Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, statalist.org is already up and running.


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Standard error of the weighted mean


From   Steve Samuels <sjsamuels@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: Standard error of the weighted mean
Date   Sat, 4 May 2013 19:04:07 -0400

Nick:


> On May 3, 2013, at 1:03 PM, nick bungy wrote:
> 
> 
> I am having issue finding the correct way to calculate the standard error of a probability weighted mean. I can successfully calculate the sample standard deviation, as taken from:
>  
>  
> s2 = {n/[W(n - 1)]} sum wi (xi - xbar)2
>  
> The Stata article suggests that the standard error of the weighted mean is then simply:
>  
> V_srs = s2/n
>  
> Where V_srs is an ‘estimate of the variance of muhat [mean] assuming a simple random sample of the same number of observations’
>  


The article doesn't suggest that at all.

The first formula shows how to estimate the population variance from a
weighted sample. The second formula gives the variance (squared SE) for
the mean of a simple random sample without replacement a mean that is not 
weighted.

As the article states, the standard error for the weighted mean "is
pretty complicated; see [SVY] variance estimation for details." 
Formula 1 (p. 180) of the Stata 12 Survey manual, specified to
a single stratum, shows that the formula involves the square of
the weights and has no connection or the population variance or
to the formula for s2 above.


Steve



> However, when I calculate V_srs and compare it to the linearized standard error as taken from the svy: commands, the linearized standard error is significantly larger.
>  
> Since my calculations and Stata agree on s^2, presumably I must be calculating V_srs (variance of the mean) incorrectly.
>  
> Can anyone shed some light on where I am going wrong? Here's an example of the disparity between my mata code and the official svy command.
>  
> sysuse auto
> gen pweight = trunk^2 + 1
>  
> mata
> st_view(y=. ,. , "mpg")
> st_view(weight =. ,., "pweight")
>  
> Weighted_mean = colsum(y:*weight) / colsum(weight)
> Weighted_var = rows(y)/(colsum(weight)*(rows(y)-1))*colsum(weight:*(y :- Weighted_mean):^2)
> Weighted_sd = sqrt(Weighted_var)
> SD_Mean = sqrt(Weighted_var / rows(y))
>  
> Weighted_sd
> SD_Mean
>  
> end
>  
> svyset [pweight = pweight]
> svy: mean mpg
> estat sd
>  


*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/


© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index