Advanced Tests in Statistics
ARE, Robustness of Nonparametric
Tests, and McNemar’s Test
Introduction to Advanced Tests
• Go beyond basic hypothesis testing (t-test, z-test, chi-
square).
• Used when parametric assumptions are violated or
data are paired/categorical.
• Examples: Nonparametric tests, McNemar’s test,
Robust methods.
Asymptotic Relative Efficiency
(ARE)
• Definition: Efficiency comparison between two tests
as sample size grows.
• Formula: ARE = lim (m/n) as n → ∞.
• Interpretation: ARE=1 (equal efficiency), ARE>1
(numerator better).
• Example: Wilcoxon vs. t-test → ARE ≈ 0.955 under
normality.
Robustness of Nonparametric Tests
• Robustness = test validity when assumptions are
violated.
• Nonparametric tests:
• • Do not assume normality.
• • Less sensitive to outliers.
• • Work with ordinal or skewed data.
• Example: Pain score study → use Wilcoxon Signed-
Rank test.
McNemar’s Test: Introduction
• Used for paired/matched nominal data in 2×2
contingency table.
• Tests equality of marginal proportions.
• Focuses only on discordant pairs (b and c).
• Applications: Before–After studies, medical research,
psychology.
McNemar’s Test: Assumptions
• Data must be paired (same subject before–after).
• Variables are categorical with two outcomes.
• Test uses discordant pairs only (b and c).
• Large sample: Chi-square approximation.
• Small sample: Exact binomial test.
Steps for McNemar’s Test
• 1. Arrange data in 2×2 paired table.
• 2. Identify discordant cells b and c.
• 3. Compute χ² = (b - c)² / (b + c).
• 4. Compare with critical χ² (df=1, α=0.05 → 3.84).
• 5. Decision: Reject or Fail to Reject H₀.
Worked Example: McNemar’s Test
• Study: 50 employees tested before and after training.
• Table: a=20, b=10, c=15, d=5.
• Discordant pairs: b=10, c=15.
• χ² = (10 - 15)² / (25) = 1.
• Critical value = 3.84 → Fail to reject H₀.
• Interpretation: No significant improvement after
training.
Summary
• ARE compares efficiency of tests under large
samples.
• Nonparametric tests are robust against assumption
violations.
• McNemar’s test is used for paired binary categorical
data.
• Key: Focus only on discordant pairs in 2×2 table.
Flowchart: McNemar’s Test Steps
Step 1: Arrange data
in 2×2 paired table
Step 2: Identify
discordant cells (b, c)
Step 3: Compute χ² =
(b - c)² / (b + c)
Step 4: Compare with
χ² critical (df=1,
α=0.05=3.84)
Step 5: Decision →
Diagram: ARE Comparison
(Wilcoxon vs t-test)
Wilcoxon Test
ARE ≈ 0.955 under
t-test
normality
Efficient under normality
Better under
skewed/heavy tails