Analyzing Data
The Analyze menu is the heart of ChakataStat. This section documents every procedure, grouped the way the menu is. Each page tells you when to use a procedure, how to fill the dialog, and how to read the output, with worked examples on the bundled sample data (File → Open Sample Dataset).
If you are not sure which test you need, start with the procedure chooser.
How every analysis dialog works
The dialogs share one shape, so once you know one you know them all:
- A source list of variables on the left; target list(s) (Variables, Factor, Dependent, …) on the right. Double-click or use the arrows to move variables across.
- An Options section (collapsible) holds the advanced knobs — confidence level, significance level (α), one/two-tailed, missing-data handling, random seed — hidden by default so the common path stays simple.
- A Run button (also Enter) dispatches the analysis; it shows a spinner while the engine works and blocks a double-submit. Esc cancels.
- Results append to the Output View as titled cards. Many dialogs also offer a Plot… checkbox that adds a chart built from the result.
Every analysis respects the active case semantics — Select, Weight and Split — described in Case Semantics & Transforms. And every run is recorded in the Syntax console.
The procedures
| Page | Menu group | Covers |
|---|---|---|
| Descriptive Statistics | Descriptive Statistics | Frequencies, Descriptives, Explore, Crosstabs |
| Compare Means | Compare Means, Bayesian | Means, the three T tests (independent with an exact/Monte-Carlo permutation test), the Equivalence T Test (TOST), One-Way ANOVA, and the Bayesian t test / correlation / ANOVA / linear regression (Bayes factors + posteriors) |
| Correlation & Regression | Correlate, Regression, GLM, Mixed | Bivariate/partial correlation (bivariate with a permutation test); linear (with regularized, robust, nonlinear and instrumental-variables variants), logistic, Poisson, negative-binomial, gamma, multinomial, ordinal, quantile, GEE and loglinear regression; mediation analysis (with moderated mediation and the interaction plot); factorial GLM/ANCOVA; linear and generalized linear mixed models; repeated measures & MANOVA |
| Panel Regression | Regression | Fixed effects (within), random effects (Swamy–Arora GLS) and pooled OLS for long-format panel data; the Hausman test; cluster-robust (by entity) standard errors; within/between/overall R² |
| Meta-Analysis | Meta-Analysis | Fixed-effect and DerSimonian-Laird random-effects pooling of per-study effect sizes; heterogeneity (Q, I², τ²); forest and funnel plots with Egger's test |
| Propensity Score | Propensity Score | Logistic propensity model; greedy 1:k nearest-neighbour matching (caliper, ATT) or IPTW weighting (ATE/ATT); standardized-mean-difference balance table and plot; saved filter/weight columns for downstream analyses |
| Nonparametric Tests | Nonparametric Tests | Mann-Whitney, Wilcoxon, Sign, Kruskal-Wallis, Friedman, Kendall's W, Runs, Binomial, Tests of Normality |
| Scale, Reduction & Classify | Scale, Dimension Reduction, Classify, ROC Curve | Reliability; intraclass correlation; factor analysis; k-means, hierarchical & two-step cluster; discriminant analysis; ROC curve / AUC |
| Method Comparison (Bland-Altman) | Scale | Agreement between two measurement methods: bias, SD of the differences and the 1.96 limits of agreement with t-based CIs; the proportional-bias regression; the mean-vs-difference plot |
| Latent Class Analysis | Classify | Unobserved subgroups from categorical indicators: EM mixture model with class-enumeration fit indices (AIC/BIC/aBIC, entropy), class profiles, classification diagnostics, saved class/posterior columns |
| Item Response Theory | Scale | Rasch / 2PL item parameters (discrimination, difficulty with SEs) for binary items by marginal maximum likelihood; item and test information curves showing where the test measures well |
| Confirmatory Factor Analysis | Dimension Reduction | Testing a hypothesized measurement structure: ML loadings with standard errors, factor correlations, and the conventional fit block (χ², CFI, TLI, RMSEA, SRMR); modification indices; multi-group measurement-invariance testing (configural → strict) with latent means; ordinal indicators by WLSMV (polychorics, thresholds, the scaled-shifted χ²) |
| Complex Samples (Survey Designs) | Data → Survey Design | Declaring strata/PSU/sampling-weight designs; design-based (Taylor-linearized) estimates, CIs on the design df and design effects in Explore, Frequencies, Crosstabs (Rao–Scott), Means and linear/logistic regression; domain estimation for subgroups |
| Survival & Time Series | Survival, Time Series | Kaplan-Meier, Cox regression; autocorrelation, ARIMA/SARIMA with forecasting, stationarity tests, exponential smoothing, seasonal decomposition, cross-correlation |
| VAR & Granger Causality | Time Series | Vector autoregression of two or more series: per-equation OLS with lag-order selection (AIC/HQ/BIC/FPE), residual covariance, and pairwise Granger causality F tests |
| Custom Tables | Tables | Custom (pivot) tables; multiple-response tables |
A note on assumptions
Many procedures carry optional assumption checks (Levene's test of equal variances, Mauchly's sphericity test, Tests of Normality, casewise influence diagnostics). Where they exist, the relevant page points them out — turning them on costs nothing and keeps you honest about whether the headline test is valid.