My sense is that "generic ballot" survey questions -- which pit the incumbent against the "generic" opponent of the other party -- have some utility in that people tend to vote by party brand in recent years. OTOH, I suspect they are a better measure of dissatisfaction than a predictor of actual voter behavior -- especially where highly polarized primaries produce more "extreme" (rather than "generic") candidates.
Would love to know of any research on the subject.
