cogpsych-rebuttal

Revision Rebuttal (cogpsych-rebuttal)

A Cognitive Psychology revision typically asks for more modeling rigor — an additional experiment, a further model comparison, parameter/model recovery, alternative priors, or reproducible code — because the contribution is a model-driven theoretical claim. The response letter must convert every reviewer and reassure the editor that the model adjudication is now airtight, while keeping the integrative argument coherent.

When to trigger

• A major/minor revision arrived and you are planning the revision + response letter
• Reviewers requested added experiments, model comparisons, recovery, or open-code changes
• A requested analysis or model would change the conclusion
• Writing the cover note to the handling editor

Strategy

1. Read the editor's letter as the rubric. Solve the decisive points first; the editor adjudicates among reviewers and decides the next round.
2. Point-by-point, every comment. Quote each comment, then respond; never skip one.
3. Strengthen the model inference, don't just defend. Many requests (fit a further rival, add recovery, cross-validate, refit hierarchically, share code) make the adjudication stronger — do them and say where. A request that exposes overfitting must be addressed, not waved away.
4. Keep the program coherent. A new experiment or model should slot into the argument; update the General Discussion so the synthesis still holds (see cogpsych-writing-style).
5. Concede or rebut with evidence. Did what was asked (cite the location), or push back respectfully with a reason (e.g., why a requested model is not identifiable) — don't add an analysis that quietly undercuts the claim without saying so.
6. Keep the modeling reproducible. New analyses must be reflected in the deposited model/analysis code and regenerate in a fresh session (see cogpsych-open-science-and-transparency).

Response-letter format

For each reviewer comment:

> [Quoted reviewer comment]

Response: [What we did / why we respectfully disagree].
Change: [Manuscript section, supplement/appendix section, table/figure, or
         deposited-code file].

Open with a short summary of the main changes to the editor; group by reviewer; end each entry with the location (note when added analyses or experiments went to the supplement/appendix).

Worked micro-example (illustrative response entries)

For the recognition-memory program, a major revision asked for a further model and recovery.

> R2: You compare UVSD and DPSD, but a mixture model might fit better -
> have you ruled it out?

Response: We agree this rival should be tested. We added a finite-mixture
SDT model, fit under matched flexibility; it does not improve penalized fit
(dBIC = 9 favoring UVSD) and model recovery confirms the comparison is
diagnostic at our design's N/trials.
Change: Results (model comparison, Table 1 expanded); recovery → Appendix B;
         fitting code updated (deposit, fit_mixture.R).

> R1: Can you recover the DPSD parameters at your trial counts?

Response: Yes - we now report parameter recovery for all three models
(recovered values within credible intervals). This is why the model
comparison is interpretable rather than an artifact of identifiability.
Change: Appendix B (recovery); deposited code recovery_sim.R; one sentence
         in Results pointing to it.

Revision triage — where each request lands

Recurring revision pushback and the venue fix

• "You only ruled out one rival" → fit the additional model(s) under matched flexibility; report the penalized comparison and recovery; never argue from a single fit.
• "Your better fit might be overfitting" → add cross-validation/penalized criteria and model recovery; if the edge does not survive, adjust the claim.
• "I couldn't reproduce your fits" → ship seeded code + a pinned environment + a fresh-session run log; reference it in the response.
• "The new analysis weakens the effect" → disclose it, interpret it, and scale the theoretical claim; concealment is the cardinal sin.

Anti-patterns

• Ignoring or merging away a comment without a visible response
• Defending a single fit instead of adding the requested comparison/recovery
• Adding an experiment or model that breaks the program's coherence without re-synthesizing
• Adding analyses that contradict the original claim without acknowledgment
• Letting deposited model code/data drift out of sync with the revision

Output format

【Editor's decisive points】addressed first? [list]
【Coverage】every reviewer comment answered? [Y/N]
【Model inference strengthened】added comparison/recovery/hierarchy? [Y/N]
【Program coherent】new experiment/model integrated into the synthesis? [Y/N]
【Reproducible】deposited code updated + fits regenerate? [Y/N]
【Next】resubmit via Editorial Manager

Supplementary resources

• ../../resources/official-source-map.md — review model, revision norms, reproducibility expectations