Have Trump's Numbers Been Overstated in the Polls?

According to the headlines, we're seeing a surge in the polls for Ben Carson, who has now over-taken Donald Trump as the new front-runner. But as soon as I saw that, I was reminded of something Steve Koczela wrote last week.

Donald Trump's poll numbers are slipping in Iowa, and a new national poll is the first in a while to show him trailing. But a closer look at the polling suggests that the Trump wave may have been overstated from the beginning. His sizeable lead has been based largely on the influence of Internet polls. Trump’s summer surge looks far less impressive in telephone polls, and polls of likely voters show his lead was always smaller and is now gone entirely.

It is interesting to note that the NBC/WSJ poll that is driving all the headlines was one of likely voters.

Here's a summary of what Koczela found from the different types of polls:

Koczela doesn't assume that we know which type of polling will turn out to be more accurate. But he does suggest this:

One potential explanation is that online polls may be sampling more non-voters. These non-voters may not be paying close attention and may simply select the candidate closest to the top of their mind, which could be Trump because of his heavy media coverage. This top-of-mind phenomenon might also explain why Trump is doing better in phone polls of registered voters than in polls of likely voters.

I just did my own check at the Huffington Post poll aggregator to see how this might be playing out.  Going back to mid-October, Trump's average among registered voters is 31.6% while his average among likely voters is 25.2%, a difference of -6.4%. For Carson, his average among registered voters is 20.5% and among likely voters it's 24.6%, a difference of +4.1%.

So, if you think that polls of registered voters are more accurate, Trump's lead is still about 11 points over Carson. But if you assume that likely voters are a better read right now, they're basically tied - and have been for awhile now.