Raoul Kamadjeu wrote: > Can someone tells me more on the design effect?
As far as I can tell, it's smoke and mirrors. I say this not to be critical, but so that someone will fill in this gap in my knowledge. When designing a survey, the question inevitably arises about how many subjects to interview. The answer is based on variance considerations. Problem is that variance estimates are usually available only for simple random samples and the survey is some clustered-or-other thing. No one knows how to calculate the variance under the the actual sampling plan, so (here comes the smoke) they take the variance for a SRS and multiply it by the 'design effect'. The mirrors has to do with the fact that the design effect is invariably 2. Others, please feel free to educate me! . . ================================================================= Instructions for joining and leaving this list, remarks about the problem of INAPPROPRIATE MESSAGES, and archives are available at: . http://jse.stat.ncsu.edu/ . =================================================================
