Active and experiential learning
You talked for ninety minutes and cannot tell what landed
The activity technically happened, the room made noise for four minutes, and you have no idea what anyone concluded.
What it is
Students doing something with the concept during class rather than receiving it. The fellowship treated this as a family of activities rather than one procedure, so what follows is the shared skeleton and then the specific techniques it named.
The skeleton matters more than the choice of activity. “Turn to your neighbour and discuss” is not an activity, and the difference between it and a structured one shows up in who participates.
Run it
- Ask something with more than one defensible answer. “Compute the odds ratio” produces one right answer and one confident student. “Would you adjust for this variable, and what would change your mind” produces a room.
- Make the first pass private, written, and silent. Ninety seconds with a pen. This reaches the students who will never speak cold, and it is the step that gets cut when you are behind.
- Structure the exchange. Name the pairing, the time, and what the pair has to produce.
- Collect something. A card, a photographed sheet, a line in a form. Collection is what tells students the private step was not decorative.
- Close it yourself. Say what the right answer was and why the attractive wrong one was attractive.
Skip step two or step four and you have a discussion, which is a different thing with a different distribution of who participates. If you run out of time, cut the sharing, not the closing.
Techniques
Documented problem solutions
Students write every step they took, not the answer, so the break in the reasoning is visible. The highest-yield of these in a quantitative course, because a correct final number conceals where the reasoning failed.
The prompt
“Do not give me the answer. Give me every step you took, in the order you took them, including the ones you abandoned. Where you compared two numbers, write down which two and why those two. If you got stuck, write down what you were looking for when you got stuck.”
What’s the principle
Six short vignettes in mixed order. Students name the threat in each before touching any numbers, and nobody solves anything. For a methods seminar the six are confounding, effect modification, mediation, collider stratification, selection, and measurement error. The diagnosis is the skill, and a problem set organized by chapter never requires it.
Application cards
At the end of a concept, one index card: name one place this idea appears in your own data, and one place it looks like it applies but does not. The second half does the work, because naming a false positive requires knowing the boundary. Two minutes of setup, five of class.
In my courses
Run in EPI 501. Planned for EPI 601, even weeks.
Each major topic (causality, measures, standardization, interaction, matching, validity) will include a hands-on activity where students apply the concept to a real published epidemiologic study. For example, students will take a published paper and identify threats to validity, propose alternative designs, or re-derive epidemiologic measures from raw data.
The 601 plan attaches a specific activity to a specific paper each week: re-deriving measures from Baylin et al., drawing the DAG for Westreich and Greenland, identifying selection bias in a published study. Used before the fellowship, adapted from EPI 501.
Evidence
Active learning as a category is well supported. Across 225 studies, exam performance improved (d = 0.47) and failure rates fell from 33.8 to 21.8 percent, with the largest effects in classes of 50 or fewer and no publication bias detected (Freeman et al., 2014). Achievement gaps for underrepresented students narrowed (Theobald et al., 2020).
The individual techniques do not inherit that. The meta-analysis pools everything from clicker questions to studio courses, and it cannot tell you which activity to run on Tuesday. Anyone citing d = 0.47 in defence of a specific activity is substituting a category for an exposure.
Of the techniques above, think-pair-share has one rigorous trial, in ninth grade, with hand-raising as the outcome (Mundelsee & Jurkowski, 2021), and an origin chapter containing no data (Lyman, 1981). The other three come from a classroom-assessment handbook that reports no controlled trials (Angelo & Cross, 1993). Cite it as a source of procedures, never as evidence that a procedure works.
Step five is the one part with a direct randomized warrant: discussion plus instructor explanation beat either alone (Smith et al., 2011). Two corrections, because both circulate as settled. The study usually offered as proof that peer discussion works has no control condition (Smith et al., 2009). And clickers versus paper flashcards produced no difference in learning gains (Lasry, 2008), so the active ingredient is the discussion rather than the device.