Clinical papers ask for a small number of specific analyses: a ROC curve with its AUC, agreement between two measurements, a survival curve with a log-rank test, a kappa between two raters, and the sensitivity and specificity of a test. Magic Stat does those from a spreadsheet, with the confidence intervals the journal expects.
The module gathers the analyses that recur in clinical and diagnostic research: ROC curves and the area under them, Bland–Altman agreement, Kaplan–Meier survival with the log-rank test and Cox regression, Cohen's kappa for inter-rater agreement, and the accuracy of a diagnostic test (sensitivity, specificity, predictive values and likelihood ratios).
Each one reports what it should. The ROC gives the AUC with a DeLong confidence interval and the Youden index; Bland–Altman gives the limits of agreement with their uncertainty and tests for trend and proportional bias; the survival tab gives the product-limit curve, the log-rank test and hazard ratios.
It is not a regulatory or comprehensive survival package. If you need competing risks, time-varying covariates or repeated events, R (survival, cmprsk) or Stata is the right tool. Magic Stat is for the standard analyses a clinical paper reports, done without a script.
ROC and diagnostic: a status column (0/1) and a score or test-result column. Bland–Altman: two measurement columns. Kappa: two rater columns, unweighted or weighted.
Survival: a time column, an event column (0/1) and an optional grouping column; Cox regression takes its predictors from the dialog.
For the standard analyses — ROC/AUC, agreement, survival with log-rank and Cox, kappa and diagnostic accuracy — yes. If the paper needs competing risks, time-varying covariates or a full survival modelling strategy, use R (survival, cmprsk) or Stata instead.
The AUC uses the DeLong method for its confidence interval and reports the Youden index. The point estimate is shown as well, since some journals still report that alone.
Yes: Kaplan–Meier with the log-rank test, and Cox proportional hazards for hazard ratios. It does not cover competing risks or time-varying covariates — those need a dedicated survival package.
No. It replaces the script plumbing for the common analyses. Interpretation, study design and the decision about which test is appropriate remain your responsibility — the software reports, it does not decide.
No — patient data stays on your machine. The analysis runs locally; the only automatic signal is an anonymous installation counter with no data in it.
Logistic regression → Power analysis → Non-parametric tests →