In short
This project takes one small question from plan to report. You'll write a dated plan, run a blinded test with real participants on strips or another suitable medium, analyze as planned, report what happened, and have someone else check it. A modest, well-reported study is worth much more than an ambitious, unclear one, and a null result counts as success if you did it well.
What you will learn
- Carry out a complete small study
- Follow your own plan, and report any deviations
- Produce a report that someone else could check
Before this lesson: Writing Up Without Overclaiming
Choosing a project
| Question | Design | Main measure |
|---|---|---|
| Can people detect a reformulation at 2 hours? | Triangle test, about 18 to 30 people | Correct identifications |
| Which dose gives the "cleanest" reading of a musk structure? | Balanced ranking of three doses | Rankings and open descriptions |
| Does a label change how people rate a fragrance? | Same sample, different neutral labels, between groups | Ratings |
| Do two grades of a material differ perceptibly in an accord? | Triangle test | Correct identifications |
| How long is a material detected on a strip, by different people? | Timed detection with blanks | Last detected time |
The label test needs mild, explained concealment, so keep it harmless and tell people at the end. Avoid anything involving mood, health or skin exposure to unassessed materials.
Stages and outputs
| Stage | Output | Lesson |
|---|---|---|
| Question | Refined question and search log | Testable question |
| Reading | Appraisals of at least three sources, or a documented search | Reading critically |
| Design | Design sheet and allocation table | Blinding |
| Numbers | Sample size, critical value, power and rule | Sample size |
| Ethics | Consent, risk list and stop rule | Ethics |
| Plan | Dated, before data collection | Data records |
| Dry run | Changes to the design, plus a fixed plan | All of the above |
| Data | Raw data and deviations | Data records |
| Analysis | Planned results, then exploratory ones | Sample size |
| Report | Up to 2,000 words | Writing up |
| Outside check | A reader's notes and your response | This lesson |
A worked outline, with an invented result
We asked whether adults can detect a replacement of musk X by musk Y at 2 hours on paper strips. We prepared two batches from one premix on one day. A helper coded and randomized 18 triangle sets. Eighteen adults from a local group, screened for stated asthma or fragrance allergy, took part once each, in a ventilated room. Our rule, written beforehand, was to report a detectable difference at 10 or more correct (p = 0.043). Nine chose correctly. This does not reach our criterion. Given a power of about 0.41 for a modest difference, the result doesn't show that the two are equivalent. We recorded one deviation: one participant left early and was replaced under the stated rule.
That result is invented, and it would be a good outcome for this project. It's honest, it's useful, and someone could check it.
An outside check
Ask someone with some research or statistics experience to do three things: recalculate your main result from the raw data, read your method and say whether they could repeat it, and read your conclusion and say whether it goes too far. Respond in writing. If you have no such contact, ask a peer on this track, and say so in your report.
Assessment
| Criterion | Meets it if |
|---|---|
| Plan | Dated before the first observation, with the decision rule stated |
| Design | Control, randomization and real blinding; key held by someone else |
| Ethics | Consent, exclusions, stop rule and suitable medium for the sample's status |
| Data | Raw data unchanged, dictionary included, deviations recorded |
| Analysis | Follows the plan; exploratory results labeled |
| Report | Numbers, group and limits are clear; the conclusion stays within the data |
| Check | Another person reproduced the main figure |
Practical exercise: Final project: a small study
Complete the project above. Allow a month, since your plan, dry run, recruitment and analysis can't sensibly be rushed.
You have done this well if you would be willing to hand your file to a skeptic. Finishing this track doesn't confer a qualification. If you think your work could be published, seek advice from someone experienced first, including on ethical approval, which is usually needed before data collection.
Common mistakes
- Picking too big a question. Answer a small one well.
- Peeking at unblinded results partway. You will be tempted to change something.
- Reporting only a favorable subgroup. Report the plan first.
Knowledge check
Written and reviewed by the RUŌOD Lab team. This is general education, not legal, regulatory, medical or safety advice. Example dilutions and exercises are for learning, are not validated products, and have not been safety-assessed. Completing a lesson confers no qualification. How we check these lessons.