NO CHANGE IN FORMULA, every thing is the same. Skewness and kurtosis are still calculated as they have been. The JB statistic is a goodness-of-fit type of testing based on skewness and kurtosis. Both skewness and kurtosis are based on mean error raised to thr 3rd and 4th power respectively. These basic formula do not change in the JB test.

LINK:

https://en.wikipedia.org/wiki/Jarque%E2%80%93Bera_test

There is no much sense to test for the normality in the case of panel data.

@Paul Louangrath thanks

@Elma Satrovic actually that's what i'm trying to ask, i don't see how normality test is needed for panel data analysis, especially for FEM which indicates that there's a different intercept for each observation. Like for example companies, each company has its own capability. I don't see how the xbar to calculate skewness and kurtosis would represent the xbar for each company. I think that each company has its own xbar especially for fixed effect model

@Elma Satrovic you are correct that there is so much sense to test for the normality in the case of panel data. Normality is not a must in running panel data and that is why you hardly see kurtosis, Skewness or JB result in any Descriptive Statistic result of a panel data.

You know the need for normality test in statistics is to determine whether a data set is modeled for normal distribution and it is needed most in time series than panel because of the number of series as relate to T and N. Use the normal pre and post estimation test in stata to model your data set.

@Jerry Kwarbai any references for your comment

Jerry Kwarbai You meant "no much sense" instead of "so much sense". right?