Course contentsShow
← Data Science
Lesson 383 of 2,14510. Hypothesis Tests for MeansFree lesson

Diagnostic Workflow: When to Proceed or Switch Tests

A practical checklist for diagnosing assumptions and deciding whether to use parametric or non-parametric tests.

Diagnostic Workflow: When to Proceed or Switch Tests

You'll learn: A practical, step-by-step checklist for checking assumptions and deciding whether your parametric t-test is safe to use—or if you need a non-parametric alternative.

The Decision Framework

Think of this workflow as a pre-flight checklist for pilots. Before taking off with a t-test, you systematically verify each assumption. If something's wrong, you either fix it or choose a different "aircraft."

Step 1: Check Independence

  • Are observations unrelated? (No duplicates, clusters, or time dependencies?)
  • If violated: This is serious. Consider specialized methods or seek expert guidance.

Step 2: Assess Normality

  • Use visual tools (Q-Q plots, histograms)
  • Consider formal tests (Shapiro-Wilk) for small samples
  • If violated: Check sample size. With n > 30-40 per group, the Central Limit Theorem often saves you (t-tests are robust). For smaller samples or severe skewness, switch to non-parametric tests.

Step 3: Test Equal Variances (for two-sample t-tests)

  • Apply Levene's or Bartlett's test
  • If violated: Use Welch's t-test instead of pooled variance t-test. Easy fix!

Step 4: Make Your Decision

  • All assumptions met? → Proceed with standard parametric t-test
  • Normality violated, small sample? → Switch to non-parametric alternative (Mann-Whitney, Wilcoxon signed-rank)
  • Unequal variances? → Use Welch's t-test
  • Independence violated? → Reconsider your analysis approach entirely

The Robustness Safety Net

Remember: t-tests are remarkably robust to mild violations, especially with balanced designs and moderate-to-large samples. Don't abandon them at the first hiccup—but do document your diagnostic process.

Key Takeaway: A systematic diagnostic workflow helps you confidently choose between parametric and non-parametric tests by evaluating independence, normality, and variance assumptions in order—ensuring your conclusions rest on solid statistical ground.