Its really staggering, they define a group of 3000 LVs by race gender etc, ask them who they are going to vote for, then ask them how they currently identify themselves, and arbitrarily throw out votes once numbers hit %3 dem. That is not polling. That is telling people what they think of themselves and telling them that only +3% think of themselves as dem at that moment.
I do not know where they got these folks, but they did not wander the halls of science. You don't start throwing out data once it pushes the edges of your model. That is not sampling. The US Army knows how to do sampling (seriously, their tables and stuff is very well regarded), they would never have conjured such a method.
I'm just wondering if they get it, or if they just think they need to pick a better target number to modify their data towards.
How about not modifying the data ... and actually measuring and recording instead?
| 