Descriptive statistics, hypothesis testing, and interactive charts — DuckDB-native computation with automatic fallback
The statistics engine prefers DuckDB-native computation for speed. When the input dataset is in DuckDB, statistical functions run directly as DuckDB aggregate queries. For non-DuckDB backends, the engine falls back to scipy. 135 tests validate accuracy across both paths.
| Method | Description | Backend Support |
|---|---|---|
| Descriptive Statistics | Mean, median, mode, std dev, variance, min/max, quartiles, skewness, kurtosis | All backends |
| Correlation | Pearson, Spearman correlation coefficients with p-values | All backends |
| T-tests | One-sample, independent (equal/unequal variance), paired | All backends |
| ANOVA | One-way analysis of variance with F-statistic and effect size | All backends |
| Chi-square | Goodness of fit and test of independence with Cramér's V | All backends |
| Linear Regression | Simple and multiple regression with coefficients, R², p-values, residuals | All backends |
All charts are interactive Plotly visualisations. New chart types added in 0.12.0: scatter, area, horizontal bar, donut, histogram, and distribution.
| Chart | Best For |
|---|---|
| Bar Chart | Comparing categories |
| Line Chart | Trends over time or sequence |
| Scatter Plot | Relationship between two numeric variables |
| Area Chart | Magnitude over time with filled area |
| Horizontal Bar | Long category labels |
| Donut Chart | Part-to-whole with centre annotation |
| Histogram | Distribution of a single numeric variable |
| Distribution | Density plot with KDE curve |
| Pie Chart | Simple part-to-whole |
The Tools → Visualisation sandbox lets you experiment with chart types and configuration before adding them to a recipe. Choose a dataset, select a chart type, configure options, and see a live preview. When you're satisfied, save the configuration to use in a recipe.