Re: st:non-constant Frequency weights

[Steven Samuels <sjsamuels@gmail.com>]

--
Stephen asked me privately for a reference to weighting  panel data..  
The method I suggested has been used in many studies, for example the  
Survey of Labour and Income Dynamics of Statistics Canada (http://dsp-psd.pwgsc.gc.ca/Collection/Statcan/75F0002MIE/75F0002MIE2004008.pdf 
, page 7)

However the computation of weights in panel surveys can be complicated  
(Johnson, 2008; Lynn, 2009, p 12). The weights for the first survey  
wave might be cross-sectional, constructed so that the data represent  
the population at the time.  But, with attrition, the respondents to  
the first survey who respond subsequently might no longer be  
representative of the original population.  To correct for this, some  
surveys create longitudinal weights in addition to the cross-sectional  
ones. To be sure he is using the correct weight, Stephen's safest  
recourse is to consult the documentation for his data set.

David Johnson (2008), Using Weights in the Analysis of Survey Data
http://help.pop.psu.edu/help-by-statistical-method/weighting/Introduction%20to%20survey%20weights%20pri%20version.ppt

Peter Lynn (2009) Methodology of Longitudinal Surveys, Wiley, NY.

Steve

Steven J. Samuels
sjsamuels@gmail.com
18 Cantine's Island
Saugerties NY 12477
USA
Voice: 845-246-0774
Fax:    206-202-4783



>
>To answer your question: If the panel is based on a sample of the  
population in the first year, then use that first year weight.  If the  
panel is a rotating panel, with individuals entering in different  
years, than for those individuals, use the weight for their first  
year.  The weights in different years are used for presenting cross- 
sectional statistics for each of the years, but not for the panel  
analysis.
>
>Also, the weights should  be specified as probability weights, not  
frequency weights.  A frequency weight equal to K, say, is intended  
for use when there are K individuals in the _sample_  with identical  
values.  If you tell a command that the weights are frequency weights,  
then you are saying that the sample size is equal to the population  
size.
>
>Regards,
>
>Steve
>
>On Oct 26, 2010, at 4:34 PM, Stephen O Neill wrote:
>
>Hi Steve,
>My data is from a survey which contains weights representing the  
number of
>inidividuals in the population represented by each sample  
observation. However
>in differnt years this frequency weight is different i.e. the number  
of
>population individuals the sample observation represents changes  
from year to
>year.
>Hope this clarifies,
>Stephen
>
>
>
>----- Original Message ----
>From: Steven Samuels <sjsamuels@gmail.com>
>To: statalist@hsphsun2.harvard.edu
>Sent: Tue, October 26, 2010 7:40:47 PM
>Subject: Re: st: RE: non-constant Frequency weights
>
>
>Stephen O Neill:
>
>Off-topic, but I'm curious:  How can you have repeated identical  
observations on
>one individual at a single time point, which is what a frequency  
weight>1 would
>imply?
>
>Steve
>sjsamuels@gmail.com
>
>
>On Oct 26, 2010, at 2:28 PM, Nick Cox wrote:
>
>Fair enough. Now you have a specific question, to which I don't know  
the answer;
>but someone else may have specific advice.
>
>Nick
>n.j.cox@durham.ac.uk
>
>Stephen O Neill
>
>When I try to run - xtreg,fe - I get the error message "weight must  
be constant
>within ID", where ID is my unique identifier for each individual and  
is constant
>for an individual over time. I also wanteed to use the user written  
- xtabond2 -
>or some other dynamic model but it appears that weights can't be  
used with
>these.
>
>Nick Cox <n.j.cox@durham.ac.uk>
>
>I don't see the connections between these statements. A variable  
containing
>frequency weights can vary between observations, in this instance  
within panels
>as well as between panels, so that is not a problem in itself.
>
>Perhaps some particular procedures have more restricted assumptions,  
but you
>don't say what they are.
>
>Stephen O Neill
>
>I am using unbalanced panel data which contains a frequency weight  
each year.
>However the weight is not constant through time with the result that  
I cannot
>weight the data using - [fweight = weight] - . I was just wondering  
if there is
>a simple work around?
>
>
>*
*
*   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/