0
$\begingroup$

I have 2 variables, which I can't be sure to call them random variables. They are aggregates from multiple measurements of two distinct phenomena. I don't have access to the raw measurements data, just the aggregates: Var A with mean $\mu_1$ and standard deviation $\sigma_1$, and Var B with mean $\mu_2$ and standard deviation $\sigma_2$.

I want to calculate Var C which given by the following formula:

$$C=\frac{a}{b}+2$$

I'd imagine for Var C's mean $\mu_3$, this boils down to

$$\mu_3=\frac{\mu_1}{\mu_2}+2$$

How about Var C's standard deviation $\sigma_3$? Is it even possible to calculate Var C mean and standard deviation given only the means and standard deviations for Var A and B?

Thanks

$\endgroup$
  • $\begingroup$ Perhaps in your case $μ_3≈μ_1/μ_2 + 2$ would be close to correct. Certainly not as a reliable general rule. For example, if $μ2\approx 0,$ then this method might give a useless result. (Same as previous Comment, but with typo in notation repaired.) $\endgroup$ – BruceET Dec 20 '17 at 7:41
0
$\begingroup$

Comment continued, with specific examples.

Suppose $X_1 \sim \mathsf{Norm}(\mu_1=200,\, \sigma_1=20)$ and $X_2 \sim \mathsf{Norm}(\mu_2=150,\, \sigma_1=20).$ Then, by simulation, $E(X_1/X_2) +2 \approx 3.36$ and $\mu_1/\mu_2 + 2 = 3.33.$ Not a horrible approximation.

x1 = rnorm(10^6, 200, 20)
x2 = rnorm(10^6, 150, 20)
mean(x1/x2) + 2
## 3.358685  # aprx E(X1/X2) + 2
200/150 + 2
## 3.333333

However, if $X_1 \sim \mathsf{Norm}(\mu_1=20,\, \sigma_1=10)$ and $X_2 \sim \mathsf{Norm}(\mu_2=15, \sigma_1=10).$ Then $E(X_1/X_2) +2$ is unstable because you are dividing by values very near 0 some of the time. (Theoretically speaking, $E(X_1/X_2)$ is undefined.) Here, $3.33$ is not a useful approximation.

$\endgroup$

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.