多重比较真的不适合用t检验吗?--对'金科玉律'的挑战

摘要

At present, the proposition which the t-test is not a proper method for the multiple comparison is a gold rule in statistics. But it is not exact. Although the increase of the test number raises the probability of the Type Ⅰ error, it does not demonstrate the proposition. We should calculate the real probability of the Type Ⅰ error with P-value not the α-value from each test.