Receiving a review letter that requests reanalysis of your data is an experience that generates anxiety in most researchers. Reviewers may ask for anything from minor adjustments (adding an effect size you forgot to report) to substantial changes that involve completely redoing the analysis with a different approach. The important thing to understand is that these requests are not necessarily a bad sign. In fact, a letter requesting revisions (major or minor) is much better than an outright rejection, and responding well to the reviewers' statistical requests can make the difference between publication and definitive rejection.
The first step, before doing anything, is to read the entire review letter calmly and without responding emotionally. Many researchers become defensive when a reviewer questions their analyses, and that defensive attitude carries over into the response, which rarely helps. Reviewers are not enemies: they are colleagues who are dedicating their time to improving your work, and most of their statistical suggestions have merit, even if they are sometimes phrased bluntly or fail to consider aspects that you know better than they do.
Types of requests and how to address them
Statistical requests from reviewers can be classified into three categories. The first includes requests that are reasonable and easy to implement: reporting missing effect sizes, adding confidence intervals, verifying assumptions that were not reported, including sensitivity analyses. These requests should be implemented without argument, because they improve the article and do not alter the conclusions. Respond briefly that you have made the requested change, indicate where it can be found in the revised manuscript, and move on.
The second category includes requests that are reasonable but substantial: changing the analytical method (for example, switching from an ANOVA to a mixed model), applying corrections for multiple comparisons that were not applied, including additional covariates, or performing complementary analyses such as bootstrapping or Bayesian analysis. These requests require more work but are usually legitimate. The ideal approach is to implement them and show that the main results hold (or honestly explain if they change). If results change substantially with the analysis proposed by the reviewer, this is important information that deserves an honest discussion in the manuscript.
The third category includes requests that you consider inappropriate or that reflect a misunderstanding on the reviewer's part. The reviewer may request an analysis that is not suitable for your type of data, demand a sample size that was impossible to obtain, or incorrectly interpret your results. In these cases, you do not have to do what the reviewer asks, but you do have to respond respectfully, with well-reasoned arguments and bibliographic references to support your position. Never ignore a reviewer's request. If you decide not to implement something, explain why in detail and with clarity.
The structure of a good response
An effective response to reviewers has a clear and consistent structure. For each reviewer comment, include three elements: the reviewer's original comment (quoted verbatim), your explanatory response, and a description of the changes made in the manuscript (with indication of the page and paragraph where they can be found). If you have added a new analysis, include the main results in the response so that the reviewer can evaluate them without having to search through the manuscript.
The tone of the response is as important as the content. Start by thanking the reviewer for their constructive comments (even if they do not seem constructive to you). Use language that demonstrates you have taken their suggestions seriously. Phrases like "we appreciate this suggestion, which has allowed us to strengthen the analysis" or "the reviewer raises an important point that we have addressed as follows" establish a collaborative tone that facilitates acceptance of the revised manuscript.
Common mistakes in responses to reviewers
The most serious mistake is not doing what you say you have done. If you claim that you have added a sensitivity analysis, make sure it is actually in the revised manuscript. Reviewers check these things, and finding discrepancies between the response letter and the revised manuscript erodes trust and practically guarantees another round of revisions or a rejection.
Another frequent mistake is performing the reanalysis requested by the reviewer and, when the results do not favor your hypothesis, simply not reporting it or minimizing it. Experienced reviewers detect this strategy. It is much better to be transparent: present the reanalysis, acknowledge that the results differ from those of the original analysis, and discuss the possible reasons. An article that demonstrates robustness under different analytical specifications is more convincing than one that only presents the most favorable version. And if results truly change depending on the analytical method, that is valuable information the scientific community should know, not something you should conceal.
Typical reviewer requests by analysis family
Over the years, the requests that come back from reviewers cluster into a small number of recurring patterns, and knowing them in advance lets you anticipate the reanalyses you may need to run. For mediation analyses, expect reviewers to request bootstrap confidence intervals (typically 5000 to 10000 resamples) instead of the Sobel test, and to ask about the assumption of no unmeasured confounding between mediator and outcome. For moderation analyses, expect simple slopes, Johnson-Neyman regions, and a request to mean-center predictors. For regression analyses, expect a check of multicollinearity (VIF), residual diagnostics, and influential cases. For ANOVA designs, expect a request to switch to a mixed-effects model when there are repeated measures, especially when sphericity is violated.
For factor analysis and SEM, expect reviewers to question your decision about extraction method (maximum likelihood vs principal axis), rotation (oblique vs orthogonal), the number of factors retained (with parallel analysis as the modern standard), and fit indices reported. For latent class or mixture models, expect requests about the number of classes (BIC, BLRT, entropy) and about the substantive interpretability of the solution. Familiarity with these patterns means that when the reviewers' letter arrives, half of the response is already drafted in your head before you open the manuscript.
Sensitivity analyses: showing your results are not fragile
Sensitivity analyses are the single most effective tool for converting a critical reviewer into a satisfied one. The logic is simple: if your conclusion holds under multiple reasonable analytical choices, your conclusion is robust. If it depends on one specific decision, the reviewer was right to be skeptical. Standard sensitivity analyses include: rerunning the model with and without outliers; comparing complete-case analysis with multiple imputation for missing data; comparing different operationalizations of the dependent variable (raw scores vs latent factor); comparing different covariate sets (minimal vs adjusted); and, for mediation, running E-values or correlated-confounder analyses to quantify how strong an unmeasured confounder would need to be to nullify the effect.
I generally recommend running these analyses preemptively, before submission, and keeping the results in a supplementary file ready to deploy if reviewers ask. This is not just defensive: it is good science. If you find that your effect disappears when you exclude three influential cases, that is something you needed to know anyway. For more guidance on this kind of preventive reanalysis, see my piece on what to do when a reviewer asks for an analysis you do not master.
Response letter language patterns that work
A few phrases recur in successful response letters and are worth keeping as templates. To acknowledge a good point: "We thank the reviewer for this thoughtful suggestion. We have now [done X], which strengthens the analysis in the following ways..." To respectfully push back: "We appreciate this comment and considered implementing [X]. However, [methodological reason with citation]. We have nevertheless added a sensitivity analysis using [alternative approach] in the supplementary materials, which shows that our main conclusion is robust to this concern." To accept a partial criticism: "The reviewer correctly notes that [limitation]. We have now acknowledged this explicitly in the Discussion (page X, paragraph Y) and tempered our conclusions accordingly."
Avoid combative phrases ("as is well known in the literature", "the reviewer seems to misunderstand", "we do not agree"). Avoid sycophantic ones ("we are deeply grateful", "the reviewer is absolutely correct"). Aim for a tone that treats the reviewer as a knowledgeable colleague offering input on a paper you both want to see published. Document every change with a clear pointer to where it appears in the revised manuscript (page, paragraph, line numbers if the journal uses them). And, when in doubt, follow the editor's guidance: in Q1 journals, the editor's framing of the reviewers' comments often signals which battles are worth fighting and which are not.
When the reanalysis a reviewer asks for uses a technique you do not master, my reviewer response service reruns the analyses and drafts the rebuttal letter with you, usually within 48-72 hours. If you have not submitted yet, anticipate the critiques with the free AI Paper Reviewer.