Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: st: weights in pooled repeated cross sections
From
Steven Samuels <[email protected]>
To
[email protected]
Subject
Re: st: weights in pooled repeated cross sections
Date
Wed, 28 Sep 2011 09:12:00 -0400
I think that the notion of an "average population" is not only useful but often unavoidable, even for single year surveys. Many of these are conducted over a period of months, sometimes, for continuous surveys, over 12 months. As such, the population represented is the "average population" for the time period.
Austin Nichols once suggested that the proper denominator for multi-year surveys should be not persons, but "person-years". This is just another version of the "average population" over the for the survey period.
Even more: many national yearly surveys provide modified weights so that population frequencies of important characteristics (age, gender, ethnic group) match the frequencies in recent censuses.
Steve
On Sep 28, 2011, at 4:36 AM, Ivica Rubil wrote:
Steve, Cameron:
According to the paper that Cameron suggested me to read:
" It is not a lways necessary to adjust
the weights when pooling the dat a .
When weights are adjusted , the
assumption is that they are being
adjusted to properly represent a
population. The problem is that when
we ight s f rom di ff e r ent t ime pe r iods
a r e combined, the r e sul t ing we ight s
do not represent the current population,
but rather, an average population that
does not exist. Consequently, creating
tot a l s wi th a combined f i l e may not
b e a p p r o p r i a t e , wh e t h e r o r n o t t h e
we ight s a r e adjus t ed. On the othe r
hand, r a t ios , propor t ions and me ans
can be regarded as useful statistics when
cons ide r ed a s pe r iod e s t ima t e s . For
these types of statistics, the results using
the original weights or the weights that
have be en adjus t ed us ing a common
factor 1/k (k= # of datasets combined) will give the same result.
This a l so holds for r e gr e s s i o n p a r a m e t e r s ,
whe r e we ight s a r e us ed in the mode l
in order to take the survey design into
account rather than to make estimates
for some f ini t e popul a t ion.
--
Ivica Rubil
Ekonomski institut / The Institute of Economics, Zagreb
Trg J. F. Kennedyja 7, 10 000 Zagreb, Croatia
tel. +385-1-2362-269
fax. +385-1-2335-165
[email protected]
www.eizg.hr
-----Original Message-----
From: [email protected]
[mailto:[email protected]] On Behalf Of Cameron
McIntosh
Sent: 28. rujan 2011 0:17
To: STATA LIST
Subject: RE: st: weights in pooled repeated cross sections
Ivaca,
Strategies for dealing with your type of situation are dealt with in:
Thomas, S., & Wannell, B. (2009). Combining cycles of the Canadian
Community Health Survey. Health Reports,
20(1).http://www.statcan.gc.ca/pub/82-003-x/2009001/article/10795-eng.pd
f
Cam
----------------------------------------
> Subject: Re: st: weights in pooled repeated cross sections
> From: [email protected]
> Date: Tue, 27 Sep 2011 17:30:10 -0400
> To: [email protected]
>
>
> Thanks, Nick. I missed that part of Ivica's question.
>
>
> To Ivica, I would say: you can usually find a description of the study
design and variables in the study manuals and reports. For example, the
2008 Survey is described in www.dzs.hr/Eng/Publication/2009/SI-1400.pdf
and contains the statement:
>
> "In the year 2008, the Household Budget Survey sample was selected in
two stages. In the first one there were 467 segments selected (segments
are groups of neighbouring enumeration areas) out of segments previously
selected for the 2008 Labour Force Survey. In the second stage, out of
each of selected segments, 10 dwellings, occupied by private households,
were selected, which were not previously selected into the 2008 Labour
Force Survey. Thus, 4 670 dwellings occupied by private households were
selected. At each selected occupied dwelling all private households were
interviewed."
>
> This says nothing about previous years, but the answer might lie in
reports about the Labor Force Survey. You should contact the Central
Bureau of Statistics. If they have a web site, the reports might be
there.
>
> A survey analysis which does not account for the design
variables-strata, PSUs- will produce standard errors and p-values which
are too small, usually much too small. Such an analysis assumes that the
observations are independent, whereas the effective sample size is more
the number of PSUs than the number of households.
>
>
> Regards,
>
> Steve
>
>
>
> On Sep 27, 2011, at 4:35 PM, Nick Cox wrote:
>
> PSUs are primary survey units.
>
> Nick
>
> On Tue, Sep 27, 2011 at 9:32 PM, Steven Samuels <[email protected]>
wrote:
>>
>> 1. Pool
>> 2. Divide weights by four (would affect only totals, but still a
good idea)
>> 3. -svyset- the whole thing.
>>
>>
>> Steve
>>
>> On Sep 27, 2011, at 10:40 AM, Ivica Rubil wrote:
>>
>> so, you're saying I just pool the 4 datasets, divide the weights by
4,
>> and apply -svyset- to the pooled dataset? Or should I rather use
>> -svyset- for each of the 4 datasets, pool them in one, and then
divide
>> the weights by 4?
>> Further, sorry for bothering you: what are PSUs? How do I check if
they
>> change form year to year?
>>
>> --
>> Ivica Rubil
>> Ekonomski institut / The Institute of Economics, Zagreb
>> Trg J. F. Kennedyja 7, 10 000 Zagreb, Croatia
>> tel. +385-1-2362-269
>> fax. +385-1-2335-165
>> [email protected]
>> www.eizg.hr
>>
>>
>> -----Original Message-----
>> From: [email protected]
>> [mailto:[email protected]] On Behalf Of Steven
>> Samuels
>> Sent: 27. rujan 2011 15:51
>> To: [email protected]
>> Subject: Re: st: weights in pooled repeated cross sections
>>
>>
>> Ivica-
>> You can use the individual weights, Ivica, but divide by 4 so that
they
>> sum to the average population total over the four years
>>
>> You still have to write the rest of the -svyset- command. If the
PSUs
>> did not change over the four years, then treat the pooled sample as
one
>> large sample, and use the same -svyset- statement that you would use
for
>> a single year. If some PSUs changed you will have to do some stratum
>> recoding. For an example see
>> http://www.cdc.gov/nchs/ppt/nchs2010/46_Moriarity.ppt
>>
>> Steve
>>
>> On Sep 26, 2011, at 12:03 PM, Ivica Rubil wrote:
>>
>> Dear all,
>>
>> I am trying to pool four repeated cross-sections of Croatian
Hpusehold
>> Budget Survey.
>> For each year that I want to pool, I have sampling weights for each
>> observation (both household and person).
>> My questions are: What should I do with the weights once I pool the
four
>> datasets? Is it wrong to use
>> dataset-specific weights in the pooled dataset and just run
estimation
>> commands with the weight option, if
>> available? I am confused. Please, help.
>>
>
> *
> * 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/
*
* 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/
*
* 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/