0

Suppose an event takes place, and there is a parameter $\theta$ with $\mathbb E[\theta] = \mu$. Suppose I take a sample to estimate the parameter by taking the average of sample $\bar X$ (i.e. $\bar X$ is a statistic). $I$ is an interval carefully chosen from our sample, and $H_0$ is a null hypothesis that $\mu \in I$, while $H_1$ is $\mu \not\in I$. Note that either $\mu \in I$ or $\mu \not\in I$, so $p$ is either $1$ or $0$. Please refer to the diagram below:

enter image description here

In Statistic I, this is what I have learned: if $\bar X \not\in I$, then I reject $H_0$. This methodology is straightforward, but I don't understand the justification that I need to reject $H_0$ if $\bar X \not\in I$ (if $p$ is either 1 or 0).

If I observe that $\bar X \not\in I$, then I reject $H_0$. This probability is given as $$\mathbb P(H_0 = 0 \mid \bar X \not\in I) = \frac{\mathbb P(H_1, \bar X \not\in I)}{\mathbb P(\bar X \not\in I)} = \frac{(1-\beta)(1-p)}{\alpha p + (1-\beta)(1-p)}. $$

Now, $p \in \{0, 1\}$. Very trivially, \begin{equation*}\mathbb P(H_1 = 1 \mid \bar X \not\in I) = \begin{cases} 0 \quad\quad \text{if $p = 1$} \\ 1 \quad\quad \text{if $p = 0$} \end{cases} \end{equation*}

Also,

\begin{equation*}\mathbb P(H_0 = 1 \mid \bar X \not\in I) = \begin{cases} 1 \quad\quad \text{if $p = 1$} \\ 0 \quad\quad \text{if $p = 0$} \end{cases} \end{equation*}

In other words, since $p \in \{0, 1\}$, the thing called "hypothesis testing" becomes trivial. Hence, is it really true that $p$ is either 1 or 0? In hypothesis testing, we are making an educated guess whether we embrace $H_0$ (or fail to reject it) or reject $H_0$. We are comparing the likelihood of $H_0$ being false or not, and if it is more likely that $H_0$ is false, then we reject it. Hence, we are assigning some probability value on $H_0$ being true or not, meaning $p$ should not be either 0 or 1. Can someone clarify my confusion?

James C
  • 141
  • Please investigate our threads on hypothesis testing and p-values. You might begin with https://stats.stackexchange.com/questions/31. There's a lot of irrelevant stuff in this formulation, starting with the introduction of the mysterious "$p,$" which doesn't seem to provide any useful information about the situation. – whuber Jan 21 '23 at 21:57
  • If you are choosing $I$ based on your sample, the hope is that your method of producing a confidence interval has a desired probability of covering the unknown population parameter. If that parameter is the population mean and you are basing your interval on the sample mean, then typically the confidence interval will start below the sample mean and end above it so the sample mean is always in the confidence interval (and with some but not all methods exactly in the middle of the confidence interval by construction) – Henry Jan 21 '23 at 22:48

0 Answers0