Statistical uncertainty of weighted mean


by Brais
Tags: statistical, uncertainty, weighted
Brais
Brais is offline
#1
Nov9-12, 11:56 AM
P: 7
Hello!

I am using physical data to do an analysis (~30k measurements). These measurements include energies, momenta, angles... of particles.
I am calculating a value (call it v) at the end after a lengthy process, and if I introduce all the data into my program I did, the result is vσ.
If, however, I "bin" my events in energy and angle (say I made four bins in total), when I calculate "v", I get v1σ1, v2σ2, v3σ3, v4σ4. Then I combine these values into one using a weighted average (c stands for combined): [itex] v_c = \sum(v_i/\sigma^2_i)/\sum(1/\sigma^2_i)[/itex], and [itex]\sigma_c = 1/\sqrt{\sum(1/\sigma^2_i)}[/itex] (as can be seen here).
When I do this, it turns out that [itex]\sigma_c \simeq \sigma/2[/itex]. How can this be? I am using the same amount of statistics!

Any reply or idea will be very welcome!!

Thank you!

Brais.
Phys.Org News Partner Science news on Phys.org
NASA's space station Robonaut finally getting legs
Free the seed: OSSI nurtures growing plants without patent barriers
Going nuts? Turkey looks to pistachios to heat new eco-city
chiro
chiro is offline
#2
Nov9-12, 05:52 PM
P: 4,570
Hey Brais.

The wiki article looks straight-forward, but perhaps you could just outline your calculations in a little more detail step by step to show the simplifications and assumptions you used.
Brais
Brais is offline
#3
Nov10-12, 04:49 AM
P: 7
Hi, thanks for your reply!
I am calculating a fit. If I put all my data together I get an error that is higher than that of fitting different sets of points separately and then combining them with a weighted mean.
I didn't do any simplification, just applied the expression seen in wikipedia.

Brais

Stephen Tashi
Stephen Tashi is offline
#4
Nov10-12, 09:28 AM
Sci Advisor
P: 3,175

Statistical uncertainty of weighted mean


Quote Quote by Brais View Post
I didn't do any simplification, just applied the expression seen in wikipedia.
Brais
You are using a formula from that article that applies when you know the standard deviations of the distributions that are involved, but I'd guess that you don't. Your [itex] \sigma_i [/itex] are probably estimators of standard deviations that you computed from the sample. (The term "standard deviation" is ambiguous. It has at least 5 different meanings in statistics, depending on the context where it appears.)
chiro
chiro is offline
#5
Nov10-12, 06:49 PM
P: 4,570
Following Stephen Tashi's post, you should probably just clarify exactly what attribute you are using.
Brais
Brais is offline
#6
Feb1-13, 09:32 AM
P: 7
Hi again!

A long time ago I had to stop this analysis and so my doubt wasn't importantr for some time :)
I use the errors that my minimization algorithm "MINUIT" gives. Unfortunatelly I cannot find anything except that it (obviously) calculates a covariance matrix and error matrix...

Thanks,

Brais


Register to reply

Related Discussions
Bevington and uncertainty of a weighted mean Set Theory, Logic, Probability, Statistics 5
Is the Uncertainty Principle Statistical? And if so... Quantum Physics 16
Statistical Analysis Question: How many measurements to reduce uncertainty? Introductory Physics Homework 0
Statistical Uncertainty for Discrete Events Set Theory, Logic, Probability, Statistics 2
Statistical form of Heisenberg Uncertainty Quantum Physics 2