[Achilles]

The Bradley effect itself has been considered by many to have irrelevant or negligible since the early 90's. The median range for poll error changed from -4% (iirc) for Bradley in 82 to a miniscule positive by the early 90's. The data I've seen in that regard encompassed many elections and would seem to be pretty solid. I really only referenced Bradley because it's a good example of an unexpected poll deviation which is what I am predicting here, simply in the other direction.

As for pollsters, the vast majority of polls only ask you who you are voting for, not why you are doing so. I was much more making the point that people who had previously voted Bush/Republican now believe they made a mistake. People may not acknowledge this change of heart publicly, either because they simply don't like admitting mistakes (most people probably fall into this category to some extent) or feel a need to remain outwardly loyal to something they once genuinely believed in even if, perhaps, they no longer do. This could be for reasons of pride or perhaps to appear steadfast in front of other Republicans in their families or neighbourhoods. They may well be inclined to ignore such pressures in the privacy of the voting booth.