental leap using a sleep log. We could find no measurement for any of the three.
Baby Sleep Planner keeps the sleep log and shows when daytime sleep has settled lower and how much sleep the day is short by. It does not name a cause — but it does show whether there is a real loss of sleep and how far to move the evening.
Try free for 7 days