Re: st: programming: automate identification of strata with one psu

[Steven Samuels <samplerx@earthlink.net>]

On Dec 16, 2006, at 8:39 PM, Brent Fulton wrote:

> Lastly, is there another way to estimate the standard errors using  
> Stata (or
> using another statistical program e.g., SUDAAN; or bootstrap) when a
> stratrum has a single psu.
1. There are methods for computing standard errors when one PSU is  
drawn from stratum each BY DESIGN.  In the simplest approach,  
adjoining strata are collapsed for the purpose of estimating the  
standard errors (the observations retain their original weights).  
This process results in conservative standard errors.

2. With so much missing data--enough to reduce the number of PSU's in  
some strata to 1 so that the percent missing in the stratum is >50%--  
you have potentially serious bias.  The collapsed stratum approach  
will not solve the missing data bias problem.   The magnitude of the  
bias problem will depend on the relative sizes of the problem  
strata.  If you have 50% missing data in strata which total only 5%  
of the population, the bias problem MIGHT be tolerable, but this  
depends on the relation between missingness and your analysis variables.

3. You asked for a way of identifying and omitting the  problem  
strata.  However ignoring them will lead to biase, to underestimation  
of the true standard error, or to both.

4. You can directly address the missing data problem by multiply- 
imputing the missing values.  Do a search for "ice"

5. With good external information on the sampled population, sample  
raking might reduce some of the bias from missing data.  Search on  
"raking" for Nick Winter's "SURVWGT" package.

In sum, there is no automatic solution to your problem. You will need  
knowledgeable assistance with all of the suggestions above. You have  
a great resource in the Survey Research Center at Berkeley.

Regards,

Steve


*
*   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/