Dear All
I have a population of 200 clusters accounting for 3000 individuals. I
know the intra-class correlation is comparatively high (say .4 as
measured using McNemar's 1957 formula) from earlier research sampling 50
of the clusters. Now, if I were to conduct a census of all 3000
individuals in all 200 clusters what would be the effective sample size?
If I use Kish's formula for the design effect (Design effect = [1 + roh
(B - 1)]) I estimate a design effect of 6.6 and an effective sample of
455.
But here's my question - Do I have to factor in some sort of finite
population correction in the design effect calculation in the same way I
would when calculating a SE? Clearly in this case, if I were to sample
everybody, the SEs would be zero.
Regards
David Millar
ERC
Dublin
|