For clinical & epidemiological researchers
StatEze maps your research question to a vetted menu of 28 + statistical tests, checks the assumptions before it runs, and returns APA 7 narrative you can paste into a manuscript. The language model never touches the math.
The problem
Proprietary suites are expensive and slow to teach. R demands months of syntax fluency before a new RA can produce a clean ANCOVA. Generic chatbots will cheerfully write code that looks right and computes the wrong thing — parametric tests on skewed data, ignored repeated measures, Bonferroni forgotten. And custom scripts rarely carry their own assumption checks.
StatEze is built to collapse that gap. Researchers describe what they want to know. Assumption-checked calculations run against the same scipy and statsmodels libraries peer reviewers already trust. The output arrives publication-ready.
How it works
Drop in a CSV, Excel workbook, or SPSS .sav file. StatEze profiles variable types, missingness, and distribution shape before you ask anything. Full datasets never leave your machine in Mode 1.
Type “compare 24-hour pain between the two protocols” or select from templated designs (independent groups, paired, ANCOVA with covariate, repeated measures, correlation…). StatEze maps the intent to a specific test.
Get the test statistic, effect size, confidence interval, a publication-ready chart, and an APA 7 narrative paragraph. Export to Word in one click.
Capabilities
Every test is backed by scipy or statsmodels. The same dataset produces the same numbers, every time.
Shapiro–Wilk per group, Levene’s homogeneity, Mauchly’s sphericity, sample-size gates. Violations trigger automatic non-parametric substitution.
Results table, effect size with interpretation, 95 % CI, and a paragraph you can paste into a manuscript. Exportable to Word.
Mode 1 offline (rule-based, air-gapped). Mode 2 hybrid (Claude validates local picks). Mode 3 Claude-native. You choose how much the model is involved.
Full datasets never leave your machine. In hybrid and cloud modes, only a 5-row preview and column metadata are sent. Every outbound payload is logged.
t-tests, ANOVA (with sphericity), Mann–Whitney, Kruskal–Wallis, Wilcoxon, Pearson, Spearman, linear & logistic regression, chi-square, Fisher’s exact, ANCOVA, Cronbach’s α, Sobel mediation, linear mixed effects, post-hoc power.
Methodology
A general-purpose chatbot that writes statistical code can be confidently wrong in ways that are invisible to a reviewer. StatEze is built differently.
Claude Sonnet 4.6 acts only as a translator: it maps your natural-language question to a specific function in a tested dispatcher. The dispatcher validates assumptions and routes to a pre-registered statistical function. The calculator — always scipy or statsmodels — computes the numbers. The LLM never emits arithmetic; it only selects a name from a finite menu.
Translator → Dispatcher → Calculator. A finite, auditable path from intent to result.
Who uses it
The research assistant
“I’m running descriptives and inferentials for a retrospective cohort — 340 charts, 22 variables. StatEze gave me the t-tests with effect sizes, flagged the one variable that needed Mann–Whitney, and pasted the paragraph into my draft. Two hours instead of two days.”
Retrospective cohort, 340 charts, 22 variables.
The methodologist
“I teach a research-methods class and I need to show 14 residents how ANCOVA changes the picture when you actually adjust for baseline. StatEze gives me a reproducible walkthrough with assumption diagnostics I can project on the wall.”
Teaching ANCOVA with live diagnostics.
The PhD candidate
“Before I wrote my discussion, I wanted to re-verify every effect size and make sure the interpretation language matched APA conventions. StatEze gave me Cohen’s d, η², and the interpretation string per comparison. My committee flagged nothing.”
Dissertation, checking effect sizes pre-defense.
Privacy & trust
Fully offline, rule-based, air-gapped. No network calls. Suitable for protected health information and IRB-constrained environments.
Only a 5-row preview and column metadata are sent to Claude. The full dataset never leaves your machine.
Every outbound payload is logged locally. You can inspect exactly what was sent, when, and to which endpoint.
A pattern library soft-warns on ingest when columns resemble identifiers (MRN, DOB, name-shaped strings). Soft warnings, not silent filtering — you stay in control.
Frequently asked
For the analyses StatEze covers — which includes the bulk of what appears in a typical clinical or epidemiological manuscript — yes. For specialized workflows (survival analysis with complex censoring, Bayesian hierarchical models, bespoke simulation), R remains the right tool. StatEze is designed to remove SPSS-shaped friction, not to replace statistical programming entirely.
Your full dataset never leaves your machine. In Mode 1, nothing is sent anywhere. In Modes 2 and 3, only a 5-row preview and column metadata (names, types, basic distribution summary) are sent to Claude so the translator can route correctly. Every outbound call is written to a local audit log.
CSV, Excel (.xls, .xlsx), and SPSS (.sav) at launch. Parquet and long-form clinical exports are on the near-term roadmap.
Yes. APA 7 narrative, the results table, and publication-ready charts export to Word with one click. You can also copy any element individually.
Mode 1 is a first-class path, not a fallback. The rule-based dispatcher can select tests from templated designs without any network connectivity at all. If you lose connection mid-session in Mode 3, StatEze offers to continue in Mode 1 against the same dataset.
At launch: independent-samples and paired t-tests, one- and two-way ANOVA (with Mauchly’s sphericity for repeated measures), ANCOVA, Mann–Whitney, Kruskal–Wallis, Wilcoxon signed-rank, Pearson and Spearman correlation, linear regression, logistic regression, chi-square of independence, Fisher’s exact, Cohen’s d, η², Cramér’s V, Cronbach’s α, post-hoc power, Sobel mediation, linear mixed effects, and more — 28 + in total.
The closed beta is free for invited researchers. Commercial pricing will be announced before general availability. If you’re a methodologist at an academic medical center, request access below — the beta is actively taking friendlies.
Early access
Early access is open to clinical and epidemiological methodologists, trial statisticians, and research coordinators at academic medical centers. Drop your email and we’ll reach out as seats open.