This illustrates the importance of detecting mixed heteroskedasticitywhich,
in our opinion, must be frequent in practice.
Unfortunately, in this respect, our test only seems to help when:
-the size of sample is not too small (30 or more)
-the standard deviation of x is large (30 when the mean of x is 50)
-z appears with power no less than 1 in nbsolute value in the formula of u.I