Skip to main content

Analyze

All tests, models, and statistical analyses.

Descriptives & basic tests

Descriptives

What it is. Mean, median, SD, min, max, quartiles, skewness, kurtosis.

When to use. First thing to run on any numeric variable.

Frequencies

What it is. Frequency table: counts, %, and cumulative % per value.

When to use. For categorical or discrete variables.

T-Test

What it is. Compare means. One-sample (vs value), Independent (two groups), Paired (repeated measures).

Options / Parameters

  • Levene's test — Checks equality of variances (important for independent-samples).
  • Cohen's d — Effect size — small 0.2, medium 0.5, large 0.8.

ANOVA

What it is. Compare means across 3+ groups.

Options / Parameters

  • One-Way — Single categorical factor.
  • Two-Way — Two crossed factors — tests main effects + interaction.
  • Repeated Measures — Same subjects measured under several conditions.
  • Post-hoc Tukey HSD — Pairwise comparisons if the ANOVA is significant.
  • Levene — Homogeneity of variance (assumption check).
  • eta² — Proportion of variance explained by the factor.

Chi-Square

What it is. Tests association between two categoricals, or goodness-of-fit to an expected distribution.

Options / Parameters

  • Independence — Two categorical vars — produces crosstab + Cramer's V.
  • Goodness of Fit — One variable — tests if observed freqs match expected (uniform or custom).

Correlation

What it is. Linear (Pearson), monotonic (Spearman), or rank-based (Kendall) association.

Options / Parameters

  • Pearson — Linear. Requires normality and interval scale.
  • Spearman — Robust to outliers; measures monotonic relationships.
  • Kendall — Better for small samples, more conservative.
  • Partial — Correlation controlling for a third variable.

Regression

Linear Regression and Variants

What it is. Model a numeric outcome as a function of predictors.

Options / Parameters

  • OLS — Standard — minimizes sum of squared residuals.
  • Ridge (L2) — Regularized — shrinks coefficients; helps with multicollinearity.
  • LASSO (L1) — Can drive coefficients to zero — automatic feature selection.
  • Elastic Net — L1+L2; alpha controls strength, l1_ratio mixes the penalties.
  • Robust (Huber) — Resistant to outliers.
  • Quantile — Estimates quantiles (e.g., median) instead of the mean.
  • Stepwise — Adds/removes predictors automatically by AIC/BIC.
  • WLS — Weighted LS — each observation has a weight (pick a weight variable).
  • FGLS — Feasible Generalized LS — corrects heteroscedasticity/serial correlation.
  • IV-2SLS — Instrumental variables — endogenous predictors replaced via instruments.
  • ARDL — Autoregressive Distributed Lag — short- and long-run effects.
  • Truncated — Y bounded (e.g., Y≥0).
  • NLS — Nonlinear LS — provide a formula and starting values (JSON).
  • Robust SE (HC3) — Heteroscedasticity-consistent standard errors.
  • Standardized β — Coefficients in SD units — compare relative importance.
tip

For a binary outcome, use Logistic (dedicated menu entry), not OLS.

Logistic Regression

What it is. Binary (0/1) outcome modeled as a probability.

Options / Parameters

  • Odds ratios — exp(β) — odds-ratio interpretation.
  • McFadden R² — Pseudo-R² equivalent.
  • Classification accuracy — How many cases are correctly classified.

Count Models

What it is. For Y that is a count (non-negative integer).

Options / Parameters

  • Poisson — Assumes E(Y)=Var(Y).
  • Negative Binomial — For data with overdispersion (Var > E).
  • ZIP — Zero-inflated Poisson — mixes structural zeros with counts.
  • ZINB — Like ZIP but with overdispersion.
  • Hurdle — Binary (zero/non-zero) + truncated count model.

Non-Parametric

Mann-Whitney U

What it is. Alternative to independent-samples t-test when normality fails.

Wilcoxon signed-rank

What it is. Alternative to paired t-test.

Kruskal-Wallis

What it is. Alternative to one-way ANOVA.

Non-Parametric Extended

What it is. Bundle of additional distribution-free tests.

Options / Parameters

  • KS 2-sample — Do two samples come from the same distribution?
  • Sign test — Simple alternative to paired t-test.
  • Runs test — Tests randomness of a sequence.
  • McNemar — Paired nominal (before/after).
  • Cochran Q — Multiple related samples with binary outcomes.
  • Kendall's W — Agreement between raters/rankings.
  • Jonckheere-Terpstra — Ordered alternative to Kruskal-Wallis.
  • Mood's median — Tests equality of medians across k groups.
  • Permutation — Exact resampling-based tests.
  • Panel Unit Root — Stationarity for panel data.
  • KDE — Kernel density estimation.
  • LOESS — Local regression / smoothing.

Dimension reduction & reliability

PCA / EFA / Canonical Correlation

What it is. Dimension reduction and latent structure.

Options / Parameters

  • PCA — Orthogonal components maximizing variance. Use Kaiser (eigenvalue>1) to pick count.
  • EFA — Factor analysis with rotation (varimax/promax/oblimin). Includes KMO + Bartlett.
  • Canonical Corr. — Correlations between TWO sets of variables.

Clustering

What it is. Group similar observations.

Options / Parameters

  • K-Means — K-cluster partition by centroids. Pick K via silhouette score.
  • Hierarchical — Agglomerative — produces a dendrogram. Linkage: ward / complete / average / single.

Reliability

What it is. Internal consistency of a scale.

Options / Parameters

  • Cronbach α — α ≥ 0.9 excellent, ≥ 0.8 good, ≥ 0.7 acceptable.
  • Split-Half — Splits the scale in halves, correlates, applies Spearman-Brown correction.

Time Series

Unit Root Tests

What it is. Tests whether a series is stationary.

Options / Parameters

  • ADF — H0 = unit root (non-stationary).
  • KPSS — Opposite null — stationarity.
  • Phillips-Perron — Non-parametric correction for serial correlation.
tip

Run ADF + KPSS together for a robust conclusion.

ARIMA

What it is. Autoregressive Integrated Moving Average model.

Options / Parameters

  • Auto-select (p,d,q) — Grid search minimizing AIC.
  • p, d, q (manual) — AR lags, differencing, MA lags.
  • Combine Forecasts — Combine forecasts from multiple models (mean/weighted).

GARCH family

What it is. Models time-varying volatility (volatility clustering).

Options / Parameters

  • GARCH(p,q) — Standard symmetric.
  • EGARCH — Exponential — captures asymmetry (leverage effect).
  • GJR-GARCH — Threshold — bad news increases volatility more.
  • DCC-GARCH — Multivariate — dynamic conditional correlations across series.
  • TAR — Threshold AR — different regimes above/below a threshold.

VAR & Granger Causality

What it is. Model multiple series jointly.

Options / Parameters

  • VAR — Vector autoregression with auto lag selection.
  • Granger test — X 'causes' Y if lags of X improve Y's forecast.
  • Cointegration — Individually non-stationary series whose linear combo is.

Panel & Causal

Panel Data (FE/RE/Hausman)

What it is. Longitudinal data (multiple entities × time).

Options / Parameters

  • Fixed Effects — Removes time-invariant unobserved heterogeneity.
  • Random Effects — Assumes effects uncorrelated with X — more efficient.
  • Hausman test — Decides FE vs RE. Rejecting H0 → use FE.

Causal Inference

What it is. Causal identification via quasi-experimental designs.

Options / Parameters

  • DiD — Difference-in-Differences. Requires parallel trends.
  • RDD — Regression Discontinuity at a running-variable cutoff.
  • RKD — Regression Kink — change in slope, not level.
  • Bunching — Estimator from mass concentrated near a threshold.
  • PSM — Propensity Score Matching — matches treated/control on score.
  • IPW/DR — Inverse-Probability Weighting / Doubly Robust.
  • Event Study — Dynamic effects over time relative to an event.
  • Synthetic Control — Builds a synthetic control from weighted units.
  • LATE — Local ATE via IV — compliers only.
  • Lee Bounds — Nonparametric bounds under sample selection.
  • Mediation — Decomposes total effect into direct + indirect.

Missing / Bootstrap / Survey / Bayesian

Missing Data

What it is. Imputation and pattern analysis.

Options / Parameters

  • Simple — mean/median/mode — fast but understates SEs.
  • KNN — Uses k nearest complete rows.
  • MICE — Multiple imputation via chained equations — pooled with Rubin's rules.
  • EM — Assumes multivariate normality.
  • PMM — Predictive Mean Matching — preserves observed distribution.
  • Patterns — Little's MCAR test + pattern visualization.
  • Sensitivity — How results change under MNAR assumptions.

Bootstrap

What it is. Resampling for SE, CI, validation.

Options / Parameters

  • SE / CI — Bootstrap standard error / CI for a statistic.
  • CI types — percentile / basic / BCa / studentized.
  • Regression — Bootstrap coefficient estimates.
  • Jackknife — Leave-one-out.
  • Permutation (2-sample) — Exact test for difference of means.
  • Bayesian bootstrap — Dirichlet weights — Bayesian variant.

Survey / Complex Samples

What it is. Design-based analysis (strata, clusters, weights).

Options / Parameters

  • Design — Specify strata / cluster / weights vars.
  • Means / Totals / Proportions — Design-based estimates.
  • DEFF — Design effect — variance inflation relative to SRS.
  • Rao-Scott χ² — Chi-square corrected for complex design.
  • Replicate Weights — BRR / JKn via replicate weight columns.

Bayesian Models

What it is. MCMC-based Bayesian inference (PyMC).

Options / Parameters

  • Linear / Logistic / Hierarchical — Base models.
  • Draws / Tuning / Chains — MCMC settings (defaults 2000/1000/4).
  • Posterior Predictive Check — Simulate data from posterior and compare to observed.
  • Prior vs Posterior — Visualize how beliefs updated.
  • Credible Interval — 94% HDI per parameter.
  • Bayes Factor — Compare evidence between two models.
  • MCMC Diagnostics — R-hat, ESS, trace plots.

ML & SHAP

ML Models

What it is. Random Forest, XGBoost, SVM, Decision Tree, Gradient Boosting.

Options / Parameters

  • Hyperparameters — Per-model — n_estimators, max_depth, etc.
  • Train/Test Predict — Fit on train, predict on held-out test (test_size configurable).
  • Tree Diagram Export — Export decision tree as Graphviz DOT for visualization.

ML Evaluation

What it is. Performance metrics.

Options / Parameters

  • Cross-Validation — K-fold CV (default 5).
  • ROC / AUC — Binary classification performance.
  • Confusion Matrix — TP/FP/TN/FN counts.
  • Feature Importance — From tree-based models.

SHAP

What it is. Explainability: per-feature contribution per prediction.

tip

Works with xgboost / random_forest / decision_tree.

Niche

Niche (DSGE / Conjoint / Neural / MIDAS)

What it is. DSGE macro, Conjoint analysis, Deep NN, MIDAS (mixed-frequency data).

tip

Accessible via Command Palette (Ctrl+K) for the less common ones.