Thomas, the weights computed in your way are proportional to the
standard weights. However the standard computation is more
informative--it shows many people represented by each sample member
(in many instances, the weights differ from person-to-person, not just
from stratum to stratum). The hallmark of standard weights is that, if
summed over the sample, they estimate the population size. The
standard weights do that. Also, weights are often computed when the
population N's are not known. Your version requires that the N's be
known in advance.
To use survey weights in Stata, you will have to use the -svyset-
command, so study that also. The documentation will tell you which
weight you should designate.
-Steve
On Wed, Sep 9, 2009 at 4:21 AM, Thomas Klausch<[email protected]> wrote:
> Hi Steve,
>
> Many thanks for your advice. I will check out the references you gave.
>
> I was suprised about the weighting that you describe, though. As far
> as I was aware a proper weight is computed as
>
> weight= % in population / % in sample
>
> For my example (large stratum) I would then have
>
> % in population = 7900/8800=.898
> % in sample = 600/1200=.5
>
> weight= .898/.5 =1.797
>
> The same follows for the small stratum.
>
> Now you suggest N/n= 7900/600=13.17 ?
>
> Is that correct? I wouldn't see the point yet...
>
> Maybe you or somebody else also knows the answer to my question,
> whether in STATA -iweight- option is the correct one to use.
>
> Many thanks.
>
> Best
> Thomas
>
>
>
>
> 2009/9/8 <[email protected]>:
>> --
>> Thomas, I want to more directly answer part of your question. I
>> believe you have two strata. The basic sample weight for observations
>> in a stratum is : N/n where N is number of population elements in
>> the stratum, and n = the number in the sample. However if you know
>> other information about the population other than that used to form
>> the strata , you can use it to improve the weights. See sections on
>> post-stratification in the books I referenced.
>>
>> Steve
>>
>>
>> On Tue, Sep 8, 2009 at 12:29 PM, <[email protected]> wrote:
>>> Thomas Your knowledge about weights is faulty--weights are usually
>>>>1, for an observation's weight is the number of population members
>>> represented by the observation (the weight can be non-integer).
>>> Weights <1 do arise, but not in the sample that you are describing.
>>> Before going any further. I suggest that you read Sharon Lohr.
>>> Sampling: Design and Analysis, Duxbury, 1999). Also, look at her
>>> section on sample size calculations. The particular sample size
>>> calculation will depend on the purpose of your survey--whether
>>> descriptive or analytic. If the latter than do not use the finite
>>> population correction. If you know a lot of about your population,
>>> you can also apply post-stratification techniques.
>>>
>>> The book "Sampling of Populations" by Levy and Lemeshow (Wiley) has
>>> some Stata examples, although they are limited to descriptive samples.
>>>
>>> Regards,
>>>
>>> -Steve
>>>
>>>
>>> On Sat, Sep 5, 2009 at 11:26 AM, Thomas Klausch<[email protected]> wrote:
>>>> Dear list members,
>>>>
>>>> I am planning to survey a stratified sample from a relatively small finite
>>>> population of which I know the size. I am not too familiar with weighting
>>>> techniques and the STATA -svy- command family, which is why I write. Maybe
>>>> somebody can give useful advice.
>>>>
>>>> My popluation consists of size n=8800, with n=900 and n=7900 sized relevant
>>>> stratas. I decided to sample n=600 from each of the STRATAs since I am
>>>> particularly interested in estimates from the smaller strata.
>>>>
>>>> My knowledge about weighting tells me that this gives rise to weights of
>>>> approx. .20 for the small strata and approx. 1.80 for the large strata. In
>>>> SPSS I would know how to use the weights function, as there is only one.
>>>> STATA provides several weight options -fweight-, -pweight-, -aweight- and
>>>> -iweight-. I have tested in STATA that it gives the same estimates using
>>>> -iweight- than SPSS does using the WEIGHT BY command.
>>>>
>>>> My first question is if -iweight- is reasonable to use for my case when
>>>> estimating regression models (in particular logistic panel models).
>>>> My second question is whether there is a, maybe better, way to use the
>>>> options provided in the -svy- command family to specify the survey design.
>>>> maybe there is also a third way I do not know of.
>>>>
>>>> If anybody could give advice or further reference for detailed information
>>>> on -svy- I'd appreciate it a lot.
>>>>
>>>> Many thanks
>>>>
>>>> Thomas
>>>>
>>>> *
>>>> * 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/
>>>>
>>>
>>>
>>>
>>> --
>>> Steven Samuels
>>> [email protected]
>>> 18 Cantine's Island
>>> Saugerties NY 12477
>>> USA
>>> 845-246-0774
>>>
>>
>>
>>
>> --
>> Steven Samuels
>> [email protected]
>> 18 Cantine's Island
>> Saugerties NY 12477
>> USA
>> 845-246-0774
>>
>> *
>> * 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/
>>
>
> *
> * 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/
>
--
Steven Samuels
[email protected]
18 Cantine's Island
Saugerties NY 12477
USA
845-246-0774
*
* 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/
