1
$\begingroup$

I observe a variable $\hat{x}$. I know that my observation is noisy due to some Gaussian measurement noise $\mathcal{N}(0, \sigma)$.

My goal is to compute the probability $p(x > 0\mid\hat{x})$ that the true underlying $x \in \mathbb{R}$ is positive. The a priori probability of any $x$ is unknown, i.e. the prior distribution is assumed to be flat. Also assume I know $\sigma$ precisely. What is $p(x > 0\mid\hat{x})$?

$\endgroup$
2
$\begingroup$

We can find the posterior distribution of $x$:

$$p(x|\hat x)\propto p(\hat x|x)p(x)\propto\exp\left[-\dfrac{(x-\hat x)}{2\sigma^2}\right]\Rightarrow x\sim\mathcal N(\hat x, \sigma^2)$$

Since $x|\hat x$ is normal, we can easily calculate $p(x>0|\hat x)$ using the standard normal CDF, $\Phi$:

$$p(x>0|\hat x)=1-\Phi\left(\dfrac{0-\hat x}{\sigma}\right)=\Phi\left(\dfrac{\hat x}{\sigma}\right)$$

Notice that, in this case, this happens to be the same as one minus the p-value for the frequentist Z-test for $x>0$.

$\endgroup$
1
  • $\begingroup$ Thanks, much appreciated! $\endgroup$ – monade Sep 30 '20 at 19:42

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.