Unfortunately, I can’t find any write-up of the full #s for this, but the NY Times was given access to a poll conducted by Civis Analytics. What’s unique about this poll is that they used an actual list of registered voters, and checked whether the people had actually voted before in past elections. Thus they can define “likely voters” by whether the person has actually voted, rather than just relying on the honesty of the respondent as to whether they’re voters. The poll was conducted Aug. 10-19.
Nate Cohn mentions just a few results here:
http://www.nytimes.com/2015/08/23/us/politics/why-donald-trump-wont-fold-polls-and-people-speak.html?ref=topics&_r=0For all registered voters who identified as Republicans, the poll found:
Trump 16%
Carson 11%
Bush 10%
Those are the only names mentioned in the story. But Cohn also says that the support of Trump, Bush, and Carson is higher among people who haven’t actually voted in recent elections, and lower among those who have. I guess that means that several other candidates are in the opposite situation (more support among people who actually vote than those who don’t).