No Newspoll today for some reason (the LP community points out the obvious – NSW public holiday. Duh!) , so instead a quick comparative look between polling and Intrade trends for estimating the number of Electoral College votes for each party in the US Election. There is an excellent site on the US election called the Princeton Election Consortium which does quite brilliant things with polling data from Electoral-Vote Pollster.com. It’s sort of like FiveThirtyEight – but a little more robust with the data as a result of not trying to go down the forecasting route.
Over on their “For fellow geeks” page, they have supplied all the bits and pieces needed to replicate their work. What struck me though as I was looking through their stuff was how their Electoral College vote allocation from polling data walked hand in hand with the Intrade data we’ve been running here on our US Election page.
A while ago I promised a Poll Bludger by the name of Jaundiced View that I’d run some stuff on comparing the polls to the Intrade markets – now that I’ve discovered the Princeton data, this is the first part of that process.
If we chart the Princeton Electoral College vote numbers for the Democrats since early September and compare it to both the Intrade State market EV allocations (where each State that the Democrats have been predicted to win at Intrade at any given time, those Electoral College votes get tallied up to the Democrats) as well as the median Electoral College vote result from our daily State Intrade market simulations - we get the promised Jaundiced View Intrade Polling Confluence.
Just click to expand.
All the metrics here clearly move together, but what’s interesting is that previously the polls and the market sims either broke out together, or the market sims generally broke out first with the polls quickly playing catch up. Yet recently that’s been reversed to some extent and it’s been the polls that have broken out first where the market sims have still yet catch up.
The actual raw EV allocations from the State markets are harder to decipher because of a few States jumping over and under 50% probability and their EV’s creating all those peaks you see with the black line, where often those jumping States move opposite to the broader Intrade market on a given day.
Later on in the week, we’ll have a look at this over a longer time span than the last month that we’ve done here, and some rather interesting issues pop up.
Another thing worth noting about the State Intrade markets is the probability spread between the States. Over on the US Election page we show the State probabilities like this:
It’s clean and simple, but also slightly deceptive as it doesn’t do the actual State of play in the market any visual justice whatsoever. If we redo that chart with a fixed probability scale, we end up with this instead:
That chart really does hammer home exactly what is going on, with each Party having a bunch of safe States and a very small number of effectively contestable States (contestable according to Intrade) sparsely populating the middle probabilities – yet where those States in the middle continue to polarise to each end of the probability spectrum and where the Democrats are enjoying a stronger pull on those middle probability States than the Republicans.
UPDATE:
Someone with a good sense of humour sent in the following image (clickable)
12 Comments
“Yet recently that’s been reversed to some extent and it’s been the polls that have broken out first where the market sims have still yet catch up.”
That’s pretty much what you’d predict. Far from an election, when polls are very rare, markets will be leading indicators, but the closer it gets to the election the more frequent the polls, and markets simply bet based on the most recent polling data.
That last graph is really quite clever, and informationally dense. It displays a lot of information in a particularly simple way to illustrate the situation very well.
Adrian,
I suppose it is what you’d expect when you think about it. I’m surprised a little though since there seems to have been be a fair bit of insider activity on the Intrade state markets previously where a State will jump a few points to be followed by a poll or two a few days later backing the movement up. Things like that.
In Australia, oddly enough, the opposite happened last year where the polls led the way a long way out and where it took the headline betting markets a while to catch up, and where the individual seat markets didnt really get their act together until the last week or so.
PJN, I was playing around this morning trying to find a way to highlight the probability gap between the two big clusters of states – that seemed to do the trick!
Glad you like
Possum
May I humbly suggest that “consilience” would be a more apposite word than “confluence”?
Consilience, or the unity of knowledge (literally a “jumping together” of knowledge), has its roots in the ancient Greek concept of an intrinsic orderliness that governs our cosmos, inherently comprehensible by logical process, a vision at odds with mystical views in many cultures that surrounded the Hellenes.
Dio’s word of the day
Like the last graph a lot, Poss. Easy on the eye and tragics can instantly tell what States each black vertical line represents by the ECV numbers.
Just two words: neato keeno:)
Nifty neologism, Dio.
I cannot help but get the feeling that polls are less accurate in a voluntary voting system. I feel that McCain will get a hammering from GOP voters who will not be bothered getting off their Jason recliners come election day while Obama supporters will be bushy tailed.
Poss, do you have any info on the accuracy of polling in past US elections?
It’s pretty hit and miss Kit. The national tracking polls are usually right – well most of them anyway, but the State polls get all over the shop.
Pretty much like what we’re seeing at the moment.
Mr Possum, could you comment on what you think may be the impact of the Bradley Effect:
“The term refers to Tom Bradley, a former black mayor of Los Angeles, who lost his 1982 bid to become governor of California, even though every poll in the state showed him leading his white opponent by substantial margins. Similar results appeared in 1989, when David Dinkins ran for mayor of New York City and Douglas Wilder sought election as governor of Virginia. Dinkins was ahead by 18 percentage points, but won by only two, and Wilder was leading by nine points, but squeaked through by only half a percent. Numerous other examples lead Hacker to offer this advice to Obama campaign offices: always subtract 7% from favorable poll results. That’s the potential Bradley effect.”
From Is This Election the Major Historical Turning Point It Seems to Be? Yes
By Chalmers Johnson, Tomdispatch.com. Posted October 8, 2008.
http://www.tomdispatch.com/post/174987/chalmers_johnson_the_ultimate_election
Bernice, in no way do I attempt to pre-empt the thoughts of Teh Furry One, however, below is the portfolio of 538 Nate’s analysis of “The Bradley Effect”. The short take is that Nate Silver doesn’t think that it will play very much in 2008, and if it does, he believes it will be offset by Team Kid’s “ground game”.
http://www.fivethirtyeight.com/search/label/bradley%20effect
And yes, was likewise appalled, but not surprised at Mc/Palin’s tacit nods to bigotry you mentioned on LP. Cheers, EC.
Bernice,
The link Ecky gave just about sums it all up except to say that if there is a Bradley Effect operating in a handful of States at a few percentage points, I think it will be almost impossible to pick up because it will probably be washed away by turnout on the one hand and the cell phone effect and the increase in the 18-35 year age group Democrat vote on the other.
Anyone that says Obama should subtract 7% from poll results – 7%!! – is quite frankly off with the fairies.We saw nothing like that in the primaries and nothing like that in the last 20 years of US politics.
Thanks for remembering to do up the Confluence chart Possum. It looks terrific. I also hope there is good punting on the state by state elections given what you say in 3 above!