I have N measurements whose uncertainties are unknown. What is the variance?

36 Views Asked by At

I measured something $N$ times using different measurement techniques. Each measurement technique $i$ has a known variance $\sigma_i^2$. So every measurement is $x_i = \hat x + \epsilon_i$ where $\hat x$ is the true value and $\epsilon_i$ is pulled from a normal distribution with mean 0 and variance $\sigma^2_i$.

Ok, I have all these measurements. Now I want to know the variance of my measurements taken together, taking into account the fact that I know the variance of all the measurement techniques. What's the best way to do this?

1

There are 1 best solutions below

9
On

You can just treat them as sum of normally distributed variables, which is extensively written down in Wikipedia.

In short, if $X \sim N(x,\sigma_x^2)$ and $Y \sim N(y,\sigma_y^2)$ and $X,Y$ are independent, then $$X+Y \sim N(x+y, \sigma_x^2 + \sigma_y^2)$$


In your case you have the measurements $X_i \sim N(x, \sigma_i^2)$, but you don't know the actual value $x$ (which is why you're doing the measurements.) Now In order to find $x$, you probably want to take the average of all of them, that is $\hat x = \frac{1}{n} \sum_{i=1}^n X_i$

If we now treat $\hat x$ as a random variable, we can determine its variance by using the formula above, and we get:

$$n \hat x = \sum_i X_i \sim N(nx, \sum_i \sigma_x^2)$$

Now as a last step you have to "solve" for $\hat x$ which is not difficult. (Use that $Var(aX) = a^2 Var(X)$, and $E[aX] = aE[x]$)