The Jarque–Bera test evaluates whether regression residuals are normally distributed by measuring skewness, or asymmetry, and kurtosis, or tail thickness. section 4.2.4
Use the test to interpret residual normality alongside a visual Q–Q check:
- Set the null hypothesis as normally distributed data. section 4.2.4
- Reject normality when the p-value is below 5%. section 4.2.4
- Treat the result as a more rigorous complement to the potentially subjective Q–Q plot. section 4.2.4
>>> from statsmodels.stats.stattools import jarque_bera >>> print(jarque_bera(lm.resid)) SignificanceResult(statistic=2.150037983280995, pvalue=0.34129127355827205)
Apply the test directly to fitted-model residuals with statsmodels; the demonstrated result is a statistic of 2.1500 and p-value of 0.3413. section 4.2.4
