The JB value calculator works out the Jarque–Bera test statistic from a sample’s skewness and kurtosis, and you enter the sample size plus the skewness and excess kurtosis values. It returns the JB statistic and the p-value, which tell you whether the data are plausibly normally distributed. Below are worked examples, the formula, and the practical details you need to interpret the result correctly.
Formula
The Jarque–Bera test statistic is defined as:
JB = (n / 6) × (S² + (K − 3)² / 4)
where:
- n = sample size (number of observations, unitless)
- S = sample skewness (unitless; 0 for a symmetric distribution)
- K = sample kurtosis (unitless; 3 for a normal distribution, so excess kurtosis K − 3 = 0)
The test statistic follows a chi-square distribution with 2 degrees of freedom under the null hypothesis that the data come from a normal distribution. The p-value is computed from that chi-square distribution; the calculator above uses the approximation p ≈ exp(−JB/2), which is accurate for JB up to about 10. For larger JB values, use a chi-square table or software for the exact p-value.
Worked Example
Suppose you have a sample of n = 100 observations, the sample skewness is 0.15, and the sample excess kurtosis is 0.30. (That means the kurtosis itself is 3.30.) Plug the numbers into the formula:
JB = (100 / 6) × (0.15² + 0.30² / 4)
First compute the squares: 0.15² = 0.0225 and 0.30² = 0.09. Then 0.09 / 4 = 0.0225. Add: 0.0225 + 0.0225 = 0.045. Multiply by 100/6 ≈ 16.6667: JB = 16.6667 × 0.045 = 0.75.
The p-value approximation: p ≈ exp(−0.75 / 2) = exp(−0.375) ≈ 0.6873. Since the p-value is far above 0.05, you do not reject normality. The data are consistent with a normal distribution.
Assumptions
- The sample is a random sample from the population.
- The observations are independent of each other.
- The sample size is large enough for the chi-square approximation to hold (typically n > 30, but more is better).
- The skewness and kurtosis are computed from the sample, not from a theoretical distribution.
What the Result Means
The JB statistic measures how far the sample’s skewness and kurtosis are from the values expected under normality (skewness = 0, excess kurtosis = 0). A JB value close to 0 means the data are close to normal. A large JB value means the data deviate from normality, either because they are skewed, have heavy tails, or both.
The p-value is the probability of observing a JB statistic at least as extreme as the one you got, assuming the data really are normal. Common practice:
- p ≥ 0.05: fail to reject normality (data are consistent with normal).
- p < 0.05: reject normality (data are not consistent with normal).
Remember that failing to reject is not proof of normality; it just means the sample does not provide strong evidence against it.
How to Compute Skewness and Kurtosis
You need the sample skewness and excess kurtosis before using the JB calculator. Most statistical software (R, Python, Excel, SPSS) can compute these directly. The formulas are:
Skewness S = (1/n) × Σ((xᵢ − x̄)³) / s³
Excess kurtosis K − 3 = (1/n) × Σ((xᵢ − x̄)⁴) / s⁴ − 3
where x̄ is the sample mean and s is the sample standard deviation (using n−1 in the denominator for the sample version). Some software reports the “excess kurtosis” directly; others report the raw kurtosis (where normal = 3). Make sure you enter the correct value: the formula above expects excess kurtosis (K − 3). If your software gives raw kurtosis, subtract 3 first.
Common Mistakes and Misconceptions
- Using raw kurtosis instead of excess kurtosis. If you enter 3.0 for a normal sample, the JB statistic becomes huge and you incorrectly reject normality. Always subtract 3 if your software reports raw kurtosis.
- Ignoring sample size. The JB statistic scales with n. A tiny deviation from normality can become significant with a large sample, even if the data are practically normal. Conversely, with a small sample, the test has low power to detect non-normality.
- Using the p-value approximation beyond JB ≈ 10. The exp(−JB/2) approximation is only good for small JB. For larger values, use a chi-square table or software.
- Applying the test to small samples. The chi-square approximation is asymptotic; for n < 30, the p-value may be inaccurate.
- Thinking the test tells you the distribution type. The JB test only checks normality; it does not tell you what distribution the data follow if not normal.
When to Use the JB Test
The Jarque–Bera test is commonly used in econometrics and finance to check whether regression residuals are normally distributed, because many statistical tests (t-tests, F-tests) assume normality. It is also used in quality control and any field where normality of data is a prerequisite for parametric tests.
However, the test is known to be sensitive to outliers and can be underpowered for small samples. Many statisticians prefer graphical methods (Q-Q plots) or other normality tests (Shapiro–Wilk) for small samples. The JB test is best used with large samples (n > 100) and as one of several checks.
Limitations and Alternatives
The JB test has limitations:
- It only uses the third and fourth moments; it may miss departures in other aspects of the distribution.
- It is not robust to outliers, which can inflate skewness and kurtosis.
- It requires a large sample for the chi-square approximation to be reliable.
Alternatives include the Shapiro–Wilk test (better for small samples), the Kolmogorov–Smirnov test (with estimated parameters), and the Anderson–Darling test. Graphical methods like histograms and Q-Q plots are always recommended alongside any formal test.
Frequently Asked Questions
What is the difference between kurtosis and excess kurtosis? Kurtosis measures the “tailedness” of a distribution. The normal distribution has a kurtosis of 3. Excess kurtosis is simply kurtosis minus 3, so the normal distribution has excess kurtosis of 0. The Jarque–Bera formula uses excess kurtosis; if you have raw kurtosis, subtract 3 before entering it.
Can I use the JB test with a small sample size? Technically yes, but the p-value may be inaccurate because the test statistic only approximately follows a chi-square distribution with large samples. For n < 30, prefer the Shapiro–Wilk test or a Q-Q plot.
What if my p-value is exactly 0.05? The 0.05 threshold is arbitrary. A p-value of 0.05 means there is a 5% chance of seeing a JB statistic this large if the data are normal. You can reject or not; the evidence is borderline. Report the exact p-value and let the reader decide.
Does the JB test work for any distribution? No, it only tests for normality. If the data are not normal, the test does not tell you what distribution they follow.
Why does the calculator use exp(−JB/2) for the p-value? The p-value from a chi-square distribution with 2 degrees of freedom is exactly e^(−x/2) for a test statistic x. That is the exact formula, not an approximation, for df=2. So the calculator gives the exact p-value for any JB statistic.
What if my skewness or kurtosis is negative? Skewness can be negative (left-tailed data) and excess kurtosis can be negative (light tails, like a uniform distribution). The formula squares them, so negative values are fine. Just enter them as-is.
Related
Formula used
in n | Sample size n | - | 100 | 3 in s | Skewness | - | 0.15 | -10000 in k | Excess kurtosis | - | 0.3 | -10000 out jb | Jarque-Bera statistic | - | (n/6) * (s^2 + (k^2)/4) | 4 out p | p-value (chi-square df=2) | - | exp(-jb/2) | 4
Each line is one input (in, sel, list) or one result (out: label, unit, formula, decimals). How the calculators work
Method: the standard formula for this calculation with the constants and conversion factors written on the page, rounded as each result states. See method and constants, the disclaimer or report an error.
More in this section
- KNO3 Normality Calculation: How To Do It0.2000 mol/L
- Lab Normality Formula & How to Calculate It0.2000 eq/L
- KMnO4 Normality Formula & How to Use It0.5000 N
- KMnO4 Normality in Titration: How to Calculate0.5000 N
- How to Test Normality of Data in R
- KH2PO4 Normality Calculation0.0859 mol/L