[转载]121.Comparison among tests for normal

The simulated results for different normality tests were reported below.
Basically, the density of the test distribution and how well the tests perform were plotted one by one.
This is the uniform distribution, symmetric.For this distribution, JB test does not perform well.When the sample size <50, the power of the test <50%. When sample size <50, none of the tests is doing well.
b
Beta(2,2) was symmetric and bell-curved. Not surprisingly, only sample size >250, tests begin to pick up the power except for KW test.
c
T distribution is notably similar to Normal distribution and is asymptotic to normal eventually when df goes to infinity. Thus no test can be very sensitive for sure.
d
e
f
when DF goes larger, none of the tests has the power more than 10%. that is, it is not testable.
g
Laplace is not very common distribution. it is symmetric.Your may also notice that JB test is the best tests for T and Laplace distribution.