The Jarque-Bera test checks whether a sample looks like it came from a normal distribution, based on its skewness and kurtosis. The Jarque-Bera calculator takes the sample size, the skewness and the kurtosis, and gives the test statistic, the p-value, the critical value at 5 % and the excess kurtosis.
How the calculation works
- Compute the sample skewness S and the sample kurtosis K of your data. A normal distribution has S = 0 and K = 3.
- The Jarque-Bera statistic is JB = n ÷ 6 × (S² + (K − 3)² ÷ 4), where n is the sample size. It grows with the departure from symmetry and from the normal tail weight.
- Under the hypothesis of normality, JB follows a chi-square distribution with 2 degrees of freedom. The p-value is 1 minus its cumulative probability, and it simplifies to exp(−JB ÷ 2).
- A p-value below 0.05, or a JB above the critical value of 5.991, rejects normality at the 5 % level.
Example
A sample of 120 values has a skewness of 0.35 and a kurtosis of 3.9. JB = 120 ÷ 6 × (0.1225 + 0.81 ÷ 4) = 20 × 0.325 = 6.5. The p-value is 0.0388, which is below 0.05 and above the critical value of 5.991. Normality is rejected at the 5 % level, mostly because of the heavy tails (excess kurtosis 0.9).
Why it matters
Many statistical methods assume normal data or normal residuals. The Jarque-Bera test is a quick check, popular in econometrics and finance, and it can be done from the summary statistics already printed by most software.
Good to know
With large samples the test is very sensitive and may reject normality for departures too small to matter. With small samples it has little power. Look at a histogram or a Q-Q plot as well, and remember that a non-significant result does not prove that the data are normal.
Related tools
The normal distribution calculator works with normal probabilities, and the Pearson skewness calculator estimates asymmetry.