2
$\begingroup$

If I have two Gaussian distribution with the same $\sigma$ but different $\mu_x$ and $\mu_y$. How to calculate the $P(x>y\ |\ \mu_x\leq\mu_y)$? I think it's a type 1 error when I set the event "$\mu_x>\mu_y$" as null hypothesis?

$\endgroup$
1
  • $\begingroup$ Another way to interperate is:what is the minimum difference between $\mu_x$ and $\mu_y$ so that we can conclude they are significantly different? So should I calculate the distribution $z = x-y$ to see if the confidence intervals of z exlude 0? $\endgroup$
    – dexter2406
    Apr 6, 2020 at 19:07

1 Answer 1

1
$\begingroup$

Since $\mu_x,\mu_y$ are constants (just like $\sigma$. If they're not, provide their distributions), the event in the given side has no additional information. It's like $1\leq 2$. So, $P(X>Y|\mu_x\leq \mu_y)=P(X>Y)$.

You'd probably wonder "if mean of x is less than y, the probability of X being larger than Y should be smaller, so it actually has information". This is already reflected in the final formula.

$T=X-Y$ will be another Gaussian RV with mean $\mu_x-\mu_y$ and variance $2\sigma^2$ if assumed independent. If $X$ and $Y$ are dependent, you need to provide the joint distribution. Then,

$$P(X-Y>0)=P(T>0)=P\left(Z>\frac{0-(\mu_x-\mu_y)}{\sqrt2\sigma}\right)=\Phi\left(\frac{\mu_x-\mu_y}{\sqrt 2\sigma}\right)$$

Where $\Phi(x)$ is the CDF of standard Gaussian RV. See that if $\mu_x\leq\mu_y$, the expression inside $\Phi$ will be negative and the probability will be smaller than $0.5$, while if $\mu_x\geq \mu_y$, it'll be positive and the probability will be larger than $0.5$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

Not the answer you're looking for? Browse other questions tagged or ask your own question.