Do you need given standard deviation when determining the critical values of a test statistic?

267 Views Asked by At

I am given the following information:

$$\alpha = 0.025$$ $$n = 15$$ $$\sigma^2 = 100$$

I need to find the critical values.

My attempt: I chose the t-test. $$+/-t_{(n-1,\frac{\alpha}{2})}$$ $$+/-t_{(14,0.0125)}$$ $$+/-2.864$$

My question: Is this the correct way to do it? Being given the variance is making me question myself and I am left wondering if I am misunderstanding the question.

1

There are 1 best solutions below

0
On

Without knowing the hypothesis being tested or the distribution of the test statistic under the null hypothesis, it's unreasonable to state the critical value of the test, so I consider this question ill-posed. For example, without knowing whether the test is one-tailed or two-tailed, the critical value is impossible to state. Even the way the sample was collected will affect the choice of test.

That said, if we assume that the test is for a location parameter and is two-tailed; e.g., $$H_0 : \mu = \mu_0 \quad \text{vs.} \quad H_a : \mu \ne \mu_0,$$ and a simple random sample of size $n = 15$ was taken, and that the observations are assumed to have been drawn from a distribution with known variance $\sigma^2 = 15$, then the test statistic $$Z \mid H_0 = \frac{\bar x - \mu_0}{\sigma/\sqrt{n}}$$ is approximately standard normal (unless the observations were drawn from a normal distribution, in which case $Z \mid H_0$ is exactly standard normal).

This test statistic will reject $H_0$ at $\alpha = 0.025$ if $|Z| > z^*_{\alpha/2}$, where $\Phi(z^*_{\alpha/2}) = 1 - \alpha/2 = 0.9875$; that is to say, $z^*_q$ is the upper $q^{\rm th}$ quantile of the standard normal distribution. Using a computer or statistical table, $$z^*_{0.0125} \approx 2.2414.$$ This is the critical value of the test.

Note that this is where the two-tailed nature of the hypothesis test comes into play. Had the test been one-tailed, e.g. $$H_a : \mu > \mu_0,$$ then you would reject $H_0$ in favor of $H_a$ if $Z > z^*_{\alpha}$, with no absolute values, and the critical value is now $$z^*_{0.025} \approx 1.95996.$$

And if the test were one-tailed in the other direction, $$H_a : \mu < \mu_0,$$ then your critical value would be $z_{0.025} = -1.95996$, and you would reject $H_0$ if $Z < z_{0.025}$.

So as you can see, I have insufficient information to uniquely identify the appropriate critical value for your test. One thing I can say is that a $t$-test is not appropriate if the variance is assumed to be known.