Guides · Writing up

When a reviewer asks for your source data

The email arrives about a figure from three years ago, and the student who made it has left. What to do in the next fortnight, and what to put in place now so it is a half-day rather than a fortnight.

Reading time
8 min
Licence
Free to copy into your own lab documentation. No signup.

It is a reasonable request, politely worded, and it is the single most stressful email a corresponding author receives: could you provide the underlying data for Figure 3, and clarify how the normalisation was done?

The stress is rarely about the science. The figure is fine. The problem is that the figure was made in a slide deck by someone who has graduated, from a spreadsheet that had already been cleaned, from instrument files in a folder named by date — and the chain from one to the next existed only in that person's head.

Who this is for

A corresponding author who has just received the email, or a PI who read that first paragraph with a sinking feeling. The second half is what to put in place so this is routine. Nothing here needs particular software.

First, answer the email

Before searching for anything, reply. A short acknowledgement with a realistic date buys you weeks and costs nothing. Silence is read as evasion, and evasion is what turns a data request into a formal query.

Say what you are doing and when you will have it. If you do not yet know whether you can produce everything asked for, say that too — at this stage nobody expects a complete answer, only a responsive one.

The thing that damages a paper is not a slow answer. It is an answer that quietly does not match the figure.

Work backwards, not forwards

The instinct is to open the data folder and start looking. That is the slow route. Start from the figure and walk back one link at a time.

  1. The figure as published. Get the exact version that went to the journal — not a similar one from a talk. Check the submitted supplementary too.
  2. The file that drew it. The chart, the script, the slide. Its modification date is the strongest clue you have.
  3. The numbers behind the chart. Usually a spreadsheet. Note whether it holds raw values or values already processed.
  4. The processing between raw and those numbers. The step most often missing, and most often what the reviewer is actually asking about.
  5. The instrument files. Match by date and sample count before believing you have found them.

Write down each link as you confirm it. That note is the reply, and it is also the thing you wish had existed.

The fortnight checklist

Responding to a data request 0 / 8
✓Reply within three working days with a dateEven if the date is three weeks out. Responsiveness is most of the impression.
✓Re-read exactly what was askedUnderlying values, the analysis, or the raw instrument files? They are three different requests.
✓Contact the person who did the work earlyThey will answer in five minutes what would take you five days. Their goodwill decays with time.
✓Reproduce the figure from the raw data yourselfBefore sending anything. This is where discrepancies surface, and you want to find them first.
✓Write down every manual step you had to repeat"Removed the first two rows by hand" is part of the method and belongs in the reply.
✓Send raw plus processed, with a note explaining the differenceRaw alone invites a second round of questions.
✓Say plainly if something cannot be foundAn honest gap is recoverable. A reconstructed file presented as original is not.
✓Keep the whole exchange with the paper's folderThe next question, if there is one, will arrive after you have forgotten this.

If a number does not match

Sometimes the reproduction comes out slightly different. This is common and usually innocent — a rounding step, an outlier removed by hand, a version of the analysis that moved on after the figure was made.

The right response is to find the cause before replying, and then state it. “The published figure excluded one replicate flagged during acquisition; the raw file includes it, which accounts for the difference” is a completely acceptable sentence. What is not acceptable, and what does real damage, is sending data that silently disagrees with the paper and waiting to see whether anyone notices.

If the difference is material rather than cosmetic, talk to your co-authors and the editor early. Corrections issued promptly by the authors are treated very differently from discrepancies found by someone else.

Why we care about this

This email is the reason woodle.cloud exists. It keeps a stated result attached to the run and the original file that produced it, so “which file made Figure 3” is a property of the record rather than a memory. Everything above works without it — it is just slower.

See how it works →

The half-hour that prevents all of this

Do this once, at submission, while everything is still open in front of you. It is the highest-value thirty minutes in the whole publication process.

Make one table — a spreadsheet is perfect — with a row for every figure and panel in the paper:

  • Figure and panel number
  • What experiment it came from, and the date
  • Who ran it
  • The file path of the raw data
  • The file that did the analysis, and the version of the protocol followed
  • Any manual step, in one line

Save it beside the manuscript. Everyone will tell you they remember all of it. In three years none of you will, and the person who ran it will be somewhere else.

What good looks like

You can answer the email in a single afternoon, from the table, without contacting anyone who has left — and the data you send reproduces the figure exactly.

That is the whole standard, and it is entirely achievable with a spreadsheet made on the day of submission.

Take it with you

The figure table and the checklist are free to copy into your own practice. No attribution needed and no signup. If you take one thing, make the table at submission — it costs half an hour once and saves a fortnight later.