Character as Purpose of Education: How to Be Good

Modern education optimises what people know and abandons what they become; the evidence says formation is learnable, and the institutions that learn this first will produce people who suffer less, live better, and create more value.

Written by the ENSI Foresight Division. Built on a library of 150 primary documents downloaded from the world’s leading institutions.

The argument, before the list

Every developed education system can tell you, to a decimal place, how well its fifteen-year-olds solve quadratic equations. Almost none can tell you whether its graduates keep their word under pressure, act when acting is costly, or hold a discipline for longer than a deadline forces them to. This is not an accident of neglect. It is a structural choice, made and remade every year, and it rests on one quiet assumption: that knowledge is what education is for, because knowledge is what education can measure. The formation gap is the distance between what schools optimise and what lives actually run on — and the central finding of the 150-document library beneath this report is that the gap is now indefensible on the evidence.

Start with what predicts a good life. The LSE life-course work found that a child’s emotional health predicts their adult life satisfaction better than their academic achievement does (Layard et al. 2013; angle 10) — the single most subversive regression in modern education economics, because it inverts the entire ranking logic of schooling. The economics of skill formation reaches the same verdict from the labour-market side: so-called non-cognitive skills — conscientiousness, self-regulation, perseverance — carry predictive weight for wages, employment, health and crime that achievement tests do not capture, and they remain malleable long after IQ has largely set (Heckman & Kautz 2014; angle 13). The OECD’s international Survey on Social and Emotional Skills now documents, across countries and cities, that skills like persistence, curiosity and stress-resistance relate systematically to life outcomes and can be assessed at scale (OECD 2021; angle 09) — and its flagship synthesis treats them explicitly as teachable, not given (OECD 2024; angle 01). The knowledge that schools rank children by is not the knowledge that ranks their lives.

Now look at how schools deliver the part of education that is supposed to shape character — where they attempt it at all. The dominant mode, everywhere from a Prague gymnázium to a US charter network, is exhortation: assemblies, ethics lessons, posters listing the school values, a well-meant lecture on integrity. Telling people to be good is the one delivery mode with the weakest evidence in the entire library. A major US Department of Education randomised evaluation of schoolwide character programmes found essentially no effects on behaviour or character outcomes (IES 2010; angle 15). KIPP — the network that put “character strengths” on its report cards — produced strong achievement effects and flat character-survey effects at scale (Mathematica 2015; angle 09). Moral psychology explains why: judgment is largely intuition first and reasoning after (Haidt 2001; angle 03), so instruction aimed at reasoning routinely fails to reach the machinery that actually generates behaviour. Institutions can even teach the opposite of what they preach — the US Army War College showed how impossible compliance demands train honourable officers to lie routinely (Wong & Gerras 2015; angle 08).

Yet the same library contains a second, far more hopeful pattern. Where formation is delivered as experience — designed struggle, error with feedback, rehearsed situations, repeated practice under mentorship, purpose connected to the work — it measurably works. Two hundred and thirteen school programmes teaching social-emotional skills through active, sequenced practice moved achievement by eleven percentile points (Durlak et al. 2011; angle 09). Ninety-four studies of if-then planning show a medium-to-large effect on converting intention into action (Gollwitzer & Sheeran 2006; angle 07). Error-management training beats error-avoidant training across a 24-study meta-analysis (Keith & Frese 2008; angle 05). A short intervention that merely reframed failure raised achievement in a national sample of over 12,000 adolescents (Yeager et al. 2019; angle 05). A single randomised mentoring programme cut drug initiation, violence and truancy (P/PV 1995; angle 11). None of this is exhortation. All of it is formation by experience: experiment, testing, mistakes, situations, self-correction — the same loop by which people learn every other skill they actually possess.

That is the coming decade’s major move in education, and this report is the evidence case for it. Throughout, we refer to a canon of thirty-two values — eight families of four, with righteousness, boldness and discipline among its flagship members — which Report 2 lays out in full; this report establishes only that such a canon can be formed deliberately, and how. The stakes are not decorative. Formed character is the strongest lever education holds over suffering, flourishing and productive capacity — and for a mid-sized European state like the Czech Republic, which cannot outspend anyone on research or out-scale anyone on talent, being the first system to close the formation gap is one of the few genuinely open strategic moves left on the board. Formation, like foresight, is a capability, not a curriculum unit — and capabilities compound for whoever builds them first.

The main points

• Character predicts life outcomes better than academic attainment — childhood emotional health beats academic achievement as a predictor of adult life satisfaction, and non-cognitive skills carry independent economic weight (Layard et al. 2013; Heckman & Kautz 2014; OECD 2021).

• Knowing the good does not produce doing the good — moral judgment is intuition-led, behaviour follows identity and situation, and whole institutions can teach dishonesty while preaching honour (Haidt 2001; Aquino & Reed 2002; Wong & Gerras 2015).

• Virtue behaves like a skill — acquired by practice, feedback and correction along a novice-to-expert path, exactly as Aristotle claimed and modern neuroscience supports (Stichter 2007; Narvaez 2006; Han 2016).

• Designed failure is the strongest teacher — productive failure, desirable difficulties and error-management training all show struggle-before-instruction outperforms instruction-before-struggle (Kapur 2015; Bjork & Bjork 2011; Keith & Frese 2008).

• Situations beat sermons — VR dilemmas, field exercises and structured dilemma discussion move moral action and measured moral competence where lectures do not (Francis et al. 2016; Lind 2021).

• Habit technology works in real schools — if-then planning and mental contrasting reliably convert intentions into behaviour, including a roughly 60 per cent rise in adolescents’ self-disciplined study (Gollwitzer & Sheeran 2006; Duckworth et al. 2011).

• Whole environments form character; bolt-on curricula mostly do not — the SAFE design conditions, the IES null and KIPP’s flat surveys together define what implementation must look like (Durlak et al. 2011; IES 2010; Jones & Bouffard 2012).

• Mentors and exemplars are the human transmission channel — randomised and meta-analytic evidence for mentoring, and attainable exemplars outperform distant saints (P/PV 1995; Rhodes Lab 2018; Han et al. 2017).

• Purpose is the motivation engine and matching matters — purpose interventions raise performance on tedious work, and values fit predicts satisfaction, commitment and performance (Yeager et al. 2014; Kristof-Brown et al. 2005).

• Character is now measurable — and measurement is dangerous — serious psychometrics exist, but reference bias, faking and Goodhart effects mean measurement must steer formation, never grade souls (OECD 2025; Stecher & Hamilton 2014).

How this report is organised

Ten findings, ranked — not by academic novelty, but by how much each should change what educators actually do. The first three establish that the target is right and reachable: character drives outcomes, knowledge alone does not deliver it, and virtue is acquired the way skills are. Findings four to six are the experiential method itself — failure, situations, habit — the operational core. Findings seven to nine are the delivery conditions: environments, mentors, purpose. Finding ten is the instrument panel, and its warning label. Each finding states the claim, the evidence from the library, the reframe — what to do differently — and what to watch next. Report 2 supplies the canon of values this machinery should form; Report 3 designs the institution and the agentic engine that runs it.

1. Character predicts life outcomes better than academics

The claim. If education’s purpose is good lives — lives with less suffering, more accomplishment, more contribution — then the evidence says schools are optimising the weaker predictor. Character-class variables — emotional health, self-regulation, conscientiousness, prosociality — predict adult well-being, employment and health at least as well as, and often better than, academic attainment. This is the finding that licenses everything else in this report, which is why it ranks first.

The evidence. The cornerstone is the LSE Centre for Economic Performance’s life-course model built on British cohort data: of the childhood factors examined, a child’s emotional health is a better predictor of adult life satisfaction than their intellectual performance (Layard et al. 2013; angle 10). The companion CEP work on well-being economics identifies mental health as the largest single driver of misery in rich countries — larger than poverty or unemployment — and makes the investment case for psychological formation in those terms (LSE CEP 2015; angle 10). From the economics side, the canonical NBER treatise shows that non-cognitive skills predict schooling completion, wages, employment and criminality; that achievement tests capture them poorly; and — crucially for educators — that they remain malleable through adolescence, later than cognitive ability (Heckman & Kautz 2014; angle 13). The standard-setting UK review of non-cognitive skills reached the same conclusion for self-control, motivation and perseverance across life outcomes (Gutman & Schoon 2013; angle 01).

The well-being literature closes the loop from the outcome side. Flourishing has been operationalised into ten measurable features and applied to some 43,000 Europeans, making cross-country comparison of formed character’s end-state possible (Huppert & So 2013; angle 10). Harvard’s flourishing programme formalises five domains of a good life and identifies education as one of only four major causal pathways into them (VanderWeele 2017; angle 10). The World Happiness Report’s 2025 edition documents benevolence and prosocial behaviour as measurable contributors to national happiness (WHR 2025; angle 10). And on the performance side, grit — perseverance for long-term goals — predicts retention in the military, the workplace, school and marriage beyond ability (Eskreis-Winkler et al. 2014; angle 07). Persistence, curiosity and stress-resistance show systematic links to life outcomes across the OECD’s international survey of ten- and fifteen-year-olds (OECD 2021; angle 09).

The reframe. Stop treating character as education’s decoration and grades as its product; the causal arrows run the other way. A ministry that ranked schools by the emotional health and self-regulation of their leavers — even imperfectly — would be ranking on the stronger predictor of national well-being than any league table now in use. For a country like the Czech Republic, whose maturita apparatus measures knowledge with bureaucratic thoroughness while no institution owns formation at all, this is a standing arbitrage: the outcome that matters most is the one nobody is accountable for.

What to watch. OECD SSES successive rounds — whether participating systems start acting on the results rather than filing them; longitudinal returns to social-emotional skill investment in the Heckman tradition; and whether any European ministry moves first to make a formation outcome a headline system indicator.

2. Knowing is not doing — the judgment-behaviour gap

The claim. The formation gap’s engine room is a psychological fact that most curricula are built to ignore: moral knowledge does not produce moral behaviour. People who can reason beautifully about ethics lie, shirk and comply their way into dishonesty at roughly the rates of everyone else, because the machinery that generates behaviour in the moment is not the machinery that passes ethics examinations. Any values education that consists of transmitting correct answers is aiming at the wrong organ.

The evidence. The developmental tradition itself documented the ceiling. Kohlberg’s six-stage account made moral judgment measurable and showed it develops (Kohlberg & Hersh 1977; angle 03), and forty years of Defining Issues Test data confirm that moral reasoning grows with education and can be moved by intervention (Thoma 2014; angle 03) — but judgment scores were never the same thing as conduct. The pivot came with the social intuitionist model: moral judgment is intuition first, reasoning after — the emotional dog wags the rational tail (Haidt 2001; angle 03). Moral Foundations Theory extended the point, mapping care, fairness, loyalty, authority and sanctity as intuitive foundations that are innate-but-educable — shaped by culture and experience, not by argument alone (Graham & Haidt 2013; angle 03). If intuitions drive behaviour and instruction reaches only reasoning, the lecture is structurally the wrong tool.

What does bridge knowing and doing is moral identity — whether being a moral person is central to how someone defines themselves. The construct was introduced and validated as precisely this motivational bridge between judgment and action (Aquino & Reed 2002; angle 03), and subsequent work shows moral identity predicting adolescents’ actual prosocial and antisocial behaviour (Hardy 2014; angle 03). Identity is not installed by exhortation; it is accreted through experience, affiliation and practice — which is why it belongs to the experiential method, not the syllabus. Neuroscience adds that the neural responses underlying moral self-correction are shaped by social context (Ellemers & Van Nunspeet 2020; angle 03), running on a distributed moral brain network built by learning (Pascual et al. 2013; angle 03).

The most damning evidence is institutional. Lying to Ourselves, the US Army War College’s landmark study, found that an organisation with arguably the world’s most elaborate honour instruction had made routine dishonesty a professional survival skill: officers signed off on impossible compliance requirements so habitually that they no longer registered it as lying (Wong & Gerras 2015; angle 08). Seven years later the authors’ retrospective found the culture largely intact (Wong & Gerras 2022; angle 08). The lesson generalises far beyond the military: when the formal curriculum says integrity and the lived institution rewards box-ticking, the institution wins — because the institution is experiential and the curriculum is verbal. Even in simulation, the gap is visible: people’s judgments about moral dilemmas in text diverge from their actions when the same dilemma is enacted around them in virtual reality (Francis et al. 2016; angle 06).

The reframe. Audit every hour of values provision by one question: does this hour change what a person does when no one is grading them? Ethics lectures, value-of-the-month posters and codes of conduct fail that test; identity-building experience, rehearsed situations and honestly-run institutions pass it. And before adding any values instruction, remove institutional hypocrisy — the impossible demands, the tolerated dishonesties — because the organisation is already teaching, continuously, whether it means to or not.

What to watch. Whether professional-ethics fields (medicine, engineering, officer training) shift assessment from knowledge tests to observed behaviour; replication of moral-identity interventions in schools; and any institution brave enough to run a Lying to Ourselves-style audit on itself — a school system that counted its own impossible compliance demands would learn more about its values education than any survey could tell it.

3. Virtue behaves like a skill

The claim. The oldest theory in the library is also the one modern evidence keeps vindicating: virtue is acquired the way skills are — by practice, feedback and correction, along a path from novice to expert, culminating in trained judgment about what a situation requires. This matters because it converts character education from a preaching problem into a training problem, and training is something institutions know how to design.

The evidence. The root text says it plainly: we become just by doing just acts, temperate by doing temperate acts — moral virtue comes about as a result of habit, and practical wisdom (phronesis) is the master capacity that tunes it to circumstance (Aristotle, Nicomachean Ethics; angle 02). Modern moral philosophy rediscovered this after a long rule-based detour: the 1958 broadside that relaunched virtue ethics argued that character, not rule-compliance, is the proper subject of ethics (Anscombe 1958; angle 02). The contemporary sharpening is the virtue-as-skill thesis: a full philosophical statement using the Dreyfus model of expertise acquisition to show that virtues are learned like crafts — through staged practice with correction, from rule-following novice to fluent expert (Stichter 2007; angle 02). Psychology built the same bridge independently: Integrative Ethical Education models moral formation explicitly as novice-to-expert development of ethical skills — sensitivity, judgment, focus, action — merging character education with expertise science (Narvaez 2006; angle 03).

The judgment layer has been made operational. The Jubilee Centre constructed a four-component model of phronesis — moral perception, adjudication between competing values, emotional regulation, and identity — together with an instrument to measure it (Jubilee Centre 2020; angle 02), and subsequent work locates practical wisdom at the core of professional ethical decision-making, exercised and cultivated rather than possessed (Kristjánsson 2024; angle 02). The standard objections — that virtue education is paternalistic, relativistic, refuted by situationist psychology — have been systematically answered (Kristjánsson 2013; angle 02). Neuroscience supplies the mechanism-level support: the way brains actually learn morality — gradual, experience-dependent, emotion-integrated plasticity — fits Aristotelian habituation better than any rival model of moral education (Han 2016; angle 02). And habituation is not a childhood-only window: the case that it runs lifelong makes adult institutions — universities, employers, armies — formation environments whether they accept the role or not (Sanderse 2018; angle 02). The teacher’s role changes accordingly — an Aristotelian pedagogy is a master-apprentice relation in judgment, not a transmission of doctrine (Kakkori & Huttunen 2007; angle 02).

The skill frame imports skill science’s discipline. Expert performance is built by structured, effortful, feedback-rich practice, not innate talent (Ericsson et al. 1993; angle 07) — with the important corrective that deliberate practice explains a meaningful but bounded share of performance variance across domains (Macnamara et al. 2014; angle 07), and Ericsson’s own final restatement of what does and does not count as deliberate practice guards against sloganeering (Ericsson & Harwell 2019; angle 07). Translated to character: repetition alone is not formation; designed practice with feedback on the virtue itself is.

The reframe. Write character objectives the way a conservatoire writes technique objectives: a named skill, a graded progression, designed practice situations, expert feedback, and an expectation of years. A school that treats discipline or honesty as a skill with a syllabus of situations — rather than a trait to be praised — has changed category: from sermon to training. The 32-value canon of Report 2 is best read exactly this way, as a curriculum of trainable skills, not a creed.

What to watch. Validation studies of the phronesis instrument; whether professional schools adopt staged novice-to-expert ethics progressions; and the emerging research question of which values respond fastest to skill-style training — the field’s equivalent of learning curves.

4. Designed failure is the strongest teacher

The claim. The single most counter-intuitive result in the library is that the most effective preparation for learning is engineered failure. Letting learners struggle, err and fail — deliberately, before instruction, with feedback after — beats polished direct instruction for deep understanding, transfer and adaptability. For value formation the implication is structural: character is forged where mistakes are possible, survivable and examined, which is precisely what most schooling is designed to prevent.

The evidence. The flagship is productive failure: learning designs in which students generate solutions to problems beyond their current knowledge — and predictably fail — before receiving instruction. The failure activates prior knowledge, exposes gaps, and prepares deeper encoding when the canonical solution arrives (Kapur 2015; angle 05). The design principles and mechanisms — activation, awareness of gaps, encoding — are codified in the canonical handbook treatment (Kapur & Roll 2018; angle 05), and the effect holds even at micro-scale for procedural algebra knowledge (Ziegler, Trninic & Kapur 2021; angle 05) and extends to introductory programming (arXiv 2024; angle 05). The memory-science twin is desirable difficulties: conditions that make learning feel harder and slower — spacing, interleaving, testing — reliably make it deeper and more durable (Bjork & Bjork 2011; angle 05). The error literature is blunt: error commission followed by corrective feedback enhances learning; errorful learning beats errorless learning (Metcalfe 2017; angle 05). Cognitive neuroscience even documents the machinery of self-correction — post-error slowing and accuracy adjustments after mistakes (Danielmeier & Ullsperger 2011; angle 05) — the micro-loop that situational formation trains.

In adult training the result is a meta-analytic fact: error management training — explicitly encouraging errors and framing them as informative — outperforms error-avoidant training across 24 studies, with the largest advantage on adaptive transfer, performing well on problems unlike those trained (Keith & Frese 2008; angle 05). Modern professional evidence extends it: error-management designs build adaptive expertise in medical image interpretation (Europe PMC 2024; angle 05). And the mindset layer scales nationally: a short online intervention teaching that ability grows — in effect, reframing failure as information rather than verdict — raised achievement among lower-achieving students in a US-national randomised sample of more than 12,000 adolescents (Yeager et al. 2019; angle 05).

The reframe. Design failure in, on purpose, with a safety envelope and a feedback ritual. A formation curriculum should specify, for each value, the failure experiences that teach it: the overreach that teaches boldness its price, the broken streak that teaches discipline its mechanics, the temptation yielded to — in simulation — that teaches honesty its pressure points. The educator’s job shifts from preventing mistakes to curating them: choosing failures that are survivable, instructive and impossible to not notice. Grading regimes that punish every error are, on this evidence, actively anti-formative; what deserves assessment is the quality of the correction, not the absence of the mistake.

What to watch. Productive-failure designs applied beyond STEM to explicitly moral and behavioural content; error-climate research in schools — the classroom conditions under which mistakes become learning rather than shame; and neuroscientific work on why struggle-before-instruction changes encoding, which is beginning to appear in the learning-sciences literature (ISLS 2022; angle 05).

5. Situations beat sermons

The claim. Values are exercised in situations — under time pressure, with incomplete information, against social gradient — and the evidence says they must therefore be trained in situations. Simulated dilemmas, field exercises, role-play and structured dilemma discussion move moral action and measured moral competence in ways that lectures demonstrably do not. The unit of values education is not the lesson; it is the situation.

The evidence. The clearest demonstration of why situations matter is the divergence between judgment and action already noted: when moral dilemmas are enacted around a person in virtual reality rather than described in text, their responses change — simulated moral action is a different psychological event from armchair moral judgment (Francis et al. 2016; angle 06). A 2025 narrative review consolidates a decade of VR moral-dilemma research and its methodological lessons for situational moral training (Europe PMC 2025; angle 06), and VR road-traffic dilemmas under time pressure show that real ethical decision behaviour can be modelled — and rehearsed — in controlled environments (Sütfeld et al. 2017; angle 06).

The institutions that cannot afford the judgment-behaviour gap already train this way. Militaries have integrated ethical scenarios directly into high-intensity field exercises, so that moral choices are made tired, stressed and mid-mission — where they will actually occur (Europe PMC 2014; angle 06). RAND’s early political-military wargames were already surfacing how moral choices appear, or fail to appear, under strategic pressure (Emery 2021; angle 06). Medicine now rehearses moral courage explicitly: the “Should I Say Something?” simulation curriculum trains clinicians to confront professionalism lapses by doing it, repeatedly, in simulation (Europe PMC 2023; angle 06). Computing ethics has working role-play designs, such as the evaluated Privacy by Design simulation game (Springer 2020; angle 06).

For schools, the operational method exists and is manualised. The Konstanz Method of Dilemma Discussion (KMDD) structures whole-class deliberation on real, emotionally live dilemmas, with a teacher-training manual and an evidence base showing moral competence can be measured and taught through it (Lind 2021; Lind 2019; angle 06). The dilemma-discussion tradition has experimental support: combining moral-development instruction with dilemma discussion measurably shifted undergraduates’ justice reasoning on the Defining Issues Test (ERIC 2015; angle 06). This is the empirical redemption of the Kohlbergian programme — not the stages, but the pedagogy: discussing genuinely contested cases, at the edge of one’s reasoning, moves the needle.

Nor do the situations need to be simulated. The philosophical root of the whole method is Dewey’s account of education through experience — continuity and interaction, each experience reshaping the learner who meets the next one (Dewey 1938; angle 04), systematised in the experiential learning cycle (Kolb, Boyatzis & Mainemelis 2001; Kolb & Kolb 2005; angle 04). The real-world evidence bears it out: a 62-study meta-analysis of service-learning found gains in self-attitudes, school engagement, civic engagement, social skills and academics — with structured reflection and student voice moderating the effects (Celio et al. 2011; angle 04), an 11-study meta confirms the learning gains (Warren 2012; angle 04), residential outdoor programmes produced gains in cooperation, conflict resolution and self-esteem (AIR 2005; angle 04), and a 66-study meta documents project-based learning’s effects on thinking skills, attitudes and achievement (Zhang & Ma 2023; angle 04). Real service, real projects, real wilderness — these are situation libraries too, and the reflection requirement is not optional: experience without examination forms nothing in particular.

The reframe. Build a situation library the way flight schools build simulator hours. For each value in the canon, a graded set of situations — discussion dilemmas for the entry level, role-play and simulation in the middle, instrumented real-world situations at the top — with rising stakes, time pressure and ambiguity. The question to put to any school claiming to teach courage or honesty is the flight-school question: how many situations has each student flown, and at what difficulty? This is also where AI changes the economics: generative agents and LLM role-play systems in the library’s fourteenth angle — believable simulated humans (Park et al. 2023; angle 14) and conflict-rehearsal systems with feedback (Shaikh et al. 2023; angle 14) — make situational practice cheap, repeatable and personal for the first time. Report 3 designs that engine in full.

What to watch. Transfer evidence — whether simulated moral action changes behaviour outside the simulator, the field’s decisive open question; adoption of KMDD-style methods beyond their German-speaking base; and the first serious school deployments of LLM-driven dilemma rehearsal.

6. Habit technology works

The claim. Between the situation and the character stands the habit — and habit formation is now an engineering discipline with meta-analytic foundations. The mechanisms that make behaviour automatic are known, the protocols are simple enough for adolescents, and they have raised self-discipline in real schools. Values that never become habits remain opinions; habit technology is how a value gets a body.

The evidence. The mechanics are settled science: habits form when a behaviour is repeated in a stable context until context cues trigger it automatically, gradually shifting control from goals to environment (Wood & Rünger 2016; angle 07). The formation timeline has been translated into practice: the widely used protocol built on the finding that automaticity takes on the order of 66 days of consistent repetition gives educators an honest planning horizon — weeks of scaffolded repetition, not a workshop (Gardner, Lally & Wardle 2012; angle 07). This alone indicts the assembly-and-poster model: nothing done once forms anything.