Chi-Square CalculatorChi-Square Calculator
Effect size

Cramér's V Calculator

A chi-square p-value says whether there is an association; Cramér's V says how strong it is. Enter χ², the total count and the table size.

Cramér's V0.1044
Cohen's w0.1044
df* = min(r − 1, c − 1)1
Table size2 x 3
Rough conventionwV with df* = 1
Small0.10.100
Medium0.30.300
Large0.50.500

By the rough conventions usually credited to Cohen (1988), this association is small. They are rules of thumb, not cutoffs; what counts as large depends on the field.

Reproduce this result in R or Python

R

x2 <- 30.0701; N <- 2757; k <- min(2 - 1, 3 - 1)
sqrt(x2 / N / k)   # Cramer's V
sqrt(x2 / N)       # Cohen's w

Python (SciPy)

import math
x2, N, k = 30.0701, 2757, min(2 - 1, 3 - 1)
print(math.sqrt(x2 / N / k))   # Cramer's V
print(math.sqrt(x2 / N))       # Cohen's w

Checked on November 21, 2026. Results computed by our statistics engine, version 1.0.0.

On this page

What Cramér's V measures

Cramér's V rescales χ² so that it runs from 0 (no association) to 1 (perfect association), whatever the sample size: V = √(χ² / (N × df*)), where df* is the smaller of rows − 1 and columns − 1. Because χ² grows with N, a huge survey can give a tiny p-value for a trivial association; V separates the strength of the association from the size of the sample.

For a 2x2 table, V equals the absolute value of phi. Cohen's w, √(χ² / N), is the same quantity without dividing by df*; it is the usual effect size for goodness of fit and for power calculations.

Worked example

The party identification table on the home page has χ² = 30.0701, N = 2,757, 2 rows and 3 columns, so df* = 1 and V = √(30.0701 / 2,757) = 0.1044. That is pre-filled above. The p-value was tiny, but V shows the association is weak: in this sample, knowing someone's gender tells you only a little about their party identification.

The OpenStax volunteer table on the test of independence page is 3x3, so df* = 2. With χ² = 12.9909 and N = 839, V = √(12.9909 / (839 × 2)) = 0.0880.

How big is big?

Conventionw, or V with df* = 1V with df* = 2V with df* = 3
Small0.10.0710.058
Medium0.30.2120.173
Large0.50.3540.289

The values 0.1, 0.3 and 0.5 for w come from Cohen's 1988 book on power analysis (Cohen 1988). For V in larger tables they shrink by √df*, because the same w spread over more dimensions gives a smaller V. Treat them as rough conventions only: what counts as important depends on the field and the question, not on a fixed cutoff.

Phi and its sign

For a 2x2 table, phi = (ad − bc) / √(product of the four margins) keeps a sign that shows the direction of the association. For R's 2x2 example on the Yates correction page, phi is 0.2103 and V is 0.2103. Cramér's V always uses the uncorrected χ²; using the Yates-corrected value would understate it. The 2x2 calculator shows signed phi directly.

Limits and cautions

  • V depends on the table's margins: two studies with the same underlying relationship but different group sizes can give different values.
  • V cannot reach 1 unless the row and column totals allow a perfect association.
  • V says nothing about which cells drive the association; look at the adjusted residuals on the test of independence page for that.
  • A confidence interval for V needs methods beyond this page; report V with the test result and N.

Common questions

Can Cramér's V be negative?

No. It runs from 0 to 1. For a 2x2 table, use signed phi to show direction.

Should I use the Yates-corrected χ²?

No. Cramér's V is defined with the uncorrected Pearson χ².

What df* should I use for a 2x5 table?

The smaller of 2 − 1 and 5 − 1, which is 1.

Sources