1
$\begingroup$

In a $t$-test for difference in means, the null is usually specified as $\mu_X-\mu_Y = 0$ and the alternative hypothesis is EITHER $\mu_X-\mu_Y>0$, or $\mu_X-\mu_Y<0$. In any testing, we normally find the observed $t$-statistic, which we call $t^*$, and then compared it using the p-value, defined as:

$$ P(t^*>t|H_0) $$

where it is defined to be the probability of observed a $t$-statistic as extreme as the one we observed from the data, GIVEN we are under the null distribution. This p-value corresponds normally to the right-tailed test, where the alternative is $\mu_X-\mu_Y>0$. For the case where $\mu_X-\mu_Y<0$, our p-value then becomes $P(t^*<t|H_0)$.

My question then is: why does $\mu_X-\mu_Y>0$ translate to finding the probability of the $t$-statistic being greater than some observed $t$ under the null distribution and vice versa? I understand that the statistic itself, $\mu_X-\mu_Y>0$ has direction, but why does it translate to $t^*$ having that direction too? Is it by the fact dividing by the standard deviation is dividing by something positive and hence directions are preserved?

Meaning, the statistic $\mu_X-\mu_Y$ is invariant in terms of sign due to the standard deviation being a positive value?

$\endgroup$
1
$\begingroup$

The P-value is the probability of a value of the test statistic as or more extreme that what was observed--in the direction, or direction(s) of the alternative. Right tail for alternative $>,$ left tail for alternative $<,$ both tails for alternative $\ne.$

Here is output for left- and two-tailed tests, done in Minitab. Same data, same null hypothesis, different P-values. Left-sided is significant at 5% level, two-sided is not.

One-Sample T 
Test of μ = 100 vs < 100

N   Mean  StDev  SE Mean  95% Upper Bound      T      P
9  91.23  13.20     4.40            99.41  -1.99  0.041


One-Sample T 
Test of μ = 100 vs ≠ 100

N   Mean  StDev  SE Mean       95% CI          T      P
9  91.23  13.20     4.40  (81.08, 101.38)  -1.99  0.081

In the following plot the P-value for the two-sided test is the area outside the two vertical lines (left line for observed $T,$ right dotted line just as far from 0 in the opposite direction).

enter image description here

$\endgroup$
  • $\begingroup$ Could I ask why the p-value is in the direction of the alternative? It seems the p-value deals with the $t$ statistic while the alternative deals with the data. Why does the data being some direction translate into the $t$ statistic being in that same direction? $\endgroup$ – user321627 Dec 14 '16 at 10:39
  • 1
    $\begingroup$ Consider $H_a: \mu < 100$ in my Answer. $\bar X$ is the estimate of $\mu.$ We will Rej $H_0: \mu = 100$ in favor of $H_a$ if $\bar X$ is much smaller than 100. That is, if $\bar X - 100$ is much smaller than 0. To decide what 'much' means we find $T=(\bar X - 100)/[S/\sqrt{n}]$ and Rej if $T < t^*,$ where $t^*$ is critical val from table. So "$\bar X - 100$ much smaller than 0" becomes" $T < t^*.$" Thus t statistic reflects data and hypothetical value of $\mu.$ $\endgroup$ – BruceET Dec 14 '16 at 16:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.