ResAIKit
Research Integrity Toolkit
Sign In

For authors and reviewers

Verify your statistical analysis before you submit

Upload your dataset and re-run the reported analyses on the real data. We check whether each result reproduces, whether the test was the right one for the design, whether its assumptions hold, and whether corrections and effect sizes were handled properly.

Drop a dataset (CSV or Excel) here or click to select

CSV, XLSX, XLS

What it checks

Reproduction

We recompute the statistic, p-value, confidence interval and effect size from your data and compare them to what was reported.

Right test choice

We check that the test fits the design and the data type, and recommend the correct one (Welch, non-parametric, logistic, Fisher, Cox, random-effects) when it does not.

Assumptions

Normality, equal variance, multicollinearity, proportional hazards, expected cell counts and more, checked on the real data.

Corrections, effect, power

Multiple-comparison corrections, effect sizes with bootstrap confidence intervals, and a guard against reading a non-significant result as no effect.

Supported analyses

Nineteen families of tests, grounded in the reporting standards SAMPL, CHAMP, STARD, PRISMA, TRIPOD, GRRAS and BARG:

  • Group comparisons (t / Welch / Mann-Whitney / Wilcoxon)
  • ANOVA and Kruskal-Wallis
  • Association (chi-square, Fisher)
  • Correlation (Pearson, Spearman)
  • Regression (linear, logistic, Poisson, negative binomial)
  • Survival (Kaplan-Meier, log-rank, Cox with proportional-hazards check)
  • Diagnostic accuracy (sensitivity, specificity, PPV, NPV, ROC and AUC)
  • Reliability and agreement (ICC, Cronbach alpha, kappa, Bland-Altman)
  • Mixed and repeated-measures models
  • Equivalence and non-inferiority (TOST)
  • Propensity-score balance
  • MANOVA
  • Prediction-model validation (discrimination and calibration)
  • Discriminant analysis (LDA, QDA)
  • Meta-analysis (fixed and random effects, heterogeneity, publication bias, subgroup, meta-regression, trim-and-fill)
  • Bayesian t-test (Bayes factor) and MCMC convergence diagnostics

Automated recomputation on the data you provide. Verdicts reflect statistical signals and the chosen mapping, not a determination of misconduct. Nothing you upload is stored.