Re: Calculating the Standard Deviation of a Weighted List
From: Ray Koopman (koopman_at_sfu.ca)
Date: 11/05/04
- Next message: romona: "Multi-parameter sufficient statistic"
- Previous message: George Kahrimanis: "Re: chi-square test goodness of fit"
- In reply to: John: "Calculating the Standard Deviation of a Weighted List"
- Messages sorted by: [ date ] [ thread ]
Date: 5 Nov 2004 14:20:44 -0800
John wrote:
> When calculating the standard deviation of a weighted list, should
> one normalize the weights such that the sum of the weights = 1, or
> such that the sum of the weights = the number of data points proper?
>
> The latter seems to be the only way by which the weighted standard
> deviation, when the weight variable has the same value for every data
> point, equals the non-weighted standard deviation.
>
> Yet nearly everything I find on the internet seems to suggest the
> former.
>
> BTW, I'm not a statistician by any stretch so please forgive any
> ignorance.
>
> TIA,
> John
Normalizing the weights is not necessary.
The weighted mean is
m = Sum Wi*Xi / Sum Wi.
The weighted variance that corresponds to using N
in the denominator of the unweighted variance is
Sum Wi(Xi - m)^2 / Sum Wi.
The weighted variance that corresponds to using N-1
in the denominator of the unweighted variance is
Sum Wi(Xi - m)^2 / (Sum Wi - (Sum Wi^2 / Sum Wi)).
- Next message: romona: "Multi-parameter sufficient statistic"
- Previous message: George Kahrimanis: "Re: chi-square test goodness of fit"
- In reply to: John: "Calculating the Standard Deviation of a Weighted List"
- Messages sorted by: [ date ] [ thread ]
Relevant Pages
|