The raw pooled result looks spectacular on paper. Using a random-effects model fitted with restricted maximum likelihood estimation, the authors calculated a standardized mean difference of Hedges g equal to 1.29, with a 95 percent confidence interval stretching from 0.73 to 1.85. In the conventional language of meta-analysis, anything above 0.8 is considered a large effect, and a value of 1.29 would suggest that moral education interventions produce gains in value acquisition and moral development that are larger than roughly ninety percent of untreated children would show. If that number were taken at face value, it would rank preschool values education among the most powerful educational interventions ever measured, rivaling or exceeding the effects reported in landmark meta-analyses of school-based social and emotional learning programs.

But the authors were careful to interrogate that number rather than celebrate it, and what they found should change how educators and policymakers read the evidence. The heterogeneity statistic, I-squared, came in at a staggering 94.8 percent, meaning that nearly all of the variability among the pooled effect sizes reflects real differences between studies rather than sampling noise. Even more telling is the prediction interval, which spans from minus 1.22 to 3.79. That enormous range means that in a future study of the same kind of intervention, the true effect could plausibly be strongly negative, null, or extraordinarily positive. When a prediction interval is that wide, the authors argue, the pooled mean does not represent a single expected effect that a practitioner could reasonably anticipate. It is a statistical average of fundamentally different quantities, not a forecast.

The search for what was driving that heterogeneity led the researchers to a striking pattern involving study design and risk of bias. When the analysis was restricted to studies judged to be at low risk or only some concerns of bias, using the Cochrane RoB 2 tool for randomized trials and ROBINS-I for non-randomized studies, the pooled effect fell to 0.79. When a single influential study, Betawi’s 2023 experiment on moral stories and integrity values, was reclassified as high risk, the estimate dropped further to 0.53. In other words, the more rigorous the evidence, the more modest the benefit, a classic signature of bias inflating apparent effectiveness. Design family alone explained 85 percent of the between-study variance in the meta-regression, an unusually high proportion that points to methodological artifacts rather than genuine differences in children or curricula.

The design split is particularly revealing. Five of the six studies testing full curricular programs, the kind of multi-week values education packages often embedded in national preschool frameworks, were rated at high risk of bias, and these were the studies producing the largest effects. By contrast, all ten of the brief moral-story experiments, in which children hear a short narrative and are then assessed on behaviors such as honesty, sharing, or helping, were rated at low risk of bias, and these produced moderate effects. The pattern is intuitively understandable: a brief story experiment with blinded outcome assessment is far easier to conduct cleanly than a semester-long curriculum delivered by teachers who know which children received it and measured with instruments completed by those same teachers or by parents aware of the program. But it means the two literatures are not really measuring the same thing, and pooling them obscures more than it reveals.

The moral-story experiments themselves form a fascinating and methodologically healthy body of work. Building on the influential finding by Lee and colleagues in Psychological Science that classic honesty stories differentially affect children’s lie-telling depending on whether the story ends with positive or negative consequences, researchers have tested narratives emphasizing protagonists’ negative emotions about misdeeds, stories promoting confession and helping behavior, and even augmented reality-enriched tales designed to encourage altruistic sharing. Recent work has refined the recipe further: studies from Chinese laboratories have confirmed that moral stories can promote honesty in young children, while other experiments have found that emotional elements such as a protagonist’s guilt over lying promote honesty in early elementary school children but not in preschoolers, suggesting developmental boundaries on when these narrative interventions take hold. Questions about whether anthropomorphized animal characters are as effective as human protagonists have also been tested, with mixed results.

The curricular studies, for their part, are not without promise, but their designs leave too much room for alternative explanations. Programs built around picture books, creative drama, music, role play, and structured activity sequences have been implemented in preschool settings from Türkiye to Jordan to Taiwan, often aligned with national early childhood frameworks. Outcomes have been measured with instruments such as the Preschool Social Values Acquisition Scale and teacher and parent report forms. Yet when raters applied modern bias assessment tools, the recurring problems were predictable: unblinded outcome assessors, attrition that was not accounted for, and selection of participants in ways that could favor the intervention group. These are exactly the design flaws that decades of research in education and clinical psychology have shown to systematically exaggerate treatment effects.

The authors also examined publication bias using funnel plot asymmetry tests, trim-and-fill analysis, and failsafe N calculations, the standard toolkit for detecting whether small or null studies have gone missing from the published record. Their broader conclusion, however, rests less on those diagnostics than on the risk-of-bias sensitivity analyses and the prediction interval, both of which counsel against treating the pooled estimate as a practical guide. The authors explicitly state that the pooled figure should not be interpreted as a single expected intervention effect, and they call for registered preschool trials with blinded assessment as the necessary next step. That recommendation echoes a long-standing theme in the moral education literature, going back to meta-analyses of moral judgment interventions in older students and comprehensive syntheses of character education programs, which have repeatedly found that enthusiasm for values education has outpaced the rigor of its evidence base.

Why does this matter beyond the seminar room? Values, character, and moral education occupy a prominent place in early childhood policy around the world, from statutory early years frameworks in the United Kingdom to national preschool curricula in Türkiye, and governments invest real resources in programs premised on the assumption that moral learning can be deliberately accelerated in the preschool years. Developmental science offers genuine grounds for that optimism: infants show social evaluation preferences before their first birthday, toddlers help spontaneously, and children across diverse societies navigate fairness and prosociality from early ages. But the new analysis suggests that the leap from that developmental foundation to specific, curriculum-driven gains has been documented far more convincingly in tightly controlled story experiments than in the classroom programs that actually reach most children. The honest reading of the evidence is that brief, well-designed narrative interventions can modestly shift specific moral behaviors in the short term, while the case for comprehensive values curricula remains, for now, built on studies too weak to bear the weight placed upon them.

The study’s authors have deposited their analytic dataset on the Open Science Framework, allowing other researchers to scrutinize and extend the work, and they report no funding or competing interests. Their transparency extends to a detailed account of generative AI use in drafting analysis code, with the authors verifying all outputs and retaining responsibility for the analysis. What they have produced is less a celebration of moral education than a methodological warning shot: in a field where the outcomes are measured by the same teachers and parents who deliver the interventions, effect sizes will inflate unless researchers build in blinding, preregistration, and independent assessment. Until such trials are conducted, parents and educators can take modest comfort in the moral-story findings, while treating grand claims about values curricula with the statistical caution this meta-analysis so carefully demonstrates.

Subject of Research: Effects of values, character, and moral education interventions on preschool children's moral development

Article Title: The effects of values, character, and moral education interventions on preschool children’s value acquisition and moral development outcomes: a systematic review and meta-analysis

Article References: Boz, M. E., & Şen, F. N. (2026). The effects of values, character, and moral education interventions on preschool children’s value acquisition and moral development outcomes: a systematic review and meta-analysis. SN Social Sciences, 6(10), Article 442. https://doi.org/10.1007/s43545-026-01742-9

Image Credits: AI Generated

DOI: 10.1007/s43545-026-01742-9

Keywords: moral education, values education, character education, preschool children, meta-analysis, systematic review, moral stories, risk of bias, early childhood education, prosocial behavior, honesty, effect size