Building on an earlier survey by Rosholm et al. (2021), Stine Larsen et al. present an updated meta-analysis of interventions targeting children conducted in Denmark in collaboration with TrygFonden’s Centre for Child Research. Given that children and adolescents constitute one of the most vulnerable groups in society, interventions aimed at improving their learning and wellbeing require particularly careful and rigorous evaluation. As the authors point out, randomised controlled trials (RCTs) are considered the gold standard in causal inference. However, quasi-experimental methods provide a valuable complement for estimating causal effects, particularly when RCTs are not feasible.
Extending the earlier analysis from purely experimental studies to include quasi-experimental evaluations is one of the article’s main contributions, alongside the provision of cost-standardised effect size estimates. The findings can be summarised as follows. First, RCTs and non-RCTs yield similar effect sizes, suggesting that quasi-experimental estimates are, on average, not systematically biased. Second, the authors find support for the Heckman curve hypothesis, which holds that estimated effect sizes of interventions tend to be larger when they are implemented earlier in life.
Third, and perhaps the most surprising finding of the paper, the authors show that the costs of interventions do not correlate with the effect sizes. Combined with the result that interventions focusing on quality improvements appear to be less effective, this raises important questions about the Nordic model, which invests heavily in children’s education and wellbeing but may not always achieve commensurate outcomes. As discussed earlier by Rosholm et al. (2021), the poor quality of educational interventions may help explain why substantial investments by Nordic welfare states do not translate into greater improvements in equality of opportunity.
Which interventions seem to be the most cost-effective? Overall, the two most cost-effective interventions are both quasi-experimental and rely on naturally occurring institutional features, showing that low-cost informational and feedback mechanisms can have substantial educational impact relative to their costs. The most cost-effective intervention is the effect of receiving national test results in mathematics and reading. This quasi-experimental study exploits an unplanned IT breakdown during the introduction of national testing in Denmark, which created random variation in whether students received information about their current performance (see Andersen and Nielsen, 2020). This design allowed the authors to identify the causal impact of performance information itself, showing that simply informing students and teachers about achievement levels can significantly improve subsequent performance at very low cost.
The second most cost-effective intervention concerns feedback based on performance categories, in which small differences in test scores lead to different evaluative labels (see Beuchert et al., 2020). This study also uses a quasi-experimental design based on the mechanical conversion of test scores into five performances. The study examines how this relatively negative feedback affects subsequent achievement, demonstrating substantial effects at minimal cost.
While the article focuses solely on Danish studies conducted in collaboration with TrygFonden’s Centre for Child Research, it provides broader insights that should be considered in all Nordic countries. The results highlight the need to prioritise rigorous evaluation, using both experimental and quasi-experimental methods when designing and scaling up interventions or actual policies. Evidence of many insignificant, and even some negative, effects suggests that well-meaning interventions can, at times, be counterproductive for young people. Extensive regional or school-level autonomy, common in Nordic countries, should not be used as a reason to forgo systematic impact evaluation.
The Larsen study, along with other recent evidence – particularly from Norway (e.g., Haaland et al., 2024) – also demonstrates how well-functioning research infrastructure and close collaboration with local and governmental authorities are crucial for generating robust, research-based evidence on early-life interventions. For example, many of the studies surveyed by Stine Larsen et al. would not have been possible without extensive infrastructure, such as national standardised assessments of student learning at different stages of schooling in Denmark. Comparable infrastructure exists in all Nordic countries – except Finland – thereby enabling evidence-based development and evaluation of early-life interventions. However, it is still too early to assess the extent to which this infrastructure is being fully utilised.