Marat Khalili wrote:
Lewis Hyatt wrote:
Note, in particular, that this error law would imply no benefit whatsoever to averaging the results from many observations together... you would get the same error as you would from any one trial.
On a positive note, is this what you need for your work? Do you need function: estimation_from_statistics(list<T> &statistics) returning average and deviation? I bet one exists already, but I can add one to the library. What do you do next with the value you receive? Do you substitute it somewhere?
Well, the boost accumulators library offers a lot of things to do this kind of calculation, but being able to keep track of the errors automatically would be nice too. Like others have pointed out, it's very hard to do this in complete generality, allowing for arbitrary correlations between all numbers. I just wanted to point out that when you see something like x +- dx, that carries a specific meaning to most people, which is that dx is roughly the 1-sigma error on x, or the standard deviation. It is true that if x and y are correlated 100%, then the error on their sum is the sum of their errors, but that is a pretty unusual case.