bin_data when errors are provided - problem in error formula?

[melissa-hobson]

I think there may be an error in the computation of the binned errors when errors are provided. The current formula is:

            y_err_bins.append(
                np.sqrt( np.sum( yerr[i: i + n_bin - 1]**2 ) / len( yerr[i: i + n_bin - 1] )  )
                            ) 

But, let's say we're using the mean because then the error propagation is straightforward. The mean is the sum of the points, divided by the number of points (which is an integer, with no error). Then by error propagation, the error of the sum is the square root of the sum of the errors squared; and we divide by the number of points to get the error of the mean. So, essentially, I think in the formula above, the square root is wrapping the number of points when it shouldn't, and the formula should be:

            y_err_bins.append(
                np.sqrt( np.sum( yerr[i: i + n_bin - 1]**2 ) ) / len( yerr[i: i + n_bin - 1] )  
                            ) 

(I admit I'm not sure about the median...)

[nespinoza]

Hi @melissa-hobson,

You are correct! This must have been a typo. Fixing on the next juliet version.

N.

[nespinoza]

Fixed on latest version (2.2.6).

N.