There’s a linear equation that could define the model.
Something that would involve party affiliation, age, income level, turnout expectation, expectation to respond to polling calls, landline responder, cellphone responder etc.
Pollsters are smart guys. There’s big money to be made. Why can’t they stay ahead of this?
There are hard demographics like race, sex, age and the like. Then there are soft demographics like party affiliation.
Pollsters will stick to models based on hard demographics because it’s the safe thing to do. Theres a point where the risk of being wrong is outweighed by the risk of other alternatives.
Start adjusting a model based on party ID for instance and you can expect an unpleasant phone call from David Axelrod and an indictment from Attorney General Holder. Better to just stick with the hard variables and eat the loss of rep from a wrong call.