How to Respond to Reviewer Comments About Your Data (Without Redoing It All)
Most reviewer comments about data ask for one of four things. A different test, a sensitivity check, an explanation of excluded cases, or a clearer figure. None of the four requires rerunning your study. All four require going back to the dataset with a specific question.
The hard part is not statistical. It is that each request has to become a concrete analysis, a concrete number, and a sentence a reader can check.
This guide covers why this stage stalls, the three approaches people use, and where each one runs out. It is a data workflow, not publication advice, so follow your target journal's own instructions.
Why reviewer comments about data are hard to answer
The first problem is translation. A reviewer writes one sentence, and it expands into several separate analyses.
"Please confirm the result is not driven by outliers" is a single line. Answering it means defining an outlier rule, rerunning the model with and without those cases, and reporting both.
The second problem is that the original analysis is rarely reproducible on demand. It exists as a workbook with manual filters, a chart built from a range that no longer matches, and steps nobody wrote down. Reopening it months later is archaeology.
The third problem is bookkeeping. PLOS ONE asks for three separate files at revision: a response document, a marked-up copy using Track Changes, and a clean copy. Every number has to agree across all three.
A fourth problem is scope creep in your own answers. Reviewer comments often invite a broader analysis than was asked for, and a revision that grows new sections invites new objections.
There is a quiet trap in that process. PLOS states that figures and supporting information from the original submission transfer automatically. They carry over "unless you choose to update and replace them." Regenerate a figure, forget to replace it, and the old one ships.
What this costs you
A second review round you could have avoided. Partial answers invite the same question again. A reviewer who asked about outliers and received a paragraph without numbers will ask twice.
Weeks of calendar time on hours of analysis. The delay is rarely the computation. It is reconstructing what the original workbook did before you can change anything.
Inconsistency between the response letter and the manuscript. The response says the coefficient became 0.31, the table still says 0.34, and one of the two was updated. This is the most common reason a revision reads as careless.
A revision is judged on whether an editor can follow your changes without recomputing them.
The workarounds people try
Option 1: Turn each comment into a numbered analysis list
Before touching the data, split every comment into the specific outputs it requires. One comment often produces three rows.
Number them to match the reviewer's own numbering, because that is what the response document has to mirror. PLOS asks authors to "address the specific points made by each reviewer" and to list the changes made.
This is the highest-value habit available, and it costs an hour. It does not make the analyses themselves any faster.
Option 2: Rebuild the analysis so it can be rerun
Rather than editing the original workbook, rebuild the calculation with the exclusion rule in a cell rather than in a filter. Then changing the rule changes the result.
For the common requests, the native functions are enough. T.TEST returns a p value directly, with the type argument selecting paired, equal-variance or unequal-variance. QUARTILE gives you the bounds for an interquartile outlier rule, and CORREL covers a requested correlation.
Report the revised figures to your style guide rather than to spreadsheet defaults. The APA statistics guide sets two decimals for t, F and correlations, and writes any value below .001 as p < .001. Our guide to reporting statistics in APA format covers the rest.
The ceiling is effort. Rebuilding takes longer than the original build, and it happens under a deadline.
Option 3: Keep a change log beside the response letter
Maintain one table with four columns: the comment, what you ran, the result, and the manuscript location you edited. Fill it as you work rather than afterwards.
This is what catches the response-versus-table mismatch. It is also what you consult when a second round asks about the first round.
The limitation is discipline rather than capability. The log is abandoned exactly when the deadline tightens and it is needed most.
The shared ceiling. All three assume the dataset is still in a state you can analyse. If the file has been edited in place since submission, the first task is reconstructing what you originally had.
How to respond to reviewer comments with Powerdrill Bloom
Step 1: Upload the dataset behind the paper
Upload the file the submitted analysis used. Powerdrill Bloom profiles the columns on arrival, so exclusions, blank blocks and type mismatches are visible before you rerun anything.
Step 2: Ask each requested analysis in natural language
Take the comments one at a time. Ask for the test the reviewer named, then ask for the same test with the flagged cases removed, so you can report both.
Ask the follow-up questions in the same pass. Ask which cases fall outside the interquartile bounds, and whether the conclusion changes when they are dropped. Then ask which subgroups have samples too small to support a claim.
Step 3: Export the chart, report, or deck
Take out a replacement figure, a table of before-and-after values, or a written summary you can adapt into the response document.
Why this beats rebuilding the analysis by hand
| Manual route | Powerdrill Bloom | |
|---|---|---|
| Rerunning a test without flagged cases | Rebuild the range and formula | Ask for it both ways |
| Sensitivity check across several rules | One workbook copy per rule | Ask for each rule in turn |
| Regenerating a figure to match | Rebuild the chart manually | Ask for the updated chart |
| Second review round | Reconstruct the work again | Reopen the same session |
The last row matters most. Revisions arrive in rounds, and the second round is where a manual rebuild has already been forgotten.
Reviewer comments in a second round also tend to reference the first. Being able to reopen the exact analysis is worth more than doing it faster.
Common mistakes
Answering in prose without numbers. A reviewer asking about outliers wants the coefficient with and without them. A paragraph of reassurance reads as avoidance.
Editing the original file in place. You lose the ability to show what changed. Work on a copy and keep the submitted version intact.
Letting the response letter drift from the manuscript. Update both in the same sitting, then check every number in the letter against the table it came from.
Forgetting to replace transferred figures. Old figures carry over automatically at some journals. A revision containing the previous chart undoes the work.
Silently dropping cases the reviewer asked about. Removing awkward rows without saying so is worse than the original problem. State the rule and report both results.
Rounding differently in the letter and the table. Pick the precision your style guide requires and apply it everywhere. Our note on statistical significance covers what those values do and do not claim.
Treating a request for clarity as a request for more. Sometimes the figure is simply unreadable. Our guide to making charts for a research paper covers the formatting side.
Conclusion
Split each comment into named analyses, rerun them so both versions are visible, and keep one log tying every answer to a manuscript location. That structure survives a second round.
The expensive part is not the statistics. It is reconstructing an analysis you can no longer rerun, under a deadline set by someone else.
If that is where your revisions stall, try Powerdrill Bloom on the dataset behind the paper. See also our guide to checking your data before submitting and the academic papers page.
Frequently asked questions
How should I structure a response to reviewer comments about data?
Mirror the reviewer's own numbering, and give each point the analysis you ran, the result, and where the manuscript changed. PLOS ONE asks authors to address the specific points made by each reviewer and to list the changes made.
Do I need to rerun my whole analysis when a reviewer questions one result?
Usually not. Most requests target one test, one exclusion rule or one figure, so the work is a targeted rerun rather than a fresh study.
What if I disagree with a reviewer about the statistics?
Answer with the analysis rather than the opinion. Run what they asked, report the result, and then explain why your original choice stands if it does.
How do I report a revised statistic correctly?
Follow your target style guide rather than the spreadsheet default. APA style sets two decimals for values such as t, F and correlations, and writes anything below .001 as p < .001.
What files does a revision usually require?
At PLOS ONE it is three: a response to reviewers, a marked-up manuscript showing changes via Track Changes, and a clean copy. Check your own journal, since requirements differ.