I have the following problem:

  • a measurement produces a certain value $x_i$ and I have an associated error $\Delta x_i$ for every measurement;
  • the $i$ measurements are not independent and there are many of them $i \in [0, N]$ with $N \approx 2000$;

I would like to calculate the standard error of the (weighted) mean $_w\mu(x_i, w_i)$, defined as:

$$_w\mu(x_i, w_i) = \frac{\sum_i w_i x_i}{\sum_i w_i}$$

where the weights are given by $w_i = 1 / \Delta x_i^2$.

Having defined the weighted variance $_w\sigma(x_i, w_i)^2$ as:

$$_w\sigma(x_i, w_i)^2 = \frac{\sum_i w_i^2 (x_i - _w\mu)^2}{\sum_i w_i^2}$$

I assume that the standard error cannot be simply obtained dividing $_w\sigma(x_i, w_i)$ by $\sqrt{N}$. To the very least I should calculate the correction for the correlation in the sample and Wikipedia suggests to estimate/calculate a correction factor $f$ defined as:

$$f = \sqrt{\frac{1+\rho}{1-\rho}}$$

where $\rho$ is an estimate of the autocorrelation coefficient.

The standard error on the weighted mean $\Delta _w\mu(x_i, w_i)$ would then become:

$$\Delta _w\mu(x_i, w_i) = \frac{_w\sigma(x_i, w_i)}{\sqrt{N}} f$$

However, I have trouble estimating $\rho$.

What I have tried was to calculate the autocorrelation function $R(\tau)$ (normalized to the variance), but I do not understand how could I compute $\rho$ from it, and what modification would be required to account for the different weights.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.