Peter Wielhouwer <[email protected]> :
You can also use a geographic id as a clustering variable if PSU is
unavailable--or even if PSU is available and PSUs are nested within
geography and you think there is clustering by geography.
. u anes2008_2009panel_dataset, clear
. svyset [pw=wgtbase]
pweight: wgtbase
VCE: linearized
Single unit: missing
Strata 1: <one>
SU 1: <observations>
FPC 1: <zero>
. svy:ta w10p4, ci
(running tabulate on estimation sample)
Number of strata = 1 Number of obs = 619
Number of PSUs = 619 Population size = 610.10636
Design df = 618
-------------------------------------------------
w10p4. |
Favor or |
oppose |
raising |
taxes on |
incomes |
over |
200k/yr | proportions lb ub
----------+--------------------------------------
1, Favor | .381 .3393 .4245
2, Oppos | .0682 .0496 .0931
3, Neith | .5508 .5066 .5943
|
Total | 1
-------------------------------------------------
Key: proportions = cell proportions
lb = lower 95% confidence bounds for cell proportions
ub = upper 95% confidence bounds for cell proportions
. egen c=group(stratum cdstate), mi label
. svyset c [pw=wgtbase]
pweight: wgtbase
VCE: linearized
Single unit: missing
Strata 1: <one>
SU 1: c
FPC 1: <zero>
. svy:ta w10p4, ci
(running tabulate on estimation sample)
Number of strata = 1 Number of obs = 619
Number of PSUs = 81 Population size = 610.10636
Design df = 80
-------------------------------------------------
w10p4. |
Favor or |
oppose |
raising |
taxes on |
incomes |
over |
200k/yr | proportions lb ub
----------+--------------------------------------
1, Favor | .381 .3383 .4255
2, Oppos | .0682 .0513 .0903
3, Neith | .5508 .5098 .5912
|
Total | 1
-------------------------------------------------
Key: proportions = cell proportions
lb = lower 95% confidence bounds for cell proportions
ub = upper 95% confidence bounds for cell proportions
On Thu, Oct 15, 2009 at 9:22 AM, Nick Winter <[email protected]> wrote:
> The ANES weights should be used as pweights.
>
> In recent studies (eg, the 2004 dataset), ANES supplies a variable that
> indicates "strata" and "psus" that can be used for BRR or Taylor-series
> approaches to variance estimation. (They don't release the complete
> sampling information to prevent possible identification of individual
> respondents -- this is discussed, eg, in the introductory materials for the
> 2004 study documentation:
> http://www.electionstudies.org/studypages/2004prepost/nes04int.txt)
>
> So until the final,complete release of the 2008 data, you are stuck with
> just using the appropriate pweight.
>
> (As an aside, in my experience, taking proper account of the stratification
> and clustering in the ANES datasets has not had dramatic effects on standard
> errors. But of course my prior experience doesn't guarantee your future
> performance....)
>
> - NW
>
> [email protected] wrote:
>>
>> It states that "comprehensive instructions will accompany the full
>> release of the Panel Study Data."
>>
>> I'm not familiar with ANES, but neither the paper Bob found nor the
>> user guide have anything about weights centered at zero. (The paper
>> does recommend that weights be scaled so that they sum to 1.) In
>> fact, weights<0 would be rejected by any survey program.
>>
>>
>> -Steve
>>
>>
>> On Thu, Oct 15, 2009 at 6:07 AM, Peter Wielhouwer
>> <[email protected]> wrote:
>>>
>>> Subject: Re: st: Using the 2008 American National Election Study with
>>> Stata
>>> v.11
>>>
>>> Thanks, Richard. I've been looking through the SVY manual, but what the
>>> ANES
>>> dataset provides are individual weighting variables (post estimation
>>> weights
>>> centered on zero, as described in the paper you pointed me to. I agree
>>> that
>>> it would be nice for ANES to be more specific about how to use Stata well
>>> with the dataset.
>>>
>>>
>>> At 04:20 PM 10/14/2009, Peter Wielhouwer wrote:
>>>
>>>> Is anyone familiar with using the 2008 NES with Stata? I have two
>>>> specific questions:
>>>>
>>>> 1. Which weight command is most appropriate for the data? Based on the
>>>> Stata UG, it seems that the -iweight- syntax is most appropriate, but is
>>>> that correct?
>>>
>>> I am not familiar with the data set, but I would be amazed if
>>> iweights were the way to go. My guess is you want pweights. googling
>>> around found this recent paper:
>>> ftp://ftp.electionstudies.org/ftp/nes/bibliography/documents/nes012427.pdf
>>>
>>> If you google around some more though, maybe you can find something
>>> easier to wade through; it is nice when a data set explicitly tells
>>> you how to set the weights in Stata.
>>>
>>>> 2. In the ANES 2008 user guide, we are advised, "due to the complex
>>>> sample design of the ANES, sampling errors and related statistics
>>>> (including confidence intervals, p-values, t-tests, and all other tests
>>>> of statistical significance) should not be calculated using methods
>>>> intended for simple random samples." In light of this, which would be
>>>> the appropriate statistics to use in Stata?
>>>
>>> I think the correct question is not what statistics should I use, but
>>> what statistical methods should I use to get the correct
>>> statistics. Since you have Stata 11, you should also have the SVY
>>> manual available in pdf form. Just click help/ PDF
>>> documentation. If bookmarks are open then on the left hand side
>>> you'll see the svy manual. After you've gone over the opening
>>> explanatory material, the section on svy estimation will highlight
>>> the many commands you have available. You'll probably want commands
>>> like svy: tabulation, svy: mean, svy: regress, svy: logit, etc.
>>>
>>>
*
* 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/