Re: st: svyset question

[Steven Samuels <ssamuels@albany.edu>]

Jeff, I recommend that you consult a sampling statistician.  You have:

CODE PROBLEMS

1. With only one district selected within each stratum, STATA has no  
replicates with which to compute a SE. You have several choices: 1.  
Combine neighboring regions to get three strata. Be warned that the  
standard error multiplier for t-statistics will be: 3.18 (compared to  
1.95 for a normal approximation).  2. If regions do not differ much- 
but how can you tell with only one obs per region?-then omit the  
stratum specification and get 6-1 5 degrees of freedom for the  
highest level of sampling. The t-multiplier in this case is 2.57,  
about 20% less.  Personally I would go with 90% intervals, so that  
the t-multiplier is 2.35 for 3  strata.

2. The weight variable must be specified in a [pweight=  ] statement  
before the comma.

CONCEPTUAL PROBLEMS

3. It looks like there are 3-4 stages of sampling:  1. District in  
strata. 2. village/ward within district. 3. hh within village/ward.  
4. person within hh.  Only the first 3 would be specified if  only  
one person is selected from each hh.

4.  It looks like there was a second level of stratification- urban  
vs rural- someplace in the design.   Your description  "PPS" sampling  
makes no sense unless this is true.

ANALYTIC ISSUES.
5. In a survey of this size, especially with no replication at the  
first stage, some post-stratification or sample raking would be  
standard practice.


Regards,

Steven

On Apr 20, 2007, at 10:24 AM, Jeff Edmeades wrote:


> Hi all,
>
> I am working with survey data with the following design (as  
> described to
> me): "Respondents were selected through stratified cluster sampling,
> with one district randomly selected from six geographic regions.  Ten
> sampling units (villages in rural areas and urban wards in urban  
> areas)
> were then selected in each district through probability  
> proportional to
> size sampling, with purposeful oversampling of urban areas to ensure
> sufficient cases for the analysis of rural-urban differences.  A
> household listing was conducted in each of the sampling units, from
> which 40 eligible individuals were randomly selected."
>
> My understanding of the correct syntax for this (following from the  
> SVY
> manual) is:
>
> svyset district, strata(region) fpc(ndistricts) || samplingunit
> [sampling weight for urban oversample] fpc(nsamplingunits)
>
> Is this correct??
>
> Many thanks,
*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/