Good manuscript statistics are not just p-values. Reviewers often ask why a test was chosen, whether assumptions were checked, how missing data were handled, and whether numbers match across the abstract, text, tables, and figures.
What we help document
Data profiling
Variable types, distributions, missingness, event counts, outliers, and feasibility notes that may affect downstream analysis.
Method rationale
Why a test or model was selected, what assumptions were considered, and what alternatives were available.
Model outputs
Regression models, survival analyses, subgroup outputs, sensitivity checks, and other agreed outputs prepared for review.
Number reconciliation
Cross-checking sample sizes, coefficients, confidence intervals, hazard ratios, p-values, and figure labels across deliverables.
Common manuscript statistics tasks
| Task | Typical outputs | Review risk to handle |
|---|---|---|
| Table 1 | Baseline characteristics, group comparisons, missingness notes. | Wrong test choice, unlabeled units, overinterpreting p-values. |
| Survival analysis | Kaplan-Meier curves, log-rank tests, Cox model summaries. | Proportional hazards assumptions, low event counts, unclear censoring. |
| Regression modeling | Linear, logistic, Cox, or other model outputs depending on outcome and design. | Confounding, overfitting, missing covariates, reference group errors. |
| Subgroup or sensitivity analysis | Planned subgroup tables, forest plots, alternative specifications. | Multiplicity, underpowered groups, post-hoc claims. |
| Methods/results wording | Draft language describing statistical methods and results. | Claims stronger than the study design supports. |
How AI is used safely
AI can accelerate checks, summarization, code drafting, table formatting, and consistency review. It should not silently decide the final statistical model or scientific claim. In our workflow, method choices and outputs should be documented so the researcher, supervisor, or statistical reviewer can inspect them.
- AI-assisted checks help flag distribution, missingness, and inconsistency patterns.
- Analysis code and method notes make outputs easier to review.
- Researcher approval is required before manuscript use or submission.
- Unsupported clinical, causal, or comparative claims are avoided unless the study design and evidence support them.
FAQ
Can data2paper.ai choose the statistical method for me?
We can propose and document candidate methods based on study design and data structure, but the final method should be reviewed and approved by the researcher or an appropriate statistical reviewer.
Can you provide reproducible code?
When included in the written scope, deliverables can include analysis code, figure scripts, and reproducibility notes.
Can this replace peer review or statistical consultation?
No. It is technical support, not a replacement for peer review, institutional review, or specialist statistical consultation where needed.
See an example analysis package
The heart-failure demo shows Table 1, survival analysis, Cox regression, forest plots, reference checks, and number reconciliation using public example data.
View case study Request assessment