The Chi-Square Distribution
The chi-square distribution is the yardstick every chi-square test is measured against. Its shape depends on a single number, the degrees of freedom.
Checked on November 21, 2026. Results computed by our statistics engine, version 1.0.0.
In short
- Mean df, variance 2 df
- Right-skewed, becoming symmetric as df grows
- Built from squared normal variables
On this page
Definition
If Z1, ..., Zk are independent standard normal variables, the sum of their squares, Z1² + ... + Zk², has a chi-square distribution with k degrees of freedom. That is why the Pearson statistic follows it approximately: each (O − E) / √E behaves roughly like a standard normal variable when expected counts are large, and the constraints imposed by the totals remove some degrees of freedom.
The distribution is a special case of the gamma distribution, and its cumulative probabilities are the regularized incomplete gamma function P(k/2, x/2) (NIST DLMF chapter 8).
Shape by degrees of freedom
With 1 or 2 degrees of freedom the density is highest at zero. From 3 upward it rises to a peak at df − 2 and then falls away in a long right tail. As df grows the curve moves right, spreads out and becomes more symmetric.
Key properties
| df | Mean | Variance | Mode | Skewness | P(X > df) |
|---|---|---|---|---|---|
| 1 | 1 | 2 | 0 | 2.828 | 0.3173 |
| 2 | 2 | 4 | 0 | 2.000 | 0.3679 |
| 3 | 3 | 6 | 1 | 1.633 | 0.3916 |
| 5 | 5 | 10 | 3 | 1.265 | 0.4159 |
| 10 | 10 | 20 | 8 | 0.894 | 0.4405 |
| 30 | 30 | 60 | 28 | 0.516 | 0.4657 |
The mean is df and the variance is 2 df. Skewness is √(8 / df), so it falls slowly. The last column shows that a χ² equal to its degrees of freedom is unremarkable: with 1 df the chance of exceeding it is 0.317, and with 30 df it is 0.466. A χ² close to its df is about what you should expect when nothing is going on.
Large degrees of freedom
As df grows the distribution approaches a normal distribution with mean df and variance 2 df. The approach is slow because of the skew. At 100 df, the upper 5% point is 124.342; the normal approximation gives 123.262. The Wilson-Hilferty cube-root transformation, df × (1 − 2/(9 df) + z√(2/(9 df)))³, is much closer at 124.340, which is why calculators use it as a starting point before refining to the exact value.
Where it appears
- Tests of independence, homogeneity and goodness of fit (chi-square calculator).
- The McNemar test and the likelihood-ratio G-test.
- Tests and confidence intervals for a single variance: (n − 1)s² / σ² has a chi-square distribution with n − 1 df when the data are normal. This test is sensitive to non-normal data, so it is used less often.
- Confidence intervals for a Poisson rate, and many likelihood-ratio tests in regression models.
Test for a variance, briefly
With a sample of 11 measurements and sample variance s², the statistic 10s² / σ0² is compared with the chi-square distribution with 10 df. For a two-sided test at the 5% level, reject when it falls below 3.247 or above 20.483. The critical value calculator gives these lower and two-tailed values for any df.
Look up any probability
Use the p-value calculator to shade any upper tail on the curve, or print the chi-square table.
Sources
- NIST Digital Library of Mathematical Functions, chapter 8: incomplete gamma and related functions checked November 21, 2026