--
Ilana,
I think that the most likely reason for the similarity of the sample
and population totals is that someone normalized the original weights
so that the revised weights would sum to the sample size. I think
there is a reference to this practice in Lu Ann Aday (1996) Designing
& Conducting Health Surveys, Second Edition, Jossey-Bass.
This would be done by multiplying each original weight by (sample
size/population size). Rounding error would account for the small
difference between the sum of the revised weights and the true sample
size These revised weights are obviously useless fo estimating
population group totals but would give estimates of means,
proportions, and regression coefficients identical to those given by
the original weights.
-Steve
On Mon, May 4, 2009 at 6:00 PM, <[email protected]> wrote:
> --
>
> Ilana, The estimated population size is the sum of the weights. The
> two totals are close, so it doesn't look like a wrong variable.
> Weights can be less than 1 and you obviously have some. You would have
> to consult the study documentation to see how if this is possible. In
> telephone surveys, for example, they can arise when the same
> respondent could have been identified through multiple telephones. In
> such a case the base weight is divided by the number of phones.
> Perhaps something like that has happened.
>
> -Steve
>
>
> On Sun, May 3, 2009 at 1:41 PM, Akresh, Ilana Redstone
> <[email protected]> wrote:
>> I am using Stata 9 and have svyset my data with a pweight. The number of observations is 8572. When I run a weighted mean (svy: mean x) I get a population size that is smaller than the sample size (population size = 8545). How is this possible? Is there something wrong with the weight or with how I have svyset the data or something else?
>>
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
