ChakataStat
Download

Glossary

Short definitions of the statistics and ChakataStat terms used throughout this guide. For "which test do I need?", see Choosing a procedure.

ChakataStat terms

  • Case: one row of the dataset; one observation (e.g. one participant).
  • Variable: one column; a measured attribute (e.g. Age).
  • .ckd: ChakataStat's native document: data plus all variable definitions in one file.
  • .ckdict: a standalone data dictionary: variable metadata without the data.
  • .cks: ChakataStat syntax, the readable command-stream script the syntax console reads and writes.
  • Measurement level (Measure): how a variable is treated: Scale (true number), Ordinal (ordered categories), Nominal (unordered categories).
  • Value label: text shown for a stored code (e.g. 1 → Male).
  • User-missing value: a code that means "no answer", excluded from analysis.
  • Case semantics: Select (which cases), Weight (how much each counts) and Split (analyse per group); applied to every analysis.
  • Tool / command: the single named operation behind every menu action, shared by the syntax console, the AI assistant and MCP.
  • Command journal: the running record of every command a session has run; what the Syntax tab shows and what a .cks file saves.
  • Syntax editor: the Syntax tab, a real editor over the journal, with highlighting, Ctrl+Space completion and errors marked before you run.
  • Chart editor: the Edit chart… panel on a chart card. It changes how a chart looks, never what it measures, and writes the change back into the chart's own command, so a restyled chart re-runs restyled.
  • User-defined tool: a saved .cks recipe registered under a name, which the syntax console, the AI assistant and MCP can then run like any built-in.
  • Concept DOI: the permanent identifier for citing ChakataStat itself (Help → How to Cite), stable across releases; each tagged version also gets its own version DOI.
  • Foreign formats: the statistical files ChakataStat reads directly: SPSS (.sav, .zsav), Stata (.dta), SAS (.sas7bdat, .xpt), plus Parquet and Arrow. Unlike CSV they carry variable and value labels with the data. See Working with data.

Common statistics terms

  • Mean / Median / Mode: average, middle value, most frequent value.
  • Standard deviation (SD): typical spread around the mean.
  • Confidence interval (CI): a range that, at a stated level (e.g. 95%), plausibly contains the true value.
  • Significance (p-value): the probability of seeing data this extreme if the null hypothesis were true; small values are evidence against the null.
  • Alpha (α): the significance threshold for a reject/retain decision (commonly 0.05).
  • One- vs two-tailed: whether the test looks for a difference in one specified direction or in either direction.
  • Null hypothesis: the "no effect / no difference" baseline a test evaluates.

Relationships and models

  • Correlation: strength and direction of association (−1 to +1). Pearson is linear; Spearman and Kendall are rank-based.
  • Partial correlation: association between two variables controlling for others.
  • Regression: predicting an outcome from one or more predictors.
  • Coefficient (B): the modelled effect of a predictor on the outcome.
  • Odds ratio / Exp(B): in logistic regression, the multiplicative change in odds per unit of a predictor.
  • Hazard ratio: in Cox regression, the multiplicative change in the event rate per unit of a predictor.
  • R²: the share of the outcome's variance the model explains.
  • Regression diagnostics: the plots that check a fitted model's assumptions: Residuals vs Fitted, Scale-Location and a Normal Q-Q of the residuals.
  • Leverage / Cook's distance: per-case measures of how unusual a case's predictors are, and how much the fit would move without it.
  • Mediation: decomposing an effect into a direct path and an indirect path through a mediator; moderated mediation lets that depend on a third variable, and the Johnson–Neyman interval reports where the effect is significant.
  • Fixed vs random effects (panel): for repeated observations on the same entities: fixed effects use only within-entity variation, random effects also use between-entity variation; the Hausman test chooses between them.
  • Propensity score: the modelled probability of being treated, used to match or weight cases so treated and control groups become comparable; balance is judged on standardised differences.
  • ATE / ATT: the average treatment effect over everyone, versus over the treated only.

Group comparisons

  • t-test: compares two means.
  • ANOVA: compares three or more means; the F statistic is its test.
  • Levene's test: tests whether groups have equal variances.
  • Post-hoc test: pairwise comparisons after a significant ANOVA (Bonferroni, Scheffé, Tukey HSD).
  • Chi-square (χ²): tests association in a contingency table.
  • Nonparametric test: a test that does not assume normality, using ranks.
  • Equivalence test (TOST): two one-sided tests asking whether a difference is small enough to be practically nil. "No significant difference" is not the same claim, which is why this test exists.
  • Permutation test: a p value built by reshuffling the data many times rather than from a theoretical distribution.

Multivariate & specialised

  • Factor analysis: finds latent factors behind many correlated variables; loadings show each variable's link to a factor; rotation (Varimax, Promax) makes them interpretable.
  • Cronbach's alpha: internal-consistency reliability of a set of items.
  • Cluster analysis: groups similar cases (k-means, hierarchical, two-step).
  • Discriminant analysis: predicts known group membership from predictors.
  • Censoring: in survival data, a case whose event had not occurred by the end of observation.
  • Kaplan-Meier: a model-free estimate of a survival curve; log-rank compares curves between groups.
  • ACF / PACF: autocorrelation and partial autocorrelation of a time series.
  • ARIMA / SARIMA: Box-Jenkins time-series models; SARIMA adds seasonal terms; both can forecast.
  • VAR / Granger causality: a vector autoregression models several series jointly; a Granger test asks whether one series' past helps predict another's future (prediction, not cause).
  • Confirmatory factor analysis (CFA): tests a factor structure you specify in advance; modification indices suggest what would improve fit, and measurement invariance checks the structure means the same thing across groups.
  • Latent class analysis (LCA): finds unobserved subgroups behind categorical responses; the cluster analysis of categorical data.
  • Item response theory (IRT): models each item's difficulty and discrimination separately from the person's ability.
  • Meta-analysis: pools effect sizes across studies; heterogeneity (Q, I², τ²) says how much they disagree, the forest plot shows every study, and the funnel plot looks for small-study bias.
  • Bland-Altman: a method-comparison plot of the mean of two measurements against their difference, with the limits of agreement.
  • LOESS: a smooth curve fitted locally through a scatterplot, with no global equation assumed.
  • Multiple imputation: filling gaps m times to preserve uncertainty, pooling the results by Rubin's rules; the FMI reports how much of the answer came from imputation rather than data.
  • Survey design: telling ChakataStat the sample was not simple random, with its strata, clusters (PSUs) and sampling weights. The everyday procedures then estimate design-based, reporting a design df and an effective sample size rather than pretending N independent observations.