Noam Scheiber points to working paper, SOCIAL DESIRABILITY BIAS IN ESTIMATED SUPPORT FOR A BLACK PRESIDENTIAL CANDIDATE, which attempts to figure out the Bradley Effect by guaging avowed vs. implied support. Mark Blumenthal of Myster Pollster has an interview where one of the authors explains the methodology and touches upon some confusing issues....
To a great extent I think this is the sort of thing which should give us caution about overreading from social science; you might not be smoking out the dynamics which you think you are smoking out. I've put the results from the most relevant table below the fold; remember that "True Support" is a calculated projection, and if you are curious about p-values just go to the paper itself.
