November 02, 2016

Horse 2185 - The Numbers Are Even More Vague Than They're Letting On

All indications thus far point to Hillary Clinton becoming the 45th President Of The United States. The polls that I have been watching over the last few weeks, point to her picking up between 44%-48% of the popular vote, with Donald Trump getting between 37%-45%.
The Presidential Election isn't a nationwide popular vote though. Owing to the fact that the United States uses a bizarre and arcane system called the Electoral College which apportions delegates according to the number of Congress members that each state has (and DC if it was a state), and most states in the Union operate on a winner takes all basis, it makes playing with the expected numbers of Electoral College votes really easy. Also because the United States has always been polarised for a long time, lots of states remain likely to vote for the same party's candidates.

As a result of all of this, the so called "battleground" states which will determine the swing of the election, really only boil down to Nevada, Ohio, North Carolina, Iowa and the biggest prize of Florida. If you watch those four states in particular, then the election result is easier to predict. Indeed in 2000, which Al Gore famously lost due to the ambiguousness of hanging chads in the punched cards which were used in the voting machines of Florida.

There is however one thing that remains difficult to predict and that is whether or not people are truthful when answering polls. There have been multiple studies which suggest that when people talk on pollsters, they might not be as likely to give an answer which is perceived as unpleasant. No racist is going to admit that they are racist to a pollster unless they are truly unapologetic. Likewise, no sexist is going to admit their misogyny or misandry openly. This is also known as the "Bradley Effect" after the 1982 gubernatorial election campaign of  Los Angeles Mayor Tom Bradley, an African-American, running in the state of California. People tended to say that they would vote for Tom Bradley to the pollster on the other end of the telephone but within the privacy of the polling booth, all of the racists could express their electoral opinion without any qualms.

What I don't know about this election is to what degree that the Bradley Effect will have. Never before has there been an election where the approval rating of both candidates been so appallingly low; and justifiably so. Donald Trump as a populist candidate, has consistently said things which under normal circumstances would have had him ejected from any major political party. Then there's the persistent and obvious shadow of everything that he's done; with the possibility that 48 days after the election (which is Boxing Day), that he could be embroiled in a rape case. Hillary Clinton is also again being reinvestigated as a result of her keeping a private email server during her time as Secretary Of State, and the possible destruction of classified information. There is a real possibility that either candidate might find themselves in prison, after winning the election.
Now I don't know what effect this has on voters but I suspect that it equates to non-zeroes in election polling results.

One of the websites that I have been looking at is Nate Silver's 538. I don't know what sort of algorithms that they use but they run many thousands of simulated elections and then make their predictions based upon the polling data from multiple sources. The thing is though that all of the polls which go into creating the simulated elections, come from either people speaking to pollsters and are thus subject to the Bradley Effect, or are online polls which tend to be self-selecting and thus contain their own inherent biases. I think that the ranges which are being reported by polls are far far fuzzier than in any previous election.

As of this morning, the chances of the winner of the vote in Ohio was predicted as Clinton 49.7% and Trump 50.3%. That would suggest that Ohio is sitting on a knife edge but the world inside predictive models and data sets which are tainted by the Bradley Effect, could be wildly different to the one on November 8 when voters are all alone in the voting booth, with the ballot in front of them. All the secret racists and sexists who wouldn't dare speak a word in the light of day, might speak entirely differently. 50.3%-49.7% which is a margin of just 0.6% could very well be 58%-42% which is 16%; in either direction.
If this is true for Ohio, then the path to the Presidency for Donald Trump might be easier than previously expected. If all of the purple states and toss up states were in favour of Trump, then playing with the Electoral College map gives several ways to get to 270. It also could be true that this is just election hype, perpetuated by news outlets to sell copy too. We shouldn't believe what is being presented because at best, all we have are guesstimates. I just don't know about the truthiness of these numbers, it all seems a bit like peering into a fog.
If you play with fuzzy numbers, you can't expect anything other than fuzzy results.

No comments: