Why would you weight on multiple variables? (For that matter, why would you use multiple regressors, when you can avoid collinearity problems by sticking to a single one?) Why would you expect weights to change the distribution of anything?
I'm reluctant to interfere with the comic value of this thread, but:
--Some of this would have been better dealt with at the selection rather than the analysis stage.
--But, given the sample you have to work with, the right question is ignorability of selection given a set of covariates, not whether Internet users or panel members are unconditionally different from non-users or non-members. (The AAPOR report is confused on this point, among many others.) It's unlikely that one or two covariates are enough to make the selection ignorable.
--There's been quite a bit of work in recent years on better methods than cell weighting or raking for dealing with these problems (calibration, propensity scores, multilevel regression post-stratification).
On Jan 5, 2011, at 7:51 AM, Lonna Atkeson wrote:
> I just wanted to take this discussion back a little bit. Why would you weight an Internet poll to population parameters? Given what we know about the electronic divide, why would we expect an Internet sample to look like the population? And, why would weighting it make it representative of the population?
>
> This creates an opportunity for me two note two somewhat recent reports from American Association for Public Opinion Research on the changing nature of survey research. One on cell phones and the other on opt-in Internet panels. Both are available at: http://www.aapor.org/AAPOR_Committee_and_Task_Force_Reports/2820.htm.
>
> Happy New Year!
>
> Lonna
>
>
> ---
> Professor Lonna Atkeson
> Department of Political Science
> University of New Mexico
> Albuquerque, NM 87131
> Phone: 505-660-8976 (mobile)
> www.unm.edu/~atkeson
>
>
>
>
>
> On Jan 4, 2011, at 11:55 PM, Matthew DeBell wrote:
>
>> A set of step-by-step instructions for creating weights (tailored for the ANES, but applicable to other surveys too) is available here:
>>
>> http://www.electionstudies.org/resources/papers/nes012427.pdf
>>
>> --
>> Matthew DeBell, Ph.D.
>> Director of Stanford Operations
>> American National Election Studies
>> Stanford University
>>
>> [log in to unmask]
>> 650-725-2239
>> www.electionstudies.org
>>
>>
>>
>> On 1/4/2011 1:34 PM, Political Methodology wrote:
>>> ---------- Forwarded message ----------
>>> Date: Tue, 4 Jan 2011 16:27:51 +0000
>>> From: "Gries, Peter H." <[log in to unmask]>
>>> To: "[log in to unmask]" <[log in to unmask]>
>>> Subject: Q: how to create "weight" variable for SPSS
>>>
>>> Under "data" in SPSS you can choose to "weight cases" but must first
>>> have created a "frequency variable" to weight cases by. Can anyone
>>> direct me to information about how to create this frequency variable? I
>>> have a nice national sample gathered on the Internet, but it is heavy on
>>> Democrats, and I'd like to weight Republicans more highly to reflect
>>> national averages, as well as balance other demographic variables like
>>> gender while I'm at it...
>>>
>>> Cheers and thanks in advance for any help anyone can provide, on or
>>> offline.
>>>
>>> Pete Gries
>>>
>>> Harold J. & Ruth Newman Chair in US-China Issues
>>> Director & Associate Professor
>>> Institute for US-China Issues
>>> The University of Oklahoma
>>> 729 Elm Ave.; Hester Hall, Room 120
>>> Norman, Oklahoma, 73019-2105
>>>
>>> www.ou.edu/uschina
>>> <http://www.ou.edu/uschina>PHONE: 405/325-1962; FAX: 405/325-7738
>>>
>>> **********************************************************
>>> Political Methodology E-Mail List
>>> Editors: Diana O'Brien <[log in to unmask]>
>>> Jon C. Rogowski <[log in to unmask]>
>>> **********************************************************
>>> Send messages to [log in to unmask]
>>> To join the list, cancel your subscription, or modify
>>> your subscription settings visit:
>>>
>>> http://polmeth.wustl.edu/polmeth.php
>>>
>>> **********************************************************
>>>
>>
>> **********************************************************
>> Political Methodology E-Mail List
>> Editors: Diana O'Brien <[log in to unmask]>
>> Jon C. Rogowski <[log in to unmask]>
>> **********************************************************
>> Send messages to [log in to unmask]
>> To join the list, cancel your subscription, or modify
>> your subscription settings visit:
>>
>> http://polmeth.wustl.edu/polmeth.php
>>
>> **********************************************************
>
> **********************************************************
> Political Methodology E-Mail List
> Editors: Diana O'Brien <[log in to unmask]>
> Jon C. Rogowski <[log in to unmask]>
> **********************************************************
> Send messages to [log in to unmask]
> To join the list, cancel your subscription, or modify
> your subscription settings visit:
>
> http://polmeth.wustl.edu/polmeth.php
>
> **********************************************************
Douglas Rivers
Professor of Political Science
Dept. of Political Science
Stanford University
Stanford, CA 94305-6049
(650) 723-2612
[log in to unmask]
**********************************************************
Political Methodology E-Mail List
Editors: Diana O'Brien <[log in to unmask]>
Jon C. Rogowski <[log in to unmask]>
**********************************************************
Send messages to [log in to unmask]
To join the list, cancel your subscription, or modify
your subscription settings visit:
http://polmeth.wustl.edu/polmeth.php
**********************************************************
|
