Methods & support boundaries
We would rather tell you "not yet" than give you a wrong answer. Below is exactly what the engine can do today, how we check it, and what is on the way.
All statistics are computed by a deterministic engine: the same data always gives the same answer. AI is never used to produce numbers.
"Validated" means the method's test statistic, degrees of freedom and p-value are checked by automated tests against reference values computed with SciPy, a widely used scientific library, to at least 6 decimal places.
Validation shows the arithmetic is right. It cannot show that a method suits your study design or that your data are accurate. That is why every result lists its assumptions, checks and limitations. We do not promise perfect accuracy.
Missing values are handled by complete-case analysis: rows missing a needed value are excluded and the number excluded is reported. All tests are two-sided at α = 0.05 with 95% confidence intervals.
Available now
A summary of one measurement.
Chosen when: One numeric variable.
Counts and percentages for categories.
Chosen when: One categorical variable.
Compares the average of two separate groups.
Chosen when: Numeric outcome, two independent groups.
Compares two separate groups using ranks (robust to skew and outliers).
Chosen when: Numeric/ordinal outcome, two independent groups, non-normal data.
Compares two measurements taken on the same people.
Chosen when: Two numeric measurements on the same units.
Paired comparison using ranks.
Chosen when: Paired measurements, skewed differences.
Compares averages across three or more groups.
Chosen when: Numeric outcome, 3+ independent groups, similar spread.
Compares averages across groups when their spreads differ.
Chosen when: Numeric outcome, 3+ independent groups, unequal variances.
Compares three or more groups using ranks.
Chosen when: Numeric/ordinal outcome, 3+ independent groups, skewed data.
Checks whether two categorical variables are related.
Chosen when: Two categorical variables.
Exact version of the chi-square test for small 2×2 tables.
Chosen when: Two binary variables, small expected counts.
Measures how strongly two measurements move together in a straight line.
Chosen when: Two numeric variables, roughly linear, no strong outliers.
Measures how consistently two measurements rise or fall together.
Chosen when: Two numeric/ordinal variables, skewed or with outliers.
Estimates how much the outcome changes per unit of a predictor.
Chosen when: Numeric outcome, one numeric predictor.
Estimates each predictor's effect while holding the others constant.
Chosen when: Numeric outcome, two or more predictors.
Estimates how predictors change the odds of a yes/no outcome.
Chosen when: Yes/no outcome, one or more predictors.
Compares three or more measurements taken on the same people.
Chosen when: 3+ numeric measurements on the same units.
Compares how long until an event between two groups.
Chosen when: Time-to-event data with a two-group comparison.
Estimates how predictors change the risk of an event over time.
Chosen when: Time-to-event outcome with one or more predictors.
Checks whether questionnaire items measure one thing consistently.
Chosen when: 3+ numeric questionnaire items.
Measures how consistently raters score the same subjects.
Chosen when: 2+ raters scoring the same subjects numerically.
Measures agreement between two raters beyond chance.
Chosen when: Two raters assigning categories to the same cases.
Shows how well a test separates people with and without a condition.
Chosen when: Numeric test result vs a binary reference standard.
Sensitivity, specificity and predictive values of a yes/no test.
Chosen when: Binary test result vs a binary reference standard.
Measures agreement among three or more raters beyond chance.
Chosen when: 3+ raters assigning categories to the same cases.
Predicts an outcome with several unordered categories.
Chosen when: Categorical outcome with 3+ unordered categories, one or more predictors.
Predicts an outcome with ordered categories like mild/moderate/severe.
Chosen when: Ordered categorical outcome, one or more predictors.
Compares averages across two grouping factors and checks whether they interact.
Chosen when: Numeric outcome, two categorical factors.
Finds the main dimensions underlying many related variables.
Chosen when: Several numeric variables measured on the same cases.
Tracks how a measurement changes over time while allowing each person their own starting level.
Chosen when: Long-format repeated measurements (subject, time, value columns).
Tracks how a measurement changes over time, allowing each person their own starting level and their own trend.
Chosen when: Long-format repeated measurements where each subject may change at their own rate.
Finds latent factors underlying many related variables, rotated for interpretability.
Chosen when: Several numeric variables thought to reflect a few underlying constructs.
Estimates the average effect on a yes/no outcome when measurements are clustered within subjects.
Chosen when: Clustered/repeated yes/no outcomes (population-average interpretation).
Models how predictors change the rate of a count outcome.
Chosen when: Count outcome (number of visits, events, cases), one or more predictors.
Like Poisson regression but allows counts that vary more than a Poisson model expects.
Chosen when: Over-dispersed count outcomes.
Estimates the average trend of a measurement when observations are clustered within subjects.
Chosen when: Clustered/repeated numeric outcomes (population-average interpretation).
Estimates the average effect on a count when observations are clustered within subjects.
Chosen when: Clustered/repeated count outcomes (population-average interpretation).
Finds latent factors underlying many related variables, allowing the factors to correlate.
Chosen when: Several numeric variables thought to reflect a few correlated underlying constructs.
Models counts where many zeros come from a separate 'never happens' process.
Chosen when: Count outcomes with far more zeros than a Poisson model expects.
Like zero-inflated Poisson but also allows the non-zero counts to be over-dispersed.
Chosen when: Over-dispersed count outcomes with excess zeros.
Tracks a measurement over time following a curved (accelerating or decelerating) trend.
Chosen when: Long-format repeated measurements with a curved common trend.
Tracks a measurement over time following a flexible smooth curve.
Chosen when: Long-format repeated measurements with an irregular trend over time.
Support boundaries
Every method listed above is available and validated. Designs outside this list — for example Bayesian models, multilevel models with three or more levels, or time-series forecasting — are not supported; when your question needs one, the app says so plainly instead of guessing.