Donald Trump's chances of winning are nearly twice what the polls have been predicting. Furthermore, this is amplified by the intensity of his support which will see much higher turnout rates for Trump supporters than Clinton supporters at the election.

The election is still wide open.





http://www.zerohedge.com/news/2016-10-22/statistician-warns-americans-ignore-cap...

Quote:
Do you see anything wrong there?  If you don't, then you have no business being around polling data.

The average margin of error on these 7 spreads shown is only 3%.  Most polls should therefore be within a few percent of the 6% average spread that is advertised by media.  But instead most are not!

For example, the difference between the highest Ms. Clinton spread and the lowest Ms. Clinton spread is >14 percentage points!  And the standard deviation among these mainstream polls is 5%.  So both have to be added together, and each is already higher than 3%!  That's an unusual, impossible outcome through luck alone.  Therefore something is misrepresented in the polls.

Also right now 2 of the 7 polls favor Donald (you just' don't hear about them), so double the 10-15% odds he is being given.  In the final analysis of this trinomial data, on November 9 we'll look back and see only one poll being correct and most were flat out wrong.  This evidence below is a breach of the probability theory behind proper polling, where most polls should see the correct spread within the margin of error interval (that's what the interval's definition must be!)  If the margins are therefore completely busted, then so too are the egregious spreads that are seen to be all over the place (and mostly untrustworthy).  Likely the correct expected spread right now is 4-5%, and the larger spreads are coming from pollsters that ironically also have the highest margin of errors (casting further suspicion on how close the election really is for Americans).  We stand by our long-running estimate that the current probability for a Donald Trump victory is about in the 20% range, or twice what mainstream media is projecting.  Of course that is low, but to some it's still a compelling 1 in 4 chance (and much different than some might expect given all the twists and turns this campaign season has brought us).  It's also a better reflection of the true odds, versus those dished out by the same inane talking heads who recently gave you the Brexit "remain" prediction, or the NeverTrump prediction!

So there you have it as clearly shown as possible.  If these margins of error are correct, then most polls would have the spreads located within a few percent of 6% (so 3%, to 9%).  Yet the majority of the polls are outside of this 3%, to 9%, interval.  Probabilistically impossible.  The idea that whatever the correct spread is determined to be on November 9, we will ultimately prove -shockingly we might add- that one poll was correct but also that most of these other polls were wrong.  Those polls (unsure right now which) are because there the correct spread will have been outside of their margin of error intervals.

The only correction anyone can make now to the failed margin of error is to enlarge it, in order to encapsulate most of the other intervals about the correct spread.  Without these overlaps, we can't discuss spreads in the media, since the data is from an entirely corrupt polling system!  The direction of unbiasing the data is also obvious.

To start with, the only correct expected value for the spread has to be reduced since that is the direction of asymmetric bias.  The largest polling spreads have become too extreme and must be brought in already.  Combined with larger margins of error.  The result of this combination is a correct spread that is lower at about 4-5%, and a margin of error that is roughly double what's been advertised (5-6%).  Implying Donald Trump's chances of winning is nearly twice what the mainstream media's been floating around... [/quote]

You might like to look at zerohedge.com, where an article highlights how the most recent poll by WaPo/ABC has been skewed out of all relevance by sample selection.