The Real Choice: Autonomy With the Right Constraints
Self-directed learning is not the same as learning without a teacher. In health professions education, SDL means that learners own meaningful decisions across a learning cycle: they diagnose needs, form goals, identify resources, choose and apply strategies, and evaluate what happened. An educator may be present throughout that cycle, provided support improves the learner’s next decision rather than taking every decision back. The practical judgment of this series is therefore not teacher versus no teacher. It is whether the design gives learners usable ownership, supplies the structure the task requires, and measures the outcome it claims to produce.
What the evidence allows us to say
The strongest supported benefit of SDL is a qualified improvement in measured learning, especially knowledge and skills assessed through tests or formal tasks. The broad health-professions review found a small-to-moderate pooled effect, Cohen’s d = 0.34 with a 95% confidence interval from 0.04 to 0.64, across 48 studies and 74 effect sizes. The estimate is meaningful, but it is not a promise that every learner, course, or profession will see the same gain. Heterogeneity was high at I² = 87%, which means the average compresses important differences among interventions, learners, outcomes, and contexts.
A narrower review of undergraduate medical education found that exam scores favored SDL by a mean difference of 2.399 points, with a 95% confidence interval from 0.121 to 4.678. Eleven of 14 meta-analyzed studies favored SDL and three favored traditional didactic learning. Yet heterogeneity was even higher, at I² = 98.56%. The result supports a positive average difference on those measured exams, not a general law of instructional superiority. The score difference is also on the exams’ own scale, not another version of the standardized Cohen’s d estimate, so the two pooled results should not be combined into a single ranking.
The outcome boundary is more important than the headline. In the broad review, 98 of 125 studies assessed Kirkpatrick Level 2 knowledge or skills, 27 assessed Level 3 behavior, and none assessed Level 4 patient or system outcomes. In the undergraduate review, the main quantitative outcome was exam performance. These studies therefore support a cautious claim about learning measured near the educational activity. They do not establish that SDL improves clinical behavior, patient outcomes, or health-system performance. The absence of Level 4 evidence is not evidence that SDL can never affect patients; it means that this evidence base did not test that question.
The evidence also explains why implementation cannot be separated from efficacy. In the broad review, learners were involved in choosing resources in only 21% of studies and in assessment in 25%. Teachers were reported as absent in 77 of 125 studies, facilitators in 39, and content sources in 9. These are reports of coded implementation, not a complete audit of every learner’s experience. Even so, they reveal a persistent gap between SDL as a theory of learner ownership and SDL as an intervention label. A positive result from a poorly described intervention cannot tell us whether the benefit came from self-direction, from the content or technology, from unreported facilitation, or from some combination.
The evidence is similarly cautious about teacher role. Facilitator designs had the largest estimated effect in the moderator analysis, but teacher role did not significantly moderate effect size. The facilitator estimate was d = 0.54, teacher-absent designs d = 0.28, and teacher-as-content-source designs d = −0.07, with wide and overlapping uncertainty. These results make facilitation a plausible design hypothesis, especially for complex learning, not a settled causal rule. Facilitators may be assigned to different tasks, used with different modalities, or supported by better resources. The studies do not isolate all of those influences.
The most defensible conclusion is therefore conditional.