Letters to the Editor
-
Exactly correct
"If I were polling to find out how the Hagee endorsement hurt or helped John McCain, I might over sample the Catholic vote."
There are two phases to poll estimation. Phase 1: Estimation of subgroup belief. Phase 2: Estimation of population preference or belief. Subgroups are OFTEN over-sampled, to get a more accurate estimate of the beliefs of that subgroup. However when we form the composite to represent the population as a whole, the results from the subgroup will be weighted to fairly represent their proportion in the population as a whole.
This happens with AA, hispanics, etc. It also happens with Dems and Republicans. Very often, the poll determines if you are a Dem or Rep, and computes a subgroup belief there. Then, the beliefs are weighted to show the population as a whole.
This is one IMPORTANT reason why polls often differ. It is not just the simple sampling difference, where different polls get different results because the people are different. Rather, the sampling weights indicating proportion of population are different. Even the same poll results in hands of different pollsters could result in different conclusions.
That's why I call it a black art. It's not scientific in the standard statistical sense, because you incorporate the population weights, which are also unknown and estimated. The process involves weighting estimates with other weights which are also estimates. Somewhat convoluted.