I know this topic has been posted on in the past.  But it keeps coming up in comment threads.  Sam Wang recently wrote up an analysis based on Dan Hopkins's study released in August, which also brought it back to my mind right at the moment.

Rather than rely on case studies to document the purported effect, Hopkins analyzes polling data and election results from gubernatorial and Senate races over the last 17 years.  The study finds that the Bradley effect--overstatement in polling results of support for a non-white candidate because of latent or unreported racial bias--has virtually disappeared during the last decade or so.

One question I find interesting is whether a presidential (general) election presents variables not considered by this study.  (For example, do poll respondents feel different pressures when responding to a national polling outfit in a high-profile presidential election than they would in, say, a statewide gubernatorial race?)  Of course, historical data for that scenario aren't available . . .

Anyway, I thought the links might be worth sharing with those interested.