Closing the wisdom gap: Learning from the failure of universal school-based mental health interventions

TL;DR.

• The gap between our wisdom and power is growing and much of what limits wise action is emotional rather than intellectual.
• Mental health programs in schools that teach children to focus on their feelings and symtoms mostly show no benefit, or even harm, while other programs show promise. Psychological science offers possible explanations to why.
• With the help of AI a new way of delivering universal interventions without uniform delivery is possible. An ambitious iterative research effort can start a change that reaches beyond mental health to wisdom, meaning and fulfillment. Effects that can be amplified by the large money already being spent on interventions without effect.
• In this post i lay out a path towards this goal and first steps that may only cost $500.000

Part one is a vision piece and part two a longer analysis of the ideas and science. The second part is expanded and updated with the help of AI, based on earlier drafts from 2017.

Part one: The vision

What if we really tried to give everyone what they need in order to face the challenges they meet in a balanced way? What would a society look like where people have both emotional maturity and a basic understanding of how they themselves, other people, groups and relationships work? With a slowly maturing field of psychology and advances in AI I think we now have a possibility that has not previously existed. My dream is not only to limit depression, anxiety and mental illness but to create a stronger drive towards meaning, happiness and fulfillment in life.

As a clinical psychologist of many years I see mostly where our culture fails and how my patients might have been spared many years of suffering (and sometimes avoided causing pain to friends, family and children). I also see the power of certain interventions to change the foundations of how a life is lived. But why only get the skills and perspectives needed when you have developed a severe mental illness?

The wisdom gap

I find the EA community focused more on mitigating risks than on closing the gap between power and wisdom. When EA talks about wisdom, it is mostly focused on reasoning and forecasting. What limits wise action is at least as much emotional. The inability to stand with uncomfortable inner experiences drives worry, rumination, addiction, procrastination, avoidance, searching for approval, the inability to act on one’s values and much more.

How quickly we become wise enough for the long reflection could be a massively important part of the puzzle toward a beautiful future. As Will MacAskill argues in the Better Futures series, making the future good may matter as much as surviving it. I think building the emotional wisdom of ordinary people and creating a structure for continuous improvement in that direction is one of the best levers for that. 

What could change

The effect would not be limited to better mental and physical health but also less destructive behaviors and people living more intentional lives. I dream about this potentially having an effect on fundamentalism that in turn could affect bioterror risk, misuse of AI, and other harmful behaviors. Thinking about the vast possible changes in culture and interactions that could result from this makes me so curious.

Perhaps there is a way to scientifically guide our culture to more wisdom in a time with increasing power, higher stakes and a more confusing world of information. 

Why now?

The field of preventive psychology is at an inflection point. School-based programs that teach children to focus on their feelings and symptoms mostly show no benefit, and sometimes harm, while other interventions show more promise. There are explanations for this that are discussed in more detail below. Some see this as a sign that we should give up, that we cannot prepare our kids for the challenges they face. I see it more as a reason to embark on a more ambitious scientific mission.

I believe psychology is mature enough to start this kind of journey. We are already running the experiment in most schools, but without the research to know what works and iterate on that.

The failed studies follow a pattern and can be explained by several models. The one I know best is Metacognitive Therapy (MCT), where training children to monitor their emotions is expected to deepen rumination or worry in those already prone to it, which is in line with the evidence from one of the largest trials.

The development in AI makes the previous hurdle of scaling working interventions more manageable as well as creating the possibility for continuous iterations and development. AI also makes the previously difficult trade-off between generality and specificity less burdensome, as AI can tailor interventions to specific individuals and needs while being deployed to all students.

New philanthropic money might arrive that can fund long-horizon work such as this. While AI is increasing our power, the money from it might go toward increasing our wisdom.

What makes me hesitate

The field may not be developed enough. If you asked 100 different psychologists and researchers about what this intervention should contain, then you would likely get different answers. But this is also an argument for the need for this kind of research. My perspective might also influence my belief that this would be transformative, since I meet the people who would have been most in need of such an effort.

What it would take

This could scale from a small team’s preparation and feasibility research to a 10–20-year scientific project and further to policy work, implementation and continuous improvement across the globe. The aspect of continuous improvement is very important considering the young age of psychological science. If feasibility and positive effects are established, other actors would engage in the field, making it self-sufficient. A coherent large effort makes it easier to stop interventions that have no benefit or cause harm. A modest philanthropic investment in feasibility and mechanism research would buy a lot of information about a question with very large stakes.

Invitation

I hope funders, founders, researchers and policymakers find inspiration in this piece and courage from the next section to improve on this idea and start taking steps toward making this a reality. Focus on the first step: contact me to discuss this, or fund the Stage 0 feasibility work (probably under 500,000 dollars).

Imagine the long-term consequences of giving generation after generation better and better capacity to face the challenges in their life and in society at large. 

Part 2: Detailed analysis

Learning from the failure of universal school-based mental health interventions

This one is for readers who want to check whether it holds up: what the trials show, why I think they failed, what a research program would cost and what could go wrong. 

This second part has been rewritten and expanded with the help of AI from a shorter handwritten version from 2017

1. The experiment is already running

In 2017 I worried that policymakers would implement different ideas without evaluating them properly. That is roughly what has happened.

In the United States, 83% of public-school principals reported that their school used a social-emotional learning (SEL) curriculum in 2023–24, up from 46% in 2017–18 (Skoog-Hoffman et al., 2024). The public-school population is about 49.6 million (NCES). So tens of millions of children receive some form of psychological programming every year in just one method in one country, before counting the UK’s whole-school mental-health initiatives and others across the globe. 

Adoption has run far ahead of the evidence. What the effort lacks is measurement, an understanding of mechanisms, and anyone to notice when a program backfires.

2. What the trials show

“Universal school-based intervention” covers at least two literatures. One is mental-health prevention delivered in classrooms: mindfulness curricula, CBT- and DBT-informed lessons, awareness campaigns, programs that ask students to attend to their own mental states. The other is the broader SEL tradition, much of which is behavioral and outward-facing: cooperative learning, social problem-solving practice, content woven into ordinary schoolwork. The recent null and negative results come almost entirely from the first group. The distinction matters because the mechanisms in §3 point at certain types of content, and not the classroom as such as the problem.

The large, well-controlled trials of the programs focused on mental health look like this:

A scoping review of 120 studies of school-based group CBT and mindfulness programs found that 10 of 112 interventions reported at least one negative outcome, and that 5 of the 15 studies at low risk of bias did (Guzman-Holst et al., 2025). All the negative findings came from universal programs (Foulkes, Guzman-Holst & Andrews, 2025). Two cautions apply. More than half of the negative findings appeared only in subgroup analyses, and nearly all studies measured several outcomes, so some of them will be chance. I read the review as a signal about direction. As a rate of harm it is too imprecise, but it agrees with the trials above.

The field knows this. Its own journal has run a formal debate on whether universal school-based prevention should be abandoned (Birrell et al., 2025; Andrews et al., 2025; Cohen et al., 2025), and researchers have published recommendations for measuring harm in future trials (Foulkes et al., 2024).

There is counter-evidence, and it deserves a fair hearing. The largest meta-analysis of universal SEL, covering 424 studies of 252 programs and 575,361 students in 53 countries, reports positive average effects on skills, behavior, wellbeing and academic outcomes (Cipriano et al., 2023). A meta-analysis of follow-up data found effects on emotional distress of about 0.16 SD between six months and 18 years after SEL programs ended (Taylor et al., 2017). Meta-analyses of school-based depression and anxiety prevention find small effects right after the programs (g = 0.21 for depression, 0.18 for anxiety), with targeted programs doing better than universal ones for depression (Werner-Seidler et al., 2021).

Two things reconcile these findings with the trials. First, the SEL averages pool content types, and much of what they contain is the behavioral, outward-facing material that §3 does not implicate. Read that way the literatures agree more than they collide. Second, there is a quality gradient: most studies behind the positive averages are of low quality, and the recent large, well-powered trials are the ones returning null and negative results (Foulkes, Guzman-Holst & Andrews, 2025). Developer-led evaluations are known to inflate effects in prevention science generally (Eisner, 2009), but the SEL-specific test of this was inconsistent (Wigelsworth et al., 2016), so I rest the argument on study quality and power.

My summary: at the current level of design and understanding, universal classroom programs built on introspective and symptom-focused content show no reliable benefit, carry a demonstrated risk of harm, and are being deployed at a large scale anyway.

3. Why the failures make theoretical sense

Nobody predicted these results in print before they arrived. But several clinical frameworks can explain them afterwards, for overlapping reasons, and make different predictions about what would happen under other designs. Here are four, roughly ordered by how much evidence supports them.

Prevalence inflation. Awareness-focused content teaches young people to notice, label and report psychological states as possible mental-health problems. Foulkes and Andrews (2023) propose that this leads to over-interpretation of normal distress and, through self-labelling, to real worsening. Work on “concept creep” shows that lay concepts of mental illness have broadened (Haslam, 2016). Universal awareness programs can therefore add to the very prevalence they aim to reduce.

The metacognitive explanation. Metacognitive therapy (Wells, 2009) locates psychopathology in how we respond to thoughts and feelings more than in their content. With negative effects of sustained self-focused attention, threat monitoring, rumination and worry, which Wells calls the Cognitive Attentional Syndrome. Many universal programs train students to notice their emotions, monitor their stress and check in with their mental state. From an MCT standpoint that is supervised practice in the attentional habits that make up emotional disorder. The model explains the MYRIAD finding that vulnerable students got worse: students already prone to rumination were handed a curriculum that legitimizes and rehearses it. It also predicts that detached and externally focused formats giving students an understanding of why they don’t need to focus that much on their emotions should do better than sustained introspective ones. This is a more specific prediction than using only the prevalence-inflation explanation.

How complex is the subject, and how much training does it take? Healthy emotional competence and wise reasoning are complex and multifaceted. A six-lesson module on mindfulness or mental health cannot give students what they need for the challenges they will meet. Many efforts are also too theoretical, while clinical science suggests that change in emotional functioning depends on practice. In therapy, how much patients practise between sessions predicts outcome (Kazantzis et al., 2016), and exposure works through new learning that happens under real emotional activation (Craske et al., 2014). Among SEL programs, those using sequenced, active, focused and explicit methods do better (Durlak et al., 2011). The school trials fit this picture. Home practice was low in MYRIAD, students who practised more in the DBT trial did better (Harvey et al., 2023), and only 22% finished the app in Future Proofing. This account predicts that even well-designed content will fail at low dose, and that practice-heavy formats will show a dose-response relationship.

Group iatrogenesis. Putting adolescents together and directing their shared attention to distress invites co-rumination (Rose, 2002) and the peer-contagion effects documented in group interventions for at-risk youth (Dishion, McCord & Poulin, 1999). This account predicts effects of format regardless of content.

Current programs combine introspective content, theoretical low-dose formats and uniform delivery to a population whose needs vary enormously. None of the four accounts implies that building emotional competence across a population is impossible. They imply that there is much more to try and that the challenges should not be underestimated. They are also only a few of many possible, testable ideas about what is going on, and they should shape the early design of a program like this one (§9, Stage 1).

4. What works, and what has never been tried

The literature that condemns universal awareness programs also shows what survives scrutiny. Effects are stronger for targeted and indicated interventions for young people with elevated risk or early symptoms (Werner-Seidler et al., 2021). EA Philippines reached a similar conclusion for low- and middle-income countries in a 2022 cause exploration on this forum: universal interventions had been ineffective in reducing symptoms of anxiety, depression and PTSD, while tiered approaches that add targeted support had improved PTSD and depressive symptoms (Javier et al., 2022). A striking long-range example is REDI, a preschool program for 356 disadvantaged children in 25 Head Start centres. Given to 4 year olds over 33 lessons by their teachers after 4 days of training. By ninth grade, the share with very high conduct problems was 6% against 17% in the comparison group, and very high emotional symptoms 3% against 15% (Bierman et al., 2021). It was targeted, behavioral, skills-based and early. Behavioral and skills components tend to do better than awareness and psychoeducation. One-to-one and indirect formats, such as working through parents and teachers or changing the environment, avoid group effects altogether.

A newer example comes from Kenya. Shamiri consists of four weekly one-hour group sessions led by trained lay providers and built on growth mindset, gratitude and value affirmation rather than on symptoms. In a trial with 413 adolescents who had elevated symptoms, it was compared with an active control that taught study skills. Both groups improved, but in analyses with imputed data depressive symptoms fell more in the Shamiri group right after the program (d = 0.35), and the difference was at least as large seven months later (d = 0.45). Anxiety followed a similar pattern (d = 0.37 and 0.44), although the confidence intervals overlap (Osborn et al., 2021). When Shamiri was tested on whole school populations during the COVID-19 pandemic, it did not beat the same active control, which the authors think the pandemic confounded (Venturo-Conerly et al., 2025). The Happier Lives Institute’s 2026 evaluation also found the benefit concentrated in students with elevated distress (McGuire, Dupret & Plant, 2026). Brief, outward-facing content helped the students who needed it. Given to everyone, the same content has not yet shown the same effect.

Two observations set up the rest of this post. The goal is universal access, which does not require uniform delivery. And even the interventions that work have been narrow: short packages aimed at specific symptoms or skills, delivered over months. 

Length alone is not the answer either. Healthy Minds, a four-year life-skills curriculum assembled from existing programmes, showed no effect on emotional wellbeing in a cluster randomised trial in 35 English secondary schools (Lordan & McGuire, 2026). Only about a third of the original design concerned emotional health, and that part relied on CBT-based resilience lessons, emotional awareness and teaching about mental illness; the rest covered relationships, sex, substances and safety (Hale, Coleman & Layard, 2011). The programme improved self-reported physical health and behaviour but not emotional wellbeing, and its effects favored boys. Years of the same kind of content that are prevalent in the universal mental health interventions that show no results or even harm did not change the emotional outcomes.

Nobody has built or tested what I have been curious about for decades: a sustained, multi-year effort that follows the evidence across content types and aims past symptom prevention at something broader. I mean emotional maturity, the capacity to act on one’s values when things are hard, and the psychological foundations children need for the challenges their lives will bring. For a population to live wiser lives, an ambitious program is needed. Whether that broader target can be reached is unknown, because it has never truly been attempted at such a scale.

5. Who is already working on this

A claim of neglect has to survive a map of existing actors, so here is mine.

The ambition is not new. Sternberg (2001) developed classroom approaches to teaching wise reasoning. The positive education movement, starting with Seligman’s collaboration with Geelong Grammar, built wellbeing curricula into schools (Seligman et al., 2009). The Penn Resiliency Program was rolled out across UK schools, where its small effect on depression symptoms was gone within a year (Challen et al., 2011). These efforts shared a model: deploy a fixed curriculum, evaluate it afterwards. None was a mechanism-driven research program that measured harm from the start and had redesign built into how it worked.

Among current actors, Wellcome launched a five-year, £200 million mental-health strategy in 2020 with a mechanism-first logic (its “active ingredients” commissions; Wellcome, 2020), but its frame is disorder: preventing and treating anxiety, depression and psychosis in young people. CASEL and the SEL industry have scale and access to schools. They are implementation and advocacy organisations, poorly placed to discover that their own products cause harm or that another product is better. The academic field is in the middle of its reckoning, but academic incentives reward small publishable trials over decade-long adaptive programs.

In effective altruism, the work done under the label of wisdom (prioritization research on wisdom and intelligence, AI for human reasoning and epistemics, the essay contest on automating wisdom; see for example Gooen, 2021) concerns reasoning, forecasting and judgment, mostly in elites and AI systems. It treats wisdom as cognition. What limits wise action in ordinary lives is at least as much emotional: an inability to allow uncomfortable inner experiences, which drives avoidance through worry, rumination, withdrawal, the search for approval, makes it hard to act on one’s values and so much more. That half of the wisdom gap has no well-funded research program.

A few posts on this forum contribute to this view. A 2022 map of projects to “raise the sanity waterline” lists self-awareness tools such as Waking Up and Mind Ease and treats education as one “conservative” means, but names no work in schools (Matěj, 2022). Jordan Arel’s Building Cooperative Viatopia series sketches a “Children’s Movement” for social, emotional, cognitive and mindfulness development, and “Human-AI Symbiosis” tools aimed at values, wisdom and wellbeing (Arel, 2025). He presents them as proof-of-concept and says implementation should follow rigorous evaluation. This post is an argument about what that evaluation needs to look like, given what the school trials have shown. Will MacAskill’s Better Futures series argues that making the future good may matter about as much as avoiding catastrophe (MacAskill, 2025). Its list of levers centres on AI and governance and reaches human reasoning only through AI tools. The emotional capacities of the people who will do the deliberating are not yet on that list.

The measurement tools exist. The WELLBY (one point of life satisfaction on a 0–10 scale, for one person, for one year) comes from wellbeing economics (Frijters et al., 2020) and is part of UK Treasury appraisal guidance, which values one WELLBY at £13,000 (HM Treasury, 2021). The Happier Lives Institute, a small team, has done most to apply it to charity evaluation.

Each of these actors holds a piece. What no funded program at scale combines is the intersection: research into emotional capability across a population, treated as a positive capacity rather than only as disorder prevention, driven by mechanisms, alert to harm, designed to keep improving, and aimed at broader human functioning. That intersection is the proposal.

6. What AI changes

Why did schools choose uniform delivery when targeting works better? Most future cases come from the low-risk majority, so targeting misses them (Rose, 1985), and screening some students out for special treatment stigmatizes the ones you most need to reach. One teacher delivering one curriculum to a whole class was a compromise forced by the delivery technology.

Digital delivery alone does not solve this. Future Proofing delivered a gamified CBT app to 6,388 students in 134 schools; 22% completed it and it had no effect on depression (Werner-Seidler et al., 2025). Engagement is not to be underestimated but there are a few things that you can try.

Engagement and alliance at scale. The first randomised trial of a generative-AI therapy agent, Therabot, reported large symptom reductions compared with a waitlist across depression, anxiety and eating-disorder risk. Participants used it for more than six hours in four weeks (about 260 messages each) and rated the working alliance on a par with human outpatient therapy (Heinz et al., 2025). This is one developer-led trial against a waitlist, in adults, over four weeks,. So it carries limited weight. What I take from it is narrower than “AI therapy works”: the engagement and alliance numbers touch the variables that have historically sunk some digital mental health, and they are harder to explain by bias than symptom change is.

Tailoring within universal access. An AI-supported layer can give every student access to the same program while delivering different content. Providing the content that interests the child when it is appropriate and able to discuss different options with the person. It can also make skills practice matched to the individual, escalation to human care where risk appears, and could limit introspective monitoring for the students it would harm. The prevention paradox and the stigma problem both lose much of their force.

Continuous improvement built in. Three families of methods make this possible. The multiphase optimization strategy (MOST) uses factorial experiments to test several components at once and keep only those that earn their place (Collins, 2018). Sequential multiple assignment randomized trials (SMART) build the “access for all, tailored delivery” structure directly: students are randomised to a first step, and those who need more are randomised again between ways of stepping up (Murphy, 2005). Platform trials then let new components enter and weak ones leave as evidence accumulates, so the program keeps learning instead of being frozen at launch. The objection that psychology is too young can now be read as an argument for this kind of research program.

For adolescents specifically, the evidence is thin but not empty. A 2025 meta-analysis of 31 randomised trials with 29,637 participants aged 15–39 found small-to-moderate effects of AI chatbots on mental distress (SMD −0.35) (Feng et al., 2025). It covers trials published  from 2014 to 2024, so some of the chatbots tested predate today’s large language models or have used older versions, and the age range stretches well beyond adolescence. A 2026 review limited to adolescents found small-to-moderate effects from few randomised trials, with low certainty (Saputra et al., 2026). Meanwhile the uncontrolled experiment has started here too: about one in eight Americans aged 12–21 already use general-purpose AI chatbots for mental-health advice (McBain et al., 2025), products built for engagement and not for wisdom or emotional capability.

Regulation is moving fast. Illinois’s Wellness and Oversight for Psychological Resources Act (August 2025) prohibits AI from providing therapy or making therapeutic decisions, while allowing administrative support for licensed professionals (Illinois Department of Financial and Professional Regulation, 2025). Utah and Nevada passed their own restrictions. The FDA’s Digital Health Advisory Committee took up generative-AI mental-health devices in November 2025 (Orrick, 2025). The Federal Trade Commission opened an inquiry into AI companion chatbots and their effects on children in September 2025 (Federal Trade Commission, 2025), and wrongful-death lawsuits against chatbot companies are under way. In the EU, Article 5(1)(f) of the AI Act bans systems that infer emotions from biometric data, such as faces and voices, in schools and workplaces, except for medical or safety reasons. The Commission’s guidelines say that inferring emotions from written text falls outside that ban (European Commission, 2025). AI used in education to evaluate learning and steer it can still count as high-risk under the Act, which brings requirements for risk management, documentation and human oversight.

These rules shape the design, and there are sound paths through them. What I propose is adaptive delivery of skills education with human professionals in the loop, inside a regulated research framework. An autonomous AI therapist for children is not part of it. The adaptation can, and on clinical grounds probably should, run on self-report, skill performance and students’ explicit choices, with no automated reading of their emotional states. That keeps clear of the biometric ban and respects the warning in §3 against machinery that watches children’s feelings for them. Jurisdiction becomes part of the feasibility question, and a European host with clear research pathways may be an advantage.

The risks are real. The adolescent evidence is thin and mostly predates current models, safety incidents are recent and documented, regulators are restricting the most ambitious uses, and a badly designed AI layer could inflate prevalence. The technology makes the research program possible without making it safe by default. That argues for doing the work carefully and philanthropically, because the visible alternative is adolescents using engagement-optimised general chatbots and schools buying EdTech wellbeing companions, with no mechanism science behind either.

7. From skills to culture (speculative)

The sections above stand on their own as a public-health argument. This one I cannot prove, but it is where the stakes may be far larger than symptom curves.

Emotional capabilities are passed on. Parents who can regulate themselves raise children in a different environment, and norms about handling conflict, fear and disagreement spread through institutions and peer networks. Deliberate, generation-scale shifts in behavior and norms have happened before: smoking, drunk driving, corporal punishment of children. Sweden’s 1979 ban on corporal punishment, together with public education, changed both behavior and attitudes within about a generation (Durrant, 1999). Experiments on social conventions suggest that a committed minority of around 25% can tip a group’s norm (Centola et al., 2018). The Flynn effect showed that environment can shift a population’s cognitive performance over decades. Its recent reversal in Norway, also shown to be environmental (Bratsberg & Rogeberg, 2018), confirms that population-level capacities respond to environment, in either direction.

Two objections deserve special consideration.

Behavioral genetics. Shared environment explains surprisingly little variance in adult personality and wellbeing (Turkheimer, 2000), so why expect school-age input to leave lasting traces? Heritability estimates describe variation under existing environments and say nothing about a new environmental input; the Flynn effect illustrates this. Trained skills may also behave differently from temperament. Literacy is fragile and “unnatural,” yet societies keep it near-universal through sustained institutional effort. And the objection may be right. Whether trained emotional capabilities persist and pass on is an open question that only long-horizon measurement can answer.

Fade-out. Effects of childhood skill interventions often shrink or vanish within a few years (Bailey et al., 2017, 2020), which threatens any scenario with long persistence. Three replies. Fade-out is best documented for the measured skills themselves, while some programs show later “sleeper” effects on adult outcomes. Perry Preschool is the classic case: its IQ gains faded, yet the program improved adult outcomes, and much of that effect ran through changes in personality skills (Heckman, Pinto & Savelyev, 2013). More importantly, the fade-out evidence comes from the kind of intervention described in §4: short, bounded packages. That literature cannot tell us what happens under sustained, multi-year work built into ordinary schooling and aimed at maturity and valued direction, because that study has not been done. Healthy Minds (§4) is the nearest test of a longer program, and its lack of effect on emotional wellbeing shows that years of lessons are not enough by themselves, the content of those lessons are crucial. Fade-out is therefore one of the program’s central research questions.

8. Importance, tractability, neglectedness, with numbers

Importance: Mental disorder is among the largest burdens on wellbeing in high-income countries by any measure, WELLBYs included. The upside scenario, lasting population gains that pass between generations, would rank among the most valuable outcomes available, though it deserves a heavy speculation discount. Even clear evidence of what gives no effect and what harms would matter, by redirecting what school systems already spend away from content that may cause harm.

Neglectedness: High in the narrow sense mapped in §5 that other actors lack parts of this proposal. Moderate in the broad sense, since neighbouring fields are reasonably well funded. Which could help in that a successful proof of concept would have obvious partners for scaling.

Tractability: Unknown. This is a case for buying information; there is no proven intervention to scale yet. That should not excuse skipping the arithmetic, so here is a deliberately crude model for others to improve.

8.1 Benchmarks

The Happier Lives Institute’s current central estimates: GiveDirectly cash transfers produce about 7.6 WELLBYs per $1,000 (about $132 per WELLBY), and StrongMinds group psychotherapy about 40 per $1,000 (about $25 per WELLBY) (HLI, 2023; HLI, 2025). These are contested. GiveWell’s own assessment of StrongMinds applied stricter discounts and landed several times lower (GiveWell, 2023). I treat $25–130 per WELLBY as competitive for philanthropy.

Two closer comparisons exist. HLI estimates that Shamiri (§4), a school-based group program, produces about 33–37 WELLBYs per $1,000, roughly $27–30 per WELLBY, after counting effects only for the 44% of students with elevated distress (McGuire, Dupret & Plant, 2026). Action for Happiness’s eight-week “Exploring What Matters” course for adults raised wellbeing and prosociality in a randomised trial, with effects lasting at least two months (Krekel et al., 2021). Founders Pledge’s best guess for it was about $188 per point-year of life satisfaction, within a range from $10 to $1,897 (Founders Pledge, 2020). Evaluators use different methods, so these are rough reference points, but they show that both a brief school program and a short course for adults can land within or near the range above.

8.2 Deployment model, if the research succeeds

Assume an AI-supported adaptive program across three school years at $30 per student per year ($90 in total, plausibly conservative at scale). Convert effects using a standard deviation of about 2 points on the 0–10 life-satisfaction scale, so an effect of d = 0.10 equals about 0.2 points.

Where do these effect sizes come from? The central value discounts the follow-up effect on emotional distress in SEL meta-analyses (0.16; Taylor et al., 2017) for the low quality of many included studies. For the optimistic value, the best-known comparison is group psychotherapy for depression. StrongMinds delivers group interpersonal therapy in weekly 90-minute sessions, historically over 12 weeks and now over 6, led by trained lay facilitators. HLI’s meta-analysis of psychotherapy in low- and middle-income countries puts the initial effect at about 0.7 SD, or 0.46 after correcting for publication bias, decaying over about 3.4 years (HLI, 2023). Independent evaluations are more sobering. In a randomised trial of StrongMinds-supported group therapy for adolescent girls in Uganda, 23% of the therapy group had recovered at 12 months against 18% of controls, and the effect had mostly disappeared by two and a half years (Baird et al., 2025). A recent pilot trial found that self-reported improvement shrank markedly when participants were assessed without facilitators present (Atuhumuza et al., 2026).

Those numbers come from people who are suffering and have room to improve. A universal population starts closer to healthy, so the same effort produces smaller average changes. In my clinical work I see dramatic effects from treatments far shorter than a school program, but in people who are suffering. The optimistic value of 0.30 therefore assumes a multi-year program considerably stronger than anything yet tested in universal form. That is what the research would try to build, so read it as an upper scenario even though I believe the upper bound on this kind of intervention when fully developed and iterated on for many years is far higher than that. 

With these values, the central scenario beats GiveDirectly and the optimistic one beats StrongMinds about threefold, while the pessimistic scenario is not competitive. All three are conditional on the research succeeding, which is why the proposal is a research program with kill-criteria. The table leaves out two things on purpose: credit for harm avoided by replacing current content, which counts even at zero positive effect, and the speculative scenarios in §7. Given the fade-out discussion, the 15-year line is exactly the kind of number the program exists to test. If we can build something that is consistently improving the effects of more optimistic long term scenarios can be vastly better. 

8.3 The leverage calculation

US schools alone sit inside an SEL market of roughly $4 billion a year reaching tens of millions of students, allocated with little regard for the evidence in §2. If the research program helps that money get spent better by moving the science forward, that alone would be a large effect.

This post’s own §1 raises an objection. Adoption rose from 46% to 83% during the same years the null results were published, so why believe evidence moves this market at all? I think the surge shows something more specific. Demand is enormous because the problem is felt everywhere: nearly every parent, teacher and principal watches children struggle and wants to give them something. That kind of demand does not wait for evidence when there is no evidence-backed alternative. The surge shows an evidence vacuum, which is different from immunity to evidence. How the market and policymakers would respond to robust evidence that one design works and another causes harm is unknown, because such evidence has never been available. That question is itself cheap to study at Stage 0.

With that caveat: suppose research at Stages 0–1 (about $10 million) did nothing except shift 1% of one country’s annual SEL spending (about $40 million a year) from programs with roughly zero net effect to programs performing at GiveDirectly’s level ($132 per WELLBY). That is about 300,000 WELLBYs a year. If the shift lasts three years, and the chance that the research is good enough and heard widely enough to cause it is 25%, the expected yield is about 225,000 WELLBYs for $10 million, or about $45 per WELLBY. That is inside the competitive range under assumptions chosen to be hard to attack from below: 1% influence, one country, three years, no positive effect of a new intervention, no credit for avoided harm and nothing from §7.

The harm channel runs in parallel. If current universal programs cost even 0.005 life-satisfaction points for one year for a quarter of the 40 million US students reached, in line with the subgroup signals in §2, that is about 50,000 WELLBYs lost per year that better allocation would avoid. Either channel alone would justify the research budget. Every parameter is in the open, and considering  the uncertain nature of new types of interventions I welcome re-parameterisations.

A related argument has been made on this forum. Falk Lieder estimated that developing better interventions for promoting prosocial behaviour could be more cost-effective than delivering the best existing psychotherapy. After commenters showed that some of his assumed effects were implausibly large, his estimate fell from about 110 to about 7.5 times the cost-effectiveness of StrongMinds (Lieder, 2022). 

9. What funding would buy

This is only an illustrative staging, not a budget. People with more insight into the costs of these kinds of interventions are welcome to contribute to better estimations. 

Stage 0: synthesis and feasibility (about $0.3–0.5 million, 12–18 months). A small team produces a systematic review through the lens of §3, a design-principles document, a refined cost-effectiveness model, a regulatory map for AI-supported programs for minors (GDPR, the AI Act, national school law), an assessment of how evidence-responsive the SEL market is (§8.3), and a choice of host environment. Sweden is a strong candidate: national registers allow decades of follow-up, curricula are national and digital readiness is high.

Stage 1: mechanism trials (about $5–15 million, 3–5 years). Factorial trials testing the accounts in §3 directly, as sketched in the table there: introspective monitoring versus external skills content, theoretical versus practice-based formats, adaptive versus uniform delivery. Harm outcomes are co-primary endpoints and follow current recommendations for measuring harm (Foulkes et al., 2024). This stage pays for itself in information whatever the result. Even a clean “no” would reshape how billions in existing school spending are allocated.

Stage 2: adaptive platform (about $20–50 million, 5–10 years). If Stage 1 passes its kill-criteria: a continuously improving, AI-supported program run as a long-horizon scientific instrument, with pre-registered follow-up through register linkage, including the persistence and transmission measurements that the objections in §7 call for.

Stage 3: policy and scaling. I don’t think that is philanthropy’s job to fund. Philanthropy’s job is to create the evidence and the partnerships (Wellcome, national school authorities) that make this the natural next step.

Kill-criteria belong in the design from day one: harm thresholds that pause delivery arms, and evidence requirements at each stage that end the program if unmet. A project pitched on epistemic honesty has to be killable.

10. Objections

Several of these were my own objections in 2017.

“Ask 100 psychologists what the program should contain and you will get 100 different answers.” Probably true. That disagreement is a reason to run factorial trials that turn competing views into testable arms. Expert disagreement is an argument for experimentation.

“The field is too immature; you would be scaling noise.” The proposal scales nothing until Stage 2. Stages 0–1 are the maturation work, with kill-criteria. Immaturity is an argument against deployment, and deployment is already happening, without the research.

“Psychological interventions are notoriously hard to scale. Therapy outcomes depend heavily on the therapist, and engagement depends on adapting to each student’s curiosity.” This was my strongest doubt in 2017, and it is the reason §6 exists. Whether AI-supported delivery with humans in the loop can provide that flexibility is a question for Stage 1.

”AI will have taken over the world before this gives fruit.” I consider this a possible outcome. But I believe that any amount of shift or further understanding of how to give people what they need can either help us navigate a changing world, be a steppingstone for AI to advance our wisdom or in other ways contribute to increasing chances of a positive future. 

“‘Wisdom’ is hopelessly vague.” As a trait word, yes. The program targets components with existing measurement traditions: emotion-regulation skills, distress tolerance, metacognitive flexibility, wise reasoning (Grossmann, 2017) and subjective wellbeing. The word stays in the title as an ambition and probably does not appear in the endpoints.

“Shared environment barely matters, so this washes out” and “fade-out will eat your effects.” Both are addressed in §7. In their strongest forms both are reasons to run the persistence and transmission measurements that nobody has run for sustained programs of this kind.

“AI plus children plus emotions is a safety nightmare.” By default, yes. That is the argument for doing this philanthropically, mechanism first, with harm as a co-primary endpoint and adaptation based on self-report and skill performance, and not leaving the territory to engagement-optimised commercial products as it is today.

“This duplicates Wellcome.” The frames differ (§5): disorder prevention versus positive capacity, and a harm-first design. The cheapest test of this objection is to collaborate with them, which a Stage 0 team would do in its first month.

“Why should anyone fund the author of a forum post?” They shouldn’t, necessarily. The post argues for a program, not a person. If it persuades a funder to commission Stage 0 from a stronger team, it has done its job. Even though i would love to be a part of the team.

11. Conclusion

Universal school-based psychological programs are an unusual case where evidence of failure is also evidence of opportunity. The failures are systematic, they make sense mechanistically, and they happen inside a system that already reaches most children in the developed world. The pieces of a serious response exist: clinical models of mechanism, design principles from targeted interventions, adaptive trial methods, AI delivery that separates universal access from uniform treatment, and registers that allow measurement across generations. They are held by actors who rarely talk to each other. Putting them together is a problem shaped for philanthropy: too long-term for academia, too unprofitable at first for industry, too hungry for evidence for governments.

Part 1 describes what I hope such a program could eventually do. This post is the case that it is worth finding out. The cost of finding out is small. The cost of never finding out is decided by default, one school curriculum at a time.

I welcome disagreement, especially on the mechanisms in §3 and the speculation in §7. If you work at a funder or research organisation close to this and see a flaw or an overlap I have missed, I would like to hear it.

How sure am i? High confidence that §2 describes the trial evidence accurately. Moderate confidence in the mechanisms in §3. The cultural claims in §7 are speculative and marked as such. The staging and costs in §9 are illustrative.

Who I am: disclosures

Who I am: A clinical psychologist with 15+ years of practice across CBT, Metacognitive Therapy (MCT), ACT, EMDR and group treatments, writing from Sweden. I have no academic affiliation. For more than a decade I have wondered what would happen if we really tried to close the wisdom gap. Disclosures are at the end.

In my spare time I create consumer audio courses on emotional skills. That is a potential bias toward believing this field matters, and it is also part of how I know the delivery problem first-hand. I live in Sweden, which I suggest as a candidate host. And as a clinician I mostly see where things go wrong, which may make me overestimate how much a program like this could change.

I used AI (Claude) to search and check the scientific literature and to help draft this post from my handwritten versions from 2017 onwards. Judgments and any errors are mine.

I hope this can inspire a more extensive debate around these questions and that my thoughts can be improved upon and eventually lead to an effort to really try to do our best to prepare the next generation for lives with wisdom, compassion and meaning. Leaving them something to improve upon.

References

Andrews, J. L., Birrell, L., Chapman, C., Teesson, M., Newton, N., Allsop, S., & Slade, T. (2023). Evaluating the effectiveness of a universal eHealth school-based prevention programme for depression and anxiety, and the moderating role of friendship network characteristics. Psychological Medicine (published online 2022). Link

Andrews, J. L., et al. (2025). Debate: Where to next for universal school-based mental health interventions? Time to move towards more effective alternatives. Child and Adolescent Mental Health, 30(1), 102–104. https://doi.org/10.1111/camh.12753

Arel, J. (2025). Building cooperative viatopia [essay series], especially “Shortlist of viatopia interventions”. EA Forum. Link

Atuhumuza, E., et al. (2026). Group interpersonal psychotherapy for depression in Uganda: intervention effects and the influence of social desirability bias in a pilot cluster randomized trial. medRxiv (preprint, not peer reviewed). Link

Bailey, D., Duncan, G. J., Odgers, C. L., & Yu, W. (2017). Persistence and fadeout in the impacts of child and adolescent interventions. Journal of Research on Educational Effectiveness, 10(1), 7–39. https://doi.org/10.1080/19345747.2016.1232459

Bailey, D. H., Duncan, G. J., Cunha, F., Foorman, B. R., & Yeager, D. S. (2020). Persistence and fade-out of educational-intervention effects: Mechanisms and potential solutions. Psychological Science in the Public Interest, 21(2), 55–97. https://doi.org/10.1177/1529100620915848

Baird, S., et al. (2025). Therapy, mental health, and human capital accumulation among adolescent girls in Uganda. Journal of Development Economics, 176, 103473. Link

Bierman, K. L., Heinrichs, B. S., Welsh, J. A., & Nix, R. L. (2021). Reducing adolescent psychopathology in socioeconomically disadvantaged children with a preschool intervention: A randomized controlled trial. American Journal of Psychiatry, 178(4), 305–312. https://doi.org/10.1176/appi.ajp.2020.20030343

Birrell, L., et al. (2025). Debate: Where to next for universal school-based mental health interventions? Child and Adolescent Mental Health, 30(1), 92–95. https://doi.org/10.1111/camh.12749

Bratsberg, B., & Rogeberg, O. (2018). Flynn effect and its reversal are both environmentally caused. Proceedings of the National Academy of Sciences, 115(26), 6674–6678. https://doi.org/10.1073/pnas.1718793115

Centola, D., Becker, J., Brackbill, D., & Baronchelli, A. (2018). Experimental evidence for tipping points in social convention. Science, 360(6393), 1116–1119. https://doi.org/10.1126/science.aas8827

Challen, A., Noden, P., West, A., & Machin, S. (2011). UK Resilience Programme evaluation: Final report (DFE-RR097). Department for Education. Link

Challen, A. R., Machin, S. J., & Gillham, J. E. (2014). The UK Resilience Programme: A school-based universal nonrandomized pragmatic controlled trial. Journal of Consulting and Clinical Psychology, 82(1), 75–89. https://doi.org/10.1037/a0034854

Cipriano, C., Strambler, M. J., Naples, L. H., et al. (2023). The state of evidence for social and emotional learning: A contemporary meta-analysis of universal school-based SEL interventions. Child Development, 94(5), 1181–1204. https://doi.org/10.1111/cdev.13968

Cohen, J., et al. (2025). Debate: Where to next for universal school-based mental health interventions? Tensions in prevention. Child and Adolescent Mental Health. https://doi.org/10.1111/camh.12758

Collins, L. M. (2018). Optimization of behavioral, biobehavioral, and biomedical interventions: The multiphase optimization strategy (MOST). Springer.

Craske, M. G., Treanor, M., Conway, C. C., Zbozinek, T., & Vervliet, B. (2014). Maximizing exposure therapy: An inhibitory learning approach. Behaviour Research and Therapy, 58, 10–23. https://doi.org/10.1016/j.brat.2014.04.006

Dishion, T. J., McCord, J., & Poulin, F. (1999). When interventions harm: Peer groups and problem behavior. American Psychologist, 54(9), 755–764. https://doi.org/10.1037/0003-066X.54.9.755

Durlak, J. A., Weissberg, R. P., Dymnicki, A. B., Taylor, R. D., & Schellinger, K. B. (2011). The impact of enhancing students’ social and emotional learning: A meta-analysis of school-based universal interventions. Child Development, 82(1), 405–432. https://doi.org/10.1111/j.1467-8624.2010.01564.x

Durrant, J. E. (1999). Evaluating the success of Sweden’s corporal punishment ban. Child Abuse & Neglect, 23(5), 435–448. https://doi.org/10.1016/S0145-2134(99)00021-6

Eisner, M. (2009). No effects in independent prevention trials: Can we reject the cynical view? Journal of Experimental Criminology, 5(2), 163–183. https://doi.org/10.1007/s11292-009-9071-y

European Commission (2025). Guidelines on prohibited artificial intelligence practices under the AI Act. Summary: Future of Privacy Forum, Red lines under the EU AI Act: emotion recognition in the workplace and education.

Federal Trade Commission (2025, September 11). FTC launches inquiry into AI chatbots acting as companions. Link

Feng, X., Tian, L., Ho, G. W. K., Yorke, J., & Hui, V. (2025). The effectiveness of AI chatbots in alleviating mental distress and promoting health behaviors among adolescents and young adults: Systematic review and meta-analysis. Journal of Medical Internet Research, 27, e79850. https://doi.org/10.2196/79850

Foulkes, L., & Andrews, J. L. (2023). Are mental health awareness efforts contributing to the rise in reported mental health problems? A call to test the prevalence inflation hypothesis. New Ideas in Psychology, 69, 101010. https://doi.org/10.1016/j.newideapsych.2023.101010

Foulkes, L., Andrews, J. L., Reardon, T., & Stringaris, A. (2024). Research recommendations for assessing potential harm from universal school-based mental health interventions. Nature Mental Health, 2(3), 270–277. https://doi.org/10.1038/s44220-024-00208-2

Foulkes, L., Guzman-Holst, C., & Andrews, J. L. (2025). Potential harm from universal school-based mental health interventions: Candidate mechanisms and future directions. Current Opinion in Psychology, 102196. https://doi.org/10.1016/j.copsyc.2025.102196

Founders Pledge (2020). Founders Pledge charity recommendation: Action for Happiness. EA Forum (linkpost). Link

Frijters, P., Clark, A. E., Krekel, C., & Layard, R. (2020). A happy choice: Wellbeing as the goal of government. Behavioural Public Policy, 4(2), 126–165. Link

GiveWell (2023). Assessment of Happier Lives Institute’s cost-effectiveness analysis of StrongMinds. Link

Gooen, O. (2021). Prioritization research for advancing wisdom and intelligence. EA Forum. Link

Grossmann, I. (2017). Wisdom in context. Perspectives on Psychological Science, 12(2), 233–257. https://doi.org/10.1177/1745691616672066

Grummitt, L., O’Dean, S., Birrell, L., et al. (2025). Efficacy of a school-based, universal prevention programme for depression and anxiety in adolescents (OurFutures Mental Health): a two-arm cluster-randomised controlled trial. eClinicalMedicine, 90, 103672. https://doi.org/10.1016/j.eclinm.2025.103672

Guzman-Holst, C., Streckfuss Davis, R., Andrews, J. L., & Foulkes, L. (2025). Scoping review: Potential harm from school-based group mental health interventions. Child and Adolescent Mental Health, 30(3), 208–222. https://doi.org/10.1111/camh.12760

Happier Lives Institute (2023). Talking through depression: The cost-effectiveness of psychotherapy in LMICs, revised and expanded. Link

Happier Lives Institute (2025). Charity evaluations and StrongMinds comprehensive summary. Link

Harvey, L. J., White, F. A., Hunt, C., & Abbott, M. (2023). Investigating the efficacy of a dialectical behaviour therapy-based universal intervention on adolescent social and emotional well-being outcomes. Behaviour Research and Therapy, 169, 104408. https://doi.org/10.1016/j.brat.2023.104408

Haslam, N. (2016). Concept creep: Psychology’s expanding concepts of harm and pathology. Psychological Inquiry, 27(1), 1–17. https://doi.org/10.1080/1047840X.2016.1082418

Heckman, J., Pinto, R., & Savelyev, P. (2013). Understanding the mechanisms through which an influential early childhood program boosted adult outcomes. American Economic Review, 103(6), 2052–2086. https://doi.org/10.1257/aer.103.6.2052

Heinz, M. V., Mackin, D. M., Trudeau, B. M., et al. (2025). Randomized trial of a generative AI chatbot for mental health treatment. NEJM AI, 2(4). https://doi.org/10.1056/AIoa2400802

HM Treasury (2021). Wellbeing guidance for appraisal: Supplementary Green Book guidance. Link

Illinois Department of Financial and Professional Regulation (2025, August 4). Gov. Pritzker signs legislation prohibiting AI therapy in Illinois. Link

Javier, S., Muñoz, M., Sunglao, A., & Superable, Z. (2022). Cause exploration: Child and adolescent mental health in LMICs. EA Forum (Effective Altruism Philippines). Link

Kazantzis, N., Whittington, C., Zelencich, L., Kyrios, M., Norton, P. J., & Hofmann, S. G. (2016). Quantity and quality of homework compliance: A meta-analysis of relations with outcome in cognitive behavior therapy. Behavior Therapy, 47(5), 755–772. https://doi.org/10.1016/j.beth.2016.05.002

Krekel, C., De Neve, J.-E., Fancourt, D., & Layard, R. (2021). A local community course that raises wellbeing and pro-sociality: Evidence from a randomised controlled trial. Journal of Economic Behavior & Organization, 188, 322–336. https://doi.org/10.1016/j.jebo.2021.05.021

Kuyken, W., Ball, S., Crane, C., et al. (2022). Effectiveness and cost-effectiveness of universal school-based mindfulness training compared with normal school provision in reducing risk of mental health problems and promoting well-being in adolescence: The MYRIAD cluster randomised controlled trial. Evidence-Based Mental Health, 25(3), 99–109. https://doi.org/10.1136/ebmental-2021-300396

Lieder, F. (2022). Developing interventions for promoting prosocial behavior might be more cost-effective at increasing well-being than providing psychotherapy. EA Forum. Link

Lordan, G., & McGuire, A. (2026). The impact on adolescent health and wellbeing from adding evidence-based soft skill lessons to the high school curriculum. Journal of the Economic Science Association. https://doi.org/10.1017/esa.2025.10017

MacAskill, W. (2025). The Better Futures series, including “How to make the future better (other than by reducing extinction risk)”. EA Forum / Forethought. Link

Matěj, R. (2022). Projects that raise sanity waterline. EA Forum. Link

McBain, R. K., Bozick, R., Diliberti, M. K., et al. (2025). Use of generative AI for mental health advice among US adolescents and young adults. JAMA Network Open, 8(11). https://doi.org/10.1001/jamanetworkopen.2025.42281

McGuire, J., Dupret, S., & Plant, M. (2026). The wellbeing cost-effectiveness of the Shamiri Institute. Happier Lives Institute. Link

Montero-Marin, J., Allwood, M., Ball, S., et al. (2022). School-based mindfulness training in early adolescence: What works, for whom and how in the MYRIAD trial? Evidence-Based Mental Health, 25(3), 117–124. https://doi.org/10.1136/ebmental-2022-300439

Murphy, S. A. (2005). An experimental design for the development of adaptive treatment strategies. Statistics in Medicine, 24(10), 1455–1481. https://doi.org/10.1002/sim.2022

National Center for Education Statistics (2024). Fast facts: Back-to-school statistics. Link

Orrick (2025). FDA’s Digital Health Advisory Committee considers generative AI therapy chatbots for depression. Link

Osborn, T. L., Venturo-Conerly, K. E., Arango, S., et al. (2021). Effect of Shamiri layperson-provided intervention vs study skills control intervention for depression and anxiety symptoms in adolescents in Kenya: A randomized clinical trial. JAMA Psychiatry, 78(8), 829–837. Link

Rose, A. J. (2002). Co-rumination in the friendships of girls and boys. Child Development, 73(6), 1830–1843. https://doi.org/10.1111/1467-8624.00509

Rose, G. (1985). Sick individuals and sick populations. International Journal of Epidemiology, 14(1), 32–38. https://doi.org/10.1093/ije/14.1.32

Saputra, R., Hayati, I. R., Lidyawati, Y., Oktasari, M., & Noviza, N. (2026). AI-based interventions for adolescent mental health: A systematic review. Journal of Behavioral and Cognitive Therapy, 36, 100595. https://doi.org/10.1016/j.jbct.2026.100595

Seligman, M. E. P., Ernst, R. M., Gillham, J., Reivich, K., & Linkins, M. (2009). Positive education: Positive psychology and classroom interventions. Oxford Review of Education, 35(3), 293–311. https://doi.org/10.1080/03054980902934563

Skoog-Hoffman, A., Miller, A. A., Plate, R. C., Meyers, D. C., Tucker, A. S., Meyers, G., Diliberti, M. K., Schwartz, H. L., Kuhfeld, M., Jagers, R. J., Steele, L., & Schlund, J. (2024). Social and emotional learning in U.S. schools: Findings from CASEL’s nationwide policy scan and the American Teacher Panel and American School Leader Panel surveys (RR-A1822-2). RAND Corporation. Link

SNS Insider (2025). Social and emotional learning market report. Industry estimate. Link

Stallard, P., Sayal, K., Phillips, R., et al. (2012). Classroom based cognitive behavioural therapy in reducing symptoms of depression in high risk adolescents: Pragmatic cluster randomised controlled trial. BMJ, 345, e6058. https://doi.org/10.1136/bmj.e6058

Sternberg, R. J. (2001). Why schools should teach for wisdom: The balance theory of wisdom in educational settings. Educational Psychologist, 36(4), 227–245. https://doi.org/10.1207/S15326985EP3604_2

Taylor, R. D., Oberle, E., Durlak, J. A., & Weissberg, R. P. (2017). Promoting positive youth development through school-based social and emotional learning interventions: A meta-analysis of follow-up effects. Child Development, 88(4), 1156–1171. https://doi.org/10.1111/cdev.12864

Turkheimer, E. (2000). Three laws of behavior genetics and what they mean. Current Directions in Psychological Science, 9(5), 160–164. https://doi.org/10.1111/1467-8721.00084

Venturo-Conerly, K. E., Osborn, T. L., Rusch, T., Ochuku, B. K., Johnson, N. E., van der Markt, A., Wasanga, C. M., & Weisz, J. R. (2025). Testing the Shamiri intervention and its components with Kenyan adolescents during the COVID-19 pandemic: Outcomes of a universal, 5-arm randomized controlled trial. Journal of the American Academy of Child & Adolescent Psychiatry, 64(7), 786–798. Link

Wellcome (2020). Mental health: anxiety and depression in young people (strategy and active ingredients commissions). Link

Wells, A. (2009). Metacognitive therapy for anxiety and depression. Guilford Press.

Werner-Seidler, A., Spanos, S., Calear, A. L., et al. (2021). School-based depression and anxiety prevention programs: An updated systematic review and meta-analysis. Clinical Psychology Review, 89, 102079. https://doi.org/10.1016/j.cpr.2021.102079

Werner-Seidler, A., et al. (2025). Future Proofing Study: A cluster randomised controlled trial evaluating the effectiveness of a universal school-based cognitive–behavioural programme for adolescent depression. BMJ Mental Health, 28(1), e301426. https://doi.org/10.1136/bmjment-2024-301426

Wigelsworth, M., Lendrum, A., Oldfield, J., Scott, A., ten Bokkel, I., Tate, K., & Emery, C. (2016). The impact of trial stage, developer involvement and international transferability on universal social and emotional learning programme outcomes: A meta-analysis. Cambridge Journal of Education, 46(3), 347–376. https://doi.org/10.1080/0305764X.2016.1195791