0
$\begingroup$

In hypothesis testing, we set up a rejection area for rejecting $H_0$ in favor of $H_1$ with $\alpha$. I don't understand why type I error (rejecting $H_0$, when $H_0$ is actually true) is the area that we choose to reject $H_0$. Why does it makes senses that we reject something from error rates?

$\endgroup$
5
  • $\begingroup$ If the null hypothesis is correct, then a rejection (i.e. being in the rejection area) is a Type I error, while a non-rejection (i.e. not being in the rejection area) is not $\endgroup$
    – Henry
    Jan 21 at 10:24
  • $\begingroup$ Then why rejecting null hypothesis when it is correct constitutes an acceptance of $H_1$ when the null hypothesis is actually correct? Isnt that an error? $\endgroup$ Jan 21 at 10:31
  • $\begingroup$ Yes - and that Type I error is precisely what happens when the null hypothesis is correct and the observation is in the rejection region. $\endgroup$
    – Henry
    Jan 21 at 10:34
  • $\begingroup$ Then why we accept an alternative hypothesis if the null hypothesis is in fact correct? To my understanding, it is just weird that we made an error (type I) and we still think alternative hypothesis is correct. I just couldnt wrap my head around this. $\endgroup$ Jan 21 at 10:45
  • $\begingroup$ You do not know which hypothesis is correct: you use the observations to make a decision, knowing that it is possible that you may make an error when you make that decision. $\endgroup$
    – Henry
    Jan 21 at 11:03

1 Answer 1

0
$\begingroup$

by wiki,

a type I error is the mistaken rejection of an actually true null hypothesis

$\alpha$ has the same value as type 1 error but you can distinguish them by going thru the following story.

In your hypothesis test of recovery rate of a drug, you first assume your $H_0$ is correct, which means the drug gives you the same recovery rate as not using the drug. In this case, you assume the recovery rate distribution of the drug is the same as the distribution of not using drug.

Then you calculate the average recovery rate of patients using the drug, and find a value $r$. Then you look at where $r$ is in the $H_0$ distribution. and you calculate the $\alpha$ by summing the area under the distribution $>=r$, so this $\alpha$ value is actually the total probability of observing a recovery rate $>=r$ assuming that $H_0$ is correct. Let's say this value is 0.021, meaning that if $H_0$ is true, you have 2.1% chance observing $r$ or greater than $r$. It is quite unlikely judging from the size of the value, but it is NOT IMPOSSIBLE.

Then you need to decide whether you want to reject $H_0$, you can reject it like a dictator or you can reject it by saying that "OK, now $\alpha$ is smaller than a threshold which I decided to be 0.05, so let's reject $H_0$.

Now we finally introduce "type 1 error".

We can reject $H_0$ because we think the chance of observing $r$ is small, so it is merely not likely, but not impossible. Therefore we can be wrong!

If we are wrong, then $H_0$ should actually be true, and how likely it is that we are wrong (error)? It is how likely it is for us to observe $r$ or greater than $r$, which is $\alpha$, which is 0.021. Therefore the error value has the same value of $\alpha$, and this error is named as type 1 error.

$\endgroup$

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.