Course contentsShow
← Data Science
Lesson 398 of 2,14510. Hypothesis Tests for MeansFree lesson

Choosing Between Parametric and Non-Parametric Tests

Develop a decision framework for selecting the appropriate test based on data characteristics and assumptions.

Choosing Between Parametric and Non-Parametric Tests

You'll learn: How to systematically decide whether to use parametric tests (like t-tests) or non-parametric alternatives (like Wilcoxon or Mann-Whitney) based on your data's characteristics.

The Decision Framework

Think of choosing a statistical test like selecting the right tool from a toolbox. Parametric tests are powerful precision instruments—they work brilliantly if conditions are right. Non-parametric tests are your reliable all-purpose tools—they work in more situations but may be slightly less efficient when conditions are ideal.

The Decision Checklist

Use Parametric Tests (t-tests, ANOVA) when:

  • Normality holds: Your data (or sampling distribution) is approximately normal, especially with small samples (n < 30)
  • Equal variances: Groups have similar spread (for two-sample tests)
  • Interval/ratio data: Meaningful numeric measurements
  • Independence: Observations don't influence each other

Switch to Non-Parametric Tests when:

  • Normality violated: Severe skewness, outliers, or small samples from non-normal populations
  • Ordinal data: Ranks matter, but exact distances don't (e.g., survey ratings)
  • Unequal variances: Groups differ dramatically in spread and sample sizes are unequal
  • Small samples: Too few observations to assess normality reliably (n < 15-20)

The Practical Workflow

  1. Check assumptions using visual tools (histograms, Q-Q plots) and tests (Shapiro-Wilk, Levene's)
  2. Consider robustness: With n > 30-40, t-tests handle mild violations well (Central Limit Theorem)
  3. Evaluate trade-offs: Parametric tests have higher power when assumptions hold; non-parametric tests are safer when assumptions are questionable
  4. Match the question: If you care about means, try parametric first. If medians or distributions matter, non-parametric may be more appropriate

Key Takeaway: Choose parametric tests when assumptions are satisfied for maximum power; switch to non-parametric alternatives when assumptions are violated or data is ordinal—but always check your assumptions systematically before deciding.