I beg to differ. Never heard of such a rule. Working hypotheses and statistical Null hypotheses

are not related this way. Otherwise one-tailed tests would be the rule. Usually, one-tailed tests

are performed if the scientist has absolutely no interest in results which go in the unexpected

direction. Which not often is the case in science (I suppose).

With kind regards

Karabiner

I never mentioned any "working hypotheses". I said: "alternative hypothesis". By definition, the

*alternative hypothesis* (usually denoted with "H1" or "Ha") is one of the two formal, statistical hypotheses which are compared by the statistical test in question. The statistical test allows one to decide whether to accept the null hypothesis (H0) or reject the null hypothesis and accept the alternative hypothesis. For example, in the comparison

H0: mu_1 <= mu_2,

H1: mu_1 > mu_2,

the 1st row describes the null hypothesis and the 2nd row describes the alternative hypothesis. The null hypothesis is always the one with sign "=", "<=", ">=". By definition.... This is Statistics 101. Well, actually, it was called "Statistics 60" at my university.