Welch's independent-samples t-test
A common starting approach when the target quantity is a difference in population means.
Explore mean comparisons →Thinking Like an Investigator · Interactive guide
Choose an Analysis Starting Point
Describe your research setup to identify methods worth considering—then examine the estimand, assumptions, design, and interpretation before choosing one.
Build the analytical situation
This guide offers common starting approaches, not a mechanically correct test. More than one method may be defensible because different methods can estimate different quantities.
The choices cover the one-outcome methods in UCLA’s statistical-test guide, including paired and factorial comparisons. Some methods do not yet have a learning module; their results are labeled accordingly. Analyses of multiple outcomes and variable sets are outside this guide’s current scope.
Common starting approach
Your choices narrow the field—they do not make the decision.
Design audit: Confirm the analytic sample, missingness, and assumptions before execution.
Before you execute
Name the estimand
What magnitude, contrast, relationship, or probability do you want to estimate?
Identify the unit
What does one row represent, and are observations actually independent?
Define the population
Who does the analytic sample represent, including after exclusions and missingness?
Audit the design
Do pairing, repeated measures, clustering, or survey design change the implementation?
Research Specification–Driven Workflow: record the intended outcome, focal predictor,
analytical sample, estimand, model family, reference groups, survey design, and validation checks
before asking AI or software to execute the analysis.