Poll problems.  I'm not talking about adjustments made for improper questions or demographic sampling techniques, though those do tend to bias the poll results in favor of one particular candidate.  (Guess who.)  No, I'm talking about the folks actually sampled, specifically registered voters vs. likely voters.

Most are aware of the RCP rolling average poll and the famous chart showing Clinton cleaning Trump's clock (though not by as much as she needs to at this point in the electoral cycle):

Problem is, I've noticed anecdotally for some time not just that the various polls in this rolling poll average were odd (even ignoring the poll construction flaws), but that those using registered voters (R.V.) only seem to usually favor Clinton – by a lot.  So I took the RCP poll history; deleted all R.V. polls, which still leaves a lot of likely voter (L.V.) polls; and plotted the result.

Tells a different story, no?  Bottom line: using L.V.-only polls, Clinton is currently 44% and Trump is currently 42% with a similar 3% margin of error, not the 7% difference reported by the RCP rolling average poll.  Moreover, both of them are increasing right now, perhaps taking back some of the defections to third-party candidates of recent.

Moral: Beware polls in general – they are all flawed.  But especially beware any poll that samples just registered voters.  They tell you things like "Hillary's bounce remains firm."  Do you see a bounce stability in my chart?  So how can a statement like that be supported?   Because today ABC and the Washington Post released a new poll showing Clinton 8% ahead based on 815...registered voters.  Hmmmmm.