Course contentsShow
← Data Science
Lesson 429 of 2,14511. Hypothesis Tests for Proportions and CategoriesFree lesson

Effect Size: Cramér's V and Phi

Understand how to measure the strength of association between categorical variables beyond significance.

Effect Size: Cramér's V and Phi

You'll learn: How to quantify the strength of association between categorical variables after finding a significant chi-squared result.

Why Effect Size Matters for Categorical Data

A chi-squared test tells you whether two categorical variables are related (statistical significance), but not how strongly they're associated. With large samples, even tiny, practically meaningless relationships become statistically significant. Effect size measures fill this gap.

Think of it like measuring a friendship: a p-value tells you whether two people are connected at all, but effect size tells you whether they're acquaintances or best friends.

Cramér's V

Cramér's V is the standard effect size measure for chi-squared tests of independence. It ranges from 0 (no association) to 1 (perfect association) and works for any size contingency table.

The formula adjusts for both sample size and table dimensions:

V = √(χ² / (n × min(r-1, c-1)))

Where:

  • χ² = your chi-squared test statistic
  • n = total sample size
  • r = number of rows, c = number of columns

Interpretation Guidelines

  • Small effect: V ≈ 0.10
  • Medium effect: V ≈ 0.30
  • Large effect: V ≈ 0.50

Phi (φ)

Phi is a special case used exclusively for 2×2 contingency tables. It's simpler and directly interpretable like a correlation coefficient:

φ = √(χ² / n)

Phi can range from 0 to 1 (or -1 to +1 if directionality matters).

When to Use Which

  • Use Cramér's V for any contingency table (2×2, 3×4, etc.)
  • Use Phi only for 2×2 tables (it's equivalent to V in this case)

Key Takeaway: Cramér's V and Phi translate statistically significant chi-squared results into meaningful effect sizes, helping you distinguish between "detectable" and "important" associations in categorical data.