It's one explanation for the big gap between the final polls and the result:

The term Bradley effect or Wilder effect refers to a phenomenon which has led to inaccurate voter opinion polls in some American political campaigns between a white candidate and a non-white candidate. Specifically, there have been instances in which statistically significant numbers of white voters tell pollsters in advance of an election that they are either genuinely undecided, or likely to vote for the non-white candidate, but those voters exhibit a different behavior when actually casting their ballots. White voters who said that they were undecided break in statistically large numbers toward the white candidate, and many of the white voters who said that they were likely to vote for the black candidate ultimately cast their ballot for the white candidate. This reluctance to give accurate polling answers has sometimes extended to post-election exit polls as well.

Researchers who have studied the issue theorize that some white voters give inaccurate responses to polling questions because of a fear that they might appear to others to be racially prejudiced. Some research has suggested that the race of the pollster conducting the interview may factor into that concern. At least one prominent researcher has suggested that with regard to pre-election polls, the discrepancy can be traced in part by the polls' failure to account for general conservative political leanings among late-deciding voters.

Tonight is the first primary - not a caucus. People get to vote in a secret ballot - not in front of their largely liberal peers, as in Iowa. They may have told the pollsters one thing about voting for a black man, but in the privacy of the voting booth, something else happens. I don't have any hard evidence for this, but the discrepancy in the polls is remarkable. David Kuo cites it. The vast discrepancy between the last polls and the result puts it on the table. I hope it's not true. But it could be.