# p value in one-sided test

$$H_0:\mu=0 \quad\quad H_1:\mu>0$$ we assume the distribution of the sample is Gaussian.
If $p$-value is very big, sample mean very small, we still accept null hypothesis.
Isn't it counterintuitive? Why are we still using it?

E.g. $x_1,x_2,\ldots,x_n$, sample mean is $-5$, variance is known as 1.
$$H_0:\mu=0 \quad\quad H_1:\mu>0$$ It's easy to see that $p$-value is very big, but we still accept mu = 0. But actually it is very intuitive that $\mu$ is less than 0 or at least not equal to zero?

• Your question is not clear - could you elaborate a little bit more? Providing an example illustrating your question would also make it more clear. – Tim Oct 20 '15 at 11:12
• p value is calculated based on a statistic(an observable random variable or a function of sample), the statistic may not only depend on $\mu$,it may also depend on standard error. – Deep North Oct 20 '15 at 11:31
• This is for your edited question. If the p value is very small, why you still accept $\mu=0$? – Deep North Oct 20 '15 at 11:32
• @DeepNorth No, sigma is known to me. This test is only for mu. – Li haonan Oct 20 '15 at 11:32
• @DeepNorth Sorry, typo. – Li haonan Oct 20 '15 at 11:33

$H_0 \mu \le 0 \\ H_1 \mu > 0$
So, in your example, where $\bar{X} = -5$ it is clearly much less than 0 and the null cannot be rejected. That's the price of a one tailed test.