3
$\begingroup$

I know that a test statistic is used to help us in hypothesis testing, etc. We compute the test statistic, and then compare it to the $\alpha$ value to reject or accept the null hypothesis.

For a normal distribution, this is easy, you just do $ Z = ((X-\mu)\sqrt n)/\sigma $, and all of these are well-defined.

So, if I'd like to find the test statistic of a binomial distribution, I thought that it was a similar process, given the Central Limit Theorem:

$ \mu = np \\ \sigma = \sqrt(np(1-p)) \\ Z = ((X-np)\sqrt n)/\sqrt(np(1-p)) $

So, why when I look up the test statistic, I see sources such as this one under the section "Normal Approximation to the Binomial Distribution", which define the test statistic differently for a binomial distribution?

What is the test statistic used for a binomial distribution?

$\endgroup$
  • 1
    $\begingroup$ 1. What are you testing, exactly? 2. Which test statistic do you mean? There's little point linking to a lengthy document and leaving us to guess which of the many things there you are taking to be a test statistic. $\endgroup$ – Glen_b Feb 14 '16 at 8:09
2
$\begingroup$

I assume that you have $N$ observations that are either ''success'' or ''failure'' and you observe $n$ successes.

If you perform a hypothesis test, then you have some a priori idea about the success probbaility and you want to test whether it is confirmed by the data. So you $H_0$ is that $p=p_0$ where $p_0$ is you a priori value. So $H_0: p=p_0$.

If $H_0$ is true then the above mentioned observation is a random outcome of a Binomial random variable $X \sim Bin(p_0;N)$. Using the Binomial probabilities you can now compute $P(X \ge n)$ and compare this to your significance level, just as with hypothesis tests based on normal random variables.

An alternative is to approximate the Binomial random variable with a normal random variable with mean $Np_0$ and variance $Np_0(1-p_0)$. The approximation works fine for very large $N$ and is better when $p_0$ is close to 0.5.

$\endgroup$
  • $\begingroup$ Here we assume that test statistic is a sum number of successes. $\endgroup$ – Anton Aug 16 '18 at 13:57
1
$\begingroup$

Does this mean that Z would be the test statistic reported in a paper (ie like t-values or F-values)?

$\endgroup$
  • $\begingroup$ This is more a comment than an answer... $\endgroup$ – Joe_74 Sep 9 '16 at 0:32

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.