then T does not have a t distribution
and P(Type I error) can differ from !, sometimes
substantially, and so we really do not have as
much control of P(Type I error) as the chosen value
of ! leads us to believe. A hypothesis test for which
P(Type I error) = ! is called an exact test. The onesample
t -test is guaranteed to be an exact test only
when the data come from a normal distribution