In short
Good fragrance experiment records let someone else see exactly what you did and recalculate your results: the raw observations stay unchanged, each column is explained, the plan is dated before the data arrive, and every analysis step can be repeated. It sounds bureaucratic, and it takes a few minutes each session. It's what separates a study from an anecdote, and it's also your defense against your own memory.
What you will learn
- Lay out raw data so it can be analyzed and checked
- Pre-register a plan in dated form
- Decide what to share, and what to protect
Before this lesson: Ethics of Testing on People
Keep one row per observation
A tidy layout has one row for each judgment and one column for each variable. This makes counting and checking simple.
| session | assessor | order | sample_code | true_version | time_min | rating | comment |
|---|---|---|---|---|---|---|---|
| S1 | P03 | 2 | 418 | B | 15 | 6 | "sharp" |
| S1 | P03 | 1 | 752 | A | 15 | 4 | |
| S1 | P07 | 1 | 418 | B | 15 | 7 | "cool" |
The true_version column gets added only after unblinding. Until then, it's sitting in the key held by someone else.
The data dictionary
| Column | Meaning | Permitted values |
|---|---|---|
| assessor | Participant code; names kept separately | P01 to P12 |
| time_min | Minutes after strip dipping | 15, 120, 480 |
| rating | Freshness rating, anchored to a stated reference | 0 to 10, whole numbers |
| comment | The person's own words, unedited | Text, or empty |
What else belongs in the record
- Materials: names, suppliers, lots, dilution dates, SDS dates.
- Samples: formula versions, batch dates, how aged and stored.
- Setting: date, time, room temperature if measured, anything odd.
- Deviations: a helper made an error, or a participant left. Write it down.
- The plan: dated before data collection. Share with a friend by email, or use a public registry if your work warrants it.
Rules for honest records
- Don't overwrite. Keep a read-only copy of the raw data. Do your calculations elsewhere.
- Don't tidy comments. Their words are data.
- Mark any exclusion. Use a column, and give a reason that follows your stated rule.
- Separate planned from exploratory. Something you noticed afterward is a lead for the next study.
- Keep the analysis repeatable. Use formulas in the spreadsheet, or a short script, with steps listed.
- Back it up. Follow the backup guide, and test the restore.
Checking
Common mistakes include mislabeled codes, shifted rows and misread handwriting. Before you analyze:
- Count rows against what you'd expect: assessors × samples × times.
- Check for values outside the permitted range.
- Have someone recheck a random 10% against the paper originals.
- Recalculate one summary figure by another means.
Sharing
Openness makes your work checkable. But some things deserve protection:
| Share | Protect |
|---|---|
| Question, method, and results, including nulls | Participant identities |
| De-identified data, if consent allowed it | A confidential formula, stating that it's withheld |
| Materials by general type, or by name where appropriate | Client details |
| Deviations and limitations | Personal health details, always |
If the formula is withheld, a reader can't replicate your work exactly. Say so. Never withhold a result merely because it's unwelcome.
Practical exercise: Build a data file and check it
- Build the data sheet and data dictionary for your own study.
- Write your dated plan, and send a copy to someone or store it where the date can be seen.
- Do a dry run with invented or test responses, marked clearly as not real.
- Calculate one summary two different ways. Check row counts and ranges.
- Ask someone to find a figure in your file, using only the dictionary.
- Delete the test data from the final file before the real study.
You have done this well if someone could recalculate your summary from your raw data, and your plan's date precedes your first observation.
Common mistakes
- Keeping only averages. You can't recheck them.
- Merging cells or color-coding meaning. Use columns.
- Quietly dropping an odd result. State it, and follow your rule.
Key terms
- Raw data
- The original, unchanged observations, before any calculation or tidying.
- Data dictionary
- A short document explaining each column in a dataset: name, meaning, units and permitted values.
- Pre-registration
- Writing down your hypothesis, method and analysis plan, with a date, before you collect data.
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.