Transparent assignments
Half the class answered a question you did not ask
Half the class answered a question you did not ask, and the half that got it right were the ones who had seen this kind of assignment before.
What it is
Three questions answered in writing, on one page, before the student begins. The workshop taught it as the TILT framework.
The version most of us hand out states the deliverable and the due date and assumes the rest is obvious. It is obvious, to students who have written this genre before. That is the mechanism, and it is why the practice reaches inexperienced students hardest.
Run it
- Purpose. What they will be able to do afterward, and where it is used later. Not “this assignment assesses your understanding of measures of association.”
- Task. What to do, in the order they will actually do it, with the step people get wrong named at the point where they will be doing it.
- Criteria. What separates full credit from partial from none, written as observable differences rather than adjectives.
Say this
Minimum viable version
Add three headings to the assignment you are about to hand out. Purpose: two sentences on what they will be able to do and where it is used later. Task: what to do, in working order. Criteria for success: per component, one line for full credit and one for the near-miss you actually see.
Copy: blank template · worked example from EPI 501
Some assignments are supposed to involve productive floundering. Do not go back to vagueness; be transparent about the withholding instead. Say that the purpose is for students to invent and test their own approach, so no procedure is provided. Students tolerate deliberate difficulty. They cannot tolerate not knowing which kind they are in.
In my courses
Used in EPI 501, where the worked example above comes from. Planned for the scaffolded study critiques and the final paper in EPI 601.
Two places in a methods course where this does the most work. Any prompt of the form “select an appropriate model and justify it” is a transparency problem in a methods costume, so write the criteria for the justification separately from the criteria for the choice; a defensible justification of a suboptimal choice is usually worth more than the reverse, and students do not know that. And the generic limitations paragraph comes from the same source: write into the criteria that naming a limitation and saying which way it biases the estimate earns credit, while “this study has limitations” does not.
Evidence
One independent replication reproduced the effect on student perceptions at d = 0.35 (Woods et al., 2024). Nobody has shown it changes what students learn.
The most-cited source appeared in a practitioner magazine without external peer review, and its effect sizes are the authors’ own characterization at a level of statistical detail no methods reviewer would accept (Winkelmes et al., 2016). Its study parameters are reportable here only because Palmer et al. (2018) supplies them.
The outcomes are self-reported confidence and belonging. Those are real constructs worth caring about, and a student who reports more confidence has not been shown to have learned more.
The populations are undergraduate, first-generation, and low-income. A doctoral cohort already survived a selection process that filtered on exactly the genre knowledge this supplies, so the effect is plausibly much smaller. It is probably not zero, because what a good qualifying-exam answer looks like is a genre nobody has seen either.