Re: st: oversampling correction

[Steven Samuels <sjsamuels@gmail.com>]

-
Nicola, look up the sections on weights in a good sampling book, for  
example Robert Groves et al., Survey Methodology; Sharon Lohr,  
Sampling: Design and Analysis;  Chapter 16 of Levy & Lemeshow,  
Sampling of Populations.

Steve

On Nov 11, 2010, at 12:30 PM, Stas Kolenikov wrote:

There was a paper in the most recent issue of Stata Journal doing
exactly what you need. See
http://stata-journal.com/article.html?article=st0196. It has its own
set of ideas in mind, but I am sure you'll be able to twist its hands
to produce weights calibrated to your totals, means or fractions
(there's an example in the paper to help you out with that). Out of
all the sampling design intricacies, you would only need to specify
weights.

On Thu, Nov 11, 2010 at 9:41 AM, Nicola Baldini <n.baldini@unibo.it>  
wrote:
> I repeatedly read some authors whose surveys received replies from a  
> sample somewhat different from the population reporting that they  
> corrected for oversampling. What does it mean (i.e. can I do the  
> same in Stata and how)? I thought it was so simple I cound find a  
> solution in the FAQ, and also a search on previous posts did not  
> provide a statisfying reply (may be this is because my survey is so  
> simple that does not have a sampling design). Indeed, I have a  
> universe of 710 individuals, but I could find the emails for sending  
> my survey only for 510 (my population) and I received a reply only  
> from 210 (my sample). The survey is a 12-item 5-point Likert scale  
> from 1 to 5. The statistical analyses include -ttest-, -anova- and - 
> factor- on Stata 9.2
> PROBLEM 1: Women are 26.5% in the population but only 16.3% in my  
> sample: how can I correct for this?
> PROBLEM 2: The population is divided in 10 organisations of unequal  
> size, and each organisation is divided in 2 groups of unequal size  
> (or it is the reverse: there are two groups of unequal size - e.g.  
> based on traveling at least once in a life to Africa or not -, and  
> group members may belong to 10 organisations). The response rate  
> varies at the organisation and at the group levels: can I correct  
> for this? and how? Can I transform the data so that I have the same  
> number of respondents in each group (e.g. 5 respondents per group x  
> 2 groups x 10 organisations = 200 respondents)? Also: can I  
> transform the data so that I have the same response rate (e.g. 40%)  
> in each group?

--
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.

*
*   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/