Moral education through questioning

Moral education through questioning is the attempt to improve how people reason about right and wrong, how sensitive they are to ethical issues, or how they behave, by asking them questions rather than telling them what to think. The questioner may be a teacher, a facilitator, a coach, a counsellor, a researcher or the person themselves. Four traditions share the idea: Socratic and dilemma-based discussion, Philosophy for Children, the question-behaviour effect (asking people about their future behaviour or identity changes what they later do), and coaching or counselling that works through questions. Their main dispute is how far the effects found on reasoning tests reach actual conduct, and whether asking can backfire. Neighbouring work asks the same of AI systems; see AI and moral improvement.

History

Berkowitz’s 1980 account of a six-year research programme on transactive discussion, in which partners engage with each other’s reasoning, defined developmental discussion as discussion after which participants’ Kohlberg stage scores rose from pretest to posttest (Berkowitz, 1980). In 1985 Schlaefli, Rest and Thoma meta-analysed 55 interventions measured with the Defining Issues Test (DIT), a measure of moral judgment (Schlaefli et al., 1985).

Values clarification, an approach to moral education, drew criticism in print. Baer (1982) argued that despite claiming neutrality it indoctrinated students in ethical subjectivism and relativism (Baer, 1982), and Ryan (1989) argued that its claim to produce virtues was not validated and that the approach never became coherent (Ryan, 1989).

The question-behaviour effect has a separate lineage in social psychology and consumer research. A 2003 voter-turnout field experiment replicated earlier self-prophecy work, in which people predict their own behaviour (Gerber et al., 2003); Wood, Wilding and Spangenberg and their colleagues each published meta-analyses in 2016 (Wood et al., 2016; Wilding et al., 2016; Spangenberg et al., 2016).

Core concepts

Socratic and dilemma-based discussion

Dilemma discussion and Socratic questioning ask learners to reason through a concrete problem, with the facilitator probing rather than instructing. The 1985 meta-analysis of 55 interventions found that dilemma discussion and psychological development programmes produced modest overall effects on moral judgment; treatments of 3 to 12 weeks were optimal and programmes with adults aged 24 and over produced larger effects, with significant effects in every group (Schlaefli et al., 1985). A 103-student nursing trial found Socratic questioning improved moral reasoning more than lecture, and both beat a control (Torabizadeh et al., 2018). A 2023 meta-analysis of 214 character education studies named Kohlberg’s Moral Dilemma Discussion among three programmes performing above average on several measures (Brown et al., 2023).

What makes discussion work is studied through transactive discussion. Berkowitz’s programme found it separated stage-advancing from other dialogues. In a dyad study of five 45-minute moral dialogues, significant stage development occurred only for subjects paired with a partner one-third of a stage above them, not for same-stage partners or partners a full stage higher, contradicting the “+1” convention; he flags the data as preliminary (Berkowitz, 1980). Merely descriptive transactive statements did not predict individual development well, and counsellor training raised only descriptive transacts, not transformational ones (Berkowitz, 1980).

The questioner’s relationship to the learner appears to matter. Children produce more self-generated transactive statements, reasoning about reasoning, with peers than with adults, which supports Piaget’s view that peer discussion is more spontaneous (Kruger & Tomasello, 1986). In moral discussions with mothers, children treated moral judgments as non-negotiable, whereas with peers they were negotiable beliefs; mothers discussed with pedagogical intent (Mammen et al., 2019).

Philosophy for Children

Philosophy for Children (P4C) runs a group “community of inquiry” in which pupils pose and pursue questions. A systematic review of ten controlled studies reported a consistent moderate mean effect across a wide range of outcome measures, not specific to moral reasoning (Trickey & Topping, 2004). A 2018 meta-analysis of ten studies (1,509 students) reports a moderate positive effect on cognitive outcomes and a larger effect on reasoning skill, with larger effects in smaller and non-Western studies; it covers only cognitive outcomes, and the authors say prosocial attitudes and behaviour have limited literature (Yan et al., 2018). The paper gives an overall effect of 0.43 (95% CI 0.33 to 0.53) in its results and d = 0.58 in its abstract, an internal inconsistency (Yan et al., 2018).

Direct moral outcomes are sparse and inconsistent. A quasi-experimental Australian study with Year 6 students found no change in pro-social behaviour or emotional well-being relative to non-participants (Youssef et al., 2016). A small Iranian quasi-experiment (n = 37, ten weekly sessions) reported significant improvement in moral judgment and social skills (Yaghoobi et al., 2026). In the school-randomised trial run for the Education Endowment Foundation, P4C pupils made about two additional months’ progress in reading and mathematics, but the primary outcomes were attainment, and the wider outcomes (confidence, listening, self-esteem) rest on teacher and pupil reports rather than measured moral outcomes (Gorard et al., 2015). A follow-up found only a small positive impact on non-cognitive outcomes, using a non-randomised comparison group; the authors say regressions suggested perhaps only a small role for the intervention and call for a larger randomised trial (Siddiqui et al., 2017).

The question-behaviour effect

The question-behaviour effect (QBE) is the finding that asking people about their intentions, or to predict their own behaviour, changes what they later do. A meta-analysis of 116 tests found a small positive effect (d+ = 0.24), larger for more desirable and less risky behaviours, easy behaviours and student samples (Wood et al., 2016). Wilding and colleagues pooled 66 studies (94 tests) and found a small significant effect that was smaller where controls were stronger and risk of bias lower (Wilding et al., 2016). The estimate across prior meta-analyses varies: Rodrigues d = 0.09, Spangenberg d = 0.28, Wood d = 0.24 and Wilding d = 0.15 (Wilding et al., 2016).

Identity-framed questions are a related device. In the original study, framing cheating as being “a cheater” rather than “cheating” reduced it; a replication atlas records the registered-report replication by Guo et al. (2020) as successful in Experiment 1, with no failed replication found (FORRT, 2013a). It records Savir and Gamliel (2018) as another successful replication, with a small effect of self-appeals relative to behaviour-appeals (FORRT, 2013b). Simple moral questions can also act as nudges: asking people to report what they thought the morally right thing to do was before a decision raised donations by about 44 percent in one context (Capraro et al., 2019). That outcome is giving, not honesty.

The proposed mechanisms are many. A 2016 meta-analysis groups them into attitudes, consistency, fluency and motivations and finds varying degrees of support for each, not one clear winner (Spangenberg et al., 2016). A 2012 study reports that self-prediction about socially normative behaviours generates cognitive dissonance, which drives later compliance (Spangenberg et al., 2012).

Self-questioning and reflection

Reflection prompts ask people to examine their own reasoning or history. Applying material from reflection on a personal experience was associated with more ethical decisions among business managers in a lab study, while positive and process-oriented reflection increased the application of case material to new problems (not directly ethicality) (Antes et al., 2012). In a 74-student nursing randomised trial, reflective practice and problem-based learning raised moral sensitivity; scores declined from post-test to three months but stayed well above baseline (reflective +18.6, problem-based +26.6 points versus a flat control) (Zia et al., 2023). Prompting participants to recall past moral transgressions lowered moral self-concept and reduced later dishonest behaviour in a Hungarian lab study (Keller & Szakal, 2025).

Coaching and counselling

Direct evidence is thin. A meta-analysis of ICU studies found that clinical ethics consultation was associated with more consensus on clinical decisions and a shorter ICU stay, with very wide confidence intervals; these are decision and resource outcomes, not measures of participants’ ethical reasoning (Au et al., 2018). A systematic review of 29 articles on clinical ethics committees found no standardised evaluation tools and mostly positive user perceptions, and the authors call for shared criteria (Crico et al., 2021). A meta-analysis of 20 executive coaching studies found a stronger effect on behavioural outcomes than on attitudes and personal characteristics, though ethics was not reported as an outcome (Nicolau et al., 2023).

Questions, controversies, and new developments

Does better reasoning become better conduct?

Moral reasoning correlates modestly with ethical behaviour (r = 0.238 across 50 studies), and negatively with immoral behaviour (Wu & Liu, 2014). Reasoning gains from a graduate accounting ethics course were still present three years later, but the outcome was reasoning, not behaviour (Welton & Guffey, 2009). A meta-analysis of science ethics instruction (150 effect sizes) reports sizable benefits, but its abstract does not say whether outcomes were actual behaviour (Watts et al., 2017). A meta-analysis of 25 business ethics programmes found a minimal overall impact on ethical perceptions, behaviour or awareness, though some moderators suggest how to improve instruction (Waples et al., 2009).

Is intellectual deliberation less effective than experiential learning?

A meta-analysis of 30 research-integrity training studies (75 effect sizes) found experiential approaches with emotional involvement most effective, and primarily intellectual deliberation about ethical problems significantly less effective (Katsarov et al., 2022). This seems to conflict with the DIT and nursing results, but part of the conflict is definitional. Katsarov et al. code “deliberative” courses as engaging with concrete problems without addressing psychological and emotional dimensions; this is not a lack of case engagement (Katsarov et al., 2022). Their finding is relative to practical courses: the judgment outcomes pooled to g = 0.41 (k = 23), and behaviour outcomes to g = 0.69 (k = 5, 121 participants), so it is not a zero effect (Katsarov et al., 2022). They note that deliberative case discussion is the conventional gold standard and that their result challenges it, and that moral judgment develops slowly, which may explain smaller gains from short courses (Katsarov et al., 2022). In their judgment model, higher-education students and non-research professionals learned less than high-school students and general citizens (Katsarov et al., 2022).

How large is the question-behaviour effect for ethical conduct?

Critics argue the QBE evidence is not robust enough for policy: analysing the studies in the Rodrigues et al. review, a different set from Wood’s, they found a prosocial effect of d = 0.05, against Wood’s d = 0.19 (Rodrigues et al., 2016). Wilding et al. found a smaller QBE in low-risk-of-bias studies (g = 0.07, k = 27) than in unclear or high-risk studies (g = 0.20, k = 67), and higher bias scores predicted larger effects; they used broader inclusion criteria than Wood et al. and, unlike them, coded risk of bias as a moderator (Wilding et al., 2016). Their prosocial outcomes were small or null: blood donation g = 0.06 (k = 5) and voting g = 0.06 (k = 8, not significant) (Wilding et al., 2016). A larger replication of the voting self-prediction study found no evidence that self-prophecy increases turnout (Gerber et al., 2003). A 2006 meta-analysis found a small-to-medium self-prophecy effect for health-related behaviours, attributed to dissonance, though the effect was not shown for ethical conduct (Sprott et al., 2006).

Can questions backfire?

Fitzsimons, Block and Williams state that asking people about vice behaviours may unintentionally increase them (Fitzsimons et al., 2007a). In the class-attendance study, students asked to report expectations about skipping class missed class at a higher rate than controls (Fitzsimons et al., 2007b). The earlier drug-use backfire finding was challenged: Schneider, Tahk and Krosnick said the t- and p-values were computed incorrectly, making the exercise effect non-significant, and that the drug-use outcome was so skewed (mostly zeros) that t-tests were invalid (Schneider et al., 2007). Two treatment participants reported 50 drug uses despite predicting the lowest likelihood; removing them made all tests non-significant. The critics said the original paper must be withdrawn but not that the effect is disproven, and did not conclude that prediction questions cannot change behaviour (Schneider et al., 2007). The original authors’ reply states that it presents replications of their own experiment and maintains their warning; the abstract gives no results (Fitzsimons et al., 2007a). An independent longitudinal study of 2,002 US students found that surveying them about past drug use did not raise later drug use, and reports that Rodrigues et al.’s meta-analysis found no effect in seven alcohol-related studies (Briney, 2017).

A different backfire involves moral self-image. Merely presenting contemplation questions, without requiring answers, did not change ethical decisions in the authors’ first experiment (Baader et al., 2025). Requiring explicit answers marginally raised ethical decisions for high moral-identity people but significantly lowered them for low moral-identity people (N = 1,322), and the authors conclude contemplation questions can backfire among those already predisposed to unethical behaviour (Baader et al., 2025). This comes from a working paper.

Does deliberation itself make people more honest?

Shalvi et al. argue that ample time lets people overcome automatic self-serving dishonesty, so more deliberation time reduces cheating (Shalvi et al., 2012). A large study found people more honest under time pressure than under time delay in a sender-receiver game, and notes earlier experiments gave mixed results on whether honesty is intuitive or needs deliberation (Capraro et al., 2018). A preregistered direct replication of Shalvi et al.’s Experiment 2 found time pressure did not increase cheating (FORRT, 2020). These tasks do not use questions, so they bear on questioning only indirectly.

Does questioning indoctrinate?

The critiques of values clarification, an approach built on questions that claims neutrality, are an objection to questioning in general: that the questioner’s stance is smuggled in as method. Baer says it indoctrinates students in subjectivism and relativism (Baer, 1982); Ryan says its virtue claims are unvalidated (Ryan, 1989).

Does character education work, and for whom?

A 2023 meta-analysis of 214 character education studies found a small positive average effect, with signs of publication selection bias that lowered corrected estimates; single-session interventions and mentoring programmes had larger effects (Brown et al., 2023).

Broader connections

Moral education by questioning relies on reasoning and on moral judgments made under the heuristics and biases covered in judgment and decision-making. The replication problems recorded here are an instance of the wider issues in replicability. The question-behaviour effect was smaller in studies with stronger controls (Wilding et al., 2016). The AI version of the topic is covered in AI and moral improvement, and conversational tutoring in AI tutoring.

Further reading

• Schlaefli, Rest and Thoma (1985) Schlaefli et al. (1985): a meta-analysis of Defining Issues Test intervention studies of dilemma discussion.
• Katsarov et al. (2022) Katsarov et al. (2022): a meta-analysis separating deliberative from experiential ethics training, with judgment and behaviour outcomes.
• Wilding et al. (2016) Wilding et al. (2016): a methodological review of the question-behaviour effect that codes risk of bias.
• Schneider, Tahk and Krosnick (2007) Schneider et al. (2007): a detailed reanalysis of the vice-question backfire finding.
• Berkowitz (1980) Berkowitz (1980): a history of the transactive-discussion research programme.
• Brown et al. (2023) Brown et al. (2023): a meta-analysis of 214 character education studies with a selection-bias correction.