Published on 17 September 2026
Parent and child sleep are correlated: what can we conclude, and what can’t we?
Sleep Medicine Reviews · 2026 ; volume 89, October 2026 : article 102338 · Dai et al.
DOI 10.1016/j.smrv.2026.102338
PMID 42424772
Scientific 79
Editorial 72
The essentials
Systematic review with meta-analysis, pre-registered protocol, PRISMA reporting: 46 studies from 49 publications, 104,877 parent-child dyads and 5,530 mother-father-child triads, 16 countries. The question is synchronicity, that is, the degree to which a parent’s sleep and a child’s sleep vary together. Every pooled correlation is positive except satisfaction: duration r = 0.18 (95% CI 0.14 to 0.23), efficiency r = 0.30 (95% CI 0.19 to 0.41), bedtime r = 0.31 (95% CI 0.21 to 0.40), wake-up time r = 0.48 (95% CI 0.24 to 0.66), number of night wakings r = 0.48 (95% CI 0.32 to 0.61), wake after sleep onset r = 0.55 (95% CI 0.45 to 0.63), subjective satisfaction r = 0.16 (95% CI -0.01 to 0.32), not significant. The strongest coupling therefore concerns not the clock but the continuity of sleep. The association is stronger in mother-child dyads for duration, efficiency and continuity, weaker with the child’s age for duration and efficiency, and weaker again when sleep is measured by actigraphy rather than by questionnaire, except for the clock domains, where these moderations are not significant. Two limitations govern the whole reading. First, the authors themselves describe their evidence base as predominantly cross-sectional and explicitly refuse any causal or directional conclusion. Second, heterogeneity is massive, I² ranging from 86.9% to 98.2% depending on the domain, which means the pooled figure poorly summarises the studies it aggregates. The authors themselves note that calendar synchronicity may partly reflect imposed schedules, school and work. No intervention is evaluated in this work.
Context
A child’s sleep complaint rarely arrives alone in consultation. It arrives with an exhausted parent, most often the mother, and the order in which questions are asked largely determines what will be found. Does one question the child alone, as the presenting complaint would suggest, or the sleep configuration of the whole household, which clinical experience has long suggested without the literature ever quantifying it?
The notion of synchronicity comes from work on family systems and co-regulation. It holds that the biological and behavioural rhythms of members of a single family are not independent. In perinatal psychiatry and child psychiatry practice, this hypothesis has an immediate consequence: if sleep is coupled, an evaluation that isolates the child from their environment misses part of the picture.
This synthesis provides a quantitative estimate of this coupling, sleep domain by sleep domain, across a large evidence base. That is its real value. It does not say what should be done, it says what should be looked at.
Mechanism
What a within-dyad correlation can reflect
A correlation between a parent’s sleep and a child’s sleep admits at least four explanations, only one of which is trivial, and none of which can be distinguished by cross-sectional data. The distinction is decisive here, because the observed profile, weak for duration and markedly stronger for the clock and for continuity, leaves the most banal explanation open for at least part of the signal.
| Explanatory pathway | Status in this synthesis |
|---|---|
| Shared schedules and external constraints | Acknowledged by the authors, not resolved |
| Finding School and work impose shared waking and bedtime hours. This is the simplest pathway. The authors write that calendar synchronicity may partly reflect these shared constraints, without quantifying how much: most included studies do not separate weekdays from weekends, which would have made this measurable. | |
| Behavioural and emotional co-regulation | Hypothesis, not directly tested |
| Finding Night wakings in the child, parental reassurance, contagion of physiological arousal. No mediator of this kind was measured or modelled as such in the pooled analyses. This pathway is interpretive here, not demonstrated. | |
| Parental mental health, particularly maternal | Cited in the narrative synthesis, never tested |
| Finding Parental depression and insomnia are obvious candidates, both as a shared factor and as a consequence. This pathway, however, appears only in the narrative synthesis of mechanisms, supported by two studies: no meta-regression addresses parental mental health. And the protocol excluded studies of parents with a mental disorder: the evidence base describes general-population families, not clinical situations. | |
| Shared genetic and environmental factors | Not explored |
| Finding Family chronotype, noise, housing, light exposure, deprivation. A parent and child share both genes and a home, two sources of covariance that no cross-sectional design can separate from mutual influence. | |
What the synthesis measures is therefore a degree of association, not a mechanism. What it does not measure are the pathways through which this association arises, nor its direction: does the parent’s sleep follow the child’s, the reverse, or both together under the effect of a third factor.
The study at a glance
| Design, population, comparison, outcomes | |
|---|---|
| Design | |
| Detail Systematic review with meta-analysis, pre-registered on PROSPERO under number CRD42024567529, PRISMA reporting. Four databases searched, PubMed, EMBASE, PsycINFO and Scopus, from inception to April 2025, with no language or design filter at the search stage but a later restriction to English-language articles; 31,898 records screened, double screening of every record. Three-level random-effects model estimated by restricted maximum likelihood, accounting for multiple effects drawn from the same study; the figure legends note, however, that a two-level analysis was used for duration, efficiency, wake after sleep onset and bedtime. Effects converted to Pearson’s r then transformed to Fisher’s z. Methodological quality assessed with the Newcastle-Ottawa scale for cohorts and intensive longitudinal protocols, a National Institutes of Health tool for cross-sectional studies, and the Joanna Briggs Institute checklist for the single qualitative study. Univariate meta-regressions, sensitivity analyses by removing outlying values, publication-bias exploration by funnel plot, Egger’s test and trim-and-fill. | |
| Population (P) | |
| Detail 104,877 parent-child dyads and 5,530 mother-father-child triads, from 46 studies reported across 49 publications, conducted in 16 countries. Children aged 0 to 18 living under the same roof as their parent, in community populations. The evidence base is dominated by the United States, 18 studies, ahead of Japan, 5, and Israel, 3. Excluded were children with special health needs, parents with a mental disorder, children not living with their parents, and hospital or institutional settings. Maternal informants predominate: only 20 of 46 studies measure the father’s sleep, and often by maternal proxy. | |
| Exposure and measurement (I) | |
| Detail Joint measurement of parent and child sleep. Twenty-nine studies rely on questionnaires or diaries alone, such as the Pittsburgh Sleep Quality Index or the Insomnia Severity Index; seven use actigraphy, one electroencephalography, one polysomnography, and the rest combine actigraphy with self-report. This is not an intervention: no experimental manipulation is involved. | |
| Comparison (C) | |
| Detail There is no control group. The comparison concerns the degree of within-dyad correlation by sleep domain, then by pre-specified or exploratory moderators: child’s age, parent’s sex, measurement type, study design. | |
| Outcomes (O) | |
| Detail Pooled correlation coefficients across four domains of the Peds B-SATED framework: duration, efficiency and continuity (efficiency, number of night wakings, wake after sleep onset), timing (bedtime, wake-up time), satisfaction. The alertness and sleep-behaviour domains were not pooled, for lack of data and because definitions were too disparate. Complementary narrative synthesis of the mechanisms proposed in the included literature. | |
| Nature of the included designs | |
| Detail Eighteen cross-sectional studies, sixteen cohorts, seven intensive longitudinal actigraphy protocols, three cross-sectional surveys followed by a seven-day sleep diary, one qualitative study. The authors nonetheless describe their evidence base as predominantly cross-sectional, on the grounds that the pooled correlations they combine are mostly point-in-time associations, including within the cohorts. This is the point that governs the entire reading: a pooled correlation cannot order two variables in time. |
Quality control
| Point checked | Judgement |
|---|---|
| Protocol and reporting | Compliant |
| Finding Prior registration on PROSPERO and PRISMA reporting. This does not guarantee the absence of deviation between protocol and publication, but it makes it verifiable. | |
| Statistical model | Appropriate |
| Finding The three-level meta-analysis correctly handles the dependence of multiple effects drawn from the same study, a frequent and often overlooked problem in this field. Reservation: the figure legends indicate that four of the seven pooled analyses were ultimately conducted at two levels, which the methods section does not announce. | |
| Study quality assessment | Performed |
| Finding Three tools used according to design, Newcastle-Ottawa, a National Institutes of Health tool, and the Joanna Briggs Institute checklist. Quality is rated good or acceptable for 94.6% of cross-sectional studies, 93.8% of cohorts, and 90% of intensive longitudinal protocols. It is introduced as a moderator: studies of acceptable quality yield higher estimates than studies of poor quality for duration, and studies of good quality yield higher estimates for wake-up time. | |
| Publication bias search | Signal detected |
| Finding Funnel plot, Egger’s test and trim-and-fill, domain by domain. Egger’s test is significant for bedtime, z = -6.65 (p < 0.001), and for wake-up time, z = -4.69 (p < 0.001); trim-and-fill imputes five and two missing studies respectively and raises the estimates to 0.43 (95% CI 0.31 to 0.54) and 0.53 (95% CI 0.32 to 0.73). For efficiency, Egger’s test is not significant but trim-and-fill imputes five studies and raises the estimate to 0.41 (95% CI 0.28 to 0.53). No asymmetry is detected on the other domains, which is not the same as an absence of bias: Egger’s test has low power. | |
| Heterogeneity | Very high |
| Finding I² of 86.85% for bedtime, 87.67% for duration, 90.07% for wake after sleep onset, 92.80% for wake-up time, 94.12% for efficiency and for night wakings alike, 98.18% for satisfaction. Almost all of the variance observed between studies exceeds what sampling error would explain. The pooled estimates should be read as orders of magnitude, not as constants. | |
| Homogeneity of instruments | Insufficient |
| Finding Actigraphy covers different devices, algorithms, epoch lengths, sampling rates and activity thresholds, without harmonisation; the authors note that only one study specifies its wake-detection threshold. Questionnaires do not measure the same thing as actigraphy, which is supported by the moderation effect observed for duration, efficiency and continuity. | |
| Source of information | Asymmetric |
| Finding Maternal reports predominate, fathers are under-represented, 20 of 46 studies. A single informant describing both their own sleep and their child’s introduces a method correlation liable to inflate the estimate, a hypothesis the authors also endorse. | |
| Cultural representativeness | Limited |
| Finding A predominantly Western, high-income evidence base; cultures where shared sleep is the norm are little represented. Sleeping arrangement does affect the estimates, but not in the expected direction: the proportion of shared sleep is negatively associated with efficiency synchronicity (beta = -1.676; 95% CI -2.033 to -1.319), a result the authors themselves call unexpected and for which they propose several competing interpretations. | |
| Independence and transparency | Satisfactory |
| Finding Funded by three Chinese public bodies, the National Natural Science Foundation of China (grant 72404063), the Guangzhou Municipal Science and Technology Bureau (2025A04J4518), and the Guangzhou Women and Children’s Medical Center (2024BS021). No conflict of interest declared. The authors state they used a generative artificial intelligence tool for language editing of the manuscript, while taking responsibility for its content. | |
The findings
| Sleep domain | Pooled correlation |
|---|---|
| Wake after sleep onset | r = 0.55 (95% CI 0.45 to 0.63) |
| Reading The strongest coupling in the synthesis, on only three studies and twelve estimates. Mother-child correlation of 0.57 against 0.42 for father-child dyads (p = 0.01). Removing the two outlying values raises the estimate to 0.62 (95% CI 0.51 to 0.72), indicating that the result does not hinge on these values. | |
| Wake-up time | r = 0.48 (95% CI 0.24 to 0.66) |
| Reading A strong association, but a wide confidence interval, from 0.24 to 0.66. Precision is therefore poor despite the sample size: the signature of substantial between-study heterogeneity, I² at 92.8%. Egger’s test also detects funnel-plot asymmetry on this domain. | |
| Number of night wakings | r = 0.48 (95% CI 0.32 to 0.61) |
| Reading Five studies, 24 estimates. Mother-child correlation of 0.53. This is the domain where a relational reading is most tempting, the child’s waking waking the parent, but the measurement cannot distinguish this pathway from a shared perception: the coupling drops when the measure switches to actigraphy. | |
| Bedtime | r = 0.31 (95% CI 0.21 to 0.40) |
| Reading A moderate association across five studies. The publication’s abstract reports, for this domain, r = 0.34 (95% CI 0.20 to 0.46), a value the results section does not repeat: the publication is internally inconsistent on this figure, and it is the results-section value that is retained here. Egger’s test detects marked asymmetry on this domain. | |
| Sleep efficiency | r = 0.30 (95% CI 0.19 to 0.41) |
| Reading Six studies, 20 estimates, on a quality parameter less directly explained by shared schedules. This is the most conceptually interesting result, and it would warrant a longitudinal replication on its own. Mother-child correlation of 0.38 against 0.21 for father-child dyads (p < 0.001). | |
| Sleep duration | r = 0.18 (95% CI 0.14 to 0.23) |
| Reading A weak but precisely estimated association, across fifteen studies and 55 estimates. An r of 0.18 corresponds to roughly 3% shared variance: at the scale of a single consultation, it predicts nothing for a given child. | |
| Sleep satisfaction | r = 0.16 (95% CI -0.01 to 0.32), not significant |
| Reading Nine studies, 23 estimates. Non-significance is not proof of an absence of coupling: the interval admits both zero and a moderate association, and heterogeneity reaches 98.18% here. Caution is warranted in both directions. The authors note that a child’s “sleep problems” here denote parental worry, not a diagnosis. | |
| Moderator: parent’s sex | Stronger coupling in mother-child dyads, except for timing |
| Reading The difference is significant for duration, efficiency, night wakings and wake after sleep onset; it is not significant for bedtime (p = 0.057), wake-up time (p = 0.074) or satisfaction (p = 0.401), contrary to what the publication’s discussion states. A result consistent with the distribution of night-time caregiving, but confounded with the fact that the mother is also the main informant. Two explanations, one measurement. | |
| Moderator: child’s age | Weaker coupling with age, for duration and efficiency |
| Reading Beta = -0.019 (95% CI -0.023 to -0.015) for duration and -0.054 (95% CI -0.064 to -0.040) for efficiency. On the other domains, age is not a significant moderator. The expected direction, physical proximity, dependence, adult-set schedules, is therefore documented on only two domains. | |
| Moderator: measurement type | Weaker coupling with actigraphy, except for timing |
| Reading Relative to self-report, beta = -0.119 (95% CI -0.159 to -0.077) for duration, -0.614 (95% CI -0.722 to -0.506) for efficiency, -0.430 (95% CI -0.634 to -0.226) for night wakings, and -0.345 (95% CI -0.554 to -0.137) for wake after sleep onset. For bedtime and wake-up time, the effect is not significant. Replacing the parent’s perception with a sensor therefore markedly reduces the association for quantity and continuity, but not for timing, which is consistent with a measurement component within the self-reported signal. | |
| Moderator: study design | Stronger coupling in cohorts, for duration and efficiency |
| Reading Beta = 0.314 (95% CI 0.239 to 0.389) for duration and 0.770 (95% CI 0.525 to 1.016) for efficiency, relative to cross-sectional studies. For night wakings, wake after sleep onset and satisfaction, the difference is not significant, and the analysis is not available for timing. A difference to interpret cautiously: these between-study comparisons are observational by construction. | |
| Heterogeneity | I² from 86.85% to 98.18% depending on domain |
| Reading The identified moderators absorb only part of this variance. A substantial fraction remains unexplained, which limits the scope of any single figure. | |
Critical appraisal
| Domain | Risk |
|---|---|
| Temporality and causality | Disqualifying |
| Finding Since the pooled correlations are essentially point-in-time associations, no inference of direction is possible. Neither does the parent’s sleep explain the child’s, nor the reverse: the data simply do not allow the question to be posed. This is not a flaw of execution, it is the nature of the available material, and the authors acknowledge it explicitly, down to their conclusion. | |
| Structural confounding | High |
| Finding School and work schedules, housing, season, latitude, socio-economic status. No systematic adjustment is possible at the level of a meta-analysis of aggregated correlations. An indeterminate share of the observed coupling is thus structural rather than relational. | |
| Common-method bias | Probable |
| Finding When the mother describes both her own sleep and her child’s, the correlation incorporates her own response style and mood state. The drop observed with actigraphy for duration, efficiency and continuity is consistent with this hypothesis and is its best internal indicator, without proving it: the authors note that actigraphy and self-report may also capture different aspects of the same phenomenon, the diary recording brief awakenings that actigraphy does not count. | |
| Unexplained heterogeneity | High |
| Finding I² between 86.85% and 98.18% depending on domain, despite multilevel modelling and meta-regressions. The confidence intervals of the pooled estimates do not describe the dispersion of true effects: a prediction interval would be more informative here, and the publication reports none. | |
| External validity | Partial |
| Finding Sixteen countries, but a predominantly Western evidence base and under-represented fathers. Transposition to single-parent and blended families, shift workers, and cultures of shared sleep is not established. | |
| Adequacy of conclusions | Respected |
| Finding Notably, the authors do not over-interpret. They name confounding by shared schedules, refuse causality, and explicitly call for longitudinal designs, including in their clinical implications section. Two reporting reservations nonetheless: the bedtime correlation differs between the abstract and the results, and several moderation coefficients cited in the text do not match those in the supplementary table. | |
| Independence | Low risk |
| Finding Chinese public funding, no declared conflict of interest, no product or device at stake. As with any team synthesising a field to which it contributes, a residual risk of theoretical allegiance remains: family systems theory serves here as the interpretive frame both before and after the results. | |
Level of evidence
Confidence is high on one point, and one only: there is a positive association between parents’ and children’s sleep, more pronounced for continuity and timing than for duration. The sample size, the method, and the consistency of the sign across domains make this difficult to dispute. This is the established point of this work. Its magnitude is less certain than its existence: for bedtime and wake-up time, the funnel plot is asymmetric, which points toward an overestimation from small studies.
Confidence is low, if not absent, on everything else. The direction of the relationship is unknown, and will remain so as long as the estimates rest on aggregated correlations. The share of the coupling attributable to genuine co-regulation, rather than to imposed schedules and to the fact that a single informant describes both sleeps, is not quantified. The drop in the signal under actigraphy for duration, efficiency and continuity suggests that this methodological share is not marginal. Finally, heterogeneity ranging from 87% to 98% depending on domain implies that the pooled figure poorly summarises very different family situations.
What amounts to expert opinion, including our own, must be named as such: the idea that a family-based approach to paediatric sleep would be preferable to a child-only approach is plausible, clinically appealing, and not demonstrated by this work. No intervention based on synchronicity has been evaluated here, and none can therefore be recommended on this basis.
The colleague test
What an experienced colleague would say if you put this work to them in two minutes, between two consultations.
“ That parents and children sleep at the same hours, I see that every day, and a wake-up correlation of 0.48 is first and foremost school time. What interests me is the night waking, 0.48 for the number of wakings and 0.55 for wake time, because there we are no longer just sharing an alarm clock. But with cross-sectional correlations, three studies for wake time and an I² above 90, I’m not changing anything in my practice. I’m waiting for longitudinal data. ”
What this means in practice: this work changes a way of looking, not a course of action. It justifies widening the history-taking to the parent’s sleep, it justifies no new protocol.
What you can take from this
- Widen the interview. Faced with a child’s sleep complaint, also ask about the sleep of the parent or parents present: actual bedtime and wake time, night wakings, perceived quality. This costs only a few extra questions and documents a configuration, not a causal chain.
- Separate what is imposed from what is relational. Before interpreting a concordance of rhythms as a family phenomenon, reconstruct the constraints: school schedules, commutes, shift work, shared custody. The authors accept this explanation for part of the calendar synchronicity, without being able to weigh it, for lack of studies distinguishing weekdays from weekends.
- Be wary of the single informant. When the same parent describes both their own sleep and the child’s, the reported concordance is in part an effect of the reporter. An objective measure, or the account of a second adult in the household, usefully corrects this reading.
- Stay alert to the parent’s mental health. A child’s sleep complaint is an opportunity to identify exhaustion, insomnia, or depressive symptoms in the parent, particularly in the perinatal period. This is a matter of screening and referral, not of an intervention justified by this work: the protocol specifically excluded families in which a parent had a mental disorder, so these situations are not represented in the evidence base.
- Promise nothing. Nothing here allows one to claim that improving a parent’s sleep will improve a child’s, or the reverse. Saying so clearly to families is part of honest information.
Frequently asked questions
If I treat the mother’s insomnia, will the child’s sleep improve?
Nothing in this work allows that claim. A cross-sectional correlation does not say what happens when one intervenes on one of the two terms. The question is legitimate and interesting, and it calls for a controlled trial, which does not exist here. Treating the parent’s insomnia is justified for the parent’s own sake, which is already reason enough.
Is a coefficient of 0.48 a lot?
It is a strong association for this type of literature, but it corresponds to roughly 23% shared variance, and its confidence interval runs from 0.24 to 0.66. At the scale of a given child, it predicts nothing useful. A correlation coefficient describes a group tendency, never an individual trajectory.
Why is the link weaker when actigraphy is used?
First a clarification: the drop is not seen everywhere. It is marked for duration, efficiency and sleep continuity, and not significant for bedtime or wake-up time, where sensor and questionnaire agree. Where it is seen, two explanations coexist and cannot be distinguished here. Either questionnaires overestimate the coupling, because a single parent describes both sleeps with the same perceptual bias, or actigraphy and self-report simply do not measure the same thing, the diary recording brief awakenings that actigraphy does not count. The first hypothesis is the more parsimonious, but it remains a hypothesis, and the authors keep both.
Why is the coupling stronger with mothers than with fathers?
The result exists, but it is narrower than it appears and its interpretation is ambiguous. The difference is significant for duration, efficiency and sleep continuity; it is not significant for bedtime, wake-up time, or satisfaction. Mothers statistically provide a greater share of night-time caregiving, which argues for a genuinely closer coupling. But they are also the informants in most included studies, which alone would suffice to produce the same figure. With fathers under-represented, twenty of forty-six studies, the comparison lacks power, and the authors themselves speak of a possibly illusory maternal exclusivity.
Should we conclude that the family should be treated rather than the child?
This is a plausible conceptual orientation, not a demonstrated conclusion. This work establishes that sleep varies together, it does not establish that acting on one modifies the other, and it evaluates no family-based care model. The authors do raise a possible interest in family-centred approaches, but they write in the same sentence that precise clinical recommendations must wait for longitudinal data on the direction of the link. Presenting this synthesis as justification for a family protocol would therefore harden what they actually write.
Annotated bibliography
Source study. Dai Y, Dong F, Chen R, Liu X, Guo L, Luo J, Wu M, Xiong Y, Deng B, Deng Y, Lin Y. Synchronicity of parent-child sleep and potential mechanisms: A systematic review and meta-analysis. Sleep Medicine Reviews 2026 ; volume 89 : article 102338. DOI 10.1016/j.smrv.2026.102338. PMID 42424772. PROSPERO registration CRD42024567529. Contribution: first quantitative estimate of parent-child sleep coupling, domain by domain, across 46 studies and 104,877 dyads. Limitation: aggregated and mostly cross-sectional correlations, heterogeneity from 87% to 98%, funnel-plot asymmetry on the two timing domains.
Context reference 1. Meltzer LJ, Mindell JA. Relationship between child sleep disturbances and maternal sleep, mood, and parenting stress: a pilot study. Journal of Family Psychology 2007 ; volume 21, issue 1 : pages 67 to 73. DOI 10.1037/0893-3200.21.1.67. Cited by the source study’s authors. Contribution: a foundational work that establishes the idea that a child’s sleep and the mother’s state are linked, giving clinical depth to the notion of synchronicity. Limitation: a pilot study, modest sample, cross-sectional design, self-reported measures, so exactly the limitations that the present meta-analysis inherits and amplifies.
Context reference 2. Sadeh A. The role and validity of actigraphy in sleep medicine: an update. Sleep Medicine Reviews 2011 ; volume 15, issue 4 : pages 259 to 267. DOI 10.1016/j.smrv.2010.10.001. Contribution: essential to understanding why measurement type changes the estimate, and why actigraphy poorly detects quiet wakefulness in bed. Limitation: a narrative review predating current algorithms, it does not resolve the lack of harmonisation between devices that weighs on the analysed evidence base, where only one study specifies its wake-detection threshold.
Context reference 3. Buysse DJ, Reynolds CF, Monk TH, Berman SR, Kupfer DJ. The Pittsburgh Sleep Quality Index: a new instrument for psychiatric practice and research. Psychiatry Research 1989 ; volume 28, issue 2 : pages 193 to 213. DOI 10.1016/0165-1781(89)90047-4. Contribution: one of the most used instruments on the parent side in the included studies, along with the Insomnia Severity Index; knowing its structure clarifies what the word quality covers. Limitation: validated in adults, it is not transposable to children, and it measures a one-month retrospective perception, which is not comparable to an objective measure of the same period.
Context reference 4. Higgins JPT, Thompson SG. Quantifying heterogeneity in a meta-analysis. Statistics in Medicine 2002 ; volume 21, issue 11 : pages 1539 to 1558. DOI 10.1002/sim.1186. Contribution: provides the framework for interpreting I² values between 87% and 98%, and recalls that such a value invalidates any reading of the pooled result as a constant. Limitation: I² is a proportion, not an absolute measure of dispersion, and its value increases mechanically with the precision of the included studies, which makes a prediction interval preferable.