A Systematic Review and Meta-Analysis of Partial Correlations Between Caregiver Cognitive Stimulation in Early Childhood and Child Development

Sample Characteristics as Potential Moderators

Associations between cognitive stimulation and child development may vary across developmental, methodological, and cultural contexts, and understanding this variation is important for identifying the conditions under which this parenting behavior is most strongly associated with development. Early childhood is a period of rapid foundational skill acquisition and heightened brain growth and plasticity, during which children may be especially receptive to environmental inputs such as cognitive stimulation (Gabard-Durnam & McLaughlin, 2020; Shonkoff & Phillips, 2000). Developmental processes and milestones vary significantly across child age, which may affect children’s receptivity to cognitive stimulation throughout early childhood. From birth to age two (i.e., infancy), caregivers in the home play an especially critical role in supporting foundational language, sensorimotor, and attentional skill development (American Academy of Pediatrics; n.d., Gee, 2016). The centrality of the home environment continues from age two to four (i.e., toddlerhood), during which children’s developmental repertoire expands to include additional skills in expressive language, problem-solving, and social-emotional functioning (American Academy of Pediatrics, n.d; Kochanska et al., 2000). By ages four to six (i.e., preschool and early school-age), children increasingly develop more complex academic, executive functioning, and self-regulation skills that support school readiness and the transition to formal schooling. During this period, cognitive stimulation at home is increasingly complemented by learning opportunities in structured educational contexts (American Academy of Pediatrics, n.d.; Tamis-LeMonda et al., 2019). These normative developmental trends suggest that the magnitude of the association between cognitive stimulation and children’s outcomes may even vary within early childhood, for example, through differential receptivity to cognitively stimulating input or through varying degrees of how strongly children’s development is shaped by home-based interactions.

Child sex may also moderate the association between cognitive stimulation and developmental outcomes. Sex differences in the timing of maturation and the identification/expression of neurocognitive abilities have been observed for some developmental domains, including girls’ earlier gains in self-regulation and lower rates of attention-deficit/hyperactivity disorder (Babinski, 2024; Matthews et al., 2009). Additionally, gendered expectations about caregiving may shape the frequency and type of cognitive stimulation children experience, with evidence suggesting small differences by child and caregiver sex in some, but not all, low- and middle-income countries (LMICs; Bornstein & Putnick, 2016; Cuartas et al., 2020). Therefore, girls and boys may (or may not) differentially benefit from high levels of cognitive stimulation due to their sensitivity and exposure to environmental inputs. The exploration of sex-based differences in the association between caregiver stimulation and child outcomes is relevant to many early childhood policies, including the United Nations Sustainable Development Goals (UN SDGs), which emphasize gender equity in children’s access to quality environments and developmental outcomes (e.g., SDG 4.2).

Finally, much of the evidence and conceptual models on caregiving and cognitive stimulation come from high-income country contexts (Nielsen et al., 2017; Thalmayer et al., 2021), where caregivers generally have greater access to resources, infrastructure, and formal education, and children have more widespread access to early education. Family structure and cultural norms may further influence the frequency and form of cognitive stimulation, as well as how this type of caregiver-child interaction shapes development (McCoy, 2022; Ogbu, 1981). In lower-income countries, caregivers may face increased structural barriers (e.g., limited access to high-quality healthcare and safe public play spaces) that constrain development or opportunities for cognitively stimulating interactions (Black et al., 2017; Britto et al., 2017). The present study examines variation in the associations between cognitive stimulation and child development by country-income level, a proxy for structural conditions.

Study Features as Potential Moderators

Methodological features of a study may also affect the strength and/or direction of the reported associations between cognitive stimulation and child development. For example, one important factor is whether studies control for family socioeconomic status (SES) in their analytic models. Although many caregivers with limited economic resources create enriched environments for their children (e.g., through teaching in everyday routines, storytelling, singing; Coley et al., 2025), ample evidence has identified average positive associations between household SES and both the frequency of cognitive stimulation and children’s developmental outcomes across both high- and low-income countries (e.g., Bradley & Corwyn, 2002; Noble et al., 2015; Obradovic et al., 2016; Votruba-Drzal, 2003). Therefore, if SES is unaccounted for, the association between cognitive stimulation and child outcomes may be mischaracterized (i.e., overestimated).

Beyond covariate adjustment, other indicators of study quality may also affect the reported associations by introducing or mitigating sources of bias. For instance, studies that employ representative or random sampling strategies, use well-established measures (i.e., psychometrically supported or widely used), and appropriately address missing data may be less biased and, in turn, more accurately portray the true association between cognitive stimulation and child outcomes. Assessing overall study quality can clarify the extent to which variation across studies may reflect methodological bias rather than true differences in associations.

Cognitive stimulation measurement modality may also shape the strength of the association with children’s development. Common cognitive stimulation measurement approaches, including retrospective caregiver-report surveys and assessor observations, may differ in their pragmatic and theoretical strengths and be subject to different forms of systematic measurement error (Hoyt, 2000; Keller, 1980; Lytton, 1971). For example, caregiver-reported measures may be influenced by social desirability or recall bias, which could inflate reported levels of cognitive stimulation and, in turn, attenuate associations with child development. At the same time, data collectors are not immune to introducing bias and have been shown to implicitly contribute to significant variation in assessments (McDermott et al., 2014). Accordingly, we examine cognitive stimulation measurement modality as a moderator to empirically assess whether this shapes the association with children’s development.

Transparency and Openness

The review protocol and analysis plan for this systematic review and meta-analysis were preregistered in PROSPERO (CRD42022314112), and we followed the Preferred Reporting Items for Systematic Reviews and Meta-Analysis for the reporting of our results (PRISMA; see Supplementary Materials S1). We deviated from our original preregistered protocol in several ways, including incorporating a study quality assessment and modifying aspects of the analytic plan (e.g., effect-size transformation); see Supplementary Materials S2 for a full description of changes. The meta-analytic data and analytic scripts can be accessed via the Open Science Framework at https://doi.org/10.17605/OSF.IO/VF5QT. A preprint of the original submission was posted here: https://osf.io/preprints/osf/9w6tg_v1. Generative artificial intelligence was used only in a trivial way (e.g., grammar and code debugging).

Inclusion and Exclusion Criteria

First, regarding caregiver cognitive stimulation, we included studies that measured the presence or frequency of cognitively stimulating interactions caregivers provided to children. We excluded studies that focused exclusively on the quality of cognitive stimulation (e.g., conversational turn-taking during a structured play activity) or that presented separate subtypes of cognitive stimulation within the same model (as these effect sizes would underestimate the overall association of cognitive stimulation). Supplementary Materials S3 list the cognitive stimulation measures included in the meta-analysis by study.

Second, in terms of child and parent characteristics, cognitive stimulation must have been measured during early childhood (mean sample age between 0–72 months) and within a neurotypical, normatively developing sample. Studies were excluded if they exclusively sampled groups who engaged in risky health behaviors (e.g., substance use during pregnancy) or had existing medical or psychiatric diagnoses. We did not exclude studies based on the type or timing of developmental outcome. Additionally, cognitive stimulation must have been provided by primary caregivers, such as parents and/or other relatives, in the home environment. We excluded studies in which cognitive stimulation was provided by teachers or childcare workers.

Third, eligible studies were quantitative, peer-reviewed, written in English, and published between January 1, 2000 and January 29, 2025. Other evidence synthesis studies were excluded. Studies had to report at least one effect size (e.g., regression coefficient, odds ratio) quantifying the relation between cognitive stimulation and any domain of children’s development. We included both cross-sectional and longitudinal data structures, as we believe both largely investigate comparable research questions in this literature. The table in Supplementary Materials S4 outlines our inclusion and exclusion criteria in detail.

Literature Search Strategy

A systematic computerized search was conducted through four electronic databases (Education Resources Information Center, APA PsycINFO, PubMed, and Web of Science¹). Four team members conducted preliminary literature searches in consultation with a university-based research librarian. After identifying exemplar studies that met our inclusion criteria, we used these studies to refine our search terms and construct flexible Boolean operators. We then iteratively pilot-tested our search terms to confirm that they returned the exemplar studies and retrieved the intended body of relevant literature. Keywords related to broadly-defined cognitive stimulation (e.g., learning activities, stimulation, nurturing care) or common measures of cognitive stimulation found in large-scale studies (e.g., UNICEF Multiple Indicator Cluster Surveys; HOME–Short Form) were used in tandem with keywords for our caregiver (e.g., maternal, parent, caregiver) and child (e.g., childhood, toddler, preschool) target populations. These keywords needed to be present in the study title, bibliographic database tags, or the abstract. Supplementary Materials S5 contains the full list of search terms. See the PRISMA flow diagram in Figure 1 for the number of studies considered at each stage of the search, screening, and extraction process.

Screening and Extraction Procedures

All abstract and full-text eligibility screening was done using Covidence (https://www.covidence.org/).

Data Preparation

Analysis Methods

To examine the characteristics of the existing literature on caregiver cognitive stimulation and child development, we used descriptive statistics to characterize the features of studies included in our dataset.

To address our second research question, we used multilevel meta-analysis to estimate the associations between early childhood cognitive stimulation and children’s developmental outcomes. Many studies reported multiple eligible effect sizes, for example, by examining multiple outcomes. To account for the hierarchical data structure, we used multilevel random-effects meta-regression models with sampling variance at level 1, within-study variance at level 2, and between-study variance at level 3. All models were estimated using restricted maximum likelihood estimation via the rma.mv function in the metafor package in R (method = “REML”; v 4.6–0, Viechtbauer, 2010). This function uses an inverse-variance weighting approach, in which studies with lower variance (i.e., higher precision) were given greater weight. Additionally, because the effect sizes between studies were not independent (due to the use of the same measures or the same representative dataset), we also employed cluster-robust standard errors at the study level using the robust() command in the clubSandwich package in R (Pustejovsky, 2025). We examined heterogeneity at each level of the multilevel model by calculating I², which quantifies the proportion of variability in estimates attributable to true differences between studies, within studies, and from sampling error. This calculation was performed using var.comp() command in the dmetar package in R (Harrer et al., 2019). We also evaluated heterogeneity using the omnibus Q-test and absolute variance components (τ₁² and τ₂²; Borenstein et al., 2017).

We examined whether the strength of the associations varied across developmental outcome domains by including dummy-coded moderators for outcome domains at the effect-size level within a multilevel analytic framework. In the first model, we included broad outcome domains (cognitive, psychosocial, physical, and composite) simultaneously, omitting the intercept such that each estimated effect size represented the pooled effect size for that developmental domain. This approach minimizes multiplicity problems (Polanin & Pigott, 2015) and increases power to detect cross-domain heterogeneity (Pigott & Polanin, 2020). We applied this same approach in the second model, focusing on the outcome subdomains. We used pairwise robust Wald tests (implemented in the Wald_test() function in the clubSandwich package, which accounts for multi-level structure; Pustejovsky, 2025) to evaluate whether the estimated pooled effect sizes were statistically different from zero and from one another (Pustejovsky, 2024; Tipton & Pustejovsky, 2015). We reported prediction intervals calculated using the predict() command in the metafor package (IntHout et al., 2016; Viechtbauer, 2010).

Our third and fourth research questions examined whether sample characteristics and study features moderate the strength of associations between cognitive stimulation and children’s development, averaging across all outcome domains. We similarly employed a multilevel meta-analytic model with cluster-robust standard errors, including moderators coded as categorical or dummy variables at the effect-size level to account for within-study variation. For example, the model testing children’s age at the time of stimulation as a moderator included a trichotomous variable: 0–23 months, 24–47 months, and 48–71 months. Each moderator—child age, child sex, country-income level, study quality, how stimulation was measured, and whether SES was included as a covariate—was tested in a separate model, consistent with our preregistered analytic plan and guidelines to ensure a minimum of five studies within each moderator for adequate power and to support interpretation (Jackson & Turner, 2017). Each model was structured such that the reference group is the intercept, and the estimated coefficients represent the difference in the effect size for that category relative to the reference group.

Research Question 1: Systematic Review and Descriptive Summary of Studies

First, we conducted a systematic review to identify studies that reported associations between early childhood cognitive stimulation and multiple developmental outcome domains. After screening 7,010 articles for duplication and exclusion criteria, our final sample comprised 128 studies. Descriptive statistics for these studies are presented in Table 2. The majority of effect sizes were drawn from studies conducted in high-income countries (n = 246; 69.5%), followed by upper-middle-income (n = 45; 12.7%), lower-middle-income (n = 38; 10.7%), and low-income countries (n = 25; 7.1%). Of the 60 nations represented in studies, the modal frequency of representation was once (Belize, Colombia, Costa Rica, Cote d’Ivoire, England, Eswatini, Ethiopia, Ghana, Guatemala, Hong Kong, India, Iraq, Israel, Japan, Kazakhstan, Kyrgyzstan, Lao, Madagascar, Mali, Mauritania, Montenegro, Nepal, North Macedonia, Papua New Guinea, Serbia, Slovenia, South Korea, Suriname, Taiwan, Togo, Tunisia, Turkey, Turkmenistan, Vanuatu, Vietnam, Zimbabwe), while the maximum frequency was 52 times (United States). Many LMICs that were represented only once were part of large comparative international studies (see Supplementary Materials S11 for the full list of countries and a map).

These 128 studies included a total of 354 effect sizes. The majority of effect sizes reported were in the cognitive outcome domain (n = 250, 70.6%), including subdomains such as academic (n = 14, 4.0%), general (n = 37, 10.5%), language (n = 79, 22.3%), literacy (n = 46, 12.9%), numeracy (n = 28, 7.9%), and self-regulation (n = 50, 14.1%) skills. The psychosocial outcome domain was the next-largest group (n = 57, 16.1%), comprising behavioral (n = 25, 7.1%) and socioemotional (n = 28, 7.9%) subdomains. Fewer effect sizes were observed in the physical development domain (n = 23, 6.5%), comprising the motor subdomain (n = 19, 5.4%) and health subdomain (n = 4, 1.1%). The composite domain also had fewer effect sizes (n = 24, 6.8%) compared to the cognitive and psychosocial domains.

Most studies (94.5%) were observational, whereas 5.5% were experimental. Even within experimental studies, effect sizes always reflected the observational association between cognitive stimulation and child outcomes rather than experimental treatment impacts. Extracted effect sizes were largely derived from covariate-adjusted correlational models (n_effects = 336, 94.9%). Although 212 (59.9%) effect sizes were drawn from longitudinal studies, only 152 (42.9%) used models in which stimulation and children’s outcomes were measured at different time points. The number of published studies increased over time—the fewest were published between 2000 and 2009 (n_studies = 18, 14.1%), followed by a notable increase between 2010 and 2019 (n_studies = 58, 45.3%). Within the past five years, the number of published studies nearly matched that of the previous decade (n_studies = 52, 40.6%).

SES was included as a covariate in 81.9% of effect sizes (n_effects = 290), and descriptive patterns show that its use has increased over time in the published literature (74.5% from 2000–2009; 80.8% from 2010–2019; and 85.5% from 2020–2025). Beyond SES, caregiver/child demographics were controlled for in 76.3% of effect sizes, lagged cognitive stimulation or developmental outcome in 29.7% of effect sizes, other parenting-related factors (e.g., well-being, cognitive skills) in 52.0% of effect sizes, early care and education attendance in 20.9% of effect sizes, and all other covariates (e.g., neighborhood characteristics, family size, birth order) in 87.0% of effect sizes. Supplementary Materials S12 shows the most commonly used combinations of covariates in the original studies.

The average age of children across extracted effect sizes was 36.0 months (range 3.5 to 71.5 months), with an evenly distributed number of effect sizes pertaining to infancy (average child age between 0–23 months; n = 109, 30.8%), toddlerhood (24–47 months; n = 106, 29.9%), and preschool age (48–71 months; n = 112, 31.6%). After accounting for overlapping datasets, we estimated 442,749 unique children across studies, of whom 49.0% were identified as female. Caregiver sex was underreported, with data unavailable for 160 effect sizes (45.2% missing); among studies that did report caregiver gender, 95.4% of respondents were female. In terms of study quality, we identified 26 effect sizes nested within 11 studies of low quality. Cognitive stimulation was most frequently measured via caregiver report (n_effects = 274, 77.4%), followed by measures combining caregiver report with assessor observation (n_effects = 78, 21.6%), followed by measures using only assessor observation (n_effects = 2, 0.6%).

Research Question 2: Associations Across Developmental Domains

Our second research question examined the strength of the associations between caregiver cognitive stimulation in early childhood and children’s developmental outcomes, first categorizing outcomes broadly (e.g., cognitive, psychosocial, physical, and composite), then disaggregating results by subdomain. As shown in Table 2, we identified a positive and statistically significant pooled association between cognitive stimulation and each of the four child outcome domains: cognitive (r_p = .12, 95% CI [0.10, 0.14], p < .001), psychosocial (r_p = .09, 95% CI [0.06, 0.13], p < .001,), physical (r_p = .17, 95% CI [0.11, 0.255], p < .001), and composite (r_p = .13, 95% CI [0.08, 0.18], p < .001). The robust Wald F-test indicated that the set of broad domain estimates was significantly different from zero (F-statistic = 49.4, p < .001), and pairwise t-tests revealed no significant differences between most broad domain effect sizes; the only exception was that the physical domain estimate was larger than the psychosocial estimate (p = .048). While the stronger associations observed for physical outcomes may reflect a true substantively stronger association in this domain compared to psychosocial outcomes, it is also possible that this finding was partly driven by a small number of disproportionately large effect sizes. These outliers all had original effect sizes reported as odds ratios and focused on health outcomes, which may have inflated these estimates (see Supplementary Materials S13 for sensitivity analyses excluding odds-ratio-derived estimates). Most of the relative heterogeneity in this model was explained by differences between studies, $I_{\mathit{\text{Level}}3}^3=55.7\%$, followed by within studies, $I_{\mathit{\text{Level}}2}^3=41.2\%$, and the remaining variance due to sampling error $I_{\mathit{\text{Level}}1}^3=3.1\%$. In absolute terms, the estimated variance between studies was τ₁² = 0.006 (τ₁ = 0.08) and within studies was τ₂² = 0.005 (τ₂ = 0.07). Interpretatively, we estimate that for about 95% of populations, true partial correlations between stimulation and child outcomes would land within .16 of our estimated effect sizes between studies, and within .14 of the estimated effect sizes within studies.

We estimated pooled effect sizes within ten developmental subdomains to better understand how cognitive stimulation relates to specific aspects of child development. Results are presented in Table 3. Consistent with patterns for the broad domains, results showed that the associations between cognitive stimulation and each of the ten subdomains were positive and statistically significant. Specifically, the average effect sizes for each subdomain were: academic (r_p = .12, 95% CI [0.09, 0.17], p < .001), general (r_p = .15, 95% CI [0.11, 0.19], p < .001,), language (r_p = .13, 95% CI [0.11, 0.16], p < .001), literacy (r_p = .11, 95% CI [0.08, 0.14], p < .001), numeracy (r_p = .12, 95% CI [0.08, 0.17], p < .001), self-regulation (r_p = .09, 95% CI [0.06, 0.11], p < .001), behavioral (r_p = .07, 95% CI [0.03, 0.11], p < .001), socioemotional (r_p = .12, 95% CI [0.08, 0.16], p < .001), motor (r_p = .17, 95% CI [0.10, 0.23], p < .001), and “other” (r_p = .15, 95% CI [0.09, 0.21], p < .001) skills. As with the broad domains, a robust Wald F-test indicated that each subdomain differs from 0 (F-statistic = 21.2, p < .001). While cognitive stimulation showed positive associations with all subdomains, pairwise t-test comparisons revealed that the estimated associations between cognitive stimulation and behavioral outcomes were statistically significantly lower than estimated effect sizes for the academic, cognitive, language, motor, and composite “other” domains (p < .05). The estimated associations between cognitive stimulation and self-regulation were also statistically significantly lower than the estimated effect sizes of the cognitive, language, motor, and composite “other” domains (p <.05). No other differences between subdomains were statistically significant. These findings demonstrate that a broad conceptualization of cognitive stimulation during early childhood is modestly, positively associated with children’s development across domains, albeit some subdomains were weaker in magnitude than others.

Research Question 3: Variation in Associations Across Sample Features

Our third research question examined whether the associations between cognitive stimulation and children’s developmental outcomes (pooled across domains) varied by characteristics of the original studies’ samples. Table 4 presents the results of three unique models. In these models, the intercept can be interpreted as the pooled effect size for the reference group, and the subsequent coefficients represent the difference for that category relative to the reference group. As shown in the first panel of Table 4, the intercept represents the estimated effect size for studies where the average child age at cognitive stimulation was between 0 and 23 months and was statistically significantly associated with child outcomes (r_p = .14, 95% CI [0.11, 0.17], p < .001), indicating a small but positive association during infancy. Compared with infants as a reference group, studies in which the average child age was between 24–47 months showed a similarly sized average partial correlation (r_p = .00, 95% CI [0.0, 0.04, p = .920) and studies in which the average child age at time of stimulation was 48–71 months showed significantly smaller associations (r_p = −.04, 95% CI [−0.108, −0.01], p = .026). These results suggest that the association between cognitive stimulation and child development is larger for infants and toddlers than for preschool and early school-aged children. Results using an alternative binary operationalization of child age (0–35 months versus 36–71 months) are presented in Supplementary Materials S14 and show no significant moderation of effect sizes.

In the second panel of Table 4, we examined whether the proportion of girls in a study sample modified the strength of the association between cognitive stimulation and child development. Here, the intercept is centered at 50%, representing samples with an even split between boys and girls. As with other models, the intercept was positive and statistically significant (r_p = .12, 95% CI [0.10, 0.14], p < .001), suggesting a positive average association for sex-equivalent samples. Increasing the percentage of female children in the sample did not significantly change the strength of the association (r_p = .00, 95% CI [−0.00, 0.00], p = .944).

The third panel in Table 4 tested the effect of country income status on the estimated associations between cognitive stimulation and children’s development. The intercept reflects the pooled effect size for high-income countries, which was positive and statistically significant (r_p = .12, 95% CI [0.10, 0.15], p < .001). The average effect size for studies conducted in upper-middle income (r_p = .01, 95% CI [−0.05, 0.07], p = .774, lower-middle income (r_p = .00, 95% CI [−0.06, 0.06], p = .959), and low-income (r_p = −.02, 95% CI [−0.07, 0.03], p = .444) countries did not statistically differ relative to the association from high-income countries.

Research Question 4: Variation in Associations Across Study Features

Our fourth and final research question examined whether associations between cognitive stimulation and children’s developmental outcomes varied by study features. These results are presented in Table 5. The first panel shows the results when study quality was examined as a moderator. The intercept represents the pooled effect size for all studies rated as higher quality and, similar to other models, indicates a positive, statistically significant association between cognitive stimulation and child development (r_p = .12, 95% CI [0.10, 0.14], p < .001). Studies rated as lower quality were included as comparators, but the difference in magnitude of the effect size was not statistically significant (r_p = .05, 95% CI [−0.04, 0.13], p = .256). This means that the strength of the association was comparable across studies of varying quality.

Second, we tested whether controlling for SES—a likely confounder—affected the strength of the association between cognitive stimulation and child outcomes, with results shown in the middle panel of Table 5. Here, the intercept—which reflects the average association for studies that did not include SES as a covariate—was statistically significant and positive (r_p = 0.11, 95% CI [0.08, 0.015], p < .001). There was no difference in the average association for studies that did include SES as a covariate, when compared to those that did not (r_p = 0.01, 95% CI [−0.03, 0.05], p = .491). Substantively, this means that the average association between cognitive stimulation and child outcomes was similar regardless of whether its estimation included SES as a covariate.

The third panel of Table 5 presents the association between cognitive stimulation and child outcomes by cognitive stimulation measurement modality, distinguishing between studies that relied exclusively on caregiver reports and those that incorporated assessor observations with caregiver reports. The intercept represents the pooled effect size across studies that used an exclusively caregiver-reported measure of stimulation and indicates a positive, significant association with child outcomes (r_p = .13, 95% CI [0.11, 0.15], p < .001). Studies that used cognitive stimulation measures with an observational component to supplement caregiver reports served as the comparator group and did not yield estimates that differed significantly from those of exclusively caregiver-reported measures (r_p = −.02, 95% CI [−0.05, 0.02], p = .336). These results suggest that cognitive stimulation measurement modality does not significantly shape the strength of estimated associations between stimulation frequency and child development.

We conducted two exploratory moderation models as sensitivity analyses related to study features—cross-sectional versus longitudinal data structure and the number of covariates included in the original analytic models—to assess the robustness of our results. Neither analysis yielded significant moderation results, indicating that the strength of associations did not vary by data structure (Supplementary Materials S15) or by number of covariates (Supplementary Materials S16).

Publication Bias Analyses

Using multiple visual and statistical approaches, we found evidence of potential publication bias. First, the contour-enhanced funnel plot (Figure 2), which displays each transformed effect size against its variance, shows clustering of estimates above zero. While this indicates a true overall positive estimate, a gap of estimates in the lower-left area—where imprecisely estimated null effect sizes would be expected—suggests that null findings from underpowered studies may have been unpublished or unreported (M. J. Page et al., 2020). Contour-enhanced funnel plots for broad developmental outcome domains (see Supplementary Materials S17) showed that the physical and composite domains did not exhibit the expected funnel shape. These plots display several outlier-estimates, reflecting within-domain heterogeneity, likely due to the aggregation of diverse outcome types within domains (e.g., motor skills and health outcomes grouped within the physical domain). In addition, some clustering of estimates around the contour (representing a significance threshold of p < .05) further indicates potential publication bias.

Second, the Egger’s test (Egger et al., 1997) confirmed the presence of asymmetry and indicated a slightly smaller estimate in the hypothetical absence of publication bias (B = 0.071, 95% CI [0.05, 0.09], p < .001). Importantly, the Egger’s test-adjusted estimate remained positive and statistically significant, supporting our interpretation that cognitive stimulation has a small, positive association with child development, even after accounting for other predictors and publication bias. Still, Egger’s test estimates are not a definitive “solution” for accounting for publication bias, particularly when effect sizes are highly heterogeneous. Therefore, to further examine the validity of our findings, we interacted study quality scores with all other developmental domain, sample characteristic, and study feature moderator variables individually (Johnson et al., 2015). No interactions were statistically significant, suggesting that moderation findings were not driven by lower-quality studies (see Supplementary Materials S18). These additional analyses further support the idea that cognitive stimulation has a small but meaningful positive association with child development, despite significant heterogeneity in sample- and study-level features.

State of the Literature

Our systematic review synthesized a growing body of research, particularly within the past five years, on the links between early childhood cognitive stimulation and children’s cognitive, psychosocial, and physical outcomes. Our meta-analysis included 128 studies with 354 unique effect sizes representing 442,749 children across 60 countries, indicating that early childhood cognitive stimulation is well-studied.

We identified several noteworthy trends in the evidence on cognitive stimulation and child development. First, cognitive skills were the most commonly studied outcome, comprising nearly three-quarters of all effect sizes, whereas fewer studies assessed psychosocial and physical developmental domains. This may, in part, reflect the conceptual alignment between cognitive stimulation and cognitive outcomes. Second, we found that evidence is concentrated in high-income countries (69.5% of effect sizes), with notably fewer studies (7.1% of effect sizes) conducted in low-income countries. The overrepresentation of evidence from the United States (and other high-income countries) and the underrepresentation of all lower-income countries reifies longstanding generalizability/representation concerns in psychological science (Nielsen et al., 2017; Thalmayer et al., 2021) and raises important questions about drawing conclusions on decidedly non-universal samples (Henrich et al., 2010).

Third, 78.1% of studies included in our meta-analysis relied solely on caregiver-reported survey measures of cognitive stimulation. This pattern partly reflects our screening criteria, which excluded structured play tasks that assessed interaction quality rather than the presence or frequency of cognitive stimulation. Nonetheless, the predominance of caregiver-reported measures highlights an opportunity for the field to diversify measurement approaches to capture whether and how often cognitively stimulating activities occur, offering additional insights and allowing for triangulation across modalities. For example, time-use diaries that quantify minutes spent on cognitively stimulating activities, or ecological momentary assessments that capture frequency across the day, still rely on caregiver reporting but may be subject to biases different from those of retrospective surveys.

Fourth, nearly all (95.4%) studies relied on reports from female caregivers. This finding may reflect gendered norms regarding the distribution of caregiving responsibilities (including providing cognitive stimulation) within households (Cuartas et al., 2020), in which mothers and other female caregivers serve as the primary informants about children’s experiences. However, the included studies did not always disentangle which caregivers engaged in cognitively stimulating activities from the caregiver reporting on cognitive stimulation. These findings point to an opportunity to improve gender representativeness in future research.

Caregiver Cognitive Stimulation is Positively Associated with Child Development

Consistent with psychological theory, our meta-analytic findings indicate that early childhood caregiver cognitive stimulation was, on average, positively associated with children’s cognitive, psychosocial, physical, and overall development, with average partial correlations ranging from .10 to .17. Prior meta-analyses relying on bivariate associations and focusing on specific activities found similar or somewhat larger positive associations (rs = .08 to .32) with domain-aligned outcomes (e.g., home literacy activities with child language development; Bus et al., 1995; Dong et al., 2020; Castro et al., 2015; Daucourt et al., 2020; James-Brabham et al., 2025; Nag et al., 2024). On the one hand, these slight differences in magnitude may reflect our prioritization of effect sizes from covariate-adjusted models. On the other hand, it could be that our broader conceptualization of cognitive stimulation obscured potential domain-specific associations.

Importantly, our findings were relatively consistent in the magnitude across outcome domains and subdomains, suggesting that “cognitively” stimulating interactions may support multiple developmental domains simultaneously and/or sequentially through shared underlying processes. For example, such caregiver-child interactions could enhance foundational cognitive skills (e.g., executive function, language), which, in turn, have cascading effects on other developmental systems and competencies (D. H. Bailey et al., 2020; R. Bailey & Jones, 2019). This consistency could also reflect the complex ways in which cognitive stimulation is delivered in the real world. Namely, caregivers’ provision of cognitive stimulation may often co-occur with other dimensions of nurturing care, including relational and affective supports that may not have been directly accounted for in the original studies but that also simultaneously support child outcomes. Similarly, because we relied on a broad conceptualization of cognitive stimulation, we are unable to say specifically whether narrower conceptualizations would have led to stronger associations with their corresponding outcome (e.g., whether literacy activities are more strongly associated with literacy than with social-emotional outcomes). Testing patterns between different cognitively stimulating activities and child outcomes is a promising direction for future research, as well as whether interventions that promote holistic cognitive stimulation would be more, less, or equally effective than those that prescribe specific activities.

Overall, we interpret our results as small (cognitive stimulation explains about 1% of the variance in children’s overall development), yet practically meaningful. Indeed, our estimated effect sizes were derived predominantly from models that adjusted for an average of eight covariates, including family SES, caregiver well-being, and early care and education attendance. These positive estimates, therefore, reflect associations that persisted after accounting for other structural and contextual conditions. At the same time, their small magnitude reflects that cognitive stimulation is one of many factors shaping children’s development. The magnitude of our average estimates suggests that efforts to promote healthy child development likely also require additional interventions, including initiatives targeting persistent structural barriers (e.g., economic hardship).

Variation by Sample Characteristics

We found significant variation in the association between cognitive stimulation and developmental outcomes by child age, with studies of infants (0–23 months) and toddlers (24–47 months) showing stronger associations than those of preschool- and early school-aged children (48–71 months). One possible explanation is that as children enter preschool or formal schooling, the relative influence of the home environment may lessen as skills become more differentiated, and as development is shaped by a broader range of learning contexts (Akshoomoff et al., 2013; Sameroff, 2010). During infancy and toddlerhood, caregiver-child interactions may play an especially prominent role in shaping children’s rapidly developing brains and foundational skills, whereas parents’ cognitive stimulation may increasingly contribute alongside schooling or peer contexts for preschool and school-aged children (e.g., Tierney & Nelson, 2009). Alternatively, age-related differences could be confounded by measurement: cognitive stimulation measures may be more sensitive and developmentally appropriate in infancy and toddlerhood and lack sufficient granularity for preschool- and early school-aged children, potentially attenuating observed associations. Prior meta-analytic evidence on age-related moderation is mixed (Bus et al., 1995; Dong et al., 2020; James-Brabham et al., 2025; Nag et al., 2024), which may reflect true sample variability or stem from analytic limitations that ultimately disguise heterogeneity. In line with the latter possibility, when child age was operationalized as a binary variable, we found no evidence of moderation (see Supplementary Materials S12). Empirically disentangling developmental, contextual, and measurement-driven explanations for age-related heterogeneity is an important area for future research.

Moderation results indicated that associations between cognitive stimulation and child outcomes did not differ based on the proportion of female children in the sample, suggesting that the benefits of early cognitive stimulation are relatively consistent across genders. It is important to note that child gender was coded at the sample level (e.g., the proportion of female children), not the individual level, which limits our ability to detect within-sample differences. Several prior studies have shown variation in cognitively stimulating practices by child sex (e.g., Bornstein & Putnick, 2016; Cuartas et al., 2020), which suggests that individual-level sociodemographic characteristics remain relevant.

Despite global disparities in access to early childhood development and education services, which may shape caregiver-child interactions (OECD, 2025), we found no evidence of moderation by country income level, suggesting a universal process that supports early development. These findings are somewhat consistent with two recent meta-analyses on home literacy and numeracy practices that also reported no systematic differences across geographic regions (James-Brabham et al., 2025; Nag et al., 2024). Still, Nag et al. (2024) found a slightly stronger positive correlation between home literacy environments and language/literacy outcomes in countries with greater within-country wealth inequality. It is possible that the operationalization of macroeconomic contexts may matter, such that country-income classifications are not sufficiently fine-grained to capture meaningful within-nation disparities in resources and caregiving supports.

Variation by Study Features

We found no significant variation in the strength of the association between cognitive stimulation and child outcomes across any of the three study features examined: general study quality, how stimulation was measured, and whether analyses included SES as a covariate. On the one hand, this may indicate that the findings are relatively robust to differences in study design and analytic rigor. On the other hand, we may have lacked sufficient statistical power to detect moderation of effect sizes, or inconsistently controlled factors across studies may have obscured true differences. For example, most effect sizes came from models that controlled for at least one SES indicator; among those that did not control for SES, 75% still adjusted for other contextual factors, such as caregivers’ cognitive abilities and early care and education attendance. Notably, despite the absence of moderation by the study-level features examined here, we observed substantial heterogeneity across studies. This suggests that unmeasured study-level characteristics, such as the quality of caregiver-child interactions, household dynamics, and caregiving social norms, may partly explain differences in effect sizes.

Limitations and Future Directions for Research

This study has several limitations worth noting, and we hope this spurs meaningful future research to address these shortcomings. First, as is common in meta-analytic approaches, we were constrained by the available literature, our search terms, and the information reported in individual studies. Studies had to reference our search terms in their titles or abstracts; therefore, studies that used cognitive stimulation as a control variable in their analytic models may have been missed during screening. Relatedly, we only included studies published in English; although a proportion of included studies come from non-English-speaking nations, these results may reflect a monolingual bias (Johnson, 2021).

Furthermore, our broad conceptualization of cognitive stimulation led us to use correspondingly broad search terms. While this approach reflects how cognitive stimulation is typically conceptualized in applied settings, it is imperfect in that it necessarily aggregates across diverse forms of activities and offers limited cultural and naturalistic specificity, both of which constrain its utility in informing targeted interventions. Readers interested in the role of narrower cognitive stimulation activities (e.g., math, reading) are encouraged to consult prior reviews (see James-Brabham et al., 2025; Daucourt et al., 2021; Nag et al., 2024 for recent reviews). The present review cannot disentangle how distinct forms and conceptualizations of cognitive stimulation may differentially relate to child outcomes; future research that explicitly compares activity-specific operationalizations with a more holistic approach can address this gap. In addition, several studies and effect sizes could not be included because the necessary information to calculate partial correlation coefficients was not reported and could not be obtained from the authors. And while we implemented a structured data coding and validation process, we did not calculate intercoder reliability to empirically assess the consistency of extraction across coders. These limitations highlight the importance of transparent reporting practices, including open data repositories, descriptive keywords for database indexing, and digital appendices, to enable better science and more accurate comparison of findings.

Additionally, our research questions and data reflect correlational associations between cognitive stimulation and children’s outcomes, limiting inferences about causality, directionality, and timing of the estimated effect sizes. Experimental designs are therefore needed to clarify whether increases in cognitive stimulation during early childhood causally affect development. For example, observational studies of cognitive stimulation rarely account for characteristics that may influence both parenting and child development, such as parent skill levels or genetic factors. Furthermore, the analytic models from which effect sizes were extracted varied in their covariate adjustment. Some extracted effect sizes were bivariate associations, which may overestimate associations, whereas others controlled for extensive sets of predictors (see Table 1 and Supplementary Materials S11), which may yield more conservative estimates. While we did not find significant moderation by the number of covariates included in a model (see Supplementary Materials S15), model specification heterogeneity may obscure the precision of estimated associations in ways that we did not detect. Early parenting intervention studies offer complementary evidence by experimentally manipulating caregiving practices and testing whether changes in cognitive stimulation impact or mediate children’s development outcomes (Jeong et al., 2019). Cognitive stimulation may also help caregivers become more attuned to their young child, thereby promoting other dimensions of nurturing care (e.g., warmth, responsiveness). Conceptually, this suggests that cognitive stimulation operates within a broader caregiving system in which children’s development is shaped through multifaceted caregiver-child interactions.

Conclusions and Implications

This systematic review and meta-analysis synthesizes evidence on early childhood cognitive stimulation and children’s developmental outcomes from 128 studies and, in doing so, provides several implications for research, policy, and practice. Across a large and growing body of research, we find consistent and positive—though small—associations between cognitive stimulation and healthy developmental outcomes. While this study cannot establish causality or pinpoint mechanisms of change, findings reinforce the promise of parenting interventions that broadly target cognitive stimulation alongside other key parenting practices (Jeong et al., 2018, 2021; Rao et al., 2017). Importantly, the small size of the associations observed in this study suggests that focusing solely on cognitive stimulation may be insufficient to optimize children’s developmental opportunities. Indeed, parenting initiatives should also attend to structural conditions (e.g., economic and political stability, safe neighborhoods, adequate time and resources) that shape caregivers’ ability to provide a wide range of enriched experiences, as well as cultural values and priorities (e.g., displays of affection, family structures) that affect how contextually optimized forms of care are expressed/organized across the world (Lansford, 2022; Rogoff, 1994). These approaches can help ensure initiatives are relevant to families’ lived realities, thereby increasing their potential effectiveness.