Hello,

I am working with others here on a tool that will help developing countries 
measure local poverty at lower cost by using a short scale.  We calibrate a 
short scale to an established scale on an older national survey and then 
conduct new surveys with the short scale.  To estimate the variances of 
estimates based on the short scale with the new surveys, I am using Rubin's 
formula for combining imputation variance (caused by not asking all scale 
items) with survey sampling variance (of survey 2 short scale).  The estimates 
of imputation variance come from imputing the full scale based on the short 
scale report in the second survey and the calibration model fit on the first 
survey.  Now I am looking for a rule of thumb for planning the size of the 
second survey.  Some sort of design effect due to increased measurement 
variance at the household level.  Anyone have any intuition on this or examples 
of similar work?  If one can obtain an 80% R-square for the full scale based on 
the short scale, would that suggest an extra design effect on the order of 1.2 
beyond design effects caused by any clustering and/or differential weighting?

--Dave Judkins
Abt Associates




________________________________
This message may contain privileged and confidential information intended 
solely for the addressee. Please do not read, disseminate or copy it unless you 
are the intended recipient. If this message has been received in error, we 
kindly ask that you notify the sender immediately by return email and delete 
all copies of the message from your system.

Reply via email to