Numbers you can verify

What we check

Every statistical test family StatEze can run is checked against reference results computed independently of StatEze's own code. The reference values are produced outside StatEze and do not share its code, so agreement is evidence, not circularity. No test is available to users without a matching reference.

Worked examples. Recompute any of them.

Do not take our word for it. Each example below is a real analysis StatEze ran on a published clinical dataset: the chart it produced, the result it computed, and the data to reproduce it. Download the data, run it in SPSS, and check every number. Some results are strong and some are small, exactly as the data is.

These are real, widely cited datasets from the published clinical and teaching literature: the diabetes study from the Vanderbilt Biostatistics collection (Frank Harrell), the melanoma cohort from Andersen et al., the heart-failure cohort from the UCI Machine Learning Repository (Ahmad et al.), and the ACS registry from the moonBook clinical teaching dataset. They are the same starter datasets we load into every new StatEze account, not synthetic toy data. The test StatEze chose for each, and why, is shown under the chart.

Conventions, stated up front

Statistical packages sometimes present the same correct result differently, for example in how degrees of freedom are reported. Where a convention differs, the pack documents the difference and tells you which output row to compare, so an apparent mismatch is never left unexplained.

Reproducibility

Run the same analysis on the same data twice and every number comes back identical, down to the last decimal. The numbers in your manuscript can be reproduced, by you, by a coauthor, or by a reviewer.