Feature tour
Stratum covers the full arc of data work — import, wrangle, explore, test, model, and present — without ever leaving a single, fast, native window.
Get data in, shape it the way you need, and keep it reproducible.
CSV/TSV with automatic delimiter detection and ≥90% type inference, fixed-width ASCII, native Excel (.xlsx), and SPSS .sav/.zsav and Stata .dta with their variable labels, value labels and missing-value codes.
114 functions — math, strings, dates, binning, recoding, lag/lead, per-column aggregates — in re-editable, live-recomputed columns.
A persistent, dockable panel with All / Any / None compound logic, multi-key sort, and a master enable switch.
Join, stack (wide↔tall), split, aggregate (group-by), deduplicate and sample — all non-destructive.
Z-score, min–max, ranks and percentile ranks; map and lump categorical levels in a click.
Save data, analyses, filters and styling together in one .stratum file.
A chart type for every question — distributions, relationships, comparisons and composition — each with live controls and clean export.
Histograms, box & violin plots, density & ridgeline, ECDF, dot/strip, and Q-Q plots.
Scatter with OLS/LOESS fits, bubble charts, trend lines, rectangular/hex heatmaps, and smooth contour plots.
Dot plots, dumbbell (DNA) charts, and slopegraphs — compare values across items, groups and stages.
Pie/donut, stacked & grouped bars, Pareto, mosaic, and treemaps for parts of a whole.
Parallel-coordinates plots for many-variable rows, plus PCA scree and biplot views.
Zoom and pan on every continuous axis — binned and fitted charts re-compute for the window rather than just scaling the picture.
One row selection shared by the datasheet and every chart. Click, ⌘/Ctrl-click, Shift-click or drag a rectangle on any chart whose marks stand for rows: point charts dim what you did not pick, and aggregate marks — bars, bins, tiles, wedges, cross-tab cells, box and violin bodies, even tree nodes — shade to show the selected share.
The tests you actually use, with parametric and nonparametric options.
Nearly 50 statistics per column — moments, quantiles, robust measures, confidence bounds; Pearson, Spearman and Kendall matrices with CIs, p-values, and Bonferroni or FDR correction.
Two-sample t / Welch / paired, Mann-Whitney & Wilcoxon, Kruskal-Wallis with Dunn; one- and two-way ANOVA with Type I/II/III sums of squares, η²/ω²/ε², Tukey-Kramer and Games-Howell; MANOVA with Wilks’, Pillai, Hotelling-Lawley and Roy.
Chi-square contingency & differences analysis; linear regression with categorical, interaction and polynomial terms, VIF, leverage, Cook's D, Durbin-Watson, PRESS and predicted R², confidence and prediction intervals, and best-subsets selection by Mallows’ Cp, AIC, BIC or adjusted R².
Normal, lognormal, Weibull and exponential fits with goodness-of-fit tests; four normality tests (Shapiro-Wilk, Anderson-Darling, Kolmogorov-Smirnov, chi-square) and Levene’s test for equal variance.
Quadratic, cubic, power, exponential and logarithmic fits, plus LOWESS, kernel smoothers and moving averages — drawn over the data with R² on the legend.
Pearson chi-square with expected counts, the likelihood-ratio G², Yates’ correction and Fisher’s exact test; Cramér’s V, phi, the contingency coefficient and Goodman-Kruskal gamma; odds ratio and relative risk with confidence intervals for 2×2.
From the factory floor to predictive models — built in.
I-MR, X-bar R/S, p/np, c/u, CUSUM and EWMA, with all eight Nelson rules and out-of-control highlighting, limits estimated or given from historical values, and Phase I/II staging.
Cp, Cpk, CPL, CPU, Cpm, Pp, Ppk, PPL, PPU, observed and expected PPM and Z-bench, each with a confidence interval from the exact noncentral t — plus non-normal capability from fitted percentiles to ISO 22514-2.
Classification and regression trees with Gini/entropy/variance splits and cross-validated cost-complexity pruning; random forests with out-of-bag error and confusion; boosted trees with early stopping — all with variable importance and prediction columns.
Partial-dependence plots per predictor, confusion matrices, misclassification rates and test AUC — so an ensemble can be explained, not just scored.
Principal components with scree & biplot, KMO, per-variable MSA and Bartlett’s test of sphericity; k-means with silhouette and elbow plots; hierarchical clustering with Ward, complete, average or single linkage and dendrograms.
Seeded training / validation / test splits, optionally stratified, scored on rows the model never saw — and competing models compared on the same partition.
Designed to stay fast and legible, and to produce work you can hand off.
A virtualized grid and lazy filtering keep large datasets responsive.
Independent text-size controls for the interface, the data grid and reports.
Charts to PNG/JPEG/TIFF/PDF; reports to CSV/TSV/Markdown/HTML/PDF — all print-ready.
Interface and help in English, German, Spanish, French, Italian, Japanese, Korean, Dutch, Portuguese (Brazil) and Simplified Chinese.
Change a filter, edit a value or rename a column and every table, chart and fitted model that depends on it recomputes — nothing to re-run by hand, and nothing left stale on screen.
Every analysis runs on your machine. Your data is never uploaded and there is no telemetry on your datasets.
Stratum