Choosing a Procedure

Not sure which analysis you need? Start from what you have and what you want to know. Find the row that matches and follow the link.

I want to describe / summarise

You have	Use
One categorical variable — how often each value occurs	Frequencies
Scale variables — mean, SD, range side by side	Descriptives
One scale variable — full distribution, normality, outliers	Explore
A breakdown of a scale variable by groups (no test)	Means
The shape of a distribution, visually	Histogram / Boxplot
The shape of a distribution, without choosing bins	ECDF
Which of several variables move together, all at once	Scatterplot Matrix
Compare values across categories that do not start at zero	Dot Plot
Two variables plus a third as mark size	Bubble Chart
Whether a logistic / count / gamma model fits	Diagnostic plots
A chart that is right but does not look right yet	Editing a chart

I want to compare groups (a scale outcome)

Groups	Assumptions OK?	Use
One mean vs a fixed value	yes	One-Sample T Test
Two independent groups	yes	Independent-Samples T Test
Two independent groups	no (non-normal)	Mann-Whitney U (a rank shift), or the t-test dialog's permutation test (still the mean, no assumptions)
Two related measurements	yes	Paired-Samples T Test
Two related measurements	no	Wilcoxon / Sign
Three+ independent groups	yes	One-Way ANOVA
Three+ independent groups	no	Kruskal-Wallis
Three+ related measurements	no	Friedman
Several factors, ± covariates	yes	GLM

I want to relate two categorical variables

→ Crosstabs (chi-square and association measures), or Custom Tables for a bespoke pivot.

I want to measure association between scale variables

→ Correlate (Pearson, Spearman, Kendall; partial to control for other variables). Not sure the relationship is even linear? Look first: a scatterplot with a LOESS smoother shows the shape before you pick a coefficient.

I want to predict an outcome from predictors

Outcome type	Use
Continuous	Linear regression
Binary (yes/no)	Binary logistic
Count	Poisson / Negative Binomial
Unordered categories	Multinomial
Ordered categories	Ordinal
Continuous, clustered/repeated data	Mixed model
Binary or count, clustered/repeated data — subject-specific effects	Generalized linear mixed model
Any family, clustered/repeated data — population-averaged effects	GEE
Continuous, with influential outliers	Robust (Huber / Tukey)
Continuous, many or collinear predictors, prediction / selection	Regularized (ridge / lasso)
A conditional quantile (median, 90th percentile) rather than the mean	Quantile
A known curve that is nonlinear in its parameters	Nonlinear
An endogenous regressor, with instruments	Instrumental variables (2SLS)
The association structure of a multi-way table of categorical variables	Loglinear
Repeated observations per unit (firms over years, patients over visits), fixed/random effects by name	Panel Regression
How does X affect Y — through a mediator? For whom?	Mediation (with a moderator: moderated mediation)
Known groups, from scale predictors	Discriminant

I want to find structure / reduce variables

Goal	Use
Do these items form one reliable scale?	Reliability
How do the items themselves measure — difficulty, discrimination, where the test is precise?	Item Response Theory
Reduce many variables to a few factors	Factor
Test a hypothesized factor structure	Confirmatory Factor Analysis
Group cases into a fixed number of clusters	K-Means
Group cases, see the structure	Hierarchical
Group cases, let the count be chosen	Two-Step
Find unobserved subgroups from categorical items, with fit indices	Latent Class Analysis

I have time-to-event or time-ordered data

You have	Use
Time until an event, with censoring	Kaplan-Meier / Cox
A series ordered in time, want its structure	Autocorrelation
A series ordered in time, want a forecast	ARIMA / SARIMA
A seasonal series — separate trend, season and remainder	Seasonal Decomposition
Two series — does one lead the other, and by how much?	Cross-Correlation
Several series over time — do they predict each other?	VAR & Granger Causality

I want to combine results across studies

→ Meta-Analysis (one row per study — an effect size and its standard error — pooled under fixed- and random-effects models, with heterogeneity statistics, a forest plot and a funnel plot with Egger's test).

I want to know whether two measurement methods agree

→ Method Comparison (Bland-Altman) (the bias and 1.96 limits of agreement of the per-subject differences, with the mean-vs-difference plot — agreement in the measurement's own units, which a correlation cannot state). For the agreement of raters as a single relative index, use Intraclass Correlation instead; for categorical ratings, Cohen's kappa in Crosstabs.

I want to compare treated and untreated cases that were not randomized

→ Propensity Score (estimate each case's probability of treatment from covariates, then match treated cases to comparable controls or weight by inverse probability; check covariate balance before and after, and run any downstream analysis on the saved matched filter or weight column).

My data is not in yet

Not an analysis question, but often the first blocker:

You have	Use
An SPSS, Stata or SAS file	Import Data — labels, value labels and missing rules arrive with the data
A text file that opens garbled, or as one column	Import Data — encodings are detected and noted; comma, semicolon, tab and pipe delimiters are sniffed from the content
Column-aligned text with no delimiter (`.fwf`, `.prn`)	Fixed-width import — place the column breaks on a preview, or type the ranges

Still unsure?

Ask the AI assistant in plain language — describe your data and your question, and it will pick and run the appropriate procedure. See also the glossary for unfamiliar terms.